targets/gba: implement interrupt handler

Make it clearer where the stack is located.

.circleci/config.yml

@@ -18,7 +18,6 @@ commands:
             wget -O - https://apt.llvm.org/llvm-snapshot.gpg.key|sudo apt-key add -
             sudo apt-get update
             sudo apt-get install \
-                python3 \
                 llvm<<parameters.llvm>>-dev \
                 clang<<parameters.llvm>> \
                 libclang<<parameters.llvm>>-dev \
@@ -44,12 +43,12 @@ commands:
     steps:
       - restore_cache:
           keys:
-            - llvm-source-8-v5
+            - llvm-source-9-v0
       - run:
           name: "Fetch LLVM source"
           command: make llvm-source
       - save_cache:
-          key: llvm-source-8-v5
+          key: llvm-source-9-v0
           paths:
             - llvm-project
   test-linux:
@@ -57,7 +56,7 @@ commands:
       - checkout
       - submodules
       - apt-dependencies:
-          llvm: "-8"
+          llvm: "-9"
       - install-node
       - restore_cache:
           keys:
@@ -65,9 +64,9 @@ commands:
             - go-cache-v2-{{ checksum "go.mod" }}
       - llvm-source-linux
       - run: go install .
-      - run: go test -v ./transform .
+      - run: go test -v ./cgo ./compileopts ./interp ./transform .
       - run: make gen-device -j4
-      - run: make smoketest RISCV=0
+      - run: make smoketest
       - save_cache:
           key: go-cache-v2-{{ checksum "go.mod" }}-{{ .Environment.CIRCLE_BUILD_NUM }}
           paths:
@@ -75,7 +74,7 @@ commands:
             - ~/.cache/tinygo
             - /go/pkg/mod
       - run: make fmt-check
-  build-linux:
+  assert-test-linux:
     steps:
       - checkout
       - submodules
@@ -83,7 +82,6 @@ commands:
           name: "Install apt dependencies"
           command: |
             sudo apt-get install \
-                python3 \
                 gcc-arm-linux-gnueabihf \
                 binutils-arm-none-eabi \
                 libc6-dev-armel-cross \
@@ -101,7 +99,65 @@ commands:
       - llvm-source-linux
       - restore_cache:
           keys:
-            - llvm-build-8-linux-v7
+            - llvm-build-9-linux-v0-assert
+      - run:
+          name: "Build LLVM"
+          command: |
+            if [ ! -f llvm-build/lib/liblldELF.a ]
+            then
+              # install dependencies
+              sudo apt-get install cmake clang ninja-build
+              # make build faster
+              export CC=clang
+              export CXX=clang++
+              # hack ninja to use less jobs
+              echo -e '#!/bin/sh\n/usr/bin/ninja -j3 "$@"' > /go/bin/ninja
+              chmod +x /go/bin/ninja
+              # build!
+              make ASSERT=1 llvm-build
+            fi
+      - save_cache:
+          key: llvm-build-9-linux-v0-assert
+          paths:
+            llvm-build
+      - run: make ASSERT=1
+      - run:
+          name: "Test TinyGo"
+          command: make ASSERT=1 test
+      - save_cache:
+          key: go-cache-v2-{{ checksum "go.mod" }}-{{ .Environment.CIRCLE_BUILD_NUM }}
+          paths:
+            - ~/.cache/go-build
+            - ~/.cache/tinygo
+            - /go/pkg/mod
+      - run: make gen-device -j4
+      - run: make smoketest TINYGO=build/tinygo
+  build-linux:
+    steps:
+      - checkout
+      - submodules
+      - run:
+          name: "Install apt dependencies"
+          command: |
+            sudo apt-get install \
+                gcc-arm-linux-gnueabihf \
+                binutils-arm-none-eabi \
+                libc6-dev-armel-cross \
+                gcc-aarch64-linux-gnu \
+                libc6-dev-arm64-cross \
+                qemu-system-arm \
+                qemu-user \
+                gcc-avr \
+                avr-libc
+      - install-node
+      - restore_cache:
+          keys:
+            - go-cache-v2-{{ checksum "go.mod" }}-{{ .Environment.CIRCLE_PREVIOUS_BUILD_NUM }}
+            - go-cache-v2-{{ checksum "go.mod" }}
+      - llvm-source-linux
+      - restore_cache:
+          keys:
+            - llvm-build-9-linux-v0
       - run:
           name: "Build LLVM"
           command: |
@@ -119,15 +175,9 @@ commands:
               make llvm-build
             fi
       - save_cache:
-          key: llvm-build-8-linux-v7
+          key: llvm-build-9-linux-v0
           paths:
             llvm-build
-      - run:
-          name: "Create LLVM symlinks"
-          command: |
-            ln -s $PWD/llvm-build/bin/clang-8 /go/bin/clang-8
-            ln -s $PWD/llvm-build/bin/ld.lld  /go/bin/ld.lld-8
-            ln -s $PWD/llvm-build/bin/wasm-ld /go/bin/wasm-ld-8
       - run:
           name: "Test TinyGo"
           command: make test
@@ -151,11 +201,6 @@ commands:
             tar -C ~/lib -xf /tmp/tinygo.linux-amd64.tar.gz
             ln -s ~/lib/tinygo/bin/tinygo /go/bin/tinygo
             tinygo version
-      - run:
-          name: "Download SiFive GNU toolchain"
-          command: |
-            curl -O https://static.dev.sifive.com/dev-tools/riscv64-unknown-elf-gcc-8.2.0-2019.05.3-x86_64-linux-ubuntu14.tar.gz
-            sudo tar -C /usr/local --strip-components=1 -xf riscv64-unknown-elf-gcc-8.2.0-2019.05.3-x86_64-linux-ubuntu14.tar.gz
       - run: make smoketest
   build-macos:
     steps:
@@ -164,8 +209,8 @@ commands:
       - run:
           name: "Install dependencies"
           command: |
-            curl https://dl.google.com/go/go1.12.5.darwin-amd64.tar.gz -o go1.12.5.darwin-amd64.tar.gz
-            sudo tar -C /usr/local -xzf go1.12.5.darwin-amd64.tar.gz
+            curl https://dl.google.com/go/go1.13.darwin-amd64.tar.gz -o go1.13.darwin-amd64.tar.gz
+            sudo tar -C /usr/local -xzf go1.13.darwin-amd64.tar.gz
             ln -s /usr/local/go/bin/go /usr/local/bin/go
             HOMEBREW_NO_AUTO_UPDATE=1 brew install qemu
       - restore_cache:
@@ -174,17 +219,17 @@ commands:
             - go-cache-macos-v2-{{ checksum "go.mod" }}
       - restore_cache:
           keys:
-            - llvm-source-8-macos-v5
+            - llvm-source-9-macos-v0
       - run:
           name: "Fetch LLVM source"
           command: make llvm-source
       - save_cache:
-          key: llvm-source-8-macos-v5
+          key: llvm-source-9-macos-v0
           paths:
             - llvm-project
       - restore_cache:
           keys:
-            - llvm-build-8-macos-v6
+            - llvm-build-9-macos-v0
       - run:
           name: "Build LLVM"
           command: |
@@ -196,13 +241,9 @@ commands:
               make llvm-build
             fi
       - save_cache:
-          key: llvm-build-8-macos-v6
+          key: llvm-build-9-macos-v0
           paths:
             llvm-build
-      - run:
-          name: "Create LLVM symlinks"
-          command: |
-            ln -s $PWD/llvm-build/bin/clang-8 /usr/local/bin/clang-8
       - run:
           name: "Test TinyGo"
           command: make test
@@ -235,19 +276,29 @@ commands:
 
 
 jobs:
-  test-llvm8-go111:
+  test-llvm9-go111:
     docker:
       - image: circleci/golang:1.11-stretch
     steps:
       - test-linux
-  test-llvm8-go112:
+  test-llvm9-go112:
     docker:
       - image: circleci/golang:1.12-stretch
     steps:
       - test-linux
+  test-llvm9-go113:
+    docker:
+      - image: circleci/golang:1.13-stretch
+    steps:
+      - test-linux
+  assert-test-linux:
+    docker:
+      - image: circleci/golang:1.13-stretch
+    steps:
+      - assert-test-linux
   build-linux:
     docker:
-      - image: circleci/golang:1.12-stretch
+      - image: circleci/golang:1.13-stretch
     steps:
       - build-linux
   build-macos:
@@ -262,7 +313,9 @@ jobs:
 workflows:
   test-all:
     jobs:
-      - test-llvm8-go111
-      - test-llvm8-go112
+      - test-llvm9-go111
+      - test-llvm9-go112
+      - test-llvm9-go113
       - build-linux
       - build-macos
+      - assert-test-linux

.gitignore

@@ -12,5 +12,5 @@ src/device/sifive/*.s
 src/device/stm32/*.go
 src/device/stm32/*.s
 vendor
-llvm
 llvm-build
+llvm-project

CHANGELOG.md

@@ -1,3 +1,125 @@
+0.11.0
+---
+* **command line**
+  - add support for QEMU in `gdb` subcommand
+  - use builtin Clang when building statically, dropping the clang-9 dependency
+  - search for default serial port on both macOS and Linux
+  - windows: support `tinygo flash` directly by using win32 wmi
+* **compiler**
+  - add location information to the IR checker
+  - make reflection sidetables constant globals
+  - improve error locations in goroutine lowering
+  - interp: improve support for maps with string keys
+  - interp: add runtime fallback for mapassign operations
+* **standard library**
+  - `machine`: add support for `SPI.Tx()` on play.tinygo.org
+  - `machine`: rename `CPU_FREQUENCY` to `CPUFrequency()`
+* **targets**
+  - `adafruit-pybadge`: add Adafruit Pybadge
+  - `arduino-nano33`: allow simulation on play.tinygo.org
+  - `arduino-nano33`: fix default SPI pin numbers to be D13/D11/D12
+  - `circuitplay-express`: allow simulation on play.tinygo.org
+  - `hifive1-qemu`: add target for testing RISC-V bare metal in QEMU
+  - `riscv`: fix heap corruption due to changes in LLVM 9
+  - `riscv`: add support for compiler-rt
+  - `qemu`: rename to `cortex-m-qemu`
+
+0.10.0
+---
+* **command line**
+  - halt GDB after flashing with `gdb` subcommand
+  - fix a crash when using `-ocd-output`
+  - add `info` subcommand
+  - add `-programmer` flag
+* **builder**
+  - macos: use llvm@8 instead of just llvm in paths
+  - add `linkerscript` key to target JSON files
+  - write a symbol table when writing out the compiler-rt lib
+  - make Clang header detection more robust
+  - switch to LLVM 9
+* **compiler**
+  - fix interface miscompilation with reflect
+  - fix miscompile of static goroutine calls to closures
+  - fix `todo: store` panic
+  - fix incorrect starting value for optimized allocations in a loop
+  - optimize coroutines on non-Cortex-M targets
+  - fix crash for programs which have heap allocations but never hit the GC
+  - add support for async interface calls
+  - fix inserting non-const values in a const global
+  - interp: improve error reporting
+  - interp: implement comparing ptrtoint to 0
+* **cgo**
+  - improve diagnostics
+  - implement the constant parser (for `#define`) as a real parser
+  - rename reserved field names such as `type`
+  - avoid `"unsafe" imported but not used` error
+  - include all enums in the CGo Go AST
+  - add support for nested structs and unions
+  - implement `#cgo CFLAGS`
+* **standard library**
+  - `reflect`: add implementation of array alignment
+  - `runtime`: improve scheduler performance when no goroutines are queued
+  - `runtime`: add blocking select
+  - `runtime`: implement interface equality in non-trivial cases
+  - `runtime`: add AdjustTimeOffset to update current time
+  - `runtime`: only implement CountString for required platforms
+  - `runtime`: use MSP/PSP registers for scheduling on Cortex-M
+* **targets**
+  - `arm`: add system timer registers
+  - `atmega`: add port C GPIO support
+  - `atsamd21`: correct handling of pins >= 32
+  - `atsamd21`: i2s initialization fixes
+  - `atsamd51`: fix clock init code
+  - `atsamd51`: correct initialization for RTC
+  - `atsamd51`: fix pin function selection
+  - `atsamd51`: pin method cleanup
+  - `atsamd51`: allow setting pin mode for each of the SPI pins
+  - `atsamd51`: correct channel init and pin map for ADC based on ItsyBitsy-M4
+  - `feather-m4`: add Adafruit Feather M4 board
+  - `hifive1b`: add support for SPI1
+  - `hifive1b`: fix compiling in simulation
+  - `linux`: fix time on arm32
+  - `metro-m4`: add support for Adafruit Metro M4 Express Airlift board
+  - `metro-m4`: fixes for UART2
+  - `pinetime-devkit0`: add support for the PineTime dev kit
+  - `x9pro`: add support for this smartwatch
+  - `pca10040-s132v6`: add support for SoftDevice
+  - `pca10056-s140v7`: add support for SoftDevice
+  - `arduino-nano33`: added SPI1 connected to NINA-W102 chip on Arduino Nano 33 IOT
+
+0.9.0
+---
+* **command line**
+  - implement 1200-baud UART bootloader reset when flashing boards that support
+    it
+  - flash using mass-storage device for boards that support it
+  - implement `tinygo env`
+  - add support for Windows (but not yet producing Windows binaries)
+  - add Go version to `tinygo env`
+  - update SVD files for up-to-date peripheral interfaces
+* **compiler**
+  - add `//go:align` pragma
+  - fix bug related to type aliases
+  - add support for buffered channels
+  - remove incorrect reflect optimization
+  - implement copying slices in init interpretation
+  - add support for constant indices with a named type
+  - add support for recursive types like linked lists
+  - fix miscompile of function nil panics
+  - fix bug related to goroutines
+* **standard library**
+  - `machine`: do not check for nil slices in `SPI.Tx`
+  - `reflectlite`: add support for Go 1.13
+  - `runtime`: implement `internal/bytealg.CountString`
+  - `sync`: properly handle nil `New` func in `sync.Pool`
+* **targets**
+  - `arduino`: fix .bss section initialization
+  - `fe310`: implement `Pin.Get`
+  - `gameboy-advance`: support directly outputting .gba files
+  - `samd`: reduce code size by avoiding reflection
+  - `samd21`: do not hardcode pin numbers for peripherals
+  - `stm32f103`: avoid issue with `time.Sleep` less than 200µs
+
 0.8.0
 ---
 * **command line**

CODE-OF-CONDUCT.md

@@ -0,0 +1,76 @@
+# Contributor Covenant Code of Conduct
+
+## Our Pledge
+
+In the interest of fostering an open and welcoming environment, we as
+contributors and maintainers pledge to make participation in our project and
+our community a harassment-free experience for everyone, regardless of age, body
+size, disability, ethnicity, sex characteristics, gender identity and expression,
+level of experience, education, socio-economic status, nationality, personal
+appearance, race, religion, or sexual identity and orientation.
+
+## Our Standards
+
+Examples of behavior that contributes to creating a positive environment
+include:
+
+* Using welcoming and inclusive language
+* Being respectful of differing viewpoints and experiences
+* Gracefully accepting constructive criticism
+* Focusing on what is best for the community
+* Showing empathy towards other community members
+
+Examples of unacceptable behavior by participants include:
+
+* The use of sexualized language or imagery and unwelcome sexual attention or
+  advances
+* Trolling, insulting/derogatory comments, and personal or political attacks
+* Public or private harassment
+* Publishing others' private information, such as a physical or electronic
+  address, without explicit permission
+* Other conduct which could reasonably be considered inappropriate in a
+  professional setting
+
+## Our Responsibilities
+
+Project maintainers are responsible for clarifying the standards of acceptable
+behavior and are expected to take appropriate and fair corrective action in
+response to any instances of unacceptable behavior.
+
+Project maintainers have the right and responsibility to remove, edit, or
+reject comments, commits, code, wiki edits, issues, and other contributions
+that are not aligned to this Code of Conduct, or to ban temporarily or
+permanently any contributor for other behaviors that they deem inappropriate,
+threatening, offensive, or harmful.
+
+## Scope
+
+This Code of Conduct applies within all project spaces, and it also applies when
+an individual is representing the project or its community in public spaces.
+Examples of representing a project or community include using an official
+project e-mail address, posting via an official social media account, or acting
+as an appointed representative at an online or offline event. Representation of
+a project may be further defined and clarified by project maintainers.
+
+## Enforcement
+
+Instances of abusive, harassing, or otherwise unacceptable behavior may be
+reported by contacting the project team at [conduct@tinygo.org](mailto:conduct@tinygo.org). All
+complaints will be reviewed and investigated and will result in a response that
+is deemed necessary and appropriate to the circumstances. The project team is
+obligated to maintain confidentiality with regard to the reporter of an incident.
+Further details of specific enforcement policies may be posted separately.
+
+Project maintainers who do not follow or enforce the Code of Conduct in good
+faith may face temporary or permanent repercussions as determined by other
+members of the project's leadership.
+
+## Attribution
+
+This Code of Conduct is adapted from the [Contributor Covenant][homepage], version 1.4,
+available at https://www.contributor-covenant.org/version/1/4/code-of-conduct.html
+
+[homepage]: https://www.contributor-covenant.org
+
+For answers to common questions about this code of conduct, see
+https://www.contributor-covenant.org/faq

Dockerfile

@@ -1,10 +1,10 @@
-# TinyGo base stage installs Go 1.12, LLVM 8 and the TinyGo compiler itself.
-FROM golang:1.12 AS tinygo-base
+# TinyGo base stage installs Go 1.13, LLVM 9 and the TinyGo compiler itself.
+FROM golang:1.13 AS tinygo-base
 
 RUN wget -O- https://apt.llvm.org/llvm-snapshot.gpg.key| apt-key add - && \
-    echo "deb http://apt.llvm.org/buster/ llvm-toolchain-buster-8 main" >> /etc/apt/sources.list && \
+    echo "deb http://apt.llvm.org/buster/ llvm-toolchain-buster-9 main" >> /etc/apt/sources.list && \
     apt-get update && \
-    apt-get install -y llvm-8-dev libclang-8-dev git
+    apt-get install -y llvm-9-dev libclang-9-dev git
 
 COPY . /tinygo
 
@@ -25,9 +25,9 @@ COPY --from=tinygo-base /tinygo/src /tinygo/src
 COPY --from=tinygo-base /tinygo/targets /tinygo/targets
 
 RUN wget -O- https://apt.llvm.org/llvm-snapshot.gpg.key| apt-key add - && \
-    echo "deb http://apt.llvm.org/buster/ llvm-toolchain-buster-8 main" >> /etc/apt/sources.list && \
+    echo "deb http://apt.llvm.org/buster/ llvm-toolchain-buster-9 main" >> /etc/apt/sources.list && \
     apt-get update && \
-    apt-get install -y libllvm8 lld-8
+    apt-get install -y libllvm9 lld-9
 
 # tinygo-avr stage installs the needed dependencies to compile TinyGo programs for AVR microcontrollers.
 FROM tinygo-base AS tinygo-avr
@@ -41,9 +41,8 @@ COPY --from=tinygo-base /tinygo/lib /tinygo/lib
 
 RUN cd /tinygo/ && \
     apt-get update && \
-    apt-get install -y apt-utils python3 make binutils-avr gcc-avr avr-libc && \
+    apt-get install -y apt-utils make binutils-avr gcc-avr avr-libc && \
     make gen-device-avr && \
-    apt-get remove -y python3 && \
     apt-get autoremove -y && \
     apt-get clean
 
@@ -59,11 +58,8 @@ COPY --from=tinygo-base /tinygo/lib /tinygo/lib
 
 RUN cd /tinygo/ && \
     apt-get update && \
-    apt-get install -y apt-utils python3 make clang-8 && \
-    make gen-device-nrf && make gen-device-stm32 && \
-    apt-get remove -y python3 && \
-    apt-get autoremove -y && \
-    apt-get clean
+    apt-get install -y apt-utils make clang-9 && \
+    make gen-device-nrf && make gen-device-stm32
 
 # tinygo-all stage installs the needed dependencies to compile TinyGo programs for all platforms.
 FROM tinygo-wasm AS tinygo-all
@@ -74,10 +70,7 @@ COPY --from=tinygo-base /tinygo/lib /tinygo/lib
 
 RUN cd /tinygo/ && \
     apt-get update && \
-    apt-get install -y apt-utils python3 make clang-8 binutils-avr gcc-avr avr-libc && \
-    make gen-device && \
-    apt-get remove -y python3 && \
-    apt-get autoremove -y && \
-    apt-get clean
+    apt-get install -y apt-utils make clang-9 binutils-avr gcc-avr avr-libc && \
+    make gen-device
 
 CMD ["tinygo"]

LICENSE

@@ -3,6 +3,9 @@ Copyright (c) 2018-2019 TinyGo Authors. All rights reserved.
 TinyGo includes portions of the Go standard library.
 Copyright (c) 2009-2019 The Go Authors. All rights reserved.
 
+TinyGo includes portions of LLVM, which is under the Apache License v2.0 with
+LLVM Exceptions. See https://llvm.org/LICENSE.txt for license information.
+
 Redistribution and use in source and binary forms, with or without
 modification, are permitted provided that the following conditions are
 met:

Makefile

@@ -1,40 +1,95 @@
 
 # aliases
 all: tinygo
-tinygo: build/tinygo
 
 # Default build and source directories, as created by `make llvm-build`.
 LLVM_BUILDDIR ?= llvm-build
 CLANG_SRC ?= llvm-project/clang
 LLD_SRC ?= llvm-project/lld
 
-.PHONY: all tinygo build/tinygo test $(LLVM_BUILDDIR) llvm-source clean fmt gen-device gen-device-nrf gen-device-avr
+# Go binary and GOROOT to select
+GO ?= go
+export GOROOT = $(shell $(GO) env GOROOT)
 
-LLVM_COMPONENTS = all-targets analysis asmparser asmprinter bitreader bitwriter codegen core coroutines debuginfodwarf executionengine instrumentation interpreter ipo irreader linker lto mc mcjit objcarcopts option profiledata scalaropts support target
+# md5sum binary
+MD5SUM = md5sum
 
-UNAME_S := $(shell uname -s)
-ifeq ($(UNAME_S),Linux)
+# tinygo binary for tests
+TINYGO ?= tinygo
+
+# Use CCACHE for LLVM if possible
+ifneq (, $(shell which ccache))
+    LLVM_OPTION += '-DLLVM_CCACHE_BUILD=ON'
+endif
+
+# Allow enabling LLVM assertions
+ifeq (1, $(ASSERT))
+    LLVM_OPTION += '-DLLVM_ENABLE_ASSERTIONS=ON'
+else
+    LLVM_OPTION += '-DLLVM_ENABLE_ASSERTIONS=OFF'
+endif
+
+.PHONY: all tinygo test $(LLVM_BUILDDIR) llvm-source clean fmt gen-device gen-device-nrf gen-device-avr
+
+LLVM_COMPONENTS = all-targets analysis asmparser asmprinter bitreader bitwriter codegen core coroutines coverage debuginfodwarf executionengine instrumentation interpreter ipo irreader linker lto mc mcjit objcarcopts option profiledata scalaropts support target
+
+ifeq ($(OS),Windows_NT)
+    EXE = .exe
+    START_GROUP = -Wl,--start-group
+    END_GROUP = -Wl,--end-group
+
+    # LLVM compiled using MinGW on Windows appears to have problems with threads.
+    # Without this flag, linking results in errors like these:
+    #     libLLVMSupport.a(Threading.cpp.obj):Threading.cpp:(.text+0x55): undefined reference to `std::thread::hardware_concurrency()'
+    LLVM_OPTION += -DLLVM_ENABLE_THREADS=OFF
+
+    CGO_LDFLAGS += -static -static-libgcc -static-libstdc++
+    CGO_LDFLAGS_EXTRA += -lversion
+
+    # Build libclang manually because the CMake-based build system on Windows
+    # doesn't allow building libclang as a static library.
+    LIBCLANG_PATH = $(abspath build/libclang-custom.a)
+    LIBCLANG_FILES = $(abspath $(wildcard $(LLVM_BUILDDIR)/tools/clang/tools/libclang/CMakeFiles/libclang.dir/*.cpp.obj))
+
+    # Add the libclang dependency to the tinygo binary target.
+tinygo: $(LIBCLANG_PATH)
+test: $(LIBCLANG_PATH)
+    # Build libclang.
+$(LIBCLANG_PATH): $(LIBCLANG_FILES)
+	@mkdir -p build
+	ar rcs $(LIBCLANG_PATH) $^
+
+else ifeq ($(shell uname -s),Darwin)
+    MD5SUM = md5
+    LIBCLANG_PATH = $(abspath $(LLVM_BUILDDIR))/lib/libclang.a
+else ifeq ($(shell uname -s),FreeBSD)
+    MD5SUM = md5
+    LIBCLANG_PATH = $(abspath $(LLVM_BUILDDIR))/lib/libclang.a
+    START_GROUP = -Wl,--start-group
+    END_GROUP = -Wl,--end-group
+else
+    LIBCLANG_PATH = $(abspath $(LLVM_BUILDDIR))/lib/libclang.a
     START_GROUP = -Wl,--start-group
     END_GROUP = -Wl,--end-group
 endif
 
-CLANG_LIBS = $(START_GROUP) $(abspath $(LLVM_BUILDDIR))/lib/libclang.a -lclangAnalysis -lclangARCMigrate -lclangAST -lclangASTMatchers -lclangBasic -lclangCodeGen -lclangCrossTU -lclangDriver -lclangDynamicASTMatchers -lclangEdit -lclangFormat -lclangFrontend -lclangFrontendTool -lclangHandleCXX -lclangHandleLLVM -lclangIndex -lclangLex -lclangParse -lclangRewrite -lclangRewriteFrontend -lclangSema -lclangSerialization -lclangStaticAnalyzerCheckers -lclangStaticAnalyzerCore -lclangStaticAnalyzerFrontend -lclangTooling -lclangToolingASTDiff -lclangToolingCore -lclangToolingInclusions $(END_GROUP) -lstdc++
+CLANG_LIBS = $(START_GROUP) -lclangAnalysis -lclangARCMigrate -lclangAST -lclangASTMatchers -lclangBasic -lclangCodeGen -lclangCrossTU -lclangDriver -lclangDynamicASTMatchers -lclangEdit -lclangFormat -lclangFrontend -lclangFrontendTool -lclangHandleCXX -lclangHandleLLVM -lclangIndex -lclangLex -lclangParse -lclangRewrite -lclangRewriteFrontend -lclangSema -lclangSerialization -lclangStaticAnalyzerCheckers -lclangStaticAnalyzerCore -lclangStaticAnalyzerFrontend -lclangTooling -lclangToolingASTDiff -lclangToolingCore -lclangToolingInclusions $(END_GROUP) -lstdc++
 
 LLD_LIBS = $(START_GROUP) -llldCOFF -llldCommon -llldCore -llldDriver -llldELF -llldMachO -llldMinGW -llldReaderWriter -llldWasm -llldYAML $(END_GROUP)
 
 
 # For static linking.
-ifneq ("$(wildcard $(LLVM_BUILDDIR)/bin/llvm-config)","")
-    CGO_CPPFLAGS=$(shell $(LLVM_BUILDDIR)/bin/llvm-config --cppflags) -I$(abspath $(CLANG_SRC))/include -I$(abspath $(LLD_SRC))/include
+ifneq ("$(wildcard $(LLVM_BUILDDIR)/bin/llvm-config*)","")
+    CGO_CPPFLAGS=$(shell $(LLVM_BUILDDIR)/bin/llvm-config --cppflags) -I$(abspath $(LLVM_BUILDDIR))/tools/clang/include -I$(abspath $(CLANG_SRC))/include -I$(abspath $(LLD_SRC))/include
     CGO_CXXFLAGS=-std=c++11
-    CGO_LDFLAGS=-L$(LLVM_BUILDDIR)/lib $(CLANG_LIBS) $(LLD_LIBS) $(shell $(LLVM_BUILDDIR)/bin/llvm-config --ldflags --libs --system-libs $(LLVM_COMPONENTS))
+    CGO_LDFLAGS+=$(LIBCLANG_PATH) -std=c++11 -L$(abspath $(LLVM_BUILDDIR)/lib) $(CLANG_LIBS) $(LLD_LIBS) $(shell $(LLVM_BUILDDIR)/bin/llvm-config --ldflags --libs --system-libs $(LLVM_COMPONENTS)) -lstdc++ $(CGO_LDFLAGS_EXTRA)
 endif
 
 
 clean:
 	@rm -rf build
 
-FMT_PATHS = ./*.go cgo compiler interp ir loader src/device/arm src/examples src/machine src/os src/reflect src/runtime src/sync src/syscall
+FMT_PATHS = ./*.go builder cgo compiler interp ir loader src/device/arm src/examples src/machine src/os src/reflect src/runtime src/sync src/syscall src/internal/reflectlite transform
 fmt:
 	@gofmt -l -w $(FMT_PATHS)
 fmt-check:
@@ -44,36 +99,40 @@ fmt-check:
 gen-device: gen-device-avr gen-device-nrf gen-device-sam gen-device-sifive gen-device-stm32
 
 gen-device-avr:
-	./tools/gen-device-avr.py lib/avr/packs/atmega src/device/avr/
-	./tools/gen-device-avr.py lib/avr/packs/tiny src/device/avr/
-	go fmt ./src/device/avr
+	$(GO) build -o ./build/gen-device-avr ./tools/gen-device-avr/
+	./build/gen-device-avr lib/avr/packs/atmega src/device/avr/
+	./build/gen-device-avr lib/avr/packs/tiny src/device/avr/
+	@GO111MODULE=off $(GO) fmt ./src/device/avr
 
-gen-device-nrf:
-	./tools/gen-device-svd.py lib/nrfx/mdk/ src/device/nrf/ --source=https://github.com/NordicSemiconductor/nrfx/tree/master/mdk
-	go fmt ./src/device/nrf
+build/gen-device-svd: ./tools/gen-device-svd/*.go
+	$(GO) build -o $@ ./tools/gen-device-svd/
 
-gen-device-sam:
-	./tools/gen-device-svd.py lib/cmsis-svd/data/Atmel/ src/device/sam/ --source=https://github.com/posborne/cmsis-svd/tree/master/data/Atmel
-	go fmt ./src/device/sam
+gen-device-nrf: build/gen-device-svd
+	./build/gen-device-svd -source=https://github.com/NordicSemiconductor/nrfx/tree/master/mdk lib/nrfx/mdk/ src/device/nrf/
+	GO111MODULE=off $(GO) fmt ./src/device/nrf
 
-gen-device-sifive:
-	./tools/gen-device-svd.py lib/cmsis-svd/data/SiFive-Community/ src/device/sifive/ --source=https://github.com/AdaCore/svd2ada/tree/master/CMSIS-SVD/SiFive-Community
-	go fmt ./src/device/sifive
+gen-device-sam: build/gen-device-svd
+	./build/gen-device-svd -source=https://github.com/posborne/cmsis-svd/tree/master/data/Atmel lib/cmsis-svd/data/Atmel/ src/device/sam/
+	GO111MODULE=off $(GO) fmt ./src/device/sam
 
-gen-device-stm32:
-	./tools/gen-device-svd.py lib/cmsis-svd/data/STMicro/ src/device/stm32/ --source=https://github.com/posborne/cmsis-svd/tree/master/data/STMicro
-	go fmt ./src/device/stm32
+gen-device-sifive: build/gen-device-svd
+	./build/gen-device-svd -source=https://github.com/posborne/cmsis-svd/tree/master/data/SiFive-Community lib/cmsis-svd/data/SiFive-Community/ src/device/sifive/
+	GO111MODULE=off $(GO) fmt ./src/device/sifive
+
+gen-device-stm32: build/gen-device-svd
+	./build/gen-device-svd -source=https://github.com/posborne/cmsis-svd/tree/master/data/STMicro lib/cmsis-svd/data/STMicro/ src/device/stm32/
+	GO111MODULE=off $(GO) fmt ./src/device/stm32
 
 
 # Get LLVM sources.
 llvm-project/README.md:
-	git clone -b release/8.x https://github.com/llvm/llvm-project
+	git clone -b release/9.x https://github.com/llvm/llvm-project
 llvm-source: llvm-project/README.md
 
 # Configure LLVM.
 TINYGO_SOURCE_DIR=$(shell pwd)
 $(LLVM_BUILDDIR)/build.ninja: llvm-source
-	mkdir -p $(LLVM_BUILDDIR); cd $(LLVM_BUILDDIR); cmake -G Ninja $(TINYGO_SOURCE_DIR)/llvm-project/llvm "-DLLVM_TARGETS_TO_BUILD=X86;ARM;AArch64;WebAssembly" "-DLLVM_EXPERIMENTAL_TARGETS_TO_BUILD=AVR;RISCV" -DCMAKE_BUILD_TYPE=Release -DLLVM_ENABLE_ASSERTIONS=OFF -DLIBCLANG_BUILD_STATIC=ON -DLLVM_ENABLE_TERMINFO=OFF -DLLVM_ENABLE_ZLIB=OFF -DLLVM_ENABLE_PROJECTS="clang;lld" -DLLVM_TOOL_CLANG_TOOLS_EXTRA_BUILD=OFF
+	mkdir -p $(LLVM_BUILDDIR); cd $(LLVM_BUILDDIR); cmake -G Ninja $(TINYGO_SOURCE_DIR)/llvm-project/llvm "-DLLVM_TARGETS_TO_BUILD=X86;ARM;AArch64;RISCV;WebAssembly" "-DLLVM_EXPERIMENTAL_TARGETS_TO_BUILD=AVR" -DCMAKE_BUILD_TYPE=Release -DLIBCLANG_BUILD_STATIC=ON -DLLVM_ENABLE_TERMINFO=OFF -DLLVM_ENABLE_ZLIB=OFF -DLLVM_ENABLE_PROJECTS="clang;lld" -DLLVM_TOOL_CLANG_TOOLS_EXTRA_BUILD=OFF $(LLVM_OPTION)
 
 # Build LLVM.
 $(LLVM_BUILDDIR): $(LLVM_BUILDDIR)/build.ninja
@@ -81,63 +140,124 @@ $(LLVM_BUILDDIR): $(LLVM_BUILDDIR)/build.ninja
 
 
 # Build the Go compiler.
-build/tinygo:
+tinygo:
 	@if [ ! -f "$(LLVM_BUILDDIR)/bin/llvm-config" ]; then echo "Fetch and build LLVM first by running:"; echo "  make llvm-source"; echo "  make $(LLVM_BUILDDIR)"; exit 1; fi
-	CGO_CPPFLAGS="$(CGO_CPPFLAGS)" CGO_CXXFLAGS="$(CGO_CXXFLAGS)" CGO_LDFLAGS="$(CGO_LDFLAGS)" go build -o build/tinygo -tags byollvm .
+	CGO_CPPFLAGS="$(CGO_CPPFLAGS)" CGO_CXXFLAGS="$(CGO_CXXFLAGS)" CGO_LDFLAGS="$(CGO_LDFLAGS)" $(GO) build -o build/tinygo$(EXE) -tags byollvm .
 
 test:
-	CGO_CPPFLAGS="$(CGO_CPPFLAGS)" CGO_CXXFLAGS="$(CGO_CXXFLAGS)" CGO_LDFLAGS="$(CGO_LDFLAGS)" go test -v -tags byollvm ./transform .
+	CGO_CPPFLAGS="$(CGO_CPPFLAGS)" CGO_CXXFLAGS="$(CGO_CXXFLAGS)" CGO_LDFLAGS="$(CGO_LDFLAGS)" $(GO) test -v -tags byollvm ./cgo ./compileopts ./interp ./transform .
 
 tinygo-test:
 	cd tests/tinygotest && tinygo test
 
 .PHONY: smoketest
 smoketest:
+	$(TINYGO) version
 	# test all examples
-	tinygo build -size short -o test.elf -target=pca10040            examples/blinky1
-	tinygo build -size short -o test.elf -target=pca10040            examples/adc
-	tinygo build -size short -o test.elf -target=pca10040            examples/blinkm
-	tinygo build -size short -o test.elf -target=pca10040            examples/blinky2
-	tinygo build -size short -o test.elf -target=pca10040            examples/button
-	tinygo build -size short -o test.elf -target=pca10040            examples/button2
-	tinygo build -size short -o test.elf -target=pca10040            examples/echo
-	tinygo build -size short -o test.elf -target=circuitplay-express examples/i2s
-	tinygo build -size short -o test.elf -target=pca10040            examples/mcp3008
-	tinygo build -size short -o test.elf -target=microbit            examples/microbit-blink
-	tinygo build -size short -o test.elf -target=pca10040            examples/pwm
-	tinygo build -size short -o test.elf -target=pca10040            examples/serial
-	tinygo build -size short -o test.elf -target=pca10040            examples/test
+	$(TINYGO) build -size short -o test.hex -target=pca10040            examples/blinky1
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=pca10040            examples/adc
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=pca10040            examples/blinkm
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=pca10040            examples/blinky2
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=pca10040            examples/button
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=pca10040            examples/button2
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=pca10040            examples/echo
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=circuitplay-express examples/i2s
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=pca10040            examples/mcp3008
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=microbit            examples/microbit-blink
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=pca10040            examples/pwm
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=pca10040            examples/serial
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=pca10040            examples/systick
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=pca10040            examples/test
+	@$(MD5SUM) test.hex
+	# test simulated boards on play.tinygo.org
+	$(TINYGO) build             -o test.wasm -tags=arduino              examples/blinky1
+	@$(MD5SUM) test.wasm
+	$(TINYGO) build             -o test.wasm -tags=hifive1b             examples/blinky1
+	@$(MD5SUM) test.wasm
+	$(TINYGO) build             -o test.wasm -tags=reelboard            examples/blinky1
+	@$(MD5SUM) test.wasm
+	$(TINYGO) build             -o test.wasm -tags=pca10040             examples/blinky2
+	@$(MD5SUM) test.wasm
+	$(TINYGO) build             -o test.wasm -tags=pca10056             examples/blinky2
+	@$(MD5SUM) test.wasm
+	$(TINYGO) build             -o test.wasm -tags=circuitplay_express  examples/blinky1
+	@$(MD5SUM) test.wasm
 	# test all targets/boards
-	tinygo build             -o test.wasm -tags=pca10040             examples/blinky2
-	tinygo build -size short -o test.elf -target=microbit            examples/echo
-	tinygo build -size short -o test.elf -target=nrf52840-mdk        examples/blinky1
-	tinygo build -size short -o test.elf -target=pca10031            examples/blinky1
-	tinygo build -size short -o test.elf -target=bluepill            examples/blinky1
-	tinygo build -size short -o test.elf -target=reelboard           examples/blinky1
-	tinygo build -size short -o test.elf -target=reelboard           examples/blinky2
-	tinygo build -size short -o test.elf -target=pca10056            examples/blinky1
-	tinygo build -size short -o test.elf -target=pca10056            examples/blinky2
-	tinygo build -size short -o test.elf -target=itsybitsy-m0        examples/blinky1
-	tinygo build -size short -o test.elf -target=feather-m0          examples/blinky1
-	tinygo build -size short -o test.elf -target=trinket-m0          examples/blinky1
-	tinygo build -size short -o test.elf -target=circuitplay-express examples/blinky1
-	tinygo build -size short -o test.elf -target=stm32f4disco        examples/blinky1
-	tinygo build -size short -o test.elf -target=stm32f4disco        examples/blinky2
-	tinygo build -size short -o test.elf -target=circuitplay-express examples/i2s
-	tinygo build -size short -o test.elf -target=gameboy-advance     examples/gba-display
-	tinygo build -size short -o test.elf -target=itsybitsy-m4        examples/blinky1
-	tinygo build -size short -o test.elf -target=nucleo-f103rb       examples/blinky1
+	$(TINYGO) build -size short -o test.hex -target=pca10040-s132v6     examples/blinky1
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=microbit            examples/echo
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=nrf52840-mdk        examples/blinky1
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=pca10031            examples/blinky1
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=bluepill            examples/blinky1
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=reelboard           examples/blinky1
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=reelboard           examples/blinky2
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=pca10056            examples/blinky1
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=pca10056            examples/blinky2
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=itsybitsy-m0        examples/blinky1
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=feather-m0          examples/blinky1
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=trinket-m0          examples/blinky1
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=circuitplay-express examples/blinky1
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=stm32f4disco        examples/blinky1
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=stm32f4disco        examples/blinky2
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=circuitplay-express examples/i2s
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.gba -target=gameboy-advance     examples/gba-display
+	@$(MD5SUM) test.gba
+	$(TINYGO) build -size short -o test.hex -target=itsybitsy-m4        examples/blinky1
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=feather-m4          examples/blinky1
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=pybadge             examples/blinky1
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=metro-m4-airlift    examples/blinky1
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=nucleo-f103rb       examples/blinky1
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=pinetime-devkit0    examples/blinky1
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=x9pro               examples/blinky1
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=pca10056-s140v7     examples/blinky1
+	@$(MD5SUM) test.hex
 ifneq ($(AVR), 0)
-	tinygo build -size short -o test.elf -target=arduino             examples/blinky1
-	tinygo build -size short -o test.elf -target=digispark           examples/blinky1
+	$(TINYGO) build -size short -o test.hex -target=arduino             examples/blinky1
+	@$(MD5SUM) test.hex
+	$(TINYGO) build -size short -o test.hex -target=digispark           examples/blinky1
+	@$(MD5SUM) test.hex
 endif
-ifneq ($(RISCV), 0)
-	tinygo build -size short -o test.elf -target=hifive1b            examples/blinky1
-endif
-	tinygo build             -o wasm.wasm -target=wasm               examples/wasm/export
-	tinygo build             -o wasm.wasm -target=wasm               examples/wasm/main
+	$(TINYGO) build -size short -o test.hex -target=hifive1b            examples/blinky1
+	@$(MD5SUM) test.hex
+	$(TINYGO) build             -o wasm.wasm -target=wasm               examples/wasm/export
+	$(TINYGO) build             -o wasm.wasm -target=wasm               examples/wasm/main
 
-release: build/tinygo gen-device
+release: tinygo gen-device
 	@mkdir -p build/release/tinygo/bin
 	@mkdir -p build/release/tinygo/lib/clang/include
 	@mkdir -p build/release/tinygo/lib/CMSIS/CMSIS
@@ -147,7 +267,7 @@ release: build/tinygo gen-device
 	@mkdir -p build/release/tinygo/pkg/armv7m-none-eabi
 	@mkdir -p build/release/tinygo/pkg/armv7em-none-eabi
 	@echo copying source files
-	@cp -p  build/tinygo                 build/release/tinygo/bin
+	@cp -p  build/tinygo$(EXE)           build/release/tinygo/bin
 	@cp -p $(abspath $(CLANG_SRC))/lib/Headers/*.h build/release/tinygo/lib/clang/include
 	@cp -rp lib/CMSIS/CMSIS/Include      build/release/tinygo/lib/CMSIS/CMSIS
 	@cp -rp lib/CMSIS/README.md          build/release/tinygo/lib/CMSIS

README.md

@@ -1,6 +1,6 @@
 # TinyGo - Go compiler for small places
 
-[![CircleCI](https://circleci.com/gh/tinygo-org/tinygo/tree/dev.svg?style=svg)](https://circleci.com/gh/tinygo-org/tinygo/tree/dev)
+[![CircleCI](https://circleci.com/gh/tinygo-org/tinygo/tree/dev.svg?style=svg)](https://circleci.com/gh/tinygo-org/tinygo/tree/dev) [![Build Status](https://dev.azure.com/tinygo/tinygo/_apis/build/status/tinygo-CI?branchName=dev)](https://dev.azure.com/tinygo/tinygo/_build/latest?definitionId=1&branchName=dev)
 
 TinyGo is a Go compiler intended for use in small places such as microcontrollers, WebAssembly (Wasm), and command-line tools.
 
@@ -43,26 +43,32 @@ See the [getting started instructions](https://tinygo.org/getting-started/) for
 
 You can compile TinyGo programs for microcontrollers, WebAssembly and Linux.
 
-The following 18 microcontroller boards are currently supported:
+The following 22 microcontroller boards are currently supported:
 
 * [Adafruit Circuit Playground Express](https://www.adafruit.com/product/3333)
 * [Adafruit Feather M0](https://www.adafruit.com/product/2772)
+* [Adafruit Feather M4](https://www.adafruit.com/product/3857)
 * [Adafruit ItsyBitsy M0](https://www.adafruit.com/product/3727)
 * [Adafruit ItsyBitsy M4](https://www.adafruit.com/product/3800)
+* [Adafruit Metro M4 Express Airlift](https://www.adafruit.com/product/4000)
+* [Adafruit PyBadge](https://www.adafruit.com/product/4200)
 * [Adafruit Trinket M0](https://www.adafruit.com/product/3500)
 * [Arduino Nano33 IoT](https://store.arduino.cc/nano-33-iot)
 * [Arduino Uno](https://store.arduino.cc/arduino-uno-rev3)
 * [BBC micro:bit](https://microbit.org/)
-* [Game Boy Advance](https://en.wikipedia.org/wiki/Game_Boy_Advance)
-* [ST Micro STM32F103XX "Bluepill"](http://wiki.stm32duino.com/index.php?title=Blue_Pill)
-* [ST Micro "Nucleo F103RB"](https://www.st.com/en/evaluation-tools/nucleo-f103rb.html)
-* [ST Micro STM32F407 "Discovery"](https://www.st.com/en/evaluation-tools/stm32f4discovery.html)
 * [Digispark](http://digistump.com/products/1)
+* [Game Boy Advance](https://en.wikipedia.org/wiki/Game_Boy_Advance)
+* [Makerdiary nRF52840-MDK](https://wiki.makerdiary.com/nrf52840-mdk/)
 * [Nordic Semiconductor PCA10031](https://www.nordicsemi.com/eng/Products/nRF51-Dongle)
 * [Nordic Semiconductor PCA10040](https://www.nordicsemi.com/eng/Products/Bluetooth-low-energy/nRF52-DK)
 * [Nordic Semiconductor PCA10056](https://www.nordicsemi.com/Software-and-Tools/Development-Kits/nRF52840-DK)
-* [Makerdiary nRF52840-MDK](https://wiki.makerdiary.com/nrf52840-mdk/)
 * [Phytec reel board](https://www.phytec.eu/product-eu/internet-of-things/reelboard/)
+* [PineTime DevKit](https://www.pine64.org/pinetime/)
+* [SiFIve HiFive1](https://www.sifive.com/boards/hifive1)
+* [ST Micro "Nucleo F103RB"](https://www.st.com/en/evaluation-tools/nucleo-f103rb.html)
+* [ST Micro STM32F103XX "Bluepill"](http://wiki.stm32duino.com/index.php?title=Blue_Pill)
+* [ST Micro STM32F407 "Discovery"](https://www.st.com/en/evaluation-tools/stm32f4discovery.html)
+* [X9 Pro smartwatch](https://github.com/curtpw/nRF5x-device-reverse-engineering/tree/master/X9-nrf52832-activity-tracker/)
 
 For more information, see [this list of boards](https://tinygo.org/microcontrollers/). Pull requests for additional support are welcome!
 
@@ -121,3 +127,5 @@ The original reasoning was: if [Python](https://micropython.org/) can run on mic
 ## License
 
 This project is licensed under the BSD 3-clause license, just like the [Go project](https://golang.org/LICENSE) itself.
+
+Some code has been copied from the LLVM project and is therefore licensed under [a variant of the Apache 2.0 license](http://releases.llvm.org/9.0.0/LICENSE.TXT). This has been clearly indicated in the header of these files.

azure-pipelines.yml

@@ -0,0 +1,74 @@
+# Avoid lengthy LLVM rebuilds on each newly pushed branch. Pull requests will
+# be built anyway.
+trigger:
+- master
+- dev
+
+jobs:
+- job: Build
+  timeoutInMinutes: 240 # 4h
+  pool:
+    vmImage: 'VS2017-Win2016'
+  steps:
+    - checkout: self
+    - task: CacheBeta@0
+      displayName: Cache LLVM source
+      inputs:
+        key: llvm-source-9-windows-v0
+        path: llvm-project
+    - task: Bash@3
+      displayName: Download LLVM source
+      inputs:
+        targetType: inline
+        script: make llvm-source
+    - task: CacheBeta@0
+      displayName: Cache LLVM build
+      inputs:
+        key: llvm-build-9-windows-v0
+        path: llvm-build
+    - task: Bash@3
+      displayName: Build LLVM
+      inputs:
+        targetType: inline
+        script: |
+          if [ ! -f llvm-build/lib/liblldELF.a ]
+          then
+            choco install ninja
+            # LLVM 9 cannot be built with MinGW 8.
+            # For details: https://reviews.llvm.org/D70266
+            choco uninstall mingw
+            choco install mingw --version=7.3.0
+            make llvm-build
+          fi
+    - task: Bash@3
+      displayName: Install QEMU
+      inputs:
+        targetType: inline
+        script: choco install qemu
+    - task: Bash@3
+      displayName: Test TinyGo
+      inputs:
+        targetType: inline
+        script: |
+          export PATH="/c/Go1.13/bin:$PATH:./llvm-build/bin:/c/Program Files/qemu"
+          unset GOROOT
+          make test
+    - task: Bash@3
+      displayName: Build TinyGo release tarball
+      inputs:
+        targetType: inline
+        script: |
+          export PATH="/c/Go1.13/bin:$PATH:./llvm-build/bin:/c/Program Files/qemu"
+          unset GOROOT
+          make release -j4
+    - publish: $(System.DefaultWorkingDirectory)/build/release.tar.gz
+      displayName: Publish tarball as artifact
+      artifact: tinygo.windows-amd64.tar.gz
+    - task: Bash@3
+      displayName: Smoke tests
+      inputs:
+        targetType: inline
+        script: |
+          export PATH="/c/Go1.13/bin:$PATH:./llvm-build/bin:/c/Program Files/qemu"
+          unset GOROOT
+          make smoketest TINYGO=build/tinygo AVR=0

builder/ar.go

@@ -0,0 +1,182 @@
+package builder
+
+import (
+	"bytes"
+	"debug/elf"
+	"encoding/binary"
+	"errors"
+	"io"
+	"os"
+	"path/filepath"
+	"time"
+
+	"github.com/blakesmith/ar"
+)
+
+// makeArchive creates an arcive for static linking from a list of object files
+// given as a parameter. It is equivalent to the following command:
+//
+//     ar -rcs <archivePath> <objs...>
+func makeArchive(archivePath string, objs []string) error {
+	// Open the archive file.
+	arfile, err := os.Create(archivePath)
+	if err != nil {
+		return err
+	}
+	defer arfile.Close()
+	arwriter := ar.NewWriter(arfile)
+	err = arwriter.WriteGlobalHeader()
+	if err != nil {
+		return &os.PathError{"write ar header", archivePath, err}
+	}
+
+	// Open all object files and read the symbols for the symbol table.
+	symbolTable := []struct {
+		name      string // symbol name
+		fileIndex int    // index into objfiles
+	}{}
+	objfiles := make([]struct {
+		file          *os.File
+		archiveOffset int32
+	}, len(objs))
+	for i, objpath := range objs {
+		objfile, err := os.Open(objpath)
+		if err != nil {
+			return err
+		}
+		objfiles[i].file = objfile
+
+		// Read the symbols and add them to the symbol table.
+		dbg, err := elf.NewFile(objfile)
+		if err != nil {
+			return err
+		}
+		symbols, err := dbg.Symbols()
+		if err != nil {
+			return err
+		}
+		for _, symbol := range symbols {
+			bind := elf.ST_BIND(symbol.Info)
+			if bind != elf.STB_GLOBAL && bind != elf.STB_WEAK {
+				// Don't include local symbols (STB_LOCAL).
+				continue
+			}
+			if elf.ST_TYPE(symbol.Info) != elf.STT_FUNC {
+				// Not a function.
+				// TODO: perhaps globals variables should also be included?
+				continue
+			}
+			// Include in archive.
+			symbolTable = append(symbolTable, struct {
+				name      string
+				fileIndex int
+			}{symbol.Name, i})
+		}
+	}
+
+	// Create the symbol table buffer.
+	// For some (sparse) details on the file format:
+	// https://en.wikipedia.org/wiki/Ar_(Unix)#System_V_(or_GNU)_variant
+	buf := &bytes.Buffer{}
+	binary.Write(buf, binary.BigEndian, int32(len(symbolTable)))
+	for range symbolTable {
+		// This is a placeholder index, it will be updated after all files have
+		// been written to the archive (see the end of this function).
+		err = binary.Write(buf, binary.BigEndian, int32(0))
+		if err != nil {
+			return err
+		}
+	}
+	for _, sym := range symbolTable {
+		_, err := buf.Write([]byte(sym.name + "\x00"))
+		if err != nil {
+			return err
+		}
+	}
+	for buf.Len()%2 != 0 {
+		// The symbol table must be aligned.
+		// This appears to be required by lld.
+		buf.WriteByte(0)
+	}
+
+	// Write the symbol table.
+	err = arwriter.WriteHeader(&ar.Header{
+		Name:    "/",
+		ModTime: time.Unix(0, 0),
+		Uid:     0,
+		Gid:     0,
+		Mode:    0,
+		Size:    int64(buf.Len()),
+	})
+	if err != nil {
+		return err
+	}
+
+	// Keep track of the start of the symbol table.
+	symbolTableStart, err := arfile.Seek(0, os.SEEK_CUR)
+	if err != nil {
+		return err
+	}
+
+	// Write symbol table contents.
+	_, err = arfile.Write(buf.Bytes())
+	if err != nil {
+		return err
+	}
+
+	// Add all object files to the archive.
+	for i, objfile := range objfiles {
+		// Store the start index, for when we'll update the symbol table with
+		// the correct file start indices.
+		offset, err := arfile.Seek(0, os.SEEK_CUR)
+		if err != nil {
+			return err
+		}
+		if int64(int32(offset)) != offset {
+			return errors.New("large archives (4GB+) not supported: " + archivePath)
+		}
+		objfiles[i].archiveOffset = int32(offset)
+
+		// Write the file header.
+		st, err := objfile.file.Stat()
+		if err != nil {
+			return err
+		}
+		err = arwriter.WriteHeader(&ar.Header{
+			Name:    filepath.Base(objfile.file.Name()),
+			ModTime: time.Unix(0, 0),
+			Uid:     0,
+			Gid:     0,
+			Mode:    0644,
+			Size:    st.Size(),
+		})
+		if err != nil {
+			return err
+		}
+
+		// Copy the file contents into the archive.
+		n, err := io.Copy(arwriter, objfile.file)
+		if err != nil {
+			return err
+		}
+		if n != st.Size() {
+			return errors.New("file modified during ar creation: " + archivePath)
+		}
+
+		// File is not needed anymore.
+		objfile.file.Close()
+	}
+
+	// Create symbol indices.
+	indicesBuf := &bytes.Buffer{}
+	for _, sym := range symbolTable {
+		err = binary.Write(indicesBuf, binary.BigEndian, objfiles[sym.fileIndex].archiveOffset)
+		if err != nil {
+			return err
+		}
+	}
+
+	// Overwrite placeholder indices.
+	_, err = arfile.WriteAt(indicesBuf.Bytes(), symbolTableStart+4)
+	return err
+}

builder/build.go

@@ -0,0 +1,217 @@
+// Package builder is the compiler driver of TinyGo. It takes in a package name
+// and an output path, and outputs an executable. It manages the entire
+// compilation pipeline in between.
+package builder
+
+import (
+	"errors"
+	"fmt"
+	"io/ioutil"
+	"os"
+	"path/filepath"
+	"strconv"
+	"strings"
+
+	"github.com/tinygo-org/tinygo/compileopts"
+	"github.com/tinygo-org/tinygo/compiler"
+	"github.com/tinygo-org/tinygo/goenv"
+	"github.com/tinygo-org/tinygo/interp"
+)
+
+// Build performs a single package to executable Go build. It takes in a package
+// name, an output path, and set of compile options and from that it manages the
+// whole compilation process.
+//
+// The error value may be of type *MultiError. Callers will likely want to check
+// for this case and print such errors individually.
+func Build(pkgName, outpath string, config *compileopts.Config, action func(string) error) error {
+	c, err := compiler.NewCompiler(pkgName, config)
+	if err != nil {
+		return err
+	}
+
+	// Compile Go code to IR.
+	errs := c.Compile(pkgName)
+	if len(errs) != 0 {
+		return newMultiError(errs)
+	}
+	if config.Options.PrintIR {
+		fmt.Println("; Generated LLVM IR:")
+		fmt.Println(c.IR())
+	}
+	if err := c.Verify(); err != nil {
+		return errors.New("verification error after IR construction")
+	}
+
+	err = interp.Run(c.Module(), config.DumpSSA())
+	if err != nil {
+		return err
+	}
+	if err := c.Verify(); err != nil {
+		return errors.New("verification error after interpreting runtime.initAll")
+	}
+
+	if config.GOOS() != "darwin" {
+		c.ApplyFunctionSections() // -ffunction-sections
+	}
+
+	// Browsers cannot handle external functions that have type i64 because it
+	// cannot be represented exactly in JavaScript (JS only has doubles). To
+	// keep functions interoperable, pass int64 types as pointers to
+	// stack-allocated values.
+	// Use -wasm-abi=generic to disable this behaviour.
+	if config.Options.WasmAbi == "js" && strings.HasPrefix(config.Triple(), "wasm") {
+		err := c.ExternalInt64AsPtr()
+		if err != nil {
+			return err
+		}
+	}
+
+	// Optimization levels here are roughly the same as Clang, but probably not
+	// exactly.
+	errs = nil
+	switch config.Options.Opt {
+	case "none:", "0":
+		errs = c.Optimize(0, 0, 0) // -O0
+	case "1":
+		errs = c.Optimize(1, 0, 0) // -O1
+	case "2":
+		errs = c.Optimize(2, 0, 225) // -O2
+	case "s":
+		errs = c.Optimize(2, 1, 225) // -Os
+	case "z":
+		errs = c.Optimize(2, 2, 5) // -Oz, default
+	default:
+		errs = []error{errors.New("unknown optimization level: -opt=" + config.Options.Opt)}
+	}
+	if len(errs) > 0 {
+		return newMultiError(errs)
+	}
+	if err := c.Verify(); err != nil {
+		return errors.New("verification failure after LLVM optimization passes")
+	}
+
+	// On the AVR, pointers can point either to flash or to RAM, but we don't
+	// know. As a temporary fix, load all global variables in RAM.
+	// In the future, there should be a compiler pass that determines which
+	// pointers are flash and which are in RAM so that pointers can have a
+	// correct address space parameter (address space 1 is for flash).
+	if strings.HasPrefix(config.Triple(), "avr") {
+		c.NonConstGlobals()
+		if err := c.Verify(); err != nil {
+			return errors.New("verification error after making all globals non-constant on AVR")
+		}
+	}
+
+	// Generate output.
+	outext := filepath.Ext(outpath)
+	switch outext {
+	case ".o":
+		return c.EmitObject(outpath)
+	case ".bc":
+		return c.EmitBitcode(outpath)
+	case ".ll":
+		return c.EmitText(outpath)
+	default:
+		// Act as a compiler driver.
+
+		// Create a temporary directory for intermediary files.
+		dir, err := ioutil.TempDir("", "tinygo")
+		if err != nil {
+			return err
+		}
+		defer os.RemoveAll(dir)
+
+		// Write the object file.
+		objfile := filepath.Join(dir, "main.o")
+		err = c.EmitObject(objfile)
+		if err != nil {
+			return err
+		}
+
+		// Load builtins library from the cache, possibly compiling it on the
+		// fly.
+		var librt string
+		if config.Target.RTLib == "compiler-rt" {
+			librt, err = loadBuiltins(config.Triple())
+			if err != nil {
+				return err
+			}
+		}
+
+		// Prepare link command.
+		executable := filepath.Join(dir, "main")
+		tmppath := executable // final file
+		ldflags := append(config.LDFlags(), "-o", executable, objfile)
+		if config.Target.RTLib == "compiler-rt" {
+			ldflags = append(ldflags, librt)
+		}
+
+		// Compile extra files.
+		root := goenv.Get("TINYGOROOT")
+		for i, path := range config.ExtraFiles() {
+			abspath := filepath.Join(root, path)
+			outpath := filepath.Join(dir, "extra-"+strconv.Itoa(i)+"-"+filepath.Base(path)+".o")
+			err := runCCompiler(config.Target.Compiler, append(config.CFlags(), "-c", "-o", outpath, abspath)...)
+			if err != nil {
+				return &commandError{"failed to build", path, err}
+			}
+			ldflags = append(ldflags, outpath)
+		}
+
+		// Compile C files in packages.
+		for i, pkg := range c.Packages() {
+			for _, file := range pkg.CFiles {
+				path := filepath.Join(pkg.Package.Dir, file)
+				outpath := filepath.Join(dir, "pkg"+strconv.Itoa(i)+"-"+file+".o")
+				err := runCCompiler(config.Target.Compiler, append(config.CFlags(), "-c", "-o", outpath, path)...)
+				if err != nil {
+					return &commandError{"failed to build", path, err}
+				}
+				ldflags = append(ldflags, outpath)
+			}
+		}
+
+		// Link the object files together.
+		err = link(config.Target.Linker, ldflags...)
+		if err != nil {
+			return &commandError{"failed to link", executable, err}
+		}
+
+		if config.Options.PrintSizes == "short" || config.Options.PrintSizes == "full" {
+			sizes, err := loadProgramSize(executable)
+			if err != nil {
+				return err
+			}
+			if config.Options.PrintSizes == "short" {
+				fmt.Printf("   code    data     bss |   flash     ram\n")
+				fmt.Printf("%7d %7d %7d | %7d %7d\n", sizes.Code, sizes.Data, sizes.BSS, sizes.Code+sizes.Data, sizes.Data+sizes.BSS)
+			} else {
+				fmt.Printf("   code  rodata    data     bss |   flash     ram | package\n")
+				for _, name := range sizes.sortedPackageNames() {
+					pkgSize := sizes.Packages[name]
+					fmt.Printf("%7d %7d %7d %7d | %7d %7d | %s\n", pkgSize.Code, pkgSize.ROData, pkgSize.Data, pkgSize.BSS, pkgSize.Flash(), pkgSize.RAM(), name)
+				}
+				fmt.Printf("%7d %7d %7d %7d | %7d %7d | (sum)\n", sizes.Sum.Code, sizes.Sum.ROData, sizes.Sum.Data, sizes.Sum.BSS, sizes.Sum.Flash(), sizes.Sum.RAM())
+				fmt.Printf("%7d       - %7d %7d | %7d %7d | (all)\n", sizes.Code, sizes.Data, sizes.BSS, sizes.Code+sizes.Data, sizes.Data+sizes.BSS)
+			}
+		}
+
+		// Get an Intel .hex file or .bin file from the .elf file.
+		if outext == ".hex" || outext == ".bin" || outext == ".gba" {
+			tmppath = filepath.Join(dir, "main"+outext)
+			err := objcopy(executable, tmppath)
+			if err != nil {
+				return err
+			}
+		} else if outext == ".uf2" {
+			// Get UF2 from the .elf file.
+			tmppath = filepath.Join(dir, "main"+outext)
+			err := convertELFFileToUF2File(executable, tmppath)
+			if err != nil {
+				return err
+			}
+		}
+		return action(tmppath)
+	}
+}

builder/buildcache.go

@@ -1,20 +1,13 @@
-package main
+package builder
 
 import (
 	"io"
 	"os"
 	"path/filepath"
 	"time"
-)
 
