tinygo-org/drivers
1
0
Fork 0
mirror of https://github.com/tinygo-org/drivers.git synced 2026-08-19 22:23:58 +00:00
Code Issues Packages Projects Releases 10 Wiki Activity

image: add support for image/jpeg and image/png (#303)

Browse Source 
* Copy from go1.17 image package
* Remove unnecessary files
* Reduce memory usage
* Add examples/ili9341/slideshow
* image: add ./image/README.md
* image: change convert2bin to . /cmd
* Makefile: add ./cmd to NOTEST
This commit is contained in:
sago35
2021-09-18 19:18:02 +09:00
committed by GitHub
parent 259593e608
commit f2e5278965
52 changed files with 19324 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files
image/internal/compress/flate/deflate.go
+748
View File
@@ -0,0 +1,748 @@
// Copyright 2009 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.
package flate
import (
"fmt"
"io"
"math"
)
const (
NoCompression = 0
BestSpeed = 1
BestCompression = 9
DefaultCompression = -1
// HuffmanOnly disables Lempel-Ziv match searching and only performs Huffman
// entropy encoding. This mode is useful in compressing data that has
// already been compressed with an LZ style algorithm (e.g. Snappy or LZ4)
// that lacks an entropy encoder. Compression gains are achieved when
// certain bytes in the input stream occur more frequently than others.
//
// Note that HuffmanOnly produces a compressed output that is
// RFC 1951 compliant. That is, any valid DEFLATE decompressor will
// continue to be able to decompress this output.
HuffmanOnly = -2
)
const (
logWindowSize = 15
windowSize = 1 << logWindowSize
windowMask = windowSize - 1
// The LZ77 step produces a sequence of literal tokens and <length, offset>
// pair tokens. The offset is also known as distance. The underlying wire
// format limits the range of lengths and offsets. For example, there are
// 256 legitimate lengths: those in the range [3, 258]. This package's
// compressor uses a higher minimum match length, enabling optimizations
// such as finding matches via 32-bit loads and compares.
baseMatchLength = 3 // The smallest match length per the RFC section 3.2.5
minMatchLength = 4 // The smallest match length that the compressor actually emits
maxMatchLength = 258 // The largest match length
baseMatchOffset = 1 // The smallest match offset
maxMatchOffset = 1 << 15 // The largest match offset
// The maximum number of tokens we put into a single flate block, just to
// stop things from getting too large.
maxFlateBlockTokens = 1 << 14
maxStoreBlockSize = 65535
hashBits = 17 // After 17 performance degrades
hashSize = 1 << hashBits
hashMask = (1 << hashBits) - 1
maxHashOffset = 1 << 24
skipNever = math.MaxInt32
)
type compressionLevel struct {
level, good, lazy, nice, chain, fastSkipHashing int
}
var levels = []compressionLevel{
{0, 0, 0, 0, 0, 0}, // NoCompression.
{1, 0, 0, 0, 0, 0}, // BestSpeed uses a custom algorithm; see deflatefast.go.
// For levels 2-3 we don't bother trying with lazy matches.
{2, 4, 0, 16, 8, 5},
{3, 4, 0, 32, 32, 6},
// Levels 4-9 use increasingly more lazy matching
// and increasingly stringent conditions for "good enough".
{4, 4, 4, 16, 16, skipNever},
{5, 8, 16, 32, 32, skipNever},
{6, 8, 16, 128, 128, skipNever},
{7, 8, 32, 128, 256, skipNever},
{8, 32, 128, 258, 1024, skipNever},
{9, 32, 258, 258, 4096, skipNever},
}
type compressor struct {
compressionLevel
w *huffmanBitWriter
bulkHasher func([]byte, []uint32)
// compression algorithm
fill func(*compressor, []byte) int // copy data to window
step func(*compressor) // process window
sync bool // requesting flush
bestSpeed *deflateFast // Encoder for BestSpeed
// Input hash chains
// hashHead[hashValue] contains the largest inputIndex with the specified hash value
// If hashHead[hashValue] is within the current window, then
// hashPrev[hashHead[hashValue] & windowMask] contains the previous index
// with the same hash value.
chainHead int
hashHead [hashSize]uint32
hashPrev [windowSize]uint32
hashOffset int
// input window: unprocessed data is window[index:windowEnd]
index int
window []byte
windowEnd int
blockStart int // window index where current tokens start
byteAvailable bool // if true, still need to process window[index-1].
// queued output tokens
tokens []token
// deflate state
length int
offset int
hash uint32
maxInsertIndex int
err error
// hashMatch must be able to contain hashes for the maximum match length.
hashMatch [maxMatchLength - 1]uint32
}
func (d *compressor) fillDeflate(b []byte) int {
if d.index >= 2*windowSize-(minMatchLength+maxMatchLength) {
// shift the window by windowSize
copy(d.window, d.window[windowSize:2*windowSize])
d.index -= windowSize
d.windowEnd -= windowSize
if d.blockStart >= windowSize {
d.blockStart -= windowSize
} else {
d.blockStart = math.MaxInt32
}
d.hashOffset += windowSize
if d.hashOffset > maxHashOffset {
delta := d.hashOffset - 1
d.hashOffset -= delta
d.chainHead -= delta
// Iterate over slices instead of arrays to avoid copying
// the entire table onto the stack (Issue #18625).
for i, v := range d.hashPrev[:] {
if int(v) > delta {
d.hashPrev[i] = uint32(int(v) - delta)
} else {
d.hashPrev[i] = 0
}
}
for i, v := range d.hashHead[:] {
if int(v) > delta {
d.hashHead[i] = uint32(int(v) - delta)
} else {
d.hashHead[i] = 0
}
}
}
}
n := copy(d.window[d.windowEnd:], b)
d.windowEnd += n
return n
}
func (d *compressor) writeBlock(tokens []token, index int) error {
if index > 0 {
var window []byte
if d.blockStart <= index {
window = d.window[d.blockStart:index]
}
d.blockStart = index
d.w.writeBlock(tokens, false, window)
return d.w.err
}
return nil
}
// fillWindow will fill the current window with the supplied
// dictionary and calculate all hashes.
// This is much faster than doing a full encode.
// Should only be used after a reset.
func (d *compressor) fillWindow(b []byte) {
// Do not fill window if we are in store-only mode.
if d.compressionLevel.level < 2 {
return
}
if d.index != 0 || d.windowEnd != 0 {
panic("internal error: fillWindow called with stale data")
}
// If we are given too much, cut it.
if len(b) > windowSize {
b = b[len(b)-windowSize:]
}
// Add all to window.
n := copy(d.window, b)
// Calculate 256 hashes at the time (more L1 cache hits)
loops := (n + 256 - minMatchLength) / 256
for j := 0; j < loops; j++ {
index := j * 256
end := index + 256 + minMatchLength - 1
if end > n {
end = n
}
toCheck := d.window[index:end]
dstSize := len(toCheck) - minMatchLength + 1
if dstSize <= 0 {
continue
}
dst := d.hashMatch[:dstSize]
d.bulkHasher(toCheck, dst)
var newH uint32
for i, val := range dst {
di := i + index
newH = val
hh := &d.hashHead[newH&hashMask]
// Get previous value with the same hash.
// Our chain should point to the previous value.
d.hashPrev[di&windowMask] = *hh
// Set the head of the hash chain to us.
*hh = uint32(di + d.hashOffset)
}
d.hash = newH
}
// Update window information.
d.windowEnd = n
d.index = n
}
// Try to find a match starting at index whose length is greater than prevSize.
// We only look at chainCount possibilities before giving up.
func (d *compressor) findMatch(pos int, prevHead int, prevLength int, lookahead int) (length, offset int, ok bool) {
minMatchLook := maxMatchLength
if lookahead < minMatchLook {
minMatchLook = lookahead
}
win := d.window[0 : pos+minMatchLook]
// We quit when we get a match that's at least nice long
nice := len(win) - pos
if d.nice < nice {
nice = d.nice
}
// If we've got a match that's good enough, only look in 1/4 the chain.
tries := d.chain
length = prevLength
if length >= d.good {
tries >>= 2
}
wEnd := win[pos+length]
wPos := win[pos:]
minIndex := pos - windowSize
for i := prevHead; tries > 0; tries-- {
if wEnd == win[i+length] {
n := matchLen(win[i:], wPos, minMatchLook)
if n > length && (n > minMatchLength || pos-i <= 4096) {
length = n
offset = pos - i
ok = true
if n >= nice {
// The match is good enough that we don't try to find a better one.
break
}
wEnd = win[pos+n]
}
}
if i == minIndex {
// hashPrev[i & windowMask] has already been overwritten, so stop now.
break
}
i = int(d.hashPrev[i&windowMask]) - d.hashOffset
if i < minIndex || i < 0 {
break
}
}
return
}
func (d *compressor) writeStoredBlock(buf []byte) error {
if d.w.writeStoredHeader(len(buf), false); d.w.err != nil {
return d.w.err
}
d.w.writeBytes(buf)
return d.w.err
}
const hashmul = 0x1e35a7bd
// hash4 returns a hash representation of the first 4 bytes
// of the supplied slice.
// The caller must ensure that len(b) >= 4.
func hash4(b []byte) uint32 {
return ((uint32(b[3]) | uint32(b[2])<<8 | uint32(b[1])<<16 | uint32(b[0])<<24) * hashmul) >> (32 - hashBits)
}
// bulkHash4 will compute hashes using the same
// algorithm as hash4
func bulkHash4(b []byte, dst []uint32) {
if len(b) < minMatchLength {
return
}
hb := uint32(b[3]) | uint32(b[2])<<8 | uint32(b[1])<<16 | uint32(b[0])<<24
dst[0] = (hb * hashmul) >> (32 - hashBits)
end := len(b) - minMatchLength + 1
for i := 1; i < end; i++ {
hb = (hb << 8) | uint32(b[i+3])
dst[i] = (hb * hashmul) >> (32 - hashBits)
}
}
// matchLen returns the number of matching bytes in a and b
// up to length 'max'. Both slices must be at least 'max'
// bytes in size.
func matchLen(a, b []byte, max int) int {
a = a[:max]
b = b[:len(a)]
for i, av := range a {
if b[i] != av {
return i
}
}
return max
}
// encSpeed will compress and store the currently added data,
// if enough has been accumulated or we at the end of the stream.
// Any error that occurred will be in d.err
func (d *compressor) encSpeed() {
// We only compress if we have maxStoreBlockSize.
if d.windowEnd < maxStoreBlockSize {
if !d.sync {
return
}
// Handle small sizes.
if d.windowEnd < 128 {
switch {
case d.windowEnd == 0:
return
case d.windowEnd <= 16:
d.err = d.writeStoredBlock(d.window[:d.windowEnd])
default:
d.w.writeBlockHuff(false, d.window[:d.windowEnd])
d.err = d.w.err
}
d.windowEnd = 0
d.bestSpeed.reset()
return
}
}
// Encode the block.
d.tokens = d.bestSpeed.encode(d.tokens[:0], d.window[:d.windowEnd])
// If we removed less than 1/16th, Huffman compress the block.
if len(d.tokens) > d.windowEnd-(d.windowEnd>>4) {
d.w.writeBlockHuff(false, d.window[:d.windowEnd])
} else {
d.w.writeBlockDynamic(d.tokens, false, d.window[:d.windowEnd])
}
d.err = d.w.err
d.windowEnd = 0
}
func (d *compressor) initDeflate() {
d.window = make([]byte, 2*windowSize)
d.hashOffset = 1
d.tokens = make([]token, 0, maxFlateBlockTokens+1)
d.length = minMatchLength - 1
d.offset = 0
d.byteAvailable = false
d.index = 0
d.hash = 0
d.chainHead = -1
d.bulkHasher = bulkHash4
}
func (d *compressor) deflate() {
if d.windowEnd-d.index < minMatchLength+maxMatchLength && !d.sync {
return
}
d.maxInsertIndex = d.windowEnd - (minMatchLength - 1)
if d.index < d.maxInsertIndex {
d.hash = hash4(d.window[d.index : d.index+minMatchLength])
}
Loop:
for {
if d.index > d.windowEnd {
panic("index > windowEnd")
}
lookahead := d.windowEnd - d.index
if lookahead < minMatchLength+maxMatchLength {
if !d.sync {
break Loop
}
if d.index > d.windowEnd {
panic("index > windowEnd")
}
if lookahead == 0 {
// Flush current output block if any.
if d.byteAvailable {
// There is still one pending token that needs to be flushed
d.tokens = append(d.tokens, literalToken(uint32(d.window[d.index-1])))
d.byteAvailable = false
}
if len(d.tokens) > 0 {
if d.err = d.writeBlock(d.tokens, d.index); d.err != nil {
return
}
d.tokens = d.tokens[:0]
}
break Loop
}
}
if d.index < d.maxInsertIndex {
// Update the hash
d.hash = hash4(d.window[d.index : d.index+minMatchLength])
hh := &d.hashHead[d.hash&hashMask]
d.chainHead = int(*hh)
d.hashPrev[d.index&windowMask] = uint32(d.chainHead)
*hh = uint32(d.index + d.hashOffset)
}
prevLength := d.length
prevOffset := d.offset
d.length = minMatchLength - 1
d.offset = 0
minIndex := d.index - windowSize
if minIndex < 0 {
minIndex = 0
}
if d.chainHead-d.hashOffset >= minIndex &&
(d.fastSkipHashing != skipNever && lookahead > minMatchLength-1 ||
d.fastSkipHashing == skipNever && lookahead > prevLength && prevLength < d.lazy) {
if newLength, newOffset, ok := d.findMatch(d.index, d.chainHead-d.hashOffset, minMatchLength-1, lookahead); ok {
d.length = newLength
d.offset = newOffset
}
}
if d.fastSkipHashing != skipNever && d.length >= minMatchLength ||
d.fastSkipHashing == skipNever && prevLength >= minMatchLength && d.length <= prevLength {
// There was a match at the previous step, and the current match is
// not better. Output the previous match.
if d.fastSkipHashing != skipNever {
d.tokens = append(d.tokens, matchToken(uint32(d.length-baseMatchLength), uint32(d.offset-baseMatchOffset)))
} else {
d.tokens = append(d.tokens, matchToken(uint32(prevLength-baseMatchLength), uint32(prevOffset-baseMatchOffset)))
}
// Insert in the hash table all strings up to the end of the match.
// index and index-1 are already inserted. If there is not enough
// lookahead, the last two strings are not inserted into the hash
// table.
if d.length <= d.fastSkipHashing {
var newIndex int
if d.fastSkipHashing != skipNever {
newIndex = d.index + d.length
} else {
newIndex = d.index + prevLength - 1
}
index := d.index
for index++; index < newIndex; index++ {
if index < d.maxInsertIndex {
d.hash = hash4(d.window[index : index+minMatchLength])
// Get previous value with the same hash.
// Our chain should point to the previous value.
hh := &d.hashHead[d.hash&hashMask]
d.hashPrev[index&windowMask] = *hh
// Set the head of the hash chain to us.
