The Actions cache of the repository was at 10.24 GB against a quota of
10 GB, so GitHub removed entries by least recent use. The LLVM entries
were among the first to go, and CI then built LLVM again although the
LLVM version did not change.
The Docker buildx layer blobs held 6.0 GB of the quota. All LLVM caches
together held 1.97 GB. Move the Docker layer cache to a registry cache in
GHCR, which does not use the Actions quota. Both jobs have the necessary
permission and login already.
Pin the LLVM commit in llvm-version.txt and fetch that commit, in place
of a clone of the branch tip tinygo_22.x. Add a hash of that file to each
LLVM cache key, so a change of the LLVM version invalidates the caches
and no other change does.
Also rename the LLVM image tags from llvm-20 to llvm-22.
* GNUmakefile: split into topic files in make/
The GNUmakefile had 1295 lines and mixed build configuration, LLVM
bootstrapping, device generation, four test suites, the smoke tests,
release packaging, and lint tools. The smoke tests alone were 500 lines.
Move each part into its own file in make/ and include them from
GNUmakefile. config.mk must be included first because the other files
use its variables in immediate assignments and conditionals.
There is no change in behavior. The parsed make database is identical
for the default build and for ASSERT=1, STATIC=1, XTENSA=0, STM32=0,
WASM=0, and CROSS=aarch64-linux-gnu, except for MAKEFILE_LIST and the
.PHONY list, which now also includes targets that were phony but not
declared.
The Dockerfile copies GNUmakefile alone before it builds LLVM, to keep
that layer independent of the source tree. It must copy make/ too.
* make/smoketest.mk: split into groups and add smoketest-quick
The smoke test was one recipe of about 500 lines with 236 builds that
always ran in sequence. Split it at the group boundaries that were
already there, so that:
- `make -j smoketest` builds the groups in parallel. A full run goes
from 325 to 91 seconds on a 32 core machine.
- CI can shard the groups across runners.
Each group writes to its own name in build/smoke/, because all builds
wrote to test.hex before and would overwrite each other in a parallel
build. The output extension selects the format, so it stays per line.
Add smoketest-quick, which builds one board for each processor
architecture. The full smoke test answers "can TinyGo build for every
board", which does not depend on the host OS, so it only needs to run
on one OS. The other jobs use smoketest-quick.
The comment above the esp32c3 group had 4 spaces of indentation instead
of a tab. That was harmless in the middle of a recipe, but it is now the
first line of a group, where it would stop the recipe from starting.
The set of build commands is unchanged for the default flags and for
XTENSA=0, STM32=0, and WASM=0. All 226 checksums are the same as before,
for a sequential build and for `make -j16`.
* ci: run the full smoke test once, on Linux only
The smoke test ran six times for each push: twice on Linux, twice on
macOS, once on Windows, and once in the compatibility test. Together
that was about 97 minutes of the CI time.
The smoke test checks that TinyGo can build a binary for each board.
That does not depend on the host OS. The only part that does is one
build behind a Windows check.
Add a smoketest-linux job that runs the full set, split across four
runners that use the tarball from the build-linux job. The groups are
balanced with the measured build time of each group. Remove the full
smoke test from test-linux-build, which the new job replaces, and use
smoketest-quick for the other four jobs.
Expected result: about 55 to 39 minutes for the slowest workflow, and
about 97 to 35 minutes of total smoke test time.
* make/smoketest.mk: build nintendoswitch in smoketest-quick
It is the only target that uses the aarch64 LLVM triple. With it,
smoketest-quick covers all 13 triples that the full smoke test uses.
This switches the Espressif fork from LLVM 19 to LLVM 20, so we can use
the improvements made between those LLVM versions. It also better aligns
with the system-LLVM build method, which currently also defaults to LLVM
20.
Note that this disables the machine outliner for RISC-V. It appears
there's a bug in there somewhere, with the machine outliner enabled the
crypto/elliptic package tests fail with -target=riscv-qemu.
This should ideally be investigated and reported upstream.
Instead of relying on a build when TinyGo is being built, do it like all
other libraries when it is needed.
This brings a few benefits:
* No more running `make wasi-libc` on the command line as a special
case for WebAssembly. The generic `git submodule update --init` step
is now enough for wasi-libc.
* It becomes much easier to customize the build per system. For
example: include/exclude malloc as needed, disable/enable bulk
memory operations per target, etc.
This adds linux/mipsle (little endian Mips) support to TinyGo.
It also adds experimental linux/mips (big-endian) support. It doesn't
quite work yet, some parts of the standard library (like the reflect
package) currently seem to assume a little-endian system.
Node.js 16 is no longer supported, so we can drop support for it as
well.
This also means updating a whole lot of GitHub Actions versions, because
they were updated to work on Node.js 20 instead. For most actions this
should be a relatively small change, but the upload-aftifact action has
had some major changes (which should generally improve things a lot).
This commit adds support for LLVM 16 and switches to it by default. That
means three LLVM versions are supported at the same time: LLVM 14, 15,
and 16.
This commit includes work by QuLogic:
* Part of this work was based on a PR by QuLogic:
https://github.com/tinygo-org/tinygo/pull/3649
But I also had parts of this already implemented in an old branch I
already made for LLVM 16.
* QuLogic also provided a CGo fix here, which is also incorporated in
this commit:
https://github.com/tinygo-org/tinygo/pull/3869
The difference with the original PR by QuLogic is that this commit is
more complete:
* It switches to LLVM 16 by default.
* It updates some things to also make it work with a self-built LLVM.
* It fixes the CGo bug in a slightly different way, and also fixes
another one not included in the original PR.
* It does not keep compiler tests passing on older LLVM versions. I
have found this to be quite burdensome and therefore don't generally
do this - the smoke tests should hopefully catch most regressions.
The Linux artifacts have clear names (linux-amd64-double-zipped etc),
but the MacOS and Windows ones didn't. This patch renames these artifact
names to be more readable, especially when downloading the artifacts.
This test only applies when using the built-in LLVM version. This way,
we have a stable LLVM version to test against. Distribution versions of
LLVM (especially Debian) tend to be patched in a way that affect the
results.
This should fix a number of concurrency/threading issues.
I had to force-disable concurrency in the linker using a hack. I'm not
entirely sure what the cause is, possibly the MinGW version (version 12
appears to work for me, while version 11 as used on the GitHub runner
image seems to be broken).
There are a few ways to fix this in a better way:
* Fix the underlying cause (possibly by upgrading to MinGW-w64 12).
* Add the `--threads` flag to the LLD MinGW linker, so we can use a
regular parameter instead of this hack.
deadprogram
Ayke van Laethem
Damian Gryski
Elliott Sales de Andrade
Randy Reddig