-// Get the cache directory, usually ~/.cache/tinygo
-func cacheDir() string {
-	dir, err := os.UserCacheDir()
-	if err != nil {
-		panic("could not find cache dir: " + err.Error())
-	}
-	return filepath.Join(dir, "tinygo")
-}
+	"github.com/tinygo-org/tinygo/goenv"
+)
 
 // Return the newest timestamp of all the file paths passed in. Used to check
 // for stale caches.
@@ -41,8 +34,7 @@ func cacheTimestamp(paths []string) (time.Time, error) {
 // TODO: the configKey is currently ignored. It is supposed to be used as extra
 // data for the cache key, like the compiler version and arguments.
 func cacheLoad(name, configKey string, sourceFiles []string) (string, error) {
-	dir := cacheDir()
-	cachepath := filepath.Join(dir, name)
+	cachepath := filepath.Join(goenv.Get("GOCACHE"), name)
 	cacheStat, err := os.Stat(cachepath)
 	if os.IsNotExist(err) {
 		return "", nil // does not exist
@@ -76,7 +68,7 @@ func cacheStore(tmppath, name, configKey string, sourceFiles []string) (string,
 
 	// TODO: check the config key
 
-	dir := cacheDir()
+	dir := goenv.Get("GOCACHE")
 	err := os.MkdirAll(dir, 0777)
 	if err != nil {
 		return "", err
@@ -117,10 +109,10 @@ func moveFile(src, dst string) error {
 		return err
 	}
 
-	err = os.Rename(dst+".tmp", dst)
+	err = outf.Close()
 	if err != nil {
 		return err
 	}
 
-	return outf.Close()
+	return os.Rename(dst+".tmp", dst)
 }

builder/builtins.go

@@ -1,15 +1,12 @@
-package main
+package builder
 
 import (
-	"errors"
-	"io"
 	"io/ioutil"
 	"os"
 	"path/filepath"
 	"strings"
-	"time"
 
-	"github.com/blakesmith/ar"
+	"github.com/tinygo-org/tinygo/goenv"
 )
 
 // These are the GENERIC_SOURCES according to CMakeList.txt.
@@ -169,13 +166,13 @@ func builtinFiles(target string) []string {
 
 // builtinsDir returns the directory where the sources for compiler-rt are kept.
 func builtinsDir() string {
-	return filepath.Join(sourceDir(), "lib", "compiler-rt", "lib", "builtins")
+	return filepath.Join(goenv.Get("TINYGOROOT"), "lib", "compiler-rt", "lib", "builtins")
 }
 
 // Get the builtins archive, possibly generating it as needed.
 func loadBuiltins(target string) (path string, err error) {
 	// Try to load a precompiled compiler-rt library.
-	precompiledPath := filepath.Join(sourceDir(), "pkg", target, "compiler-rt.a")
+	precompiledPath := filepath.Join(goenv.Get("TINYGOROOT"), "pkg", target, "compiler-rt.a")
 	if _, err := os.Stat(precompiledPath); err == nil {
 		// Found a precompiled compiler-rt for this OS/architecture. Return the
 		// path directly.
@@ -196,7 +193,7 @@ func loadBuiltins(target string) (path string, err error) {
 	}
 
 	var cachepath string
-	err = compileBuiltins(target, func(path string) error {
+	err = CompileBuiltins(target, func(path string) error {
 		path, err := cacheStore(path, outfile, commands["clang"][0], srcs)
 		cachepath = path
 		return err
@@ -204,11 +201,11 @@ func loadBuiltins(target string) (path string, err error) {
 	return cachepath, err
 }
 
-// compileBuiltins compiles builtins from compiler-rt into a static library.
+// CompileBuiltins compiles builtins from compiler-rt into a static library.
 // When it succeeds, it will call the callback with the resulting path. The path
 // will be removed after callback returns. If callback returns an error, this is
 // passed through to the return value of this function.
-func compileBuiltins(target string, callback func(path string) error) error {
+func CompileBuiltins(target string, callback func(path string) error) error {
 	builtinsDir := builtinsDir()
 
 	builtins := builtinFiles(target)
@@ -239,56 +236,25 @@ func compileBuiltins(target string, callback func(path string) error) error {
 		// Note: -fdebug-prefix-map is necessary to make the output archive
 		// reproducible. Otherwise the temporary directory is stored in the
 		// archive itself, which varies each run.
-		err := execCommand(commands["clang"], "-c", "-Oz", "-g", "-Werror", "-Wall", "-std=c11", "-fshort-enums", "-nostdlibinc", "-ffunction-sections", "-fdata-sections", "--target="+target, "-fdebug-prefix-map="+dir+"="+remapDir, "-o", objpath, srcpath)
+		args := []string{"-c", "-Oz", "-g", "-Werror", "-Wall", "-std=c11", "-fshort-enums", "-nostdlibinc", "-ffunction-sections", "-fdata-sections", "-Wno-macro-redefined", "--target=" + target, "-fdebug-prefix-map=" + dir + "=" + remapDir}
+		if strings.HasPrefix(target, "riscv32-") {
+			args = append(args, "-march=rv32imac", "-mabi=ilp32", "-fforce-enable-int128")
+		}
+		err := runCCompiler("clang", append(args, "-o", objpath, srcpath)...)
 		if err != nil {
 			return &commandError{"failed to build", srcpath, err}
 		}
 	}
 
-	// Put all builtins in an archive to link as a static library.
-	// Note: this does not create a symbol index, but ld.lld doesn't seem to
-	// care.
+	// Put all the object files in a single archive. This archive file will be
+	// used to statically link compiler-rt.
 	arpath := filepath.Join(dir, "librt.a")
-	arfile, err := os.Create(arpath)
+	err = makeArchive(arpath, objs)
 	if err != nil {
 		return err
 	}
-	defer arfile.Close()
-	arwriter := ar.NewWriter(arfile)
-	err = arwriter.WriteGlobalHeader()
-	if err != nil {
-		return &os.PathError{"write ar header", arpath, err}
-	}
-	for _, objpath := range objs {
-		name := filepath.Base(objpath)
-		objfile, err := os.Open(objpath)
-		if err != nil {
-			return err
-		}
-		defer objfile.Close()
-		st, err := objfile.Stat()
-		if err != nil {
-			return err
-		}
-		arwriter.WriteHeader(&ar.Header{
-			Name:    name,
-			ModTime: time.Unix(0, 0),
-			Uid:     0,
-			Gid:     0,
-			Mode:    0644,
-			Size:    st.Size(),
-		})
-		n, err := io.Copy(arwriter, objfile)
-		if err != nil {
-			return err
-		}
-		if n != st.Size() {
-			return errors.New("file modified during ar creation: " + arpath)
-		}
-	}
 
 	// Give the caller the resulting file. The callback must copy the file,
 	// because after it returns the temporary directory will be removed.
-	arfile.Close()
 	return callback(arpath)
 }

builder/cc1as.cpp

@@ -0,0 +1,504 @@
+// +build byollvm
+
+//===-- cc1as.cpp - Clang Assembler  --------------------------------------===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+//
+// This is the entry point to the clang -cc1as functionality, which implements
+// the direct interface to the LLVM MC based assembler.
+//
+//===----------------------------------------------------------------------===//
+
+#include "clang/Basic/Diagnostic.h"
+#include "clang/Basic/DiagnosticOptions.h"
+#include "clang/Driver/DriverDiagnostic.h"
+#include "clang/Driver/Options.h"
+#include "clang/Frontend/FrontendDiagnostic.h"
+#include "clang/Frontend/TextDiagnosticPrinter.h"
+#include "clang/Frontend/Utils.h"
+#include "llvm/ADT/STLExtras.h"
+#include "llvm/ADT/StringSwitch.h"
+#include "llvm/ADT/Triple.h"
+#include "llvm/IR/DataLayout.h"
+#include "llvm/MC/MCAsmBackend.h"
+#include "llvm/MC/MCAsmInfo.h"
+#include "llvm/MC/MCCodeEmitter.h"
+#include "llvm/MC/MCContext.h"
+#include "llvm/MC/MCInstrInfo.h"
+#include "llvm/MC/MCObjectFileInfo.h"
+#include "llvm/MC/MCObjectWriter.h"
+#include "llvm/MC/MCParser/MCAsmParser.h"
+#include "llvm/MC/MCParser/MCTargetAsmParser.h"
+#include "llvm/MC/MCRegisterInfo.h"
+#include "llvm/MC/MCSectionMachO.h"
+#include "llvm/MC/MCStreamer.h"
+#include "llvm/MC/MCSubtargetInfo.h"
+#include "llvm/MC/MCTargetOptions.h"
+#include "llvm/Option/Arg.h"
+#include "llvm/Option/ArgList.h"
+#include "llvm/Option/OptTable.h"
+#include "llvm/Support/CommandLine.h"
+#include "llvm/Support/ErrorHandling.h"
+#include "llvm/Support/FileSystem.h"
+#include "llvm/Support/FormattedStream.h"
+#include "llvm/Support/Host.h"
+#include "llvm/Support/MemoryBuffer.h"
+#include "llvm/Support/Path.h"
+#include "llvm/Support/Signals.h"
+#include "llvm/Support/SourceMgr.h"
+#include "llvm/Support/TargetRegistry.h"
+#include "llvm/Support/TargetSelect.h"
+#include "llvm/Support/Timer.h"
+#include "llvm/Support/raw_ostream.h"
+#include <memory>
+#include <system_error>
+using namespace clang;
+using namespace clang::driver;
+using namespace clang::driver::options;
+using namespace llvm;
+using namespace llvm::opt;
+
+#include "cc1as.h"
+
+bool AssemblerInvocation::CreateFromArgs(AssemblerInvocation &Opts,
+                                         ArrayRef<const char *> Argv,
+                                         DiagnosticsEngine &Diags) {
+  bool Success = true;
+
+  // Parse the arguments.
+  std::unique_ptr<OptTable> OptTbl(createDriverOptTable());
+
+  const unsigned IncludedFlagsBitmask = options::CC1AsOption;
+  unsigned MissingArgIndex, MissingArgCount;
+  InputArgList Args = OptTbl->ParseArgs(Argv, MissingArgIndex, MissingArgCount,
+                                        IncludedFlagsBitmask);
+
+  // Check for missing argument error.
+  if (MissingArgCount) {
+    Diags.Report(diag::err_drv_missing_argument)
+        << Args.getArgString(MissingArgIndex) << MissingArgCount;
+    Success = false;
+  }
+
+  // Issue errors on unknown arguments.
+  for (const Arg *A : Args.filtered(OPT_UNKNOWN)) {
+    auto ArgString = A->getAsString(Args);
+    std::string Nearest;
+    if (OptTbl->findNearest(ArgString, Nearest, IncludedFlagsBitmask) > 1)
+      Diags.Report(diag::err_drv_unknown_argument) << ArgString;
+    else
+      Diags.Report(diag::err_drv_unknown_argument_with_suggestion)
+          << ArgString << Nearest;
+    Success = false;
+  }
+
+  // Construct the invocation.
+
+  // Target Options
+  Opts.Triple = llvm::Triple::normalize(Args.getLastArgValue(OPT_triple));
+  Opts.CPU = Args.getLastArgValue(OPT_target_cpu);
+  Opts.Features = Args.getAllArgValues(OPT_target_feature);
+
+  // Use the default target triple if unspecified.
+  if (Opts.Triple.empty())
+    Opts.Triple = llvm::sys::getDefaultTargetTriple();
+
+  // Language Options
+  Opts.IncludePaths = Args.getAllArgValues(OPT_I);
+  Opts.NoInitialTextSection = Args.hasArg(OPT_n);
+  Opts.SaveTemporaryLabels = Args.hasArg(OPT_msave_temp_labels);
+  // Any DebugInfoKind implies GenDwarfForAssembly.
+  Opts.GenDwarfForAssembly = Args.hasArg(OPT_debug_info_kind_EQ);
+
+  if (const Arg *A = Args.getLastArg(OPT_compress_debug_sections,
+                                     OPT_compress_debug_sections_EQ)) {
+    if (A->getOption().getID() == OPT_compress_debug_sections) {
+      // TODO: be more clever about the compression type auto-detection
+      Opts.CompressDebugSections = llvm::DebugCompressionType::GNU;
+    } else {
+      Opts.CompressDebugSections =
+          llvm::StringSwitch<llvm::DebugCompressionType>(A->getValue())
+              .Case("none", llvm::DebugCompressionType::None)
+              .Case("zlib", llvm::DebugCompressionType::Z)
+              .Case("zlib-gnu", llvm::DebugCompressionType::GNU)
+              .Default(llvm::DebugCompressionType::None);
+    }
+  }
+
+  Opts.RelaxELFRelocations = Args.hasArg(OPT_mrelax_relocations);
+  Opts.DwarfVersion = getLastArgIntValue(Args, OPT_dwarf_version_EQ, 2, Diags);
+  Opts.DwarfDebugFlags = Args.getLastArgValue(OPT_dwarf_debug_flags);
+  Opts.DwarfDebugProducer = Args.getLastArgValue(OPT_dwarf_debug_producer);
+  Opts.DebugCompilationDir = Args.getLastArgValue(OPT_fdebug_compilation_dir);
+  Opts.MainFileName = Args.getLastArgValue(OPT_main_file_name);
+
+  for (const auto &Arg : Args.getAllArgValues(OPT_fdebug_prefix_map_EQ))
+    Opts.DebugPrefixMap.insert(StringRef(Arg).split('='));
+
+  // Frontend Options
+  if (Args.hasArg(OPT_INPUT)) {
+    bool First = true;
+    for (const Arg *A : Args.filtered(OPT_INPUT)) {
+      if (First) {
+        Opts.InputFile = A->getValue();
+        First = false;
+      } else {
+        Diags.Report(diag::err_drv_unknown_argument) << A->getAsString(Args);
+        Success = false;
+      }
+    }
+  }
+  Opts.LLVMArgs = Args.getAllArgValues(OPT_mllvm);
+  Opts.OutputPath = Args.getLastArgValue(OPT_o);
+  Opts.SplitDwarfOutput = Args.getLastArgValue(OPT_split_dwarf_output);
+  if (Arg *A = Args.getLastArg(OPT_filetype)) {
+    StringRef Name = A->getValue();
+    unsigned OutputType = StringSwitch<unsigned>(Name)
+      .Case("asm", FT_Asm)
+      .Case("null", FT_Null)
+      .Case("obj", FT_Obj)
+      .Default(~0U);
+    if (OutputType == ~0U) {
+      Diags.Report(diag::err_drv_invalid_value) << A->getAsString(Args) << Name;
+      Success = false;
+    } else
+      Opts.OutputType = FileType(OutputType);
+  }
+  Opts.ShowHelp = Args.hasArg(OPT_help);
+  Opts.ShowVersion = Args.hasArg(OPT_version);
+
+  // Transliterate Options
+  Opts.OutputAsmVariant =
+      getLastArgIntValue(Args, OPT_output_asm_variant, 0, Diags);
+  Opts.ShowEncoding = Args.hasArg(OPT_show_encoding);
+  Opts.ShowInst = Args.hasArg(OPT_show_inst);
+
+  // Assemble Options
+  Opts.RelaxAll = Args.hasArg(OPT_mrelax_all);
+  Opts.NoExecStack = Args.hasArg(OPT_mno_exec_stack);
+  Opts.FatalWarnings = Args.hasArg(OPT_massembler_fatal_warnings);
+  Opts.RelocationModel = Args.getLastArgValue(OPT_mrelocation_model, "pic");
+  Opts.TargetABI = Args.getLastArgValue(OPT_target_abi);
+  Opts.IncrementalLinkerCompatible =
+      Args.hasArg(OPT_mincremental_linker_compatible);
+  Opts.SymbolDefs = Args.getAllArgValues(OPT_defsym);
+
+  // EmbedBitcode Option. If -fembed-bitcode is enabled, set the flag.
+  // EmbedBitcode behaves the same for all embed options for assembly files.
+  if (auto *A = Args.getLastArg(OPT_fembed_bitcode_EQ)) {
+    Opts.EmbedBitcode = llvm::StringSwitch<unsigned>(A->getValue())
+                            .Case("all", 1)
+                            .Case("bitcode", 1)
+                            .Case("marker", 1)
+                            .Default(0);
+  }
+
+  return Success;
+}
+
+static std::unique_ptr<raw_fd_ostream>
+getOutputStream(StringRef Path, DiagnosticsEngine &Diags, bool Binary) {
+  // Make sure that the Out file gets unlinked from the disk if we get a
+  // SIGINT.
+  if (Path != "-")
+    sys::RemoveFileOnSignal(Path);
+
+  std::error_code EC;
+  auto Out = llvm::make_unique<raw_fd_ostream>(
+      Path, EC, (Binary ? sys::fs::F_None : sys::fs::F_Text));
+  if (EC) {
+    Diags.Report(diag::err_fe_unable_to_open_output) << Path << EC.message();
+    return nullptr;
+  }
+
+  return Out;
+}
+
+bool ExecuteAssembler(AssemblerInvocation &Opts, DiagnosticsEngine &Diags) {
+  // Get the target specific parser.
+  std::string Error;
+  const Target *TheTarget = TargetRegistry::lookupTarget(Opts.Triple, Error);
+  if (!TheTarget)
+    return Diags.Report(diag::err_target_unknown_triple) << Opts.Triple;
+
+  ErrorOr<std::unique_ptr<MemoryBuffer>> Buffer =
+      MemoryBuffer::getFileOrSTDIN(Opts.InputFile);
+
+  if (std::error_code EC = Buffer.getError()) {
+    Error = EC.message();
+    return Diags.Report(diag::err_fe_error_reading) << Opts.InputFile;
+  }
+
+  SourceMgr SrcMgr;
+
+  // Tell SrcMgr about this buffer, which is what the parser will pick up.
+  unsigned BufferIndex = SrcMgr.AddNewSourceBuffer(std::move(*Buffer), SMLoc());
+
+  // Record the location of the include directories so that the lexer can find
+  // it later.
+  SrcMgr.setIncludeDirs(Opts.IncludePaths);
+
+  std::unique_ptr<MCRegisterInfo> MRI(TheTarget->createMCRegInfo(Opts.Triple));
+  assert(MRI && "Unable to create target register info!");
+
+  std::unique_ptr<MCAsmInfo> MAI(TheTarget->createMCAsmInfo(*MRI, Opts.Triple));
+  assert(MAI && "Unable to create target asm info!");
+
+  // Ensure MCAsmInfo initialization occurs before any use, otherwise sections
+  // may be created with a combination of default and explicit settings.
+  MAI->setCompressDebugSections(Opts.CompressDebugSections);
+
+  MAI->setRelaxELFRelocations(Opts.RelaxELFRelocations);
+
+  bool IsBinary = Opts.OutputType == AssemblerInvocation::FT_Obj;
+  if (Opts.OutputPath.empty())
+    Opts.OutputPath = "-";
+  std::unique_ptr<raw_fd_ostream> FDOS =
+      getOutputStream(Opts.OutputPath, Diags, IsBinary);
+  if (!FDOS)
+    return true;
+  std::unique_ptr<raw_fd_ostream> DwoOS;
+  if (!Opts.SplitDwarfOutput.empty())
+    DwoOS = getOutputStream(Opts.SplitDwarfOutput, Diags, IsBinary);
+
+  // FIXME: This is not pretty. MCContext has a ptr to MCObjectFileInfo and
+  // MCObjectFileInfo needs a MCContext reference in order to initialize itself.
+  std::unique_ptr<MCObjectFileInfo> MOFI(new MCObjectFileInfo());
+
+  MCContext Ctx(MAI.get(), MRI.get(), MOFI.get(), &SrcMgr);
+
+  bool PIC = false;
+  if (Opts.RelocationModel == "static") {
+    PIC = false;
+  } else if (Opts.RelocationModel == "pic") {
+    PIC = true;
+  } else {
+    assert(Opts.RelocationModel == "dynamic-no-pic" &&
+           "Invalid PIC model!");
+    PIC = false;
+  }
+
+  MOFI->InitMCObjectFileInfo(Triple(Opts.Triple), PIC, Ctx);
+  if (Opts.SaveTemporaryLabels)
+    Ctx.setAllowTemporaryLabels(false);
+  if (Opts.GenDwarfForAssembly)
+    Ctx.setGenDwarfForAssembly(true);
+  if (!Opts.DwarfDebugFlags.empty())
+    Ctx.setDwarfDebugFlags(StringRef(Opts.DwarfDebugFlags));
+  if (!Opts.DwarfDebugProducer.empty())
+    Ctx.setDwarfDebugProducer(StringRef(Opts.DwarfDebugProducer));
+  if (!Opts.DebugCompilationDir.empty())
+    Ctx.setCompilationDir(Opts.DebugCompilationDir);
+  else {
+    // If no compilation dir is set, try to use the current directory.
+    SmallString<128> CWD;
+    if (!sys::fs::current_path(CWD))
+      Ctx.setCompilationDir(CWD);
+  }
+  if (!Opts.DebugPrefixMap.empty())
+    for (const auto &KV : Opts.DebugPrefixMap)
+      Ctx.addDebugPrefixMapEntry(KV.first, KV.second);
+  if (!Opts.MainFileName.empty())
+    Ctx.setMainFileName(StringRef(Opts.MainFileName));
+  Ctx.setDwarfVersion(Opts.DwarfVersion);
+  if (Opts.GenDwarfForAssembly)
+    Ctx.setGenDwarfRootFile(Opts.InputFile,
+                            SrcMgr.getMemoryBuffer(BufferIndex)->getBuffer());
+
+  // Build up the feature string from the target feature list.
+  std::string FS;
+  if (!Opts.Features.empty()) {
+    FS = Opts.Features[0];
+    for (unsigned i = 1, e = Opts.Features.size(); i != e; ++i)
+      FS += "," + Opts.Features[i];
+  }
+
+  std::unique_ptr<MCStreamer> Str;
+
+  std::unique_ptr<MCInstrInfo> MCII(TheTarget->createMCInstrInfo());
+  std::unique_ptr<MCSubtargetInfo> STI(
+      TheTarget->createMCSubtargetInfo(Opts.Triple, Opts.CPU, FS));
+
+  raw_pwrite_stream *Out = FDOS.get();
+  std::unique_ptr<buffer_ostream> BOS;
+
+  MCTargetOptions MCOptions;
+  MCOptions.ABIName = Opts.TargetABI;
+
+  // FIXME: There is a bit of code duplication with addPassesToEmitFile.
+  if (Opts.OutputType == AssemblerInvocation::FT_Asm) {
+    MCInstPrinter *IP = TheTarget->createMCInstPrinter(
+        llvm::Triple(Opts.Triple), Opts.OutputAsmVariant, *MAI, *MCII, *MRI);
+
+    std::unique_ptr<MCCodeEmitter> CE;
+    if (Opts.ShowEncoding)
+      CE.reset(TheTarget->createMCCodeEmitter(*MCII, *MRI, Ctx));
+    std::unique_ptr<MCAsmBackend> MAB(
+        TheTarget->createMCAsmBackend(*STI, *MRI, MCOptions));
+
+    auto FOut = llvm::make_unique<formatted_raw_ostream>(*Out);
+    Str.reset(TheTarget->createAsmStreamer(
+        Ctx, std::move(FOut), /*asmverbose*/ true,
+        /*useDwarfDirectory*/ true, IP, std::move(CE), std::move(MAB),
+        Opts.ShowInst));
+  } else if (Opts.OutputType == AssemblerInvocation::FT_Null) {
+    Str.reset(createNullStreamer(Ctx));
+  } else {
+    assert(Opts.OutputType == AssemblerInvocation::FT_Obj &&
+           "Invalid file type!");
+    if (!FDOS->supportsSeeking()) {
+      BOS = make_unique<buffer_ostream>(*FDOS);
+      Out = BOS.get();
+    }
+
+    std::unique_ptr<MCCodeEmitter> CE(
+        TheTarget->createMCCodeEmitter(*MCII, *MRI, Ctx));
+    std::unique_ptr<MCAsmBackend> MAB(
+        TheTarget->createMCAsmBackend(*STI, *MRI, MCOptions));
+    std::unique_ptr<MCObjectWriter> OW =
+        DwoOS ? MAB->createDwoObjectWriter(*Out, *DwoOS)
+              : MAB->createObjectWriter(*Out);
+
+    Triple T(Opts.Triple);
+    Str.reset(TheTarget->createMCObjectStreamer(
+        T, Ctx, std::move(MAB), std::move(OW), std::move(CE), *STI,
+        Opts.RelaxAll, Opts.IncrementalLinkerCompatible,
+        /*DWARFMustBeAtTheEnd*/ true));
+    Str.get()->InitSections(Opts.NoExecStack);
+  }
+
+  // When -fembed-bitcode is passed to clang_as, a 1-byte marker
+  // is emitted in __LLVM,__asm section if the object file is MachO format.
+  if (Opts.EmbedBitcode && Ctx.getObjectFileInfo()->getObjectFileType() ==
+                               MCObjectFileInfo::IsMachO) {
+    MCSection *AsmLabel = Ctx.getMachOSection(
+        "__LLVM", "__asm", MachO::S_REGULAR, 4, SectionKind::getReadOnly());
+    Str.get()->SwitchSection(AsmLabel);
+    Str.get()->EmitZeros(1);
+  }
+
+  // Assembly to object compilation should leverage assembly info.
+  Str->setUseAssemblerInfoForParsing(true);
+
+  bool Failed = false;
+
+  std::unique_ptr<MCAsmParser> Parser(
+      createMCAsmParser(SrcMgr, Ctx, *Str.get(), *MAI));
+
+  // FIXME: init MCTargetOptions from sanitizer flags here.
+  std::unique_ptr<MCTargetAsmParser> TAP(
+      TheTarget->createMCAsmParser(*STI, *Parser, *MCII, MCOptions));
+  if (!TAP)
+    Failed = Diags.Report(diag::err_target_unknown_triple) << Opts.Triple;
+
+  // Set values for symbols, if any.
+  for (auto &S : Opts.SymbolDefs) {
+    auto Pair = StringRef(S).split('=');
+    auto Sym = Pair.first;
+    auto Val = Pair.second;
+    int64_t Value;
+    // We have already error checked this in the driver.
+    Val.getAsInteger(0, Value);
+    Ctx.setSymbolValue(Parser->getStreamer(), Sym, Value);
+  }
+
+  if (!Failed) {
+    Parser->setTargetParser(*TAP.get());
+    Failed = Parser->Run(Opts.NoInitialTextSection);
+  }
+
+  // Close Streamer first.
+  // It might have a reference to the output stream.
+  Str.reset();
+  // Close the output stream early.
+  BOS.reset();
+  FDOS.reset();
+
+  // Delete output file if there were errors.
+  if (Failed) {
+    if (Opts.OutputPath != "-")
+      sys::fs::remove(Opts.OutputPath);
+    if (!Opts.SplitDwarfOutput.empty() && Opts.SplitDwarfOutput != "-")
+      sys::fs::remove(Opts.SplitDwarfOutput);
+  }
+
+  return Failed;
+}
+
+static void LLVMErrorHandler(void *UserData, const std::string &Message,
+                             bool GenCrashDiag) {
+  DiagnosticsEngine &Diags = *static_cast<DiagnosticsEngine*>(UserData);
+
+  Diags.Report(diag::err_fe_error_backend) << Message;
+
+  // We cannot recover from llvm errors.
+  exit(1);
+}
+
+int cc1as_main(ArrayRef<const char *> Argv, const char *Argv0, void *MainAddr) {
+  // Initialize targets and assembly printers/parsers.
+  InitializeAllTargetInfos();
+  InitializeAllTargetMCs();
+  InitializeAllAsmParsers();
+
+  // Construct our diagnostic client.
+  IntrusiveRefCntPtr<DiagnosticOptions> DiagOpts = new DiagnosticOptions();
+  TextDiagnosticPrinter *DiagClient
+    = new TextDiagnosticPrinter(errs(), &*DiagOpts);
+  DiagClient->setPrefix("clang -cc1as");
+  IntrusiveRefCntPtr<DiagnosticIDs> DiagID(new DiagnosticIDs());
+  DiagnosticsEngine Diags(DiagID, &*DiagOpts, DiagClient);
+
+  // Set an error handler, so that any LLVM backend diagnostics go through our
+  // error handler.
+  ScopedFatalErrorHandler FatalErrorHandler
+    (LLVMErrorHandler, static_cast<void*>(&Diags));
+
+  // Parse the arguments.
+  AssemblerInvocation Asm;
+  if (!AssemblerInvocation::CreateFromArgs(Asm, Argv, Diags))
+    return 1;
+
+  if (Asm.ShowHelp) {
+    std::unique_ptr<OptTable> Opts(driver::createDriverOptTable());
+    Opts->PrintHelp(llvm::outs(), "clang -cc1as [options] file...",
+                    "Clang Integrated Assembler",
+                    /*Include=*/driver::options::CC1AsOption, /*Exclude=*/0,
+                    /*ShowAllAliases=*/false);
+    return 0;
+  }
+
+  // Honor -version.
+  //
+  // FIXME: Use a better -version message?
+  if (Asm.ShowVersion) {
+    llvm::cl::PrintVersionMessage();
+    return 0;
+  }
+
+  // Honor -mllvm.
+  //
+  // FIXME: Remove this, one day.
+  if (!Asm.LLVMArgs.empty()) {
+    unsigned NumArgs = Asm.LLVMArgs.size();
+    auto Args = llvm::make_unique<const char*[]>(NumArgs + 2);
+    Args[0] = "clang (LLVM option parsing)";
+    for (unsigned i = 0; i != NumArgs; ++i)
+      Args[i + 1] = Asm.LLVMArgs[i].c_str();
+    Args[NumArgs + 1] = nullptr;
+    llvm::cl::ParseCommandLineOptions(NumArgs + 1, Args.get());
+  }
+
+  // Execute the invocation, unless there were parsing errors.
+  bool Failed = Diags.hasErrorOccurred() || ExecuteAssembler(Asm, Diags);
+
+  // If any timers were active but haven't been destroyed yet, print their
+  // results now.
+  TimerGroup::printAll(errs());
+
+  return !!Failed;
+}

builder/cc1as.h

@@ -0,0 +1,120 @@
+//===-- cc1as.h - Clang Assembler  ----------------------------------------===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+//
+// This is the entry point to the clang -cc1as functionality, which implements
+// the direct interface to the LLVM MC based assembler.
+//
+//===----------------------------------------------------------------------===//
+
+#include "llvm/ADT/ArrayRef.h"
+
+/// Helper class for representing a single invocation of the assembler.
+struct AssemblerInvocation {
+  /// @name Target Options
+  /// @{
+
+  /// The name of the target triple to assemble for.
+  std::string Triple;
+
+  /// If given, the name of the target CPU to determine which instructions
+  /// are legal.
+  std::string CPU;
+
+  /// The list of target specific features to enable or disable -- this should
+  /// be a list of strings starting with '+' or '-'.
+  std::vector<std::string> Features;
+
+  /// The list of symbol definitions.
+  std::vector<std::string> SymbolDefs;
+
+  /// @}
+  /// @name Language Options
+  /// @{
+
+  std::vector<std::string> IncludePaths;
+  unsigned NoInitialTextSection : 1;
+  unsigned SaveTemporaryLabels : 1;
+  unsigned GenDwarfForAssembly : 1;
+  unsigned RelaxELFRelocations : 1;
+  unsigned DwarfVersion;
+  std::string DwarfDebugFlags;
+  std::string DwarfDebugProducer;
+  std::string DebugCompilationDir;
+  std::map<const std::string, const std::string> DebugPrefixMap;
+  llvm::DebugCompressionType CompressDebugSections =
+      llvm::DebugCompressionType::None;
+  std::string MainFileName;
+  std::string SplitDwarfOutput;
+
+  /// @}
+  /// @name Frontend Options
+  /// @{
+
+  std::string InputFile;
+  std::vector<std::string> LLVMArgs;
+  std::string OutputPath;
+  enum FileType {
+    FT_Asm,  ///< Assembly (.s) output, transliterate mode.
+    FT_Null, ///< No output, for timing purposes.
+    FT_Obj   ///< Object file output.
+  };
+  FileType OutputType;
+  unsigned ShowHelp : 1;
+  unsigned ShowVersion : 1;
+
+  /// @}
+  /// @name Transliterate Options
+  /// @{
+
+  unsigned OutputAsmVariant;
+  unsigned ShowEncoding : 1;
+  unsigned ShowInst : 1;
+
+  /// @}
+  /// @name Assembler Options
+  /// @{
+
+  unsigned RelaxAll : 1;
+  unsigned NoExecStack : 1;
+  unsigned FatalWarnings : 1;
+  unsigned IncrementalLinkerCompatible : 1;
+  unsigned EmbedBitcode : 1;
+
+  /// The name of the relocation model to use.
+  std::string RelocationModel;
+
+  /// The ABI targeted by the backend. Specified using -target-abi. Empty
+  /// otherwise.
+  std::string TargetABI;
+
+  /// @}
+
+public:
+  AssemblerInvocation() {
+    Triple = "";
+    NoInitialTextSection = 0;
+    InputFile = "-";
+    OutputPath = "-";
+    OutputType = FT_Asm;
+    OutputAsmVariant = 0;
+    ShowInst = 0;
+    ShowEncoding = 0;
+    RelaxAll = 0;
+    NoExecStack = 0;
+    FatalWarnings = 0;
+    IncrementalLinkerCompatible = 0;
+    DwarfVersion = 0;
+    EmbedBitcode = 0;
+  }
+
+  static bool CreateFromArgs(AssemblerInvocation &Res,
+                             ArrayRef<const char *> Argv,
+                             DiagnosticsEngine &Diags);
+};
+
+bool ExecuteAssembler(AssemblerInvocation &Opts, DiagnosticsEngine &Diags);

builder/clang.cpp

@@ -0,0 +1,97 @@
+// +build byollvm
+
+#include <clang/Basic/DiagnosticOptions.h>
+#include <clang/CodeGen/CodeGenAction.h>
+#include <clang/Driver/Compilation.h>
+#include <clang/Driver/Driver.h>
+#include <clang/Frontend/CompilerInstance.h>
+#include <clang/Frontend/CompilerInvocation.h>
+#include <clang/Frontend/FrontendDiagnostic.h>
+#include <clang/Frontend/TextDiagnosticPrinter.h>
+#include <clang/FrontendTool/Utils.h>
+#include <llvm/ADT/IntrusiveRefCntPtr.h>
+#include <llvm/Option/Option.h>
+
+using namespace llvm;
+using namespace clang;
+
+#include "cc1as.h"
+
+// This file provides C wrappers for the builtin tools cc1 and cc1as
+// provided by Clang, and calls them as the driver would call them.
+
+extern "C" {
+
+bool tinygo_clang_driver(int argc, char **argv) {
+	std::vector<const char*> args(argv, argv + argc);
+
+	// The compiler invocation needs a DiagnosticsEngine so it can report problems
+	llvm::IntrusiveRefCntPtr<clang::DiagnosticOptions> DiagOpts = new clang::DiagnosticOptions();
+	clang::TextDiagnosticPrinter DiagnosticPrinter(llvm::errs(), &*DiagOpts);
+	clang::DiagnosticsEngine Diags(llvm::IntrusiveRefCntPtr<clang::DiagnosticIDs>(new clang::DiagnosticIDs()), &*DiagOpts, &DiagnosticPrinter, false);
+
+	// Create the clang driver
+	clang::driver::Driver TheDriver(args[0], llvm::sys::getDefaultTargetTriple(), Diags);
+
+	// Create the set of actions to perform
+	std::unique_ptr<clang::driver::Compilation> C(TheDriver.BuildCompilation(args));
+	if (!C) {
+		return false;
+	}
+	const clang::driver::JobList &Jobs = C->getJobs();
+
+	// There may be more than one job, for example for .S files
+	// (preprocessor + assembler).
+	for (auto Cmd : Jobs) {
+		// Select the tool: cc1 or cc1as.
+		const llvm::opt::ArgStringList &CCArgs = Cmd.getArguments();
+
+		if (strcmp(*CCArgs.data(), "-cc1") == 0) {
+			// This is the C frontend.
+			// Initialize a compiler invocation object from the clang (-cc1) arguments.
+			std::unique_ptr<clang::CompilerInstance> Clang(new clang::CompilerInstance());
+			bool success = clang::CompilerInvocation::CreateFromArgs(
+					Clang->getInvocation(),
+					const_cast<const char **>(CCArgs.data()),
+					const_cast<const char **>(CCArgs.data()) + CCArgs.size(),
+					Diags);
+			if (!success) {
+				return false;
+			}
+
+			// Create the actual diagnostics engine.
+			Clang->createDiagnostics();
+			if (!Clang->hasDiagnostics()) {
+				return false;
+			}
+
+			// Execute the frontend actions.
+			success = ExecuteCompilerInvocation(Clang.get());
+			if (!success) {
+				return false;
+			}
+
+		} else if (strcmp(*CCArgs.data(), "-cc1as") == 0) {
+			// This is the assembler frontend. Parse the arguments.
+			AssemblerInvocation Asm;
+			if (!AssemblerInvocation::CreateFromArgs(Asm, llvm::ArrayRef<const char*>(CCArgs).slice(1), Diags))
+				return false;
+
+			// Execute the invocation, unless there were parsing errors.
+			bool failed = Diags.hasErrorOccurred() || ExecuteAssembler(Asm, Diags);
+			if (failed) {
+				return false;
+			}
+
+		} else {
+			// Unknown tool, print the tool and exit.
+			fprintf(stderr, "unknown tool: %s\n", *CCArgs.data());
+			return false;
+		}
+	}
+
+	// Commands executed successfully.
+	return true;
+}
+
+} // extern "C"

builder/commands.go

@@ -0,0 +1,61 @@
+package builder
+
+import (
+	"errors"
+	"os"
+	"os/exec"
+	"runtime"
+	"strings"
+)
+
+// Commands lists command alternatives for various operating systems. These
+// commands may have a slightly different name across operating systems and
+// distributions or may not even exist in $PATH, in which case absolute paths
+// may be used.
+var commands = map[string][]string{
+	"clang":   {"clang-9"},
+	"ld.lld":  {"ld.lld-9", "ld.lld"},
+	"wasm-ld": {"wasm-ld-9", "wasm-ld"},
+}
+
+func init() {
+	// Add the path to a Homebrew-installed LLVM 9 for ease of use (no need to
+	// manually set $PATH).
+	if runtime.GOOS == "darwin" {
+		commands["clang"] = append(commands["clang"], "/usr/local/opt/llvm@9/bin/clang-9")
+		commands["ld.lld"] = append(commands["ld.lld"], "/usr/local/opt/llvm@9/bin/ld.lld")
+		commands["wasm-ld"] = append(commands["wasm-ld"], "/usr/local/opt/llvm@9/bin/wasm-ld")
+	}
+	// Add the path for when LLVM was installed with the installer from
+	// llvm.org, which by default doesn't add LLVM to the $PATH environment
+	// variable.
+	if runtime.GOOS == "windows" {
+		commands["clang"] = append(commands["clang"], "clang", "C:\\Program Files\\LLVM\\bin\\clang.exe")
+		commands["ld.lld"] = append(commands["ld.lld"], "lld", "C:\\Program Files\\LLVM\\bin\\lld.exe")
+		commands["wasm-ld"] = append(commands["wasm-ld"], "C:\\Program Files\\LLVM\\bin\\wasm-ld.exe")
+	}
+	// Add the path to the llvm90 installed from ports
+	if runtime.GOOS == "freebsd" {
+		commands["clang"] = append(commands["clang"], "/usr/local/llvm90/bin/clang-9")
+		commands["ld.lld"] = append(commands["ld.lld"], "/usr/local/llvm90/bin/ld.lld")
+		commands["wasm-ld"] = append(commands["wasm-ld"], "/usr/local/llvm90/bin/wasm-ld")
+	}
+}
+
+func execCommand(cmdNames []string, args ...string) error {
+	for _, cmdName := range cmdNames {
+		cmd := exec.Command(cmdName, args...)
+		cmd.Stdout = os.Stdout
+		cmd.Stderr = os.Stderr
+		err := cmd.Run()
+		if err != nil {
+			if err, ok := err.(*exec.Error); ok && (err.Err == exec.ErrNotFound || err.Err.Error() == "file does not exist") {
+				// this command was not found, try the next
+				continue
+			}
+			return err
+		}
+		return nil
+	}
+	return errors.New("none of these commands were found in your $PATH: " + strings.Join(cmdNames, " "))
+}

builder/compiler-builtin.go

@@ -0,0 +1,60 @@
+// +build byollvm
+
+package builder
+
+import (
+	"errors"
+	"os"
+	"os/exec"
+	"unsafe"
+
+	"github.com/tinygo-org/tinygo/goenv"
+)
+
+/*
+#cgo CXXFLAGS: -fno-rtti
+#include <stdbool.h>
+#include <stdlib.h>
+bool tinygo_clang_driver(int argc, char **argv);
+*/
+import "C"
+
+// runCCompiler invokes a C compiler with the given arguments.
+//
+// This version invokes the built-in Clang when trying to run the Clang compiler.
+func runCCompiler(command string, flags ...string) error {
+	switch command {
+	case "clang":
+		// Compile this with the internal Clang compiler.
+		headerPath := getClangHeaderPath(goenv.Get("TINYGOROOT"))
+		if headerPath == "" {
+			return errors.New("could not locate Clang headers")
+		}
+		flags = append(flags, "-I"+headerPath)
+		flags = append([]string{"tinygo:" + command}, flags...)
+		var cflag *C.char
+		buf := C.calloc(C.size_t(len(flags)), C.size_t(unsafe.Sizeof(cflag)))
+		cflags := (*[1 << 10]*C.char)(unsafe.Pointer(buf))[:len(flags):len(flags)]
+		for i, flag := range flags {
+			cflag := C.CString(flag)
+			cflags[i] = cflag
+			defer C.free(unsafe.Pointer(cflag))
+		}
+		ok := C.tinygo_clang_driver(C.int(len(flags)), (**C.char)(buf))
+		if !ok {
+			return errors.New("failed to compile using built-in clang")
+		}
+		return nil
+	default:
+		// Running some other compiler. Maybe it has been defined in the
+		// commands map (unlikely).
+		if cmdNames, ok := commands[command]; ok {
+			return execCommand(cmdNames, flags...)
+		}
+		// Alternatively, run the compiler directly.
+		cmd := exec.Command(command, flags...)
+		cmd.Stdout = os.Stdout
+		cmd.Stderr = os.Stderr
+		return cmd.Run()
+	}
+}

builder/compiler-external.go

@@ -0,0 +1,24 @@
+// +build !byollvm
+
+package builder
+
+// This file provides a way to a C compiler as an external command. See also:
+// clang-external.go
+
+import (
+	"os"
+	"os/exec"
+)
+
+// runCCompiler invokes a C compiler with the given arguments.
+//
+// This version always runs the compiler as an external command.
+func runCCompiler(command string, flags ...string) error {
+	if cmdNames, ok := commands[command]; ok {
+		return execCommand(cmdNames, flags...)
+	}
+	cmd := exec.Command(command, flags...)
+	cmd.Stdout = os.Stdout
+	cmd.Stderr = os.Stderr
+	return cmd.Run()
+}

builder/config.go

@@ -0,0 +1,39 @@
+package builder
+
+import (
+	"errors"
+	"fmt"
+
+	"github.com/tinygo-org/tinygo/compileopts"
+	"github.com/tinygo-org/tinygo/goenv"
+)
+
+// NewConfig builds a new Config object from a set of compiler options. It also
+// loads some information from the environment while doing that. For example, it
+// uses the currently active GOPATH (from the goenv package) to determine the Go
+// version to use.
+func NewConfig(options *compileopts.Options) (*compileopts.Config, error) {
+	spec, err := compileopts.LoadTarget(options.Target)
+	if err != nil {
+		return nil, err
+	}
+	goroot := goenv.Get("GOROOT")
+	if goroot == "" {
+		return nil, errors.New("cannot locate $GOROOT, please set it manually")
+	}
+	major, minor, err := getGorootVersion(goroot)
+	if err != nil {
+		return nil, fmt.Errorf("could not read version from GOROOT (%v): %v", goroot, err)
+	}
+	if major != 1 || (minor != 11 && minor != 12 && minor != 13) {
+		return nil, fmt.Errorf("requires go version 1.11, 1.12, or 1.13, got go%d.%d", major, minor)
+	}
+	clangHeaderPath := getClangHeaderPath(goenv.Get("TINYGOROOT"))
+	return &compileopts.Config{
+		Options:        options,
+		Target:         spec,
+		GoMinorVersion: minor,
+		ClangHeaders:   clangHeaderPath,
+		TestConfig:     options.TestConfig,
+	}, nil
+}

builder/env.go

@@ -0,0 +1,127 @@
+package builder
+
+import (
+	"errors"
+	"fmt"
+	"io"
+	"io/ioutil"
+	"os"
+	"os/exec"
+	"path/filepath"
+	"regexp"
+	"sort"
+	"strings"
+)
+
+// getGorootVersion returns the major and minor version for a given GOROOT path.
+// If the goroot cannot be determined, (0, 0) is returned.
+func getGorootVersion(goroot string) (major, minor int, err error) {
+	s, err := GorootVersionString(goroot)
+	if err != nil {
+		return 0, 0, err
+	}
+
+	if s == "" || s[:2] != "go" {
+		return 0, 0, errors.New("could not parse Go version: version does not start with 'go' prefix")
+	}
+
+	parts := strings.Split(s[2:], ".")
+	if len(parts) < 2 {
+		return 0, 0, errors.New("could not parse Go version: version has less than two parts")
+	}
+
+	// Ignore the errors, we don't really handle errors here anyway.
+	var trailing string
+	n, err := fmt.Sscanf(s, "go%d.%d%s", &major, &minor, &trailing)
+	if n == 2 && err == io.EOF {
+		// Means there were no trailing characters (i.e., not an alpha/beta)
+		err = nil
+	}
+	if err != nil {
+		return 0, 0, fmt.Errorf("failed to parse version: %s", err)
+	}
+	return
+}
+
+// GorootVersionString returns the version string as reported by the Go
+// toolchain for the given GOROOT path. It is usually of the form `go1.x.y` but
+// can have some variations (for beta releases, for example).
+func GorootVersionString(goroot string) (string, error) {
+	if data, err := ioutil.ReadFile(filepath.Join(
+		goroot, "src", "runtime", "internal", "sys", "zversion.go")); err == nil {
+
+		r := regexp.MustCompile("const TheVersion = `(.*)`")
+		matches := r.FindSubmatch(data)
+		if len(matches) != 2 {
+			return "", errors.New("Invalid go version output:\n" + string(data))
+		}
+
+		return string(matches[1]), nil
+
+	} else if data, err := ioutil.ReadFile(filepath.Join(goroot, "VERSION")); err == nil {
+		return string(data), nil
+
+	} else {
+		return "", err
+	}
+}
+
+// getClangHeaderPath returns the path to the built-in Clang headers. It tries
+// multiple locations, which should make it find the directory when installed in
+// various ways.
+func getClangHeaderPath(TINYGOROOT string) string {
+	// Check whether we're running from the source directory.
+	path := filepath.Join(TINYGOROOT, "llvm-project", "clang", "lib", "Headers")
+	if _, err := os.Stat(path); !os.IsNotExist(err) {
+		return path
+	}
+
+	// Check whether we're running from the installation directory.
+	path = filepath.Join(TINYGOROOT, "lib", "clang", "include")
+	if _, err := os.Stat(path); !os.IsNotExist(err) {
+		return path
+	}
+
+	// It looks like we are built with a system-installed LLVM. Do a last
+	// attempt: try to use Clang headers relative to the clang binary.
+	for _, cmdName := range commands["clang"] {
+		binpath, err := exec.LookPath(cmdName)
+		if err == nil {
+			// This should be the command that will also be used by
+			// execCommand. To avoid inconsistencies, make sure we use the
+			// headers relative to this command.
+			binpath, err = filepath.EvalSymlinks(binpath)
+			if err != nil {
+				// Unexpected.
+				return ""
+			}
+			// Example executable:
+			//     /usr/lib/llvm-9/bin/clang
+			// Example include path:
+			//     /usr/lib/llvm-9/lib/clang/9.0.1/include/
+			llvmRoot := filepath.Dir(filepath.Dir(binpath))
+			clangVersionRoot := filepath.Join(llvmRoot, "lib", "clang")
+			dirs, err := ioutil.ReadDir(clangVersionRoot)
+			if err != nil {
+				// Unexpected.
+				continue
+			}
+			dirnames := make([]string, len(dirs))
+			for i, d := range dirs {
+				dirnames[i] = d.Name()
+			}
+			sort.Strings(dirnames)
+			// Check for the highest version first.
+			for i := len(dirnames) - 1; i >= 0; i-- {
+				path := filepath.Join(clangVersionRoot, dirnames[i], "include")
+				_, err := os.Stat(filepath.Join(path, "stdint.h"))
+				if err == nil {
+					return path
+				}
+			}
+		}
+	}
+
+	// Could not find it.
+	return ""
+}

builder/error.go

@@ -0,0 +1,39 @@
+package builder
+
+// MultiError is a list of multiple errors (actually: diagnostics) returned
+// during LLVM IR generation.
+type MultiError struct {
+	Errs []error
+}
+
+func (e *MultiError) Error() string {
+	// Return the first error, to conform to the error interface. Clients should
+	// really do a type-assertion on *MultiError.
+	return e.Errs[0].Error()
+}
+
+// newMultiError returns a *MultiError if there is more than one error, or
+// returns that error directly when there is only one. Passing an empty slice
+// will lead to a panic.
+func newMultiError(errs []error) error {
+	switch len(errs) {
+	case 0:
+		panic("attempted to create empty MultiError")
+	case 1:
+		return errs[0]
+	default:
+		return &MultiError{errs}
+	}
+}
+
+// commandError is an error type to wrap os/exec.Command errors. This provides
+// some more information regarding what went wrong while running a command.
+type commandError struct {
+	Msg  string
+	File string
+	Err  error
+}
+
+func (e *commandError) Error() string {
+	return e.Msg + " " + e.File + ": " + e.Err.Error()
+}

builder/linker-builtin.go

@@ -1,6 +1,6 @@
 // +build byollvm
 
-package main
+package builder
 
 // This file provides a Link() function that uses the bundled lld if possible.
 