*hh = uint32(index + d.hashOffset)
}
}
d.index = index
if d.fastSkipHashing == skipNever {
d.byteAvailable = false
d.length = minMatchLength - 1
}
} else {
// For matches this long, we don't bother inserting each individual
// item into the table.
d.index += d.length
if d.index < d.maxInsertIndex {
d.hash = hash4(d.window[d.index : d.index+minMatchLength])
}
}
if len(d.tokens) == maxFlateBlockTokens {
// The block includes the current character
if d.err = d.writeBlock(d.tokens, d.index); d.err != nil {
return
}
d.tokens = d.tokens[:0]
}
} else {
if d.fastSkipHashing != skipNever || d.byteAvailable {
i := d.index - 1
if d.fastSkipHashing != skipNever {
i = d.index
}
d.tokens = append(d.tokens, literalToken(uint32(d.window[i])))
if len(d.tokens) == maxFlateBlockTokens {
if d.err = d.writeBlock(d.tokens, i+1); d.err != nil {
return
}
d.tokens = d.tokens[:0]
}
}
d.index++
if d.fastSkipHashing == skipNever {
d.byteAvailable = true
}
}
}
}
func (d *compressor) fillStore(b []byte) int {
n := copy(d.window[d.windowEnd:], b)
d.windowEnd += n
return n
}
func (d *compressor) store() {
if d.windowEnd > 0 && (d.windowEnd == maxStoreBlockSize || d.sync) {
d.err = d.writeStoredBlock(d.window[:d.windowEnd])
d.windowEnd = 0
}
}
// storeHuff compresses and stores the currently added data
// when the d.window is full or we are at the end of the stream.
// Any error that occurred will be in d.err
func (d *compressor) storeHuff() {
if d.windowEnd < len(d.window) && !d.sync || d.windowEnd == 0 {
return
}
d.w.writeBlockHuff(false, d.window[:d.windowEnd])
d.err = d.w.err
d.windowEnd = 0
}
func (d *compressor) write(b []byte) (n int, err error) {
if d.err != nil {
return 0, d.err
}
n = len(b)
for len(b) > 0 {
d.step(d)
b = b[d.fill(d, b):]
if d.err != nil {
return 0, d.err
}
}
return n, nil
}
func (d *compressor) syncFlush() error {
if d.err != nil {
return d.err
}
d.sync = true
d.step(d)
if d.err == nil {
d.w.writeStoredHeader(0, false)
d.w.flush()
d.err = d.w.err
}
d.sync = false
return d.err
}
func (d *compressor) init(w io.Writer, level int) (err error) {
d.w = newHuffmanBitWriter(w)
switch {
case level == NoCompression:
d.window = make([]byte, maxStoreBlockSize)
d.fill = (*compressor).fillStore
d.step = (*compressor).store
case level == HuffmanOnly:
d.window = make([]byte, maxStoreBlockSize)
d.fill = (*compressor).fillStore
d.step = (*compressor).storeHuff
case level == BestSpeed:
d.compressionLevel = levels[level]
d.window = make([]byte, maxStoreBlockSize)
d.fill = (*compressor).fillStore
d.step = (*compressor).encSpeed
d.bestSpeed = newDeflateFast()
d.tokens = make([]token, maxStoreBlockSize)
case level == DefaultCompression:
level = 6
fallthrough
case 2 <= level && level <= 9:
d.compressionLevel = levels[level]
d.initDeflate()
d.fill = (*compressor).fillDeflate
d.step = (*compressor).deflate
default:
return fmt.Errorf("flate: invalid compression level %d: want value in range [-2, 9]", level)
}
return nil
}
func (d *compressor) reset(w io.Writer) {
d.w.reset(w)
d.sync = false
d.err = nil
switch d.compressionLevel.level {
case NoCompression:
d.windowEnd = 0
case BestSpeed:
d.windowEnd = 0
d.tokens = d.tokens[:0]
d.bestSpeed.reset()
default:
d.chainHead = -1
for i := range d.hashHead {
d.hashHead[i] = 0
}
for i := range d.hashPrev {
d.hashPrev[i] = 0
}
d.hashOffset = 1
d.index, d.windowEnd = 0, 0
d.blockStart, d.byteAvailable = 0, false
d.tokens = d.tokens[:0]
d.length = minMatchLength - 1
d.offset = 0
d.hash = 0
d.maxInsertIndex = 0
}
}
func (d *compressor) close() error {
if d.err != nil {
return d.err
}
d.sync = true
d.step(d)
if d.err != nil {
return d.err
}
if d.w.writeStoredHeader(0, true); d.w.err != nil {
return d.w.err
}
d.w.flush()
return d.w.err
}
// NewWriter returns a new Writer compressing data at the given level.
// Following zlib, levels range from 1 (BestSpeed) to 9 (BestCompression);
// higher levels typically run slower but compress more. Level 0
// (NoCompression) does not attempt any compression; it only adds the
// necessary DEFLATE framing.
// Level -1 (DefaultCompression) uses the default compression level.
// Level -2 (HuffmanOnly) will use Huffman compression only, giving
// a very fast compression for all types of input, but sacrificing considerable
// compression efficiency.
//
// If level is in the range [-2, 9] then the error returned will be nil.
// Otherwise the error returned will be non-nil.
func NewWriter(w io.Writer, level int) (*Writer, error) {
var dw Writer
if err := dw.d.init(w, level); err != nil {
return nil, err
}
return &dw, nil
}
// NewWriterDict is like NewWriter but initializes the new
// Writer with a preset dictionary. The returned Writer behaves
// as if the dictionary had been written to it without producing
// any compressed output. The compressed data written to w
// can only be decompressed by a Reader initialized with the
// same dictionary.
func NewWriterDict(w io.Writer, level int, dict []byte) (*Writer, error) {
dw := &dictWriter{w}
zw, err := NewWriter(dw, level)
if err != nil {
return nil, err
}
zw.d.fillWindow(dict)
zw.dict = append(zw.dict, dict...) // duplicate dictionary for Reset method.
return zw, err
}
type dictWriter struct {
w io.Writer
}
func (w *dictWriter) Write(b []byte) (n int, err error) {
return w.w.Write(b)
}
// A Writer takes data written to it and writes the compressed
// form of that data to an underlying writer (see NewWriter).
type Writer struct {
d compressor
dict []byte
}
// Write writes data to w, which will eventually write the
// compressed form of data to its underlying writer.
func (w *Writer) Write(data []byte) (n int, err error) {
return w.d.write(data)
}
// Flush flushes any pending data to the underlying writer.
// It is useful mainly in compressed network protocols, to ensure that
// a remote reader has enough data to reconstruct a packet.
// Flush does not return until the data has been written.
// Calling Flush when there is no pending data still causes the Writer
// to emit a sync marker of at least 4 bytes.
// If the underlying writer returns an error, Flush returns that error.
//
// In the terminology of the zlib library, Flush is equivalent to Z_SYNC_FLUSH.
func (w *Writer) Flush() error {
// For more about flushing:
// https://www.bolet.org/~pornin/deflate-flush.html
return w.d.syncFlush()
}
// Close flushes and closes the writer.
func (w *Writer) Close() error {
return w.d.close()
}
// Reset discards the writer's state and makes it equivalent to
// the result of NewWriter or NewWriterDict called with dst
// and w's level and dictionary.
func (w *Writer) Reset(dst io.Writer) {
if dw, ok := w.d.w.writer.(*dictWriter); ok {
// w was created with NewWriterDict
dw.w = dst
w.d.reset(dw)
w.d.fillWindow(w.dict)
} else {
// w was created with NewWriter
w.d.reset(dst)
}
}
image/internal/compress/flate/deflate_test.go
+984
View File
@@ -0,0 +1,984 @@
// Copyright 2009 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.
package flate
import (
"bytes"
"errors"
"fmt"
"internal/testenv"
"io"
"math/rand"
"os"
"reflect"
"runtime/debug"
"sync"
"testing"
)
type deflateTest struct {
in []byte
level int
out []byte
}
type deflateInflateTest struct {
in []byte
}
type reverseBitsTest struct {
in uint16
bitCount uint8
out uint16
}
var deflateTests = []*deflateTest{
{[]byte{}, 0, []byte{1, 0, 0, 255, 255}},
{[]byte{0x11}, -1, []byte{18, 4, 4, 0, 0, 255, 255}},
{[]byte{0x11}, DefaultCompression, []byte{18, 4, 4, 0, 0, 255, 255}},
{[]byte{0x11}, 4, []byte{18, 4, 4, 0, 0, 255, 255}},
{[]byte{0x11}, 0, []byte{0, 1, 0, 254, 255, 17, 1, 0, 0, 255, 255}},
{[]byte{0x11, 0x12}, 0, []byte{0, 2, 0, 253, 255, 17, 18, 1, 0, 0, 255, 255}},
{[]byte{0x11, 0x11, 0x11, 0x11, 0x11, 0x11, 0x11, 0x11}, 0,
[]byte{0, 8, 0, 247, 255, 17, 17, 17, 17, 17, 17, 17, 17, 1, 0, 0, 255, 255},
},
{[]byte{}, 2, []byte{1, 0, 0, 255, 255}},
{[]byte{0x11}, 2, []byte{18, 4, 4, 0, 0, 255, 255}},
{[]byte{0x11, 0x12}, 2, []byte{18, 20, 2, 4, 0, 0, 255, 255}},
{[]byte{0x11, 0x11, 0x11, 0x11, 0x11, 0x11, 0x11, 0x11}, 2, []byte{18, 132, 2, 64, 0, 0, 0, 255, 255}},
{[]byte{}, 9, []byte{1, 0, 0, 255, 255}},
{[]byte{0x11}, 9, []byte{18, 4, 4, 0, 0, 255, 255}},
{[]byte{0x11, 0x12}, 9, []byte{18, 20, 2, 4, 0, 0, 255, 255}},
{[]byte{0x11, 0x11, 0x11, 0x11, 0x11, 0x11, 0x11, 0x11}, 9, []byte{18, 132, 2, 64, 0, 0, 0, 255, 255}},
}
var deflateInflateTests = []*deflateInflateTest{
{[]byte{}},
{[]byte{0x11}},
{[]byte{0x11, 0x12}},
{[]byte{0x11, 0x11, 0x11, 0x11, 0x11, 0x11, 0x11, 0x11}},
{[]byte{0x11, 0x10, 0x13, 0x41, 0x21, 0x21, 0x41, 0x13, 0x87, 0x78, 0x13}},
{largeDataChunk()},
}
var reverseBitsTests = []*reverseBitsTest{
{1, 1, 1},
{1, 2, 2},
{1, 3, 4},
{1, 4, 8},
{1, 5, 16},
{17, 5, 17},
{257, 9, 257},
{29, 5, 23},
}
func largeDataChunk() []byte {
result := make([]byte, 100000)
for i := range result {
result[i] = byte(i * i & 0xFF)
}
return result
}
func TestBulkHash4(t *testing.T) {
for _, x := range deflateTests {
y := x.out
if len(y) < minMatchLength {
continue
}
y = append(y, y...)
for j := 4; j < len(y); j++ {
y := y[:j]
dst := make([]uint32, len(y)-minMatchLength+1)
for i := range dst {
dst[i] = uint32(i + 100)
}
bulkHash4(y, dst)
for i, got := range dst {
want := hash4(y[i:])
if got != want && got == uint32(i)+100 {
t.Errorf("Len:%d Index:%d, want 0x%08x but not modified", len(y), i, want)
} else if got != want {
t.Errorf("Len:%d Index:%d, got 0x%08x want:0x%08x", len(y), i, got, want)
}
}
}
}
}
func TestDeflate(t *testing.T) {
for _, h := range deflateTests {
var buf bytes.Buffer
w, err := NewWriter(&buf, h.level)
if err != nil {
t.Errorf("NewWriter: %v", err)
continue
}
w.Write(h.in)
w.Close()
if !bytes.Equal(buf.Bytes(), h.out) {
t.Errorf("Deflate(%d, %x) = \n%#v, want \n%#v", h.level, h.in, buf.Bytes(), h.out)
}
}
}
// A sparseReader returns a stream consisting of 0s followed by 1<<16 1s.
// This tests missing hash references in a very large input.
type sparseReader struct {
l int64
cur int64
}
func (r *sparseReader) Read(b []byte) (n int, err error) {
if r.cur >= r.l {
return 0, io.EOF
}
n = len(b)
cur := r.cur + int64(n)
if cur > r.l {
n -= int(cur - r.l)
cur = r.l
}
for i := range b[0:n] {
if r.cur+int64(i) >= r.l-1<<16 {
b[i] = 1
} else {
b[i] = 0
}
}
r.cur = cur
return
}
func TestVeryLongSparseChunk(t *testing.T) {
if testing.Short() {
t.Skip("skipping sparse chunk during short test")
}
w, err := NewWriter(io.Discard, 1)
if err != nil {
t.Errorf("NewWriter: %v", err)
return
}
if _, err = io.Copy(w, &sparseReader{l: 23e8}); err != nil {
t.Errorf("Compress failed: %v", err)
return
}
}
type syncBuffer struct {
buf bytes.Buffer
mu sync.RWMutex
closed bool
ready chan bool
}
func newSyncBuffer() *syncBuffer {
return &syncBuffer{ready: make(chan bool, 1)}
}
func (b *syncBuffer) Read(p []byte) (n int, err error) {
for {
b.mu.RLock()
n, err = b.buf.Read(p)
b.mu.RUnlock()
if n > 0 || b.closed {
return
}
<-b.ready
}
}
func (b *syncBuffer) signal() {
select {
case b.ready <- true:
default:
}
}
func (b *syncBuffer) Write(p []byte) (n int, err error) {
n, err = b.buf.Write(p)
b.signal()
return
}
func (b *syncBuffer) WriteMode() {
b.mu.Lock()
}
func (b *syncBuffer) ReadMode() {
b.mu.Unlock()
b.signal()
}
func (b *syncBuffer) Close() error {
b.closed = true
b.signal()
return nil
}
func testSync(t *testing.T, level int, input []byte, name string) {
if len(input) == 0 {
return
}
t.Logf("--testSync %d, %d, %s", level, len(input), name)
buf := newSyncBuffer()
buf1 := new(bytes.Buffer)
buf.WriteMode()
w, err := NewWriter(io.MultiWriter(buf, buf1), level)
if err != nil {
t.Errorf("NewWriter: %v", err)
return
}
r := NewReader(buf)
// Write half the input and read back.
for i := 0; i < 2; i++ {
var lo, hi int
if i == 0 {
lo, hi = 0, (len(input)+1)/2
} else {
lo, hi = (len(input)+1)/2, len(input)
}
t.Logf("#%d: write %d-%d", i, lo, hi)
if _, err := w.Write(input[lo:hi]); err != nil {
t.Errorf("testSync: write: %v", err)
return
}
if i == 0 {
if err := w.Flush(); err != nil {
t.Errorf("testSync: flush: %v", err)
return
}
} else {
if err := w.Close(); err != nil {
t.Errorf("testSync: close: %v", err)
}
}
buf.ReadMode()
out := make([]byte, hi-lo+1)
m, err := io.ReadAtLeast(r, out, hi-lo)
t.Logf("#%d: read %d", i, m)
if m != hi-lo || err != nil {
t.Errorf("testSync/%d (%d, %d, %s): read %d: %d, %v (%d left)", i, level, len(input), name, hi-lo, m, err, buf.buf.Len())
return
}
if !bytes.Equal(input[lo:hi], out[:hi-lo]) {
t.Errorf("testSync/%d: read wrong bytes: %x vs %x", i, input[lo:hi], out[:hi-lo])
return
}
// This test originally checked that after reading
// the first half of the input, there was nothing left
// in the read buffer (buf.buf.Len() != 0) but that is
// not necessarily the case: the write Flush may emit
// some extra framing bits that are not necessary
// to process to obtain the first half of the uncompressed
// data. The test ran correctly most of the time, because
// the background goroutine had usually read even
// those extra bits by now, but it's not a useful thing to
// check.