@@ -9,6 +9,8 @@ import (
 	"os"
 	"os/exec"
 	"unsafe"
+
+	"github.com/tinygo-org/tinygo/goenv"
 )
 
 /*
@@ -19,10 +21,10 @@ bool tinygo_link_wasm(int argc, char **argv);
 */
 import "C"
 
-// Link invokes a linker with the given name and flags.
+// link invokes a linker with the given name and flags.
 //
 // This version uses the built-in linker when trying to use lld.
-func Link(linker string, flags ...string) error {
+func link(linker string, flags ...string) error {
 	switch linker {
 	case "ld.lld":
 		flags = append([]string{"tinygo:" + linker}, flags...)
@@ -63,7 +65,7 @@ func Link(linker string, flags ...string) error {
 		cmd := exec.Command(linker, flags...)
 		cmd.Stdout = os.Stdout
 		cmd.Stderr = os.Stderr
-		cmd.Dir = sourceDir()
+		cmd.Dir = goenv.Get("TINYGOROOT")
 		return cmd.Run()
 	}
 }

builder/linker-external.go

@@ -1,6 +1,6 @@
 // +build !byollvm
 
-package main
+package builder
 
 // This file provides a Link() function that always runs an external command. It
 // is provided for when tinygo is built without linking to liblld.
@@ -8,18 +8,20 @@ package main
 import (
 	"os"
 	"os/exec"
+
+	"github.com/tinygo-org/tinygo/goenv"
 )
 
-// Link invokes a linker with the given name and arguments.
+// link invokes a linker with the given name and arguments.
 //
 // This version always runs the linker as an external command.
-func Link(linker string, flags ...string) error {
+func link(linker string, flags ...string) error {
 	if cmdNames, ok := commands[linker]; ok {
 		return execCommand(cmdNames, flags...)
 	}
 	cmd := exec.Command(linker, flags...)
 	cmd.Stdout = os.Stdout
 	cmd.Stderr = os.Stderr
-	cmd.Dir = sourceDir()
+	cmd.Dir = goenv.Get("TINYGOROOT")
 	return cmd.Run()
 }

builder/objcopy.go

@@ -1,4 +1,4 @@
-package main
+package builder
 
 import (
 	"debug/elf"
@@ -10,31 +10,31 @@ import (
 	"github.com/marcinbor85/gohex"
 )
 
-// ObjcopyError is an error returned by functions that act like objcopy.
-type ObjcopyError struct {
+// objcopyError is an error returned by functions that act like objcopy.
+type objcopyError struct {
 	Op  string
 	Err error
 }
 
-func (e ObjcopyError) Error() string {
+func (e objcopyError) Error() string {
 	if e.Err == nil {
 		return e.Op
 	}
 	return e.Op + ": " + e.Err.Error()
 }
 
-type ProgSlice []*elf.Prog
+type progSlice []*elf.Prog
 
-func (s ProgSlice) Len() int           { return len(s) }
-func (s ProgSlice) Less(i, j int) bool { return s[i].Paddr < s[j].Paddr }
-func (s ProgSlice) Swap(i, j int)      { s[i], s[j] = s[j], s[i] }
+func (s progSlice) Len() int           { return len(s) }
+func (s progSlice) Less(i, j int) bool { return s[i].Paddr < s[j].Paddr }
+func (s progSlice) Swap(i, j int)      { s[i], s[j] = s[j], s[i] }
 
-// ExtractROM extracts a firmware image and the first load address from the
+// extractROM extracts a firmware image and the first load address from the
 // given ELF file. It tries to emulate the behavior of objcopy.
-func ExtractROM(path string) (uint64, []byte, error) {
+func extractROM(path string) (uint64, []byte, error) {
 	f, err := elf.Open(path)
 	if err != nil {
-		return 0, nil, ObjcopyError{"failed to open ELF file to extract text segment", err}
+		return 0, nil, objcopyError{"failed to open ELF file to extract text segment", err}
 	}
 	defer f.Close()
 
@@ -56,7 +56,7 @@ func ExtractROM(path string) (uint64, []byte, error) {
 		}
 	}
 
-	progs := make(ProgSlice, 0, 2)
+	progs := make(progSlice, 0, 2)
 	for _, prog := range f.Progs {
 		if prog.Type != elf.PT_LOAD || prog.Filesz == 0 {
 			continue
@@ -64,18 +64,18 @@ func ExtractROM(path string) (uint64, []byte, error) {
 		progs = append(progs, prog)
 	}
 	if len(progs) == 0 {
-		return 0, nil, ObjcopyError{"file does not contain ROM segments: " + path, nil}
+		return 0, nil, objcopyError{"file does not contain ROM segments: " + path, nil}
 	}
 	sort.Sort(progs)
 
 	var rom []byte
 	for _, prog := range progs {
 		if prog.Paddr != progs[0].Paddr+uint64(len(rom)) {
-			return 0, nil, ObjcopyError{"ROM segments are non-contiguous: " + path, nil}
+			return 0, nil, objcopyError{"ROM segments are non-contiguous: " + path, nil}
 		}
 		data, err := ioutil.ReadAll(prog.Open())
 		if err != nil {
-			return 0, nil, ObjcopyError{"failed to extract segment from ELF file: " + path, err}
+			return 0, nil, objcopyError{"failed to extract segment from ELF file: " + path, err}
 		}
 		rom = append(rom, data...)
 	}
@@ -91,9 +91,9 @@ func ExtractROM(path string) (uint64, []byte, error) {
 	}
 }
 
-// Objcopy converts an ELF file to a different (simpler) output file format:
+// objcopy converts an ELF file to a different (simpler) output file format:
 // .bin or .hex. It extracts only the .text section.
-func Objcopy(infile, outfile string) error {
+func objcopy(infile, outfile string) error {
 	f, err := os.OpenFile(outfile, os.O_RDWR|os.O_CREATE|os.O_TRUNC, 0666)
 	if err != nil {
 		return err
@@ -101,13 +101,17 @@ func Objcopy(infile, outfile string) error {
 	defer f.Close()
 
 	// Read the .text segment.
-	addr, data, err := ExtractROM(infile)
+	addr, data, err := extractROM(infile)
 	if err != nil {
 		return err
 	}
 
 	// Write to the file, in the correct format.
 	switch filepath.Ext(outfile) {
+	case ".gba":
+		// The address is not stored in a .gba file.
+		_, err := f.Write(data)
+		return err
 	case ".bin":
 		// The address is not stored in a .bin file (therefore you
 		// should use .hex files in most cases).
@@ -117,7 +121,7 @@ func Objcopy(infile, outfile string) error {
 		mem := gohex.NewMemory()
 		err := mem.AddBinary(uint32(addr), data)
 		if err != nil {
-			return ObjcopyError{"failed to create .hex file", err}
+			return objcopyError{"failed to create .hex file", err}
 		}
 		mem.DumpIntelHex(f, 16) // TODO: handle error
 		return nil

builder/sizes.go

@@ -1,4 +1,4 @@
-package main
+package builder
 
 import (
 	"debug/elf"
@@ -6,18 +6,18 @@ import (
 	"strings"
 )
 
-// Statistics about code size in a program.
-type ProgramSize struct {
-	Packages map[string]*PackageSize
-	Sum      *PackageSize
+// programSize contains size statistics per package of a compiled program.
+type programSize struct {
+	Packages map[string]*packageSize
+	Sum      *packageSize
 	Code     uint64
 	Data     uint64
 	BSS      uint64
 }
 
-// Return the list of package names (ProgramSize.Packages) sorted
-// alphabetically.
-func (ps *ProgramSize) SortedPackageNames() []string {
+// sortedPackageNames returns the list of package names (ProgramSize.Packages)
+// sorted alphabetically.
+func (ps *programSize) sortedPackageNames() []string {
 	names := make([]string, 0, len(ps.Packages))
 	for name := range ps.Packages {
 		names = append(names, name)
@@ -26,8 +26,9 @@ func (ps *ProgramSize) SortedPackageNames() []string {
 	return names
 }
 
-// The size of a package, calculated from the linked object file.
-type PackageSize struct {
+// packageSize contains the size of a package, calculated from the linked object
+// file.
+type packageSize struct {
 	Code   uint64
 	ROData uint64
 	Data   uint64
@@ -35,12 +36,12 @@ type PackageSize struct {
 }
 
 // Flash usage in regular microcontrollers.
-func (ps *PackageSize) Flash() uint64 {
+func (ps *packageSize) Flash() uint64 {
 	return ps.Code + ps.ROData + ps.Data
 }
 
 // Static RAM usage in regular microcontrollers.
-func (ps *PackageSize) RAM() uint64 {
+func (ps *packageSize) RAM() uint64 {
 	return ps.Data + ps.BSS
 }
 
@@ -64,8 +65,9 @@ func (l symbolList) Swap(i, j int) {
 	l[i], l[j] = l[j], l[i]
 }
 
-// Calculate program/data size breakdown of each package for a given ELF file.
-func Sizes(path string) (*ProgramSize, error) {
+// loadProgramSize calculate a program/data size breakdown of each package for a
+// given ELF file.
+func loadProgramSize(path string) (*programSize, error) {
 	file, err := elf.Open(path)
 	if err != nil {
 		return nil, err
@@ -115,7 +117,7 @@ func Sizes(path string) (*ProgramSize, error) {
 	}
 	sort.Sort(symbolList(symbols))
 
-	sizes := map[string]*PackageSize{}
+	sizes := map[string]*packageSize{}
 	var lastSymbolValue uint64
 	for _, symbol := range symbols {
 		symType := elf.ST_TYPE(symbol.Info)
@@ -129,7 +131,7 @@ func Sizes(path string) (*ProgramSize, error) {
 		}
 		pkgSize := sizes[pkgName]
 		if pkgSize == nil {
-			pkgSize = &PackageSize{}
+			pkgSize = &packageSize{}
 			sizes[pkgName] = pkgSize
 		}
 		if lastSymbolValue != symbol.Value || lastSymbolValue == 0 {
@@ -148,7 +150,7 @@ func Sizes(path string) (*ProgramSize, error) {
 		lastSymbolValue = symbol.Value
 	}
 
-	sum := &PackageSize{}
+	sum := &packageSize{}
 	for _, pkg := range sizes {
 		sum.Code += pkg.Code
 		sum.ROData += pkg.ROData
@@ -156,5 +158,5 @@ func Sizes(path string) (*ProgramSize, error) {
 		sum.BSS += pkg.BSS
 	}
 
-	return &ProgramSize{Packages: sizes, Code: sumCode, Data: sumData, BSS: sumBSS, Sum: sum}, nil
+	return &programSize{Packages: sizes, Code: sumCode, Data: sumData, BSS: sumBSS, Sum: sum}, nil
 }

builder/uf2.go

@@ -1,10 +1,11 @@
-// Converts firmware files from BIN to UF2 format before flashing.
+package builder
+
+// This file converts firmware files from BIN to UF2 format before flashing.
 //
 // For more information about the UF2 firmware file format, please see:
 // https://github.com/Microsoft/uf2
 //
 //
-package main
 
 import (
 	"bytes"
@@ -12,24 +13,24 @@ import (
 	"io/ioutil"
 )
 
-// ConvertELFFileToUF2File converts an ELF file to a UF2 file.
-func ConvertELFFileToUF2File(infile, outfile string) error {
+// convertELFFileToUF2File converts an ELF file to a UF2 file.
+func convertELFFileToUF2File(infile, outfile string) error {
 	// Read the .text segment.
-	targetAddress, data, err := ExtractROM(infile)
+	targetAddress, data, err := extractROM(infile)
 	if err != nil {
 		return err
 	}
 
-	output, _ := ConvertBinToUF2(data, uint32(targetAddress))
+	output, _ := convertBinToUF2(data, uint32(targetAddress))
 	return ioutil.WriteFile(outfile, output, 0644)
 }
 
-// ConvertBinToUF2 converts the binary bytes in input to UF2 formatted data.
-func ConvertBinToUF2(input []byte, targetAddr uint32) ([]byte, int) {
+// convertBinToUF2 converts the binary bytes in input to UF2 formatted data.
+func convertBinToUF2(input []byte, targetAddr uint32) ([]byte, int) {
 	blocks := split(input, 256)
 	output := make([]byte, 0)
 
-	bl := NewUF2Block(targetAddr)
+	bl := newUF2Block(targetAddr)
 	bl.SetNumBlocks(len(blocks))
 
 	for i := 0; i < len(blocks); i++ {
@@ -49,8 +50,8 @@ const (
 	uf2MagicEnd    = 0x0AB16F30 // Ditto
 )
 
-// UF2Block is the structure used for each UF2 code block sent to device.
-type UF2Block struct {
+// uf2Block is the structure used for each UF2 code block sent to device.
+type uf2Block struct {
 	magicStart0 uint32
 	magicStart1 uint32
 	flags       uint32
@@ -63,9 +64,9 @@ type UF2Block struct {
 	magicEnd    uint32
 }
 
-// NewUF2Block returns a new UF2Block struct that has been correctly populated
-func NewUF2Block(targetAddr uint32) *UF2Block {
-	return &UF2Block{magicStart0: uf2MagicStart0,
+// newUF2Block returns a new uf2Block struct that has been correctly populated
+func newUF2Block(targetAddr uint32) *uf2Block {
+	return &uf2Block{magicStart0: uf2MagicStart0,
 		magicStart1: uf2MagicStart1,
 		magicEnd:    uf2MagicEnd,
 		targetAddr:  targetAddr,
@@ -76,8 +77,8 @@ func NewUF2Block(targetAddr uint32) *UF2Block {
 	}
 }
 
-// Bytes converts the UF2Block to a slice of bytes that can be written to file.
-func (b *UF2Block) Bytes() []byte {
+// Bytes converts the uf2Block to a slice of bytes that can be written to file.
+func (b *uf2Block) Bytes() []byte {
 	buf := bytes.NewBuffer(make([]byte, 0, 512))
 	binary.Write(buf, binary.LittleEndian, b.magicStart0)
 	binary.Write(buf, binary.LittleEndian, b.magicStart1)
@@ -94,23 +95,23 @@ func (b *UF2Block) Bytes() []byte {
 }
 
 // IncrementAddress moves the target address pointer forward by count bytes.
-func (b *UF2Block) IncrementAddress(count uint32) {
+func (b *uf2Block) IncrementAddress(count uint32) {
 	b.targetAddr += b.payloadSize
 }
 
 // SetData sets the data to be used for the current block.
-func (b *UF2Block) SetData(d []byte) {
+func (b *uf2Block) SetData(d []byte) {
 	b.data = make([]byte, 476)
 	copy(b.data[:], d)
 }
 
 // SetBlockNo sets the current block number to be used.
-func (b *UF2Block) SetBlockNo(bn int) {
+func (b *uf2Block) SetBlockNo(bn int) {
 	b.blockNo = uint32(bn)
 }
 
 // SetNumBlocks sets the total number of blocks for this UF2 file.
-func (b *UF2Block) SetNumBlocks(total int) {
+func (b *uf2Block) SetNumBlocks(total int) {
 	b.numBlocks = uint32(total)
 }
 

cgo/cgo.go

@@ -12,12 +12,14 @@ package cgo
 // source file parsing.
 
 import (
+	"fmt"
 	"go/ast"
 	"go/token"
 	"sort"
 	"strconv"
 	"strings"
 
+	"github.com/google/shlex"
 	"golang.org/x/tools/go/ast/astutil"
 )
 
@@ -42,7 +44,7 @@ type cgoPackage struct {
 // constantInfo stores some information about a CGo constant found by libclang
 // and declared in the Go AST.
 type constantInfo struct {
-	expr *ast.BasicLit
+	expr ast.Expr
 	pos  token.Pos
 }
 
@@ -69,9 +71,11 @@ type typedefInfo struct {
 // elaboratedTypeInfo contains some information about an elaborated type
 // (struct, union) found in the C AST.
 type elaboratedTypeInfo struct {
-	typeExpr  *ast.StructType
-	pos       token.Pos
-	bitfields []bitfieldInfo
+	typeExpr   *ast.StructType
+	pos        token.Pos
+	bitfields  []bitfieldInfo
+	unionSize  int64 // union size in bytes, nonzero when union getters/setters should be created
+	unionAlign int64 // union alignment in bytes
 }
 
 // bitfieldInfo contains information about a single bitfield in a struct. It
@@ -131,6 +135,7 @@ var builtinAliases = map[string]struct{}{
 // somehow from C. This is done by adding some typedefs to get the size of each
 // type.
 const cgoTypes = `
+# 1 "<cgo>"
 typedef char                _Cgo_char;
 typedef signed char         _Cgo_schar;
 typedef unsigned char       _Cgo_uchar;
@@ -184,6 +189,7 @@ func Process(files []*ast.File, dir string, fset *token.FileSet, cflags []string
 			Name:    files[0].Name.Name,
 		},
 		Decls: []ast.Decl{
+			// import "unsafe"
 			&ast.GenDecl{
 				TokPos: p.generatedPos,
 				Tok:    token.IMPORT,
@@ -191,6 +197,32 @@ func Process(files []*ast.File, dir string, fset *token.FileSet, cflags []string
 					unsafeImport,
 				},
 			},
+			// var _ unsafe.Pointer
+			// This avoids type errors when the unsafe package is never used.
+			&ast.GenDecl{
+				Tok: token.VAR,
+				Specs: []ast.Spec{
+					&ast.ValueSpec{
+						Names: []*ast.Ident{
+							&ast.Ident{
+								Name: "_",
+								Obj: &ast.Object{
+									Kind: ast.Var,
+									Name: "_",
+								},
+							},
+						},
+						Type: &ast.SelectorExpr{
+							X: &ast.Ident{
+								Name: "unsafe",
+							},
+							Sel: &ast.Ident{
+								Name: "Pointer",
+							},
+						},
+					},
+				},
+			},
 		},
 		Imports: []*ast.ImportSpec{unsafeImport},
 	}
@@ -204,6 +236,7 @@ func Process(files []*ast.File, dir string, fset *token.FileSet, cflags []string
 	}
 
 	// Find `import "C"` statements in the file.
+	var statements []*ast.GenDecl
 	for _, f := range files {
 		for i := 0; i < len(f.Decls); i++ {
 			decl := f.Decls[i]
@@ -220,19 +253,16 @@ func Process(files []*ast.File, dir string, fset *token.FileSet, cflags []string
 			}
 			path, err := strconv.Unquote(spec.Path.Value)
 			if err != nil {
+				// This should not happen. An import path that is not properly
+				// quoted should not exist in a correct AST.
 				panic("could not parse import path: " + err.Error())
 			}
 			if path != "C" {
 				continue
 			}
-			cgoComment := genDecl.Doc.Text()
 
-			pos := genDecl.Pos()
-			if genDecl.Doc != nil {
-				pos = genDecl.Doc.Pos()
-			}
-			position := fset.PositionFor(pos, true)
-			p.parseFragment(cgoComment+cgoTypes, cflags, position.Filename, position.Line)
+			// Found a CGo statement.
+			statements = append(statements, genDecl)
 
 			// Remove this import declaration.
 			f.Decls = append(f.Decls[:i], f.Decls[i+1:]...)
@@ -243,6 +273,93 @@ func Process(files []*ast.File, dir string, fset *token.FileSet, cflags []string
 		//ast.Print(fset, f)
 	}
 
+	// Find all #cgo lines.
+	for _, genDecl := range statements {
+		if genDecl.Doc == nil {
+			continue
+		}
+		for _, comment := range genDecl.Doc.List {
+			for {
+				// Extract the #cgo line, and replace it with spaces.
+				// Replacing with spaces makes sure that error locations are
+				// still correct, while not interfering with parsing in any way.
+				lineStart := strings.Index(comment.Text, "#cgo ")
+				if lineStart < 0 {
+					break
+				}
+				lineLen := strings.IndexByte(comment.Text[lineStart:], '\n')
+				if lineLen < 0 {
+					lineLen = len(comment.Text) - lineStart
+				}
+				lineEnd := lineStart + lineLen
+				line := comment.Text[lineStart:lineEnd]
+				spaces := make([]byte, len(line))
+				for i := range spaces {
+					spaces[i] = ' '
+				}
+				lenBefore := len(comment.Text)
+				comment.Text = comment.Text[:lineStart] + string(spaces) + comment.Text[lineEnd:]
+				if len(comment.Text) != lenBefore {
+					println(lenBefore, len(comment.Text))
+					panic("length of preamble changed!")
+				}
+
+				// Get the text before the colon in the #cgo directive.
+				colon := strings.IndexByte(line, ':')
+				if colon < 0 {
+					p.addErrorAfter(comment.Slash, comment.Text[:lineStart], "missing colon in #cgo line")
+					continue
+				}
+
+				// Extract the fields before the colon. These fields are a list
+				// of build tags and the C environment variable.
+				fields := strings.Fields(line[4:colon])
+				if len(fields) == 0 {
+					p.addErrorAfter(comment.Slash, comment.Text[:lineStart+colon-1], "invalid #cgo line")
+					continue
+				}
+
+				if len(fields) > 1 {
+					p.addErrorAfter(comment.Slash, comment.Text[:lineStart+5], "not implemented: build constraints in #cgo line")
+					continue
+				}
+
+				name := fields[len(fields)-1]
+				value := line[colon+1:]
+				switch name {
+				case "CFLAGS":
+					flags, err := shlex.Split(value)
+					if err != nil {
+						// TODO: find the exact location where the error happened.
+						p.addErrorAfter(comment.Slash, comment.Text[:lineStart+colon+1], "failed to parse flags in #cgo line: "+err.Error())
+						continue
+					}
+					if err := checkCompilerFlags(name, flags); err != nil {
+						p.addErrorAfter(comment.Slash, comment.Text[:lineStart+colon+1], err.Error())
+						continue
+					}
+					cflags = append(cflags, flags...)
+				default:
+					startPos := strings.LastIndex(line[4:colon], name) + 4
+					p.addErrorAfter(comment.Slash, comment.Text[:lineStart+startPos], "invalid #cgo line: "+name)
+					continue
+				}
+			}
+		}
+	}
+
+	// Process all CGo imports.
+	for _, genDecl := range statements {
+		cgoComment := genDecl.Doc.Text()
+
+		pos := genDecl.Pos()
+		if genDecl.Doc != nil {
+			pos = genDecl.Doc.Pos()
+		}
+		position := fset.PositionFor(pos, true)
+		p.parseFragment(cgoComment+cgoTypes, cflags, position.Filename, position.Line)
+	}
+
 	// Declare functions found by libclang.
 	p.addFuncDecls()
 
@@ -343,18 +460,18 @@ func (p *cgoPackage) addFuncPtrDecls() {
 	if len(p.functions) == 0 {
 		return
 	}
-	gen := &ast.GenDecl{
-		TokPos: token.NoPos,
-		Tok:    token.VAR,
-		Lparen: token.NoPos,
-		Rparen: token.NoPos,
-	}
 	names := make([]string, 0, len(p.functions))
 	for name := range p.functions {
 		names = append(names, name)
 	}
 	sort.Strings(names)
 	for _, name := range names {
+		gen := &ast.GenDecl{
+			TokPos: token.NoPos,
+			Tok:    token.VAR,
+			Lparen: token.NoPos,
+			Rparen: token.NoPos,
+		}
 		fn := p.functions[name]
 		obj := &ast.Object{
 			Kind: ast.Typ,
@@ -379,34 +496,32 @@ func (p *cgoPackage) addFuncPtrDecls() {
 		}
 		obj.Decl = valueSpec
 		gen.Specs = append(gen.Specs, valueSpec)
+		p.generated.Decls = append(p.generated.Decls, gen)
 	}
-	p.generated.Decls = append(p.generated.Decls, gen)
 }
 
 // addConstDecls declares external C constants in the Go source.
 // It adds code like the following to the AST:
 //
-//     const (
-//         C.CONST_INT = 5
-//         C.CONST_FLOAT = 5.8
-//         // ...
-//     )
+//     const C.CONST_INT = 5
+//     const C.CONST_FLOAT = 5.8
+//     // ...
 func (p *cgoPackage) addConstDecls() {
 	if len(p.constants) == 0 {
 		return
 	}
-	gen := &ast.GenDecl{
-		TokPos: token.NoPos,
-		Tok:    token.CONST,
-		Lparen: token.NoPos,
-		Rparen: token.NoPos,
-	}
 	names := make([]string, 0, len(p.constants))
 	for name := range p.constants {
 		names = append(names, name)
 	}
 	sort.Strings(names)
 	for _, name := range names {
+		gen := &ast.GenDecl{
+			TokPos: token.NoPos,
+			Tok:    token.CONST,
+			Lparen: token.NoPos,
+			Rparen: token.NoPos,
+		}
 		constVal := p.constants[name]
 		obj := &ast.Object{
 			Kind: ast.Con,
@@ -422,34 +537,32 @@ func (p *cgoPackage) addConstDecls() {
 		}
 		obj.Decl = valueSpec
 		gen.Specs = append(gen.Specs, valueSpec)
+		p.generated.Decls = append(p.generated.Decls, gen)
 	}
-	p.generated.Decls = append(p.generated.Decls, gen)
 }
 
 // addVarDecls declares external C globals in the Go source.
 // It adds code like the following to the AST:
 //
-//     var (
-//         C.globalInt  int
-//         C.globalBool bool
-//         // ...
-//     )
+//     var C.globalInt  int
+//     var C.globalBool bool
+//     // ...
 func (p *cgoPackage) addVarDecls() {
 	if len(p.globals) == 0 {
 		return
 	}
-	gen := &ast.GenDecl{
-		TokPos: token.NoPos,
-		Tok:    token.VAR,
-		Lparen: token.NoPos,
-		Rparen: token.NoPos,
-	}
 	names := make([]string, 0, len(p.globals))
 	for name := range p.globals {
 		names = append(names, name)
 	}
 	sort.Strings(names)
 	for _, name := range names {
+		gen := &ast.GenDecl{
+			TokPos: token.NoPos,
+			Tok:    token.VAR,
+			Lparen: token.NoPos,
+			Rparen: token.NoPos,
+		}
 		global := p.globals[name]
 		obj := &ast.Object{
 			Kind: ast.Var,
@@ -465,31 +578,29 @@ func (p *cgoPackage) addVarDecls() {
 		}
 		obj.Decl = valueSpec
 		gen.Specs = append(gen.Specs, valueSpec)
+		p.generated.Decls = append(p.generated.Decls, gen)
 	}
-	p.generated.Decls = append(p.generated.Decls, gen)
 }
 
 // addTypeAliases aliases some built-in Go types with their equivalent C types.
 // It adds code like the following to the AST:
 //
-//     type (
-//         C.int8_t  = int8
-//         C.int16_t = int16
-//         // ...
-//     )
+//     type C.int8_t  = int8
+//     type C.int16_t = int16
+//     // ...
 func (p *cgoPackage) addTypeAliases() {
 	aliasKeys := make([]string, 0, len(cgoAliases))
 	for key := range cgoAliases {
 		aliasKeys = append(aliasKeys, key)
 	}
 	sort.Strings(aliasKeys)
-	gen := &ast.GenDecl{
-		TokPos: token.NoPos,
-		Tok:    token.TYPE,
-		Lparen: token.NoPos,
-		Rparen: token.NoPos,
-	}
 	for _, typeName := range aliasKeys {
+		gen := &ast.GenDecl{
+			TokPos: token.NoPos,
+			Tok:    token.TYPE,
+			Lparen: token.NoPos,
+			Rparen: token.NoPos,
+		}
 		goTypeName := cgoAliases[typeName]
 		obj := &ast.Object{
 			Kind: ast.Typ,
@@ -509,24 +620,24 @@ func (p *cgoPackage) addTypeAliases() {
 		}
 		obj.Decl = typeSpec
 		gen.Specs = append(gen.Specs, typeSpec)
+		p.generated.Decls = append(p.generated.Decls, gen)
 	}
-	p.generated.Decls = append(p.generated.Decls, gen)
 }
 
 func (p *cgoPackage) addTypedefs() {
 	if len(p.typedefs) == 0 {
 		return
 	}
-	gen := &ast.GenDecl{
-		TokPos: token.NoPos,
-		Tok:    token.TYPE,
-	}
 	names := make([]string, 0, len(p.typedefs))
 	for name := range p.typedefs {
 		names = append(names, name)
 	}
 	sort.Strings(names)
 	for _, name := range names {
+		gen := &ast.GenDecl{
+			TokPos: token.NoPos,
+			Tok:    token.TYPE,
+		}
 		typedef := p.typedefs[name]
 		typeName := "C." + name
 		isAlias := true
@@ -555,8 +666,8 @@ func (p *cgoPackage) addTypedefs() {
 		}
 		obj.Decl = typeSpec
 		gen.Specs = append(gen.Specs, typeSpec)
+		p.generated.Decls = append(p.generated.Decls, gen)
 	}
-	p.generated.Decls = append(p.generated.Decls, gen)
 }
 
 // addElaboratedTypes adds C elaborated types as aliases. These are the "struct
@@ -568,29 +679,47 @@ func (p *cgoPackage) addElaboratedTypes() {
 	if len(p.elaboratedTypes) == 0 {
 		return
 	}
-	gen := &ast.GenDecl{
-		TokPos: token.NoPos,
-		Tok:    token.TYPE,
-	}
 	names := make([]string, 0, len(p.elaboratedTypes))
 	for name := range p.elaboratedTypes {
 		names = append(names, name)
 	}
 	sort.Strings(names)
 	for _, name := range names {
+		gen := &ast.GenDecl{
+			TokPos: token.NoPos,
+			Tok:    token.TYPE,
+		}
 		typ := p.elaboratedTypes[name]
 		typeName := "C." + name
 		obj := &ast.Object{
 			Kind: ast.Typ,
 			Name: typeName,
 		}
+		typeExpr := typ.typeExpr
+		if typ.unionSize != 0 {
+			// Create getters/setters.
+			for _, field := range typ.typeExpr.Fields.List {
+				if len(field.Names) != 1 {
+					p.addError(typ.pos, fmt.Sprintf("union must have field with a single name, it has %d names", len(field.Names)))
+					continue
+				}
+				p.createUnionAccessor(field, typeName)
+			}
+			// Convert to a single-field struct type.
+			typeExpr = p.makeUnionField(typ)
+			if typeExpr == nil {
+				// There was an error, that was already added to the list of
+				// errors.
+				continue
+			}
+		}
 		typeSpec := &ast.TypeSpec{
 			Name: &ast.Ident{
 				NamePos: typ.pos,
 				Name:    typeName,
 				Obj:     obj,
 			},
-			Type: typ.typeExpr,
+			Type: typeExpr,
 		}
 		obj.Decl = typeSpec
 		gen.Specs = append(gen.Specs, typeSpec)
@@ -599,8 +728,185 @@ func (p *cgoPackage) addElaboratedTypes() {
 			p.createBitfieldGetter(bitfield, typeName)
 			p.createBitfieldSetter(bitfield, typeName)
 		}
+		p.generated.Decls = append(p.generated.Decls, gen)
 	}
-	p.generated.Decls = append(p.generated.Decls, gen)
+}
+
+// makeUnionField creates a new struct from an existing *elaboratedTypeInfo,
+// that has just a single field that must be accessed through special accessors.
+// It returns nil when there is an error. In case of an error, that error has
+// already been added to the list of errors using p.addError.
+func (p *cgoPackage) makeUnionField(typ *elaboratedTypeInfo) *ast.StructType {
+	unionFieldTypeName, ok := map[int64]string{
+		1: "uint8",
+		2: "uint16",
+		4: "uint32",
+		8: "uint64",
+	}[typ.unionAlign]
+	if !ok {
+		p.addError(typ.typeExpr.Struct, fmt.Sprintf("expected union alignment to be one of 1, 2, 4, or 8, but got %d", typ.unionAlign))
+		return nil
+	}
+	var unionFieldType ast.Expr = &ast.Ident{
+		NamePos: token.NoPos,
+		Name:    unionFieldTypeName,
+	}
+	if typ.unionSize != typ.unionAlign {
+		// A plain struct{uintX} isn't enough, we have to make a
+		// struct{[N]uintX} to make the union big enough.
+		if typ.unionSize/typ.unionAlign*typ.unionAlign != typ.unionSize {
+			p.addError(typ.typeExpr.Struct, fmt.Sprintf("union alignment (%d) must be a multiple of union alignment (%d)", typ.unionSize, typ.unionAlign))
+			return nil
+		}
+		unionFieldType = &ast.ArrayType{
+			Len: &ast.BasicLit{
+				Kind:  token.INT,
+				Value: strconv.FormatInt(typ.unionSize/typ.unionAlign, 10),
+			},
+			Elt: unionFieldType,
+		}
+	}
+	return &ast.StructType{
+		Struct: typ.typeExpr.Struct,
+		Fields: &ast.FieldList{
+			Opening: typ.typeExpr.Fields.Opening,
+			List: []*ast.Field{&ast.Field{
+				Names: []*ast.Ident{
+					&ast.Ident{
+						NamePos: typ.typeExpr.Fields.Opening,
+						Name:    "$union",
+					},
+				},
+				Type: unionFieldType,
+			}},
+			Closing: typ.typeExpr.Fields.Closing,
+		},
+	}
+}
+
+// createUnionAccessor creates a function that returns a typed pointer to a
+// union field for each field in a union. For example:
+//
+//     func (union *C.union_1) unionfield_d() *float64 {
+//         return (*float64)(unsafe.Pointer(&union.$union))
+//     }
+//
+// Where C.union_1 is defined as:
+//
+//     type C.union_1 struct{
+//         $union uint64
+//     }
+//
+// The returned pointer can be used to get or set the field, or get the pointer
+// to a subfield.
+func (p *cgoPackage) createUnionAccessor(field *ast.Field, typeName string) {
+	if len(field.Names) != 1 {
+		panic("number of names in union field must be exactly 1")
+	}
+	fieldName := field.Names[0]
+	pos := fieldName.NamePos
+
+	// The method receiver.
+	receiver := &ast.SelectorExpr{
+		X: &ast.Ident{
+			NamePos: pos,
+			Name:    "union",
+			Obj:     nil,
+		},
+		Sel: &ast.Ident{
+			NamePos: pos,
+			Name:    "$union",
+		},
+	}
+
+	// Get the address of the $union field.
+	receiverPtr := &ast.UnaryExpr{
+		Op: token.AND,
+		X:  receiver,
+	}
+
+	// Cast to unsafe.Pointer.
+	sourcePointer := &ast.CallExpr{
+		Fun: &ast.SelectorExpr{
+			X:   &ast.Ident{Name: "unsafe"},
+			Sel: &ast.Ident{Name: "Pointer"},
+		},
+		Args: []ast.Expr{receiverPtr},
+	}
+
+	// Cast to the target pointer type.
+	targetPointer := &ast.CallExpr{
+		Lparen: pos,
+		Fun: &ast.ParenExpr{
+			Lparen: pos,
+			X: &ast.StarExpr{
+				X: field.Type,
+			},
+			Rparen: pos,
+		},
+		Args:   []ast.Expr{sourcePointer},
+		Rparen: pos,
+	}
+
+	// Create the accessor function.
+	accessor := &ast.FuncDecl{
+		Recv: &ast.FieldList{
+			Opening: pos,
+			List: []*ast.Field{
+				&ast.Field{
+					Names: []*ast.Ident{
+						&ast.Ident{
+							NamePos: pos,
+							Name:    "union",
+						},
+					},
+					Type: &ast.StarExpr{
+						Star: pos,
+						X: &ast.Ident{
+							NamePos: pos,
+							Name:    typeName,
+							Obj:     nil,
+						},
+					},
+				},
+			},
+			Closing: pos,
+		},
+		Name: &ast.Ident{
+			NamePos: pos,
+			Name:    "unionfield_" + fieldName.Name,
+		},
+		Type: &ast.FuncType{
+			Func: pos,
+			Params: &ast.FieldList{
+				Opening: pos,
+				Closing: pos,
+			},
+			Results: &ast.FieldList{
+				List: []*ast.Field{
+					&ast.Field{
+						Type: &ast.StarExpr{
+							Star: pos,
+							X:    field.Type,
+						},
+					},
+				},
+			},
+		},
+		Body: &ast.BlockStmt{
+			Lbrace: pos,
+			List: []ast.Stmt{
+				&ast.ReturnStmt{
+					Return: pos,
+					Results: []ast.Expr{
+						targetPointer,
+					},
+				},
+			},
+			Rbrace: pos,
+		},
+	}
+	p.generated.Decls = append(p.generated.Decls, accessor)
 }
 
 // createBitfieldGetter creates a bitfield getter function like the following:
@@ -905,16 +1211,16 @@ func (p *cgoPackage) addEnumTypes() {
 	if len(p.enums) == 0 {
 		return
 	}
-	gen := &ast.GenDecl{
-		TokPos: token.NoPos,
-		Tok:    token.TYPE,
-	}
 	names := make([]string, 0, len(p.enums))
 	for name := range p.enums {
 		names = append(names, name)
 	}
 	sort.Strings(names)
 	for _, name := range names {
+		gen := &ast.GenDecl{
+			TokPos: token.NoPos,
+			Tok:    token.TYPE,
+		}
 		typ := p.enums[name]
 		typeName := "C.enum_" + name
 		obj := &ast.Object{
@@ -931,8 +1237,8 @@ func (p *cgoPackage) addEnumTypes() {
 		}
 		obj.Decl = typeSpec
 		gen.Specs = append(gen.Specs, typeSpec)
+		p.generated.Decls = append(p.generated.Decls, gen)
 	}
-	p.generated.Decls = append(p.generated.Decls, gen)
 }
 
 // findMissingCGoNames traverses the AST and finds all C.something names. Only
@@ -995,3 +1301,29 @@ func (p *cgoPackage) walker(cursor *astutil.Cursor) bool {
 	}
 	return true
 }
+
+// renameFieldKeywords renames all reserved words in Go to some other field name
+// with a "_" prefix. For example, it renames `type` to `_type`.
+//
+// See: https://golang.org/cmd/cgo/#hdr-Go_references_to_C
+func renameFieldKeywords(fieldList *ast.FieldList) {
+	renameFieldName(fieldList, "type")
+}
+
+// renameFieldName renames a given field name to a name with a "_" prepended. It
+// makes sure to do the same thing for any field sharing the same name.
+func renameFieldName(fieldList *ast.FieldList, name string) {
+	var ident *ast.Ident
+	for _, f := range fieldList.List {
+		for _, n := range f.Names {
+			if n.Name == name {
+				ident = n
+			}
+		}
+	}
+	if ident == nil {
+		return
+	}
+	renameFieldName(fieldList, "_"+name)
+	ident.Name = "_" + ident.Name
+}

cgo/cgo_test.go

@@ -0,0 +1,129 @@
+package cgo
+
+import (
+	"bytes"
+	"flag"
+	"fmt"
+	"go/ast"
+	"go/format"
+	"go/parser"
+	"go/token"
+	"go/types"
+	"io/ioutil"
+	"path/filepath"
+	"runtime"
+	"strings"
+	"testing"
+)
+
+// Pass -update to go test to update the output of the test files.
+var flagUpdate = flag.Bool("update", false, "Update images based on test output.")
+
+func TestCGo(t *testing.T) {
+	var cflags = []string{"--target=armv6m-none-eabi"}
+
+	for _, name := range []string{"basic", "errors", "types", "flags", "const"} {
+		name := name // avoid a race condition
+		t.Run(name, func(t *testing.T) {
+			// Read the AST in memory.
+			path := filepath.Join("testdata", name+".go")
+			fset := token.NewFileSet()
+			f, err := parser.ParseFile(fset, path, nil, parser.ParseComments)
+			if err != nil {
+				t.Fatal("could not parse Go source file:", err)
+			}
+
+			// Process the AST with CGo.
+			cgoAST, cgoErrors := Process([]*ast.File{f}, "testdata", fset, cflags)
+
+			// Check the AST for type errors.
+			var typecheckErrors []error
+			config := types.Config{
+				Error: func(err error) {
+					typecheckErrors = append(typecheckErrors, err)
+				},
+				Importer: simpleImporter{},
+				Sizes:    types.SizesFor("gccgo", "arm"),
+			}
+			_, err = config.Check("", fset, []*ast.File{f, cgoAST}, nil)
+			if err != nil && len(typecheckErrors) == 0 {
+				// Only report errors when no type errors are found (an
+				// unexpected condition).
+				t.Error(err)
+			}
+
+			// Store the (formatted) output in a buffer. Format it, so it
+			// becomes easier to read (and will hopefully change less with CGo
+			// changes).
+			buf := &bytes.Buffer{}
+			if len(cgoErrors) != 0 {
+				buf.WriteString("// CGo errors:\n")
+				for _, err := range cgoErrors {
+					buf.WriteString(formatDiagnostic(err))
+				}
+				buf.WriteString("\n")
+			}
+			if len(typecheckErrors) != 0 {
+				buf.WriteString("// Type checking errors after CGo processing:\n")
+				for _, err := range typecheckErrors {
+					buf.WriteString(formatDiagnostic(err))
+				}
+				buf.WriteString("\n")
+			}
+			err = format.Node(buf, fset, cgoAST)
+			if err != nil {
+				t.Errorf("could not write out CGo AST: %v", err)
+			}
+			actual := strings.Replace(string(buf.Bytes()), "\r\n", "\n", -1)
+
+			// Read the file with the expected output, to compare against.
+			outfile := filepath.Join("testdata", name+".out.go")
+			expectedBytes, err := ioutil.ReadFile(outfile)
+			if err != nil {
+				t.Fatalf("could not read expected output: %v", err)
+			}
+			expected := strings.Replace(string(expectedBytes), "\r\n", "\n", -1)
+
+			// Check whether the output is as expected.
+			if expected != actual {
+				// It is not. Test failed.
+				if *flagUpdate {
+					// Update the file with the expected data.
+					err := ioutil.WriteFile(outfile, []byte(actual), 0666)
+					if err != nil {
+						t.Error("could not write updated output file:", err)
+					}
+					return
+				}
+				t.Errorf("output did not match:\n%s", string(actual))
+			}
+		})
+	}
+}
+
+// simpleImporter implements the types.Importer interface, but only allows
+// importing the unsafe package.
+type simpleImporter struct {
+}
+
+// Import implements the Importer interface. For testing usage only: it only
+// supports importing the unsafe package.
+func (i simpleImporter) Import(path string) (*types.Package, error) {
+	switch path {
+	case "unsafe":
+		return types.Unsafe, nil
+	default:
+		return nil, fmt.Errorf("importer not implemented for package %s", path)
+	}
+}
+
+// formatDiagnostics formats the error message to be an indented comment. It
+// also fixes Windows path name issues (backward slashes).
+func formatDiagnostic(err error) string {
+	msg := err.Error()
+	if runtime.GOOS == "windows" {
+		// Fix Windows path slashes.
+		msg = strings.Replace(msg, "testdata\\", "testdata/", -1)
+	}
+	return "//     " + msg + "\n"
+}

cgo/const.go

@@ -0,0 +1,218 @@
+package cgo
+
+// This file implements a parser of a subset of the C language, just enough to
+// parse common #define statements to Go constant expressions.
+
+import (
+	"fmt"
+	"go/ast"
+	"go/scanner"
+	"go/token"
+	"strings"
+)
+
+// parseConst parses the given string as a C constant.
+func parseConst(pos token.Pos, fset *token.FileSet, value string) (ast.Expr, *scanner.Error) {
+	t := newTokenizer(pos, fset, value)
+	expr, err := parseConstExpr(t)
+	if t.token != token.EOF {
+		return nil, &scanner.Error{
+			Pos: t.fset.Position(t.pos),
+			Msg: "unexpected token " + t.token.String(),
+		}
+	}
+	return expr, err
+}
+
+// parseConstExpr parses a stream of C tokens to a Go expression.
+func parseConstExpr(t *tokenizer) (ast.Expr, *scanner.Error) {
+	switch t.token {
+	case token.LPAREN:
+		lparen := t.pos
+		t.Next()
+		x, err := parseConstExpr(t)
+		if err != nil {
+			return nil, err
+		}
+		if t.token != token.RPAREN {
+			return nil, unexpectedToken(t, token.RPAREN)
+		}
+		expr := &ast.ParenExpr{
+			Lparen: lparen,
+			X:      x,
+			Rparen: t.pos,
+		}
+		t.Next()
+		return expr, nil
+	case token.INT, token.FLOAT, token.STRING, token.CHAR:
+		expr := &ast.BasicLit{
+			ValuePos: t.pos,
+			Kind:     t.token,
+			Value:    t.value,
+		}
+		t.Next()
+		return expr, nil
+	case token.IDENT:
+		expr := &ast.Ident{
+			NamePos: t.pos,
+			Name:    "C." + t.value,
+		}
+		t.Next()
+		return expr, nil
+	case token.EOF:
+		return nil, &scanner.Error{
+			Pos: t.fset.Position(t.pos),
+			Msg: "empty constant",
+		}
+	default:
+		return nil, &scanner.Error{
+			Pos: t.fset.Position(t.pos),
+			Msg: fmt.Sprintf("unexpected token %s", t.token),
+		}
+	}
+}
+
+// unexpectedToken returns an error of the form "unexpected token FOO, expected
+// BAR".
+func unexpectedToken(t *tokenizer, expected token.Token) *scanner.Error {
+	return &scanner.Error{
+		Pos: t.fset.Position(t.pos),
+		Msg: fmt.Sprintf("unexpected token %s, expected %s", t.token, expected),
+	}
+}
+
+// tokenizer reads C source code and converts it to Go tokens.
+type tokenizer struct {
+	pos   token.Pos
+	fset  *token.FileSet
+	token token.Token
+	value string
+	buf   string
+}
+
+// newTokenizer initializes a new tokenizer, positioned at the first token in
+// the string.
+func newTokenizer(start token.Pos, fset *token.FileSet, buf string) *tokenizer {
+	t := &tokenizer{
+		pos:   start,
+		fset:  fset,
+		buf:   buf,
+		token: token.ILLEGAL,
+	}
+	t.Next() // Parse the first token.
+	return t
+}
+
+// Next consumes the next token in the stream. There is no return value, read
+// the next token from the pos, token and value properties.
+func (t *tokenizer) Next() {
+	t.pos += token.Pos(len(t.value))
+	for {
+		if len(t.buf) == 0 {
+			t.token = token.EOF
+			return
+		}
+		c := t.buf[0]
+		switch {
+		case c == ' ' || c == '\f' || c == '\n' || c == '\r' || c == '\t' || c == '\v':
+			// Skip whitespace.
+			// Based on this source, not sure whether it represents C whitespace:
+			// https://en.cppreference.com/w/cpp/string/byte/isspace
+			t.pos++
+			t.buf = t.buf[1:]
+		case c == '(' || c == ')':
+			// Single-character tokens.
+			switch c {
+			case '(':
+				t.token = token.LPAREN
+			case ')':
+				t.token = token.RPAREN
+			}
+			t.value = t.buf[:1]
+			t.buf = t.buf[1:]
+			return
+		case c >= '0' && c <= '9':
+			// Numeric constant (int, float, etc.).
+			// Find the last non-numeric character.
+			tokenLen := len(t.buf)
+			hasDot := false
+			for i, c := range t.buf {
+				if c == '.' {
+					hasDot = true
+				}
+				if c >= '0' && c <= '9' || c == '.' || c == '_' || c >= 'a' && c <= 'z' || c >= 'A' && c <= 'Z' {
+					tokenLen = i + 1
+				} else {
+					break
+				}
+			}
+			t.value = t.buf[:tokenLen]
+			t.buf = t.buf[tokenLen:]
+			if hasDot {
+				// Integer constants are more complicated than this but this is
+				// a close approximation.
+				// https://en.cppreference.com/w/cpp/language/integer_literal
+				t.token = token.FLOAT
+				t.value = strings.TrimRight(t.value, "f")
+			} else {
+				t.token = token.INT
+				t.value = strings.TrimRight(t.value, "uUlL")
+			}
+			return
+		case c >= 'A' && c <= 'Z' || c >= 'a' && c <= 'z' || c == '_':
+			// Identifier. Find all remaining tokens that are part of this
+			// identifier.
+			tokenLen := len(t.buf)
+			for i, c := range t.buf {
+				if c >= '0' && c <= '9' || c >= 'A' && c <= 'Z' || c >= 'a' && c <= 'z' || c == '_' {
+					tokenLen = i + 1
+				} else {
+					break
+				}
+			}
+			t.value = t.buf[:tokenLen]
+			t.buf = t.buf[tokenLen:]
+			t.token = token.IDENT
+			return
+		case c == '"':
+			// String constant. Find the first '"' character that is not
+			// preceded by a backslash.
+			escape := false
+			tokenLen := len(t.buf)
+			for i, c := range t.buf {
+				if i != 0 && c == '"' && !escape {
+					tokenLen = i + 1
+					break
+				}
+				if !escape {
+					escape = c == '\\'
+				}
+			}
+			t.token = token.STRING
+			t.value = t.buf[:tokenLen]
+			t.buf = t.buf[tokenLen:]
+			return
+		case c == '\'':
+			// Char (rune) constant. Find the first '\'' character that is not
+			// preceded by a backslash.
+			escape := false
+			tokenLen := len(t.buf)
+			for i, c := range t.buf {
+				if i != 0 && c == '\'' && !escape {
+					tokenLen = i + 1
+					break
+				}
+				if !escape {
+					escape = c == '\\'
+				}
+			}
+			t.token = token.CHAR
+			t.value = t.buf[:tokenLen]
+			t.buf = t.buf[tokenLen:]
+			return
+		default:
+			t.token = token.ILLEGAL
+			return
+		}
+	}
+}

cgo/const_test.go

@@ -0,0 +1,60 @@
+package cgo
+
+import (
+	"bytes"
+	"go/format"
+	"go/token"
+	"strings"
+	"testing"
+)
+
+func TestParseConst(t *testing.T) {
+	// Test converting a C constant to a Go constant.
+	for _, tc := range []struct {
+		C  string
+		Go string
+	}{
+		{`5`, `5`},
+		{`(5)`, `(5)`},
+		{`(((5)))`, `(5)`},
+		{`)`, `error: 1:1: unexpected token )`},
+		{`5)`, `error: 1:2: unexpected token )`},
+		{"  \t)", `error: 1:4: unexpected token )`},
+		{`5.8f`, `5.8`},
+		{`foo`, `C.foo`},
+		{``, `error: 1:1: empty constant`}, // empty constants not allowed in Go
+		{`"foo"`, `"foo"`},
+		{`"a\\n"`, `"a\\n"`},
+		{`"a\n"`, `"a\n"`},
+		{`"a\""`, `"a\""`},
+		{`'a'`, `'a'`},
+		{`0b10`, `0b10`},
+		{`0x1234_5678`, `0x1234_5678`},
+		{`5 5`, `error: 1:3: unexpected token INT`}, // test for a bugfix
+	} {
+		fset := token.NewFileSet()
+		startPos := fset.AddFile("", -1, 1000).Pos(0)
+		expr, err := parseConst(startPos, fset, tc.C)
+		s := "<invalid>"
+		if err != nil {
+			if !strings.HasPrefix(tc.Go, "error: ") {
+				t.Errorf("expected value %#v for C constant %#v but got error %#v", tc.Go, tc.C, err.Error())
+				continue
+			}
+			s = "error: " + err.Error()
+		} else if expr != nil {
+			// Serialize the Go constant to a string, for more readable test
+			// cases.
+			buf := &bytes.Buffer{}
+			err := format.Node(buf, fset, expr)
+			if err != nil {
+				t.Errorf("could not format expr from C constant %#v: %v", tc.C, err)
+				continue
+			}
+			s = buf.String()
+		}
+		if s != tc.Go {
+			t.Errorf("C constant %#v was parsed to %#v while expecting %#v", tc.C, s, tc.Go)
+		}
+	}
+}

cgo/libclang.go

@@ -15,7 +15,7 @@ import (
 )
 
 /*
-#include <clang-c/Index.h> // if this fails, install libclang-8-dev
+#include <clang-c/Index.h> // if this fails, install libclang-9-dev
 #include <stdlib.h>
 #include <stdint.h>
 
@@ -109,7 +109,8 @@ func (p *cgoPackage) parseFragment(fragment string, cflags []string, posFilename
 		C.CXTranslationUnit_DetailedPreprocessingRecord,
 		&unit)
 	if errCode != 0 {
-		panic("loader: failed to parse source with libclang")
+		// This is probably a bug in the usage of libclang.
+		panic("cgo: failed to parse source with libclang")
 	}
 	defer C.clang_disposeTranslationUnit(unit)
 
@@ -118,27 +119,8 @@ func (p *cgoPackage) parseFragment(fragment string, cflags []string, posFilename
 			spelling := getString(C.clang_getDiagnosticSpelling(diagnostic))
 			severity := diagnosticSeverity[C.clang_getDiagnosticSeverity(diagnostic)]
 			location := C.clang_getDiagnosticLocation(diagnostic)
-			var libclangFilename C.CXString
-			var line C.unsigned
-			var column C.unsigned
-			C.clang_getPresumedLocation(location, &libclangFilename, &line, &column)
-			filename := getString(libclangFilename)
-			if filepath.IsAbs(filename) {
-				// Relative paths for readability, like other Go parser errors.
-				relpath, err := filepath.Rel(p.dir, filename)
-				if err == nil {
-					filename = relpath
-				}
-			}
-			p.errors = append(p.errors, &scanner.Error{
-				Pos: token.Position{
-					Filename: filename,
-					Offset:   0, // not provided by clang_getPresumedLocation
-					Line:     int(line),
-					Column:   int(column),
-				},
-				Msg: severity + ": " + spelling,
-			})
+			pos := p.getClangLocationPosition(location, unit)
+			p.addError(pos, severity+": "+spelling)
 		}
 		for i := 0; i < numDiagnostics; i++ {
 			diagnostic := C.clang_getDiagnostic(unit, C.uint(i))
@@ -150,7 +132,6 @@ func (p *cgoPackage) parseFragment(fragment string, cflags []string, posFilename
 				addDiagnostic(C.clang_getDiagnosticInSet(diagnostics, C.uint(j)))
 			}
 		}
-		return
 	}
 
 	ref := storedRefs.Put(p)
@@ -237,14 +218,17 @@ func tinygo_clang_globals_visitor(c, parent C.GoCXCursor, client_data C.CXClient
 		var startOffset, endOffset C.unsigned
 		C.clang_getExpansionLocation(start, &file, nil, nil, &startOffset)
 		if file == nil {
-			panic("could not find file where macro is defined")
+			p.addError(pos, "internal error: could not find file where macro is defined")
+			break
 		}
 		C.clang_getExpansionLocation(end, &endFile, nil, nil, &endOffset)
 		if file != endFile {
-			panic("expected start and end location of a #define to be in the same file")
+			p.addError(pos, "internal error: expected start and end location of a macro to be in the same file")
+			break
 		}
 		if startOffset > endOffset {
-			panic("startOffset > endOffset")
+			p.addError(pos, "internal error: start offset of macro is after end offset")
+			break
 		}
 
 		// read file contents and extract the relevant byte range
@@ -252,59 +236,29 @@ func tinygo_clang_globals_visitor(c, parent C.GoCXCursor, client_data C.CXClient
 		var size C.size_t
 		sourcePtr := C.clang_getFileContents(tu, file, &size)
 		if endOffset >= C.uint(size) {
-			panic("endOffset lies after end of file")
+			p.addError(pos, "internal error: end offset of macro lies after end of file")
+			break
 		}
 		source := string(((*[1 << 28]byte)(unsafe.Pointer(sourcePtr)))[startOffset:endOffset:endOffset])
 		if !strings.HasPrefix(source, name) {
-			panic(fmt.Sprintf("expected #define value to start with %#v, got %#v", name, source))
+			p.addError(pos, fmt.Sprintf("internal error: expected macro value to start with %#v, got %#v", name, source))
+			break
 		}
-		value := strings.TrimSpace(source[len(name):])
-		for len(value) != 0 && value[0] == '(' && value[len(value)-1] == ')' {
-			value = strings.TrimSpace(value[1 : len(value)-1])
+		value := source[len(name):]
+		// Try to convert this #define into a Go constant expression.
+		expr, err := parseConst(pos+token.Pos(len(name)), p.fset, value)
+		if err != nil {
+			p.errors = append(p.errors, err)
 		}
-		if len(value) == 0 {
-			// Pretend it doesn't exist at all.
-			return C.CXChildVisit_Continue
-		}
-		// For information about integer literals:
-		// https://en.cppreference.com/w/cpp/language/integer_literal
-		if value[0] == '"' {
-			// string constant
-			p.constants[name] = constantInfo{&ast.BasicLit{pos, token.STRING, value}, pos}
-			return C.CXChildVisit_Continue
-		}
-		if value[0] == '\'' {
-			// char constant
-			p.constants[name] = constantInfo{&ast.BasicLit{pos, token.CHAR, value}, pos}
-			return C.CXChildVisit_Continue
-		}
-		// assume it's a number (int or float)
-		value = strings.Replace(value, "'", "", -1) // remove ' chars
-		value = strings.TrimRight(value, "lu")      // remove llu suffixes etc.
-		// find the first non-number
-		nonnum := byte(0)
-		for i := 0; i < len(value); i++ {
-			if value[i] < '0' || value[i] > '9' {
-				nonnum = value[i]
-				break
-			}
-		}
-		// determine number type based on the first non-number
-		switch nonnum {
-		case 0:
-			// no non-number found, must be an integer
-			p.constants[name] = constantInfo{&ast.BasicLit{pos, token.INT, value}, pos}
-		case 'x', 'X':
-			// hex integer constant
-			// TODO: may also be a floating point number per C++17.
-			p.constants[name] = constantInfo{&ast.BasicLit{pos, token.INT, value}, pos}
-		case '.', 'e':
-			// float constant
-			value = strings.TrimRight(value, "fFlL")
-			p.constants[name] = constantInfo{&ast.BasicLit{pos, token.FLOAT, value}, pos}
-		default:
-			// unknown type, ignore
+		if expr != nil {
+			// Parsing was successful.
+			p.constants[name] = constantInfo{expr, pos}
 		}
+	case C.CXCursor_EnumDecl:
+		// Visit all enums, because the fields may be used even when the enum
+		// type itself is not.
+		typ := C.tinygo_clang_getCursorType(c)
+		p.makeASTType(typ, pos)
 	}
 	return C.CXChildVisit_Continue
 }
@@ -316,11 +270,16 @@ func getString(clangString C.CXString) (s string) {
 	return
 }
 
-// getCursorPosition returns a usable token.Pos from a libclang cursor. If the
-// file for this cursor has not been seen before, it is read from libclang
-// (which already has the file in memory) and added to the token.FileSet.
+// getCursorPosition returns a usable token.Pos from a libclang cursor.
 func (p *cgoPackage) getCursorPosition(cursor C.GoCXCursor) token.Pos {
-	location := C.tinygo_clang_getCursorLocation(cursor)
+	return p.getClangLocationPosition(C.tinygo_clang_getCursorLocation(cursor), C.tinygo_clang_Cursor_getTranslationUnit(cursor))
+}
+
+// getClangLocationPosition returns a usable token.Pos based on a libclang
+// location and translation unit. If the file for this cursor has not been seen
+// before, it is read from libclang (which already has the file in memory) and
+// added to the token.FileSet.
+func (p *cgoPackage) getClangLocationPosition(location C.CXSourceLocation, tu C.CXTranslationUnit) token.Pos {
 	var file C.CXFile
 	var line C.unsigned
 	var column C.unsigned
@@ -334,7 +293,6 @@ func (p *cgoPackage) getCursorPosition(cursor C.GoCXCursor) token.Pos {
 	if _, ok := p.tokenFiles[filename]; !ok {
 		// File has not been seen before in this package, add line information
 		// now by reading the file from libclang.
-		tu := C.tinygo_clang_Cursor_getTranslationUnit(cursor)
 		var size C.size_t
 		sourcePtr := C.clang_getFileContents(tu, file, &size)
 		source := ((*[1 << 28]byte)(unsafe.Pointer(sourcePtr)))[:size:size]
@@ -348,7 +306,62 @@ func (p *cgoPackage) getCursorPosition(cursor C.GoCXCursor) token.Pos {
 		f.SetLines(lines)
 		p.tokenFiles[filename] = f
 	}
-	return p.tokenFiles[filename].Pos(int(offset))
+	positionFile := p.tokenFiles[filename]
+
+	// Check for alternative line/column information (set with a line directive).
+	var filename2String C.CXString
+	var line2 C.unsigned
+	var column2 C.unsigned
+	C.clang_getPresumedLocation(location, &filename2String, &line2, &column2)
+	filename2 := getString(filename2String)
+	if filename2 != filename || line2 != line || column2 != column {
+		// The location was changed with a preprocessor directive.
+		// TODO: this only works for locations that are added in order. Adding
+		// line/column info to a file that already has line/column info after
+		// the given offset is ignored.
+		positionFile.AddLineColumnInfo(int(offset), filename2, int(line2), int(column2))
+	}
+
+	return positionFile.Pos(int(offset))
+}
+
+// addError is a utility function to add an error to the list of errors. It will
+// convert the token position to a line/column position first, and call
+// addErrorAt.
+func (p *cgoPackage) addError(pos token.Pos, msg string) {
+	p.addErrorAt(p.fset.PositionFor(pos, true), msg)
+}
+
+// addErrorAfter is like addError, but adds the text `after` to the source
+// location.
+func (p *cgoPackage) addErrorAfter(pos token.Pos, after, msg string) {
+	position := p.fset.PositionFor(pos, true)
+	lines := strings.Split(after, "\n")
+	if len(lines) != 1 {
+		// Adjust lines.
+		// For why we can't just do pos+token.Pos(len(after)), see:
+		// https://github.com/golang/go/issues/35803
+		position.Line += len(lines) - 1
+		position.Column = len(lines[len(lines)-1]) + 1
+	} else {
+		position.Column += len(after)
+	}
+	p.addErrorAt(position, msg)
+}
+
+// addErrorAt is a utility function to add an error to the list of errors.
+func (p *cgoPackage) addErrorAt(position token.Position, msg string) {
+	if filepath.IsAbs(position.Filename) {
+		// Relative paths for readability, like other Go parser errors.
+		relpath, err := filepath.Rel(p.dir, position.Filename)
+		if err == nil {
+			position.Filename = relpath
+		}
+	}
+	p.errors = append(p.errors, scanner.Error{
+		Pos: position,
+		Msg: msg,
+	})
 }
 
 // makeASTType return the ast.Expr for the given libclang type. In other words,
@@ -459,7 +472,7 @@ func (p *cgoPackage) makeASTType(typ C.CXType, pos token.Pos) ast.Expr {
 						// This happens for some very special purpose architectures
 						// (DSPs etc.) that are not currently targeted.
 						// https://www.embecosm.com/2017/04/18/non-8-bit-char-support-in-clang-and-llvm/
-						panic("unknown char width")
+						p.addError(pos, fmt.Sprintf("unknown char width: %d", typeSize))
 					}
 					switch underlyingType.kind {
 					case C.CXType_Char_S:
@@ -508,48 +521,44 @@ func (p *cgoPackage) makeASTType(typ C.CXType, pos token.Pos) ast.Expr {
 		case C.CXType_Enum:
 			return p.makeASTType(underlying, pos)
 		default:
-			panic("unknown elaborated type")
+			typeKindSpelling := getString(C.clang_getTypeKindSpelling(underlying.kind))
+			p.addError(pos, fmt.Sprintf("unknown elaborated type (libclang type kind %s)", typeKindSpelling))
+			typeName = "<unknown>"
 		}
 	case C.CXType_Record:
 		cursor := C.tinygo_clang_getTypeDeclaration(typ)
 		name := getString(C.tinygo_clang_getCursorSpelling(cursor))
+		var cgoRecordPrefix string
+		switch C.tinygo_clang_getCursorKind(cursor) {
+		case C.CXCursor_StructDecl:
+			cgoRecordPrefix = "struct_"
+		case C.CXCursor_UnionDecl:
+			cgoRecordPrefix = "union_"
+		default:
+			// makeASTRecordType will create an appropriate error.
+			cgoRecordPrefix = "record_"
+		}
 		if name == "" {
 			// Anonymous record, probably inside a typedef.
-			typeExpr, bitfieldList := p.makeASTRecordType(cursor, pos)
-			if bitfieldList != nil {
-				// This struct has bitfields, so we have to declare it as a
-				// named type (for bitfield getters/setters to work).
+			typeInfo := p.makeASTRecordType(cursor, pos)
+			if typeInfo.bitfields != nil || typeInfo.unionSize != 0 {
+				// This record is a union or is a struct with bitfields, so we
+				// have to declare it as a named type (for getters/setters to
+				// work).
 				p.anonStructNum++
-				cgoName := "struct_" + strconv.Itoa(p.anonStructNum)
-				p.elaboratedTypes[cgoName] = &elaboratedTypeInfo{
-					typeExpr:  typeExpr,
-					pos:       pos,
-					bitfields: bitfieldList,
-				}
+				cgoName := cgoRecordPrefix + strconv.Itoa(p.anonStructNum)
+				p.elaboratedTypes[cgoName] = typeInfo
 				return &ast.Ident{
 					NamePos: pos,
 					Name:    "C." + cgoName,
 				}
 			}
-			return typeExpr
+			return typeInfo.typeExpr
 		} else {
-			var cgoName string
-			switch C.tinygo_clang_getCursorKind(cursor) {
-			case C.CXCursor_StructDecl:
-				cgoName = "struct_" + name
-			case C.CXCursor_UnionDecl:
-				cgoName = "union_" + name
-			default:
-				panic("unknown record declaration")
-			}
+			cgoName := cgoRecordPrefix + name
 			if _, ok := p.elaboratedTypes[cgoName]; !ok {
 				p.elaboratedTypes[cgoName] = nil // predeclare (to avoid endless recursion)
-				typeExpr, bitfieldList := p.makeASTRecordType(cursor, pos)
-				p.elaboratedTypes[cgoName] = &elaboratedTypeInfo{
-					typeExpr:  typeExpr,
-					pos:       pos,
-					bitfields: bitfieldList,
-				}
+				p.elaboratedTypes[cgoName] = p.makeASTRecordType(cursor, pos)
 			}
 			return &ast.Ident{
 				NamePos: pos,
@@ -585,14 +594,10 @@ func (p *cgoPackage) makeASTType(typ C.CXType, pos token.Pos) ast.Expr {
 	}
 	if typeName == "" {
 		// Report this as an error.
-		spelling := getString(C.clang_getTypeSpelling(typ))
-		p.errors = append(p.errors, scanner.Error{
-			Pos: p.fset.PositionFor(pos, true),
-			Msg: fmt.Sprintf("unknown C type: %v (libclang type kind %d)", spelling, typ.kind),
-		})
-		// Fallback, probably incorrect but at least the error points to an odd
-		// type name.
-		typeName = "C." + spelling
+		typeSpelling := getString(C.clang_getTypeSpelling(typ))
+		typeKindSpelling := getString(C.clang_getTypeKindSpelling(typ.kind))
+		p.addError(pos, fmt.Sprintf("unknown C type: %v (libclang type kind %s)", typeSpelling, typeKindSpelling))
+		typeName = "C.<unknown>"
 	}
 	return &ast.Ident{
 		NamePos: pos,
@@ -601,9 +606,8 @@ func (p *cgoPackage) makeASTType(typ C.CXType, pos token.Pos) ast.Expr {
 }
 
 // makeASTRecordType parses a C record (struct or union) and translates it into
-// a Go struct type. Unions are implemented by setting the first field to a
-// zero-lengt "C union" field, which cannot be written in Go directly.
-func (p *cgoPackage) makeASTRecordType(cursor C.GoCXCursor, pos token.Pos) (*ast.StructType, []bitfieldInfo) {
+// a Go struct type.
+func (p *cgoPackage) makeASTRecordType(cursor C.GoCXCursor, pos token.Pos) *elaboratedTypeInfo {
 	fieldList := &ast.FieldList{
 		Opening: pos,
 		Closing: pos,
@@ -620,53 +624,53 @@ func (p *cgoPackage) makeASTRecordType(cursor C.GoCXCursor, pos token.Pos) (*ast
 	}{fieldList, p, &inBitfield, &bitfieldNum, &bitfieldList})
 	defer storedRefs.Remove(ref)
 	C.tinygo_clang_visitChildren(cursor, C.CXCursorVisitor(C.tinygo_clang_struct_visitor), C.CXClientData(ref))
+	renameFieldKeywords(fieldList)
 	switch C.tinygo_clang_getCursorKind(cursor) {
 	case C.CXCursor_StructDecl:
-		return &ast.StructType{
-			Struct: pos,
-			Fields: fieldList,
-		}, bitfieldList
+		return &elaboratedTypeInfo{
+			typeExpr: &ast.StructType{
+				Struct: pos,
+				Fields: fieldList,
+			},
+			pos:       pos,
+			bitfields: bitfieldList,
+		}
 	case C.CXCursor_UnionDecl:
+		typeInfo := &elaboratedTypeInfo{
+			typeExpr: &ast.StructType{
+				Struct: pos,
+				Fields: fieldList,
+			},
+			pos:       pos,
+			bitfields: bitfieldList,
+		}
+		if len(fieldList.List) <= 1 {
+			// Useless union, treat it as a regular struct.
+			return typeInfo
+		}
 		if bitfieldList != nil {
 			// This is valid C... but please don't do this.
-			p.errors = append(p.errors, scanner.Error{
-				Pos: p.fset.PositionFor(pos, true),
-				Msg: fmt.Sprintf("bitfield in a union is not supported"),
-			})
+			p.addError(pos, "bitfield in a union is not supported")
 		}
-		if len(fieldList.List) > 1 {
-			// Insert a special field at the front (of zero width) as a
-			// marker that this is struct is actually a union. This is done
-			// by giving the field a name that cannot be expressed directly
-			// in Go.
-			// Other parts of the compiler look at the first element in a
-			// struct (of size > 2) to know whether this is a union.
-			// Note that we don't have to insert it for single-element
-			// unions as they're basically equivalent to a struct.
-			unionMarker := &ast.Field{
-				Type: &ast.StructType{
-					Struct: pos,
-				},
-			}
-			unionMarker.Names = []*ast.Ident{
-				&ast.Ident{
-					NamePos: pos,
-					Name:    "C union",
-					Obj: &ast.Object{
-						Kind: ast.Var,
-						Name: "C union",
-						Decl: unionMarker,
-					},
-				},
-			}
-			fieldList.List = append([]*ast.Field{unionMarker}, fieldList.List...)
+		typ := C.tinygo_clang_getCursorType(cursor)
+		alignInBytes := int64(C.clang_Type_getAlignOf(typ))
+		sizeInBytes := int64(C.clang_Type_getSizeOf(typ))
+		if sizeInBytes == 0 {
+			p.addError(pos, "zero-length union is not supported")
 		}
-		return &ast.StructType{
-			Struct: pos,
-			Fields: fieldList,
-		}, bitfieldList
+		typeInfo.unionSize = sizeInBytes
+		typeInfo.unionAlign = alignInBytes
+		return typeInfo
 	default:
-		panic("unknown record declaration")
+		cursorKind := C.tinygo_clang_getCursorKind(cursor)
+		cursorKindSpelling := getString(C.clang_getCursorKindSpelling(cursorKind))
+		p.addError(pos, fmt.Sprintf("expected StructDecl or UnionDecl, not %s", cursorKindSpelling))
+		return &elaboratedTypeInfo{
+			typeExpr: &ast.StructType{
+				Struct: pos,
+			},
+			pos: pos,
+		}
 	}
 }
 
@@ -684,8 +688,17 @@ func tinygo_clang_struct_visitor(c, parent C.GoCXCursor, client_data C.CXClientD
 	inBitfield := passed.inBitfield
 	bitfieldNum := passed.bitfieldNum
 	bitfieldList := passed.bitfieldList
-	if C.tinygo_clang_getCursorKind(c) != C.CXCursor_FieldDecl {
-		panic("expected field inside cursor")
+	pos := p.getCursorPosition(c)
+	switch cursorKind := C.tinygo_clang_getCursorKind(c); cursorKind {
+	case C.CXCursor_FieldDecl:
+		// Expected. This is a regular field.
+	case C.CXCursor_StructDecl, C.CXCursor_UnionDecl:
+		// Ignore. The next field will be the struct/union itself.
+		return C.CXChildVisit_Continue
+	default:
+		cursorKindSpelling := getString(C.clang_getCursorKindSpelling(cursorKind))
+		p.addError(pos, fmt.Sprintf("expected FieldDecl in struct or union, not %s", cursorKindSpelling))
+		return C.CXChildVisit_Continue
 	}
 	name := getString(C.tinygo_clang_getCursorSpelling(c))
 	if name == "" {
@@ -694,7 +707,6 @@ func tinygo_clang_struct_visitor(c, parent C.GoCXCursor, client_data C.CXClientD
 		return C.CXChildVisit_Continue
 	}
 	typ := C.tinygo_clang_getCursorType(c)
-	pos := p.getCursorPosition(c)
 	field := &ast.Field{
 		Type: p.makeASTType(typ, p.getCursorPosition(c)),
 	}
@@ -703,7 +715,8 @@ func tinygo_clang_struct_visitor(c, parent C.GoCXCursor, client_data C.CXClientD
 	bitfieldOffset := offsetof % alignOf
 	if bitfieldOffset != 0 {
 		if C.tinygo_clang_Cursor_isBitField(c) != 1 {
-			panic("expected a bitfield")
+			p.addError(pos, "expected a bitfield")
+			return C.CXChildVisit_Continue
 		}
 		if !*inBitfield {
 			*bitfieldNum++

cgo/libclang_config.go

@@ -3,9 +3,11 @@
 package cgo
 
 /*
-#cgo linux  CFLAGS: -I/usr/lib/llvm-8/include
-#cgo darwin CFLAGS: -I/usr/local/opt/llvm/include
-#cgo linux  LDFLAGS: -L/usr/lib/llvm-8/lib -lclang
-#cgo darwin LDFLAGS: -L/usr/local/opt/llvm/lib -lclang -lffi
+#cgo linux  CFLAGS: -I/usr/lib/llvm-9/include
+#cgo darwin CFLAGS: -I/usr/local/opt/llvm@9/include
+#cgo freebsd CFLAGS: -I/usr/local/llvm90/include
+#cgo linux  LDFLAGS: -L/usr/lib/llvm-9/lib -lclang
+#cgo darwin LDFLAGS: -L/usr/local/opt/llvm@9/lib -lclang -lffi
+#cgo freebsd LDFLAGS: -L/usr/local/llvm90/lib -lclang
 */
 import "C"

cgo/libclang_stubs.c

@@ -3,7 +3,7 @@
 // are slightly different from the ones defined in libclang.go, but they
 // should be ABI compatible.
 
-#include <clang-c/Index.h> // if this fails, install libclang-8-dev
+#include <clang-c/Index.h> // if this fails, install libclang-9-dev
 
 CXCursor tinygo_clang_getTranslationUnitCursor(CXTranslationUnit tu) {
 	return clang_getTranslationUnitCursor(tu);

cgo/security.go

@@ -0,0 +1,301 @@
+// Copyright 2018 The Go Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+// This file has been copied from the Go 1.13 release tree.
+
+// Checking of compiler and linker flags.
+// We must avoid flags like -fplugin=, which can allow
+// arbitrary code execution during the build.
+// Do not make changes here without carefully
+// considering the implications.
+// (That's why the code is isolated in a file named security.go.)
+//
+// Note that -Wl,foo means split foo on commas and pass to
+// the linker, so that -Wl,-foo,bar means pass -foo bar to
+// the linker. Similarly -Wa,foo for the assembler and so on.
+// If any of these are permitted, the wildcard portion must
+// disallow commas.
+//
+// Note also that GNU binutils accept any argument @foo
+// as meaning "read more flags from the file foo", so we must
+// guard against any command-line argument beginning with @,
+// even things like "-I @foo".
+// We use safeArg (which is even more conservative)
+// to reject these.
+//
+// Even worse, gcc -I@foo (one arg) turns into cc1 -I @foo (two args),
+// so although gcc doesn't expand the @foo, cc1 will.
+// So out of paranoia, we reject @ at the beginning of every
+// flag argument that might be split into its own argument.
+
+package cgo
+
+import (
+	"fmt"
+	"os"
+	"regexp"
+	"strings"
+	"unicode/utf8"
+)
+
+var re = regexp.MustCompile
+
+var validCompilerFlags = []*regexp.Regexp{
+	re(`-D([A-Za-z_].*)`),
+	re(`-F([^@\-].*)`),
+	re(`-I([^@\-].*)`),
+	re(`-O`),
+	re(`-O([^@\-].*)`),
+	re(`-W`),
+	re(`-W([^@,]+)`), // -Wall but not -Wa,-foo.
+	re(`-Wa,-mbig-obj`),
+	re(`-Wp,-D([A-Za-z_].*)`),
+	re(`-ansi`),
+	re(`-f(no-)?asynchronous-unwind-tables`),
+	re(`-f(no-)?blocks`),
+	re(`-f(no-)builtin-[a-zA-Z0-9_]*`),
+	re(`-f(no-)?common`),
+	re(`-f(no-)?constant-cfstrings`),
+	re(`-fdiagnostics-show-note-include-stack`),
+	re(`-f(no-)?eliminate-unused-debug-types`),
+	re(`-f(no-)?exceptions`),
+	re(`-f(no-)?fast-math`),
+	re(`-f(no-)?inline-functions`),
+	re(`-finput-charset=([^@\-].*)`),
+	re(`-f(no-)?fat-lto-objects`),
+	re(`-f(no-)?keep-inline-dllexport`),
+	re(`-f(no-)?lto`),
+	re(`-fmacro-backtrace-limit=(.+)`),
+	re(`-fmessage-length=(.+)`),
+	re(`-f(no-)?modules`),
+	re(`-f(no-)?objc-arc`),
+	re(`-f(no-)?objc-nonfragile-abi`),
+	re(`-f(no-)?objc-legacy-dispatch`),
+	re(`-f(no-)?omit-frame-pointer`),
+	re(`-f(no-)?openmp(-simd)?`),
+	re(`-f(no-)?permissive`),
+	re(`-f(no-)?(pic|PIC|pie|PIE)`),
+	re(`-f(no-)?plt`),
+	re(`-f(no-)?rtti`),
+	re(`-f(no-)?split-stack`),
+	re(`-f(no-)?stack-(.+)`),
+	re(`-f(no-)?strict-aliasing`),
+	re(`-f(un)signed-char`),
+	re(`-f(no-)?use-linker-plugin`), // safe if -B is not used; we don't permit -B
+	re(`-f(no-)?visibility-inlines-hidden`),
+	re(`-fsanitize=(.+)`),
+	re(`-ftemplate-depth-(.+)`),
+	re(`-fvisibility=(.+)`),
+	re(`-g([^@\-].*)?`),
+	re(`-m32`),
+	re(`-m64`),
+	re(`-m(abi|arch|cpu|fpu|tune)=([^@\-].*)`),
+	re(`-m(no-)?v?aes`),
+	re(`-marm`),
+	re(`-m(no-)?avx[0-9a-z]*`),
+	re(`-mfloat-abi=([^@\-].*)`),
+	re(`-mfpmath=[0-9a-z,+]*`),
+	re(`-m(no-)?avx[0-9a-z.]*`),
+	re(`-m(no-)?ms-bitfields`),
+	re(`-m(no-)?stack-(.+)`),
+	re(`-mmacosx-(.+)`),
+	re(`-mios-simulator-version-min=(.+)`),
+	re(`-miphoneos-version-min=(.+)`),
+	re(`-mtvos-simulator-version-min=(.+)`),
+	re(`-mtvos-version-min=(.+)`),
+	re(`-mwatchos-simulator-version-min=(.+)`),
+	re(`-mwatchos-version-min=(.+)`),
+	re(`-mnop-fun-dllimport`),
+	re(`-m(no-)?sse[0-9.]*`),
+	re(`-m(no-)?ssse3`),
+	re(`-mthumb(-interwork)?`),
+	re(`-mthreads`),
+	re(`-mwindows`),
+	re(`--param=ssp-buffer-size=[0-9]*`),
+	re(`-pedantic(-errors)?`),
+	re(`-pipe`),
+	re(`-pthread`),
+	re(`-?-std=([^@\-].*)`),
+	re(`-?-stdlib=([^@\-].*)`),
+	re(`--sysroot=([^@\-].*)`),
+	re(`-w`),
+	re(`-x([^@\-].*)`),
+	re(`-v`),
+}
+
+var validCompilerFlagsWithNextArg = []string{
+	"-arch",
+	"-D",
+	"-I",
+	"-framework",
+	"-isysroot",
+	"-isystem",
+	"--sysroot",
+	"-target",
+	"-x",
+}
+
+var validLinkerFlags = []*regexp.Regexp{
+	re(`-F([^@\-].*)`),
+	re(`-l([^@\-].*)`),
+	re(`-L([^@\-].*)`),
+	re(`-O`),
+	re(`-O([^@\-].*)`),
+	re(`-f(no-)?(pic|PIC|pie|PIE)`),
+	re(`-f(no-)?openmp(-simd)?`),
+	re(`-fsanitize=([^@\-].*)`),
+	re(`-flat_namespace`),
+	re(`-g([^@\-].*)?`),
+	re(`-headerpad_max_install_names`),
+	re(`-m(abi|arch|cpu|fpu|tune)=([^@\-].*)`),
+	re(`-mfloat-abi=([^@\-].*)`),
+	re(`-mmacosx-(.+)`),
+	re(`-mios-simulator-version-min=(.+)`),
+	re(`-miphoneos-version-min=(.+)`),
+	re(`-mthreads`),
+	re(`-mwindows`),
+	re(`-(pic|PIC|pie|PIE)`),
+	re(`-pthread`),
+	re(`-rdynamic`),
+	re(`-shared`),
+	re(`-?-static([-a-z0-9+]*)`),
+	re(`-?-stdlib=([^@\-].*)`),
+	re(`-v`),
+
+	// Note that any wildcards in -Wl need to exclude comma,
+	// since -Wl splits its argument at commas and passes
+	// them all to the linker uninterpreted. Allowing comma
+	// in a wildcard would allow tunnelling arbitrary additional
+	// linker arguments through one of these.
+	re(`-Wl,--(no-)?allow-multiple-definition`),
+	re(`-Wl,--(no-)?allow-shlib-undefined`),
+	re(`-Wl,--(no-)?as-needed`),
+	re(`-Wl,-Bdynamic`),
+	re(`-Wl,-berok`),
+	re(`-Wl,-Bstatic`),
+	re(`-WL,-O([^@,\-][^,]*)?`),
+	re(`-Wl,-d[ny]`),
+	re(`-Wl,--disable-new-dtags`),
+	re(`-Wl,-e[=,][a-zA-Z0-9]*`),
+	re(`-Wl,--enable-new-dtags`),
+	re(`-Wl,--end-group`),
+	re(`-Wl,--(no-)?export-dynamic`),
+	re(`-Wl,-framework,[^,@\-][^,]+`),
+	re(`-Wl,-headerpad_max_install_names`),
+	re(`-Wl,--no-undefined`),
+	re(`-Wl,-R([^@\-][^,@]*$)`),
+	re(`-Wl,--just-symbols[=,]([^,@\-][^,@]+)`),
+	re(`-Wl,-rpath(-link)?[=,]([^,@\-][^,]+)`),
+	re(`-Wl,-s`),
+	re(`-Wl,-search_paths_first`),
+	re(`-Wl,-sectcreate,([^,@\-][^,]+),([^,@\-][^,]+),([^,@\-][^,]+)`),
+	re(`-Wl,--start-group`),
+	re(`-Wl,-?-static`),
+	re(`-Wl,-?-subsystem,(native|windows|console|posix|xbox)`),
+	re(`-Wl,-syslibroot[=,]([^,@\-][^,]+)`),
+	re(`-Wl,-undefined[=,]([^,@\-][^,]+)`),
+	re(`-Wl,-?-unresolved-symbols=[^,]+`),
+	re(`-Wl,--(no-)?warn-([^,]+)`),
+	re(`-Wl,-z,(no)?execstack`),
+	re(`-Wl,-z,relro`),
+
+	re(`[a-zA-Z0-9_/].*\.(a|o|obj|dll|dylib|so)`), // direct linker inputs: x.o or libfoo.so (but not -foo.o or @foo.o)
+	re(`\./.*\.(a|o|obj|dll|dylib|so)`),
+}
+
+var validLinkerFlagsWithNextArg = []string{
+	"-arch",
+	"-F",
+	"-l",
+	"-L",
+	"-framework",
+	"-isysroot",
+	"--sysroot",
+	"-target",
+	"-Wl,-framework",
+	"-Wl,-rpath",
+	"-Wl,-R",
+	"-Wl,--just-symbols",
+	"-Wl,-undefined",
+}
+
+func checkCompilerFlags(name string, list []string) error {
+	return checkFlags(name, list, validCompilerFlags, validCompilerFlagsWithNextArg)
+}
+
+func checkLinkerFlags(name string, list []string) error {
+	return checkFlags(name, list, validLinkerFlags, validLinkerFlagsWithNextArg)
+}
+
+func checkFlags(name string, list []string, valid []*regexp.Regexp, validNext []string) error {
+	// Let users override rules with $CGO_CFLAGS_ALLOW, $CGO_CFLAGS_DISALLOW, etc.
+	var (
+		allow    *regexp.Regexp
+		disallow *regexp.Regexp
+	)
+	if env := os.Getenv("CGO_" + name + "_ALLOW"); env != "" {
+		r, err := regexp.Compile(env)
+		if err != nil {
+			return fmt.Errorf("parsing $CGO_%s_ALLOW: %v", name, err)
+		}
+		allow = r
+	}
+	if env := os.Getenv("CGO_" + name + "_DISALLOW"); env != "" {
+		r, err := regexp.Compile(env)
+		if err != nil {
+			return fmt.Errorf("parsing $CGO_%s_DISALLOW: %v", name, err)
+		}
+		disallow = r
+	}
+
+Args:
+	for i := 0; i < len(list); i++ {
+		arg := list[i]
+		if disallow != nil && disallow.FindString(arg) == arg {
+			goto Bad
+		}
+		if allow != nil && allow.FindString(arg) == arg {
+			continue Args
+		}
+		for _, re := range valid {
+			if re.FindString(arg) == arg { // must be complete match
+				continue Args
+			}
+		}
+		for _, x := range validNext {
+			if arg == x {
+				if i+1 < len(list) && safeArg(list[i+1]) {
+					i++
+					continue Args
+				}
+
+				// Permit -Wl,-framework -Wl,name.
+				if i+1 < len(list) &&
+					strings.HasPrefix(arg, "-Wl,") &&
+					strings.HasPrefix(list[i+1], "-Wl,") &&
+					safeArg(list[i+1][4:]) &&
+					!strings.Contains(list[i+1][4:], ",") {
+					i++
+					continue Args
+				}
+
+				if i+1 < len(list) {
+					return fmt.Errorf("invalid flag: %s %s (see https://golang.org/s/invalidflag)", arg, list[i+1])
+				}
+				return fmt.Errorf("invalid flag: %s without argument (see https://golang.org/s/invalidflag)", arg)
+			}
+		}
+	Bad:
+		return fmt.Errorf("invalid flag: %s", arg)
+	}
+	return nil
+}
+
+func safeArg(name string) bool {
+	if name == "" {
+		return false
+	}
+	c := name[0]
+	return '0' <= c && c <= '9' || 'A' <= c && c <= 'Z' || 'a' <= c && c <= 'z' || c == '.' || c == '_' || c == '/' || c >= utf8.RuneSelf
+}

cgo/security_test.go

@@ -0,0 +1,260 @@
+// Copyright 2018 The Go Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+// This file has been copied from the Go 1.13 release tree.
+
+package cgo
+
+import (
+	"os"
+	"testing"
+)
+
+var goodCompilerFlags = [][]string{
+	{"-DFOO"},
+	{"-Dfoo=bar"},
+	{"-F/Qt"},
+	{"-I/"},
+	{"-I/etc/passwd"},
+	{"-I."},
+	{"-O"},
+	{"-O2"},
+	{"-Osmall"},
+	{"-W"},
+	{"-Wall"},
+	{"-fobjc-arc"},
+	{"-fno-objc-arc"},
+	{"-fomit-frame-pointer"},
+	{"-fno-omit-frame-pointer"},
+	{"-fpic"},
+	{"-fno-pic"},
+	{"-fPIC"},
+	{"-fno-PIC"},
+	{"-fpie"},
+	{"-fno-pie"},
+	{"-fPIE"},
+	{"-fno-PIE"},
+	{"-fsplit-stack"},
+	{"-fno-split-stack"},
+	{"-fstack-xxx"},
+	{"-fno-stack-xxx"},
+	{"-fsanitize=hands"},
+	{"-g"},
+	{"-ggdb"},
+	{"-march=souza"},
+	{"-mcpu=123"},
+	{"-mfpu=123"},
+	{"-mtune=happybirthday"},
+	{"-mstack-overflow"},
+	{"-mno-stack-overflow"},
+	{"-mmacosx-version"},
+	{"-mnop-fun-dllimport"},
+	{"-pthread"},
+	{"-std=c99"},
+	{"-xc"},
+	{"-D", "FOO"},
+	{"-D", "foo=bar"},
+	{"-I", "."},
+	{"-I", "/etc/passwd"},
+	{"-I", "世界"},
+	{"-framework", "Chocolate"},
+	{"-x", "c"},
+	{"-v"},
+}
+
+var badCompilerFlags = [][]string{
+	{"-D@X"},
+	{"-D-X"},
+	{"-F@dir"},
+	{"-F-dir"},
+	{"-I@dir"},
+	{"-I-dir"},
+	{"-O@1"},
+	{"-Wa,-foo"},
+	{"-W@foo"},
+	{"-g@gdb"},
+	{"-g-gdb"},
+	{"-march=@dawn"},
+	{"-march=-dawn"},
+	{"-std=@c99"},
+	{"-std=-c99"},
+	{"-x@c"},
+	{"-x-c"},
+	{"-D", "@foo"},
+	{"-D", "-foo"},
+	{"-I", "@foo"},
+	{"-I", "-foo"},
+	{"-framework", "-Caffeine"},
+	{"-framework", "@Home"},
+	{"-x", "--c"},
+	{"-x", "@obj"},
+}
+
+func TestCheckCompilerFlags(t *testing.T) {
+	for _, f := range goodCompilerFlags {
+		if err := checkCompilerFlags("test", f); err != nil {
+			t.Errorf("unexpected error for %q: %v", f, err)
+		}
+	}
+	for _, f := range badCompilerFlags {
+		if err := checkCompilerFlags("test", f); err == nil {
+			t.Errorf("missing error for %q", f)
+		}
+	}
+}
+
+var goodLinkerFlags = [][]string{
+	{"-Fbar"},
+	{"-lbar"},
+	{"-Lbar"},
+	{"-fpic"},
+	{"-fno-pic"},
+	{"-fPIC"},
+	{"-fno-PIC"},
+	{"-fpie"},
+	{"-fno-pie"},
+	{"-fPIE"},
+	{"-fno-PIE"},
+	{"-fsanitize=hands"},
+	{"-g"},
+	{"-ggdb"},
+	{"-march=souza"},
+	{"-mcpu=123"},
+	{"-mfpu=123"},
+	{"-mtune=happybirthday"},
+	{"-pic"},
+	{"-pthread"},
+	{"-Wl,-rpath,foo"},
+	{"-Wl,-rpath,$ORIGIN/foo"},
+	{"-Wl,-R", "/foo"},
+	{"-Wl,-R", "foo"},
+	{"-Wl,-R,foo"},
+	{"-Wl,--just-symbols=foo"},
+	{"-Wl,--just-symbols,foo"},
+	{"-Wl,--warn-error"},
+	{"-Wl,--no-warn-error"},
+	{"foo.so"},
+	{"_世界.dll"},
+	{"./x.o"},
+	{"libcgosotest.dylib"},
+	{"-F", "framework"},
+	{"-l", "."},
+	{"-l", "/etc/passwd"},
+	{"-l", "世界"},
+	{"-L", "framework"},
+	{"-framework", "Chocolate"},
+	{"-v"},
+	{"-Wl,-framework", "-Wl,Chocolate"},
+	{"-Wl,-framework,Chocolate"},
+	{"-Wl,-unresolved-symbols=ignore-all"},
+}
+
+var badLinkerFlags = [][]string{
+	{"-DFOO"},
+	{"-Dfoo=bar"},
+	{"-W"},
+	{"-Wall"},
+	{"-fobjc-arc"},
+	{"-fno-objc-arc"},
+	{"-fomit-frame-pointer"},
+	{"-fno-omit-frame-pointer"},
+	{"-fsplit-stack"},
+	{"-fno-split-stack"},
+	{"-fstack-xxx"},
+	{"-fno-stack-xxx"},
+	{"-mstack-overflow"},
+	{"-mno-stack-overflow"},
+	{"-mnop-fun-dllimport"},
+	{"-std=c99"},
+	{"-xc"},
+	{"-D", "FOO"},
+	{"-D", "foo=bar"},
+	{"-I", "FOO"},
+	{"-L", "@foo"},
+	{"-L", "-foo"},
+	{"-x", "c"},
+	{"-D@X"},
+	{"-D-X"},
+	{"-I@dir"},
+	{"-I-dir"},
+	{"-O@1"},
+	{"-Wa,-foo"},
+	{"-W@foo"},
+	{"-g@gdb"},
+	{"-g-gdb"},
+	{"-march=@dawn"},
+	{"-march=-dawn"},
+	{"-std=@c99"},
+	{"-std=-c99"},
+	{"-x@c"},
+	{"-x-c"},
+	{"-D", "@foo"},
+	{"-D", "-foo"},
+	{"-I", "@foo"},
+	{"-I", "-foo"},
+	{"-l", "@foo"},
+	{"-l", "-foo"},
+	{"-framework", "-Caffeine"},
+	{"-framework", "@Home"},
+	{"-Wl,-framework,-Caffeine"},
+	{"-Wl,-framework", "-Wl,@Home"},
+	{"-Wl,-framework", "@Home"},
+	{"-Wl,-framework,Chocolate,@Home"},
+	{"-x", "--c"},
+	{"-x", "@obj"},
+	{"-Wl,-rpath,@foo"},
+	{"-Wl,-R,foo,bar"},
+	{"-Wl,-R,@foo"},
+	{"-Wl,--just-symbols,@foo"},
+	{"../x.o"},
+}
+
+func TestCheckLinkerFlags(t *testing.T) {
+	for _, f := range goodLinkerFlags {
+		if err := checkLinkerFlags("test", f); err != nil {
+			t.Errorf("unexpected error for %q: %v", f, err)
+		}
+	}
+	for _, f := range badLinkerFlags {
+		if err := checkLinkerFlags("test", f); err == nil {
+			t.Errorf("missing error for %q", f)
+		}
+	}
+}
+
+func TestCheckFlagAllowDisallow(t *testing.T) {
+	if err := checkCompilerFlags("TEST", []string{"-disallow"}); err == nil {
+		t.Fatalf("missing error for -disallow")
+	}
+	os.Setenv("CGO_TEST_ALLOW", "-disallo")
+	if err := checkCompilerFlags("TEST", []string{"-disallow"}); err == nil {
+		t.Fatalf("missing error for -disallow with CGO_TEST_ALLOW=-disallo")
+	}
+	os.Setenv("CGO_TEST_ALLOW", "-disallow")
+	if err := checkCompilerFlags("TEST", []string{"-disallow"}); err != nil {
+		t.Fatalf("unexpected error for -disallow with CGO_TEST_ALLOW=-disallow: %v", err)
+	}
+	os.Unsetenv("CGO_TEST_ALLOW")
+
+	if err := checkCompilerFlags("TEST", []string{"-Wall"}); err != nil {
+		t.Fatalf("unexpected error for -Wall: %v", err)
+	}
+	os.Setenv("CGO_TEST_DISALLOW", "-Wall")
+	if err := checkCompilerFlags("TEST", []string{"-Wall"}); err == nil {
+		t.Fatalf("missing error for -Wall with CGO_TEST_DISALLOW=-Wall")
+	}
+	os.Setenv("CGO_TEST_ALLOW", "-Wall") // disallow wins
+	if err := checkCompilerFlags("TEST", []string{"-Wall"}); err == nil {
+		t.Fatalf("missing error for -Wall with CGO_TEST_DISALLOW=-Wall and CGO_TEST_ALLOW=-Wall")
+	}
+
+	os.Setenv("CGO_TEST_ALLOW", "-fplugin.*")
+	os.Setenv("CGO_TEST_DISALLOW", "-fplugin=lint.so")
+	if err := checkCompilerFlags("TEST", []string{"-fplugin=faster.so"}); err != nil {
+		t.Fatalf("unexpected error for -fplugin=faster.so: %v", err)
+	}
+	if err := checkCompilerFlags("TEST", []string{"-fplugin=lint.so"}); err == nil {
+		t.Fatalf("missing error for -fplugin=lint.so: %v", err)
+	}
+}

cgo/testdata/basic.go

@@ -0,0 +1,3 @@
+package main
+
+import "C"

cgo/testdata/basic.out.go

@@ -0,0 +1,26 @@
+package main
+
+import "unsafe"
+
+var _ unsafe.Pointer
+
+type C.int16_t = int16
+type C.int32_t = int32
+type C.int64_t = int64
+type C.int8_t = int8
+type C.uint16_t = uint16
+type C.uint32_t = uint32
+type C.uint64_t = uint64
+type C.uint8_t = uint8
+type C.uintptr_t = uintptr
+type C.char uint8
+type C.int int32
+type C.long int32
+type C.longlong int64
+type C.schar int8
+type C.short int16
+type C.uchar uint8
+type C.uint uint32
+type C.ulong uint32
+type C.ulonglong uint64
+type C.ushort uint16

cgo/testdata/const.go

@@ -0,0 +1,12 @@
+package main
+
+/*
+#define foo 3
+#define bar foo
+*/
+import "C"
+
+const (
+	Foo = C.foo
+	Bar = C.bar
+)

cgo/testdata/const.out.go

@@ -0,0 +1,29 @@
+package main
+
+import "unsafe"
+
+var _ unsafe.Pointer
+
+const C.bar = C.foo
+const C.foo = 3
+
+type C.int16_t = int16
+type C.int32_t = int32
+type C.int64_t = int64
+type C.int8_t = int8
+type C.uint16_t = uint16
+type C.uint32_t = uint32
+type C.uint64_t = uint64
+type C.uint8_t = uint8
+type C.uintptr_t = uintptr
+type C.char uint8
+type C.int int32
+type C.long int32
+type C.longlong int64
+type C.schar int8
+type C.short int16
+type C.uchar uint8
+type C.uint uint32
+type C.ulong uint32
+type C.ulonglong uint64
+type C.ushort uint16

cgo/testdata/errors.go

@@ -0,0 +1,33 @@
+package main
+
+/*
+#warning some warning
+
+typedef struct {
+	int x;
+	int y;
+} point_t;
+
+typedef someType noType; // undefined type
+
+#define SOME_CONST_1 5) // invalid const syntax
+#define SOME_CONST_2 6) // const not used (so no error)
+#define SOME_CONST_3 1234 // const too large for byte
+*/
+import "C"
+
+// Make sure that errors for the following lines won't change with future
+// additions to the CGo preamble.
+//line errors.go:100
+var (
+	// constant too large
+	_ C.uint8_t = 2 << 10
+
+	// z member does not exist
+	_ C.point_t = C.point_t{z: 3}
+
+	// constant has syntax error
+	_ = C.SOME_CONST_1
+
+	_ byte = C.SOME_CONST_3
+)

cgo/testdata/errors.out.go

@@ -0,0 +1,43 @@
+// CGo errors:
+//     testdata/errors.go:4:2: warning: some warning
+//     testdata/errors.go:11:9: error: unknown type name 'someType'
+//     testdata/errors.go:13:23: unexpected token )
+
+// Type checking errors after CGo processing:
+//     testdata/errors.go:102: 2 << 10 (untyped int constant 2048) overflows uint8
+//     testdata/errors.go:105: unknown field z in struct literal
+//     testdata/errors.go:108: undeclared name: C.SOME_CONST_1
+//     testdata/errors.go:110: C.SOME_CONST_3 (untyped int constant 1234) overflows byte
+
+package main
+
+import "unsafe"
+
+var _ unsafe.Pointer
+
+const C.SOME_CONST_3 = 1234
+
+type C.int16_t = int16
+type C.int32_t = int32
+type C.int64_t = int64
+type C.int8_t = int8
+type C.uint16_t = uint16
+type C.uint32_t = uint32
+type C.uint64_t = uint64
+type C.uint8_t = uint8
+type C.uintptr_t = uintptr
+type C.char uint8
+type C.int int32
+type C.long int32
+type C.longlong int64
+type C.schar int8
+type C.short int16
+type C.uchar uint8
+type C.uint uint32
+type C.ulong uint32
+type C.ulonglong uint64
+type C.ushort uint16
+type C.point_t = struct {
+	x C.int
+	y C.int
+}

cgo/testdata/flags.go

@@ -0,0 +1,26 @@
+package main
+
+/*
+// this name doesn't exist
+#cgo  NOFLAGS: -foo
+
+// unknown flag
+#cgo CFLAGS: -fdoes-not-exist -DNOTDEFINED
+
+#cgo CFLAGS: -DFOO
+
+#if defined(FOO)
+#define BAR 3
+#else
+#define BAR 5
+#endif
+
+#if defined(NOTDEFINED)
+#warning flag must not be defined
+#endif
+*/
+import "C"
+
+var (
+	_ = C.BAR
+)

cgo/testdata/flags.out.go

@@ -0,0 +1,32 @@
+// CGo errors:
+//     testdata/flags.go:5:7: invalid #cgo line: NOFLAGS
+//     testdata/flags.go:8:13: invalid flag: -fdoes-not-exist
+
+package main
+
+import "unsafe"
+
+var _ unsafe.Pointer
+
+const C.BAR = 3
+
+type C.int16_t = int16
+type C.int32_t = int32
+type C.int64_t = int64
+type C.int8_t = int8
+type C.uint16_t = uint16
+type C.uint32_t = uint32
+type C.uint64_t = uint64
+type C.uint8_t = uint8
+type C.uintptr_t = uintptr
+type C.char uint8
+type C.int int32
+type C.long int32
+type C.longlong int64
+type C.schar int8
+type C.short int16
+type C.uchar uint8
+type C.uint uint32
+type C.ulong uint32
+type C.ulonglong uint64
+type C.ushort uint16

cgo/testdata/types.go

@@ -0,0 +1,165 @@
+package main
+
+/*
+// Simple typedef.
+typedef int myint;
+
+// Structs, with or without name.
+typedef struct {
+	int x;
+	int y;
+} point2d_t;
+typedef struct point3d {
+	int x;
+	int y;
+	int z;
+} point3d_t;
+
+// Structs with reserved field names.
+struct type1 {
+	// All these fields should be renamed.
+	int type;
+	int _type;
+	int __type;
+};
+struct type2 {
+	// This field should not be renamed.
+	int _type;
+};
+
+// Unions.
+typedef union {
+	// Union should be treated as a struct.
+	int i;
+} union1_t;
+typedef union {
+	// Union must contain a single field and have special getters/setters.
+	int    i;
+	double d;
+	short  s;
+} union3_t;
+typedef union union2d {
+	int i;
+	double d[2];
+} union2d_t;
+typedef union {
+	unsigned char arr[10];
+} unionarray_t;
+
+// Nested structs and unions.
+typedef struct {
+	point2d_t begin;
+	point2d_t end;
+	int       tag;
+	union {
+		point2d_t area;
+		point3d_t solid;
+	} coord;
+} struct_nested_t;
+typedef union {
+	point3d_t    point;
+	unionarray_t array;
+	union3_t     thing;
+} union_nested_t;
+
+// Enums. These define constant numbers. All these constants must be given the
+// correct number.
+typedef enum option {
+	optionA,
+	optionB,
+	optionC = -5,
+	optionD,
+	optionE = 10,
+	optionF,
+	optionG,
+} option_t;
+enum unused {
+	unused1 = 5,
+};
+
+// Anonymous enum.
+typedef enum {
+	option2A = 20,
+} option2_t;
+
+// Various types that are usually translated directly to Go types, but storing
+// them in a struct reveals them.
+typedef struct {
+	float  f;
+	double d;
+	int    *ptr;
+} types_t;
+
+// Arrays.
+typedef int myIntArray[10];
+
+// Bitfields.
+typedef struct {
+	unsigned char start;
+	unsigned char a : 5;
+	unsigned char b : 1;
+	unsigned char c : 2;
+	unsigned char :0; // new field
+	unsigned char d : 6;
+	unsigned char e : 3;
+	// Note that C++ allows bitfields bigger than the underlying type.
+} bitfield_t;
+*/
+import "C"
+
+var (
+	// Simple typedefs.
+	_ C.myint
+
+	// Structs.
+	_ C.point2d_t
+	_ C.point3d_t
+	_ C.struct_point3d
+
+	// Structs with reserved field names.
+	_ C.struct_type1
+	_ C.struct_type2
+
+	// Unions.
+	_ C.union1_t
+	_ C.union3_t
+	_ C.union2d_t
+	_ C.unionarray_t
+
+	// Nested structs and unions.
+	_ C.struct_nested_t
+	_ C.union_nested_t
+
+	// Enums (anonymous and named).
+	_ C.option_t
+	_ C.enum_option
+	_ C.option2_t
+
+	// Various types.
+	_ C.types_t
+
+	// Arrays.
+	_ C.myIntArray
+)
+
+// Test bitfield accesses.
+func accessBitfields() {
+	var x C.bitfield_t
+	x.start = 3
+	x.set_bitfield_a(4)
+	x.set_bitfield_b(1)
+	x.set_bitfield_c(2)
+	x.d = 10
+	x.e = 5
+	var _ C.uchar = x.bitfield_a()
+}
+
+// Test union accesses.
+func accessUnion() {
+	var union1 C.union1_t
+	union1.i = 5
+
+	var union2d C.union2d_t
+	var _ *C.int = union2d.unionfield_i()
+	var _ *[2]float64 = union2d.unionfield_d()
+}

cgo/testdata/types.out.go

@@ -0,0 +1,119 @@
+package main
+
+import "unsafe"
+
+var _ unsafe.Pointer
+
+const C.option2A = 20
+const C.optionA = 0
+const C.optionB = 1
+const C.optionC = -5
+const C.optionD = -4
+const C.optionE = 10
+const C.optionF = 11
+const C.optionG = 12
+const C.unused1 = 5
+
+type C.int16_t = int16
+type C.int32_t = int32
+type C.int64_t = int64
+type C.int8_t = int8
+type C.uint16_t = uint16
+type C.uint32_t = uint32
+type C.uint64_t = uint64
+type C.uint8_t = uint8
+type C.uintptr_t = uintptr
+type C.char uint8
+type C.int int32
+type C.long int32
+type C.longlong int64
+type C.schar int8
+type C.short int16
+type C.uchar uint8
+type C.uint uint32
+type C.ulong uint32
+type C.ulonglong uint64
+type C.ushort uint16
+type C.bitfield_t = C.struct_4
+type C.myIntArray = [10]C.int
+type C.myint = C.int
+type C.option2_t = C.uint
+type C.option_t = C.enum_option
+type C.point2d_t = struct {
+	x C.int
+	y C.int
+}
+type C.point3d_t = C.struct_point3d
+type C.struct_nested_t = struct {
+	begin C.point2d_t
+	end   C.point2d_t
+	tag   C.int
+
+	coord C.union_2
+}
+type C.types_t = struct {
+	f   float32
+	d   float64
+	ptr *C.int
+}
+type C.union1_t = struct{ i C.int }
+type C.union2d_t = C.union_union2d
+type C.union3_t = C.union_1
+type C.union_nested_t = C.union_3
+type C.unionarray_t = struct{ arr [10]C.uchar }
+
+func (s *C.struct_4) bitfield_a() C.uchar          { return s.__bitfield_1 & 0x1f }
+func (s *C.struct_4) set_bitfield_a(value C.uchar) { s.__bitfield_1 = s.__bitfield_1&^0x1f | value&0x1f<<0 }
+func (s *C.struct_4) bitfield_b() C.uchar {
+	return s.__bitfield_1 >> 5 & 0x1
+}
+func (s *C.struct_4) set_bitfield_b(value C.uchar) { s.__bitfield_1 = s.__bitfield_1&^0x20 | value&0x1<<5 }
+func (s *C.struct_4) bitfield_c() C.uchar {
+	return s.__bitfield_1 >> 6
+}
+func (s *C.struct_4) set_bitfield_c(value C.uchar,
+
+) { s.__bitfield_1 = s.__bitfield_1&0x3f | value<<6 }
+
+type C.struct_4 struct {
+	start        C.uchar
+	__bitfield_1 C.uchar
+
+	d C.uchar
+	e C.uchar
+}
+type C.struct_point3d struct {
+	x C.int
+	y C.int
+	z C.int
+}
+type C.struct_type1 struct {
+	_type   C.int
+	__type  C.int
+	___type C.int
+}
+type C.struct_type2 struct{ _type C.int }
+
+func (union *C.union_1) unionfield_i() *C.int   { return (*C.int)(unsafe.Pointer(&union.$union)) }
+func (union *C.union_1) unionfield_d() *float64 { return (*float64)(unsafe.Pointer(&union.$union)) }
+func (union *C.union_1) unionfield_s() *C.short { return (*C.short)(unsafe.Pointer(&union.$union)) }
+
+type C.union_1 struct{ $union uint64 }
+
+func (union *C.union_2) unionfield_area() *C.point2d_t  { return (*C.point2d_t)(unsafe.Pointer(&union.$union)) }
+func (union *C.union_2) unionfield_solid() *C.point3d_t { return (*C.point3d_t)(unsafe.Pointer(&union.$union)) }
+
+type C.union_2 struct{ $union [3]uint32 }
+
+func (union *C.union_3) unionfield_point() *C.point3d_t    { return (*C.point3d_t)(unsafe.Pointer(&union.$union)) }
+func (union *C.union_3) unionfield_array() *C.unionarray_t { return (*C.unionarray_t)(unsafe.Pointer(&union.$union)) }
+func (union *C.union_3) unionfield_thing() *C.union3_t     { return (*C.union3_t)(unsafe.Pointer(&union.$union)) }
+
+type C.union_3 struct{ $union [2]uint64 }
+
+func (union *C.union_union2d) unionfield_i() *C.int      { return (*C.int)(unsafe.Pointer(&union.$union)) }
+func (union *C.union_union2d) unionfield_d() *[2]float64 { return (*[2]float64)(unsafe.Pointer(&union.$union)) }
+
+type C.union_union2d struct{ $union [2]uint64 }
+type C.enum_option C.int
+type C.enum_unused C.uint

commands.go

@@ -1,45 +0,0 @@
-package main
-
-import (
-	"errors"
-	"os"
-	"os/exec"
-	"runtime"
-	"strings"
-)
-
-// Commands used by the compilation process might have different file names
-// across operating systems and distributions.
-var commands = map[string][]string{
-	"clang":   {"clang-8"},
-	"ld.lld":  {"ld.lld-8", "ld.lld"},
-	"wasm-ld": {"wasm-ld-8", "wasm-ld"},
-}
-
-func init() {
-	// Add the path to a Homebrew-installed LLVM 8 for ease of use (no need to
-	// manually set $PATH).
-	if runtime.GOOS == "darwin" {
-		commands["clang"] = append(commands["clang"], "/usr/local/opt/llvm/bin/clang-8")
-		commands["ld.lld"] = append(commands["ld.lld"], "/usr/local/opt/llvm/bin/ld.lld")
-		commands["wasm-ld"] = append(commands["wasm-ld"], "/usr/local/opt/llvm/bin/wasm-ld")
-	}
-}
-
-func execCommand(cmdNames []string, args ...string) error {
-	for _, cmdName := range cmdNames {
-		cmd := exec.Command(cmdName, args...)
-		cmd.Stdout = os.Stdout
-		cmd.Stderr = os.Stderr
-		err := cmd.Run()
-		if err != nil {
-			if err, ok := err.(*exec.Error); ok && err.Err == exec.ErrNotFound {
-				// this command was not found, try the next
-				continue
-			}
-			return err
-		}
-		return nil
-	}
-	return errors.New("none of these commands were found in your $PATH: " + strings.Join(cmdNames, " "))
-}

compileopts/config.go

@@ -0,0 +1,221 @@
+// Package compileopts contains the configuration for a single to-be-built
+// binary.
+package compileopts
+
+import (
+	"errors"
+	"fmt"
+	"regexp"
+	"strconv"
+	"strings"
+
+	"github.com/tinygo-org/tinygo/goenv"
+)
+
+// Config keeps all configuration affecting the build in a single struct.
+type Config struct {
+	Options        *Options
+	Target         *TargetSpec
+	GoMinorVersion int
+	ClangHeaders   string // Clang built-in header include path
+	TestConfig     TestConfig
+}
+
+// Triple returns the LLVM target triple, like armv6m-none-eabi.
+func (c *Config) Triple() string {
+	return c.Target.Triple
+}
+
+// CPU returns the LLVM CPU name, like atmega328p or arm7tdmi. It may return an
+// empty string if the CPU name is not known.
+func (c *Config) CPU() string {
+	return c.Target.CPU
+}
+
+// Features returns a list of features this CPU supports. For example, for a
+// RISC-V processor, that could be ["+a", "+c", "+m"]. For many targets, an
+// empty list will be returned.
+func (c *Config) Features() []string {
+	return c.Target.Features
+}
+
+// GOOS returns the GOOS of the target. This might not always be the actual OS:
+// for example, bare-metal targets will usually pretend to be linux to get the
+// standard library to compile.
+func (c *Config) GOOS() string {
+	return c.Target.GOOS
+}
+
+// GOARCH returns the GOARCH of the target. This might not always be the actual
+// archtecture: for example, the AVR target is not supported by the Go standard
+// library so such targets will usually pretend to be linux/arm.
+func (c *Config) GOARCH() string {
+	return c.Target.GOARCH
+}
+
+// BuildTags returns the complete list of build tags used during this build.
+func (c *Config) BuildTags() []string {
+	tags := append(c.Target.BuildTags, []string{"tinygo", "gc." + c.GC(), "scheduler." + c.Scheduler()}...)
+	for i := 1; i <= c.GoMinorVersion; i++ {
+		tags = append(tags, fmt.Sprintf("go1.%d", i))
+	}
+	if extraTags := strings.Fields(c.Options.Tags); len(extraTags) != 0 {
+		tags = append(tags, extraTags...)
+	}
+	return tags
+}
+
+// GC returns the garbage collection strategy in use on this platform. Valid
+// values are "none", "leaking", and "conservative".
+func (c *Config) GC() string {
+	if c.Options.GC != "" {
+		return c.Options.GC
+	}
+	if c.Target.GC != "" {
+		return c.Target.GC
+	}
+	return "conservative"
+}
+
+// NeedsStackObjects returns true if the compiler should insert stack objects
+// that can be traced by the garbage collector.
+func (c *Config) NeedsStackObjects() bool {
+	if c.GC() != "conservative" {
+		return false
+	}
+	for _, tag := range c.BuildTags() {
+		if tag == "baremetal" {
+			return false
+		}
+	}
+
+	return true
+}
+
+// Scheduler returns the scheduler implementation. Valid values are "coroutines"
+// and "tasks".
+func (c *Config) Scheduler() string {
+	if c.Options.Scheduler != "" {
+		return c.Options.Scheduler
+	}
+	if c.Target.Scheduler != "" {
+		return c.Target.Scheduler
+	}
+	// Fall back to coroutines, which are supported everywhere.
+	return "coroutines"
+}
+
+// PanicStrategy returns the panic strategy selected for this target. Valid
+// values are "print" (print the panic value, then exit) or "trap" (issue a trap
+// instruction).
+func (c *Config) PanicStrategy() string {
+	return c.Options.PanicStrategy
+}
+
+// CFlags returns the flags to pass to the C compiler. This is necessary for CGo
+// preprocessing.
+func (c *Config) CFlags() []string {
+	cflags := append([]string{}, c.Options.CFlags...)
+	for _, flag := range c.Target.CFlags {
+		cflags = append(cflags, strings.Replace(flag, "{root}", goenv.Get("TINYGOROOT"), -1))
+	}
+	return cflags
+}
+
+// LDFlags returns the flags to pass to the linker. A few more flags are needed
+// (like the one for the compiler runtime), but this represents the majority of
+// the flags.
+func (c *Config) LDFlags() []string {
+	root := goenv.Get("TINYGOROOT")
+	// Merge and adjust LDFlags.
+	ldflags := append([]string{}, c.Options.LDFlags...)
+	for _, flag := range c.Target.LDFlags {
+		ldflags = append(ldflags, strings.Replace(flag, "{root}", root, -1))
+	}
+	ldflags = append(ldflags, "-L", root)
+	if c.Target.GOARCH == "wasm" {
+		// Round heap size to next multiple of 65536 (the WebAssembly page
+		// size).
+		heapSize := (c.Options.HeapSize + (65536 - 1)) &^ (65536 - 1)
+		ldflags = append(ldflags, "--initial-memory="+strconv.FormatInt(heapSize, 10))
+	}
+	if c.Target.LinkerScript != "" {
+		ldflags = append(ldflags, "-T", c.Target.LinkerScript)
+	}
+	return ldflags
+}
+
+// ExtraFiles returns the list of extra files to be built and linked with the
+// executable. This can include extra C and assembly files.
+func (c *Config) ExtraFiles() []string {
+	return c.Target.ExtraFiles
+}
+
+// DumpSSA returns whether to dump Go SSA while compiling (-dumpssa flag). Only
+// enable this for debugging.
+func (c *Config) DumpSSA() bool {
+	return c.Options.DumpSSA
+}
+
+// VerifyIR returns whether to run extra checks on the IR. This is normally
+// disabled but enabled during testing.
+func (c *Config) VerifyIR() bool {
+	return c.Options.VerifyIR
+}
+
+// Debug returns whether to add debug symbols to the IR, for debugging with GDB
+// and similar.
+func (c *Config) Debug() bool {
+	return c.Options.Debug
+}
+
+// Programmer returns the flash method and OpenOCD interface name given a
+// particular configuration. It may either be all configured in the target JSON
+// file or be modified using the -programmmer command-line option.
+func (c *Config) Programmer() (method, openocdInterface string) {
+	switch c.Options.Programmer {
+	case "":
+		// No configuration supplied.
+		return c.Target.FlashMethod, c.Target.OpenOCDInterface
+	case "openocd", "msd", "command":
+		// The -programmer flag only specifies the flash method.
+		return c.Options.Programmer, c.Target.OpenOCDInterface
+	default:
+		// The -programmer flag specifies something else, assume it specifies
+		// the OpenOCD interface name.
+		return "openocd", c.Options.Programmer
+	}
+}
+
+// OpenOCDConfiguration returns a list of command line arguments to OpenOCD.
+// This list of command-line arguments is based on the various OpenOCD-related
+// flags in the target specification.
+func (c *Config) OpenOCDConfiguration() (args []string, err error) {
+	_, openocdInterface := c.Programmer()
+	if openocdInterface == "" {
+		return nil, errors.New("OpenOCD programmer not set")
+	}
+	if !regexp.MustCompile("^[\\p{L}0-9_-]+$").MatchString(openocdInterface) {
+		return nil, fmt.Errorf("OpenOCD programmer has an invalid name: %#v", openocdInterface)
+	}
+	if c.Target.OpenOCDTarget == "" {
+		return nil, errors.New("OpenOCD chip not set")
+	}
+	if !regexp.MustCompile("^[\\p{L}0-9_-]+$").MatchString(c.Target.OpenOCDTarget) {
+		return nil, fmt.Errorf("OpenOCD target has an invalid name: %#v", c.Target.OpenOCDTarget)
+	}
+	if c.Target.OpenOCDTransport != "" && c.Target.OpenOCDTransport != "swd" {
+		return nil, fmt.Errorf("unknown OpenOCD transport: %#v", c.Target.OpenOCDTransport)
+	}
+	args = []string{"-f", "interface/" + openocdInterface + ".cfg"}
+	if c.Target.OpenOCDTransport != "" {
+		args = append(args, "-c", "transport select "+c.Target.OpenOCDTransport)
+	}
+	args = append(args, "-f", "target/"+c.Target.OpenOCDTarget+".cfg")
+	return args, nil
+}
+
+type TestConfig struct {
+	CompileTestBinary bool
+	// TODO: Filter the test functions to run, include verbose flag, etc
+}

compileopts/options.go

@@ -0,0 +1,23 @@
+package compileopts
+
+// Options contains extra options to give to the compiler. These options are
+// usually passed from the command line.
+type Options struct {
+	Target        string
+	Opt           string
+	GC            string
+	PanicStrategy string
+	Scheduler     string
+	PrintIR       bool
+	DumpSSA       bool
+	VerifyIR      bool
+	Debug         bool
+	PrintSizes    string
+	CFlags        []string
+	LDFlags       []string
+	Tags          string
+	WasmAbi       string
+	HeapSize      int64
+	TestConfig    TestConfig
+	Programmer    string
+}