buf.WriteMode()
}
buf.ReadMode()
out := make([]byte, 10)
if n, err := r.Read(out); n > 0 || err != io.EOF {
t.Errorf("testSync (%d, %d, %s): final Read: %d, %v (hex: %x)", level, len(input), name, n, err, out[0:n])
}
if buf.buf.Len() != 0 {
t.Errorf("testSync (%d, %d, %s): extra data at end", level, len(input), name)
}
r.Close()
// stream should work for ordinary reader too
r = NewReader(buf1)
out, err = io.ReadAll(r)
if err != nil {
t.Errorf("testSync: read: %s", err)
return
}
r.Close()
if !bytes.Equal(input, out) {
t.Errorf("testSync: decompress(compress(data)) != data: level=%d input=%s", level, name)
}
}
func testToFromWithLevelAndLimit(t *testing.T, level int, input []byte, name string, limit int) {
var buffer bytes.Buffer
w, err := NewWriter(&buffer, level)
if err != nil {
t.Errorf("NewWriter: %v", err)
return
}
w.Write(input)
w.Close()
if limit > 0 && buffer.Len() > limit {
t.Errorf("level: %d, len(compress(data)) = %d > limit = %d", level, buffer.Len(), limit)
return
}
if limit > 0 {
t.Logf("level: %d, size:%.2f%%, %d b\n", level, float64(buffer.Len()*100)/float64(limit), buffer.Len())
}
r := NewReader(&buffer)
out, err := io.ReadAll(r)
if err != nil {
t.Errorf("read: %s", err)
return
}
r.Close()
if !bytes.Equal(input, out) {
t.Errorf("decompress(compress(data)) != data: level=%d input=%s", level, name)
return
}
testSync(t, level, input, name)
}
func testToFromWithLimit(t *testing.T, input []byte, name string, limit [11]int) {
for i := 0; i < 10; i++ {
testToFromWithLevelAndLimit(t, i, input, name, limit[i])
}
// Test HuffmanCompression
testToFromWithLevelAndLimit(t, -2, input, name, limit[10])
}
func TestDeflateInflate(t *testing.T) {
t.Parallel()
for i, h := range deflateInflateTests {
if testing.Short() && len(h.in) > 10000 {
continue
}
testToFromWithLimit(t, h.in, fmt.Sprintf("#%d", i), [11]int{})
}
}
func TestReverseBits(t *testing.T) {
for _, h := range reverseBitsTests {
if v := reverseBits(h.in, h.bitCount); v != h.out {
t.Errorf("reverseBits(%v,%v) = %v, want %v",
h.in, h.bitCount, v, h.out)
}
}
}
type deflateInflateStringTest struct {
filename string
label string
limit [11]int
}
var deflateInflateStringTests = []deflateInflateStringTest{
{
"../testdata/e.txt",
"2.718281828...",
[...]int{100018, 50650, 50960, 51150, 50930, 50790, 50790, 50790, 50790, 50790, 43683},
},
{
"../../testdata/Isaac.Newton-Opticks.txt",
"Isaac.Newton-Opticks",
[...]int{567248, 218338, 198211, 193152, 181100, 175427, 175427, 173597, 173422, 173422, 325240},
},
}
func TestDeflateInflateString(t *testing.T) {
t.Parallel()
if testing.Short() && testenv.Builder() == "" {
t.Skip("skipping in short mode")
}
for _, test := range deflateInflateStringTests {
gold, err := os.ReadFile(test.filename)
if err != nil {
t.Error(err)
}
testToFromWithLimit(t, gold, test.label, test.limit)
if testing.Short() {
break
}
}
}
func TestReaderDict(t *testing.T) {
const (
dict = "hello world"
text = "hello again world"
)
var b bytes.Buffer
w, err := NewWriter(&b, 5)
if err != nil {
t.Fatalf("NewWriter: %v", err)
}
w.Write([]byte(dict))
w.Flush()
b.Reset()
w.Write([]byte(text))
w.Close()
r := NewReaderDict(&b, []byte(dict))
data, err := io.ReadAll(r)
if err != nil {
t.Fatal(err)
}
if string(data) != "hello again world" {
t.Fatalf("read returned %q want %q", string(data), text)
}
}
func TestWriterDict(t *testing.T) {
const (
dict = "hello world"
text = "hello again world"
)
var b bytes.Buffer
w, err := NewWriter(&b, 5)
if err != nil {
t.Fatalf("NewWriter: %v", err)
}
w.Write([]byte(dict))
w.Flush()
b.Reset()
w.Write([]byte(text))
w.Close()
var b1 bytes.Buffer
w, _ = NewWriterDict(&b1, 5, []byte(dict))
w.Write([]byte(text))
w.Close()
if !bytes.Equal(b1.Bytes(), b.Bytes()) {
t.Fatalf("writer wrote %q want %q", b1.Bytes(), b.Bytes())
}
}
// See https://golang.org/issue/2508
func TestRegression2508(t *testing.T) {
if testing.Short() {
t.Logf("test disabled with -short")
return
}
w, err := NewWriter(io.Discard, 1)
if err != nil {
t.Fatalf("NewWriter: %v", err)
}
buf := make([]byte, 1024)
for i := 0; i < 131072; i++ {
if _, err := w.Write(buf); err != nil {
t.Fatalf("writer failed: %v", err)
}
}
w.Close()
}
func TestWriterReset(t *testing.T) {
t.Parallel()
for level := 0; level <= 9; level++ {
if testing.Short() && level > 1 {
break
}
w, err := NewWriter(io.Discard, level)
if err != nil {
t.Fatalf("NewWriter: %v", err)
}
buf := []byte("hello world")
n := 1024
if testing.Short() {
n = 10
}
for i := 0; i < n; i++ {
w.Write(buf)
}
w.Reset(io.Discard)
wref, err := NewWriter(io.Discard, level)
if err != nil {
t.Fatalf("NewWriter: %v", err)
}
// DeepEqual doesn't compare functions.
w.d.fill, wref.d.fill = nil, nil
w.d.step, wref.d.step = nil, nil
w.d.bulkHasher, wref.d.bulkHasher = nil, nil
w.d.bestSpeed, wref.d.bestSpeed = nil, nil
// hashMatch is always overwritten when used.
copy(w.d.hashMatch[:], wref.d.hashMatch[:])
if len(w.d.tokens) != 0 {
t.Errorf("level %d Writer not reset after Reset. %d tokens were present", level, len(w.d.tokens))
}
// As long as the length is 0, we don't care about the content.
w.d.tokens = wref.d.tokens
// We don't care if there are values in the window, as long as it is at d.index is 0
w.d.window = wref.d.window
if !reflect.DeepEqual(w, wref) {
t.Errorf("level %d Writer not reset after Reset", level)
}
}
levels := []int{0, 1, 2, 5, 9}
for _, level := range levels {
t.Run(fmt.Sprint(level), func(t *testing.T) {
testResetOutput(t, level, nil)
})
}
t.Run("dict", func(t *testing.T) {
for _, level := range levels {
t.Run(fmt.Sprint(level), func(t *testing.T) {
testResetOutput(t, level, nil)
})
}
})
}
func testResetOutput(t *testing.T, level int, dict []byte) {
writeData := func(w *Writer) {
msg := []byte("now is the time for all good gophers")
w.Write(msg)
w.Flush()
hello := []byte("hello world")
for i := 0; i < 1024; i++ {
w.Write(hello)
}
fill := bytes.Repeat([]byte("x"), 65000)
w.Write(fill)
}
buf := new(bytes.Buffer)
var w *Writer
var err error
if dict == nil {
w, err = NewWriter(buf, level)
} else {
w, err = NewWriterDict(buf, level, dict)
}
if err != nil {
t.Fatalf("NewWriter: %v", err)
}
writeData(w)
w.Close()
out1 := buf.Bytes()
buf2 := new(bytes.Buffer)
w.Reset(buf2)
writeData(w)
w.Close()
out2 := buf2.Bytes()
if len(out1) != len(out2) {
t.Errorf("got %d, expected %d bytes", len(out2), len(out1))
return
}
if !bytes.Equal(out1, out2) {
mm := 0
for i, b := range out1[:len(out2)] {
if b != out2[i] {
t.Errorf("mismatch index %d: %#02x, expected %#02x", i, out2[i], b)
}
mm++
if mm == 10 {
t.Fatal("Stopping")
}
}
}
t.Logf("got %d bytes", len(out1))
}
// TestBestSpeed tests that round-tripping through deflate and then inflate
// recovers the original input. The Write sizes are near the thresholds in the
// compressor.encSpeed method (0, 16, 128), as well as near maxStoreBlockSize
// (65535).
func TestBestSpeed(t *testing.T) {
t.Parallel()
abc := make([]byte, 128)
for i := range abc {
abc[i] = byte(i)
}
abcabc := bytes.Repeat(abc, 131072/len(abc))
var want []byte
testCases := [][]int{
{65536, 0},
{65536, 1},
{65536, 1, 256},
{65536, 1, 65536},
{65536, 14},
{65536, 15},
{65536, 16},
{65536, 16, 256},
{65536, 16, 65536},
{65536, 127},
{65536, 128},
{65536, 128, 256},
{65536, 128, 65536},
{65536, 129},
{65536, 65536, 256},
{65536, 65536, 65536},
}
for i, tc := range testCases {
if i >= 3 && testing.Short() {
break
}
for _, firstN := range []int{1, 65534, 65535, 65536, 65537, 131072} {
tc[0] = firstN
outer:
for _, flush := range []bool{false, true} {
buf := new(bytes.Buffer)
want = want[:0]
w, err := NewWriter(buf, BestSpeed)
if err != nil {
t.Errorf("i=%d, firstN=%d, flush=%t: NewWriter: %v", i, firstN, flush, err)
continue
}
for _, n := range tc {
want = append(want, abcabc[:n]...)
if _, err := w.Write(abcabc[:n]); err != nil {
t.Errorf("i=%d, firstN=%d, flush=%t: Write: %v", i, firstN, flush, err)
continue outer
}
if !flush {
continue
}
if err := w.Flush(); err != nil {
t.Errorf("i=%d, firstN=%d, flush=%t: Flush: %v", i, firstN, flush, err)
continue outer
}
}
if err := w.Close(); err != nil {
t.Errorf("i=%d, firstN=%d, flush=%t: Close: %v", i, firstN, flush, err)
continue
}
r := NewReader(buf)
got, err := io.ReadAll(r)
if err != nil {
t.Errorf("i=%d, firstN=%d, flush=%t: ReadAll: %v", i, firstN, flush, err)
continue
}
r.Close()
if !bytes.Equal(got, want) {
t.Errorf("i=%d, firstN=%d, flush=%t: corruption during deflate-then-inflate", i, firstN, flush)
continue
}
}
}
}
}
var errIO = errors.New("IO error")
// failWriter fails with errIO exactly at the nth call to Write.
type failWriter struct{ n int }
func (w *failWriter) Write(b []byte) (int, error) {
w.n--
if w.n == -1 {
return 0, errIO
}
return len(b), nil
}
func TestWriterPersistentError(t *testing.T) {
t.Parallel()
d, err := os.ReadFile("../../testdata/Isaac.Newton-Opticks.txt")
if err != nil {
t.Fatalf("ReadFile: %v", err)
}
d = d[:10000] // Keep this test short
zw, err := NewWriter(nil, DefaultCompression)
if err != nil {
t.Fatalf("NewWriter: %v", err)
}
// Sweep over the threshold at which an error is returned.
// The variable i makes it such that the ith call to failWriter.Write will
// return errIO. Since failWriter errors are not persistent, we must ensure
// that flate.Writer errors are persistent.
for i := 0; i < 1000; i++ {
fw := &failWriter{i}
zw.Reset(fw)
_, werr := zw.Write(d)
cerr := zw.Close()
if werr != errIO && werr != nil {
t.Errorf("test %d, mismatching Write error: got %v, want %v", i, werr, errIO)
}
if cerr != errIO && fw.n < 0 {
t.Errorf("test %d, mismatching Close error: got %v, want %v", i, cerr, errIO)
}
if fw.n >= 0 {
// At this point, the failure threshold was sufficiently high enough
// that we wrote the whole stream without any errors.
return
}
}
}
func TestBestSpeedMatch(t *testing.T) {
t.Parallel()
cases := []struct {
previous, current []byte
t, s, want int32
}{{
previous: []byte{0, 0, 0, 1, 2},
current: []byte{3, 4, 5, 0, 1, 2, 3, 4, 5},
t: -3,
s: 3,
want: 6,
}, {
previous: []byte{0, 0, 0, 1, 2},
current: []byte{2, 4, 5, 0, 1, 2, 3, 4, 5},
t: -3,
s: 3,
want: 3,
}, {
previous: []byte{0, 0, 0, 1, 1},
current: []byte{3, 4, 5, 0, 1, 2, 3, 4, 5},
t: -3,
s: 3,
want: 2,
}, {
previous: []byte{0, 0, 0, 1, 2},
current: []byte{2, 2, 2, 2, 1, 2, 3, 4, 5},
t: -1,
s: 0,
want: 4,
}, {
previous: []byte{0, 0, 0, 1, 2, 3, 4, 5, 2, 2},
current: []byte{2, 2, 2, 2, 1, 2, 3, 4, 5},
t: -7,
s: 4,
want: 5,
}, {
previous: []byte{9, 9, 9, 9, 9},
current: []byte{2, 2, 2, 2, 1, 2, 3, 4, 5},
t: -1,
s: 0,
want: 0,
}, {
previous: []byte{9, 9, 9, 9, 9},
current: []byte{9, 2, 2, 2, 1, 2, 3, 4, 5},
t: 0,
s: 1,
want: 0,
}, {
previous: []byte{},
current: []byte{9, 2, 2, 2, 1, 2, 3, 4, 5},
t: -5,
s: 1,
want: 0,
}, {
previous: []byte{},
current: []byte{9, 2, 2, 2, 1, 2, 3, 4, 5},
t: -1,
s: 1,
want: 0,
}, {
previous: []byte{},
current: []byte{2, 2, 2, 2, 1, 2, 3, 4, 5},
t: 0,
s: 1,
want: 3,
}, {
previous: []byte{3, 4, 5},
current: []byte{3, 4, 5},
t: -3,
s: 0,
want: 3,
}, {
previous: make([]byte, 1000),
current: make([]byte, 1000),
t: -1000,
s: 0,
want: maxMatchLength - 4,
}, {
previous: make([]byte, 200),
current: make([]byte, 500),
t: -200,
s: 0,
want: maxMatchLength - 4,
}, {
previous: make([]byte, 200),
current: make([]byte, 500),
t: 0,
s: 1,
want: maxMatchLength - 4,
}, {
previous: make([]byte, maxMatchLength-4),
current: make([]byte, 500),
t: -(maxMatchLength - 4),
s: 0,
want: maxMatchLength - 4,
}, {
previous: make([]byte, 200),
current: make([]byte, 500),
t: -200,
s: 400,
want: 100,
}, {
previous: make([]byte, 10),
current: make([]byte, 500),
t: 200,
s: 400,
want: 100,
}}
for i, c := range cases {
e := deflateFast{prev: c.previous}
got := e.matchLen(c.s, c.t, c.current)
if got != c.want {
t.Errorf("Test %d: match length, want %d, got %d", i, c.want, got)
}
}
}
func TestBestSpeedMaxMatchOffset(t *testing.T) {
t.Parallel()
const abc, xyz = "abcdefgh", "stuvwxyz"
for _, matchBefore := range []bool{false, true} {
for _, extra := range []int{0, inputMargin - 1, inputMargin, inputMargin + 1, 2 * inputMargin} {
for offsetAdj := -5; offsetAdj <= +5; offsetAdj++ {
report := func(desc string, err error) {
t.Errorf("matchBefore=%t, extra=%d, offsetAdj=%d: %s%v",
matchBefore, extra, offsetAdj, desc, err)
}
offset := maxMatchOffset + offsetAdj
// Make src to be a []byte of the form
// "%s%s%s%s%s" % (abc, zeros0, xyzMaybe, abc, zeros1)
// where:
// zeros0 is approximately maxMatchOffset zeros.