compileopts/target.go

@@ -0,0 +1,277 @@
+package compileopts
+
+// This file loads a target specification from a JSON file.
+
+import (
+	"encoding/json"
+	"errors"
+	"io"
+	"os"
+	"path/filepath"
+	"runtime"
+	"strings"
+
+	"github.com/tinygo-org/tinygo/goenv"
+)
+
+// Target specification for a given target. Used for bare metal targets.
+//
+// The target specification is mostly inspired by Rust:
+// https://doc.rust-lang.org/nightly/nightly-rustc/rustc_target/spec/struct.TargetOptions.html
+// https://github.com/shepmaster/rust-arduino-blink-led-no-core-with-cargo/blob/master/blink/arduino.json
+type TargetSpec struct {
+	Inherits         []string `json:"inherits"`
+	Triple           string   `json:"llvm-target"`
+	CPU              string   `json:"cpu"`
+	Features         []string `json:"features"`
+	GOOS             string   `json:"goos"`
+	GOARCH           string   `json:"goarch"`
+	BuildTags        []string `json:"build-tags"`
+	GC               string   `json:"gc"`
+	Scheduler        string   `json:"scheduler"`
+	Compiler         string   `json:"compiler"`
+	Linker           string   `json:"linker"`
+	RTLib            string   `json:"rtlib"` // compiler runtime library (libgcc, compiler-rt)
+	CFlags           []string `json:"cflags"`
+	LDFlags          []string `json:"ldflags"`
+	LinkerScript     string   `json:"linkerscript"`
+	ExtraFiles       []string `json:"extra-files"`
+	Emulator         []string `json:"emulator"`
+	FlashCommand     string   `json:"flash-command"`
+	GDB              string   `json:"gdb"`
+	PortReset        string   `json:"flash-1200-bps-reset"`
+	FlashMethod      string   `json:"flash-method"`
+	FlashVolume      string   `json:"msd-volume-name"`
+	FlashFilename    string   `json:"msd-firmware-name"`
+	OpenOCDInterface string   `json:"openocd-interface"`
+	OpenOCDTarget    string   `json:"openocd-target"`
+	OpenOCDTransport string   `json:"openocd-transport"`
+}
+
+// copyProperties copies all properties that are set in spec2 into itself.
+func (spec *TargetSpec) copyProperties(spec2 *TargetSpec) {
+	// TODO: simplify this using reflection? Inherits and BuildTags are special
+	// cases, but the rest can simply be copied if set.
+	spec.Inherits = append(spec.Inherits, spec2.Inherits...)
+	if spec2.Triple != "" {
+		spec.Triple = spec2.Triple
+	}
+	if spec2.CPU != "" {
+		spec.CPU = spec2.CPU
+	}
+	spec.Features = append(spec.Features, spec2.Features...)
+	if spec2.GOOS != "" {
+		spec.GOOS = spec2.GOOS
+	}
+	if spec2.GOARCH != "" {
+		spec.GOARCH = spec2.GOARCH
+	}
+	spec.BuildTags = append(spec.BuildTags, spec2.BuildTags...)
+	if spec2.GC != "" {
+		spec.GC = spec2.GC
+	}
+	if spec2.Scheduler != "" {
+		spec.Scheduler = spec2.Scheduler
+	}
+	if spec2.Compiler != "" {
+		spec.Compiler = spec2.Compiler
+	}
+	if spec2.Linker != "" {
+		spec.Linker = spec2.Linker
+	}
+	if spec2.RTLib != "" {
+		spec.RTLib = spec2.RTLib
+	}
+	spec.CFlags = append(spec.CFlags, spec2.CFlags...)
+	spec.LDFlags = append(spec.LDFlags, spec2.LDFlags...)
+	if spec2.LinkerScript != "" {
+		spec.LinkerScript = spec2.LinkerScript
+	}
+	spec.ExtraFiles = append(spec.ExtraFiles, spec2.ExtraFiles...)
+	if len(spec2.Emulator) != 0 {
+		spec.Emulator = spec2.Emulator
+	}
+	if spec2.FlashCommand != "" {
+		spec.FlashCommand = spec2.FlashCommand
+	}
+	if spec2.GDB != "" {
+		spec.GDB = spec2.GDB
+	}
+	if spec2.PortReset != "" {
+		spec.PortReset = spec2.PortReset
+	}
+	if spec2.FlashMethod != "" {
+		spec.FlashMethod = spec2.FlashMethod
+	}
+	if spec2.FlashVolume != "" {
+		spec.FlashVolume = spec2.FlashVolume
+	}
+	if spec2.FlashFilename != "" {
+		spec.FlashFilename = spec2.FlashFilename
+	}
+	if spec2.OpenOCDInterface != "" {
+		spec.OpenOCDInterface = spec2.OpenOCDInterface
+	}
+	if spec2.OpenOCDTarget != "" {
+		spec.OpenOCDTarget = spec2.OpenOCDTarget
+	}
+	if spec2.OpenOCDTransport != "" {
+		spec.OpenOCDTransport = spec2.OpenOCDTransport
+	}
+}
+
+// load reads a target specification from the JSON in the given io.Reader. It
+// may load more targets specified using the "inherits" property.
+func (spec *TargetSpec) load(r io.Reader) error {
+	err := json.NewDecoder(r).Decode(spec)
+	if err != nil {
+		return err
+	}
+
+	return nil
+}
+
+// loadFromGivenStr loads the TargetSpec from the given string that could be:
+// - targets/ directory inside the compiler sources
+// - a relative or absolute path to custom (project specific) target specification .json file;
+//   the Inherits[] could contain the files from target folder (ex. stm32f4disco)
+//   as well as path to custom files (ex. myAwesomeProject.json)
+func (spec *TargetSpec) loadFromGivenStr(str string) error {
+	path := ""
+	if strings.HasSuffix(str, ".json") {
+		path, _ = filepath.Abs(str)
+	} else {
+		path = filepath.Join(goenv.Get("TINYGOROOT"), "targets", strings.ToLower(str)+".json")
+	}
+	fp, err := os.Open(path)
+	if err != nil {
+		return err
+	}
+	defer fp.Close()
+	return spec.load(fp)
+}
+
+// resolveInherits loads inherited targets, recursively.
+func (spec *TargetSpec) resolveInherits() error {
+	// First create a new spec with all the inherited properties.
+	newSpec := &TargetSpec{}
+	for _, name := range spec.Inherits {
+		subtarget := &TargetSpec{}
+		err := subtarget.loadFromGivenStr(name)
+		if err != nil {
+			return err
+		}
+		err = subtarget.resolveInherits()
+		if err != nil {
+			return err
+		}
+		newSpec.copyProperties(subtarget)
+	}
+
+	// When all properties are loaded, make sure they are properly inherited.
+	newSpec.copyProperties(spec)
+	*spec = *newSpec
+
+	return nil
+}
+
+// Load a target specification.
+func LoadTarget(target string) (*TargetSpec, error) {
+	if target == "" {
+		// Configure based on GOOS/GOARCH environment variables (falling back to
+		// runtime.GOOS/runtime.GOARCH), and generate a LLVM target based on it.
+		goos := goenv.Get("GOOS")
+		goarch := goenv.Get("GOARCH")
+		llvmos := goos
+		llvmarch := map[string]string{
+			"386":   "i386",
+			"amd64": "x86_64",
+			"arm64": "aarch64",
+		}[goarch]
+		if llvmarch == "" {
+			llvmarch = goarch
+		}
+		target = llvmarch + "--" + llvmos
+		if goarch == "arm" {
+			target += "-gnueabihf"
+		}
+		return defaultTarget(goos, goarch, target)
+	}
+
+	// See whether there is a target specification for this target (e.g.
+	// Arduino).
+	spec := &TargetSpec{}
+	err := spec.loadFromGivenStr(target)
+	if err == nil {
+		// Successfully loaded this target from a built-in .json file. Make sure
+		// it includes all parents as specified in the "inherits" key.
+		err = spec.resolveInherits()
+		if err != nil {
+			return nil, err
+		}
+		return spec, nil
+	} else if !os.IsNotExist(err) {
+		// Expected a 'file not found' error, got something else. Report it as
+		// an error.
+		return nil, err
+	} else {
+		// Load target from given triple, ignore GOOS/GOARCH environment
+		// variables.
+		tripleSplit := strings.Split(target, "-")
+		if len(tripleSplit) < 3 {
+			return nil, errors.New("expected a full LLVM target or a custom target in -target flag")
+		}
+		goos := tripleSplit[2]
+		if strings.HasPrefix(goos, "darwin") {
+			goos = "darwin"
+		}
+		goarch := map[string]string{ // map from LLVM arch to Go arch
+			"i386":    "386",
+			"x86_64":  "amd64",
+			"aarch64": "arm64",
+		}[tripleSplit[0]]
+		if goarch == "" {
+			goarch = tripleSplit[0]
+		}
+		return defaultTarget(goos, goarch, target)
+	}
+}
+
+func defaultTarget(goos, goarch, triple string) (*TargetSpec, error) {
+	// No target spec available. Use the default one, useful on most systems
+	// with a regular OS.
+	spec := TargetSpec{
+		Triple:      triple,
+		GOOS:        goos,
+		GOARCH:      goarch,
+		BuildTags:   []string{goos, goarch},
+		Compiler:    "clang",
+		Linker:      "cc",
+		GDB:         "gdb",
+		PortReset:   "false",
+		FlashMethod: "native",
+	}
+	if goos == "darwin" {
+		spec.LDFlags = append(spec.LDFlags, "-Wl,-dead_strip")
+	} else {
+		spec.LDFlags = append(spec.LDFlags, "-no-pie", "-Wl,--gc-sections") // WARNING: clang < 5.0 requires -nopie
+	}
+	if goarch != runtime.GOARCH {
+		// Some educated guesses as to how to invoke helper programs.
+		if goarch == "arm" && goos == "linux" {
+			spec.Linker = "arm-linux-gnueabihf-gcc"
+			spec.GDB = "arm-linux-gnueabihf-gdb"
+			spec.Emulator = []string{"qemu-arm", "-L", "/usr/arm-linux-gnueabihf"}
+		}
+		if goarch == "arm64" && goos == "linux" {
+			spec.Linker = "aarch64-linux-gnu-gcc"
+			spec.GDB = "aarch64-linux-gnu-gdb"
+			spec.Emulator = []string{"qemu-aarch64", "-L", "/usr/aarch64-linux-gnu"}
+		}
+		if goarch == "386" {
+			spec.CFlags = []string{"-m32"}
+			spec.LDFlags = []string{"-m32"}
+		}
+	}
+	return &spec, nil
+}

compileopts/target_test.go

@@ -1,4 +1,4 @@
-package main
+package compileopts
 
 import "testing"
 

compiler/asserts.go

@@ -22,7 +22,7 @@ func (c *Compiler) emitLookupBoundsCheck(frame *Frame, arrayLen, index llvm.Valu
 	if index.Type().IntTypeWidth() < arrayLen.Type().IntTypeWidth() {
 		// Sometimes, the index can be e.g. an uint8 or int8, and we have to
 		// correctly extend that type.
-		if indexType.(*types.Basic).Info()&types.IsUnsigned == 0 {
+		if indexType.Underlying().(*types.Basic).Info()&types.IsUnsigned == 0 {
 			index = c.builder.CreateZExt(index, arrayLen.Type(), "")
 		} else {
 			index = c.builder.CreateSExt(index, arrayLen.Type(), "")

compiler/calls.go

@@ -1,6 +1,7 @@
 package compiler
 
 import (
+	"fmt"
 	"golang.org/x/tools/go/ssa"
 	"tinygo.org/x/go-llvm"
 )
@@ -20,6 +21,9 @@ func (c *Compiler) createRuntimeCall(fnName string, args []llvm.Value, name stri
 		panic("trying to call runtime." + fnName)
 	}
 	fn := c.ir.GetFunction(member.(*ssa.Function))
+	if fn.LLVMFn.IsNil() {
+		panic(fmt.Errorf("function %s does not appear in LLVM IR", fnName))
+	}
 	if !fn.IsExported() {
 		args = append(args, llvm.Undef(c.i8ptrType))            // unused context parameter
 		args = append(args, llvm.ConstPointerNull(c.i8ptrType)) // coroutine handle

compiler/channel.go

@@ -4,32 +4,18 @@ package compiler
 // or pseudo-operations that are lowered during goroutine lowering.
 
 import (
-	"fmt"
 	"go/types"
 
+	"github.com/tinygo-org/tinygo/compiler/llvmutil"
 	"golang.org/x/tools/go/ssa"
 	"tinygo.org/x/go-llvm"
 )
 
-// emitMakeChan returns a new channel value for the given channel type.
-func (c *Compiler) emitMakeChan(expr *ssa.MakeChan) (llvm.Value, error) {
-	chanType := c.getLLVMType(expr.Type())
-	size := c.targetData.TypeAllocSize(chanType.ElementType())
-	sizeValue := llvm.ConstInt(c.uintptrType, size, false)
-	ptr := c.createRuntimeCall("alloc", []llvm.Value{sizeValue}, "chan.alloc")
-	ptr = c.builder.CreateBitCast(ptr, chanType, "chan")
-	// Set the elementSize field
-	elementSizePtr := c.builder.CreateGEP(ptr, []llvm.Value{
-		llvm.ConstInt(c.ctx.Int32Type(), 0, false),
-		llvm.ConstInt(c.ctx.Int32Type(), 0, false),
-	}, "")
+func (c *Compiler) emitMakeChan(frame *Frame, expr *ssa.MakeChan) llvm.Value {
 	elementSize := c.targetData.TypeAllocSize(c.getLLVMType(expr.Type().(*types.Chan).Elem()))
-	if elementSize > 0xffff {
-		return ptr, c.makeError(expr.Pos(), fmt.Sprintf("element size is %d bytes, which is bigger than the maximum of %d bytes", elementSize, 0xffff))
-	}
-	elementSizeValue := llvm.ConstInt(c.ctx.Int16Type(), elementSize, false)
-	c.builder.CreateStore(elementSizeValue, elementSizePtr)
-	return ptr, nil
+	elementSizeValue := llvm.ConstInt(c.uintptrType, elementSize, false)
+	bufSize := c.getValue(frame, expr.Size)
+	return c.createRuntimeCall("chanMake", []llvm.Value{elementSizeValue, bufSize}, "")
 }
 
 // emitChanSend emits a pseudo chan send operation. It is lowered to the actual
@@ -44,8 +30,7 @@ func (c *Compiler) emitChanSend(frame *Frame, instr *ssa.Send) {
 	c.builder.CreateStore(chanValue, valueAlloca)
 
 	// Do the send.
-	coroutine := c.createRuntimeCall("getCoroutine", nil, "")
-	c.createRuntimeCall("chanSend", []llvm.Value{coroutine, ch, valueAllocaCast}, "")
+	c.createRuntimeCall("chanSend", []llvm.Value{ch, valueAllocaCast}, "")
 
 	// End the lifetime of the alloca.
 	// This also works around a bug in CoroSplit, at least in LLVM 8:
@@ -63,14 +48,11 @@ func (c *Compiler) emitChanRecv(frame *Frame, unop *ssa.UnOp) llvm.Value {
 	valueAlloca, valueAllocaCast, valueAllocaSize := c.createTemporaryAlloca(valueType, "chan.value")
 
 	// Do the receive.
-	coroutine := c.createRuntimeCall("getCoroutine", nil, "")
-	c.createRuntimeCall("chanRecv", []llvm.Value{coroutine, ch, valueAllocaCast}, "")
+	commaOk := c.createRuntimeCall("chanRecv", []llvm.Value{ch, valueAllocaCast}, "")
 	received := c.builder.CreateLoad(valueAlloca, "chan.received")
 	c.emitLifetimeEnd(valueAllocaCast, valueAllocaSize)
 
 	if unop.CommaOk {
-		commaOk := c.createRuntimeCall("getTaskStateData", []llvm.Value{coroutine}, "chan.commaOk.wide")
-		commaOk = c.builder.CreateTrunc(commaOk, c.ctx.Int1Type(), "chan.commaOk")
 		tuple := llvm.Undef(c.ctx.StructType([]llvm.Type{valueType, c.ctx.Int1Type()}, false))
 		tuple = c.builder.CreateInsertValue(tuple, received, 0, "")
 		tuple = c.builder.CreateInsertValue(tuple, commaOk, 1, "")
@@ -141,7 +123,7 @@ func (c *Compiler) emitSelect(frame *Frame, expr *ssa.Select) llvm.Value {
 			// Store this value in an alloca and put a pointer to this alloca
 			// in the send state.
 			sendValue := c.getValue(frame, state.Send)
-			alloca := c.createEntryBlockAlloca(sendValue.Type(), "select.send.value")
+			alloca := llvmutil.CreateEntryBlockAlloca(c.builder, sendValue.Type(), "select.send.value")
 			c.builder.CreateStore(sendValue, alloca)
 			ptr := c.builder.CreateBitCast(alloca, c.i8ptrType, "")
 			selectState = c.builder.CreateInsertValue(selectState, ptr, 1, "")
@@ -155,7 +137,7 @@ func (c *Compiler) emitSelect(frame *Frame, expr *ssa.Select) llvm.Value {
 	recvbuf := llvm.Undef(c.i8ptrType)
 	if hasReceives {
 		allocaType := llvm.ArrayType(c.ctx.Int8Type(), int(recvbufSize))
-		recvbufAlloca := c.builder.CreateAlloca(allocaType, "select.recvbuf.alloca")
+		recvbufAlloca, _, _ := c.createTemporaryAlloca(allocaType, "select.recvbuf.alloca")
 		recvbufAlloca.SetAlignment(recvbufAlign)
 		recvbuf = c.builder.CreateGEP(recvbufAlloca, []llvm.Value{
 			llvm.ConstInt(c.ctx.Int32Type(), 0, false),
@@ -165,7 +147,7 @@ func (c *Compiler) emitSelect(frame *Frame, expr *ssa.Select) llvm.Value {
 
 	// Create the states slice (allocated on the stack).
 	statesAllocaType := llvm.ArrayType(chanSelectStateType, len(selectStates))
-	statesAlloca := c.builder.CreateAlloca(statesAllocaType, "select.states.alloca")
+	statesAlloca, statesI8, statesSize := c.createTemporaryAlloca(statesAllocaType, "select.states.alloca")
 	for i, state := range selectStates {
 		// Set each slice element to the appropriate channel.
 		gep := c.builder.CreateGEP(statesAlloca, []llvm.Value{
@@ -180,19 +162,36 @@ func (c *Compiler) emitSelect(frame *Frame, expr *ssa.Select) llvm.Value {
 	}, "select.states")
 	statesLen := llvm.ConstInt(c.uintptrType, uint64(len(selectStates)), false)
 
-	// Convert the 'blocking' flag on this select into a LLVM value.
-	blockingInt := uint64(0)
-	if expr.Blocking {
-		blockingInt = 1
-	}
-	blockingValue := llvm.ConstInt(c.ctx.Int1Type(), blockingInt, false)
-
 	// Do the select in the runtime.
-	results := c.createRuntimeCall("chanSelect", []llvm.Value{
-		recvbuf,
-		statesPtr, statesLen, statesLen, // []chanSelectState
-		blockingValue,
-	}, "")
+	var results llvm.Value
+	if expr.Blocking {
+		// Stack-allocate operation structures.
+		// If these were simply created as a slice, they would heap-allocate.
+		chBlockAllocaType := llvm.ArrayType(c.getLLVMRuntimeType("channelBlockedList"), len(selectStates))
+		chBlockAlloca, chBlockAllocaPtr, chBlockSize := c.createTemporaryAlloca(chBlockAllocaType, "select.block.alloca")
+		chBlockLen := llvm.ConstInt(c.uintptrType, uint64(len(selectStates)), false)
+		chBlockPtr := c.builder.CreateGEP(chBlockAlloca, []llvm.Value{
+			llvm.ConstInt(c.ctx.Int32Type(), 0, false),
+			llvm.ConstInt(c.ctx.Int32Type(), 0, false),
+		}, "select.block")
+
+		results = c.createRuntimeCall("chanSelect", []llvm.Value{
+			recvbuf,
+			statesPtr, statesLen, statesLen, // []chanSelectState
+			chBlockPtr, chBlockLen, chBlockLen, // []channelBlockList
+		}, "select.result")
+
+		// Terminate the lifetime of the operation structures.
+		c.emitLifetimeEnd(chBlockAllocaPtr, chBlockSize)
+	} else {
+		results = c.createRuntimeCall("tryChanSelect", []llvm.Value{
+			recvbuf,
+			statesPtr, statesLen, statesLen, // []chanSelectState
+		}, "select.result")
+	}
+
+	// Terminate the lifetime of the states alloca.
+	c.emitLifetimeEnd(statesI8, statesSize)
 
 	// The result value does not include all the possible received values,
 	// because we can't load them in advance. Instead, the *ssa.Extract

compiler/check.go

@@ -10,14 +10,10 @@ import (
 	"tinygo.org/x/go-llvm"
 )
 
-func (c *Compiler) checkType(t llvm.Type, checked map[llvm.Type]struct{}, specials map[llvm.TypeKind]llvm.Type) {
-	if t.IsNil() {
-		panic(t)
-	}
-
+func (c *Compiler) checkType(t llvm.Type, checked map[llvm.Type]struct{}, specials map[llvm.TypeKind]llvm.Type) error {
 	// prevent infinite recursion for self-referential types
 	if _, ok := checked[t]; ok {
-		return
+		return nil
 	}
 	checked[t] = struct{}{}
 
@@ -27,10 +23,10 @@ func (c *Compiler) checkType(t llvm.Type, checked map[llvm.Type]struct{}, specia
 		// this is correct
 	case t.Context() == llvm.GlobalContext():
 		// somewhere we accidentally used the global context instead of a real context
-		panic(fmt.Errorf("type %q uses global context", t.String()))
+		return fmt.Errorf("type %q uses global context", t.String())
 	default:
 		// we used some other context by accident
-		panic(fmt.Errorf("type %q uses context %v instead of the main context %v", t.Context(), c.ctx))
+		return fmt.Errorf("type %q uses context %v instead of the main context %v", t.Context(), c.ctx)
 	}
 
 	// if this is a composite type, check the components of the type
@@ -40,7 +36,7 @@ func (c *Compiler) checkType(t llvm.Type, checked map[llvm.Type]struct{}, specia
 		if s, ok := specials[t.TypeKind()]; !ok {
 			specials[t.TypeKind()] = t
 		} else if s != t {
-			panic(fmt.Errorf("duplicate special type %q: %v and %v", t.TypeKind().String(), t, s))
+			return fmt.Errorf("duplicate special type %q: %v and %v", t.TypeKind().String(), t, s)
 		}
 	case llvm.FloatTypeKind, llvm.DoubleTypeKind, llvm.X86_FP80TypeKind, llvm.FP128TypeKind, llvm.PPC_FP128TypeKind:
 		// floating point numbers are primitives - nothing to recurse
@@ -48,81 +44,129 @@ func (c *Compiler) checkType(t llvm.Type, checked map[llvm.Type]struct{}, specia
 		// integers are primitives - nothing to recurse
 	case llvm.FunctionTypeKind:
 		// check arguments and return(s)
-		for _, v := range t.ParamTypes() {
-			c.checkType(v, checked, specials)
+		for i, v := range t.ParamTypes() {
+			if err := c.checkType(v, checked, specials); err != nil {
+				return fmt.Errorf("failed to verify argument %d of type %s: %s", i, t.String(), err.Error())
+			}
+		}
+		if err := c.checkType(t.ReturnType(), checked, specials); err != nil {
+			return fmt.Errorf("failed to verify return type of type %s: %s", t.String(), err.Error())
 		}
-		c.checkType(t.ReturnType(), checked, specials)
 	case llvm.StructTypeKind:
 		// check all elements
-		for _, v := range t.StructElementTypes() {
-			c.checkType(v, checked, specials)
+		for i, v := range t.StructElementTypes() {
+			if err := c.checkType(v, checked, specials); err != nil {
+				return fmt.Errorf("failed to verify type of field %d of struct type %s: %s", i, t.String(), err.Error())
+			}
 		}
 	case llvm.ArrayTypeKind:
 		// check element type
-		c.checkType(t.ElementType(), checked, specials)
+		if err := c.checkType(t.ElementType(), checked, specials); err != nil {
+			return fmt.Errorf("failed to verify element type of array type %s: %s", t.String(), err.Error())
+		}
 	case llvm.PointerTypeKind:
 		// check underlying type
-		c.checkType(t.ElementType(), checked, specials)
+		if err := c.checkType(t.ElementType(), checked, specials); err != nil {
+			return fmt.Errorf("failed to verify underlying type of pointer type %s: %s", t.String(), err.Error())
+		}
 	case llvm.VectorTypeKind:
 		// check element type
-		c.checkType(t.ElementType(), checked, specials)
+		if err := c.checkType(t.ElementType(), checked, specials); err != nil {
+			return fmt.Errorf("failed to verify element type of vector type %s: %s", t.String(), err.Error())
+		}
+	default:
+		return fmt.Errorf("unrecognized kind %q of type %s", t.TypeKind(), t.String())
 	}
+
+	return nil
 }
 
-func (c *Compiler) checkValue(v llvm.Value, types map[llvm.Type]struct{}, specials map[llvm.TypeKind]llvm.Type) {
+func (c *Compiler) checkValue(v llvm.Value, types map[llvm.Type]struct{}, specials map[llvm.TypeKind]llvm.Type) error {
 	// check type
-	c.checkType(v.Type(), types, specials)
+	if err := c.checkType(v.Type(), types, specials); err != nil {
+		return fmt.Errorf("failed to verify type of value: %s", err.Error())
+	}
+
+	// check if this is an undefined void
+	if v.IsUndef() && v.Type().TypeKind() == llvm.VoidTypeKind {
+		return errors.New("encountered undefined void value")
+	}
+
+	return nil
 }
 
-func (c *Compiler) checkInstruction(inst llvm.Value, types map[llvm.Type]struct{}, specials map[llvm.TypeKind]llvm.Type) {
+func (c *Compiler) checkInstruction(inst llvm.Value, types map[llvm.Type]struct{}, specials map[llvm.TypeKind]llvm.Type) error {
 	// check value properties
-	c.checkValue(inst, types, specials)
+	if err := c.checkValue(inst, types, specials); err != nil {
+		return errorAt(inst, err.Error())
+	}
 
 	// check operands
 	for i := 0; i < inst.OperandsCount(); i++ {
-		c.checkValue(inst.Operand(i), types, specials)
+		if err := c.checkValue(inst.Operand(i), types, specials); err != nil {
+			return errorAt(inst, fmt.Sprintf("failed to validate operand %d of instruction %q: %s", i, inst.Name(), err.Error()))
+		}
 	}
+
+	return nil
 }
 
-func (c *Compiler) checkBasicBlock(bb llvm.BasicBlock, types map[llvm.Type]struct{}, specials map[llvm.TypeKind]llvm.Type) {
+func (c *Compiler) checkBasicBlock(bb llvm.BasicBlock, types map[llvm.Type]struct{}, specials map[llvm.TypeKind]llvm.Type) []error {
 	// check basic block value and type
-	c.checkValue(bb.AsValue(), types, specials)
+	var errs []error
+	if err := c.checkValue(bb.AsValue(), types, specials); err != nil {
+		errs = append(errs, errorAt(bb.Parent(), fmt.Sprintf("failed to validate value of basic block %s: %v", bb.AsValue().Name(), err)))
+	}
 
 	// check instructions
 	for inst := bb.FirstInstruction(); !inst.IsNil(); inst = llvm.NextInstruction(inst) {
-		c.checkInstruction(inst, types, specials)
+		if err := c.checkInstruction(inst, types, specials); err != nil {
+			errs = append(errs, err)
+		}
 	}
+
+	return errs
 }
 
-func (c *Compiler) checkFunction(fn llvm.Value, types map[llvm.Type]struct{}, specials map[llvm.TypeKind]llvm.Type) {
+func (c *Compiler) checkFunction(fn llvm.Value, types map[llvm.Type]struct{}, specials map[llvm.TypeKind]llvm.Type) []error {
 	// check function value and type
-	c.checkValue(fn, types, specials)
+	var errs []error
+	if err := c.checkValue(fn, types, specials); err != nil {
+		errs = append(errs, fmt.Errorf("failed to validate value of function %s: %s", fn.Name(), err.Error()))
+	}
 
 	// check basic blocks
 	for bb := fn.FirstBasicBlock(); !bb.IsNil(); bb = llvm.NextBasicBlock(bb) {
-		c.checkBasicBlock(bb, types, specials)
+		errs = append(errs, c.checkBasicBlock(bb, types, specials)...)
 	}
+
+	return errs
 }
 
-func (c *Compiler) checkModule() {
+func (c *Compiler) checkModule() []error {
 	// check for any context mismatches
+	var errs []error
 	switch {
 	case c.mod.Context() == c.ctx:
 		// this is correct
 	case c.mod.Context() == llvm.GlobalContext():
 		// somewhere we accidentally used the global context instead of a real context
-		panic(errors.New("module uses global context"))
+		errs = append(errs, errors.New("module uses global context"))
 	default:
 		// we used some other context by accident
-		panic(fmt.Errorf("module uses context %v instead of the main context %v", c.mod.Context(), c.ctx))
+		errs = append(errs, fmt.Errorf("module uses context %v instead of the main context %v", c.mod.Context(), c.ctx))
 	}
 
 	types := map[llvm.Type]struct{}{}
 	specials := map[llvm.TypeKind]llvm.Type{}
 	for fn := c.mod.FirstFunction(); !fn.IsNil(); fn = llvm.NextFunction(fn) {
-		c.checkFunction(fn, types, specials)
+		errs = append(errs, c.checkFunction(fn, types, specials)...)
 	}
 	for g := c.mod.FirstGlobal(); !g.IsNil(); g = llvm.NextGlobal(g) {
-		c.checkValue(g, types, specials)
+		if err := c.checkValue(g, types, specials); err != nil {
+			errs = append(errs, fmt.Errorf("failed to verify global %s of module: %s", g.Name(), err.Error()))
+		}
 	}
+
+	return errs
 }

compiler/compiler.go

@@ -1,6 +1,7 @@
 package compiler
 
 import (
+	"debug/dwarf"
 	"errors"
 	"fmt"
 	"go/ast"
@@ -13,6 +14,9 @@ import (
 	"strconv"
 	"strings"
 
+	"github.com/tinygo-org/tinygo/compileopts"
+	"github.com/tinygo-org/tinygo/compiler/llvmutil"
+	"github.com/tinygo-org/tinygo/goenv"
 	"github.com/tinygo-org/tinygo/ir"
 	"github.com/tinygo-org/tinygo/loader"
 	"golang.org/x/tools/go/ssa"
@@ -36,51 +40,35 @@ const tinygoPath = "github.com/tinygo-org/tinygo"
 var functionsUsedInTransforms = []string{
 	"runtime.alloc",
 	"runtime.free",
-	"runtime.sleepTask",
-	"runtime.sleepCurrentTask",
-	"runtime.setTaskStatePtr",
-	"runtime.getTaskStatePtr",
-	"runtime.activateTask",
 	"runtime.scheduler",
+	"runtime.nilPanic",
+}
+
+var taskFunctionsUsedInTransforms = []string{
 	"runtime.startGoroutine",
 }
 
-// Configure the compiler.
-type Config struct {
-	Triple        string   // LLVM target triple, e.g. x86_64-unknown-linux-gnu (empty string means default)
-	CPU           string   // LLVM CPU name, e.g. atmega328p (empty string means default)
-	Features      []string // LLVM CPU features
-	GOOS          string   //
-	GOARCH        string   //
-	GC            string   // garbage collection strategy
-	Scheduler     string   // scheduler implementation ("coroutines" or "tasks")
-	PanicStrategy string   // panic strategy ("print" or "trap")
-	CFlags        []string // cflags to pass to cgo
-	LDFlags       []string // ldflags to pass to cgo
-	ClangHeaders  string   // Clang built-in header include path
-	DumpSSA       bool     // dump Go SSA, for compiler debugging
-	VerifyIR      bool     // run extra checks on the IR
-	Debug         bool     // add debug symbols for gdb
-	GOROOT        string   // GOROOT
-	TINYGOROOT    string   // GOROOT for TinyGo
-	GOPATH        string   // GOPATH, like `go env GOPATH`
-	BuildTags     []string // build tags for TinyGo (empty means {Config.GOOS/Config.GOARCH})
-	TestConfig    TestConfig
-}
-
-type TestConfig struct {
-	CompileTestBinary bool
-	// TODO: Filter the test functions to run, include verbose flag, etc
+var coroFunctionsUsedInTransforms = []string{
+	"runtime.avrSleep",
+	"runtime.getFakeCoroutine",
+	"runtime.setTaskStatePtr",
+	"runtime.getTaskStatePtr",
+	"runtime.activateTask",
+	"runtime.noret",
+	"runtime.getParentHandle",
+	"runtime.getCoroutine",
+	"runtime.llvmCoroRefHolder",
 }
 
 type Compiler struct {
-	Config
+	*compileopts.Config
 	mod                     llvm.Module
 	ctx                     llvm.Context
 	builder                 llvm.Builder
 	dibuilder               *llvm.DIBuilder
 	cu                      llvm.Metadata
 	difiles                 map[string]llvm.Metadata
+	ditypes                 map[types.Type]llvm.Metadata
 	machine                 llvm.TargetMachine
 	targetData              llvm.TargetData
 	intType                 llvm.Type
@@ -116,35 +104,27 @@ type Phi struct {
 	llvm llvm.Value
 }
 
-func NewCompiler(pkgName string, config Config) (*Compiler, error) {
-	if config.Triple == "" {
-		config.Triple = llvm.DefaultTargetTriple()
-	}
-	if len(config.BuildTags) == 0 {
-		config.BuildTags = []string{config.GOOS, config.GOARCH}
-	}
+func NewCompiler(pkgName string, config *compileopts.Config) (*Compiler, error) {
 	c := &Compiler{
 		Config:  config,
 		difiles: make(map[string]llvm.Metadata),
+		ditypes: make(map[types.Type]llvm.Metadata),
 	}
 
-	target, err := llvm.GetTargetFromTriple(config.Triple)
+	target, err := llvm.GetTargetFromTriple(config.Triple())
 	if err != nil {
 		return nil, err
 	}
-	features := ""
-	if len(config.Features) > 0 {
-		features = strings.Join(config.Features, `,`)
-	}
-	c.machine = target.CreateTargetMachine(config.Triple, config.CPU, features, llvm.CodeGenLevelDefault, llvm.RelocStatic, llvm.CodeModelDefault)
+	features := strings.Join(config.Features(), ",")
+	c.machine = target.CreateTargetMachine(config.Triple(), config.CPU(), features, llvm.CodeGenLevelDefault, llvm.RelocStatic, llvm.CodeModelDefault)
 	c.targetData = c.machine.CreateTargetData()
 
 	c.ctx = llvm.NewContext()
 	c.mod = c.ctx.NewModule(pkgName)
-	c.mod.SetTarget(config.Triple)
+	c.mod.SetTarget(config.Triple())
 	c.mod.SetDataLayout(c.targetData.String())
 	c.builder = c.ctx.NewBuilder()
-	if c.Debug {
+	if c.Debug() {
 		c.dibuilder = llvm.NewDIBuilder(c.mod)
 	}
 
@@ -177,28 +157,18 @@ func (c *Compiler) Module() llvm.Module {
 	return c.mod
 }
 
-// Return the LLVM target data object. Only valid after a successful compile.
-func (c *Compiler) TargetData() llvm.TargetData {
-	return c.targetData
-}
-
-// selectGC picks an appropriate GC strategy if none was provided.
-func (c *Compiler) selectGC() string {
-	if c.GC != "" {
-		return c.GC
+// getFunctionsUsedInTransforms gets a list of all special functions that should be preserved during transforms and optimization.
+func (c *Compiler) getFunctionsUsedInTransforms() []string {
+	fnused := functionsUsedInTransforms
+	switch c.Scheduler() {
+	case "coroutines":
+		fnused = append(append([]string{}, fnused...), coroFunctionsUsedInTransforms...)
+	case "tasks":
+		fnused = append(append([]string{}, fnused...), taskFunctionsUsedInTransforms...)
+	default:
+		panic(fmt.Errorf("invalid scheduler %q", c.Scheduler()))
 	}
-	return "conservative"
-}
-
-// selectScheduler picks an appropriate scheduler for the target if none was
-// given.
-func (c *Compiler) selectScheduler() string {
-	if c.Scheduler != "" {
-		// A scheduler was specified in the target description.
-		return c.Scheduler
-	}
-	// Fall back to coroutines, which are supported everywhere.
-	return "coroutines"
+	return fnused
 }
 
 // Compile the given package path or .go file path. Return an error when this
@@ -206,38 +176,37 @@ func (c *Compiler) selectScheduler() string {
 func (c *Compiler) Compile(mainPath string) []error {
 	// Prefix the GOPATH with the system GOROOT, as GOROOT is already set to
 	// the TinyGo root.
-	overlayGopath := c.GOPATH
+	overlayGopath := goenv.Get("GOPATH")
 	if overlayGopath == "" {
-		overlayGopath = c.GOROOT
+		overlayGopath = goenv.Get("GOROOT")
 	} else {
-		overlayGopath = c.GOROOT + string(filepath.ListSeparator) + overlayGopath
+		overlayGopath = goenv.Get("GOROOT") + string(filepath.ListSeparator) + overlayGopath
 	}
 
 	wd, err := os.Getwd()
 	if err != nil {
 		return []error{err}
 	}
-	buildTags := append([]string{"tinygo", "gc." + c.selectGC(), "scheduler." + c.selectScheduler()}, c.BuildTags...)
 	lprogram := &loader.Program{
 		Build: &build.Context{
-			GOARCH:      c.GOARCH,
-			GOOS:        c.GOOS,
-			GOROOT:      c.GOROOT,
-			GOPATH:      c.GOPATH,
+			GOARCH:      c.GOARCH(),
+			GOOS:        c.GOOS(),
+			GOROOT:      goenv.Get("GOROOT"),
+			GOPATH:      goenv.Get("GOPATH"),
 			CgoEnabled:  true,
 			UseAllFiles: false,
 			Compiler:    "gc", // must be one of the recognized compilers
-			BuildTags:   buildTags,
+			BuildTags:   c.BuildTags(),
 		},
 		OverlayBuild: &build.Context{
-			GOARCH:      c.GOARCH,
-			GOOS:        c.GOOS,
-			GOROOT:      c.TINYGOROOT,
+			GOARCH:      c.GOARCH(),
+			GOOS:        c.GOOS(),
+			GOROOT:      goenv.Get("TINYGOROOT"),
 			GOPATH:      overlayGopath,
 			CgoEnabled:  true,
 			UseAllFiles: false,
 			Compiler:    "gc", // must be one of the recognized compilers
-			BuildTags:   buildTags,
+			BuildTags:   c.BuildTags(),
 		},
 		OverlayPath: func(path string) string {
 			// Return the (overlay) import path when it should be overlaid, and
@@ -248,13 +217,13 @@ func (c *Compiler) Compile(mainPath string) []error {
 				path = path[len(tinygoPath+"/src/"):]
 			}
 			switch path {
-			case "machine", "os", "reflect", "runtime", "runtime/volatile", "sync", "testing":
+			case "machine", "os", "reflect", "runtime", "runtime/interrupt", "runtime/volatile", "sync", "testing", "internal/reflectlite":
 				return path
 			default:
 				if strings.HasPrefix(path, "device/") || strings.HasPrefix(path, "examples/") {
 					return path
 				} else if path == "syscall" {
-					for _, tag := range c.BuildTags {
+					for _, tag := range c.BuildTags() {
 						if tag == "baremetal" || tag == "darwin" {
 							return path
 						}
@@ -271,8 +240,8 @@ func (c *Compiler) Compile(mainPath string) []error {
 			},
 		},
 		Dir:          wd,
-		TINYGOROOT:   c.TINYGOROOT,
-		CFlags:       c.CFlags,
+		TINYGOROOT:   goenv.Get("TINYGOROOT"),
+		CFlags:       c.CFlags(),
 		ClangHeaders: c.ClangHeaders,
 	}
 
@@ -282,13 +251,17 @@ func (c *Compiler) Compile(mainPath string) []error {
 			return []error{err}
 		}
 	} else {
-		_, err = lprogram.Import(mainPath, wd)
+		_, err = lprogram.Import(mainPath, wd, token.Position{
+			Filename: "build command-line-arguments",
+		})
 		if err != nil {
 			return []error{err}
 		}
 	}
 
-	_, err = lprogram.Import("runtime", "")
+	_, err = lprogram.Import("runtime", "", token.Position{
+		Filename: "build default import",
+	})
 	if err != nil {
 		return []error{err}
 	}
@@ -304,7 +277,7 @@ func (c *Compiler) Compile(mainPath string) []error {
 	c.ir.SimpleDCE()
 
 	// Initialize debug information.
-	if c.Debug {
+	if c.Debug() {
 		c.cu = c.dibuilder.CreateCompileUnit(llvm.DICompileUnit{
 			Language:  0xb, // DW_LANG_C99 (0xc, off-by-one?)
 			File:      mainPath,
@@ -348,7 +321,7 @@ func (c *Compiler) Compile(mainPath string) []error {
 	initFn := c.ir.GetFunction(c.ir.Program.ImportedPackage("runtime").Members["initAll"].(*ssa.Function))
 	initFn.LLVMFn.SetLinkage(llvm.InternalLinkage)
 	initFn.LLVMFn.SetUnnamedAddr(true)
-	if c.Debug {
+	if c.Debug() {
 		difunc := c.attachDebugInfo(initFn)
 		pos := c.ir.Program.Fset.Position(initFn.Pos())
 		c.builder.SetCurrentDebugLocation(uint(pos.Line), uint(pos.Column), difunc, llvm.Metadata{})
@@ -366,10 +339,10 @@ func (c *Compiler) Compile(mainPath string) []error {
 	realMain.SetLinkage(llvm.ExternalLinkage) // keep alive until goroutine lowering
 
 	// Make sure these functions are kept in tact during TinyGo transformation passes.
-	for _, name := range functionsUsedInTransforms {
+	for _, name := range c.getFunctionsUsedInTransforms() {
 		fn := c.mod.NamedFunction(name)
 		if fn.IsNil() {
-			continue
+			panic(fmt.Errorf("missing core function %q", name))
 		}
 		fn.SetLinkage(llvm.ExternalLinkage)
 	}
@@ -412,7 +385,7 @@ func (c *Compiler) Compile(mainPath string) []error {
 	}
 
 	// see: https://reviews.llvm.org/D18355
-	if c.Debug {
+	if c.Debug() {
 		c.mod.AddNamedMetadataOperand("llvm.module.flags",
 			c.ctx.MDNode([]llvm.Metadata{
 				llvm.ConstInt(c.ctx.Int32Type(), 1, false).ConstantAsMetadata(), // Error on mismatch
@@ -525,33 +498,6 @@ func (c *Compiler) getLLVMType(goType types.Type) llvm.Type {
 		for i := 0; i < typ.NumFields(); i++ {
 			members[i] = c.getLLVMType(typ.Field(i).Type())
 		}
-		if len(members) > 2 && typ.Field(0).Name() == "C union" {
-			// Not a normal struct but a C union emitted by cgo.
-			// Such a field name cannot be entered in regular Go code, this must
-			// be manually inserted in the AST so this is safe.
-			maxAlign := 0
-			maxSize := uint64(0)
-			mainType := members[0]
-			for _, member := range members {
-				align := c.targetData.ABITypeAlignment(member)
-				size := c.targetData.TypeAllocSize(member)
-				if align > maxAlign {
-					maxAlign = align
-					mainType = member
-				} else if align == maxAlign && size > maxSize {
-					maxAlign = align
-					maxSize = size
-					mainType = member
-				} else if size > maxSize {
-					maxSize = size
-				}
-			}
-			members = []llvm.Type{mainType}
-			mainTypeSize := c.targetData.TypeAllocSize(mainType)
-			if mainTypeSize < maxSize {
-				members = append(members, llvm.ArrayType(c.ctx.Int8Type(), int(maxSize-mainTypeSize)))
-			}
-		}
 		return c.ctx.StructType(members, false)
 	case *types.Tuple:
 		members := make([]llvm.Type, typ.Len())
@@ -577,6 +523,17 @@ func isPointer(typ types.Type) bool {
 
 // Get the DWARF type for this Go type.
 func (c *Compiler) getDIType(typ types.Type) llvm.Metadata {
+	if md, ok := c.ditypes[typ]; ok {
+		return md
+	}
+	md := c.createDIType(typ)
+	c.ditypes[typ] = md
+	return md
+}
+
+// createDIType creates a new DWARF type. Don't call this function directly,
+// call getDIType instead.
+func (c *Compiler) createDIType(typ types.Type) llvm.Metadata {
 	llvmType := c.getLLVMType(typ)
 	sizeInBytes := c.targetData.TypeAllocSize(llvmType)
 	switch typ := typ.(type) {
@@ -713,14 +670,17 @@ func (c *Compiler) getDIType(typ types.Type) llvm.Metadata {
 			},
 		})
 	case *types.Struct:
+		// Placeholder metadata node, to be replaced afterwards.
+		temporaryMDNode := c.dibuilder.CreateReplaceableCompositeType(llvm.Metadata{}, llvm.DIReplaceableCompositeType{
+			Tag:         dwarf.TagStructType,
+			SizeInBits:  sizeInBytes * 8,
+			AlignInBits: uint32(c.targetData.ABITypeAlignment(llvmType)) * 8,
+		})
+		c.ditypes[typ] = temporaryMDNode
 		elements := make([]llvm.Metadata, typ.NumFields())
 		for i := range elements {
 			field := typ.Field(i)
 			fieldType := field.Type()
-			if _, ok := fieldType.Underlying().(*types.Pointer); ok {
-				// XXX hack to avoid recursive types
-				fieldType = types.Typ[types.UnsafePointer]
-			}
 			llvmField := c.getLLVMType(fieldType)
 			elements[i] = c.dibuilder.CreateMemberType(llvm.Metadata{}, llvm.DIMemberType{
 				Name:         field.Name(),
@@ -730,11 +690,13 @@ func (c *Compiler) getDIType(typ types.Type) llvm.Metadata {
 				Type:         c.getDIType(fieldType),
 			})
 		}
-		return c.dibuilder.CreateStructType(llvm.Metadata{}, llvm.DIStructType{
+		md := c.dibuilder.CreateStructType(llvm.Metadata{}, llvm.DIStructType{
 			SizeInBits:  sizeInBytes * 8,
 			AlignInBits: uint32(c.targetData.ABITypeAlignment(llvmType)) * 8,
 			Elements:    elements,
 		})
+		temporaryMDNode.ReplaceAllUsesWith(md)
+		return md
 	default:
 		panic("unknown type while generating DWARF debug type: " + typ.String())
 	}
@@ -809,14 +771,6 @@ func (c *Compiler) attachDebugInfo(f *ir.Function) llvm.Metadata {
 }
 
 func (c *Compiler) attachDebugInfoRaw(f *ir.Function, llvmFn llvm.Value, suffix, filename string, line int) llvm.Metadata {
-	if _, ok := c.difiles[filename]; !ok {
-		dir, file := filepath.Split(filename)
-		if dir != "" {
-			dir = dir[:len(dir)-1]
-		}
-		c.difiles[filename] = c.dibuilder.CreateFile(file, dir)
-	}
-
 	// Debug info for this function.
 	diparams := make([]llvm.Metadata, 0, len(f.Params))
 	for _, param := range f.Params {
@@ -830,7 +784,7 @@ func (c *Compiler) attachDebugInfoRaw(f *ir.Function, llvmFn llvm.Value, suffix,
 	difunc := c.dibuilder.CreateFunction(c.difiles[filename], llvm.DIFunction{
 		Name:         f.RelString(nil) + suffix,
 		LinkageName:  f.LinkName() + suffix,
-		File:         c.difiles[filename],
+		File:         c.getDIFile(filename),
 		Line:         line,
 		Type:         diFuncType,
 		LocalToUnit:  true,
@@ -843,21 +797,37 @@ func (c *Compiler) attachDebugInfoRaw(f *ir.Function, llvmFn llvm.Value, suffix,
 	return difunc
 }
 
+// getDIFile returns a DIFile metadata node for the given filename. It tries to
+// use one that was already created, otherwise it falls back to creating a new
+// one.
+func (c *Compiler) getDIFile(filename string) llvm.Metadata {
+	if _, ok := c.difiles[filename]; !ok {
+		dir, file := filepath.Split(filename)
+		if dir != "" {
+			dir = dir[:len(dir)-1]
+		}
+		c.difiles[filename] = c.dibuilder.CreateFile(file, dir)
+	}
+	return c.difiles[filename]
+}
+
 func (c *Compiler) parseFunc(frame *Frame) {
-	if c.DumpSSA {
+	if c.DumpSSA() {
 		fmt.Printf("\nfunc %s:\n", frame.fn.Function)
 	}
 	if !frame.fn.LLVMFn.IsDeclaration() {
-		c.addError(frame.fn.Pos(), "function is already defined:"+frame.fn.LLVMFn.Name())
+		errValue := frame.fn.LLVMFn.Name() + " redeclared in this program"
+		fnPos := getPosition(frame.fn.LLVMFn)
+		if fnPos.IsValid() {
+			errValue += "\n\tprevious declaration at " + fnPos.String()
+		}
+		c.addError(frame.fn.Pos(), errValue)
 		return
 	}
 	if !frame.fn.IsExported() {
 		frame.fn.LLVMFn.SetLinkage(llvm.InternalLinkage)
 		frame.fn.LLVMFn.SetUnnamedAddr(true)
 	}
-	if frame.fn.IsInterrupt() && strings.HasPrefix(c.Triple, "avr") {
-		frame.fn.LLVMFn.SetFunctionCallConv(85) // CallingConv::AVR_SIGNAL
-	}
 
 	// Some functions have a pragma controlling the inlining level.
 	switch frame.fn.Inline() {
@@ -872,7 +842,7 @@ func (c *Compiler) parseFunc(frame *Frame) {
 	}
 
 	// Add debug info, if needed.
-	if c.Debug {
+	if c.Debug() {
 		if frame.fn.Synthetic == "package initializer" {
 			// Package initializers have no debug info. Create some fake debug
 			// info to at least have *something*.
@@ -906,7 +876,7 @@ func (c *Compiler) parseFunc(frame *Frame) {
 		frame.locals[param] = c.collapseFormalParam(llvmType, fields)
 
 		// Add debug information to this parameter (if available)
-		if c.Debug && frame.fn.Syntax() != nil {
+		if c.Debug() && frame.fn.Syntax() != nil {
 			pos := c.ir.Program.Fset.Position(frame.fn.Syntax().Pos())
 			diType := c.getDIType(param.Type())
 			dbgParam := c.dibuilder.CreateParameterVariable(frame.difunc, llvm.DIParameterVariable{
@@ -968,7 +938,7 @@ func (c *Compiler) parseFunc(frame *Frame) {
 
 	// Fill blocks with instructions.
 	for _, block := range frame.fn.DomPreorder() {
-		if c.DumpSSA {
+		if c.DumpSSA() {
 			fmt.Printf("%d: %s:\n", block.Index, block.Comment)
 		}
 		c.builder.SetInsertPointAtEnd(frame.blockEntries[block])
@@ -977,7 +947,7 @@ func (c *Compiler) parseFunc(frame *Frame) {
 			if _, ok := instr.(*ssa.DebugRef); ok {
 				continue
 			}
-			if c.DumpSSA {
+			if c.DumpSSA() {
 				if val, ok := instr.(ssa.Value); ok && val.Name() != "" {
 					fmt.Printf("\t%s = %s\n", val.Name(), val.String())
 				} else {
@@ -1003,7 +973,7 @@ func (c *Compiler) parseFunc(frame *Frame) {
 }
 
 func (c *Compiler) parseInstr(frame *Frame, instr ssa.Instruction) {
-	if c.Debug {
+	if c.Debug() {
 		pos := c.ir.Program.Fset.Position(instr.Pos())
 		c.builder.SetCurrentDebugLocation(uint(pos.Line), uint(pos.Column), frame.difunc, llvm.Metadata{})
 	}
@@ -1021,7 +991,7 @@ func (c *Compiler) parseInstr(frame *Frame, instr ssa.Instruction) {
 			frame.locals[instr] = llvm.Undef(c.getLLVMType(instr.Type()))
 		} else {
 			frame.locals[instr] = value
-			if len(*instr.Referrers()) != 0 && c.needsStackObjects() {
+			if len(*instr.Referrers()) != 0 && c.NeedsStackObjects() {
 				c.trackExpr(frame, instr, value)
 			}
 		}
@@ -1041,28 +1011,34 @@ func (c *Compiler) parseInstr(frame *Frame, instr ssa.Instruction) {
 			// Static callee is known. This makes it easier to start a new
 			// goroutine.
 			calleeFn := c.ir.GetFunction(callee)
-			if !calleeFn.IsExported() && c.selectScheduler() != "tasks" {
-				// For coroutine scheduling, this is only required when calling
-				// an external function.
-				// For tasks, because all params are stored in a single object,
-				// no unnecessary parameters should be stored anyway.
-				params = append(params, llvm.Undef(c.i8ptrType))            // context parameter
-				params = append(params, llvm.ConstPointerNull(c.i8ptrType)) // parent coroutine handle
+			var context llvm.Value
+			switch value := instr.Call.Value.(type) {
+			case *ssa.Function:
+				// Goroutine call is regular function call. No context is necessary.
+				context = llvm.Undef(c.i8ptrType)
+			case *ssa.MakeClosure:
+				// A goroutine call on a func value, but the callee is trivial to find. For
+				// example: immediately applied functions.
+				funcValue := c.getValue(frame, value)
+				context = c.extractFuncContext(funcValue)
+			default:
+				panic("StaticCallee returned an unexpected value")
 			}
+			params = append(params, context) // context parameter
 			c.emitStartGoroutine(calleeFn.LLVMFn, params)
 		} else if !instr.Call.IsInvoke() {
 			// This is a function pointer.
 			// At the moment, two extra params are passed to the newly started
 			// goroutine:
 			//   * The function context, for closures.
-			//   * The parent handle (for coroutines) or the function pointer
-			//     itself (for tasks).
+			//   * The function pointer (for tasks).
 			funcPtr, context := c.decodeFuncValue(c.getValue(frame, instr.Call.Value), instr.Call.Value.Type().(*types.Signature))
 			params = append(params, context) // context parameter
-			switch c.selectScheduler() {
+			switch c.Scheduler() {
 			case "coroutines":
-				params = append(params, llvm.ConstPointerNull(c.i8ptrType)) // parent coroutine handle
+				// There are no additional parameters needed for the goroutine start operation.
 			case "tasks":
+				// Add the function pointer as a parameter to start the goroutine.
 				params = append(params, funcPtr)
 			default:
 				panic("unknown scheduler type")
@@ -1333,6 +1309,8 @@ func (c *Compiler) parseCall(frame *Frame, instr *ssa.CallCommon) (llvm.Value, e
 			return c.emitVolatileLoad(frame, instr)
 		case strings.HasPrefix(name, "runtime/volatile.Store"):
 			return c.emitVolatileStore(frame, instr)
+		case name == "runtime/interrupt.New" && len(fn.Blocks) == 0:
+			return c.emitInterruptGlobal(frame, instr)
 		}
 
 		targetFunc := c.ir.GetFunction(fn)
@@ -1423,7 +1401,7 @@ func (c *Compiler) parseExpr(frame *Frame, expr ssa.Value) (llvm.Value, error) {
 			buf = c.builder.CreateBitCast(buf, llvm.PointerType(typ, 0), "")
 			return buf, nil
 		} else {
-			buf := c.createEntryBlockAlloca(typ, expr.Comment)
+			buf := llvmutil.CreateEntryBlockAlloca(c.builder, typ, expr.Comment)
 			if c.targetData.TypeAllocSize(typ) != 0 {
 				c.builder.CreateStore(llvm.ConstNull(typ), buf) // zero-initialize var
 			}
@@ -1488,18 +1466,6 @@ func (c *Compiler) parseExpr(frame *Frame, expr ssa.Value) (llvm.Value, error) {
 		return c.builder.CreateExtractValue(value, expr.Index, ""), nil
 	case *ssa.Field:
 		value := c.getValue(frame, expr.X)
-		if s := expr.X.Type().Underlying().(*types.Struct); s.NumFields() > 2 && s.Field(0).Name() == "C union" {
-			// Extract a field from a CGo union.
-			// This could be done directly, but as this is a very infrequent
-			// operation it's much easier to bitcast it through an alloca.
-			resultType := c.getLLVMType(expr.Type())
-			alloca, allocaPtr, allocaSize := c.createTemporaryAlloca(value.Type(), "union.alloca")
-			c.builder.CreateStore(value, alloca)
-			bitcast := c.builder.CreateBitCast(alloca, llvm.PointerType(resultType, 0), "union.bitcast")
-			result := c.builder.CreateLoad(bitcast, "union.result")
-			c.emitLifetimeEnd(allocaPtr, allocaSize)
-			return result, nil
-		}
 		result := c.builder.CreateExtractValue(value, expr.Field, "")
 		return result, nil
 	case *ssa.FieldAddr:
@@ -1509,20 +1475,12 @@ func (c *Compiler) parseExpr(frame *Frame, expr ssa.Value) (llvm.Value, error) {
 		// > pointer of type *T to x. [...] If the evaluation of x would cause a
 		// > run-time panic, then the evaluation of &x does too.
 		c.emitNilCheck(frame, val, "gep")
-		if s := expr.X.Type().(*types.Pointer).Elem().Underlying().(*types.Struct); s.NumFields() > 2 && s.Field(0).Name() == "C union" {
-			// This is not a regular struct but actually an union.
-			// That simplifies things, as we can just bitcast the pointer to the
-			// right type.
-			ptrType := c.getLLVMType(expr.Type())
-			return c.builder.CreateBitCast(val, ptrType, ""), nil
-		} else {
-			// Do a GEP on the pointer to get the field address.
-			indices := []llvm.Value{
-				llvm.ConstInt(c.ctx.Int32Type(), 0, false),
-				llvm.ConstInt(c.ctx.Int32Type(), uint64(expr.Field), false),
-			}
-			return c.builder.CreateInBoundsGEP(val, indices, ""), nil
+		// Do a GEP on the pointer to get the field address.
+		indices := []llvm.Value{
+			llvm.ConstInt(c.ctx.Int32Type(), 0, false),
+			llvm.ConstInt(c.ctx.Int32Type(), uint64(expr.Field), false),
 		}
+		return c.builder.CreateInBoundsGEP(val, indices, ""), nil
 	case *ssa.Function:
 		panic("function is not an expression")
 	case *ssa.Global:
@@ -1618,7 +1576,7 @@ func (c *Compiler) parseExpr(frame *Frame, expr ssa.Value) (llvm.Value, error) {
 			panic("unknown lookup type: " + expr.String())
 		}
 	case *ssa.MakeChan:
-		return c.emitMakeChan(expr)
+		return c.emitMakeChan(frame, expr), nil
 	case *ssa.MakeClosure:
 		return c.parseMakeClosure(frame, expr)
 	case *ssa.MakeInterface:

compiler/defer.go

@@ -14,6 +14,7 @@ package compiler
 //     frames.
 
 import (
+	"github.com/tinygo-org/tinygo/compiler/llvmutil"
 	"github.com/tinygo-org/tinygo/ir"
 	"golang.org/x/tools/go/ssa"
 	"tinygo.org/x/go-llvm"
@@ -34,6 +35,40 @@ func (c *Compiler) deferInitFunc(frame *Frame) {
 	c.builder.CreateStore(llvm.ConstPointerNull(deferType), frame.deferPtr)
 }
 
+// isInLoop checks if there is a path from a basic block to itself.
+func isInLoop(start *ssa.BasicBlock) bool {
+	// Use a breadth-first search to scan backwards through the block graph.
+	queue := []*ssa.BasicBlock{start}
+	checked := map[*ssa.BasicBlock]struct{}{}
+
+	for len(queue) > 0 {
+		// pop a block off of the queue
+		block := queue[len(queue)-1]
+		queue = queue[:len(queue)-1]
+
+		// Search through predecessors.
+		// Searching backwards means that this is pretty fast when the block is close to the start of the function.
+		// Defers are often placed near the start of the function.
+		for _, pred := range block.Preds {
+			if pred == start {
+				// cycle found
+				return true
+			}
+
+			if _, ok := checked[pred]; ok {
+				// block already checked
+				continue
+			}
+
+			// add to queue and checked map
+			queue = append(queue, pred)
+			checked[pred] = struct{}{}
+		}
+	}
+
+	return false
+}
+
 // emitDefer emits a single defer instruction, to be run when this function
 // returns.
 func (c *Compiler) emitDefer(frame *Frame, instr *ssa.Defer) {
@@ -127,12 +162,22 @@ func (c *Compiler) emitDefer(frame *Frame, instr *ssa.Defer) {
 		deferFrame = c.builder.CreateInsertValue(deferFrame, value, i, "")
 	}
 
-	// Put this struct in an alloca.
-	alloca := c.builder.CreateAlloca(deferFrameType, "defer.alloca")
-	c.builder.CreateStore(deferFrame, alloca)
-	if c.needsStackObjects() {
+	// Put this struct in an allocation.
+	var alloca llvm.Value
+	if !isInLoop(instr.Block()) {
+		// This can safely use a stack allocation.
+		alloca = llvmutil.CreateEntryBlockAlloca(c.builder, deferFrameType, "defer.alloca")
+	} else {
+		// This may be hit a variable number of times, so use a heap allocation.
+		size := c.targetData.TypeAllocSize(deferFrameType)
+		sizeValue := llvm.ConstInt(c.uintptrType, size, false)
+		allocCall := c.createRuntimeCall("alloc", []llvm.Value{sizeValue}, "defer.alloc.call")
+		alloca = c.builder.CreateBitCast(allocCall, llvm.PointerType(deferFrameType, 0), "defer.alloc")
+	}
+	if c.NeedsStackObjects() {
 		c.trackPointer(alloca)
 	}
+	c.builder.CreateStore(deferFrame, alloca)
 