// xyzMaybe is either xyz or the empty string.
// zeros1 is between 0 and 30 zeros.
// The difference between the two abc's will be offset, which
// is maxMatchOffset plus or minus a small adjustment.
src := make([]byte, offset+len(abc)+extra)
copy(src, abc)
if !matchBefore {
copy(src[offset-len(xyz):], xyz)
}
copy(src[offset:], abc)
buf := new(bytes.Buffer)
w, err := NewWriter(buf, BestSpeed)
if err != nil {
report("NewWriter: ", err)
continue
}
if _, err := w.Write(src); err != nil {
report("Write: ", err)
continue
}
if err := w.Close(); err != nil {
report("Writer.Close: ", err)
continue
}
r := NewReader(buf)
dst, err := io.ReadAll(r)
r.Close()
if err != nil {
report("ReadAll: ", err)
continue
}
if !bytes.Equal(dst, src) {
report("", fmt.Errorf("bytes differ after round-tripping"))
continue
}
}
}
}
}
func TestBestSpeedShiftOffsets(t *testing.T) {
// Test if shiftoffsets properly preserves matches and resets out-of-range matches
// seen in https://github.com/golang/go/issues/4142
enc := newDeflateFast()
// testData may not generate internal matches.
testData := make([]byte, 32)
rng := rand.New(rand.NewSource(0))
for i := range testData {
testData[i] = byte(rng.Uint32())
}
// Encode the testdata with clean state.
// Second part should pick up matches from the first block.
wantFirstTokens := len(enc.encode(nil, testData))
wantSecondTokens := len(enc.encode(nil, testData))
if wantFirstTokens <= wantSecondTokens {
t.Fatalf("test needs matches between inputs to be generated")
}
// Forward the current indicator to before wraparound.
enc.cur = bufferReset - int32(len(testData))
// Part 1 before wrap, should match clean state.
got := len(enc.encode(nil, testData))
if wantFirstTokens != got {
t.Errorf("got %d, want %d tokens", got, wantFirstTokens)
}
// Verify we are about to wrap.
if enc.cur != bufferReset {
t.Errorf("got %d, want e.cur to be at bufferReset (%d)", enc.cur, bufferReset)
}
// Part 2 should match clean state as well even if wrapped.
got = len(enc.encode(nil, testData))
if wantSecondTokens != got {
t.Errorf("got %d, want %d token", got, wantSecondTokens)
}
// Verify that we wrapped.
if enc.cur >= bufferReset {
t.Errorf("want e.cur to be < bufferReset (%d), got %d", bufferReset, enc.cur)
}
// Forward the current buffer, leaving the matches at the bottom.
enc.cur = bufferReset
enc.shiftOffsets()
// Ensure that no matches were picked up.
got = len(enc.encode(nil, testData))
if wantFirstTokens != got {
t.Errorf("got %d, want %d tokens", got, wantFirstTokens)
}
}
func TestMaxStackSize(t *testing.T) {
// This test must not run in parallel with other tests as debug.SetMaxStack
// affects all goroutines.
n := debug.SetMaxStack(1 << 16)
defer debug.SetMaxStack(n)
var wg sync.WaitGroup
defer wg.Wait()
b := make([]byte, 1<<20)
for level := HuffmanOnly; level <= BestCompression; level++ {
// Run in separate goroutine to increase probability of stack regrowth.
wg.Add(1)
go func(level int) {
defer wg.Done()
zw, err := NewWriter(io.Discard, level)
if err != nil {
t.Errorf("level %d, NewWriter() = %v, want nil", level, err)
}
if n, err := zw.Write(b); n != len(b) || err != nil {
t.Errorf("level %d, Write() = (%d, %v), want (%d, nil)", level, n, err, len(b))
}
if err := zw.Close(); err != nil {
t.Errorf("level %d, Close() = %v, want nil", level, err)
}
zw.Reset(io.Discard)
}(level)
}
}
image/internal/compress/flate/deflatefast.go
+309
View File
@@ -0,0 +1,309 @@
// Copyright 2016 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.
package flate
import "math"
// This encoding algorithm, which prioritizes speed over output size, is
// based on Snappy's LZ77-style encoder: github.com/golang/snappy
const (
tableBits = 14 // Bits used in the table.
tableSize = 1 << tableBits // Size of the table.
tableMask = tableSize - 1 // Mask for table indices. Redundant, but can eliminate bounds checks.
tableShift = 32 - tableBits // Right-shift to get the tableBits most significant bits of a uint32.
// Reset the buffer offset when reaching this.
// Offsets are stored between blocks as int32 values.
// Since the offset we are checking against is at the beginning
// of the buffer, we need to subtract the current and input
// buffer to not risk overflowing the int32.
bufferReset = math.MaxInt32 - maxStoreBlockSize*2
)
func load32(b []byte, i int32) uint32 {
b = b[i : i+4 : len(b)] // Help the compiler eliminate bounds checks on the next line.
return uint32(b[0]) | uint32(b[1])<<8 | uint32(b[2])<<16 | uint32(b[3])<<24
}
func load64(b []byte, i int32) uint64 {
b = b[i : i+8 : len(b)] // Help the compiler eliminate bounds checks on the next line.
return uint64(b[0]) | uint64(b[1])<<8 | uint64(b[2])<<16 | uint64(b[3])<<24 |
uint64(b[4])<<32 | uint64(b[5])<<40 | uint64(b[6])<<48 | uint64(b[7])<<56
}
func hash(u uint32) uint32 {
return (u * 0x1e35a7bd) >> tableShift
}
// These constants are defined by the Snappy implementation so that its
// assembly implementation can fast-path some 16-bytes-at-a-time copies. They
// aren't necessary in the pure Go implementation, as we don't use those same
// optimizations, but using the same thresholds doesn't really hurt.
const (
inputMargin = 16 - 1
minNonLiteralBlockSize = 1 + 1 + inputMargin
)
type tableEntry struct {
val uint32 // Value at destination
offset int32
}
// deflateFast maintains the table for matches,
// and the previous byte block for cross block matching.
type deflateFast struct {
table [tableSize]tableEntry
prev []byte // Previous block, zero length if unknown.
cur int32 // Current match offset.
}
func newDeflateFast() *deflateFast {
return &deflateFast{cur: maxStoreBlockSize, prev: make([]byte, 0, maxStoreBlockSize)}
}
// encode encodes a block given in src and appends tokens
// to dst and returns the result.
func (e *deflateFast) encode(dst []token, src []byte) []token {
// Ensure that e.cur doesn't wrap.
if e.cur >= bufferReset {
e.shiftOffsets()
}
// This check isn't in the Snappy implementation, but there, the caller
// instead of the callee handles this case.
if len(src) < minNonLiteralBlockSize {
e.cur += maxStoreBlockSize
e.prev = e.prev[:0]
return emitLiteral(dst, src)
}
// sLimit is when to stop looking for offset/length copies. The inputMargin
// lets us use a fast path for emitLiteral in the main loop, while we are
// looking for copies.
sLimit := int32(len(src) - inputMargin)
// nextEmit is where in src the next emitLiteral should start from.
nextEmit := int32(0)
s := int32(0)
cv := load32(src, s)
nextHash := hash(cv)
for {
// Copied from the C++ snappy implementation:
//
// Heuristic match skipping: If 32 bytes are scanned with no matches
// found, start looking only at every other byte. If 32 more bytes are
// scanned (or skipped), look at every third byte, etc.. When a match
// is found, immediately go back to looking at every byte. This is a
// small loss (~5% performance, ~0.1% density) for compressible data
// due to more bookkeeping, but for non-compressible data (such as
// JPEG) it's a huge win since the compressor quickly "realizes" the
// data is incompressible and doesn't bother looking for matches
// everywhere.
//
// The "skip" variable keeps track of how many bytes there are since
// the last match; dividing it by 32 (ie. right-shifting by five) gives
// the number of bytes to move ahead for each iteration.
skip := int32(32)
nextS := s
var candidate tableEntry
for {
s = nextS
bytesBetweenHashLookups := skip >> 5
nextS = s + bytesBetweenHashLookups
skip += bytesBetweenHashLookups
if nextS > sLimit {
goto emitRemainder
}
candidate = e.table[nextHash&tableMask]
now := load32(src, nextS)
e.table[nextHash&tableMask] = tableEntry{offset: s + e.cur, val: cv}
nextHash = hash(now)
offset := s - (candidate.offset - e.cur)
if offset > maxMatchOffset || cv != candidate.val {
// Out of range or not matched.
cv = now
continue
}
break
}
// A 4-byte match has been found. We'll later see if more than 4 bytes
// match. But, prior to the match, src[nextEmit:s] are unmatched. Emit
// them as literal bytes.
dst = emitLiteral(dst, src[nextEmit:s])
// Call emitCopy, and then see if another emitCopy could be our next
// move. Repeat until we find no match for the input immediately after
// what was consumed by the last emitCopy call.
//
// If we exit this loop normally then we need to call emitLiteral next,
// though we don't yet know how big the literal will be. We handle that
// by proceeding to the next iteration of the main loop. We also can
// exit this loop via goto if we get close to exhausting the input.
for {
// Invariant: we have a 4-byte match at s, and no need to emit any
// literal bytes prior to s.
// Extend the 4-byte match as long as possible.
//
s += 4
t := candidate.offset - e.cur + 4
l := e.matchLen(s, t, src)
// matchToken is flate's equivalent of Snappy's emitCopy. (length,offset)
dst = append(dst, matchToken(uint32(l+4-baseMatchLength), uint32(s-t-baseMatchOffset)))
s += l
nextEmit = s
if s >= sLimit {
goto emitRemainder
}
// We could immediately start working at s now, but to improve
// compression we first update the hash table at s-1 and at s. If
// another emitCopy is not our next move, also calculate nextHash
// at s+1. At least on GOARCH=amd64, these three hash calculations
// are faster as one load64 call (with some shifts) instead of
// three load32 calls.
x := load64(src, s-1)
prevHash := hash(uint32(x))
e.table[prevHash&tableMask] = tableEntry{offset: e.cur + s - 1, val: uint32(x)}
x >>= 8
currHash := hash(uint32(x))
candidate = e.table[currHash&tableMask]
e.table[currHash&tableMask] = tableEntry{offset: e.cur + s, val: uint32(x)}
offset := s - (candidate.offset - e.cur)
if offset > maxMatchOffset || uint32(x) != candidate.val {
cv = uint32(x >> 8)
nextHash = hash(cv)
s++
break
}
}
}
emitRemainder:
if int(nextEmit) < len(src) {
dst = emitLiteral(dst, src[nextEmit:])
}
e.cur += int32(len(src))
e.prev = e.prev[:len(src)]
copy(e.prev, src)
return dst
}
func emitLiteral(dst []token, lit []byte) []token {
for _, v := range lit {
dst = append(dst, literalToken(uint32(v)))
}
return dst
}
// matchLen returns the match length between src[s:] and src[t:].
// t can be negative to indicate the match is starting in e.prev.
// We assume that src[s-4:s] and src[t-4:t] already match.
func (e *deflateFast) matchLen(s, t int32, src []byte) int32 {
s1 := int(s) + maxMatchLength - 4
if s1 > len(src) {
s1 = len(src)
}
// If we are inside the current block
if t >= 0 {
b := src[t:]
a := src[s:s1]
b = b[:len(a)]
// Extend the match to be as long as possible.
for i := range a {
if a[i] != b[i] {
return int32(i)
}
}
return int32(len(a))
}
// We found a match in the previous block.
tp := int32(len(e.prev)) + t
if tp < 0 {
return 0
}
// Extend the match to be as long as possible.
a := src[s:s1]
b := e.prev[tp:]
if len(b) > len(a) {
b = b[:len(a)]
}
a = a[:len(b)]
for i := range b {
if a[i] != b[i] {
return int32(i)
}
}
// If we reached our limit, we matched everything we are
// allowed to in the previous block and we return.
n := int32(len(b))
if int(s+n) == s1 {
return n
}
// Continue looking for more matches in the current block.
a = src[s+n : s1]
b = src[:len(a)]
for i := range a {
if a[i] != b[i] {
return int32(i) + n
}
}
return int32(len(a)) + n
}
// Reset resets the encoding history.
// This ensures that no matches are made to the previous block.
func (e *deflateFast) reset() {
e.prev = e.prev[:0]
// Bump the offset, so all matches will fail distance check.
// Nothing should be >= e.cur in the table.
e.cur += maxMatchOffset
// Protect against e.cur wraparound.
if e.cur >= bufferReset {
e.shiftOffsets()
}
}
// shiftOffsets will shift down all match offset.
// This is only called in rare situations to prevent integer overflow.
//
// See https://golang.org/issue/18636 and https://github.com/golang/go/issues/34121.
func (e *deflateFast) shiftOffsets() {
if len(e.prev) == 0 {
// We have no history; just clear the table.
for i := range e.table[:] {
e.table[i] = tableEntry{}
}
e.cur = maxMatchOffset + 1
return
}
// Shift down everything in the table that isn't already too far away.
for i := range e.table[:] {
v := e.table[i].offset - e.cur + maxMatchOffset + 1
if v < 0 {
// We want to reset e.cur to maxMatchOffset + 1, so we need to shift
// all table entries down by (e.cur - (maxMatchOffset + 1)).
// Because we ignore matches > maxMatchOffset, we can cap
// any negative offsets at 0.
v = 0
}
e.table[i].offset = v
}
e.cur = maxMatchOffset + 1
}
image/internal/compress/flate/dict_decoder.go
+186
View File
@@ -0,0 +1,186 @@
// Copyright 2016 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.
package flate
// dictDecoder implements the LZ77 sliding dictionary as used in decompression.
// LZ77 decompresses data through sequences of two forms of commands:
//
// * Literal insertions: Runs of one or more symbols are inserted into the data
// stream as is. This is accomplished through the writeByte method for a
// single symbol, or combinations of writeSlice/writeMark for multiple symbols.
// Any valid stream must start with a literal insertion if no preset dictionary
// is used.
//
// * Backward copies: Runs of one or more symbols are copied from previously
// emitted data. Backward copies come as the tuple (dist, length) where dist
// determines how far back in the stream to copy from and length determines how
// many bytes to copy. Note that it is valid for the length to be greater than
// the distance. Since LZ77 uses forward copies, that situation is used to
// perform a form of run-length encoding on repeated runs of symbols.
// The writeCopy and tryWriteCopy are used to implement this command.
//
// For performance reasons, this implementation performs little to no sanity
// checks about the arguments. As such, the invariants documented for each
// method call must be respected.
type dictDecoder struct {
hist []byte // Sliding window history
// Invariant: 0 <= rdPos <= wrPos <= len(hist)
wrPos int // Current output position in buffer
rdPos int // Have emitted hist[:rdPos] already
full bool // Has a full window length been written yet?