 	// Push it on top of the linked list by replacing deferPtr.
 	allocaCast := c.builder.CreateBitCast(alloca, next.Type(), "defer.alloca.cast")

compiler/errors.go

@@ -1,8 +1,12 @@
 package compiler
 
 import (
+	"go/scanner"
 	"go/token"
 	"go/types"
+	"path/filepath"
+
+	"tinygo.org/x/go-llvm"
 )
 
 func (c *Compiler) makeError(pos token.Pos, msg string) types.Error {
@@ -16,3 +20,42 @@ func (c *Compiler) makeError(pos token.Pos, msg string) types.Error {
 func (c *Compiler) addError(pos token.Pos, msg string) {
 	c.diagnostics = append(c.diagnostics, c.makeError(pos, msg))
 }
+
+// errorAt returns an error value at the location of the instruction.
+// The location information may not be complete as it depends on debug
+// information in the IR.
+func errorAt(inst llvm.Value, msg string) scanner.Error {
+	return scanner.Error{
+		Pos: getPosition(inst),
+		Msg: msg,
+	}
+}
+
+// getPosition returns the position information for the given value, as far as
+// it is available.
+func getPosition(val llvm.Value) token.Position {
+	if !val.IsAInstruction().IsNil() {
+		loc := val.InstructionDebugLoc()
+		if loc.IsNil() {
+			return token.Position{}
+		}
+		file := loc.LocationScope().ScopeFile()
+		return token.Position{
+			Filename: filepath.Join(file.FileDirectory(), file.FileFilename()),
+			Line:     int(loc.LocationLine()),
+			Column:   int(loc.LocationColumn()),
+		}
+	} else if !val.IsAFunction().IsNil() {
+		loc := val.Subprogram()
+		if loc.IsNil() {
+			return token.Position{}
+		}
+		file := loc.ScopeFile()
+		return token.Position{
+			Filename: filepath.Join(file.FileDirectory(), file.FileFilename()),
+			Line:     int(loc.SubprogramLine()),
+		}
+	} else {
+		return token.Position{}
+	}
+}

compiler/func.go

@@ -34,7 +34,7 @@ const (
 func (c *Compiler) funcImplementation() funcValueImplementation {
 	// Always pick the switch implementation, as it allows the use of blocking
 	// inside a function that is used as a func value.
-	switch c.selectScheduler() {
+	switch c.Scheduler() {
 	case "coroutines":
 		return funcValueSwitch
 	case "tasks":

compiler/gc.go

@@ -5,27 +5,11 @@ package compiler
 
 import (
 	"go/token"
-	"math/big"
 
 	"golang.org/x/tools/go/ssa"
 	"tinygo.org/x/go-llvm"
 )
 
-// needsStackObjects returns true if the compiler should insert stack objects
-// that can be traced by the garbage collector.
-func (c *Compiler) needsStackObjects() bool {
-	if c.selectGC() != "conservative" {
-		return false
-	}
-	for _, tag := range c.BuildTags {
-		if tag == "baremetal" {
-			return false
-		}
-	}
-
-	return true
-}
-
 // trackExpr inserts pointer tracking intrinsics for the GC if the expression is
 // one of the expressions that need this.
 func (c *Compiler) trackExpr(frame *Frame, expr ssa.Value, value llvm.Value) {
@@ -120,352 +104,3 @@ func typeHasPointers(t llvm.Type) bool {
 		return false
 	}
 }
-
-// makeGCStackSlots converts all calls to runtime.trackPointer to explicit
-// stores to stack slots that are scannable by the GC.
-func (c *Compiler) makeGCStackSlots() bool {
-	// Check whether there are allocations at all.
-	alloc := c.mod.NamedFunction("runtime.alloc")
-	if alloc.IsNil() {
-		// Nothing to. Make sure all remaining bits and pieces for stack
-		// chains are neutralized.
-		for _, call := range getUses(c.mod.NamedFunction("runtime.trackPointer")) {
-			call.EraseFromParentAsInstruction()
-		}
-		stackChainStart := c.mod.NamedGlobal("runtime.stackChainStart")
-		if !stackChainStart.IsNil() {
-			stackChainStart.SetInitializer(llvm.ConstNull(stackChainStart.Type().ElementType()))
-			stackChainStart.SetGlobalConstant(true)
-		}
-	}
-
-	trackPointer := c.mod.NamedFunction("runtime.trackPointer")
-	if trackPointer.IsNil() || trackPointer.FirstUse().IsNil() {
-		return false // nothing to do
-	}
-
-	// Look at *all* functions to see whether they are free of function pointer
-	// calls.
-	// This takes less than 5ms for ~100kB of WebAssembly but would perhaps be
-	// faster when written in C++ (to avoid the CGo overhead).
-	funcsWithFPCall := map[llvm.Value]struct{}{}
-	n := 0
-	for fn := c.mod.FirstFunction(); !fn.IsNil(); fn = llvm.NextFunction(fn) {
-		n++
-		if _, ok := funcsWithFPCall[fn]; ok {
-			continue // already found
-		}
-		done := false
-		for bb := fn.FirstBasicBlock(); !bb.IsNil() && !done; bb = llvm.NextBasicBlock(bb) {
-			for call := bb.FirstInstruction(); !call.IsNil() && !done; call = llvm.NextInstruction(call) {
-				if call.IsACallInst().IsNil() {
-					continue // only looking at calls
-				}
-				called := call.CalledValue()
-				if !called.IsAFunction().IsNil() {
-					continue // only looking for function pointers
-				}
-				funcsWithFPCall[fn] = struct{}{}
-				markParentFunctions(funcsWithFPCall, fn)
-				done = true
-			}
-		}
-	}
-
-	// Determine which functions need stack objects. Many leaf functions don't
-	// need it: it only causes overhead for them.
-	// Actually, in one test it was only able to eliminate stack object from 12%
-	// of functions that had a call to runtime.trackPointer (8 out of 68
-	// functions), so this optimization is not as big as it may seem.
-	allocatingFunctions := map[llvm.Value]struct{}{} // set of allocating functions
-
-	// Work from runtime.alloc and trace all parents to check which functions do
-	// a heap allocation (and thus which functions do not).
-	markParentFunctions(allocatingFunctions, alloc)
-
-	// Also trace all functions that call a function pointer.
-	for fn := range funcsWithFPCall {
-		// Assume that functions that call a function pointer do a heap
-		// allocation as a conservative guess because the called function might
-		// do a heap allocation.
-		allocatingFunctions[fn] = struct{}{}
-		markParentFunctions(allocatingFunctions, fn)
-	}
-
-	// Collect some variables used below in the loop.
-	stackChainStart := c.mod.NamedGlobal("runtime.stackChainStart")
-	if stackChainStart.IsNil() {
-		panic("stack chain start not found!")
-	}
-	stackChainStartType := stackChainStart.Type().ElementType()
-	stackChainStart.SetInitializer(llvm.ConstNull(stackChainStartType))
-
-	// Iterate until runtime.trackPointer has no uses left.
-	for use := trackPointer.FirstUse(); !use.IsNil(); use = trackPointer.FirstUse() {
-		// Pick the first use of runtime.trackPointer.
-		call := use.User()
-		if call.IsACallInst().IsNil() {
-			panic("expected runtime.trackPointer use to be a call")
-		}
-
-		// Pick the parent function.
-		fn := call.InstructionParent().Parent()
-
-		if _, ok := allocatingFunctions[fn]; !ok {
-			// This function nor any of the functions it calls (recursively)
-			// allocate anything from the heap, so it will not trigger a garbage
-			// collection cycle. Thus, it does not need to track local pointer
-			// values.
-			// This is a useful optimization but not as big as you might guess,
-			// as described above (it avoids stack objects for ~12% of
-			// functions).
-			call.EraseFromParentAsInstruction()
-			continue
-		}
-
-		// Find all calls to runtime.trackPointer in this function.
-		var calls []llvm.Value
-		var returns []llvm.Value
-		for bb := fn.FirstBasicBlock(); !bb.IsNil(); bb = llvm.NextBasicBlock(bb) {
-			for inst := bb.FirstInstruction(); !inst.IsNil(); inst = llvm.NextInstruction(inst) {
-				switch inst.InstructionOpcode() {
-				case llvm.Call:
-					if inst.CalledValue() == trackPointer {
-						calls = append(calls, inst)
-					}
-				case llvm.Ret:
-					returns = append(returns, inst)
-				}
-			}
-		}
-
-		// Determine what to do with each call.
-		var allocas, pointers []llvm.Value
-		for _, call := range calls {
-			ptr := call.Operand(0)
-			call.EraseFromParentAsInstruction()
-			if ptr.IsAInstruction().IsNil() {
-				continue
-			}
-
-			// Some trivial optimizations.
-			if ptr.IsAInstruction().IsNil() {
-				continue
-			}
-			switch ptr.InstructionOpcode() {
-			case llvm.PHI, llvm.GetElementPtr:
-				// These values do not create new values: the values already
-				// existed locally in this function so must have been tracked
-				// already.
-				continue
-			case llvm.ExtractValue, llvm.BitCast:
-				// These instructions do not create new values, but their
-				// original value may not be tracked. So keep tracking them for
-				// now.
-				// With more analysis, it should be possible to optimize a
-				// significant chunk of these away.
-			case llvm.Call, llvm.Load, llvm.IntToPtr:
-				// These create new values so must be stored locally. But
-				// perhaps some of these can be fused when they actually refer
-				// to the same value.
-			default:
-				// Ambiguous. These instructions are uncommon, but perhaps could
-				// be optimized if needed.
-			}
-
-			if !ptr.IsAAllocaInst().IsNil() {
-				if typeHasPointers(ptr.Type().ElementType()) {
-					allocas = append(allocas, ptr)
-				}
-			} else {
-				pointers = append(pointers, ptr)
-			}
-		}
-
-		if len(allocas) == 0 && len(pointers) == 0 {
-			// This function does not need to keep track of stack pointers.
-			continue
-		}
-
-		// Determine the type of the required stack slot.
-		fields := []llvm.Type{
-			stackChainStartType, // Pointer to parent frame.
-			c.uintptrType,       // Number of elements in this frame.
-		}
-		for _, alloca := range allocas {
-			fields = append(fields, alloca.Type().ElementType())
-		}
-		for _, ptr := range pointers {
-			fields = append(fields, ptr.Type())
-		}
-		stackObjectType := c.ctx.StructType(fields, false)
-
-		// Create the stack object at the function entry.
-		c.builder.SetInsertPointBefore(fn.EntryBasicBlock().FirstInstruction())
-		stackObject := c.builder.CreateAlloca(stackObjectType, "gc.stackobject")
-		initialStackObject := llvm.ConstNull(stackObjectType)
-		numSlots := (c.targetData.TypeAllocSize(stackObjectType) - c.targetData.TypeAllocSize(c.i8ptrType)*2) / uint64(c.targetData.ABITypeAlignment(c.uintptrType))
-		numSlotsValue := llvm.ConstInt(c.uintptrType, numSlots, false)
-		initialStackObject = llvm.ConstInsertValue(initialStackObject, numSlotsValue, []uint32{1})
-		c.builder.CreateStore(initialStackObject, stackObject)
-
-		// Update stack start.
-		parent := c.builder.CreateLoad(stackChainStart, "")
-		gep := c.builder.CreateGEP(stackObject, []llvm.Value{
-			llvm.ConstInt(c.ctx.Int32Type(), 0, false),
-			llvm.ConstInt(c.ctx.Int32Type(), 0, false),
-		}, "")
-		c.builder.CreateStore(parent, gep)
-		stackObjectCast := c.builder.CreateBitCast(stackObject, stackChainStartType, "")
-		c.builder.CreateStore(stackObjectCast, stackChainStart)
-
-		// Replace all independent allocas with GEPs in the stack object.
-		for i, alloca := range allocas {
-			gep := c.builder.CreateGEP(stackObject, []llvm.Value{
-				llvm.ConstInt(c.ctx.Int32Type(), 0, false),
-				llvm.ConstInt(c.ctx.Int32Type(), uint64(2+i), false),
-			}, "")
-			alloca.ReplaceAllUsesWith(gep)
-			alloca.EraseFromParentAsInstruction()
-		}
-
-		// Do a store to the stack object after each new pointer that is created.
-		for i, ptr := range pointers {
-			c.builder.SetInsertPointBefore(llvm.NextInstruction(ptr))
-			gep := c.builder.CreateGEP(stackObject, []llvm.Value{
-				llvm.ConstInt(c.ctx.Int32Type(), 0, false),
-				llvm.ConstInt(c.ctx.Int32Type(), uint64(2+len(allocas)+i), false),
-			}, "")
-			c.builder.CreateStore(ptr, gep)
-		}
-
-		// Make sure this stack object is popped from the linked list of stack
-		// objects at return.
-		for _, ret := range returns {
-			c.builder.SetInsertPointBefore(ret)
-			c.builder.CreateStore(parent, stackChainStart)
-		}
-	}
-
-	return true
-}
-
-func (c *Compiler) addGlobalsBitmap() bool {
-	if c.mod.NamedGlobal("runtime.trackedGlobalsStart").IsNil() {
-		return false // nothing to do: no GC in use
-	}
-
-	var trackedGlobals []llvm.Value
-	var trackedGlobalTypes []llvm.Type
-	for global := c.mod.FirstGlobal(); !global.IsNil(); global = llvm.NextGlobal(global) {
-		if global.IsDeclaration() {
-			continue
-		}
-		typ := global.Type().ElementType()
-		ptrs := c.getPointerBitmap(typ, global.Name())
-		if ptrs.BitLen() == 0 {
-			continue
-		}
-		trackedGlobals = append(trackedGlobals, global)
-		trackedGlobalTypes = append(trackedGlobalTypes, typ)
-	}
-
-	globalsBundleType := c.ctx.StructType(trackedGlobalTypes, false)
-	globalsBundle := llvm.AddGlobal(c.mod, globalsBundleType, "tinygo.trackedGlobals")
-	globalsBundle.SetLinkage(llvm.InternalLinkage)
-	globalsBundle.SetUnnamedAddr(true)
-	initializer := llvm.Undef(globalsBundleType)
-	for i, global := range trackedGlobals {
-		initializer = llvm.ConstInsertValue(initializer, global.Initializer(), []uint32{uint32(i)})
-		gep := llvm.ConstGEP(globalsBundle, []llvm.Value{
-			llvm.ConstInt(c.ctx.Int32Type(), 0, false),
-			llvm.ConstInt(c.ctx.Int32Type(), uint64(i), false),
-		})
-		global.ReplaceAllUsesWith(gep)
-		global.EraseFromParentAsGlobal()
-	}
-	globalsBundle.SetInitializer(initializer)
-
-	trackedGlobalsStart := llvm.ConstPtrToInt(globalsBundle, c.uintptrType)
-	c.mod.NamedGlobal("runtime.trackedGlobalsStart").SetInitializer(trackedGlobalsStart)
-
-	alignment := c.targetData.PrefTypeAlignment(c.i8ptrType)
-	trackedGlobalsLength := llvm.ConstInt(c.uintptrType, c.targetData.TypeAllocSize(globalsBundleType)/uint64(alignment), false)
-	c.mod.NamedGlobal("runtime.trackedGlobalsLength").SetInitializer(trackedGlobalsLength)
-
-	bitmapBytes := c.getPointerBitmap(globalsBundleType, "globals bundle").Bytes()
-	bitmapValues := make([]llvm.Value, len(bitmapBytes))
-	for i, b := range bitmapBytes {
-		bitmapValues[len(bitmapBytes)-i-1] = llvm.ConstInt(c.ctx.Int8Type(), uint64(b), false)
-	}
-	bitmapArray := llvm.ConstArray(c.ctx.Int8Type(), bitmapValues)
-	bitmapNew := llvm.AddGlobal(c.mod, bitmapArray.Type(), "runtime.trackedGlobalsBitmap.tmp")
-	bitmapOld := c.mod.NamedGlobal("runtime.trackedGlobalsBitmap")
-	bitmapOld.ReplaceAllUsesWith(llvm.ConstBitCast(bitmapNew, bitmapOld.Type()))
-	bitmapNew.SetInitializer(bitmapArray)
-	bitmapNew.SetName("runtime.trackedGlobalsBitmap")
-
-	return true // the IR was changed
-}
-
-func (c *Compiler) getPointerBitmap(typ llvm.Type, name string) *big.Int {
-	alignment := c.targetData.PrefTypeAlignment(c.i8ptrType)
-	switch typ.TypeKind() {
-	case llvm.IntegerTypeKind, llvm.FloatTypeKind, llvm.DoubleTypeKind:
-		return big.NewInt(0)
-	case llvm.PointerTypeKind:
-		return big.NewInt(1)
-	case llvm.StructTypeKind:
-		ptrs := big.NewInt(0)
-		for i, subtyp := range typ.StructElementTypes() {
-			subptrs := c.getPointerBitmap(subtyp, name)
-			if subptrs.BitLen() == 0 {
-				continue
-			}
-			offset := c.targetData.ElementOffset(typ, i)
-			if offset%uint64(alignment) != 0 {
-				panic("precise GC: global contains unaligned pointer: " + name)
-			}
-			subptrs.Lsh(subptrs, uint(offset)/uint(alignment))
-			ptrs.Or(ptrs, subptrs)
-		}
-		return ptrs
-	case llvm.ArrayTypeKind:
-		subtyp := typ.ElementType()
-		subptrs := c.getPointerBitmap(subtyp, name)
-		ptrs := big.NewInt(0)
-		if subptrs.BitLen() == 0 {
-			return ptrs
-		}
-		elementSize := c.targetData.TypeAllocSize(subtyp)
-		for i := 0; i < typ.ArrayLength(); i++ {
-			ptrs.Lsh(ptrs, uint(elementSize)/uint(alignment))
-			ptrs.Or(ptrs, subptrs)
-		}
-		return ptrs
-	default:
-		panic("unknown type kind of global: " + name)
-	}
-}
-
-// markParentFunctions traverses all parent function calls (recursively) and
-// adds them to the set of marked functions. It only considers function calls:
-// any other uses of such a function is ignored.
-func markParentFunctions(marked map[llvm.Value]struct{}, fn llvm.Value) {
-	worklist := []llvm.Value{fn}
-	for len(worklist) != 0 {
-		fn := worklist[len(worklist)-1]
-		worklist = worklist[:len(worklist)-1]
-		for _, use := range getUses(fn) {
-			if use.IsACallInst().IsNil() || use.CalledValue() != fn {
-				// Not the parent function.
-				continue
-			}
-			parent := use.InstructionParent().Parent()
-			if _, ok := marked[parent]; !ok {
-				marked[parent] = struct{}{}
-				worklist = append(worklist, parent)
-			}
-		}
-	}
-}

compiler/goroutine.go

@@ -7,10 +7,13 @@ import "tinygo.org/x/go-llvm"
 
 // emitStartGoroutine starts a new goroutine with the provided function pointer
 // and parameters.
+// In general, you should pass all regular parameters plus the context parameter.
+// There is one exception: the task-based scheduler needs to have the function
+// pointer passed in as a parameter too in addition to the context.
 //
 // Because a go statement doesn't return anything, return undef.
 func (c *Compiler) emitStartGoroutine(funcPtr llvm.Value, params []llvm.Value) llvm.Value {
-	switch c.selectScheduler() {
+	switch c.Scheduler() {
 	case "tasks":
 		paramBundle := c.emitPointerPack(params)
 		paramBundle = c.builder.CreatePtrToInt(paramBundle, c.uintptrType, "")
@@ -24,7 +27,7 @@ func (c *Compiler) emitStartGoroutine(funcPtr llvm.Value, params []llvm.Value) l
 		calleeValue := c.builder.CreatePtrToInt(funcPtr, c.uintptrType, "")
 		calleeValue = c.createRuntimeCall("makeGoroutine", []llvm.Value{calleeValue}, "")
 		calleeValue = c.builder.CreateIntToPtr(calleeValue, funcPtr.Type(), "")
-		c.createCall(calleeValue, params, "")
+		c.createCall(calleeValue, append(params, llvm.ConstPointerNull(c.i8ptrType)), "")
 	default:
 		panic("unreachable")
 	}
@@ -74,8 +77,8 @@ func (c *Compiler) createGoroutineStartWrapper(fn llvm.Value) llvm.Value {
 
 		// Create the list of params for the call.
 		paramTypes := fn.Type().ElementType().ParamTypes()
-		params := c.emitPointerUnpack(wrapper.Param(0), paramTypes[:len(paramTypes)-2])
-		params = append(params, llvm.Undef(c.i8ptrType), llvm.ConstPointerNull(c.i8ptrType))
+		params := c.emitPointerUnpack(wrapper.Param(0), paramTypes[:len(paramTypes)-1])
+		params = append(params, llvm.Undef(c.i8ptrType))
 
 		// Create the call.
 		c.builder.CreateCall(fn, params, "")

compiler/interface.go

@@ -210,10 +210,6 @@ func getTypeCodeName(t types.Type) string {
 		return "slice:" + getTypeCodeName(t.Elem())
 	case *types.Struct:
 		elems := make([]string, t.NumFields())
-		if t.NumFields() > 2 && t.Field(0).Name() == "C union" {
-			// TODO: report this as a normal error instead of panicking.
-			panic("cgo unions are not allowed in interfaces")
-		}
 		for i := 0; i < t.NumFields(); i++ {
 			embedded := ""
 			if t.Field(i).Embedded() {
@@ -283,14 +279,14 @@ func (c *Compiler) getInterfaceMethodSet(typ *types.Named) llvm.Value {
 		return llvm.ConstGEP(global, []llvm.Value{zero, zero})
 	}
 
-	// Every method is a *i16 reference indicating the signature of this method.
+	// Every method is a *i8 reference indicating the signature of this method.
 	methods := make([]llvm.Value, typ.Underlying().(*types.Interface).NumMethods())
 	for i := range methods {
 		method := typ.Underlying().(*types.Interface).Method(i)
 		methods[i] = c.getMethodSignature(method)
 	}
 
-	value := llvm.ConstArray(methods[0].Type(), methods)
+	value := llvm.ConstArray(c.i8ptrType, methods)
 	global = llvm.AddGlobal(c.mod, value.Type(), typ.String()+"$interface")
 	global.SetInitializer(value)
 	global.SetGlobalConstant(true)
@@ -299,7 +295,7 @@ func (c *Compiler) getInterfaceMethodSet(typ *types.Named) llvm.Value {
 }
 
 // getMethodSignature returns a global variable which is a reference to an
-// external *i16 indicating the indicating the signature of this method. It is
+// external *i8 indicating the indicating the signature of this method. It is
 // used during the interface lowering pass.
 func (c *Compiler) getMethodSignature(method *types.Func) llvm.Value {
 	signature := ir.MethodSignature(method)
@@ -465,6 +461,9 @@ func (c *Compiler) getInterfaceInvokeWrapper(f *ir.Function) llvm.Value {
 	paramTypes := append([]llvm.Type{c.i8ptrType}, fnType.ParamTypes()[len(expandedReceiverType):]...)
 	wrapFnType := llvm.FunctionType(fnType.ReturnType(), paramTypes, false)
 	wrapper = llvm.AddFunction(c.mod, wrapperName, wrapFnType)
+	if f.LLVMFn.LastParam().Name() == "parentHandle" {
+		wrapper.LastParam().SetName("parentHandle")
+	}
 	c.interfaceInvokeWrappers = append(c.interfaceInvokeWrappers, interfaceInvokeWrapper{
 		fn:           f,
 		wrapper:      wrapper,
@@ -483,7 +482,7 @@ func (c *Compiler) createInterfaceInvokeWrapper(state interfaceInvokeWrapper) {
 	wrapper.SetUnnamedAddr(true)
 
 	// add debug info if needed
-	if c.Debug {
+	if c.Debug() {
 		pos := c.ir.Program.Fset.Position(fn.Pos())
 		difunc := c.attachDebugInfoRaw(fn, wrapper, "$invoke", pos.Filename, pos.Line)
 		c.builder.SetCurrentDebugLocation(uint(pos.Line), uint(pos.Column), difunc, llvm.Metadata{})

compiler/interrupt.go

@@ -0,0 +1,91 @@
+package compiler
+
+import (
+	"strconv"
+	"strings"
+
+	"golang.org/x/tools/go/ssa"
+	"tinygo.org/x/go-llvm"
+)
+
+// emitInterruptGlobal creates a new runtime/interrupt.Interrupt struct that
+// will be lowered to a real interrupt during interrupt lowering.
+//
+// This two-stage approach allows unused interrupts to be optimized away if
+// necessary.
+func (c *Compiler) emitInterruptGlobal(frame *Frame, instr *ssa.CallCommon) (llvm.Value, error) {
+	// Get the interrupt number, which must be a compile-time constant.
+	id, ok := instr.Args[0].(*ssa.Const)
+	if !ok {
+		return llvm.Value{}, c.makeError(instr.Pos(), "interrupt ID is not a constant")
+	}
+
+	// Get the func value, which also must be a compile time constant.
+	// Note that bound functions are allowed if the function has a pointer
+	// receiver and is a global. This is rather strict but still allows for
+	// idiomatic Go code.
+	funcValue := c.getValue(frame, instr.Args[1])
+	if funcValue.IsAConstant().IsNil() {
+		// Try to determine the cause of the non-constantness for a nice error
+		// message.
+		switch instr.Args[1].(type) {
+		case *ssa.MakeClosure:
+			// This may also be a bound method.
+			return llvm.Value{}, c.makeError(instr.Pos(), "closures are not supported in interrupt.New")
+		}
+		// Fall back to a generic error.
+		return llvm.Value{}, c.makeError(instr.Pos(), "interrupt function must be constant")
+	}
+
+	// Create a new global of type runtime/interrupt.handle. Globals of this
+	// type are lowered in the interrupt lowering pass.
+	globalType := c.ir.Program.ImportedPackage("runtime/interrupt").Type("handle").Type()
+	globalLLVMType := c.getLLVMType(globalType)
+	globalName := "runtime/interrupt.$interrupt" + strconv.FormatInt(id.Int64(), 10)
+	if global := c.mod.NamedGlobal(globalName); !global.IsNil() {
+		return llvm.Value{}, c.makeError(instr.Pos(), "interrupt redeclared in this program")
+	}
+	global := llvm.AddGlobal(c.mod, globalLLVMType, globalName)
+	global.SetLinkage(llvm.PrivateLinkage)
+	global.SetGlobalConstant(true)
+	initializer := llvm.ConstNull(globalLLVMType)
+	initializer = llvm.ConstInsertValue(initializer, funcValue, []uint32{0})
+	initializer = llvm.ConstInsertValue(initializer, llvm.ConstInt(c.intType, uint64(id.Int64()), true), []uint32{1, 0})
+	global.SetInitializer(initializer)
+
+	// Add debug info to the interrupt global.
+	if c.Debug() {
+		pos := c.ir.Program.Fset.Position(instr.Pos())
+		diglobal := c.dibuilder.CreateGlobalVariableExpression(c.difiles[pos.Filename], llvm.DIGlobalVariableExpression{
+			Name:        "interrupt" + strconv.FormatInt(id.Int64(), 10),
+			LinkageName: globalName,
+			File:        c.getDIFile(pos.Filename),
+			Line:        pos.Line,
+			Type:        c.getDIType(globalType),
+			Expr:        c.dibuilder.CreateExpression(nil),
+			LocalToUnit: false,
+		})
+		global.AddMetadata(0, diglobal)
+	}
+
+	// Create the runtime/interrupt.Interrupt type. It is a struct with a single
+	// member of type int.
+	num := llvm.ConstPtrToInt(global, c.intType)
+	interrupt := llvm.ConstNamedStruct(c.mod.GetTypeByName("runtime/interrupt.Interrupt"), []llvm.Value{num})
+
+	// Add dummy "use" call for AVR, because interrupts may be used even though
+	// they are never referenced again. This is unlike Cortex-M or the RISC-V
+	// PLIC where each interrupt must be enabled using the interrupt number, and
+	// thus keeps the Interrupt object alive.
+	// This call is removed during interrupt lowering.
+	if strings.HasPrefix(c.Triple(), "avr") {
+		useFn := c.mod.NamedFunction("runtime/interrupt.use")
+		if useFn.IsNil() {
+			useFnType := llvm.FunctionType(c.ctx.VoidType(), []llvm.Type{interrupt.Type()}, false)
+			useFn = llvm.AddFunction(c.mod, "runtime/interrupt.use", useFnType)
+		}
+		c.builder.CreateCall(useFn, []llvm.Value{interrupt}, "")
+	}
+
+	return interrupt, nil
+}

compiler/llvm.go

@@ -1,8 +1,7 @@
 package compiler
 
 import (
-	"reflect"
-
+	"github.com/tinygo-org/tinygo/compiler/llvmutil"
 	"tinygo.org/x/go-llvm"
 )
 
@@ -24,22 +23,6 @@ func getUses(value llvm.Value) []llvm.Value {
 	return uses
 }
 
-// createEntryBlockAlloca creates a new alloca in the entry block, even though
-// the IR builder is located elsewhere. It assumes that the insert point is
-// at the end of the current block.
-func (c *Compiler) createEntryBlockAlloca(t llvm.Type, name string) llvm.Value {
-	currentBlock := c.builder.GetInsertBlock()
-	entryBlock := currentBlock.Parent().EntryBasicBlock()
-	if entryBlock.FirstInstruction().IsNil() {
-		c.builder.SetInsertPointAtEnd(entryBlock)
-	} else {
-		c.builder.SetInsertPointBefore(entryBlock.FirstInstruction())
-	}
-	alloca := c.builder.CreateAlloca(t, name)
-	c.builder.SetInsertPointAtEnd(currentBlock)
-	return alloca
-}
-
 // createTemporaryAlloca creates a new alloca in the entry block and adds
 // lifetime start infromation in the IR signalling that the alloca won't be used
 // before this point.
@@ -47,155 +30,40 @@ func (c *Compiler) createEntryBlockAlloca(t llvm.Type, name string) llvm.Value {
 // This is useful for creating temporary allocas for intrinsics. Don't forget to
 // end the lifetime using emitLifetimeEnd after you're done with it.
 func (c *Compiler) createTemporaryAlloca(t llvm.Type, name string) (alloca, bitcast, size llvm.Value) {
-	alloca = c.createEntryBlockAlloca(t, name)
-	bitcast = c.builder.CreateBitCast(alloca, c.i8ptrType, name+".bitcast")
-	size = llvm.ConstInt(c.ctx.Int64Type(), c.targetData.TypeAllocSize(t), false)
-	c.builder.CreateCall(c.getLifetimeStartFunc(), []llvm.Value{size, bitcast}, "")
-	return
+	return llvmutil.CreateTemporaryAlloca(c.builder, c.mod, t, name)
 }
 
 // emitLifetimeEnd signals the end of an (alloca) lifetime by calling the
 // llvm.lifetime.end intrinsic. It is commonly used together with
 // createTemporaryAlloca.
 func (c *Compiler) emitLifetimeEnd(ptr, size llvm.Value) {
-	c.builder.CreateCall(c.getLifetimeEndFunc(), []llvm.Value{size, ptr}, "")
+	llvmutil.EmitLifetimeEnd(c.builder, c.mod, ptr, size)
 }
 
-// getLifetimeStartFunc returns the llvm.lifetime.start intrinsic and creates it
-// first if it doesn't exist yet.
-func (c *Compiler) getLifetimeStartFunc() llvm.Value {
-	fn := c.mod.NamedFunction("llvm.lifetime.start.p0i8")
-	if fn.IsNil() {
-		fnType := llvm.FunctionType(c.ctx.VoidType(), []llvm.Type{c.ctx.Int64Type(), c.i8ptrType}, false)
-		fn = llvm.AddFunction(c.mod, "llvm.lifetime.start.p0i8", fnType)
-	}
-	return fn
+// emitPointerPack packs the list of values into a single pointer value using
+// bitcasts, or else allocates a value on the heap if it cannot be packed in the
+// pointer value directly. It returns the pointer with the packed data.
+func (c *Compiler) emitPointerPack(values []llvm.Value) llvm.Value {
+	return llvmutil.EmitPointerPack(c.builder, c.mod, c.Config, values)
 }
 
-// getLifetimeEndFunc returns the llvm.lifetime.end intrinsic and creates it
-// first if it doesn't exist yet.
-func (c *Compiler) getLifetimeEndFunc() llvm.Value {
-	fn := c.mod.NamedFunction("llvm.lifetime.end.p0i8")
-	if fn.IsNil() {
-		fnType := llvm.FunctionType(c.ctx.VoidType(), []llvm.Type{c.ctx.Int64Type(), c.i8ptrType}, false)
-		fn = llvm.AddFunction(c.mod, "llvm.lifetime.end.p0i8", fnType)
-	}
-	return fn
-}
-
-// splitBasicBlock splits a LLVM basic block into two parts. All instructions
-// after afterInst are moved into a new basic block (created right after the
-// current one) with the given name.
-func (c *Compiler) splitBasicBlock(afterInst llvm.Value, insertAfter llvm.BasicBlock, name string) llvm.BasicBlock {
-	oldBlock := afterInst.InstructionParent()
-	newBlock := c.ctx.InsertBasicBlock(insertAfter, name)
-	var nextInstructions []llvm.Value // values to move
-
-	// Collect to-be-moved instructions.
-	inst := afterInst
-	for {
-		inst = llvm.NextInstruction(inst)
-		if inst.IsNil() {
-			break
-		}
-		nextInstructions = append(nextInstructions, inst)
-	}
-
-	// Move instructions.
-	c.builder.SetInsertPointAtEnd(newBlock)
-	for _, inst := range nextInstructions {
-		inst.RemoveFromParentAsInstruction()
-		c.builder.Insert(inst)
-	}
-
-	// Find PHI nodes to update.
-	var phiNodes []llvm.Value // PHI nodes to update
-	for bb := insertAfter.Parent().FirstBasicBlock(); !bb.IsNil(); bb = llvm.NextBasicBlock(bb) {
-		for inst := bb.FirstInstruction(); !inst.IsNil(); inst = llvm.NextInstruction(inst) {
-			if inst.IsAPHINode().IsNil() {
-				continue
-			}
-			needsUpdate := false
-			incomingCount := inst.IncomingCount()
-			for i := 0; i < incomingCount; i++ {
-				if inst.IncomingBlock(i) == oldBlock {
-					needsUpdate = true
-					break
-				}
-			}
-			if !needsUpdate {
-				// PHI node has no incoming edge from the old block.
-				continue
-			}
-			phiNodes = append(phiNodes, inst)
-		}
-	}
-
-	// Update PHI nodes.
-	for _, phi := range phiNodes {
-		c.builder.SetInsertPointBefore(phi)
-		newPhi := c.builder.CreatePHI(phi.Type(), "")
-		incomingCount := phi.IncomingCount()
-		incomingVals := make([]llvm.Value, incomingCount)
-		incomingBlocks := make([]llvm.BasicBlock, incomingCount)
-		for i := 0; i < incomingCount; i++ {
-			value := phi.IncomingValue(i)
-			block := phi.IncomingBlock(i)
-			if block == oldBlock {
-				block = newBlock
-			}
-			incomingVals[i] = value
-			incomingBlocks[i] = block
-		}
-		newPhi.AddIncoming(incomingVals, incomingBlocks)
-		phi.ReplaceAllUsesWith(newPhi)
-		phi.EraseFromParentAsInstruction()
-	}
-
-	return newBlock
+// emitPointerUnpack extracts a list of values packed using emitPointerPack.
+func (c *Compiler) emitPointerUnpack(ptr llvm.Value, valueTypes []llvm.Type) []llvm.Value {
+	return llvmutil.EmitPointerUnpack(c.builder, c.mod, ptr, valueTypes)
 }
 
 // makeGlobalArray creates a new LLVM global with the given name and integers as
 // contents, and returns the global.
 // Note that it is left with the default linkage etc., you should set
 // linkage/constant/etc properties yourself.
-func (c *Compiler) makeGlobalArray(bufItf interface{}, name string, elementType llvm.Type) llvm.Value {
-	buf := reflect.ValueOf(bufItf)
-	globalType := llvm.ArrayType(elementType, buf.Len())
+func (c *Compiler) makeGlobalArray(buf []byte, name string, elementType llvm.Type) llvm.Value {
+	globalType := llvm.ArrayType(elementType, len(buf))
 	global := llvm.AddGlobal(c.mod, globalType, name)
 	value := llvm.Undef(globalType)
-	for i := 0; i < buf.Len(); i++ {
-		ch := buf.Index(i).Uint()
+	for i := 0; i < len(buf); i++ {
+		ch := uint64(buf[i])
 		value = llvm.ConstInsertValue(value, llvm.ConstInt(elementType, ch, false), []uint32{uint32(i)})
 	}
 	global.SetInitializer(value)
 	return global
 }
-
-// getGlobalBytes returns the slice contained in the array of the provided
-// global. It can recover the bytes originally created using makeGlobalArray, if
-// makeGlobalArray was given a byte slice.
-func getGlobalBytes(global llvm.Value) []byte {
-	value := global.Initializer()
-	buf := make([]byte, value.Type().ArrayLength())
-	for i := range buf {
-		buf[i] = byte(llvm.ConstExtractValue(value, []uint32{uint32(i)}).ZExtValue())
-	}
-	return buf
-}
-
-// replaceGlobalByteWithArray replaces a global integer type in the module with
-// an integer array, using a GEP to make the types match. It is a convenience
-// function used for creating reflection sidetables, for example.
-func (c *Compiler) replaceGlobalIntWithArray(name string, buf interface{}) llvm.Value {
-	oldGlobal := c.mod.NamedGlobal(name)
-	global := c.makeGlobalArray(buf, name+".tmp", oldGlobal.Type().ElementType())
-	gep := llvm.ConstGEP(global, []llvm.Value{
-		llvm.ConstInt(c.ctx.Int32Type(), 0, false),
-		llvm.ConstInt(c.ctx.Int32Type(), 0, false),
-	})
-	oldGlobal.ReplaceAllUsesWith(gep)
-	oldGlobal.EraseFromParentAsGlobal()
-	global.SetName(name)
-	return global
-}

compiler/llvmutil/llvm.go

@@ -0,0 +1,168 @@
+// Package llvmutil contains utility functions used across multiple compiler
+// packages. For example, they may be used by both the compiler pacakge and
+// transformation packages.
+//
+// Normally, utility packages are avoided. However, in this case, the utility
+// functions are non-trivial and hard to get right. Copying them to multiple
+// places would be a big risk if only one of them is updated.
+package llvmutil
+
+import "tinygo.org/x/go-llvm"
+
+// CreateEntryBlockAlloca creates a new alloca in the entry block, even though
+// the IR builder is located elsewhere. It assumes that the insert point is
+// at the end of the current block.
+func CreateEntryBlockAlloca(builder llvm.Builder, t llvm.Type, name string) llvm.Value {
+	currentBlock := builder.GetInsertBlock()
+	entryBlock := currentBlock.Parent().EntryBasicBlock()
+	if entryBlock.FirstInstruction().IsNil() {
+		builder.SetInsertPointAtEnd(entryBlock)
+	} else {
+		builder.SetInsertPointBefore(entryBlock.FirstInstruction())
+	}
+	alloca := builder.CreateAlloca(t, name)
+	builder.SetInsertPointAtEnd(currentBlock)
+	return alloca
+}
+
+// CreateTemporaryAlloca creates a new alloca in the entry block and adds
+// lifetime start infromation in the IR signalling that the alloca won't be used
+// before this point.
+//
+// This is useful for creating temporary allocas for intrinsics. Don't forget to
+// end the lifetime using emitLifetimeEnd after you're done with it.
+func CreateTemporaryAlloca(builder llvm.Builder, mod llvm.Module, t llvm.Type, name string) (alloca, bitcast, size llvm.Value) {
+	ctx := t.Context()
+	targetData := llvm.NewTargetData(mod.DataLayout())
+	i8ptrType := llvm.PointerType(ctx.Int8Type(), 0)
+	alloca = CreateEntryBlockAlloca(builder, t, name)
+	bitcast = builder.CreateBitCast(alloca, i8ptrType, name+".bitcast")
+	size = llvm.ConstInt(ctx.Int64Type(), targetData.TypeAllocSize(t), false)
+	builder.CreateCall(getLifetimeStartFunc(mod), []llvm.Value{size, bitcast}, "")
+	return
+}
+
+// CreateInstructionAlloca creates an alloca in the entry block, and places lifetime control intrinsics around the instruction
+func CreateInstructionAlloca(builder llvm.Builder, mod llvm.Module, t llvm.Type, inst llvm.Value, name string) llvm.Value {
+	ctx := mod.Context()
+	targetData := llvm.NewTargetData(mod.DataLayout())
+	i8ptrType := llvm.PointerType(ctx.Int8Type(), 0)
+
+	alloca := CreateEntryBlockAlloca(builder, t, name)
+	builder.SetInsertPointBefore(inst)
+	bitcast := builder.CreateBitCast(alloca, i8ptrType, name+".bitcast")
+	size := llvm.ConstInt(ctx.Int64Type(), targetData.TypeAllocSize(t), false)
+	builder.CreateCall(getLifetimeStartFunc(mod), []llvm.Value{size, bitcast}, "")
+	if next := llvm.NextInstruction(inst); !next.IsNil() {
+		builder.SetInsertPointBefore(next)
+	} else {
+		builder.SetInsertPointAtEnd(inst.InstructionParent())
+	}
+	builder.CreateCall(getLifetimeEndFunc(mod), []llvm.Value{size, bitcast}, "")
+	return alloca
+}
+
+// EmitLifetimeEnd signals the end of an (alloca) lifetime by calling the
+// llvm.lifetime.end intrinsic. It is commonly used together with
+// createTemporaryAlloca.
+func EmitLifetimeEnd(builder llvm.Builder, mod llvm.Module, ptr, size llvm.Value) {
+	builder.CreateCall(getLifetimeEndFunc(mod), []llvm.Value{size, ptr}, "")
+}
+
+// getLifetimeStartFunc returns the llvm.lifetime.start intrinsic and creates it
+// first if it doesn't exist yet.
+func getLifetimeStartFunc(mod llvm.Module) llvm.Value {
+	fn := mod.NamedFunction("llvm.lifetime.start.p0i8")
+	ctx := mod.Context()
+	i8ptrType := llvm.PointerType(ctx.Int8Type(), 0)
+	if fn.IsNil() {
+		fnType := llvm.FunctionType(ctx.VoidType(), []llvm.Type{ctx.Int64Type(), i8ptrType}, false)
+		fn = llvm.AddFunction(mod, "llvm.lifetime.start.p0i8", fnType)
+	}
+	return fn
+}
+
+// getLifetimeEndFunc returns the llvm.lifetime.end intrinsic and creates it
+// first if it doesn't exist yet.
+func getLifetimeEndFunc(mod llvm.Module) llvm.Value {
+	fn := mod.NamedFunction("llvm.lifetime.end.p0i8")
+	ctx := mod.Context()
+	i8ptrType := llvm.PointerType(ctx.Int8Type(), 0)
+	if fn.IsNil() {
+		fnType := llvm.FunctionType(ctx.VoidType(), []llvm.Type{ctx.Int64Type(), i8ptrType}, false)
+		fn = llvm.AddFunction(mod, "llvm.lifetime.end.p0i8", fnType)
+	}
+	return fn
+}
+
+// SplitBasicBlock splits a LLVM basic block into two parts. All instructions
+// after afterInst are moved into a new basic block (created right after the
+// current one) with the given name.
+func SplitBasicBlock(builder llvm.Builder, afterInst llvm.Value, insertAfter llvm.BasicBlock, name string) llvm.BasicBlock {
+	oldBlock := afterInst.InstructionParent()
+	newBlock := afterInst.Type().Context().InsertBasicBlock(insertAfter, name)
+	var nextInstructions []llvm.Value // values to move
+
+	// Collect to-be-moved instructions.
+	inst := afterInst
+	for {
+		inst = llvm.NextInstruction(inst)
+		if inst.IsNil() {
+			break
+		}
+		nextInstructions = append(nextInstructions, inst)
+	}
+
+	// Move instructions.
+	builder.SetInsertPointAtEnd(newBlock)
+	for _, inst := range nextInstructions {
+		inst.RemoveFromParentAsInstruction()
+		builder.Insert(inst)
+	}
+
+	// Find PHI nodes to update.
+	var phiNodes []llvm.Value // PHI nodes to update
+	for bb := insertAfter.Parent().FirstBasicBlock(); !bb.IsNil(); bb = llvm.NextBasicBlock(bb) {
+		for inst := bb.FirstInstruction(); !inst.IsNil(); inst = llvm.NextInstruction(inst) {
+			if inst.IsAPHINode().IsNil() {
+				continue
+			}
+			needsUpdate := false
+			incomingCount := inst.IncomingCount()
+			for i := 0; i < incomingCount; i++ {
+				if inst.IncomingBlock(i) == oldBlock {
+					needsUpdate = true
+					break
+				}
+			}
+			if !needsUpdate {
+				// PHI node has no incoming edge from the old block.
+				continue
+			}
+			phiNodes = append(phiNodes, inst)
+		}
+	}
+
+	// Update PHI nodes.
+	for _, phi := range phiNodes {
+		builder.SetInsertPointBefore(phi)
+		newPhi := builder.CreatePHI(phi.Type(), "")
+		incomingCount := phi.IncomingCount()
+		incomingVals := make([]llvm.Value, incomingCount)
+		incomingBlocks := make([]llvm.BasicBlock, incomingCount)
+		for i := 0; i < incomingCount; i++ {
+			value := phi.IncomingValue(i)
+			block := phi.IncomingBlock(i)
+			if block == oldBlock {
+				block = newBlock
+			}
+			incomingVals[i] = value
+			incomingBlocks[i] = block
+		}
+		newPhi.AddIncoming(incomingVals, incomingBlocks)
+		phi.ReplaceAllUsesWith(newPhi)
+		phi.EraseFromParentAsInstruction()
+	}
+
+	return newBlock
+}

compiler/llvmutil/wordpack.go

@@ -0,0 +1,141 @@
+package llvmutil
+
+// This file contains utility functions to pack and unpack sets of values. It
+// can take in a list of values and tries to store it efficiently in the pointer
+// itself if possible and legal.
+
+import (
+	"github.com/tinygo-org/tinygo/compileopts"
+	"tinygo.org/x/go-llvm"
+)
+
+// EmitPointerPack packs the list of values into a single pointer value using
+// bitcasts, or else allocates a value on the heap if it cannot be packed in the
+// pointer value directly. It returns the pointer with the packed data.
+func EmitPointerPack(builder llvm.Builder, mod llvm.Module, config *compileopts.Config, values []llvm.Value) llvm.Value {
+	ctx := mod.Context()
+	targetData := llvm.NewTargetData(mod.DataLayout())
+	i8ptrType := llvm.PointerType(mod.Context().Int8Type(), 0)
+	uintptrType := ctx.IntType(llvm.NewTargetData(mod.DataLayout()).PointerSize() * 8)
+
+	valueTypes := make([]llvm.Type, len(values))
+	for i, value := range values {
+		valueTypes[i] = value.Type()
+	}
+	packedType := ctx.StructType(valueTypes, false)
+
+	// Allocate memory for the packed data.
+	var packedAlloc, packedHeapAlloc llvm.Value
+	size := targetData.TypeAllocSize(packedType)
+	if size == 0 {
+		return llvm.ConstPointerNull(i8ptrType)
+	} else if len(values) == 1 && values[0].Type().TypeKind() == llvm.PointerTypeKind {
+		return builder.CreateBitCast(values[0], i8ptrType, "pack.ptr")
+	} else if size <= targetData.TypeAllocSize(i8ptrType) {
+		// Packed data fits in a pointer, so store it directly inside the
+		// pointer.
+		if len(values) == 1 && values[0].Type().TypeKind() == llvm.IntegerTypeKind {
+			// Try to keep this cast in SSA form.
+			return builder.CreateIntToPtr(values[0], i8ptrType, "pack.int")
+		}
+		// Because packedType is a struct and we have to cast it to a *i8, store
+		// it in an alloca first for bitcasting (store+bitcast+load).
+		packedAlloc, _, _ = CreateTemporaryAlloca(builder, mod, packedType, "")
+	} else {
+		// Packed data is bigger than a pointer, so allocate it on the heap.
+		sizeValue := llvm.ConstInt(uintptrType, size, false)
+		alloc := mod.NamedFunction("runtime.alloc")
+		packedHeapAlloc = builder.CreateCall(alloc, []llvm.Value{
+			sizeValue,
+			llvm.Undef(i8ptrType),            // unused context parameter
+			llvm.ConstPointerNull(i8ptrType), // coroutine handle
+		}, "")
+		if config.NeedsStackObjects() {
+			trackPointer := mod.NamedFunction("runtime.trackPointer")
+			builder.CreateCall(trackPointer, []llvm.Value{
+				packedHeapAlloc,
+				llvm.Undef(i8ptrType),            // unused context parameter
+				llvm.ConstPointerNull(i8ptrType), // coroutine handle
+			}, "")
+		}
+		packedAlloc = builder.CreateBitCast(packedHeapAlloc, llvm.PointerType(packedType, 0), "")
+	}
+	// Store all values in the alloca or heap pointer.
+	for i, value := range values {
+		indices := []llvm.Value{
+			llvm.ConstInt(ctx.Int32Type(), 0, false),
+			llvm.ConstInt(ctx.Int32Type(), uint64(i), false),
+		}
+		gep := builder.CreateInBoundsGEP(packedAlloc, indices, "")
+		builder.CreateStore(value, gep)
+	}
+
+	if packedHeapAlloc.IsNil() {
+		// Load value (as *i8) from the alloca.
+		packedAlloc = builder.CreateBitCast(packedAlloc, llvm.PointerType(i8ptrType, 0), "")
+		result := builder.CreateLoad(packedAlloc, "")
+		packedPtr := builder.CreateBitCast(packedAlloc, i8ptrType, "")
+		packedSize := llvm.ConstInt(ctx.Int64Type(), targetData.TypeAllocSize(packedAlloc.Type()), false)
+		EmitLifetimeEnd(builder, mod, packedPtr, packedSize)
+		return result
+	} else {
+		// Get the original heap allocation pointer, which already is an *i8.
+		return packedHeapAlloc
+	}
+}
+
+// EmitPointerUnpack extracts a list of values packed using EmitPointerPack.
+func EmitPointerUnpack(builder llvm.Builder, mod llvm.Module, ptr llvm.Value, valueTypes []llvm.Type) []llvm.Value {
+	ctx := mod.Context()
+	targetData := llvm.NewTargetData(mod.DataLayout())
+	i8ptrType := llvm.PointerType(mod.Context().Int8Type(), 0)
+	uintptrType := ctx.IntType(llvm.NewTargetData(mod.DataLayout()).PointerSize() * 8)
+
+	packedType := ctx.StructType(valueTypes, false)
+
+	// Get a correctly-typed pointer to the packed data.
+	var packedAlloc, packedRawAlloc llvm.Value
+	size := targetData.TypeAllocSize(packedType)
+	if size == 0 {
+		// No data to unpack.
+	} else if len(valueTypes) == 1 && valueTypes[0].TypeKind() == llvm.PointerTypeKind {
+		// A single pointer is always stored directly.
+		return []llvm.Value{builder.CreateBitCast(ptr, valueTypes[0], "unpack.ptr")}
+	} else if size <= targetData.TypeAllocSize(i8ptrType) {
+		// Packed data stored directly in pointer.
+		if len(valueTypes) == 1 && valueTypes[0].TypeKind() == llvm.IntegerTypeKind {
+			// Keep this cast in SSA form.
+			return []llvm.Value{builder.CreatePtrToInt(ptr, valueTypes[0], "unpack.int")}
+		}
+		// Fallback: load it using an alloca.
+		packedRawAlloc, _, _ = CreateTemporaryAlloca(builder, mod, llvm.PointerType(i8ptrType, 0), "unpack.raw.alloc")
+		packedRawValue := builder.CreateBitCast(ptr, llvm.PointerType(i8ptrType, 0), "unpack.raw.value")
+		builder.CreateStore(packedRawValue, packedRawAlloc)
+		packedAlloc = builder.CreateBitCast(packedRawAlloc, llvm.PointerType(packedType, 0), "unpack.alloc")
+	} else {
+		// Packed data stored on the heap. Bitcast the passed-in pointer to the
+		// correct pointer type.
+		packedAlloc = builder.CreateBitCast(ptr, llvm.PointerType(packedType, 0), "unpack.raw.ptr")
+	}
+	// Load each value from the packed data.
+	values := make([]llvm.Value, len(valueTypes))
+	for i, valueType := range valueTypes {
+		if targetData.TypeAllocSize(valueType) == 0 {
+			// This value has length zero, so there's nothing to load.
+			values[i] = llvm.ConstNull(valueType)
+			continue
+		}
+		indices := []llvm.Value{
+			llvm.ConstInt(ctx.Int32Type(), 0, false),
+			llvm.ConstInt(ctx.Int32Type(), uint64(i), false),
+		}
+		gep := builder.CreateInBoundsGEP(packedAlloc, indices, "")
+		values[i] = builder.CreateLoad(gep, "")
+	}
+	if !packedRawAlloc.IsNil() {
+		allocPtr := builder.CreateBitCast(packedRawAlloc, i8ptrType, "")
+		allocSize := llvm.ConstInt(ctx.Int64Type(), targetData.TypeAllocSize(uintptrType), false)
+		EmitLifetimeEnd(builder, mod, allocPtr, allocSize)
+	}
+	return values
+}

compiler/optimizer.go

@@ -9,7 +9,7 @@ import (
 
 // Run the LLVM optimizer over the module.
 // The inliner can be disabled (if necessary) by passing 0 to the inlinerThreshold.
-func (c *Compiler) Optimize(optLevel, sizeLevel int, inlinerThreshold uint) error {
+func (c *Compiler) Optimize(optLevel, sizeLevel int, inlinerThreshold uint) []error {
 	builder := llvm.NewPassManagerBuilder()
 	defer builder.Dispose()
 	builder.SetOptLevel(optLevel)
@@ -19,13 +19,16 @@ func (c *Compiler) Optimize(optLevel, sizeLevel int, inlinerThreshold uint) erro
 	}
 	builder.AddCoroutinePassesToExtensionPoints()
 
-	if c.PanicStrategy == "trap" {
-		c.replacePanicsWithTrap() // -panic=trap
+	if c.PanicStrategy() == "trap" {
+		transform.ReplacePanicsWithTrap(c.mod) // -panic=trap
 	}
 
 	// run a check of all of our code
-	if c.VerifyIR {
-		c.checkModule()
+	if c.VerifyIR() {
+		errs := c.checkModule()
+		if errs != nil {
+			return errs
+		}
 	}
 
 	// Run function passes for each function.
@@ -42,6 +45,7 @@ func (c *Compiler) Optimize(optLevel, sizeLevel int, inlinerThreshold uint) erro
 		// Run some preparatory passes for the Go optimizer.
 		goPasses := llvm.NewPassManager()
 		defer goPasses.Dispose()
+		goPasses.AddGlobalDCEPass()
 		goPasses.AddGlobalOptimizerPass()
 		goPasses.AddConstantPropagationPass()
 		goPasses.AddAggressiveDCEPass()
@@ -50,10 +54,18 @@ func (c *Compiler) Optimize(optLevel, sizeLevel int, inlinerThreshold uint) erro
 
 		// Run Go-specific optimization passes.
 		transform.OptimizeMaps(c.mod)
-		c.OptimizeStringToBytes()
+		transform.OptimizeStringToBytes(c.mod)
 		transform.OptimizeAllocs(c.mod)
-		c.LowerInterfaces()
-		c.LowerFuncValues()
+		transform.LowerInterfaces(c.mod)
+
+		errs := transform.LowerInterruptRegistrations(c.mod)
+		if len(errs) > 0 {
+			return errs
+		}
+
+		if c.funcImplementation() == funcValueSwitch {
+			transform.LowerFuncValues(c.mod)
+		}
 
 		// After interfaces are lowered, there are many more opportunities for
 		// interprocedural optimizations. To get them to work, function
@@ -62,7 +74,7 @@ func (c *Compiler) Optimize(optLevel, sizeLevel int, inlinerThreshold uint) erro
 
 		// Run TinyGo-specific interprocedural optimizations.
 		transform.OptimizeAllocs(c.mod)
-		c.OptimizeStringToBytes()
+		transform.OptimizeStringToBytes(c.mod)
 
 		// Lower runtime.isnil calls to regular nil comparisons.
 		isnil := c.mod.NamedFunction("runtime.isnil")
@@ -82,19 +94,30 @@ func (c *Compiler) Optimize(optLevel, sizeLevel int, inlinerThreshold uint) erro
 
 		err := c.LowerGoroutines()
 		if err != nil {
-			return err
+			return []error{err}
 		}
 	} else {
 		// Must be run at any optimization level.
-		c.LowerInterfaces()
-		c.LowerFuncValues()
+		transform.LowerInterfaces(c.mod)
+		if c.funcImplementation() == funcValueSwitch {
+			transform.LowerFuncValues(c.mod)
+		}
 		err := c.LowerGoroutines()
 		if err != nil {
-			return err
+			return []error{err}
+		}
+		errs := transform.LowerInterruptRegistrations(c.mod)
+		if len(errs) > 0 {
+			return errs
+		}
+	}
+	if c.VerifyIR() {
+		if errs := c.checkModule(); errs != nil {
+			return errs
 		}
 	}
 	if err := c.Verify(); err != nil {
-		return errors.New("optimizations caused a verification failure")
+		return []error{errors.New("optimizations caused a verification failure")}
 	}
 
 	if sizeLevel >= 2 {
@@ -108,7 +131,7 @@ func (c *Compiler) Optimize(optLevel, sizeLevel int, inlinerThreshold uint) erro
 	}
 
 	// After TinyGo-specific transforms have finished, undo exporting these functions.
-	for _, name := range functionsUsedInTransforms {
+	for _, name := range c.getFunctionsUsedInTransforms() {
 		fn := c.mod.NamedFunction(name)
 		if fn.IsNil() {
 			continue
@@ -129,132 +152,13 @@ func (c *Compiler) Optimize(optLevel, sizeLevel int, inlinerThreshold uint) erro
 	builder.Populate(modPasses)
 	modPasses.Run(c.mod)
 