}
// To minimize the memory usage in TinyGo, it is defined as a fixed array
// instead of a make().
var ddHistBuf [1 << 15]byte
// init initializes dictDecoder to have a sliding window dictionary of the given
// size. If a preset dict is provided, it will initialize the dictionary with
// the contents of dict.
func (dd *dictDecoder) init(size int, dict []byte) {
*dd = dictDecoder{hist: dd.hist}
if cap(dd.hist) < size {
dd.hist = ddHistBuf[:size]
}
dd.hist = dd.hist[:size]
if len(dict) > len(dd.hist) {
dict = dict[len(dict)-len(dd.hist):]
}
dd.wrPos = copy(dd.hist, dict)
if dd.wrPos == len(dd.hist) {
dd.wrPos = 0
dd.full = true
}
dd.rdPos = dd.wrPos
}
// histSize reports the total amount of historical data in the dictionary.
func (dd *dictDecoder) histSize() int {
if dd.full {
return len(dd.hist)
}
return dd.wrPos
}
// availRead reports the number of bytes that can be flushed by readFlush.
func (dd *dictDecoder) availRead() int {
return dd.wrPos - dd.rdPos
}
// availWrite reports the available amount of output buffer space.
func (dd *dictDecoder) availWrite() int {
return len(dd.hist) - dd.wrPos
}
// writeSlice returns a slice of the available buffer to write data to.
//
// This invariant will be kept: len(s) <= availWrite()
func (dd *dictDecoder) writeSlice() []byte {
return dd.hist[dd.wrPos:]
}
// writeMark advances the writer pointer by cnt.
//
// This invariant must be kept: 0 <= cnt <= availWrite()
func (dd *dictDecoder) writeMark(cnt int) {
dd.wrPos += cnt
}
// writeByte writes a single byte to the dictionary.
//
// This invariant must be kept: 0 < availWrite()
func (dd *dictDecoder) writeByte(c byte) {
dd.hist[dd.wrPos] = c
dd.wrPos++
}
// writeCopy copies a string at a given (dist, length) to the output.
// This returns the number of bytes copied and may be less than the requested
// length if the available space in the output buffer is too small.
//
// This invariant must be kept: 0 < dist <= histSize()
func (dd *dictDecoder) writeCopy(dist, length int) int {
dstBase := dd.wrPos
dstPos := dstBase
srcPos := dstPos - dist
endPos := dstPos + length
if endPos > len(dd.hist) {
endPos = len(dd.hist)
}
// Copy non-overlapping section after destination position.
//
// This section is non-overlapping in that the copy length for this section
// is always less than or equal to the backwards distance. This can occur
// if a distance refers to data that wraps-around in the buffer.
// Thus, a backwards copy is performed here; that is, the exact bytes in
// the source prior to the copy is placed in the destination.
if srcPos < 0 {
srcPos += len(dd.hist)
dstPos += copy(dd.hist[dstPos:endPos], dd.hist[srcPos:])
srcPos = 0
}
// Copy possibly overlapping section before destination position.
//
// This section can overlap if the copy length for this section is larger
// than the backwards distance. This is allowed by LZ77 so that repeated
// strings can be succinctly represented using (dist, length) pairs.
// Thus, a forwards copy is performed here; that is, the bytes copied is
// possibly dependent on the resulting bytes in the destination as the copy
// progresses along. This is functionally equivalent to the following:
//
// for i := 0; i < endPos-dstPos; i++ {
// dd.hist[dstPos+i] = dd.hist[srcPos+i]
// }
// dstPos = endPos
//
for dstPos < endPos {
dstPos += copy(dd.hist[dstPos:endPos], dd.hist[srcPos:dstPos])
}
dd.wrPos = dstPos
return dstPos - dstBase
}
// tryWriteCopy tries to copy a string at a given (distance, length) to the
// output. This specialized version is optimized for short distances.
//
// This method is designed to be inlined for performance reasons.
//
// This invariant must be kept: 0 < dist <= histSize()
func (dd *dictDecoder) tryWriteCopy(dist, length int) int {
dstPos := dd.wrPos
endPos := dstPos + length
if dstPos < dist || endPos > len(dd.hist) {
return 0
}
dstBase := dstPos
srcPos := dstPos - dist
// Copy possibly overlapping section before destination position.
for dstPos < endPos {
dstPos += copy(dd.hist[dstPos:endPos], dd.hist[srcPos:dstPos])
}
dd.wrPos = dstPos
return dstPos - dstBase
}
// readFlush returns a slice of the historical buffer that is ready to be
// emitted to the user. The data returned by readFlush must be fully consumed
// before calling any other dictDecoder methods.
func (dd *dictDecoder) readFlush() []byte {
toRead := dd.hist[dd.rdPos:dd.wrPos]
dd.rdPos = dd.wrPos
if dd.wrPos == len(dd.hist) {
dd.wrPos, dd.rdPos = 0, 0
dd.full = true
}
return toRead
}
image/internal/compress/flate/dict_decoder_test.go
+139
View File
@@ -0,0 +1,139 @@
// Copyright 2016 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.
package flate
import (
"bytes"
"strings"
"testing"
)
func TestDictDecoder(t *testing.T) {
const (
abc = "ABC\n"
fox = "The quick brown fox jumped over the lazy dog!\n"
poem = "The Road Not Taken\nRobert Frost\n" +
"\n" +
"Two roads diverged in a yellow wood,\n" +
"And sorry I could not travel both\n" +
"And be one traveler, long I stood\n" +
"And looked down one as far as I could\n" +
"To where it bent in the undergrowth;\n" +
"\n" +
"Then took the other, as just as fair,\n" +
"And having perhaps the better claim,\n" +
"Because it was grassy and wanted wear;\n" +
"Though as for that the passing there\n" +
"Had worn them really about the same,\n" +
"\n" +
"And both that morning equally lay\n" +
"In leaves no step had trodden black.\n" +
"Oh, I kept the first for another day!\n" +
"Yet knowing how way leads on to way,\n" +
"I doubted if I should ever come back.\n" +
"\n" +
"I shall be telling this with a sigh\n" +
"Somewhere ages and ages hence:\n" +
"Two roads diverged in a wood, and I-\n" +
"I took the one less traveled by,\n" +
"And that has made all the difference.\n"
)
var poemRefs = []struct {
dist int // Backward distance (0 if this is an insertion)
length int // Length of copy or insertion
}{
{0, 38}, {33, 3}, {0, 48}, {79, 3}, {0, 11}, {34, 5}, {0, 6}, {23, 7},
{0, 8}, {50, 3}, {0, 2}, {69, 3}, {34, 5}, {0, 4}, {97, 3}, {0, 4},
{43, 5}, {0, 6}, {7, 4}, {88, 7}, {0, 12}, {80, 3}, {0, 2}, {141, 4},
{0, 1}, {196, 3}, {0, 3}, {157, 3}, {0, 6}, {181, 3}, {0, 2}, {23, 3},
{77, 3}, {28, 5}, {128, 3}, {110, 4}, {70, 3}, {0, 4}, {85, 6}, {0, 2},
{182, 6}, {0, 4}, {133, 3}, {0, 7}, {47, 5}, {0, 20}, {112, 5}, {0, 1},
{58, 3}, {0, 8}, {59, 3}, {0, 4}, {173, 3}, {0, 5}, {114, 3}, {0, 4},
{92, 5}, {0, 2}, {71, 3}, {0, 2}, {76, 5}, {0, 1}, {46, 3}, {96, 4},
{130, 4}, {0, 3}, {360, 3}, {0, 3}, {178, 5}, {0, 7}, {75, 3}, {0, 3},
{45, 6}, {0, 6}, {299, 6}, {180, 3}, {70, 6}, {0, 1}, {48, 3}, {66, 4},
{0, 3}, {47, 5}, {0, 9}, {325, 3}, {0, 1}, {359, 3}, {318, 3}, {0, 2},
{199, 3}, {0, 1}, {344, 3}, {0, 3}, {248, 3}, {0, 10}, {310, 3}, {0, 3},
{93, 6}, {0, 3}, {252, 3}, {157, 4}, {0, 2}, {273, 5}, {0, 14}, {99, 4},
{0, 1}, {464, 4}, {0, 2}, {92, 4}, {495, 3}, {0, 1}, {322, 4}, {16, 4},
{0, 3}, {402, 3}, {0, 2}, {237, 4}, {0, 2}, {432, 4}, {0, 1}, {483, 5},
{0, 2}, {294, 4}, {0, 2}, {306, 3}, {113, 5}, {0, 1}, {26, 4}, {164, 3},
{488, 4}, {0, 1}, {542, 3}, {248, 6}, {0, 5}, {205, 3}, {0, 8}, {48, 3},
{449, 6}, {0, 2}, {192, 3}, {328, 4}, {9, 5}, {433, 3}, {0, 3}, {622, 25},
{615, 5}, {46, 5}, {0, 2}, {104, 3}, {475, 10}, {549, 3}, {0, 4}, {597, 8},
{314, 3}, {0, 1}, {473, 6}, {317, 5}, {0, 1}, {400, 3}, {0, 3}, {109, 3},
{151, 3}, {48, 4}, {0, 4}, {125, 3}, {108, 3}, {0, 2},
}
var got, want bytes.Buffer
var dd dictDecoder
dd.init(1<<11, nil)
var writeCopy = func(dist, length int) {
for length > 0 {
cnt := dd.tryWriteCopy(dist, length)
if cnt == 0 {
cnt = dd.writeCopy(dist, length)
}
length -= cnt
if dd.availWrite() == 0 {
got.Write(dd.readFlush())
}
}
}
var writeString = func(str string) {
for len(str) > 0 {
cnt := copy(dd.writeSlice(), str)
str = str[cnt:]
dd.writeMark(cnt)
if dd.availWrite() == 0 {
got.Write(dd.readFlush())
}
}
}
writeString(".")
want.WriteByte('.')
str := poem
for _, ref := range poemRefs {
if ref.dist == 0 {
writeString(str[:ref.length])
} else {
writeCopy(ref.dist, ref.length)
}
str = str[ref.length:]
}
want.WriteString(poem)
writeCopy(dd.histSize(), 33)
want.Write(want.Bytes()[:33])
writeString(abc)
writeCopy(len(abc), 59*len(abc))
want.WriteString(strings.Repeat(abc, 60))
writeString(fox)
writeCopy(len(fox), 9*len(fox))
want.WriteString(strings.Repeat(fox, 10))
writeString(".")
writeCopy(1, 9)
want.WriteString(strings.Repeat(".", 10))
writeString(strings.ToUpper(poem))
writeCopy(len(poem), 7*len(poem))
want.WriteString(strings.Repeat(strings.ToUpper(poem), 8))
writeCopy(dd.histSize(), 10)
want.Write(want.Bytes()[want.Len()-dd.histSize():][:10])
got.Write(dd.readFlush())
if got.String() != want.String() {
t.Errorf("final string mismatch:\ngot %q\nwant %q", got.String(), want.String())
}
}
image/internal/compress/flate/example_test.go
+243
View File
@@ -0,0 +1,243 @@
// Copyright 2016 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.
package flate_test
import (
"bytes"
"compress/flate"
"fmt"
"io"
"log"
"os"
"strings"
"sync"
)
// In performance critical applications, Reset can be used to discard the
// current compressor or decompressor state and reinitialize them quickly
// by taking advantage of previously allocated memory.
func Example_reset() {
proverbs := []string{
"Don't communicate by sharing memory, share memory by communicating.\n",
"Concurrency is not parallelism.\n",
"The bigger the interface, the weaker the abstraction.\n",
"Documentation is for users.\n",
}
var r strings.Reader
var b bytes.Buffer
buf := make([]byte, 32<<10)
zw, err := flate.NewWriter(nil, flate.DefaultCompression)
if err != nil {
log.Fatal(err)
}
zr := flate.NewReader(nil)
for _, s := range proverbs {
r.Reset(s)
b.Reset()
// Reset the compressor and encode from some input stream.
zw.Reset(&b)
if _, err := io.CopyBuffer(zw, &r, buf); err != nil {
log.Fatal(err)
}
if err := zw.Close(); err != nil {
log.Fatal(err)
}
// Reset the decompressor and decode to some output stream.
if err := zr.(flate.Resetter).Reset(&b, nil); err != nil {
log.Fatal(err)
}
if _, err := io.CopyBuffer(os.Stdout, zr, buf); err != nil {
log.Fatal(err)
}
if err := zr.Close(); err != nil {
log.Fatal(err)
}
}
// Output:
// Don't communicate by sharing memory, share memory by communicating.
// Concurrency is not parallelism.
// The bigger the interface, the weaker the abstraction.
// Documentation is for users.
}
// A preset dictionary can be used to improve the compression ratio.
// The downside to using a dictionary is that the compressor and decompressor
// must agree in advance what dictionary to use.
func Example_dictionary() {
// The dictionary is a string of bytes. When compressing some input data,
// the compressor will attempt to substitute substrings with matches found
// in the dictionary. As such, the dictionary should only contain substrings
// that are expected to be found in the actual data stream.
const dict = `<?xml version="1.0"?>` + `<book>` + `<data>` + `<meta name="` + `" content="`
// The data to compress should (but is not required to) contain frequent
// substrings that match those in the dictionary.
const data = `<?xml version="1.0"?>
<book>
<meta name="title" content="The Go Programming Language"/>
<meta name="authors" content="Alan Donovan and Brian Kernighan"/>
<meta name="published" content="2015-10-26"/>
<meta name="isbn" content="978-0134190440"/>
<data>...</data>
</book>
`
var b bytes.Buffer
// Compress the data using the specially crafted dictionary.
zw, err := flate.NewWriterDict(&b, flate.DefaultCompression, []byte(dict))
if err != nil {
log.Fatal(err)
}
if _, err := io.Copy(zw, strings.NewReader(data)); err != nil {
log.Fatal(err)
}
if err := zw.Close(); err != nil {
log.Fatal(err)
}
// The decompressor must use the same dictionary as the compressor.
// Otherwise, the input may appear as corrupted.
fmt.Println("Decompressed output using the dictionary:")
zr := flate.NewReaderDict(bytes.NewReader(b.Bytes()), []byte(dict))
if _, err := io.Copy(os.Stdout, zr); err != nil {
log.Fatal(err)
}
if err := zr.Close(); err != nil {
log.Fatal(err)
}
fmt.Println()
// Substitute all of the bytes in the dictionary with a '#' to visually
// demonstrate the approximate effectiveness of using a preset dictionary.
fmt.Println("Substrings matched by the dictionary are marked with #:")
hashDict := []byte(dict)
for i := range hashDict {
hashDict[i] = '#'
}
zr = flate.NewReaderDict(&b, hashDict)
if _, err := io.Copy(os.Stdout, zr); err != nil {
log.Fatal(err)
}
if err := zr.Close(); err != nil {
log.Fatal(err)
}
// Output:
// Decompressed output using the dictionary:
// <?xml version="1.0"?>
// <book>
// <meta name="title" content="The Go Programming Language"/>
// <meta name="authors" content="Alan Donovan and Brian Kernighan"/>
// <meta name="published" content="2015-10-26"/>
// <meta name="isbn" content="978-0134190440"/>
// <data>...</data>
// </book>
//
// Substrings matched by the dictionary are marked with #:
// #####################
// ######
// ############title###########The Go Programming Language"/#
// ############authors###########Alan Donovan and Brian Kernighan"/#
// ############published###########2015-10-26"/#
// ############isbn###########978-0134190440"/#
// ######...</#####
// </#####
}
// DEFLATE is suitable for transmitting compressed data across the network.
func Example_synchronization() {
var wg sync.WaitGroup
defer wg.Wait()
// Use io.Pipe to simulate a network connection.
// A real network application should take care to properly close the
// underlying connection.
rp, wp := io.Pipe()
// Start a goroutine to act as the transmitter.
wg.Add(1)
go func() {
defer wg.Done()
zw, err := flate.NewWriter(wp, flate.BestSpeed)
if err != nil {
log.Fatal(err)
}
b := make([]byte, 256)
for _, m := range strings.Fields("A long time ago in a galaxy far, far away...") {
// We use a simple framing format where the first byte is the
// message length, followed the message itself.
b[0] = uint8(copy(b[1:], m))
if _, err := zw.Write(b[:1+len(m)]); err != nil {
log.Fatal(err)
}
// Flush ensures that the receiver can read all data sent so far.
if err := zw.Flush(); err != nil {
log.Fatal(err)
}
}
if err := zw.Close(); err != nil {
log.Fatal(err)
}
}()
// Start a goroutine to act as the receiver.
wg.Add(1)
go func() {
defer wg.Done()
zr := flate.NewReader(rp)
b := make([]byte, 256)
for {
// Read the message length.
// This is guaranteed to return for every corresponding
// Flush and Close on the transmitter side.
if _, err := io.ReadFull(zr, b[:1]); err != nil {
if err == io.EOF {
break // The transmitter closed the stream
}
log.Fatal(err)
}
// Read the message content.
n := int(b[0])
if _, err := io.ReadFull(zr, b[:n]); err != nil {
log.Fatal(err)
}
fmt.Printf("Received %d bytes: %s\n", n, b[:n])
}
fmt.Println()
if err := zr.Close(); err != nil {
log.Fatal(err)
}
}()
// Output:
// Received 1 bytes: A
// Received 4 bytes: long
// Received 4 bytes: time
// Received 3 bytes: ago
// Received 2 bytes: in
// Received 1 bytes: a
// Received 6 bytes: galaxy
// Received 4 bytes: far,
// Received 3 bytes: far
// Received 7 bytes: away...
}
image/internal/compress/flate/flate_test.go
+352
View File
@@ -0,0 +1,352 @@
// Copyright 2009 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 test tests some internals of the flate package.
// The tests in package compress/gzip serve as the
// end-to-end test of the decompressor.
package flate
import (
"bytes"
"encoding/hex"
"io"
"strings"
"testing"
)
// The following test should not panic.
func TestIssue5915(t *testing.T) {
bits := []int{4, 0, 0, 6, 4, 3, 2, 3, 3, 4, 4, 5, 0, 0, 0, 0, 5, 5, 6,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 11, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 7, 8, 6, 0, 11, 0, 8, 0, 6, 6, 10, 8}
var h huffmanDecoder
if h.init(bits) {
t.Fatalf("Given sequence of bits is bad, and should not succeed.")
}
}
// The following test should not panic.
func TestIssue5962(t *testing.T) {
bits := []int{4, 0, 0, 6, 4, 3, 2, 3, 3, 4, 4, 5, 0, 0, 0, 0,
5, 5, 6, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 11}
var h huffmanDecoder
if h.init(bits) {
t.Fatalf("Given sequence of bits is bad, and should not succeed.")
}
}
// The following test should not panic.
func TestIssue6255(t *testing.T) {
bits1 := []int{1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 11}
bits2 := []int{11, 13}
var h huffmanDecoder
if !h.init(bits1) {
t.Fatalf("Given sequence of bits is good and should succeed.")
}
if h.init(bits2) {
t.Fatalf("Given sequence of bits is bad and should not succeed.")
}
}
func TestInvalidEncoding(t *testing.T) {
// Initialize Huffman decoder to recognize "0".
var h huffmanDecoder
if !h.init([]int{1}) {
t.Fatal("Failed to initialize Huffman decoder")
}
// Initialize decompressor with invalid Huffman coding.
var f decompressor
f.r = bytes.NewReader([]byte{0xff})
_, err := f.huffSym(&h)
if err == nil {
t.Fatal("Should have rejected invalid bit sequence")
}
}
func TestInvalidBits(t *testing.T) {
oversubscribed := []int{1, 2, 3, 4, 4, 5}
incomplete := []int{1, 2, 4, 4}
var h huffmanDecoder
if h.init(oversubscribed) {
t.Fatal("Should reject oversubscribed bit-length set")
}
if h.init(incomplete) {
t.Fatal("Should reject incomplete bit-length set")
}
}
func TestStreams(t *testing.T) {
// To verify any of these hexstrings as valid or invalid flate streams
// according to the C zlib library, you can use the Python wrapper library:
// >>> hex_string = "010100feff11"
// >>> import zlib
// >>> zlib.decompress(hex_string.decode("hex"), -15) # Negative means raw DEFLATE
// '\x11'
testCases := []struct {
desc string // Description of the stream
stream string // Hexstring of the input DEFLATE stream
want string // Expected result. Use "fail" to expect failure
}{{
"degenerate HCLenTree",
"05e0010000000000100000000000000000000000000000000000000000000000" +
"00000000000000000004",
"fail",
}, {
"complete HCLenTree, empty HLitTree, empty HDistTree",
"05e0010400000000000000000000000000000000000000000000000000000000" +
"00000000000000000010",
"fail",
}, {
"empty HCLenTree",
"05e0010000000000000000000000000000000000000000000000000000000000" +
"00000000000000000010",
"fail",
}, {
"complete HCLenTree, complete HLitTree, empty HDistTree, use missing HDist symbol",
"000100feff000de0010400000000100000000000000000000000000000000000" +
"0000000000000000000000000000002c",
"fail",
}, {
"complete HCLenTree, complete HLitTree, degenerate HDistTree, use missing HDist symbol",
"000100feff000de0010000000000000000000000000000000000000000000000" +
"00000000000000000610000000004070",
"fail",
}, {
"complete HCLenTree, empty HLitTree, empty HDistTree",
"05e0010400000000100400000000000000000000000000000000000000000000" +
"0000000000000000000000000008",
"fail",
}, {
"complete HCLenTree, empty HLitTree, degenerate HDistTree",
"05e0010400000000100400000000000000000000000000000000000000000000" +
"0000000000000000000800000008",
"fail",
}, {
"complete HCLenTree, degenerate HLitTree, degenerate HDistTree, use missing HLit symbol",
"05e0010400000000100000000000000000000000000000000000000000000000" +
"0000000000000000001c",
"fail",
}, {
"complete HCLenTree, complete HLitTree, too large HDistTree",
"edff870500000000200400000000000000000000000000000000000000000000" +
"000000000000000000080000000000000004",
"fail",
}, {
"complete HCLenTree, complete HLitTree, empty HDistTree, excessive repeater code",
"edfd870500000000200400000000000000000000000000000000000000000000" +
"000000000000000000e8b100",
"fail",
}, {
"complete HCLenTree, complete HLitTree, empty HDistTree of normal length 30",
"05fd01240000000000f8ffffffffffffffffffffffffffffffffffffffffffff" +
"ffffffffffffffffff07000000fe01",
"",
}, {
"complete HCLenTree, complete HLitTree, empty HDistTree of excessive length 31",
"05fe01240000000000f8ffffffffffffffffffffffffffffffffffffffffffff" +
"ffffffffffffffffff07000000fc03",
"fail",
}, {
"complete HCLenTree, over-subscribed HLitTree, empty HDistTree",
"05e001240000000000fcffffffffffffffffffffffffffffffffffffffffffff" +
"ffffffffffffffffff07f00f",
"fail",
}, {
"complete HCLenTree, under-subscribed HLitTree, empty HDistTree",
"05e001240000000000fcffffffffffffffffffffffffffffffffffffffffffff" +
"fffffffffcffffffff07f00f",
"fail",
}, {
"complete HCLenTree, complete HLitTree with single code, empty HDistTree",
"05e001240000000000f8ffffffffffffffffffffffffffffffffffffffffffff" +
"ffffffffffffffffff07f00f",
"01",
}, {
"complete HCLenTree, complete HLitTree with multiple codes, empty HDistTree",
"05e301240000000000f8ffffffffffffffffffffffffffffffffffffffffffff" +
"ffffffffffffffffff07807f",
"01",
}, {
"complete HCLenTree, complete HLitTree, degenerate HDistTree, use valid HDist symbol",
"000100feff000de0010400000000100000000000000000000000000000000000" +
"0000000000000000000000000000003c",
"00000000",
}, {
"complete HCLenTree, degenerate HLitTree, degenerate HDistTree",
"05e0010400000000100000000000000000000000000000000000000000000000" +
"0000000000000000000c",
"",
}, {
"complete HCLenTree, degenerate HLitTree, empty HDistTree",
"05e0010400000000100000000000000000000000000000000000000000000000" +
"00000000000000000004",
"",
}, {
"complete HCLenTree, complete HLitTree, empty HDistTree, spanning repeater code",
"edfd870500000000200400000000000000000000000000000000000000000000" +
"000000000000000000e8b000",
"",
}, {
"complete HCLenTree with length codes, complete HLitTree, empty HDistTree",
"ede0010400000000100000000000000000000000000000000000000000000000" +
"0000000000000000000400004000",
"",
}, {
"complete HCLenTree, complete HLitTree, degenerate HDistTree, use valid HLit symbol 284 with count 31",
"000100feff00ede0010400000000100000000000000000000000000000000000" +
"000000000000000000000000000000040000407f00",
"0000000000000000000000000000000000000000000000000000000000000000" +
"0000000000000000000000000000000000000000000000000000000000000000" +
"0000000000000000000000000000000000000000000000000000000000000000" +
"0000000000000000000000000000000000000000000000000000000000000000" +
"0000000000000000000000000000000000000000000000000000000000000000" +
"0000000000000000000000000000000000000000000000000000000000000000" +
"0000000000000000000000000000000000000000000000000000000000000000" +
"0000000000000000000000000000000000000000000000000000000000000000" +
"000000",
}, {
"complete HCLenTree, complete HLitTree, degenerate HDistTree, use valid HLit and HDist symbols",
"0cc2010d00000082b0ac4aff0eb07d27060000ffff",
"616263616263",
}, {
"fixed block, use reserved symbol 287",
"33180700",
"fail",
}, {
"raw block",
"010100feff11",
"11",
}, {
"issue 10426 - over-subscribed HCLenTree causes a hang",
"344c4a4e494d4b070000ff2e2eff2e2e2e2e2eff",
"fail",
}, {
"issue 11030 - empty HDistTree unexpectedly leads to error",
"05c0070600000080400fff37a0ca",
"",
}, {
"issue 11033 - empty HDistTree unexpectedly leads to error",
"050fb109c020cca5d017dcbca044881ee1034ec149c8980bbc413c2ab35be9dc" +
"b1473449922449922411202306ee97b0383a521b4ffdcf3217f9f7d3adb701",
"3130303634342068652e706870005d05355f7ed957ff084a90925d19e3ebc6d0" +
"c6d7",
}}
for i, tc := range testCases {
data, err := hex.DecodeString(tc.stream)
if err != nil {
t.Fatal(err)
}
data, err = io.ReadAll(NewReader(bytes.NewReader(data)))
if tc.want == "fail" {
if err == nil {
t.Errorf("#%d (%s): got nil error, want non-nil", i, tc.desc)
}
} else {
if err != nil {
t.Errorf("#%d (%s): %v", i, tc.desc, err)
continue
}
if got := hex.EncodeToString(data); got != tc.want {
t.Errorf("#%d (%s):\ngot %q\nwant %q", i, tc.desc, got, tc.want)
}
}
}
}
func TestTruncatedStreams(t *testing.T) {
const data = "\x00\f\x00\xf3\xffhello, world\x01\x00\x00\xff\xff"
for i := 0; i < len(data)-1; i++ {
r := NewReader(strings.NewReader(data[:i]))
_, err := io.Copy(io.Discard, r)
if err != io.ErrUnexpectedEOF {
t.Errorf("io.Copy(%d) on truncated stream: got %v, want %v", i, err, io.ErrUnexpectedEOF)
}
}
}
// Verify that flate.Reader.Read returns (n, io.EOF) instead
// of (n, nil) + (0, io.EOF) when possible.
//
// This helps net/http.Transport reuse HTTP/1 connections more
// aggressively.
//
// See https://github.com/google/go-github/pull/317 for background.
func TestReaderEarlyEOF(t *testing.T) {
t.Parallel()
testSizes := []int{
1, 2, 3, 4, 5, 6, 7, 8,
100, 1000, 10000, 100000,
128, 1024, 16384, 131072,
// Testing multiples of windowSize triggers the case
// where Read will fail to return an early io.EOF.
windowSize * 1, windowSize * 2, windowSize * 3,
}
var maxSize int
for _, n := range testSizes {
if maxSize < n {
maxSize = n
}
}
readBuf := make([]byte, 40)
data := make([]byte, maxSize)
for i := range data {
data[i] = byte(i)
}
for _, sz := range testSizes {
if testing.Short() && sz > windowSize {
continue
}
for _, flush := range []bool{true, false} {
earlyEOF := true // Do we expect early io.EOF?
var buf bytes.Buffer
w, _ := NewWriter(&buf, 5)
w.Write(data[:sz])
if flush {
// If a Flush occurs after all the actual data, the flushing
// semantics dictate that we will observe a (0, io.EOF) since
// Read must return data before it knows that the stream ended.
w.Flush()
earlyEOF = false
}
w.Close()
r := NewReader(&buf)
for {
n, err := r.Read(readBuf)
if err == io.EOF {
// If the availWrite == windowSize, then that means that the
// previous Read returned because the write buffer was full
// and it just so happened that the stream had no more data.
// This situation is rare, but unavoidable.
if r.(*decompressor).dict.availWrite() == windowSize {
earlyEOF = false
}
if n == 0 && earlyEOF {
t.Errorf("On size:%d flush:%v, Read() = (0, io.EOF), want (n, io.EOF)", sz, flush)
}
if n != 0 && !earlyEOF {
t.Errorf("On size:%d flush:%v, Read() = (%d, io.EOF), want (0, io.EOF)", sz, flush, n)
}
break
}
if err != nil {
t.Fatal(err)
}
}
}
}
}
image/internal/compress/flate/huffman_bit_writer.go
+704
View File
@@ -0,0 +1,704 @@
// Copyright 2009 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.
package flate
import (
"io"
)
const (
// The largest offset code.
offsetCodeCount = 30
// The special code used to mark the end of a block.
endBlockMarker = 256
// The first length code.
lengthCodesStart = 257
// The number of codegen codes.
codegenCodeCount = 19
badCode = 255
// bufferFlushSize indicates the buffer size
// after which bytes are flushed to the writer.
// Should preferably be a multiple of 6, since
// we accumulate 6 bytes between writes to the buffer.
bufferFlushSize = 240
// bufferSize is the actual output byte buffer size.