-	hasGCPass := c.addGlobalsBitmap()
-	hasGCPass = c.makeGCStackSlots() || hasGCPass
+	hasGCPass := transform.AddGlobalsBitmap(c.mod)
+	hasGCPass = transform.MakeGCStackSlots(c.mod) || hasGCPass
 	if hasGCPass {
 		if err := c.Verify(); err != nil {
-			return errors.New("GC pass caused a verification failure")
+			return []error{errors.New("GC pass caused a verification failure")}
 		}
 	}
 
 	return nil
 }
-
-// Replace panic calls with calls to llvm.trap, to reduce code size. This is the
-// -panic=trap intrinsic.
-func (c *Compiler) replacePanicsWithTrap() {
-	trap := c.mod.NamedFunction("llvm.trap")
-	for _, name := range []string{"runtime._panic", "runtime.runtimePanic"} {
-		fn := c.mod.NamedFunction(name)
-		if fn.IsNil() {
-			continue
-		}
-		for _, use := range getUses(fn) {
-			if use.IsACallInst().IsNil() || use.CalledValue() != fn {
-				panic("expected use of a panic function to be a call")
-			}
-			c.builder.SetInsertPointBefore(use)
-			c.builder.CreateCall(trap, nil, "")
-		}
-	}
-}
-
-// Transform runtime.stringToBytes(...) calls into const []byte slices whenever
-// possible. This optimizes the following pattern:
-//     w.Write([]byte("foo"))
-// where Write does not store to the slice.
-func (c *Compiler) OptimizeStringToBytes() {
-	stringToBytes := c.mod.NamedFunction("runtime.stringToBytes")
-	if stringToBytes.IsNil() {
-		// nothing to optimize
-		return
-	}
-
-	for _, call := range getUses(stringToBytes) {
-		strptr := call.Operand(0)
-		strlen := call.Operand(1)
-
-		// strptr is always constant because strings are always constant.
-
-		convertedAllUses := true
-		for _, use := range getUses(call) {
-			nilValue := llvm.Value{}
-			if use.IsAExtractValueInst() == nilValue {
-				convertedAllUses = false
-				continue
-			}
-			switch use.Type().TypeKind() {
-			case llvm.IntegerTypeKind:
-				// A length (len or cap). Propagate the length value.
-				use.ReplaceAllUsesWith(strlen)
-				use.EraseFromParentAsInstruction()
-			case llvm.PointerTypeKind:
-				// The string pointer itself.
-				if !c.isReadOnly(use) {
-					convertedAllUses = false
-					continue
-				}
-				use.ReplaceAllUsesWith(strptr)
-				use.EraseFromParentAsInstruction()
-			default:
-				// should not happen
-				panic("unknown return type of runtime.stringToBytes: " + use.Type().String())
-			}
-		}
-		if convertedAllUses {
-			// Call to runtime.stringToBytes can be eliminated: both the input
-			// and the output is constant.
-			call.EraseFromParentAsInstruction()
-		}
-	}
-}
-
-// Check whether the given value (which is of pointer type) is never stored to.
-func (c *Compiler) isReadOnly(value llvm.Value) bool {
-	uses := getUses(value)
-	for _, use := range uses {
-		nilValue := llvm.Value{}
-		if use.IsAGetElementPtrInst() != nilValue {
-			if !c.isReadOnly(use) {
-				return false
-			}
-		} else if use.IsACallInst() != nilValue {
-			if !c.hasFlag(use, value, "readonly") {
-				return false
-			}
-		} else {
-			// Unknown instruction, might not be readonly.
-			return false
-		}
-	}
-	return true
-}
-
-// Check whether all uses of this param as parameter to the call have the given
-// flag. In most cases, there will only be one use but a function could take the
-// same parameter twice, in which case both must have the flag.
-// A flag can be any enum flag, like "readonly".
-func (c *Compiler) hasFlag(call, param llvm.Value, kind string) bool {
-	fn := call.CalledValue()
-	nilValue := llvm.Value{}
-	if fn.IsAFunction() == nilValue {
-		// This is not a function but something else, like a function pointer.
-		return false
-	}
-	kindID := llvm.AttributeKindID(kind)
-	for i := 0; i < fn.ParamsCount(); i++ {
-		if call.Operand(i) != param {
-			// This is not the parameter we're checking.
-			continue
-		}
-		index := i + 1 // param attributes start at 1
-		attr := fn.GetEnumAttributeAtIndex(index, kindID)
-		nilAttribute := llvm.Attribute{}
-		if attr == nilAttribute {
-			// At least one parameter doesn't have the flag (there may be
-			// multiple).
-			return false
-		}
-	}
-	return true
-}

compiler/sizes.go

@@ -63,12 +63,6 @@ func (s *StdSizes) Alignof(T types.Type) int64 {
 
 func (s *StdSizes) Offsetsof(fields []*types.Var) []int64 {
 	offsets := make([]int64, len(fields))
-	if len(fields) > 1 && fields[0].Name() == "C union" {
-		// This struct contains the magic "C union" field which indicates that
-		// this is actually a union from CGo.
-		// All fields in the union start at 0 so return that.
-		return offsets // all fields are still set to 0
-	}
 	var o int64
 	for i, f := range fields {
 		a := s.Alignof(f.Type())
@@ -116,6 +110,9 @@ func (s *StdSizes) Sizeof(T types.Type) int64 {
 		if k == types.UnsafePointer {
 			return s.PtrSize
 		}
+		if k == types.Invalid {
+			return 0 // only relevant when there is a type error somewhere
+		}
 		panic("unknown basic type: " + t.String())
 	case *types.Array:
 		n := t.Len()
@@ -143,29 +140,12 @@ func (s *StdSizes) Sizeof(T types.Type) int64 {
 				maxAlign = al
 			}
 		}
-		if fields[0].Name() == "C union" {
-			// Magic field that indicates this is a CGo union and not a struct.
-			// The size is the biggest element, aligned to the element with the
-			// biggest alignment. This is not necessarily the same, for example
-			// in the following union:
-			//     union { int32_t l; int16_t s[3] }
-			maxSize := int64(0)
-			for _, field := range fields[1:] {
-				si := s.Sizeof(field.Type())
-				if si > maxSize {
-					maxSize = si
-				}
-			}
-			return align(maxSize, maxAlign)
-		} else {
-			// This is a regular struct.
-			// Pick the size that fits this struct and add some alignment. Some
-			// structs have some extra padding at the end which should also be
-			// taken care of:
-			//     struct { int32 n; byte b }
-			offsets := s.Offsetsof(fields)
-			return align(offsets[n-1]+s.Sizeof(fields[n-1].Type()), maxAlign)
-		}
+		// Pick the size that fits this struct and add some alignment. Some
+		// structs have some extra padding at the end which should also be taken
+		// care of:
+		//     struct { int32 n; byte b }
+		offsets := s.Offsetsof(fields)
+		return align(offsets[n-1]+s.Sizeof(fields[n-1].Type()), maxAlign)
 	case *types.Interface:
 		return s.PtrSize * 2
 	case *types.Pointer:

compiler/symbol.go

@@ -7,6 +7,7 @@ import (
 	"go/ast"
 	"go/token"
 	"go/types"
+	"strconv"
 	"strings"
 
 	"github.com/tinygo-org/tinygo/loader"
@@ -20,6 +21,7 @@ import (
 type globalInfo struct {
 	linkName string // go:extern
 	extern   bool   // go:extern
+	align    int    // go:align
 }
 
 // loadASTComments loads comments on globals from the AST, for use later in the
@@ -58,12 +60,45 @@ func (c *Compiler) getGlobal(g *ssa.Global) llvm.Value {
 	info := c.getGlobalInfo(g)
 	llvmGlobal := c.mod.NamedGlobal(info.linkName)
 	if llvmGlobal.IsNil() {
-		llvmType := c.getLLVMType(g.Type().(*types.Pointer).Elem())
+		typ := g.Type().(*types.Pointer).Elem()
+		llvmType := c.getLLVMType(typ)
 		llvmGlobal = llvm.AddGlobal(c.mod, llvmType, info.linkName)
 		if !info.extern {
 			llvmGlobal.SetInitializer(llvm.ConstNull(llvmType))
 			llvmGlobal.SetLinkage(llvm.InternalLinkage)
 		}
+
+		// Set alignment from the //go:align comment.
+		var alignInBits uint32
+		if info.align < 0 || info.align&(info.align-1) != 0 {
+			// Check for power-of-two (or 0).
+			// See: https://stackoverflow.com/a/108360
+			c.addError(g.Pos(), "global variable alignment must be a positive power of two")
+		} else {
+			// Set the alignment only when it is a power of two.
+			alignInBits = uint32(info.align) ^ uint32(info.align-1)
+			if info.align > c.targetData.ABITypeAlignment(llvmType) {
+				llvmGlobal.SetAlignment(info.align)
+			}
+		}
+
+		if c.Debug() {
+			// Add debug info.
+			// TODO: this should be done for every global in the program, not just
+			// the ones that are referenced from some code.
+			pos := c.ir.Program.Fset.Position(g.Pos())
+			diglobal := c.dibuilder.CreateGlobalVariableExpression(c.difiles[pos.Filename], llvm.DIGlobalVariableExpression{
+				Name:        g.RelString(nil),
+				LinkageName: info.linkName,
+				File:        c.getDIFile(pos.Filename),
+				Line:        pos.Line,
+				Type:        c.getDIType(typ),
+				LocalToUnit: false,
+				Expr:        c.dibuilder.CreateExpression(nil),
+				AlignInBits: alignInBits,
+			})
+			llvmGlobal.AddMetadata(0, diglobal)
+		}
 	}
 	return llvmGlobal
 }
@@ -102,6 +137,11 @@ func (info *globalInfo) parsePragmas(doc *ast.CommentGroup) {
 			if len(parts) == 2 {
 				info.linkName = parts[1]
 			}
+		case "//go:align":
+			align, err := strconv.Atoi(parts[1])
+			if err == nil {
+				info.align = align
+			}
 		}
 	}
 }

compiler/syscall.go

@@ -16,8 +16,8 @@ func (c *Compiler) emitSyscall(frame *Frame, call *ssa.CallCommon) (llvm.Value,
 	num := c.getValue(frame, call.Args[0])
 	var syscallResult llvm.Value
 	switch {
-	case c.GOARCH == "amd64":
-		if c.GOOS == "darwin" {
+	case c.GOARCH() == "amd64":
+		if c.GOOS() == "darwin" {
 			// Darwin adds this magic number to system call numbers:
 			//
 			// > Syscall classes for 64-bit system call entry.
@@ -58,7 +58,7 @@ func (c *Compiler) emitSyscall(frame *Frame, call *ssa.CallCommon) (llvm.Value,
 		fnType := llvm.FunctionType(c.uintptrType, argTypes, false)
 		target := llvm.InlineAsm(fnType, "syscall", constraints, true, false, llvm.InlineAsmDialectIntel)
 		syscallResult = c.builder.CreateCall(target, args, "")
-	case c.GOARCH == "386" && c.GOOS == "linux":
+	case c.GOARCH() == "386" && c.GOOS() == "linux":
 		// Sources:
 		//   syscall(2) man page
 		//   https://stackoverflow.com/a/2538212
@@ -84,7 +84,7 @@ func (c *Compiler) emitSyscall(frame *Frame, call *ssa.CallCommon) (llvm.Value,
 		fnType := llvm.FunctionType(c.uintptrType, argTypes, false)
 		target := llvm.InlineAsm(fnType, "int 0x80", constraints, true, false, llvm.InlineAsmDialectIntel)
 		syscallResult = c.builder.CreateCall(target, args, "")
-	case c.GOARCH == "arm" && c.GOOS == "linux":
+	case c.GOARCH() == "arm" && c.GOOS() == "linux":
 		// Implement the EABI system call convention for Linux.
 		// Source: syscall(2) man page.
 		args := []llvm.Value{}
@@ -116,7 +116,7 @@ func (c *Compiler) emitSyscall(frame *Frame, call *ssa.CallCommon) (llvm.Value,
 		fnType := llvm.FunctionType(c.uintptrType, argTypes, false)
 		target := llvm.InlineAsm(fnType, "svc #0", constraints, true, false, 0)
 		syscallResult = c.builder.CreateCall(target, args, "")
-	case c.GOARCH == "arm64" && c.GOOS == "linux":
+	case c.GOARCH() == "arm64" && c.GOOS() == "linux":
 		// Source: syscall(2) man page.
 		args := []llvm.Value{}
 		argTypes := []llvm.Type{}
@@ -149,10 +149,10 @@ func (c *Compiler) emitSyscall(frame *Frame, call *ssa.CallCommon) (llvm.Value,
 		target := llvm.InlineAsm(fnType, "svc #0", constraints, true, false, 0)
 		syscallResult = c.builder.CreateCall(target, args, "")
 	default:
-		return llvm.Value{}, c.makeError(call.Pos(), "unknown GOOS/GOARCH for syscall: "+c.GOOS+"/"+c.GOARCH)
+		return llvm.Value{}, c.makeError(call.Pos(), "unknown GOOS/GOARCH for syscall: "+c.GOOS()+"/"+c.GOARCH())
 	}
-	switch c.GOOS {
-	case "linux":
+	switch c.GOOS() {
+	case "linux", "freebsd":
 		// Return values: r0, r1 uintptr, err Errno
 		// Pseudocode:
 		//     var err uintptr
@@ -187,6 +187,6 @@ func (c *Compiler) emitSyscall(frame *Frame, call *ssa.CallCommon) (llvm.Value,
 		retval = c.builder.CreateInsertValue(retval, errResult, 2, "")
 		return retval, nil
 	default:
-		return llvm.Value{}, c.makeError(call.Pos(), "unknown GOOS/GOARCH for syscall: "+c.GOOS+"/"+c.GOARCH)
+		return llvm.Value{}, c.makeError(call.Pos(), "unknown GOOS/GOARCH for syscall: "+c.GOOS()+"/"+c.GOARCH())
 	}
 }

compiler/wordpack.go

@@ -1,120 +0,0 @@
-package compiler
-
-// This file contains utility functions to pack and unpack sets of values. It
-// can take in a list of values and tries to store it efficiently in the pointer
-// itself if possible and legal.
-
-import (
-	"tinygo.org/x/go-llvm"
-)
-
-// emitPointerPack packs the list of values into a single pointer value using
-// bitcasts, or else allocates a value on the heap if it cannot be packed in the
-// pointer value directly. It returns the pointer with the packed data.
-func (c *Compiler) emitPointerPack(values []llvm.Value) llvm.Value {
-	valueTypes := make([]llvm.Type, len(values))
-	for i, value := range values {
-		valueTypes[i] = value.Type()
-	}
-	packedType := c.ctx.StructType(valueTypes, false)
-
-	// Allocate memory for the packed data.
-	var packedAlloc, packedHeapAlloc llvm.Value
-	size := c.targetData.TypeAllocSize(packedType)
-	if size == 0 {
-		return llvm.ConstPointerNull(c.i8ptrType)
-	} else if len(values) == 1 && values[0].Type().TypeKind() == llvm.PointerTypeKind {
-		return c.builder.CreateBitCast(values[0], c.i8ptrType, "pack.ptr")
-	} else if size <= c.targetData.TypeAllocSize(c.i8ptrType) {
-		// Packed data fits in a pointer, so store it directly inside the
-		// pointer.
-		if len(values) == 1 && values[0].Type().TypeKind() == llvm.IntegerTypeKind {
-			// Try to keep this cast in SSA form.
-			return c.builder.CreateIntToPtr(values[0], c.i8ptrType, "pack.int")
-		}
-		// Because packedType is a struct and we have to cast it to a *i8, store
-		// it in an alloca first for bitcasting (store+bitcast+load).
-		packedAlloc, _, _ = c.createTemporaryAlloca(packedType, "")
-	} else {
-		// Packed data is bigger than a pointer, so allocate it on the heap.
-		sizeValue := llvm.ConstInt(c.uintptrType, size, false)
-		packedHeapAlloc = c.createRuntimeCall("alloc", []llvm.Value{sizeValue}, "")
-		if c.needsStackObjects() {
-			c.trackPointer(packedHeapAlloc)
-		}
-		packedAlloc = c.builder.CreateBitCast(packedHeapAlloc, llvm.PointerType(packedType, 0), "")
-	}
-	// Store all values in the alloca or heap pointer.
-	for i, value := range values {
-		indices := []llvm.Value{
-			llvm.ConstInt(c.ctx.Int32Type(), 0, false),
-			llvm.ConstInt(c.ctx.Int32Type(), uint64(i), false),
-		}
-		gep := c.builder.CreateInBoundsGEP(packedAlloc, indices, "")
-		c.builder.CreateStore(value, gep)
-	}
-
-	if packedHeapAlloc.IsNil() {
-		// Load value (as *i8) from the alloca.
-		packedAlloc = c.builder.CreateBitCast(packedAlloc, llvm.PointerType(c.i8ptrType, 0), "")
-		result := c.builder.CreateLoad(packedAlloc, "")
-		packedPtr := c.builder.CreateBitCast(packedAlloc, c.i8ptrType, "")
-		packedSize := llvm.ConstInt(c.ctx.Int64Type(), c.targetData.TypeAllocSize(packedAlloc.Type()), false)
-		c.emitLifetimeEnd(packedPtr, packedSize)
-		return result
-	} else {
-		// Get the original heap allocation pointer, which already is an *i8.
-		return packedHeapAlloc
-	}
-}
-
-// emitPointerUnpack extracts a list of values packed using emitPointerPack.
-func (c *Compiler) emitPointerUnpack(ptr llvm.Value, valueTypes []llvm.Type) []llvm.Value {
-	packedType := c.ctx.StructType(valueTypes, false)
-
-	// Get a correctly-typed pointer to the packed data.
-	var packedAlloc, packedRawAlloc llvm.Value
-	size := c.targetData.TypeAllocSize(packedType)
-	if size == 0 {
-		// No data to unpack.
-	} else if len(valueTypes) == 1 && valueTypes[0].TypeKind() == llvm.PointerTypeKind {
-		// A single pointer is always stored directly.
-		return []llvm.Value{c.builder.CreateBitCast(ptr, valueTypes[0], "unpack.ptr")}
-	} else if size <= c.targetData.TypeAllocSize(c.i8ptrType) {
-		// Packed data stored directly in pointer.
-		if len(valueTypes) == 1 && valueTypes[0].TypeKind() == llvm.IntegerTypeKind {
-			// Keep this cast in SSA form.
-			return []llvm.Value{c.builder.CreatePtrToInt(ptr, valueTypes[0], "unpack.int")}
-		}
-		// Fallback: load it using an alloca.
-		packedRawAlloc, _, _ = c.createTemporaryAlloca(llvm.PointerType(c.i8ptrType, 0), "unpack.raw.alloc")
-		packedRawValue := c.builder.CreateBitCast(ptr, llvm.PointerType(c.i8ptrType, 0), "unpack.raw.value")
-		c.builder.CreateStore(packedRawValue, packedRawAlloc)
-		packedAlloc = c.builder.CreateBitCast(packedRawAlloc, llvm.PointerType(packedType, 0), "unpack.alloc")
-	} else {
-		// Packed data stored on the heap. Bitcast the passed-in pointer to the
-		// correct pointer type.
-		packedAlloc = c.builder.CreateBitCast(ptr, llvm.PointerType(packedType, 0), "unpack.raw.ptr")
-	}
-	// Load each value from the packed data.
-	values := make([]llvm.Value, len(valueTypes))
-	for i, valueType := range valueTypes {
-		if c.targetData.TypeAllocSize(valueType) == 0 {
-			// This value has length zero, so there's nothing to load.
-			values[i] = llvm.ConstNull(valueType)
-			continue
-		}
-		indices := []llvm.Value{
-			llvm.ConstInt(c.ctx.Int32Type(), 0, false),
-			llvm.ConstInt(c.ctx.Int32Type(), uint64(i), false),
-		}
-		gep := c.builder.CreateInBoundsGEP(packedAlloc, indices, "")
-		values[i] = c.builder.CreateLoad(gep, "")
-	}
-	if !packedRawAlloc.IsNil() {
-		allocPtr := c.builder.CreateBitCast(packedRawAlloc, c.i8ptrType, "")
-		allocSize := llvm.ConstInt(c.ctx.Int64Type(), c.targetData.TypeAllocSize(c.uintptrType), false)
-		c.emitLifetimeEnd(allocPtr, allocSize)
-	}
-	return values
-}

go.mod

@@ -4,7 +4,11 @@ go 1.11
 
 require (
 	github.com/blakesmith/ar v0.0.0-20150311145944-8bd4349a67f2
+	github.com/creack/goselect v0.1.0 // indirect
+	github.com/google/shlex v0.0.0-20181106134648-c34317bd91bf
 	github.com/marcinbor85/gohex v0.0.0-20180128172054-7a43cd876e46
+	go.bug.st/serial.v1 v0.0.0-20180827123349-5f7892a7bb45
+	golang.org/x/sys v0.0.0-20191010194322-b09406accb47 // indirect
 	golang.org/x/tools v0.0.0-20190227180812-8dcc6e70cdef
-	tinygo.org/x/go-llvm v0.0.0-20190818154551-95bc4ffe1add
+	tinygo.org/x/go-llvm v0.0.0-20200104190746-1ff21df33566
 )

go.sum

@@ -1,11 +1,19 @@
 github.com/blakesmith/ar v0.0.0-20150311145944-8bd4349a67f2 h1:oMCHnXa6CCCafdPDbMh/lWRhRByN0VFLvv+g+ayx1SI=
 github.com/blakesmith/ar v0.0.0-20150311145944-8bd4349a67f2/go.mod h1:PkYb9DJNAwrSvRx5DYA+gUcOIgTGVMNkfSCbZM8cWpI=
+github.com/creack/goselect v0.1.0 h1:4QiXIhcpSQF50XGaBsFzesjwX/1qOY5bOveQPmN9CXY=
+github.com/creack/goselect v0.1.0/go.mod h1:gHrIcH/9UZDn2qgeTUeW5K9eZsVYCH6/60J/FHysWyE=
 github.com/golang/protobuf v1.2.0/go.mod h1:6lQm79b+lXiMfvg/cZm0SGofjICqVBUtrP5yJMmIC1U=
+github.com/google/shlex v0.0.0-20181106134648-c34317bd91bf h1:7+FW5aGwISbqUtkfmIpZJGRgNFg2ioYPvFaUxdqpDsg=
+github.com/google/shlex v0.0.0-20181106134648-c34317bd91bf/go.mod h1:RpwtwJQFrIEPstU94h88MWPXP2ektJZ8cZ0YntAmXiE=
 github.com/marcinbor85/gohex v0.0.0-20180128172054-7a43cd876e46 h1:wXG2bA8fO7Vv7lLk2PihFMTqmbT173Tje39oKzQ50Mo=
 github.com/marcinbor85/gohex v0.0.0-20180128172054-7a43cd876e46/go.mod h1:Pb6XcsXyropB9LNHhnqaknG/vEwYztLkQzVCHv8sQ3M=
+go.bug.st/serial.v1 v0.0.0-20180827123349-5f7892a7bb45 h1:mACY1anK6HNCZtm/DK2Rf2ZPHggVqeB0+7rY9Gl6wyI=
+go.bug.st/serial.v1 v0.0.0-20180827123349-5f7892a7bb45/go.mod h1:dRSl/CVCTf56CkXgJMDOdSwNfo2g1orOGE/gBGdvjZw=
 golang.org/x/net v0.0.0-20180724234803-3673e40ba225/go.mod h1:mL1N/T3taQHkDXs73rZJwtUhF3w3ftmwwsq0BUmARs4=
 golang.org/x/net v0.0.0-20190213061140-3a22650c66bd/go.mod h1:mL1N/T3taQHkDXs73rZJwtUhF3w3ftmwwsq0BUmARs4=
 golang.org/x/sync v0.0.0-20181108010431-42b317875d0f/go.mod h1:RxMgew5VJxzue5/jJTE5uejpjVlOe/izrB70Jof72aM=
+golang.org/x/sys v0.0.0-20191010194322-b09406accb47 h1:/XfQ9z7ib8eEJX2hdgFTZJ/ntt0swNk5oYBziWeTCvY=
+golang.org/x/sys v0.0.0-20191010194322-b09406accb47/go.mod h1:h1NjWce9XRLGQEsW7wpKNCjG9DtNlClVuFLEZdDNbEs=
 golang.org/x/text v0.3.0/go.mod h1:NqM8EUOU14njkJ3fqMW+pc6Ldnwhi/IjpwHt7yyuwOQ=
 golang.org/x/tools v0.0.0-20190227180812-8dcc6e70cdef h1:ymc9FeDom3RIEA3coKokSllBB1hRcMT0tZ1W3Jf9Ids=
 golang.org/x/tools v0.0.0-20190227180812-8dcc6e70cdef/go.mod h1:9Yl7xja0Znq3iFh3HoIrodX9oNMXvdceNzlUR8zjMvY=
@@ -14,3 +22,15 @@ tinygo.org/x/go-llvm v0.0.0-20190224120431-7707ae5d1261 h1:rJS2Hga39YAnm7DE4qrPm
 tinygo.org/x/go-llvm v0.0.0-20190224120431-7707ae5d1261/go.mod h1:fv1F0BSNpxMfCL0zF3M4OPFbgYHnhtB6ST0HvUtu/LE=
 tinygo.org/x/go-llvm v0.0.0-20190818154551-95bc4ffe1add h1:dFjMH1sLhYADg8UQm7DB56B7e+TfvAmWmEZLhyv3r/w=
 tinygo.org/x/go-llvm v0.0.0-20190818154551-95bc4ffe1add/go.mod h1:fv1F0BSNpxMfCL0zF3M4OPFbgYHnhtB6ST0HvUtu/LE=
+tinygo.org/x/go-llvm v0.0.0-20191103182207-90b6e4bdc0b9 h1:d6rAX39a3C0pKrY5HcojEGyN8w9ocU0v7X28lC/TRKU=
+tinygo.org/x/go-llvm v0.0.0-20191103182207-90b6e4bdc0b9/go.mod h1:fv1F0BSNpxMfCL0zF3M4OPFbgYHnhtB6ST0HvUtu/LE=
+tinygo.org/x/go-llvm v0.0.0-20191103200204-37e93e3f04e2 h1:Q5Hv3e5cLMGkiYwYgZL1Zrv6nb/EY+DJpRWrdO6ws6o=
+tinygo.org/x/go-llvm v0.0.0-20191103200204-37e93e3f04e2/go.mod h1:fv1F0BSNpxMfCL0zF3M4OPFbgYHnhtB6ST0HvUtu/LE=
+tinygo.org/x/go-llvm v0.0.0-20191113125529-bad6d01809e8 h1:9Bfvso+tTVQg16UzOA614NaYA4x8vsRBNtd3eBrXwp0=
+tinygo.org/x/go-llvm v0.0.0-20191113125529-bad6d01809e8/go.mod h1:fv1F0BSNpxMfCL0zF3M4OPFbgYHnhtB6ST0HvUtu/LE=
+tinygo.org/x/go-llvm v0.0.0-20191124211856-b2db3df3f257 h1:o8VDylrMN7gWemBMu8rEyuogKPhcLTdx5KrUAp9macc=
+tinygo.org/x/go-llvm v0.0.0-20191124211856-b2db3df3f257/go.mod h1:fv1F0BSNpxMfCL0zF3M4OPFbgYHnhtB6ST0HvUtu/LE=
+tinygo.org/x/go-llvm v0.0.0-20191215173731-ad71f3d24aae h1:s8J5EyxCkHxXB08UI3gk9W9IS/ekizRvSX+PfZxnAB0=
+tinygo.org/x/go-llvm v0.0.0-20191215173731-ad71f3d24aae/go.mod h1:fv1F0BSNpxMfCL0zF3M4OPFbgYHnhtB6ST0HvUtu/LE=
+tinygo.org/x/go-llvm v0.0.0-20200104190746-1ff21df33566 h1:a4y30bTf7U0zDA75v2PTL+XQ2OzJetj19gK8XwQpUNY=
+tinygo.org/x/go-llvm v0.0.0-20200104190746-1ff21df33566/go.mod h1:fv1F0BSNpxMfCL0zF3M4OPFbgYHnhtB6ST0HvUtu/LE=

goenv/goenv.go

@@ -0,0 +1,189 @@
+// Package goenv returns environment variables that are used in various parts of
+// the compiler. You can query it manually with the `tinygo env` subcommand.
+package goenv
+
+import (
+	"fmt"
+	"os"
+	"os/exec"
+	"os/user"
+	"path/filepath"
+	"runtime"
+)
+
+// Keys is a slice of all available environment variable keys.
+var Keys = []string{
+	"GOOS",
+	"GOARCH",
+	"GOROOT",
+	"GOPATH",
+	"GOCACHE",
+	"TINYGOROOT",
+}
+
+// TINYGOROOT is the path to the final location for checking tinygo files. If
+// unset (by a -X ldflag), then sourceDir() will fallback to the original build
+// directory.
+var TINYGOROOT string
+
+// Get returns a single environment variable, possibly calculating it on-demand.
+// The empty string is returned for unknown environment variables.
+func Get(name string) string {
+	switch name {
+	case "GOOS":
+		if dir := os.Getenv("GOOS"); dir != "" {
+			return dir
+		}
+		return runtime.GOOS
+	case "GOARCH":
+		if dir := os.Getenv("GOARCH"); dir != "" {
+			return dir
+		}
+		return runtime.GOARCH
+	case "GOROOT":
+		return getGoroot()
+	case "GOPATH":
+		if dir := os.Getenv("GOPATH"); dir != "" {
+			return dir
+		}
+
+		// fallback
+		home := getHomeDir()
+		return filepath.Join(home, "go")
+	case "GOCACHE":
+		// Get the cache directory, usually ~/.cache/tinygo
+		dir, err := os.UserCacheDir()
+		if err != nil {
+			panic("could not find cache dir: " + err.Error())
+		}
+		return filepath.Join(dir, "tinygo")
+	case "TINYGOROOT":
+		return sourceDir()
+	default:
+		return ""
+	}
+}
+
+// Return the TINYGOROOT, or exit with an error.
+func sourceDir() string {
+	// Use $TINYGOROOT as root, if available.
+	root := os.Getenv("TINYGOROOT")
+	if root != "" {
+		if !isSourceDir(root) {
+			fmt.Fprintln(os.Stderr, "error: $TINYGOROOT was not set to the correct root")
+			os.Exit(1)
+		}
+		return root
+	}
+
+	if TINYGOROOT != "" {
+		if !isSourceDir(TINYGOROOT) {
+			fmt.Fprintln(os.Stderr, "error: TINYGOROOT was not set to the correct root")
+			os.Exit(1)
+		}
+		return TINYGOROOT
+	}
+
+	// Find root from executable path.
+	path, err := os.Executable()
+	if err != nil {
+		// Very unlikely. Bail out if it happens.
+		panic("could not get executable path: " + err.Error())
+	}
+	root = filepath.Dir(filepath.Dir(path))
+	if isSourceDir(root) {
+		return root
+	}
+
+	// Fallback: use the original directory from where it was built
+	// https://stackoverflow.com/a/32163888/559350
+	_, path, _, _ = runtime.Caller(0)
+	root = filepath.Dir(filepath.Dir(path))
+	if isSourceDir(root) {
+		return root
+	}
+
+	fmt.Fprintln(os.Stderr, "error: could not autodetect root directory, set the TINYGOROOT environment variable to override")
+	os.Exit(1)
+	panic("unreachable")
+}
+
+// isSourceDir returns true if the directory looks like a TinyGo source directory.
+func isSourceDir(root string) bool {
+	_, err := os.Stat(filepath.Join(root, "src/runtime/internal/sys/zversion.go"))
+	if err != nil {
+		return false
+	}
+	_, err = os.Stat(filepath.Join(root, "src/device/arm/arm.go"))
+	return err == nil
+}
+
+func getHomeDir() string {
+	u, err := user.Current()
+	if err != nil {
+		panic("cannot get current user: " + err.Error())
+	}
+	if u.HomeDir == "" {
+		// This is very unlikely, so panic here.
+		// Not the nicest solution, however.
+		panic("could not find home directory")
+	}
+	return u.HomeDir
+}
+
+// getGoroot returns an appropriate GOROOT from various sources. If it can't be
+// found, it returns an empty string.
+func getGoroot() string {
+	goroot := os.Getenv("GOROOT")
+	if goroot != "" {
+		// An explicitly set GOROOT always has preference.
+		return goroot
+	}
+
+	// Check for the location of the 'go' binary and base GOROOT on that.
+	binpath, err := exec.LookPath("go")
+	if err == nil {
+		binpath, err = filepath.EvalSymlinks(binpath)
+		if err == nil {
+			goroot := filepath.Dir(filepath.Dir(binpath))
+			if isGoroot(goroot) {
+				return goroot
+			}
+		}
+	}
+
+	// Check what GOROOT was at compile time.
+	if isGoroot(runtime.GOROOT()) {
+		return runtime.GOROOT()
+	}
+
+	// Check for some standard locations, as a last resort.
+	var candidates []string
+	switch runtime.GOOS {
+	case "linux":
+		candidates = []string{
+			"/usr/local/go", // manually installed
+			"/usr/lib/go",   // from the distribution
+		}
+	case "darwin":
+		candidates = []string{
+			"/usr/local/go",             // manually installed
+			"/usr/local/opt/go/libexec", // from Homebrew
+		}
+	}
+
+	for _, candidate := range candidates {
+		if isGoroot(candidate) {
+			return candidate
+		}
+	}
+
+	// Can't find GOROOT...
+	return ""
+}
+
+// isGoroot checks whether the given path looks like a GOROOT.
+func isGoroot(goroot string) bool {
+	_, err := os.Stat(filepath.Join(goroot, "src", "runtime", "internal", "sys", "zversion.go"))
+	return err == nil
+}

interp/errors.go

@@ -3,15 +3,92 @@ package interp
 // This file provides useful types for errors encountered during IR evaluation.
 
 import (
+	"errors"
+	"go/scanner"
+	"go/token"
+	"path/filepath"
+
 	"tinygo.org/x/go-llvm"
 )
 
+// errUnreachable is returned when an unreachable instruction is executed. This
+// error should not be visible outside of the interp package.
+var errUnreachable = errors.New("interp: unreachable executed")
+
+// Unsupported is the specific error that is returned when an unsupported
+// instruction is hit while trying to interpret all initializers.
 type Unsupported struct {
-	Inst llvm.Value
+	ImportPath string
+	Inst       llvm.Value
+	Pos        token.Position
 }
 
 func (e Unsupported) Error() string {
 	// TODO: how to return the actual instruction string?
 	// It looks like LLVM provides no function for that...
-	return "interp: unsupported instruction"
+	return scanner.Error{
+		Pos: e.Pos,
+		Msg: "interp: unsupported instruction",
+	}.Error()
+}
+
+// unsupportedInstructionError returns a new "unsupported instruction" error for
+// the given instruction. It includes source location information, when
+// available.
+func (e *evalPackage) unsupportedInstructionError(inst llvm.Value) *Unsupported {
+	return &Unsupported{
+		ImportPath: e.packagePath,
+		Inst:       inst,
+		Pos:        getPosition(inst),
+	}
+}
+
+// Error encapsulates compile-time interpretation errors with an associated
+// import path. The errors may not have a precise location attached.
+type Error struct {
+	ImportPath string
+	Errs       []scanner.Error
+}
+
+// Error returns the string of the first error in the list of errors.
+func (e Error) Error() string {
+	return e.Errs[0].Error()
+}
+
+// errorAt returns an error value for the currently interpreted package at the
+// location of the instruction. The location information may not be complete as
+// it depends on debug information in the IR.
+func (e *evalPackage) errorAt(inst llvm.Value, msg string) Error {
+	return Error{
+		ImportPath: e.packagePath,
+		Errs:       []scanner.Error{errorAt(inst, msg)},
+	}
+}
+
+// errorAt returns an error value at the location of the instruction.
+// The location information may not be complete as it depends on debug
+// information in the IR.
+func errorAt(inst llvm.Value, msg string) scanner.Error {
+	return scanner.Error{
+		Pos: getPosition(inst),
+		Msg: msg,
+	}
+}
+
+// getPosition returns the position information for the given instruction, as
+// far as it is available.
+func getPosition(inst llvm.Value) token.Position {
+	if inst.IsAInstruction().IsNil() {
+		return token.Position{}
+	}
+	loc := inst.InstructionDebugLoc()
+	if loc.IsNil() {
+		return token.Position{}
+	}
+	file := loc.LocationScope().ScopeFile()
+	return token.Position{
+		Filename: filepath.Join(file.FileDirectory(), file.FileFilename()),
+		Line:     int(loc.LocationLine()),
+		Column:   int(loc.LocationColumn()),
+	}
 }

interp/frame.go

@@ -4,21 +4,17 @@ package interp
 // functions.
 
 import (
-	"errors"
 	"strings"
 
 	"tinygo.org/x/go-llvm"
 )
 
 type frame struct {
-	*Eval
-	fn      llvm.Value
-	pkgName string
-	locals  map[llvm.Value]Value
+	*evalPackage
+	fn     llvm.Value
+	locals map[llvm.Value]Value
 }
 
-var ErrUnreachable = errors.New("interp: unreachable executed")
-
 // evalBasicBlock evaluates a single basic block, returning the return value (if
 // ending with a ret instruction), a list of outgoing basic blocks (if not
 // ending with a ret instruction), or an error on failure.
@@ -79,13 +75,13 @@ func (fr *frame) evalBasicBlock(bb, incoming llvm.BasicBlock, indent string) (re
 				fr.locals[inst] = &LocalValue{fr.Eval, fr.builder.CreateXor(lhs, rhs, "")}
 
 			default:
-				return nil, nil, &Unsupported{inst}
+				return nil, nil, fr.unsupportedInstructionError(inst)
 			}
 
 		// Memory operators
 		case !inst.IsAAllocaInst().IsNil():
 			allocType := inst.Type().ElementType()
-			alloca := llvm.AddGlobal(fr.Mod, allocType, fr.pkgName+"$alloca")
+			alloca := llvm.AddGlobal(fr.Mod, allocType, fr.packagePath+"$alloca")
 			alloca.SetInitializer(llvm.ConstNull(allocType))
 			alloca.SetLinkage(llvm.InternalLinkage)
 			fr.locals[inst] = &LocalValue{
@@ -101,7 +97,7 @@ func (fr *frame) evalBasicBlock(bb, incoming llvm.BasicBlock, indent string) (re
 				value = operand.Load()
 			}
 			if value.Type() != inst.Type() {
-				panic("interp: load: type does not match")
+				return nil, nil, fr.errorAt(inst, "interp: load: type does not match")
 			}
 			fr.locals[inst] = fr.getValue(value)
 		case !inst.IsAStoreInst().IsNil():
@@ -125,15 +121,13 @@ func (fr *frame) evalBasicBlock(bb, incoming llvm.BasicBlock, indent string) (re
 					// Not a constant operation.
 					// This should be detected by the scanner, but isn't at the
 					// moment.
-					panic("todo: non-const gep")
+					return nil, nil, fr.errorAt(inst, "todo: non-const gep")
 				}
 				indices[i] = uint32(operand.Value().ZExtValue())
 			}
 			result := value.GetElementPtr(indices)
 			if result.Type() != inst.Type() {
-				println(" expected:", inst.Type().String())
-				println(" actual:  ", result.Type().String())
-				panic("interp: gep: type does not match")
+				return nil, nil, fr.errorAt(inst, "interp: gep: type does not match")
 			}
 			fr.locals[inst] = result
 
@@ -188,7 +182,7 @@ func (fr *frame) evalBasicBlock(bb, incoming llvm.BasicBlock, indent string) (re
 					}
 				}
 				// It is not possible in Go to bitcast a map value to a pointer.
-				panic("unimplemented: bitcast of map")
+				return nil, nil, fr.errorAt(inst, "unimplemented: bitcast of map")
 			}
 			value := fr.getLocal(operand).(*LocalValue)
 			fr.locals[inst] = &LocalValue{fr.Eval, fr.builder.CreateBitCast(value.Value(), inst.Type(), "")}
@@ -198,20 +192,28 @@ func (fr *frame) evalBasicBlock(bb, incoming llvm.BasicBlock, indent string) (re
 			lhs := fr.getLocal(inst.Operand(0)).(*LocalValue).Underlying
 			rhs := fr.getLocal(inst.Operand(1)).(*LocalValue).Underlying
 			predicate := inst.IntPredicate()
-			if predicate == llvm.IntEQ && lhs.Type().TypeKind() == llvm.PointerTypeKind {
-				// Unfortunately, the const propagation in the IR builder
-				// doesn't handle pointer compares of inttoptr values. So we
-				// implement it manually here.
-				lhsNil, ok1 := isPointerNil(lhs)
-				rhsNil, ok2 := isPointerNil(rhs)
+			if predicate == llvm.IntEQ {
+				var lhsZero, rhsZero bool
+				var ok1, ok2 bool
+				if lhs.Type().TypeKind() == llvm.PointerTypeKind {
+					// Unfortunately, the const propagation in the IR builder
+					// doesn't handle pointer compares of inttoptr values. So we
+					// implement it manually here.
+					lhsZero, ok1 = isPointerNil(lhs)
+					rhsZero, ok2 = isPointerNil(rhs)
+				}
+				if lhs.Type().TypeKind() == llvm.IntegerTypeKind {
+					lhsZero, ok1 = isZero(lhs)
+					rhsZero, ok2 = isZero(rhs)
+				}
 				if ok1 && ok2 {
-					if lhsNil && rhsNil {
-						// Both are nil, so this icmp is always evaluated to true.
+					if lhsZero && rhsZero {
+						// Both are zero, so this icmp is always evaluated to true.
 						fr.locals[inst] = &LocalValue{fr.Eval, llvm.ConstInt(fr.Mod.Context().Int1Type(), 1, false)}
 						continue
 					}
-					if lhsNil != rhsNil {
-						// Only one of them is nil, so this comparison must return false.
+					if lhsZero != rhsZero {
+						// Only one of them is zero, so this comparison must return false.
 						fr.locals[inst] = &LocalValue{fr.Eval, llvm.ConstInt(fr.Mod.Context().Int1Type(), 0, false)}
 						continue
 					}
@@ -247,12 +249,12 @@ func (fr *frame) evalBasicBlock(bb, incoming llvm.BasicBlock, indent string) (re
 				if size != typeSize {
 					// allocate an array
 					if size%typeSize != 0 {
-						return nil, nil, &Unsupported{inst}
+						return nil, nil, fr.unsupportedInstructionError(inst)
 					}
 					elementCount = int(size / typeSize)
 					allocType = llvm.ArrayType(allocType, elementCount)
 				}
-				alloc := llvm.AddGlobal(fr.Mod, allocType, fr.pkgName+"$alloc")
+				alloc := llvm.AddGlobal(fr.Mod, allocType, fr.packagePath+"$alloc")
 				alloc.SetInitializer(llvm.ConstNull(allocType))
 				alloc.SetLinkage(llvm.InternalLinkage)
 				result := &LocalValue{
@@ -270,24 +272,55 @@ func (fr *frame) evalBasicBlock(bb, incoming llvm.BasicBlock, indent string) (re
 				valueSize := inst.Operand(1).ZExtValue()
 				fr.locals[inst] = &MapValue{
 					Eval:      fr.Eval,
-					PkgName:   fr.pkgName,
+					PkgName:   fr.packagePath,
 					KeySize:   int(keySize),
 					ValueSize: int(valueSize),
 				}
 			case callee.Name() == "runtime.hashmapStringSet":
 				// set a string key in the map
-				m := fr.getLocal(inst.Operand(0)).(*MapValue)
-				// "key" is a Go string value, which in the TinyGo calling convention is split up
-				// into separate pointer and length parameters.
 				keyBuf := fr.getLocal(inst.Operand(1)).(*LocalValue)
 				keyLen := fr.getLocal(inst.Operand(2)).(*LocalValue)
 				valPtr := fr.getLocal(inst.Operand(3)).(*LocalValue)
+				m, ok := fr.getLocal(inst.Operand(0)).(*MapValue)
+				if !ok || !keyBuf.IsConstant() || !keyLen.IsConstant() || !valPtr.IsConstant() {
+					// The mapassign operation could not be done at compile
+					// time. Do it at runtime instead.
+					m := fr.getLocal(inst.Operand(0)).Value()
+					fr.markDirty(m)
+					llvmParams := []llvm.Value{
+						m,                                    // *runtime.hashmap
+						fr.getLocal(inst.Operand(1)).Value(), // key.ptr
+						fr.getLocal(inst.Operand(2)).Value(), // key.len
+						fr.getLocal(inst.Operand(3)).Value(), // value (unsafe.Pointer)
+						fr.getLocal(inst.Operand(4)).Value(), // context
+						fr.getLocal(inst.Operand(5)).Value(), // parentHandle
+					}
+					fr.builder.CreateCall(callee, llvmParams, "")
+					continue
+				}
+				// "key" is a Go string value, which in the TinyGo calling convention is split up
+				// into separate pointer and length parameters.
 				m.PutString(keyBuf, keyLen, valPtr)
 			case callee.Name() == "runtime.hashmapBinarySet":
 				// set a binary (int etc.) key in the map
-				m := fr.getLocal(inst.Operand(0)).(*MapValue)
 				keyBuf := fr.getLocal(inst.Operand(1)).(*LocalValue)
 				valPtr := fr.getLocal(inst.Operand(2)).(*LocalValue)
+				m, ok := fr.getLocal(inst.Operand(0)).(*MapValue)
+				if !ok || !keyBuf.IsConstant() || !valPtr.IsConstant() {
+					// The mapassign operation could not be done at compile
+					// time. Do it at runtime instead.
+					m := fr.getLocal(inst.Operand(0)).Value()
+					fr.markDirty(m)
+					llvmParams := []llvm.Value{
+						m,                                    // *runtime.hashmap
+						fr.getLocal(inst.Operand(1)).Value(), // key
+						fr.getLocal(inst.Operand(2)).Value(), // value
+						fr.getLocal(inst.Operand(3)).Value(), // context
+						fr.getLocal(inst.Operand(4)).Value(), // parentHandle
+					}
+					fr.builder.CreateCall(callee, llvmParams, "")
+					continue
+				}
 				m.PutBinary(keyBuf, valPtr)
 			case callee.Name() == "runtime.stringConcat":
 				// adding two strings together
@@ -304,7 +337,7 @@ func (fr *frame) evalBasicBlock(bb, incoming llvm.BasicBlock, indent string) (re
 				}
 				globalType := llvm.ArrayType(fr.Mod.Context().Int8Type(), len(result))
 				globalValue := llvm.ConstArray(fr.Mod.Context().Int8Type(), vals)
-				global := llvm.AddGlobal(fr.Mod, globalType, fr.pkgName+"$stringconcat")
+				global := llvm.AddGlobal(fr.Mod, globalType, fr.packagePath+"$stringconcat")
 				global.SetInitializer(globalValue)
 				global.SetLinkage(llvm.InternalLinkage)
 				global.SetGlobalConstant(true)
@@ -316,6 +349,49 @@ func (fr *frame) evalBasicBlock(bb, incoming llvm.BasicBlock, indent string) (re
 				ret = llvm.ConstInsertValue(ret, retPtr, []uint32{0})
 				ret = llvm.ConstInsertValue(ret, retLen, []uint32{1})
 				fr.locals[inst] = &LocalValue{fr.Eval, ret}
+			case callee.Name() == "runtime.sliceCopy":
+				elementSize := fr.getLocal(inst.Operand(4)).(*LocalValue).Value().ZExtValue()
+				dstArray := fr.getLocal(inst.Operand(0)).(*LocalValue).stripPointerCasts()
+				srcArray := fr.getLocal(inst.Operand(1)).(*LocalValue).stripPointerCasts()
+				dstLen := fr.getLocal(inst.Operand(2)).(*LocalValue)
+				srcLen := fr.getLocal(inst.Operand(3)).(*LocalValue)
+				if elementSize != 1 && dstArray.Type().ElementType().TypeKind() == llvm.ArrayTypeKind && srcArray.Type().ElementType().TypeKind() == llvm.ArrayTypeKind {
+					// Slice data pointers are created by adding a global array
+					// and getting the address of the first element using a GEP.
+					// However, before the compiler can pass it to
+					// runtime.sliceCopy, it has to perform a bitcast to a *i8,
+					// to make it a unsafe.Pointer. Now, when the IR builder
+					// sees a bitcast of a GEP with zero indices, it will make
+					// a bitcast of the original array instead of the GEP,
+					// which breaks our assumptions.
+					// Re-add this GEP, in the hope that it it is then of the correct type...
+					dstArray = dstArray.GetElementPtr([]uint32{0, 0}).(*LocalValue)
+					srcArray = srcArray.GetElementPtr([]uint32{0, 0}).(*LocalValue)
+				}
+				if fr.Eval.TargetData.TypeAllocSize(dstArray.Type().ElementType()) != elementSize {
+					return nil, nil, fr.errorAt(inst, "interp: slice dst element size does not match pointer type")
+				}
+				if fr.Eval.TargetData.TypeAllocSize(srcArray.Type().ElementType()) != elementSize {
+					return nil, nil, fr.errorAt(inst, "interp: slice src element size does not match pointer type")
+				}
+				if dstArray.Type() != srcArray.Type() {
+					return nil, nil, fr.errorAt(inst, "interp: slice element types don't match")
+				}
+				length := dstLen.Value().SExtValue()
+				if srcLength := srcLen.Value().SExtValue(); srcLength < length {
+					length = srcLength
+				}
+				if length < 0 {
+					return nil, nil, fr.errorAt(inst, "interp: trying to copy a slice with negative length?")
+				}
+				for i := int64(0); i < length; i++ {
+					// *dst = *src
+					dstArray.Store(srcArray.Load())
+					// dst++
+					dstArray = dstArray.GetElementPtr([]uint32{1}).(*LocalValue)
+					// src++
+					srcArray = srcArray.GetElementPtr([]uint32{1}).(*LocalValue)
+				}
 			case callee.Name() == "runtime.stringToBytes":
 				// convert a string to a []byte
 				bufPtr := fr.getLocal(inst.Operand(0))
@@ -327,7 +403,7 @@ func (fr *frame) evalBasicBlock(bb, incoming llvm.BasicBlock, indent string) (re
 				}
 				globalType := llvm.ArrayType(fr.Mod.Context().Int8Type(), len(result))
 				globalValue := llvm.ConstArray(fr.Mod.Context().Int8Type(), vals)
-				global := llvm.AddGlobal(fr.Mod, globalType, fr.pkgName+"$bytes")
+				global := llvm.AddGlobal(fr.Mod, globalType, fr.packagePath+"$bytes")
 				global.SetInitializer(globalValue)
 				global.SetLinkage(llvm.InternalLinkage)
 				global.SetGlobalConstant(true)
@@ -344,16 +420,16 @@ func (fr *frame) evalBasicBlock(bb, incoming llvm.BasicBlock, indent string) (re
 				typecode := fr.getLocal(inst.Operand(0)).(*LocalValue).Underlying
 				interfaceMethodSet := fr.getLocal(inst.Operand(1)).(*LocalValue).Underlying
 				if typecode.IsAConstantExpr().IsNil() || typecode.Opcode() != llvm.PtrToInt {
-					panic("interp: expected typecode to be a ptrtoint")
+					return nil, nil, fr.errorAt(inst, "interp: expected typecode to be a ptrtoint")
 				}
 				typecode = typecode.Operand(0)
 				if interfaceMethodSet.IsAConstantExpr().IsNil() || interfaceMethodSet.Opcode() != llvm.GetElementPtr {
-					panic("interp: expected method set in runtime.interfaceImplements to be a constant gep")
+					return nil, nil, fr.errorAt(inst, "interp: expected method set in runtime.interfaceImplements to be a constant gep")
 				}
 				interfaceMethodSet = interfaceMethodSet.Operand(0).Initializer()
 				methodSet := llvm.ConstExtractValue(typecode.Initializer(), []uint32{1})
 				if methodSet.IsAConstantExpr().IsNil() || methodSet.Opcode() != llvm.GetElementPtr {
-					panic("interp: expected method set to be a constant gep")
+					return nil, nil, fr.errorAt(inst, "interp: expected method set to be a constant gep")
 				}
 				methodSet = methodSet.Operand(0).Initializer()
 
@@ -382,6 +458,8 @@ func (fr *frame) evalBasicBlock(bb, incoming llvm.BasicBlock, indent string) (re
 				fr.locals[inst] = &LocalValue{fr.Eval, llvm.ConstInt(fr.Mod.Context().Int64Type(), 0, false)}
 			case callee.Name() == "llvm.dbg.value":
 				// do nothing
+			case strings.HasPrefix(callee.Name(), "llvm.lifetime."):
+				// do nothing
 			case callee.Name() == "runtime.trackPointer":
 				// do nothing
 			case strings.HasPrefix(callee.Name(), "runtime.print") || callee.Name() == "runtime._panic":
@@ -423,7 +501,10 @@ func (fr *frame) evalBasicBlock(bb, incoming llvm.BasicBlock, indent string) (re
 					params = append(params, local)
 				}
 				var ret Value
-				scanResult := fr.Eval.hasSideEffects(callee)
+				scanResult, err := fr.hasSideEffects(callee)
+				if err != nil {
+					return nil, nil, err
+				}
 				if scanResult.severity == sideEffectLimited || dirtyParams && scanResult.severity != sideEffectAll {
 					// Side effect is bounded. This means the operation invokes
 					// side effects (like calling an external function) but it
@@ -446,7 +527,7 @@ func (fr *frame) evalBasicBlock(bb, incoming llvm.BasicBlock, indent string) (re
 					//     compile time.
 					//   * Unbounded: cannot call at runtime so we'll try to
 					//     interpret anyway and hope for the best.
-					ret, err = fr.function(callee, params, fr.pkgName, indent+"    ")
+					ret, err = fr.function(callee, params, indent+"    ")
 					if err != nil {
 						return nil, nil, err
 					}
@@ -456,7 +537,7 @@ func (fr *frame) evalBasicBlock(bb, incoming llvm.BasicBlock, indent string) (re
 				}
 			default:
 				// function pointers, etc.
-				return nil, nil, &Unsupported{inst}
+				return nil, nil, fr.unsupportedInstructionError(inst)
 			}
 		case !inst.IsAExtractValueInst().IsNil():
 			agg := fr.getLocal(inst.Operand(0)).(*LocalValue) // must be constant
@@ -466,7 +547,7 @@ func (fr *frame) evalBasicBlock(bb, incoming llvm.BasicBlock, indent string) (re
 				fr.locals[inst] = fr.getValue(newValue)
 			} else {
 				if len(indices) != 1 {
-					return nil, nil, errors.New("cannot handle extractvalue with not exactly 1 index")
+					return nil, nil, fr.errorAt(inst, "interp: cannot handle extractvalue with not exactly 1 index")
 				}
 				fr.locals[inst] = &LocalValue{fr.Eval, fr.builder.CreateExtractValue(agg.Underlying, int(indices[0]), inst.Name())}
 			}
@@ -479,7 +560,7 @@ func (fr *frame) evalBasicBlock(bb, incoming llvm.BasicBlock, indent string) (re
 				fr.locals[inst] = &LocalValue{fr.Eval, newValue}
 			} else {
 				if len(indices) != 1 {
-					return nil, nil, errors.New("cannot handle insertvalue with not exactly 1 index")
+					return nil, nil, fr.errorAt(inst, "interp: cannot handle insertvalue with not exactly 1 index")
 				}
 				fr.locals[inst] = &LocalValue{fr.Eval, fr.builder.CreateInsertValue(agg.Underlying, val.Value(), int(indices[0]), inst.Name())}
 			}
@@ -492,17 +573,17 @@ func (fr *frame) evalBasicBlock(bb, incoming llvm.BasicBlock, indent string) (re
 			// conditional branch (if/then/else)
 			cond := fr.getLocal(inst.Operand(0)).Value()
 			if cond.Type() != fr.Mod.Context().Int1Type() {
-				panic("expected an i1 in a branch instruction")
+				return nil, nil, fr.errorAt(inst, "expected an i1 in a branch instruction")
 			}
 			thenBB := inst.Operand(1)
 			elseBB := inst.Operand(2)
 			if !cond.IsAInstruction().IsNil() {
-				return nil, nil, errors.New("interp: branch on a non-constant")
+				return nil, nil, fr.errorAt(inst, "interp: branch on a non-constant")
 			}
 			if !cond.IsAConstantExpr().IsNil() {
 				// This may happen when the instruction builder could not
 				// const-fold some instructions.
-				return nil, nil, errors.New("interp: branch on a non-const-propagated constant expression")
+				return nil, nil, fr.errorAt(inst, "interp: branch on a non-const-propagated constant expression")
 			}
 			switch cond {
 			case llvm.ConstInt(fr.Mod.Context().Int1Type(), 0, false): // false
@@ -510,7 +591,7 @@ func (fr *frame) evalBasicBlock(bb, incoming llvm.BasicBlock, indent string) (re
 			case llvm.ConstInt(fr.Mod.Context().Int1Type(), 1, false): // true
 				return nil, []llvm.Value{elseBB}, nil // else
 			default:
-				panic("branch was not true or false")
+				return nil, nil, fr.errorAt(inst, "branch was not true or false")
 			}
 		case !inst.IsABranchInst().IsNil() && inst.OperandsCount() == 1:
 			// unconditional branch (goto)
@@ -518,10 +599,11 @@ func (fr *frame) evalBasicBlock(bb, incoming llvm.BasicBlock, indent string) (re
 		case !inst.IsAUnreachableInst().IsNil():
 			// Unreachable was reached (e.g. after a call to panic()).
 			// Report this as an error, as it is not supposed to happen.
-			return nil, nil, ErrUnreachable
+			// This is a sentinel error value.
+			return nil, nil, errUnreachable
 
 		default:
-			return nil, nil, &Unsupported{inst}
+			return nil, nil, fr.unsupportedInstructionError(inst)
 		}
 	}
 
@@ -535,6 +617,7 @@ func (fr *frame) getLocal(v llvm.Value) Value {
 	} else if value := fr.getValue(v); value != nil {
 		return value
 	} else {
+		// This should not happen under normal circumstances.
 		panic("cannot find value")
 	}
 }

interp/interp.go

@@ -7,7 +7,6 @@ package interp
 // methods.
 
 import (
-	"errors"
 	"strings"
 
 	"tinygo.org/x/go-llvm"
@@ -22,9 +21,17 @@ type Eval struct {
 	sideEffectFuncs map[llvm.Value]*sideEffectResult // cache of side effect scan results
 }
 
+// evalPackage encapsulates the Eval type for just a single package. The Eval
+// type keeps state across the whole program, the evalPackage type keeps extra
+// state for the currently interpreted package.
+type evalPackage struct {
+	*Eval
+	packagePath string
+}
+
 // Run evaluates the function with the given name and then eliminates all
 // callers.
-func Run(mod llvm.Module, targetData llvm.TargetData, debug bool) error {
+func Run(mod llvm.Module, debug bool) error {
 	if debug {
 		println("\ncompile-time evaluation:")
 	}
@@ -32,7 +39,7 @@ func Run(mod llvm.Module, targetData llvm.TargetData, debug bool) error {
 	name := "runtime.initAll"
 	e := &Eval{
 		Mod:          mod,
-		TargetData:   targetData,
+		TargetData:   llvm.NewTargetData(mod.DataLayout()),
 		Debug:        debug,
 		dirtyGlobals: map[llvm.Value]struct{}{},
 	}
@@ -56,7 +63,7 @@ func Run(mod llvm.Module, targetData llvm.TargetData, debug bool) error {
 			break // ret void
 		}
 		if inst.IsACallInst().IsNil() || inst.CalledValue().IsAFunction().IsNil() {
-			return errors.New("expected all instructions in " + name + " to be direct calls")
+			return errorAt(inst, "interp: expected all instructions in "+name+" to be direct calls")
 		}
 		initCalls = append(initCalls, inst)
 	}
@@ -66,13 +73,17 @@ func Run(mod llvm.Module, targetData llvm.TargetData, debug bool) error {
 	for _, call := range initCalls {
 		initName := call.CalledValue().Name()
 		if !strings.HasSuffix(initName, ".init") {
-			return errors.New("expected all instructions in " + name + " to be *.init() calls")
+			return errorAt(call, "interp: expected all instructions in "+name+" to be *.init() calls")
 		}
 		pkgName := initName[:len(initName)-5]
 		fn := call.CalledValue()
 		call.EraseFromParentAsInstruction()
-		_, err := e.Function(fn, []Value{&LocalValue{e, undefPtr}, &LocalValue{e, undefPtr}}, pkgName)
-		if err == ErrUnreachable {
+		evalPkg := evalPackage{
+			Eval:        e,
+			packagePath: pkgName,
+		}
+		_, err := evalPkg.function(fn, []Value{&LocalValue{e, undefPtr}, &LocalValue{e, undefPtr}}, "")
+		if err == errUnreachable {
 			break
 		}
 		if err != nil {
@@ -83,16 +94,14 @@ func Run(mod llvm.Module, targetData llvm.TargetData, debug bool) error {
 	return nil
 }
 