// It must have additional headroom for a flush
// which can contain up to 8 bytes.
bufferSize = bufferFlushSize + 8
)
// The number of extra bits needed by length code X - LENGTH_CODES_START.
var lengthExtraBits = []int8{
/* 257 */ 0, 0, 0,
/* 260 */ 0, 0, 0, 0, 0, 1, 1, 1, 1, 2,
/* 270 */ 2, 2, 2, 3, 3, 3, 3, 4, 4, 4,
/* 280 */ 4, 5, 5, 5, 5, 0,
}
// The length indicated by length code X - LENGTH_CODES_START.
var lengthBase = []uint32{
0, 1, 2, 3, 4, 5, 6, 7, 8, 10,
12, 14, 16, 20, 24, 28, 32, 40, 48, 56,
64, 80, 96, 112, 128, 160, 192, 224, 255,
}
// offset code word extra bits.
var offsetExtraBits = []int8{
0, 0, 0, 0, 1, 1, 2, 2, 3, 3,
4, 4, 5, 5, 6, 6, 7, 7, 8, 8,
9, 9, 10, 10, 11, 11, 12, 12, 13, 13,
}
var offsetBase = []uint32{
0x000000, 0x000001, 0x000002, 0x000003, 0x000004,
0x000006, 0x000008, 0x00000c, 0x000010, 0x000018,
0x000020, 0x000030, 0x000040, 0x000060, 0x000080,
0x0000c0, 0x000100, 0x000180, 0x000200, 0x000300,
0x000400, 0x000600, 0x000800, 0x000c00, 0x001000,
0x001800, 0x002000, 0x003000, 0x004000, 0x006000,
}
// The odd order in which the codegen code sizes are written.
var codegenOrder = []uint32{16, 17, 18, 0, 8, 7, 9, 6, 10, 5, 11, 4, 12, 3, 13, 2, 14, 1, 15}
type huffmanBitWriter struct {
// writer is the underlying writer.
// Do not use it directly; use the write method, which ensures
// that Write errors are sticky.
writer io.Writer
// Data waiting to be written is bytes[0:nbytes]
// and then the low nbits of bits. Data is always written
// sequentially into the bytes array.
bits uint64
nbits uint
bytes [bufferSize]byte
codegenFreq [codegenCodeCount]int32
nbytes int
literalFreq []int32
offsetFreq []int32
codegen []uint8
literalEncoding *huffmanEncoder
offsetEncoding *huffmanEncoder
codegenEncoding *huffmanEncoder
err error
}
func newHuffmanBitWriter(w io.Writer) *huffmanBitWriter {
return &huffmanBitWriter{
writer: w,
literalFreq: make([]int32, maxNumLit),
offsetFreq: make([]int32, offsetCodeCount),
codegen: make([]uint8, maxNumLit+offsetCodeCount+1),
literalEncoding: newHuffmanEncoder(maxNumLit),
codegenEncoding: newHuffmanEncoder(codegenCodeCount),
offsetEncoding: newHuffmanEncoder(offsetCodeCount),
}
}
func (w *huffmanBitWriter) reset(writer io.Writer) {
w.writer = writer
w.bits, w.nbits, w.nbytes, w.err = 0, 0, 0, nil
}
func (w *huffmanBitWriter) flush() {
if w.err != nil {
w.nbits = 0
return
}
n := w.nbytes
for w.nbits != 0 {
w.bytes[n] = byte(w.bits)
w.bits >>= 8
if w.nbits > 8 { // Avoid underflow
w.nbits -= 8
} else {
w.nbits = 0
}
n++
}
w.bits = 0
w.write(w.bytes[:n])
w.nbytes = 0
}
func (w *huffmanBitWriter) write(b []byte) {
if w.err != nil {
return
}
_, w.err = w.writer.Write(b)
}
func (w *huffmanBitWriter) writeBits(b int32, nb uint) {
if w.err != nil {
return
}
w.bits |= uint64(b) << w.nbits
w.nbits += nb
if w.nbits >= 48 {
bits := w.bits
w.bits >>= 48
w.nbits -= 48
n := w.nbytes
bytes := w.bytes[n : n+6]
bytes[0] = byte(bits)
bytes[1] = byte(bits >> 8)
bytes[2] = byte(bits >> 16)
bytes[3] = byte(bits >> 24)
bytes[4] = byte(bits >> 32)
bytes[5] = byte(bits >> 40)
n += 6
if n >= bufferFlushSize {
w.write(w.bytes[:n])
n = 0
}
w.nbytes = n
}
}
func (w *huffmanBitWriter) writeBytes(bytes []byte) {
if w.err != nil {
return
}
n := w.nbytes
if w.nbits&7 != 0 {
w.err = InternalError("writeBytes with unfinished bits")
return
}
for w.nbits != 0 {
w.bytes[n] = byte(w.bits)
w.bits >>= 8
w.nbits -= 8
n++
}
if n != 0 {
w.write(w.bytes[:n])
}
w.nbytes = 0
w.write(bytes)
}
// RFC 1951 3.2.7 specifies a special run-length encoding for specifying
// the literal and offset lengths arrays (which are concatenated into a single
// array). This method generates that run-length encoding.
//
// The result is written into the codegen array, and the frequencies
// of each code is written into the codegenFreq array.
// Codes 0-15 are single byte codes. Codes 16-18 are followed by additional
// information. Code badCode is an end marker
//
// numLiterals The number of literals in literalEncoding
// numOffsets The number of offsets in offsetEncoding
// litenc, offenc The literal and offset encoder to use
func (w *huffmanBitWriter) generateCodegen(numLiterals int, numOffsets int, litEnc, offEnc *huffmanEncoder) {
for i := range w.codegenFreq {
w.codegenFreq[i] = 0
}
// Note that we are using codegen both as a temporary variable for holding
// a copy of the frequencies, and as the place where we put the result.
// This is fine because the output is always shorter than the input used
// so far.
codegen := w.codegen // cache
// Copy the concatenated code sizes to codegen. Put a marker at the end.
cgnl := codegen[:numLiterals]
for i := range cgnl {
cgnl[i] = uint8(litEnc.codes[i].len)
}
cgnl = codegen[numLiterals : numLiterals+numOffsets]
for i := range cgnl {
cgnl[i] = uint8(offEnc.codes[i].len)
}
codegen[numLiterals+numOffsets] = badCode
size := codegen[0]
count := 1
outIndex := 0
for inIndex := 1; size != badCode; inIndex++ {
// INVARIANT: We have seen "count" copies of size that have not yet
// had output generated for them.
nextSize := codegen[inIndex]
if nextSize == size {
count++
continue
}
// We need to generate codegen indicating "count" of size.
if size != 0 {
codegen[outIndex] = size
outIndex++
w.codegenFreq[size]++
count--
for count >= 3 {
n := 6
if n > count {
n = count
}
codegen[outIndex] = 16
outIndex++
codegen[outIndex] = uint8(n - 3)
outIndex++
w.codegenFreq[16]++
count -= n
}
} else {
for count >= 11 {
n := 138
if n > count {
n = count
}
codegen[outIndex] = 18
outIndex++
codegen[outIndex] = uint8(n - 11)
outIndex++
w.codegenFreq[18]++
count -= n
}
if count >= 3 {
// count >= 3 && count <= 10
codegen[outIndex] = 17
outIndex++
codegen[outIndex] = uint8(count - 3)
outIndex++
w.codegenFreq[17]++
count = 0
}
}
count--
for ; count >= 0; count-- {
codegen[outIndex] = size
outIndex++
w.codegenFreq[size]++
}
// Set up invariant for next time through the loop.
size = nextSize
count = 1
}
// Marker indicating the end of the codegen.
codegen[outIndex] = badCode
}
// dynamicSize returns the size of dynamically encoded data in bits.
func (w *huffmanBitWriter) dynamicSize(litEnc, offEnc *huffmanEncoder, extraBits int) (size, numCodegens int) {
numCodegens = len(w.codegenFreq)
for numCodegens > 4 && w.codegenFreq[codegenOrder[numCodegens-1]] == 0 {
numCodegens--
}
header := 3 + 5 + 5 + 4 + (3 * numCodegens) +
w.codegenEncoding.bitLength(w.codegenFreq[:]) +
int(w.codegenFreq[16])*2 +
int(w.codegenFreq[17])*3 +
int(w.codegenFreq[18])*7
size = header +
litEnc.bitLength(w.literalFreq) +
offEnc.bitLength(w.offsetFreq) +
extraBits
return size, numCodegens
}
// fixedSize returns the size of dynamically encoded data in bits.
func (w *huffmanBitWriter) fixedSize(extraBits int) int {
return 3 +
fixedLiteralEncoding.bitLength(w.literalFreq) +
fixedOffsetEncoding.bitLength(w.offsetFreq) +
extraBits
}
// storedSize calculates the stored size, including header.
// The function returns the size in bits and whether the block
// fits inside a single block.
func (w *huffmanBitWriter) storedSize(in []byte) (int, bool) {
if in == nil {
return 0, false
}
if len(in) <= maxStoreBlockSize {
return (len(in) + 5) * 8, true
}
return 0, false
}
func (w *huffmanBitWriter) writeCode(c hcode) {
if w.err != nil {
return
}
w.bits |= uint64(c.code) << w.nbits
w.nbits += uint(c.len)
if w.nbits >= 48 {
bits := w.bits
w.bits >>= 48
w.nbits -= 48
n := w.nbytes
bytes := w.bytes[n : n+6]
bytes[0] = byte(bits)
bytes[1] = byte(bits >> 8)
bytes[2] = byte(bits >> 16)
bytes[3] = byte(bits >> 24)
bytes[4] = byte(bits >> 32)
bytes[5] = byte(bits >> 40)
n += 6
if n >= bufferFlushSize {
w.write(w.bytes[:n])
n = 0
}
w.nbytes = n
}
}
// Write the header of a dynamic Huffman block to the output stream.
//
// numLiterals The number of literals specified in codegen
// numOffsets The number of offsets specified in codegen
// numCodegens The number of codegens used in codegen
func (w *huffmanBitWriter) writeDynamicHeader(numLiterals int, numOffsets int, numCodegens int, isEof bool) {
if w.err != nil {
return
}
var firstBits int32 = 4
if isEof {
firstBits = 5
}
w.writeBits(firstBits, 3)
w.writeBits(int32(numLiterals-257), 5)
w.writeBits(int32(numOffsets-1), 5)
w.writeBits(int32(numCodegens-4), 4)
for i := 0; i < numCodegens; i++ {
value := uint(w.codegenEncoding.codes[codegenOrder[i]].len)
w.writeBits(int32(value), 3)
}
i := 0
for {
var codeWord int = int(w.codegen[i])
i++
if codeWord == badCode {
break
}
w.writeCode(w.codegenEncoding.codes[uint32(codeWord)])
switch codeWord {
case 16:
w.writeBits(int32(w.codegen[i]), 2)
i++
break
case 17:
w.writeBits(int32(w.codegen[i]), 3)
i++
break
case 18:
w.writeBits(int32(w.codegen[i]), 7)
i++
break
}
}
}
func (w *huffmanBitWriter) writeStoredHeader(length int, isEof bool) {
if w.err != nil {
return
}
var flag int32
if isEof {
flag = 1
}
w.writeBits(flag, 3)
w.flush()
w.writeBits(int32(length), 16)
w.writeBits(int32(^uint16(length)), 16)
}
func (w *huffmanBitWriter) writeFixedHeader(isEof bool) {
if w.err != nil {
return
}
// Indicate that we are a fixed Huffman block
var value int32 = 2
if isEof {
value = 3
}
w.writeBits(value, 3)
}
// writeBlock will write a block of tokens with the smallest encoding.
// The original input can be supplied, and if the huffman encoded data
// is larger than the original bytes, the data will be written as a
// stored block.
// If the input is nil, the tokens will always be Huffman encoded.
func (w *huffmanBitWriter) writeBlock(tokens []token, eof bool, input []byte) {
if w.err != nil {
return
}
tokens = append(tokens, endBlockMarker)
numLiterals, numOffsets := w.indexTokens(tokens)
var extraBits int
storedSize, storable := w.storedSize(input)
if storable {
// We only bother calculating the costs of the extra bits required by
// the length of offset fields (which will be the same for both fixed
// and dynamic encoding), if we need to compare those two encodings
// against stored encoding.
for lengthCode := lengthCodesStart + 8; lengthCode < numLiterals; lengthCode++ {
// First eight length codes have extra size = 0.
extraBits += int(w.literalFreq[lengthCode]) * int(lengthExtraBits[lengthCode-lengthCodesStart])
}
for offsetCode := 4; offsetCode < numOffsets; offsetCode++ {
// First four offset codes have extra size = 0.
extraBits += int(w.offsetFreq[offsetCode]) * int(offsetExtraBits[offsetCode])
}
}
// Figure out smallest code.
// Fixed Huffman baseline.
var literalEncoding = fixedLiteralEncoding
var offsetEncoding = fixedOffsetEncoding
var size = w.fixedSize(extraBits)
// Dynamic Huffman?
var numCodegens int
// Generate codegen and codegenFrequencies, which indicates how to encode
// the literalEncoding and the offsetEncoding.
w.generateCodegen(numLiterals, numOffsets, w.literalEncoding, w.offsetEncoding)
w.codegenEncoding.generate(w.codegenFreq[:], 7)
dynamicSize, numCodegens := w.dynamicSize(w.literalEncoding, w.offsetEncoding, extraBits)
if dynamicSize < size {
size = dynamicSize
literalEncoding = w.literalEncoding
offsetEncoding = w.offsetEncoding
}
// Stored bytes?
if storable && storedSize < size {
w.writeStoredHeader(len(input), eof)
w.writeBytes(input)
return
}
// Huffman.
if literalEncoding == fixedLiteralEncoding {
w.writeFixedHeader(eof)
} else {
w.writeDynamicHeader(numLiterals, numOffsets, numCodegens, eof)
}
// Write the tokens.
w.writeTokens(tokens, literalEncoding.codes, offsetEncoding.codes)
}
// writeBlockDynamic encodes a block using a dynamic Huffman table.
// This should be used if the symbols used have a disproportionate
// histogram distribution.
// If input is supplied and the compression savings are below 1/16th of the
// input size the block is stored.
func (w *huffmanBitWriter) writeBlockDynamic(tokens []token, eof bool, input []byte) {
if w.err != nil {
return
}
tokens = append(tokens, endBlockMarker)
numLiterals, numOffsets := w.indexTokens(tokens)
// Generate codegen and codegenFrequencies, which indicates how to encode
// the literalEncoding and the offsetEncoding.
w.generateCodegen(numLiterals, numOffsets, w.literalEncoding, w.offsetEncoding)
w.codegenEncoding.generate(w.codegenFreq[:], 7)
size, numCodegens := w.dynamicSize(w.literalEncoding, w.offsetEncoding, 0)
// Store bytes, if we don't get a reasonable improvement.
if ssize, storable := w.storedSize(input); storable && ssize < (size+size>>4) {
w.writeStoredHeader(len(input), eof)
w.writeBytes(input)
return
}
// Write Huffman table.
w.writeDynamicHeader(numLiterals, numOffsets, numCodegens, eof)
// Write the tokens.
w.writeTokens(tokens, w.literalEncoding.codes, w.offsetEncoding.codes)
}
// indexTokens indexes a slice of tokens, and updates
// literalFreq and offsetFreq, and generates literalEncoding
// and offsetEncoding.
// The number of literal and offset tokens is returned.
func (w *huffmanBitWriter) indexTokens(tokens []token) (numLiterals, numOffsets int) {
for i := range w.literalFreq {
w.literalFreq[i] = 0
}
for i := range w.offsetFreq {
w.offsetFreq[i] = 0
}
for _, t := range tokens {
if t < matchType {
w.literalFreq[t.literal()]++
continue
}
length := t.length()
offset := t.offset()
w.literalFreq[lengthCodesStart+lengthCode(length)]++
w.offsetFreq[offsetCode(offset)]++
}
// get the number of literals
numLiterals = len(w.literalFreq)
for w.literalFreq[numLiterals-1] == 0 {
numLiterals--
}
// get the number of offsets
numOffsets = len(w.offsetFreq)
for numOffsets > 0 && w.offsetFreq[numOffsets-1] == 0 {
numOffsets--
}
if numOffsets == 0 {
// We haven't found a single match. If we want to go with the dynamic encoding,
// we should count at least one offset to be sure that the offset huffman tree could be encoded.
w.offsetFreq[0] = 1
numOffsets = 1
}
w.literalEncoding.generate(w.literalFreq, 15)
w.offsetEncoding.generate(w.offsetFreq, 15)
return
}
// writeTokens writes a slice of tokens to the output.