-func (e *Eval) Function(fn llvm.Value, params []Value, pkgName string) (Value, error) {
-	return e.function(fn, params, pkgName, "")
-}
-
-func (e *Eval) function(fn llvm.Value, params []Value, pkgName, indent string) (Value, error) {
+// function interprets the given function. The params are the function params
+// and the indent is the string indentation to use when dumping all interpreted
+// instructions.
+func (e *evalPackage) function(fn llvm.Value, params []Value, indent string) (Value, error) {
 	fr := frame{
-		Eval:    e,
-		fn:      fn,
-		pkgName: pkgName,
-		locals:  make(map[llvm.Value]Value),
+		evalPackage: e,
+		fn:          fn,
+		locals:      make(map[llvm.Value]Value),
 	}
 	for i, param := range fn.Params() {
 		fr.locals[param] = params[i]

interp/interp_test.go

@@ -0,0 +1,102 @@
+package interp
+
+import (
+	"io/ioutil"
+	"os"
+	"strings"
+	"testing"
+
+	"tinygo.org/x/go-llvm"
+)
+
+func TestInterp(t *testing.T) {
+	for _, name := range []string{
+		"basic",
+		"slice-copy",
+		"consteval",
+		"map",
+	} {
+		name := name // make tc local to this closure
+		t.Run(name, func(t *testing.T) {
+			t.Parallel()
+			runTest(t, "testdata/"+name)
+		})
+	}
+}
+
+func runTest(t *testing.T, pathPrefix string) {
+	// Read the input IR.
+	ctx := llvm.NewContext()
+	buf, err := llvm.NewMemoryBufferFromFile(pathPrefix + ".ll")
+	os.Stat(pathPrefix + ".ll") // make sure this file is tracked by `go test` caching
+	if err != nil {
+		t.Fatalf("could not read file %s: %v", pathPrefix+".ll", err)
+	}
+	mod, err := ctx.ParseIR(buf)
+	if err != nil {
+		t.Fatalf("could not load module:\n%v", err)
+	}
+
+	// Perform the transform.
+	err = Run(mod, false)
+	if err != nil {
+		t.Fatal(err)
+	}
+
+	// Run some cleanup passes to get easy-to-read outputs.
+	pm := llvm.NewPassManager()
+	defer pm.Dispose()
+	pm.AddGlobalOptimizerPass()
+	pm.AddDeadStoreEliminationPass()
+	pm.Run(mod)
+
+	// Read the expected output IR.
+	out, err := ioutil.ReadFile(pathPrefix + ".out.ll")
+	if err != nil {
+		t.Fatalf("could not read output file %s: %v", pathPrefix+".out.ll", err)
+	}
+
+	// See whether the transform output matches with the expected output IR.
+	expected := string(out)
+	actual := mod.String()
+	if !fuzzyEqualIR(expected, actual) {
+		t.Logf("output does not match expected output:\n%s", actual)
+		t.Fail()
+	}
+}
+
+// fuzzyEqualIR returns true if the two LLVM IR strings passed in are roughly
+// equal. That means, only relevant lines are compared (excluding comments
+// etc.).
+func fuzzyEqualIR(s1, s2 string) bool {
+	lines1 := filterIrrelevantIRLines(strings.Split(s1, "\n"))
+	lines2 := filterIrrelevantIRLines(strings.Split(s2, "\n"))
+	if len(lines1) != len(lines2) {
+		return false
+	}
+	for i, line := range lines1 {
+		if line != lines2[i] {
+			return false
+		}
+	}
+
+	return true
+}
+
+// filterIrrelevantIRLines removes lines from the input slice of strings that
+// are not relevant in comparing IR. For example, empty lines and comments are
+// stripped out.
+func filterIrrelevantIRLines(lines []string) []string {
+	var out []string
+	for _, line := range lines {
+		line = strings.TrimSpace(line) // drop '\r' on Windows
+		if line == "" || line[0] == ';' {
+			continue
+		}
+		if strings.HasPrefix(line, "source_filename = ") {
+			continue
+		}
+		out = append(out, line)
+	}
+	return out
+}

interp/scan.go

@@ -1,11 +1,28 @@
 package interp
 
 import (
+	"strings"
+
 	"tinygo.org/x/go-llvm"
 )
 
 type sideEffectSeverity int
 
+func (severity sideEffectSeverity) String() string {
+	switch severity {
+	case sideEffectInProgress:
+		return "in progress"
+	case sideEffectNone:
+		return "none"
+	case sideEffectLimited:
+		return "limited"
+	case sideEffectAll:
+		return "all"
+	default:
+		return "unknown"
+	}
+}
+
 const (
 	sideEffectInProgress sideEffectSeverity = iota // computing side effects is in progress (for recursive functions)
 	sideEffectNone                                 // no side effects at all (pure)
@@ -23,28 +40,36 @@ type sideEffectResult struct {
 // hasSideEffects scans this function and all descendants, recursively. It
 // returns whether this function has side effects and if it does, which globals
 // it mentions anywhere in this function or any called functions.
-func (e *Eval) hasSideEffects(fn llvm.Value) *sideEffectResult {
-	switch fn.Name() {
-	case "runtime.alloc":
+func (e *evalPackage) hasSideEffects(fn llvm.Value) (*sideEffectResult, error) {
+	name := fn.Name()
+	switch {
+	case name == "runtime.alloc":
 		// Cannot be scanned but can be interpreted.
-		return &sideEffectResult{severity: sideEffectNone}
-	case "runtime.nanotime":
+		return &sideEffectResult{severity: sideEffectNone}, nil
+	case name == "runtime.nanotime":
 		// Fixed value at compile time.
-		return &sideEffectResult{severity: sideEffectNone}
-	case "runtime._panic":
-		return &sideEffectResult{severity: sideEffectLimited}
-	case "runtime.interfaceImplements":
-		return &sideEffectResult{severity: sideEffectNone}
-	case "runtime.trackPointer":
-		return &sideEffectResult{severity: sideEffectNone}
-	case "llvm.dbg.value":
-		return &sideEffectResult{severity: sideEffectNone}
+		return &sideEffectResult{severity: sideEffectNone}, nil
+	case name == "runtime._panic":
+		return &sideEffectResult{severity: sideEffectLimited}, nil
+	case name == "runtime.interfaceImplements":
+		return &sideEffectResult{severity: sideEffectNone}, nil
+	case name == "runtime.sliceCopy":
+		return &sideEffectResult{severity: sideEffectNone}, nil
+	case name == "runtime.trackPointer":
+		return &sideEffectResult{severity: sideEffectNone}, nil
+	case name == "llvm.dbg.value":
+		return &sideEffectResult{severity: sideEffectNone}, nil
+	case strings.HasPrefix(name, "llvm.lifetime."):
+		return &sideEffectResult{severity: sideEffectNone}, nil
+	}
+	if fn.IsDeclaration() {
+		return &sideEffectResult{severity: sideEffectLimited}, nil
 	}
 	if e.sideEffectFuncs == nil {
 		e.sideEffectFuncs = make(map[llvm.Value]*sideEffectResult)
 	}
 	if se, ok := e.sideEffectFuncs[fn]; ok {
-		return se
+		return se, nil
 	}
 	result := &sideEffectResult{
 		severity:        sideEffectInProgress,
@@ -55,6 +80,7 @@ func (e *Eval) hasSideEffects(fn llvm.Value) *sideEffectResult {
 	for bb := fn.EntryBasicBlock(); !bb.IsNil(); bb = llvm.NextBasicBlock(bb) {
 		for inst := bb.FirstInstruction(); !inst.IsNil(); inst = llvm.NextInstruction(inst) {
 			if inst.IsAInstruction().IsNil() {
+				// Should not happen in valid IR.
 				panic("not an instruction")
 			}
 
@@ -71,7 +97,7 @@ func (e *Eval) hasSideEffects(fn llvm.Value) *sideEffectResult {
 			switch inst.InstructionOpcode() {
 			case llvm.IndirectBr, llvm.Invoke:
 				// Not emitted by the compiler.
-				panic("unknown instructions")
+				return nil, e.errorAt(inst, "unknown instructions")
 			case llvm.Call:
 				child := inst.CalledValue()
 				if !child.IsAInlineAsm().IsNil() {
@@ -97,7 +123,10 @@ func (e *Eval) hasSideEffects(fn llvm.Value) *sideEffectResult {
 					}
 					continue
 				}
-				childSideEffects := e.hasSideEffects(child)
+				childSideEffects, err := e.hasSideEffects(child)
+				if err != nil {
+					return nil, err
+				}
 				switch childSideEffects.severity {
 				case sideEffectInProgress, sideEffectNone:
 					// no side effects or recursive function - continue scanning
@@ -139,7 +168,7 @@ func (e *Eval) hasSideEffects(fn llvm.Value) *sideEffectResult {
 		// No side effect was reported for this function.
 		result.severity = sideEffectNone
 	}
-	return result
+	return result, nil
 }
 
 // hasLocalSideEffects checks whether the given instruction flows into a branch
@@ -154,6 +183,7 @@ func (e *Eval) hasLocalSideEffects(dirtyLocals map[llvm.Value]struct{}, inst llv
 	for use := inst.FirstUse(); !use.IsNil(); use = use.NextUse() {
 		user := use.User()
 		if user.IsAInstruction().IsNil() {
+			// Should not happen in valid IR.
 			panic("user not an instruction")
 		}
 		switch user.InstructionOpcode() {
@@ -173,10 +203,13 @@ func (e *Eval) hasLocalSideEffects(dirtyLocals map[llvm.Value]struct{}, inst llv
 				// Already handled in (*Eval).hasSideEffects.
 				continue
 			}
-			// But a store might also store to an alloca, in which case all uses
-			// of the alloca (possibly indirect through a GEP, bitcast, etc.)
-			// must be marked dirty.
-			panic("todo: store")
+			// This store might affect all kinds of values. While it is
+			// certainly possible to traverse through all of them, the easiest
+			// option right now is to just assume the worst and say that this
+			// function has side effects.
+			// TODO: traverse through all stores and mark all relevant allocas /
+			// globals dirty.
+			return true
 		default:
 			// All instructions that take 0 or more operands (1 or more if it
 			// was a use) and produce a result.

interp/scan_test.go

@@ -0,0 +1,93 @@
+package interp
+
+import (
+	"os"
+	"sort"
+	"testing"
+
+	"tinygo.org/x/go-llvm"
+)
+
+var scanTestTable = []struct {
+	name            string
+	severity        sideEffectSeverity
+	mentionsGlobals []string
+}{
+	{"returnsConst", sideEffectNone, nil},
+	{"returnsArg", sideEffectNone, nil},
+	{"externalCallOnly", sideEffectNone, nil},
+	{"externalCallAndReturn", sideEffectLimited, nil},
+	{"externalCallBranch", sideEffectLimited, nil},
+	{"readCleanGlobal", sideEffectNone, []string{"cleanGlobalInt"}},
+	{"readDirtyGlobal", sideEffectLimited, []string{"dirtyGlobalInt"}},
+	{"callFunctionPointer", sideEffectAll, []string{"functionPointer"}},
+	{"getDirtyPointer", sideEffectLimited, nil},
+	{"storeToPointer", sideEffectLimited, nil},
+}
+
+func TestScan(t *testing.T) {
+	t.Parallel()
+
+	// Read the input IR.
+	path := "testdata/scan.ll"
+	ctx := llvm.NewContext()
+	buf, err := llvm.NewMemoryBufferFromFile(path)
+	os.Stat(path) // make sure this file is tracked by `go test` caching
+	if err != nil {
+		t.Fatalf("could not read file %s: %v", path, err)
+	}
+	mod, err := ctx.ParseIR(buf)
+	if err != nil {
+		t.Fatalf("could not load module:\n%v", err)
+	}
+
+	// Check all to-be-tested functions.
+	for _, tc := range scanTestTable {
+		// Create an eval object, for testing.
+		e := &Eval{
+			Mod:          mod,
+			TargetData:   llvm.NewTargetData(mod.DataLayout()),
+			dirtyGlobals: map[llvm.Value]struct{}{},
+		}
+
+		// Mark some globals dirty, for testing.
+		e.markDirty(mod.NamedGlobal("dirtyGlobalInt"))
+
+		// Scan for side effects.
+		fn := mod.NamedFunction(tc.name)
+		if fn.IsNil() {
+			t.Errorf("scan test: could not find tested function %s in the IR", tc.name)
+			continue
+		}
+		evalPkg := &evalPackage{e, "testdata"}
+		result, err := evalPkg.hasSideEffects(fn)
+		if err != nil {
+			t.Errorf("scan test: failed to scan %s for side effects: %v", fn.Name(), err)
+		}
+
+		// Check whether the result is what we expect.
+		if result.severity != tc.severity {
+			t.Errorf("scan test: function %s should have severity %s but it has %s", tc.name, tc.severity, result.severity)
+		}
+
+		// Check whether the mentioned globals match with what we'd expect.
+		mentionsGlobalNames := make([]string, 0, len(result.mentionsGlobals))
+		for global := range result.mentionsGlobals {
+			mentionsGlobalNames = append(mentionsGlobalNames, global.Name())
+		}
+		sort.Strings(mentionsGlobalNames)
+		globalsMismatch := false
+		if len(result.mentionsGlobals) != len(tc.mentionsGlobals) {
+			globalsMismatch = true
+		} else {
+			for i, globalName := range mentionsGlobalNames {
+				if tc.mentionsGlobals[i] != globalName {
+					globalsMismatch = true
+				}
+			}
+		}
+		if globalsMismatch {
+			t.Errorf("scan test: expected %s to mention globals %v, but it mentions globals %v", tc.name, tc.mentionsGlobals, mentionsGlobalNames)
+		}
+	}
+}

interp/testdata/basic.ll

@@ -0,0 +1,60 @@
+target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
+target triple = "x86_64--linux"
+
+@main.v1 = internal global i64 0
+@main.nonConst1 = global [4 x i64] zeroinitializer
+@main.nonConst2 = global i64 0
+
+declare void @runtime.printint64(i64) unnamed_addr
+
+declare void @runtime.printnl() unnamed_addr
+
+define void @runtime.initAll() unnamed_addr {
+entry:
+  call void @runtime.init()
+  call void @main.init()
+  ret void
+}
+
+define void @main() unnamed_addr {
+entry:
+  %0 = load i64, i64* @main.v1
+  call void @runtime.printint64(i64 %0)
+  call void @runtime.printnl()
+  ret void
+}
+
+define internal void @runtime.init() unnamed_addr {
+entry:
+  ret void
+}
+
+define internal void @main.init() unnamed_addr {
+entry:
+  store i64 3, i64* @main.v1
+  call void @"main.init#1"()
+
+  ; test the following pattern:
+  ;     func someValue() int // extern function
+  ;     var nonConst1 = [4]int{someValue(), 0, 0, 0}
+  %value1 = call i64 @someValue()
+  %gep1 = getelementptr [4 x i64], [4 x i64]* @main.nonConst1, i32 0, i32 0
+  store i64 %value1, i64* %gep1
+
+  ; Test that the global really is marked dirty:
+  ;     var nonConst2 = nonConst1[0]
+  %gep2 = getelementptr [4 x i64], [4 x i64]* @main.nonConst1, i32 0, i32 0
+  %value2 = load i64, i64* %gep2
+  store i64 %value2, i64* @main.nonConst2
+
+  ret void
+}
+
+define internal void @"main.init#1"() unnamed_addr {
+entry:
+  call void @runtime.printint64(i64 5)
+  call void @runtime.printnl()
+  ret void
+}
+
+declare i64 @someValue()

interp/testdata/basic.out.ll

@@ -0,0 +1,29 @@
+target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
+target triple = "x86_64--linux"
+
+@main.nonConst1 = local_unnamed_addr global [4 x i64] zeroinitializer
+@main.nonConst2 = local_unnamed_addr global i64 0
+
+declare void @runtime.printint64(i64) unnamed_addr
+
+declare void @runtime.printnl() unnamed_addr
+
+define void @runtime.initAll() unnamed_addr {
+entry:
+  call void @runtime.printint64(i64 5)
+  call void @runtime.printnl()
+  %value1 = call i64 @someValue()
+  store i64 %value1, i64* getelementptr inbounds ([4 x i64], [4 x i64]* @main.nonConst1, i32 0, i32 0)
+  %value2 = load i64, i64* getelementptr inbounds ([4 x i64], [4 x i64]* @main.nonConst1, i32 0, i32 0)
+  store i64 %value2, i64* @main.nonConst2
+  ret void
+}
+
+define void @main() unnamed_addr {
+entry:
+  call void @runtime.printint64(i64 3)
+  call void @runtime.printnl()
+  ret void
+}
+
+declare i64 @someValue() local_unnamed_addr

interp/testdata/consteval.ll

@@ -0,0 +1,42 @@
+target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
+target triple = "x86_64--linux"
+
+@intToPtrResult = global i8 0
+@ptrToIntResult = global i8 0
+
+define void @runtime.initAll() {
+  call void @main.init()
+  ret void
+}
+
+define internal void @main.init() {
+  call void @testIntToPtr()
+  call void @testPtrToInt()
+  ret void
+}
+
+define internal void @testIntToPtr() {
+  %nil = icmp eq i8* inttoptr (i64 1024 to i8*), null
+  br i1 %nil, label %a, label %b
+a:
+  ; should not be reached
+  store i8 1, i8* @intToPtrResult
+  ret void
+b:
+  ; should be reached
+  store i8 2, i8* @intToPtrResult
+  ret void
+}
+
+define internal void @testPtrToInt() {
+  %zero = icmp eq i64 ptrtoint (i8* @ptrToIntResult to i64), 0
+  br i1 %zero, label %a, label %b
+a:
+  ; should not be reached
+  store i8 1, i8* @ptrToIntResult
+  ret void
+b:
+  ; should be reached
+  store i8 2, i8* @ptrToIntResult
+  ret void
+}

interp/testdata/consteval.out.ll

@@ -0,0 +1,9 @@
+target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
+target triple = "x86_64--linux"
+
+@intToPtrResult = local_unnamed_addr global i8 2
+@ptrToIntResult = local_unnamed_addr global i8 2
+
+define void @runtime.initAll() local_unnamed_addr {
+  ret void
+}

interp/testdata/map.ll

@@ -0,0 +1,76 @@
+target datalayout = "e-m:e-p:32:32-Fi8-i64:64-v128:64:128-a:0:32-n32-S64"
+target triple = "armv6m-none-eabi"
+
+%runtime._string = type { i8*, i32 }
+%runtime.hashmap = type { %runtime.hashmap*, i8*, i32, i8, i8, i8 }
+
+@main.m = global %runtime.hashmap* null
+@main.binaryMap = global %runtime.hashmap* null
+@main.stringMap = global %runtime.hashmap* null
+@main.init.string = internal unnamed_addr constant [7 x i8] c"CONNECT"
+
+declare %runtime.hashmap* @runtime.hashmapMake(i8, i8, i32, i8* %context, i8* %parentHandle)
+declare void @runtime.hashmapBinarySet(%runtime.hashmap*, i8*, i8*, i8* %context, i8* %parentHandle)
+declare void @runtime.hashmapStringSet(%runtime.hashmap*, i8*, i32, i8*, i8* %context, i8* %parentHandle)
+declare void @llvm.lifetime.end.p0i8(i64, i8*)
+declare void @llvm.lifetime.start.p0i8(i64, i8*)
+
+define void @runtime.initAll() unnamed_addr {
+entry:
+  call void @main.init(i8* undef, i8* null)
+  ret void
+}
+
+define internal void @main.init(i8* %context, i8* %parentHandle) unnamed_addr {
+entry:
+; Test that hashmap optimizations generally work (even with lifetimes).
+  %hashmap.key = alloca i8
+  %hashmap.value = alloca %runtime._string
+  %0 = call %runtime.hashmap* @runtime.hashmapMake(i8 1, i8 8, i32 1, i8* undef, i8* null)
+  %hashmap.value.bitcast = bitcast %runtime._string* %hashmap.value to i8*
+  call void @llvm.lifetime.start.p0i8(i64 8, i8* %hashmap.value.bitcast)
+  store %runtime._string { i8* getelementptr inbounds ([7 x i8], [7 x i8]* @main.init.string, i32 0, i32 0), i32 7 }, %runtime._string* %hashmap.value
+  call void @llvm.lifetime.start.p0i8(i64 1, i8* %hashmap.key)
+  store i8 1, i8* %hashmap.key
+  call void @runtime.hashmapBinarySet(%runtime.hashmap* %0, i8* %hashmap.key, i8* %hashmap.value.bitcast, i8* undef, i8* null)
+  call void @llvm.lifetime.end.p0i8(i64 1, i8* %hashmap.key)
+  call void @llvm.lifetime.end.p0i8(i64 8, i8* %hashmap.value.bitcast)
+  store %runtime.hashmap* %0, %runtime.hashmap** @main.m
+
+  ; Other tests, that can be done in a separate function.
+  call void @main.testNonConstantBinarySet()
+  call void @main.testNonConstantStringSet()
+  ret void
+}
+
+; Test that a map loaded from a global can still be used for mapassign
+; operations (with binary keys).
+define internal void @main.testNonConstantBinarySet() {
+  %hashmap.key = alloca i8
+  %hashmap.value = alloca i8
+  ; Create hashmap from global. This breaks the normal hashmapBinarySet
+  ; optimization, to test the fallback.
+  %map.new = call %runtime.hashmap* @runtime.hashmapMake(i8 1, i8 1, i32 1, i8* undef, i8* null)
+  store %runtime.hashmap* %map.new, %runtime.hashmap** @main.binaryMap
+  %map = load %runtime.hashmap*, %runtime.hashmap** @main.binaryMap
+  ; Do the binary set to the newly loaded map.
+  store i8 1, i8* %hashmap.key
+  store i8 2, i8* %hashmap.value
+  call void @runtime.hashmapBinarySet(%runtime.hashmap* %map, i8* %hashmap.key, i8* %hashmap.value, i8* undef, i8* null)
+  ret void
+}
+
+; Test that a map loaded from a global can still be used for mapassign
+; operations (with string keys).
+define internal void @main.testNonConstantStringSet() {
+  %hashmap.value = alloca i8
+  ; Create hashmap from global. This breaks the normal hashmapStringSet
+  ; optimization, to test the fallback.
+  %map.new = call %runtime.hashmap* @runtime.hashmapMake(i8 8, i8 1, i32 1, i8* undef, i8* null)
+  store %runtime.hashmap* %map.new, %runtime.hashmap** @main.stringMap
+  %map = load %runtime.hashmap*, %runtime.hashmap** @main.stringMap
+  ; Do the string set to the newly loaded map.
+  store i8 2, i8* %hashmap.value
+  call void @runtime.hashmapStringSet(%runtime.hashmap* %map, i8* getelementptr inbounds ([7 x i8], [7 x i8]* @main.init.string, i32 0, i32 0), i32 7, i8* %hashmap.value, i8* undef, i8* null)
+  ret void
+}

interp/testdata/map.out.ll

@@ -0,0 +1,28 @@
+target datalayout = "e-m:e-p:32:32-Fi8-i64:64-v128:64:128-a:0:32-n32-S64"
+target triple = "armv6m-none-eabi"
+
+%runtime.hashmap = type { %runtime.hashmap*, i8*, i32, i8, i8, i8 }
+%runtime._string = type { i8*, i32 }
+
+@main.m = local_unnamed_addr global %runtime.hashmap* @"main$map"
+@main.binaryMap = local_unnamed_addr global %runtime.hashmap* @"main$map.4"
+@main.stringMap = local_unnamed_addr global %runtime.hashmap* @"main$map.6"
+@main.init.string = internal unnamed_addr constant [7 x i8] c"CONNECT"
+@"main$mapbucket" = internal unnamed_addr global { [8 x i8], i8*, [8 x i8], [8 x %runtime._string] } { [8 x i8] c"\04\00\00\00\00\00\00\00", i8* null, [8 x i8] c"\01\00\00\00\00\00\00\00", [8 x %runtime._string] [%runtime._string { i8* getelementptr inbounds ([7 x i8], [7 x i8]* @main.init.string, i32 0, i32 0), i32 7 }, %runtime._string zeroinitializer, %runtime._string zeroinitializer, %runtime._string zeroinitializer, %runtime._string zeroinitializer, %runtime._string zeroinitializer, %runtime._string zeroinitializer, %runtime._string zeroinitializer] }
+@"main$map" = internal unnamed_addr global %runtime.hashmap { %runtime.hashmap* null, i8* getelementptr inbounds ({ [8 x i8], i8*, [8 x i8], [8 x %runtime._string] }, { [8 x i8], i8*, [8 x i8], [8 x %runtime._string] }* @"main$mapbucket", i32 0, i32 0, i32 0), i32 1, i8 1, i8 8, i8 0 }
+@"main$alloca.2" = internal global i8 1
+@"main$alloca.3" = internal global i8 2
+@"main$map.4" = internal unnamed_addr global %runtime.hashmap { %runtime.hashmap* null, i8* null, i32 0, i8 1, i8 1, i8 0 }
+@"main$alloca.5" = internal global i8 2
+@"main$map.6" = internal unnamed_addr global %runtime.hashmap { %runtime.hashmap* null, i8* null, i32 0, i8 8, i8 1, i8 0 }
+
+declare void @runtime.hashmapBinarySet(%runtime.hashmap*, i8*, i8*, i8*, i8*) local_unnamed_addr
+
+declare void @runtime.hashmapStringSet(%runtime.hashmap*, i8*, i32, i8*, i8*, i8*) local_unnamed_addr
+
+define void @runtime.initAll() unnamed_addr {
+entry:
+  call void @runtime.hashmapBinarySet(%runtime.hashmap* @"main$map.4", i8* @"main$alloca.2", i8* @"main$alloca.3", i8* undef, i8* null)
+  call void @runtime.hashmapStringSet(%runtime.hashmap* @"main$map.6", i8* getelementptr inbounds ([7 x i8], [7 x i8]* @main.init.string, i32 0, i32 0), i32 7, i8* @"main$alloca.5", i8* undef, i8* null)
+  ret void
+}

interp/testdata/scan.ll

@@ -0,0 +1,61 @@
+target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
+target triple = "x86_64--linux"
+
+define i64 @returnsConst() {
+  ret i64 0
+}
+
+define i64 @returnsArg(i64 %arg) {
+  ret i64 %arg
+}
+
+declare i64 @externalCall()
+
+define i64 @externalCallOnly() {
+  %result = call i64 @externalCall()
+  ret i64 0
+}
+
+define i64 @externalCallAndReturn() {
+  %result = call i64 @externalCall()
+  ret i64 %result
+}
+
+define i64 @externalCallBranch() {
+  %result = call i64 @externalCall()
+  %zero = icmp eq i64 %result, 0
+  br i1 %zero, label %if.then, label %if.done
+
+if.then:
+  ret i64 2
+
+if.done:
+  ret i64 4
+}
+
+@cleanGlobalInt = global i64 5
+define i64 @readCleanGlobal() {
+  %global = load i64, i64* @cleanGlobalInt
+  ret i64 %global
+}
+
+@dirtyGlobalInt = global i64 5
+define i64 @readDirtyGlobal() {
+  %global = load i64, i64* @dirtyGlobalInt
+  ret i64 %global
+}
+
+declare i64* @getDirtyPointer()
+
+define void @storeToPointer() {
+  %ptr = call i64* @getDirtyPointer()
+  store i64 3, i64* %ptr
+  ret void
+}
+
+@functionPointer = global i64()* null
+define i64 @callFunctionPointer() {
+  %fp = load i64()*, i64()** @functionPointer
+  %result = call i64 %fp()
+  ret i64 %result
+}

interp/testdata/slice-copy.ll

@@ -0,0 +1,86 @@
+target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
+target triple = "x86_64--linux"
+
+@main.uint8SliceSrc.buf = internal global [2 x i8] c"\03d"
+@main.uint8SliceSrc = internal unnamed_addr global { i8*, i64, i64 } { i8* getelementptr inbounds ([2 x i8], [2 x i8]* @main.uint8SliceSrc.buf, i32 0, i32 0), i64 2, i64 2 }
+@main.uint8SliceDst = internal unnamed_addr global { i8*, i64, i64 } zeroinitializer
+@main.int16SliceSrc.buf = internal global [3 x i16] [i16 5, i16 123, i16 1024]
+@main.int16SliceSrc = internal unnamed_addr global { i16*, i64, i64 } { i16* getelementptr inbounds ([3 x i16], [3 x i16]* @main.int16SliceSrc.buf, i32 0, i32 0), i64 3, i64 3 }
+@main.int16SliceDst = internal unnamed_addr global { i16*, i64, i64 } zeroinitializer
+
+declare i64 @runtime.sliceCopy(i8* %dst, i8* %src, i64 %dstLen, i64 %srcLen, i64 %elemSize) unnamed_addr
+
+declare i8* @runtime.alloc(i64) unnamed_addr
+
+declare void @runtime.printuint8(i8)
+
+declare void @runtime.printint16(i16)
+
+define void @runtime.initAll() unnamed_addr {
+entry:
+  call void @main.init()
+  ret void
+}
+
+define void @main() unnamed_addr {
+entry:
+  ; print(uintSliceSrc[0])
+  %uint8SliceSrc.buf = load i8*, i8** getelementptr inbounds ({ i8*, i64, i64 }, { i8*, i64, i64 }* @main.uint8SliceSrc, i64 0, i32 0)
+  %uint8SliceSrc.val = load i8, i8* %uint8SliceSrc.buf
+  call void @runtime.printuint8(i8 %uint8SliceSrc.val)
+
+  ; print(uintSliceDst[0])
+  %uint8SliceDst.buf = load i8*, i8** getelementptr inbounds ({ i8*, i64, i64 }, { i8*, i64, i64 }* @main.uint8SliceDst, i64 0, i32 0)
+  %uint8SliceDst.val = load i8, i8* %uint8SliceDst.buf
+  call void @runtime.printuint8(i8 %uint8SliceDst.val)
+
+  ; print(int16SliceSrc[0])
+  %int16SliceSrc.buf = load i16*, i16** getelementptr inbounds ({ i16*, i64, i64 }, { i16*, i64, i64 }* @main.int16SliceSrc, i64 0, i32 0)
+  %int16SliceSrc.val = load i16, i16* %int16SliceSrc.buf
+  call void @runtime.printint16(i16 %int16SliceSrc.val)
+
+  ; print(int16SliceDst[0])
+  %int16SliceDst.buf = load i16*, i16** getelementptr inbounds ({ i16*, i64, i64 }, { i16*, i64, i64 }* @main.int16SliceDst, i64 0, i32 0)
+  %int16SliceDst.val = load i16, i16* %int16SliceDst.buf
+  call void @runtime.printint16(i16 %int16SliceDst.val)
+  ret void
+}
+
+define internal void @main.init() unnamed_addr {
+entry:
+  ; equivalent of:
+  ;     uint8SliceDst = make([]uint8, len(uint8SliceSrc))
+  %uint8SliceSrc = load { i8*, i64, i64 }, { i8*, i64, i64 }* @main.uint8SliceSrc
+  %uint8SliceSrc.len = extractvalue { i8*, i64, i64 } %uint8SliceSrc, 1
+  %uint8SliceDst.buf = call i8* @runtime.alloc(i64 %uint8SliceSrc.len)
+  %0 = insertvalue { i8*, i64, i64 } undef, i8* %uint8SliceDst.buf, 0
+  %1 = insertvalue { i8*, i64, i64 } %0, i64 %uint8SliceSrc.len, 1
+  %2 = insertvalue { i8*, i64, i64 } %1, i64 %uint8SliceSrc.len, 2
+  store { i8*, i64, i64 } %2, { i8*, i64, i64 }* @main.uint8SliceDst
+
+  ; equivalent of:
+  ;     copy(uint8SliceDst, uint8SliceSrc)
+  %uint8SliceSrc.buf = extractvalue { i8*, i64, i64 } %uint8SliceSrc, 0
+  %copy.n = call i64 @runtime.sliceCopy(i8* %uint8SliceDst.buf, i8* %uint8SliceSrc.buf, i64 %uint8SliceSrc.len, i64 %uint8SliceSrc.len, i64 1)
+
+  ; equivalent of:
+  ;     int16SliceDst = make([]int16, len(int16SliceSrc))
+  %int16SliceSrc = load { i16*, i64, i64 }, { i16*, i64, i64 }* @main.int16SliceSrc
+  %int16SliceSrc.len = extractvalue { i16*, i64, i64 } %int16SliceSrc, 1
+  %int16SliceSrc.len.bytes = mul i64 %int16SliceSrc.len, 2
+  %int16SliceDst.buf.raw = call i8* @runtime.alloc(i64 %int16SliceSrc.len.bytes)
+  %int16SliceDst.buf = bitcast i8* %int16SliceDst.buf.raw to i16*
+  %3 = insertvalue { i16*, i64, i64 } undef, i16* %int16SliceDst.buf, 0
+  %4 = insertvalue { i16*, i64, i64 } %3, i64 %int16SliceSrc.len, 1
+  %5 = insertvalue { i16*, i64, i64 } %4, i64 %int16SliceSrc.len, 2
+  store { i16*, i64, i64 } %5, { i16*, i64, i64 }* @main.int16SliceDst
+
+  ; equivalent of:
+  ;     copy(int16SliceDst, int16SliceSrc)
+  %int16SliceSrc.buf = extractvalue { i16*, i64, i64 } %int16SliceSrc, 0
+  %int16SliceSrc.buf.i8ptr = bitcast i16* %int16SliceSrc.buf to i8*
+  %int16SliceDst.buf.i8ptr = bitcast i16* %int16SliceDst.buf to i8*
+  %copy.n2 = call i64 @runtime.sliceCopy(i8* %int16SliceDst.buf.i8ptr, i8* %int16SliceSrc.buf.i8ptr, i64 %int16SliceSrc.len, i64 %int16SliceSrc.len, i64 2)
+
+  ret void
+}

interp/testdata/slice-copy.out.ll

@@ -0,0 +1,20 @@
+target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
+target triple = "x86_64--linux"
+
+declare void @runtime.printuint8(i8) local_unnamed_addr
+
+declare void @runtime.printint16(i16) local_unnamed_addr
+
+define void @runtime.initAll() unnamed_addr {
+entry:
+  ret void
+}
+
+define void @main() unnamed_addr {
+entry:
+  call void @runtime.printuint8(i8 3)
+  call void @runtime.printuint8(i8 3)
+  call void @runtime.printint16(i16 5)
+  call void @runtime.printint16(i16 5)
+  ret void
+}

interp/utils.go

@@ -60,9 +60,9 @@ func isPointerNil(v llvm.Value) (result bool, ok bool) {
 		case llvm.IntToPtr:
 			// Whether a constant inttoptr is nil is easy to
 			// determine.
-			operand := v.Operand(0)
-			if operand.IsConstant() {
-				return operand.ZExtValue() == 0, true
+			result, ok = isZero(v.Operand(0))
+			if ok {
+				return
 			}
 		case llvm.BitCast, llvm.GetElementPtr:
 			// These const instructions are just a kind of wrappers for the
@@ -74,5 +74,44 @@ func isPointerNil(v llvm.Value) (result bool, ok bool) {
 		// A constant pointer null is always null, of course.
 		return true, true
 	}
+	if !v.IsAGlobalValue().IsNil() {
+		// A global value is never null.
+		return false, true
+	}
 	return false, false // not valid
 }
+
+// isZero returns whether the value in v is the integer zero, and whether that
+// can be known right now.
+func isZero(v llvm.Value) (result bool, ok bool) {
+	if !v.IsAConstantExpr().IsNil() {
+		switch v.Opcode() {
+		case llvm.PtrToInt:
+			return isPointerNil(v.Operand(0))
+		}
+	}
+	if !v.IsAConstantInt().IsNil() {
+		val := v.ZExtValue()
+		return val == 0, true
+	}
+	return false, false // not valid
+}
+
+// unwrap returns the underlying value, with GEPs removed. This can be useful to
+// get the underlying global of a GEP pointer.
+func unwrap(value llvm.Value) llvm.Value {
+	for {
+		if !value.IsAConstantExpr().IsNil() {
+			switch value.Opcode() {
+			case llvm.GetElementPtr:
+				value = value.Operand(0)
+				continue
+			}
+		} else if !value.IsAGetElementPtrInst().IsNil() {
+			value = value.Operand(0)
+			continue
+		}
+		break
+	}
+	return value
+}

interp/values.go

@@ -36,7 +36,7 @@ func (v *LocalValue) Type() llvm.Type {
 }
 
 func (v *LocalValue) IsConstant() bool {
-	if _, ok := v.Eval.dirtyGlobals[v.Underlying]; ok {
+	if _, ok := v.Eval.dirtyGlobals[unwrap(v.Underlying)]; ok {
 		return false
 	}
 	return v.Underlying.IsConstant()
@@ -75,6 +75,11 @@ func (v *LocalValue) Store(value llvm.Value) {
 		}
 		return
 	}
+	if !value.IsConstant() {
+		v.MarkDirty()
+		v.Eval.builder.CreateStore(value, v.Underlying)
+		return
+	}
 	switch v.Underlying.Opcode() {
 	case llvm.GetElementPtr:
 		indices := v.getConstGEPIndices()
@@ -98,13 +103,33 @@ func (v *LocalValue) GetElementPtr(indices []uint32) Value {
 		gep := llvm.ConstGEP(v.Underlying, getLLVMIndices(int32Type, indices))
 		return &LocalValue{v.Eval, gep}
 	}
-	switch v.Underlying.Opcode() {
-	case llvm.GetElementPtr, llvm.IntToPtr:
-		int32Type := v.Underlying.Type().Context().Int32Type()
-		llvmIndices := getLLVMIndices(int32Type, indices)
-		return &LocalValue{v.Eval, llvm.ConstGEP(v.Underlying, llvmIndices)}
-	default:
-		panic("interp: GEP on a constant")
+	if !v.Underlying.IsAConstantExpr().IsNil() {
+		switch v.Underlying.Opcode() {
+		case llvm.GetElementPtr, llvm.IntToPtr, llvm.BitCast:
+			int32Type := v.Underlying.Type().Context().Int32Type()
+			llvmIndices := getLLVMIndices(int32Type, indices)
+			return &LocalValue{v.Eval, llvm.ConstGEP(v.Underlying, llvmIndices)}
+		}
+	}
+	panic("interp: unknown GEP")
+}
+
+// stripPointerCasts removes all const bitcasts from pointer values, if there
+// are any.
+func (v *LocalValue) stripPointerCasts() *LocalValue {
+	value := v.Underlying
+	for {
+		if !value.IsAConstantExpr().IsNil() {
+			switch value.Opcode() {
+			case llvm.BitCast:
+				value = value.Operand(0)
+				continue
+			}
+		}
+		return &LocalValue{
+			Eval:       v.Eval,
+			Underlying: value,
+		}
 	}
 }
 
@@ -130,13 +155,14 @@ func (v *LocalValue) getConstGEPIndices() []uint32 {
 // MarkDirty marks this global as dirty, meaning that every load from and store
 // to this global (from now on) must be performed at runtime.
 func (v *LocalValue) MarkDirty() {
-	if v.Underlying.IsAGlobalVariable().IsNil() {
+	underlying := unwrap(v.Underlying)
+	if underlying.IsAGlobalVariable().IsNil() {
 		panic("trying to mark a non-global as dirty")
 	}
 	if !v.IsConstant() {
 		return // already dirty
 	}
-	v.Eval.dirtyGlobals[v.Underlying] = struct{}{}
+	v.Eval.dirtyGlobals[underlying] = struct{}{}
 }
 
 // MapValue implements a Go map which is created at compile time and stored as a
@@ -341,10 +367,12 @@ func (v *MapValue) PutBinary(keyPtr, valPtr *LocalValue) {
 		}
 	}
 
-	if keyPtr.Underlying.Opcode() == llvm.BitCast {
-		keyPtr = &LocalValue{v.Eval, keyPtr.Underlying.Operand(0)}
-	} else if keyPtr.Underlying.Opcode() == llvm.GetElementPtr {
-		keyPtr = &LocalValue{v.Eval, keyPtr.Underlying.Operand(0)}
+	if !keyPtr.Underlying.IsAConstantExpr().IsNil() {
+		if keyPtr.Underlying.Opcode() == llvm.BitCast {
+			keyPtr = &LocalValue{v.Eval, keyPtr.Underlying.Operand(0)}
+		} else if keyPtr.Underlying.Opcode() == llvm.GetElementPtr {
+			keyPtr = &LocalValue{v.Eval, keyPtr.Underlying.Operand(0)}
+		}
 	}
 	key := keyPtr.Load()
 	if v.KeyType.IsNil() {

ir/ir.go

@@ -27,14 +27,13 @@ type Program struct {
 // Function or method.
 type Function struct {
 	*ssa.Function
-	LLVMFn    llvm.Value
-	module    string     // go:wasm-module
-	linkName  string     // go:linkname, go:export, go:interrupt
-	exported  bool       // go:export
-	nobounds  bool       // go:nobounds
-	flag      bool       // used by dead code elimination
-	interrupt bool       // go:interrupt
-	inline    InlineType // go:inline
+	LLVMFn   llvm.Value
+	module   string     // go:wasm-module
+	linkName string     // go:linkname, go:export
+	exported bool       // go:export
+	nobounds bool       // go:nobounds
+	flag     bool       // used by dead code elimination
+	inline   InlineType // go:inline
 }
 
 // Interface type that is at some point used in a type assert (to check whether
@@ -178,6 +177,9 @@ func (p *Program) AddPackage(pkg *ssa.Package) {
 }
 
 func (p *Program) addFunction(ssaFn *ssa.Function) {
+	if _, ok := p.functionMap[ssaFn]; ok {
+		return
+	}
 	f := &Function{Function: ssaFn}
 	f.parsePragmas()
 	p.Functions = append(p.Functions, f)
@@ -240,18 +242,6 @@ func (f *Function) parsePragmas() {
 				f.inline = InlineHint
 			case "//go:noinline":
 				f.inline = InlineNone
-			case "//go:interrupt":
-				if len(parts) != 2 {
-					continue
-				}
-				name := parts[1]
-				if strings.HasSuffix(name, "_vect") {
-					// AVR vector naming
-					name = "__vector_" + name[:len(name)-5]
-				}
-				f.linkName = name
-				f.exported = true
-				f.interrupt = true
 			case "//go:linkname":
 				if len(parts) != 3 || parts[1] != f.Name() {
 					continue
@@ -285,14 +275,6 @@ func (f *Function) IsExported() bool {
 	return f.exported || f.CName() != ""
 }
 
-// Return true for functions annotated with //go:interrupt. The function name is
-// already customized in LinkName() to hook up in the interrupt vector.
-//
-// On some platforms (like AVR), interrupts need a special compiler flag.
-func (f *Function) IsInterrupt() bool {
-	return f.interrupt
-}
-
 // Return the inline directive of this function.
 func (f *Function) Inline() InlineType {
 	return f.inline

loader/loader.go

@@ -6,6 +6,7 @@ import (
 	"go/ast"
 	"go/build"
 	"go/parser"
+	"go/scanner"
 	"go/token"
 	"go/types"
 	"os"
@@ -46,7 +47,7 @@ type Package struct {
 
 // Import loads the given package relative to srcDir (for the vendor directory).
 // It only loads the current package without recursion.
-func (p *Program) Import(path, srcDir string) (*Package, error) {
+func (p *Program) Import(path, srcDir string, pos token.Position) (*Package, error) {
 	if p.Packages == nil {
 		p.Packages = make(map[string]*Package)
 	}
@@ -59,7 +60,10 @@ func (p *Program) Import(path, srcDir string) (*Package, error) {
 	}
 	buildPkg, err := ctx.Import(path, srcDir, build.ImportComment)
 	if err != nil {
-		return nil, err
+		return nil, scanner.Error{
+			Pos: pos,
+			Msg: err.Error(), // TODO: define a new error type that will wrap the inner error
+		}
 	}
 	if existingPkg, ok := p.Packages[buildPkg.ImportPath]; ok {
 		// Already imported, or at least started the import.
@@ -317,9 +321,9 @@ func (p *Program) parseFile(path string, mode parser.Mode) (*ast.File, error) {
 	defer rd.Close()
 	relpath := path
 	if filepath.IsAbs(path) {
-		relpath, err = filepath.Rel(p.Dir, path)
-		if err != nil {
-			return nil, err
+		rp, err := filepath.Rel(p.Dir, path)
+		if err == nil {
+			relpath = rp
 		}
 	}
 	return parser.ParseFile(p.fset, relpath, rd, mode)
@@ -467,7 +471,14 @@ func (p *Package) importRecursively(includeTests bool) error {
 		if _, ok := p.Imports[to]; ok {
 			continue
 		}
-		importedPkg, err := p.Program.Import(to, p.Package.Dir)
+		// Find error location.
+		var pos token.Position
+		if len(p.Package.ImportPos[to]) > 0 {
+			pos = p.Package.ImportPos[to][0]
+		} else {
+			pos = token.Position{Filename: p.Package.ImportPath}
+		}
+		importedPkg, err := p.Program.Import(to, p.Package.Dir, pos)
 		if err != nil {
 			if err, ok := err.(*ImportCycleError); ok {
 				err.Packages = append([]string{p.ImportPath}, err.Packages...)

main_test.go

@@ -12,8 +12,11 @@ import (
 	"path/filepath"
 	"runtime"
 	"sort"
+	"sync"
 	"testing"
+	"time"
 
+	"github.com/tinygo-org/tinygo/compileopts"
 	"github.com/tinygo-org/tinygo/loader"
 )
 
@@ -38,17 +41,9 @@ func TestCompiler(t *testing.T) {
 
 	sort.Strings(matches)
 
-	// Create a temporary directory for test output files.
-	tmpdir, err := ioutil.TempDir("", "tinygo-test")
-	if err != nil {
-		t.Fatal("could not create temporary directory:", err)
-	}
-	defer os.RemoveAll(tmpdir)
-
-	t.Log("running tests on host...")
-	for _, path := range matches {
-		t.Run(path, func(t *testing.T) {
-			runTest(path, tmpdir, "", t)
+	if runtime.GOOS != "windows" {
+		t.Run("Host", func(t *testing.T) {
+			runPlatTests("", matches, t)
 		})
 	}
 
@@ -56,73 +51,100 @@ func TestCompiler(t *testing.T) {
 		return
 	}
 
-	t.Log("running tests for emulated cortex-m3...")
-	for _, path := range matches {
-		t.Run(path, func(t *testing.T) {
-			runTest(path, tmpdir, "qemu", t)
-		})
-	}
+	t.Run("EmulatedCortexM3", func(t *testing.T) {
+		runPlatTests("cortex-m-qemu", matches, t)
+	})
 
 	if runtime.GOOS == "linux" {
-		t.Log("running tests for linux/arm...")
-		for _, path := range matches {
-			if path == filepath.Join("testdata", "cgo")+string(filepath.Separator) {
-				continue // TODO: improve CGo
-			}
-			t.Run(path, func(t *testing.T) {
-				runTest(path, tmpdir, "arm--linux-gnueabihf", t)
-			})
-		}
-
-		t.Log("running tests for linux/arm64...")
-		for _, path := range matches {
-			if path == filepath.Join("testdata", "cgo")+string(filepath.Separator) {
-				continue // TODO: improve CGo
-			}
-			t.Run(path, func(t *testing.T) {
-				runTest(path, tmpdir, "aarch64--linux-gnu", t)
-			})
-		}
-
-		t.Log("running tests for WebAssembly...")
-		for _, path := range matches {
-			if path == filepath.Join("testdata", "gc.go") {
-				continue // known to fail
-			}
-			t.Run(path, func(t *testing.T) {
-				runTest(path, tmpdir, "wasm", t)
-			})
-		}
+		t.Run("ARMLinux", func(t *testing.T) {
+			runPlatTests("arm--linux-gnueabihf", matches, t)
+		})
+		t.Run("ARM64Linux", func(t *testing.T) {
+			runPlatTests("aarch64--linux-gnu", matches, t)
+		})
+		t.Run("WebAssembly", func(t *testing.T) {
+			runPlatTests("wasm", matches, t)
+		})
 	}
 }
 
-func runTest(path, tmpdir string, target string, t *testing.T) {
+func runPlatTests(target string, matches []string, t *testing.T) {
+	t.Parallel()
+
+	for _, path := range matches {
+		switch {
+		case target == "wasm":
+			// testdata/gc.go is known not to work on WebAssembly
+			if path == filepath.Join("testdata", "gc.go") {
+				continue
+			}
+		case target == "":
+			// run all tests on host
+		case target == "cortex-m-qemu":
+			// all tests are supported
+		default:
+			// cross-compilation of cgo is not yet supported
+			if path == filepath.Join("testdata", "cgo")+string(filepath.Separator) {
+				continue
+			}
+		}
+
+		t.Run(filepath.Base(path), func(t *testing.T) {
+			t.Parallel()
+
+			runTest(path, target, t)
+		})
+	}
+}
+
+// Due to some problems with LLD, we cannot run links in parallel, or in parallel with compiles.
+// Therefore, we put a lock around builds and run everything else in parallel.
+var buildLock sync.Mutex
+
+// runBuild is a thread-safe wrapper around Build.
+func runBuild(src, out string, opts *compileopts.Options) error {
+	buildLock.Lock()
+	defer buildLock.Unlock()
+
+	return Build(src, out, opts)
+}
+
+func runTest(path, target string, t *testing.T) {
 	// Get the expected output for this test.
 	txtpath := path[:len(path)-3] + ".txt"
 	if path[len(path)-1] == os.PathSeparator {
 		txtpath = path + "out.txt"
 	}
-	f, err := os.Open(txtpath)
-	if err != nil {
-		t.Fatal("could not open expected output file:", err)
-	}
-	expected, err := ioutil.ReadAll(f)
+	expected, err := ioutil.ReadFile(txtpath)
 	if err != nil {
 		t.Fatal("could not read expected output file:", err)
 	}
 
+	// Create a temporary directory for test output files.
+	tmpdir, err := ioutil.TempDir("", "tinygo-test")
+	if err != nil {
+		t.Fatal("could not create temporary directory:", err)
+	}
+	defer func() {
+		rerr := os.RemoveAll(tmpdir)
+		if rerr != nil {
+			t.Errorf("failed to remove temporary directory %q: %s", tmpdir, rerr.Error())
+		}
+	}()
+
 	// Build the test binary.
-	config := &BuildConfig{
-		opt:        "z",
-		printIR:    false,
-		dumpSSA:    false,
-		verifyIR:   true,
-		debug:      false,
-		printSizes: "",
-		wasmAbi:    "js",
+	config := &compileopts.Options{
+		Target:     target,
+		Opt:        "z",
+		PrintIR:    false,
+		DumpSSA:    false,
+		VerifyIR:   true,
+		Debug:      false,
+		PrintSizes: "",
+		WasmAbi:    "js",
 	}
 	binary := filepath.Join(tmpdir, "test")
-	err = Build("./"+path, binary, target, config)
+	err = runBuild("./"+path, binary, config)
 	if err != nil {
 		if errLoader, ok := err.(loader.Errors); ok {
 			for _, err := range errLoader.Errs {
@@ -136,11 +158,13 @@ func runTest(path, tmpdir string, target string, t *testing.T) {
 	}
 
 	// Run the test.
+	runComplete := make(chan struct{})
 	var cmd *exec.Cmd
+	ranTooLong := false
 	if target == "" {
 		cmd = exec.Command(binary)
 	} else {
-		spec, err := LoadTarget(target)
+		spec, err := compileopts.LoadTarget(target)
 		if err != nil {
 			t.Fatal("failed to load target spec:", err)
 		}
@@ -152,16 +176,39 @@ func runTest(path, tmpdir string, target string, t *testing.T) {
 	}
 	stdout := &bytes.Buffer{}
 	cmd.Stdout = stdout
-	if target != "" {
-		cmd.Stderr = os.Stderr
+	cmd.Stderr = os.Stderr
+	err = cmd.Start()
+	if err != nil {
+		t.Fatal("failed to start:", err)
 	}
-	err = cmd.Run()
+	go func() {
+		// Terminate the process if it runs too long.
+		timer := time.NewTimer(1 * time.Second)
+		select {
+		case <-runComplete:
+			timer.Stop()
+		case <-timer.C:
+			ranTooLong = true
+			if runtime.GOOS == "windows" {
+				cmd.Process.Signal(os.Kill) // Windows doesn't support SIGINT.
+			} else {
+				cmd.Process.Signal(os.Interrupt)
+			}
+		}
+	}()
+	err = cmd.Wait()
 	if _, ok := err.(*exec.ExitError); ok && target != "" {
 		err = nil // workaround for QEMU
 	}
+	close(runComplete)
+
+	if ranTooLong {
+		stdout.WriteString("--- test ran too long, terminating...\n")
+	}
 
 	// putchar() prints CRLF, convert it to LF.
 	actual := bytes.Replace(stdout.Bytes(), []byte{'\r', '\n'}, []byte{'\n'}, -1)
+	expected = bytes.Replace(expected, []byte{'\r', '\n'}, []byte{'\n'}, -1) // for Windows
 
 	// Check whether the command ran successfully.
 	fail := false

src/device/arm/arm.go

@@ -30,6 +30,7 @@
 package arm
 
 import (
+	"errors"
 	"runtime/volatile"
 	"unsafe"
 )
@@ -72,6 +73,7 @@ func SVCall4(num uintptr, a1, a2, a3, a4 interface{}) uintptr
 
 const (
 	SCS_BASE  = 0xE000E000
+	SYST_BASE = SCS_BASE + 0x0010
 	NVIC_BASE = SCS_BASE + 0x0100
 	SCB_BASE  = SCS_BASE + 0x0D00
 )
@@ -119,6 +121,53 @@ type NVIC_Type struct {
 
 var NVIC = (*NVIC_Type)(unsafe.Pointer(uintptr(NVIC_BASE)))
 
+// System Timer (SYST)
+//
+// Source: https://static.docs.arm.com/ddi0403/e/DDI0403E_d_armv7m_arm.pdf B3.3
+type SYST_Type struct {
+	SYST_CSR   volatile.Register32
+	SYST_RVR   volatile.Register32
+	SYST_CVR   volatile.Register32
+	SYST_CALIB volatile.Register32
+}
+
+var SYST = (*SYST_Type)(unsafe.Pointer(uintptr(SYST_BASE)))
+
+// Bitfields for SYST: System Timer
+const (
+	// SYST.SYST_CSR: SysTick Control and Status Register
+	SYST_CSR_ENABLE_Pos    = 0x0     // Position of ENABLE field.
+	SYST_CSR_ENABLE_Msk    = 0x1     // Bit mask of ENABLE field.
+	SYST_CSR_ENABLE        = 0x1     // Bit ENABLE.
+	SYST_CSR_TICKINT_Pos   = 0x1     // Position of TICKINT field.
+	SYST_CSR_TICKINT_Msk   = 0x2     // Bit mask of TICKINT field.
+	SYST_CSR_TICKINT       = 0x2     // Bit TICKINT.
+	SYST_CSR_CLKSOURCE_Pos = 0x2     // Position of CLKSOURCE field.
+	SYST_CSR_CLKSOURCE_Msk = 0x4     // Bit mask of CLKSOURCE field.
+	SYST_CSR_CLKSOURCE     = 0x4     // Bit CLKSOURCE.
+	SYST_CSR_COUNTFLAG_Pos = 0x10    // Position of COUNTFLAG field.
+	SYST_CSR_COUNTFLAG_Msk = 0x10000 // Bit mask of COUNTFLAG field.
+	SYST_CSR_COUNTFLAG     = 0x10000 // Bit COUNTFLAG.
+
+	// SYST.SYST_RVR: SysTick Reload Value Register
+	SYST_RVR_RELOAD_Pos = 0x0      // Position of RELOAD field.
+	SYST_RVR_RELOAD_Msk = 0xffffff // Bit mask of RELOAD field.
+
+	// SYST.SYST_CVR: SysTick Current Value Register
+	SYST_CVR_CURRENT_Pos = 0x0      // Position of CURRENT field.
+	SYST_CVR_CURRENT_Msk = 0xffffff // Bit mask of CURRENT field.
+
+	// SYST.SYST_CALIB: SysTick Calibration Value Register
+	SYST_CALIB_TENMS_Pos = 0x0        // Position of TENMS field.
+	SYST_CALIB_TENMS_Msk = 0xffffff   // Bit mask of TENMS field.
+	SYST_CALIB_SKEW_Pos  = 0x1e       // Position of SKEW field.
+	SYST_CALIB_SKEW_Msk  = 0x40000000 // Bit mask of SKEW field.
+	SYST_CALIB_SKEW      = 0x40000000 // Bit SKEW.
+	SYST_CALIB_NOREF_Pos = 0x1f       // Position of NOREF field.
+	SYST_CALIB_NOREF_Msk = 0x80000000 // Bit mask of NOREF field.
+	SYST_CALIB_NOREF     = 0x80000000 // Bit NOREF.
+)
+
 // Enable the given interrupt number.
 func EnableIRQ(irq uint32) {
 	NVIC.ISER[irq>>5].Set(1 << (irq & 0x1F))
@@ -169,3 +218,28 @@ func SystemReset() {
 		Asm("wfi")
 	}
 }
+
+// Set up the system timer to generate periodic tick events.
+// This will cause SysTick_Handler to fire once per tick.
+// The cyclecount parameter is a counter value which can range from 0 to
+// 0xffffff.  A value of 0 disables the timer.
+func SetupSystemTimer(cyclecount uint32) error {
+	// turn it off
+	SYST.SYST_CSR.ClearBits(SYST_CSR_TICKINT | SYST_CSR_ENABLE)
+	if cyclecount == 0 {
+		// leave the system timer turned off.
+		return nil
+	}
+	if cyclecount&SYST_RVR_RELOAD_Msk != cyclecount {
+		// The cycle refresh register is only 24 bits wide.  The user-specified value will overflow.
+		return errors.New("requested cycle count is too large, overflows 24 bit counter")
+	}
+
+	// set refresh count
+	SYST.SYST_RVR.Set(cyclecount)
+	// set current counter value
+	SYST.SYST_CVR.Set(cyclecount)
+	// enable clock, enable SysTick interrupt when clock reaches 0, run it off of the processor clock
+	SYST.SYST_CSR.SetBits(SYST_CSR_TICKINT | SYST_CSR_ENABLE | SYST_CSR_CLKSOURCE)
+	return nil
+}

src/device/arm/cortexm.s

@@ -19,3 +19,12 @@ HardFault_Handler:
 
     // Continue handling this error in Go.
     bl handleHardFault
+
+// This is a convenience function for semihosting support.
+// At some point, this should be replaced by inline assembly.
+.section .text.SemihostingCall
+.global  SemihostingCall
+.type    SemihostingCall, %function
+SemihostingCall:
+    bkpt 0xab
+    bx   lr