// codes for literal and offset encoding must be supplied.
func (w *huffmanBitWriter) writeTokens(tokens []token, leCodes, oeCodes []hcode) {
if w.err != nil {
return
}
for _, t := range tokens {
if t < matchType {
w.writeCode(leCodes[t.literal()])
continue
}
// Write the length
length := t.length()
lengthCode := lengthCode(length)
w.writeCode(leCodes[lengthCode+lengthCodesStart])
extraLengthBits := uint(lengthExtraBits[lengthCode])
if extraLengthBits > 0 {
extraLength := int32(length - lengthBase[lengthCode])
w.writeBits(extraLength, extraLengthBits)
}
// Write the offset
offset := t.offset()
offsetCode := offsetCode(offset)
w.writeCode(oeCodes[offsetCode])
extraOffsetBits := uint(offsetExtraBits[offsetCode])
if extraOffsetBits > 0 {
extraOffset := int32(offset - offsetBase[offsetCode])
w.writeBits(extraOffset, extraOffsetBits)
}
}
}
// huffOffset is a static offset encoder used for huffman only encoding.
// It can be reused since we will not be encoding offset values.
var huffOffset *huffmanEncoder
func init() {
offsetFreq := make([]int32, offsetCodeCount)
offsetFreq[0] = 1
huffOffset = newHuffmanEncoder(offsetCodeCount)
huffOffset.generate(offsetFreq, 15)
}
// writeBlockHuff encodes a block of bytes as either
// Huffman encoded literals or uncompressed bytes if the
// results only gains very little from compression.
func (w *huffmanBitWriter) writeBlockHuff(eof bool, input []byte) {
if w.err != nil {
return
}
// Clear histogram
for i := range w.literalFreq {
w.literalFreq[i] = 0
}
// Add everything as literals
histogram(input, w.literalFreq)
w.literalFreq[endBlockMarker] = 1
const numLiterals = endBlockMarker + 1
w.offsetFreq[0] = 1
const numOffsets = 1
w.literalEncoding.generate(w.literalFreq, 15)
// Figure out smallest code.
// Always use dynamic Huffman or Store
var numCodegens int
// Generate codegen and codegenFrequencies, which indicates how to encode
// the literalEncoding and the offsetEncoding.
w.generateCodegen(numLiterals, numOffsets, w.literalEncoding, huffOffset)
w.codegenEncoding.generate(w.codegenFreq[:], 7)
size, numCodegens := w.dynamicSize(w.literalEncoding, huffOffset, 0)
// Store bytes, if we don't get a reasonable improvement.
if ssize, storable := w.storedSize(input); storable && ssize < (size+size>>4) {
w.writeStoredHeader(len(input), eof)
w.writeBytes(input)
return
}
// Huffman.
w.writeDynamicHeader(numLiterals, numOffsets, numCodegens, eof)
encoding := w.literalEncoding.codes[:257]
n := w.nbytes
for _, t := range input {
// Bitwriting inlined, ~30% speedup
c := encoding[t]
w.bits |= uint64(c.code) << w.nbits
w.nbits += uint(c.len)
if w.nbits < 48 {
continue
}
// Store 6 bytes
bits := w.bits
w.bits >>= 48
w.nbits -= 48
bytes := w.bytes[n : n+6]
bytes[0] = byte(bits)
bytes[1] = byte(bits >> 8)
bytes[2] = byte(bits >> 16)
bytes[3] = byte(bits >> 24)
bytes[4] = byte(bits >> 32)
bytes[5] = byte(bits >> 40)
n += 6
if n < bufferFlushSize {
continue
}
w.write(w.bytes[:n])
if w.err != nil {
return // Return early in the event of write failures
}
n = 0
}
w.nbytes = n
w.writeCode(encoding[endBlockMarker])
}
// histogram accumulates a histogram of b in h.
//
// len(h) must be >= 256, and h's elements must be all zeroes.
func histogram(b []byte, h []int32) {
h = h[:256]
for _, t := range b {
h[t]++
}
}
image/internal/compress/flate/huffman_bit_writer_test.go
+365
View File
File diff suppressed because one or more lines are too long
image/internal/compress/flate/huffman_code.go
+349
View File
@@ -0,0 +1,349 @@
// Copyright 2009 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.
package flate
import (
"math"
"math/bits"
"sort"
)
// hcode is a huffman code with a bit code and bit length.
type hcode struct {
code, len uint16
}
type huffmanEncoder struct {
codes []hcode
freqcache []literalNode
bitCount [17]int32
lns byLiteral // stored to avoid repeated allocation in generate
lfs byFreq // stored to avoid repeated allocation in generate
}
type literalNode struct {
literal uint16
freq int32
}
// A levelInfo describes the state of the constructed tree for a given depth.
type levelInfo struct {
// Our level. for better printing
level int32
// The frequency of the last node at this level
lastFreq int32
// The frequency of the next character to add to this level
nextCharFreq int32
// The frequency of the next pair (from level below) to add to this level.
// Only valid if the "needed" value of the next lower level is 0.
nextPairFreq int32
// The number of chains remaining to generate for this level before moving
// up to the next level
needed int32
}
// set sets the code and length of an hcode.
func (h *hcode) set(code uint16, length uint16) {
h.len = length
h.code = code
}
func maxNode() literalNode { return literalNode{math.MaxUint16, math.MaxInt32} }
func newHuffmanEncoder(size int) *huffmanEncoder {
return &huffmanEncoder{codes: make([]hcode, size)}
}
// Generates a HuffmanCode corresponding to the fixed literal table
func generateFixedLiteralEncoding() *huffmanEncoder {
h := newHuffmanEncoder(maxNumLit)
codes := h.codes
var ch uint16
for ch = 0; ch < maxNumLit; ch++ {
var bits uint16
var size uint16
switch {
case ch < 144:
// size 8, 000110000 .. 10111111
bits = ch + 48
size = 8
break
case ch < 256:
// size 9, 110010000 .. 111111111
bits = ch + 400 - 144
size = 9
break
case ch < 280:
// size 7, 0000000 .. 0010111
bits = ch - 256
size = 7
break
default:
// size 8, 11000000 .. 11000111
bits = ch + 192 - 280
size = 8
}
codes[ch] = hcode{code: reverseBits(bits, byte(size)), len: size}
}
return h
}
func generateFixedOffsetEncoding() *huffmanEncoder {
h := newHuffmanEncoder(30)
codes := h.codes
for ch := range codes {
codes[ch] = hcode{code: reverseBits(uint16(ch), 5), len: 5}
}
return h
}
var fixedLiteralEncoding *huffmanEncoder = generateFixedLiteralEncoding()
var fixedOffsetEncoding *huffmanEncoder = generateFixedOffsetEncoding()
func (h *huffmanEncoder) bitLength(freq []int32) int {
var total int
for i, f := range freq {
if f != 0 {
total += int(f) * int(h.codes[i].len)
}
}
return total
}
const maxBitsLimit = 16
// Return the number of literals assigned to each bit size in the Huffman encoding
//
// This method is only called when list.length >= 3
// The cases of 0, 1, and 2 literals are handled by special case code.
//
// list An array of the literals with non-zero frequencies
// and their associated frequencies. The array is in order of increasing
// frequency, and has as its last element a special element with frequency
// MaxInt32
// maxBits The maximum number of bits that should be used to encode any literal.
// Must be less than 16.
// return An integer array in which array[i] indicates the number of literals
// that should be encoded in i bits.
func (h *huffmanEncoder) bitCounts(list []literalNode, maxBits int32) []int32 {
if maxBits >= maxBitsLimit {
panic("flate: maxBits too large")
}
n := int32(len(list))
list = list[0 : n+1]
list[n] = maxNode()
// The tree can't have greater depth than n - 1, no matter what. This
// saves a little bit of work in some small cases
if maxBits > n-1 {
maxBits = n - 1
}
// Create information about each of the levels.
// A bogus "Level 0" whose sole purpose is so that
// level1.prev.needed==0. This makes level1.nextPairFreq
// be a legitimate value that never gets chosen.
var levels [maxBitsLimit]levelInfo
// leafCounts[i] counts the number of literals at the left
// of ancestors of the rightmost node at level i.
// leafCounts[i][j] is the number of literals at the left
// of the level j ancestor.
var leafCounts [maxBitsLimit][maxBitsLimit]int32
for level := int32(1); level <= maxBits; level++ {
// For every level, the first two items are the first two characters.
// We initialize the levels as if we had already figured this out.
levels[level] = levelInfo{
level: level,
lastFreq: list[1].freq,
nextCharFreq: list[2].freq,
nextPairFreq: list[0].freq + list[1].freq,
}
leafCounts[level][level] = 2
if level == 1 {
levels[level].nextPairFreq = math.MaxInt32
}
}
// We need a total of 2*n - 2 items at top level and have already generated 2.
levels[maxBits].needed = 2*n - 4
level := maxBits
for {
l := &levels[level]
if l.nextPairFreq == math.MaxInt32 && l.nextCharFreq == math.MaxInt32 {
// We've run out of both leafs and pairs.
// End all calculations for this level.
// To make sure we never come back to this level or any lower level,
// set nextPairFreq impossibly large.
l.needed = 0
levels[level+1].nextPairFreq = math.MaxInt32
level++
continue
}
prevFreq := l.lastFreq
if l.nextCharFreq < l.nextPairFreq {
// The next item on this row is a leaf node.
n := leafCounts[level][level] + 1
l.lastFreq = l.nextCharFreq
// Lower leafCounts are the same of the previous node.
leafCounts[level][level] = n
l.nextCharFreq = list[n].freq
} else {
// The next item on this row is a pair from the previous row.
// nextPairFreq isn't valid until we generate two
// more values in the level below
l.lastFreq = l.nextPairFreq
// Take leaf counts from the lower level, except counts[level] remains the same.
copy(leafCounts[level][:level], leafCounts[level-1][:level])
levels[l.level-1].needed = 2
}
if l.needed--; l.needed == 0 {
// We've done everything we need to do for this level.
// Continue calculating one level up. Fill in nextPairFreq
// of that level with the sum of the two nodes we've just calculated on
// this level.
if l.level == maxBits {
// All done!
break
}
levels[l.level+1].nextPairFreq = prevFreq + l.lastFreq
level++
} else {
// If we stole from below, move down temporarily to replenish it.
for levels[level-1].needed > 0 {
level--
}
}
}
// Somethings is wrong if at the end, the top level is null or hasn't used
// all of the leaves.
if leafCounts[maxBits][maxBits] != n {
panic("leafCounts[maxBits][maxBits] != n")
}
bitCount := h.bitCount[:maxBits+1]
bits := 1
counts := &leafCounts[maxBits]
for level := maxBits; level > 0; level-- {
// chain.leafCount gives the number of literals requiring at least "bits"
// bits to encode.
bitCount[bits] = counts[level] - counts[level-1]
bits++
}
return bitCount
}
// Look at the leaves and assign them a bit count and an encoding as specified
// in RFC 1951 3.2.2
func (h *huffmanEncoder) assignEncodingAndSize(bitCount []int32, list []literalNode) {
code := uint16(0)
for n, bits := range bitCount {
code <<= 1
if n == 0 || bits == 0 {
continue
}
// The literals list[len(list)-bits] .. list[len(list)-bits]
// are encoded using "bits" bits, and get the values
// code, code + 1, .... The code values are
// assigned in literal order (not frequency order).
chunk := list[len(list)-int(bits):]
h.lns.sort(chunk)
for _, node := range chunk {
h.codes[node.literal] = hcode{code: reverseBits(code, uint8(n)), len: uint16(n)}
code++
}
list = list[0 : len(list)-int(bits)]
}
}
// Update this Huffman Code object to be the minimum code for the specified frequency count.
//
// freq An array of frequencies, in which frequency[i] gives the frequency of literal i.
// maxBits The maximum number of bits to use for any literal.
func (h *huffmanEncoder) generate(freq []int32, maxBits int32) {
if h.freqcache == nil {
// Allocate a reusable buffer with the longest possible frequency table.
// Possible lengths are codegenCodeCount, offsetCodeCount and maxNumLit.
// The largest of these is maxNumLit, so we allocate for that case.
h.freqcache = make([]literalNode, maxNumLit+1)
}
list := h.freqcache[:len(freq)+1]
// Number of non-zero literals
count := 0
// Set list to be the set of all non-zero literals and their frequencies
for i, f := range freq {
if f != 0 {
list[count] = literalNode{uint16(i), f}
count++
} else {
list[count] = literalNode{}
h.codes[i].len = 0
}
}
list[len(freq)] = literalNode{}
list = list[:count]
if count <= 2 {
// Handle the small cases here, because they are awkward for the general case code. With
// two or fewer literals, everything has bit length 1.
for i, node := range list {
// "list" is in order of increasing literal value.
h.codes[node.literal].set(uint16(i), 1)
}
return
}
h.lfs.sort(list)
// Get the number of literals for each bit count
bitCount := h.bitCounts(list, maxBits)
// And do the assignment
h.assignEncodingAndSize(bitCount, list)
}
type byLiteral []literalNode
func (s *byLiteral) sort(a []literalNode) {
*s = byLiteral(a)
sort.Sort(s)
}
func (s byLiteral) Len() int { return len(s) }
func (s byLiteral) Less(i, j int) bool {
return s[i].literal < s[j].literal
}
func (s byLiteral) Swap(i, j int) { s[i], s[j] = s[j], s[i] }
type byFreq []literalNode
func (s *byFreq) sort(a []literalNode) {
*s = byFreq(a)
sort.Sort(s)
}
func (s byFreq) Len() int { return len(s) }
func (s byFreq) Less(i, j int) bool {
if s[i].freq == s[j].freq {
return s[i].literal < s[j].literal
}
return s[i].freq < s[j].freq
}
func (s byFreq) Swap(i, j int) { s[i], s[j] = s[j], s[i] }
func reverseBits(number uint16, bitLength byte) uint16 {
return bits.Reverse16(number << (16 - bitLength))
}
image/internal/compress/flate/inflate.go
+825
View File
@@ -0,0 +1,825 @@
// Copyright 2009 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.
// Package flate implements the DEFLATE compressed data format, described in
// RFC 1951. The gzip and zlib packages implement access to DEFLATE-based file
// formats.
package flate
import (
"bufio"
"io"
"math/bits"
"strconv"
"sync"
)
const (
maxCodeLen = 16 // max length of Huffman code
// The next three numbers come from the RFC section 3.2.7, with the
// additional proviso in section 3.2.5 which implies that distance codes
// 30 and 31 should never occur in compressed data.
maxNumLit = 286
maxNumDist = 30
numCodes = 19 // number of codes in Huffman meta-code
)
// Initialize the fixedHuffmanDecoder only once upon first use.
var fixedOnce sync.Once
var fixedHuffmanDecoder huffmanDecoder
// A CorruptInputError reports the presence of corrupt input at a given offset.
type CorruptInputError int64
func (e CorruptInputError) Error() string {
return "flate: corrupt input before offset " + strconv.FormatInt(int64(e), 10)
}
// An InternalError reports an error in the flate code itself.
type InternalError string
func (e InternalError) Error() string { return "flate: internal error: " + string(e) }
// A ReadError reports an error encountered while reading input.
//
// Deprecated: No longer returned.
type ReadError struct {
Offset int64 // byte offset where error occurred
Err error // error returned by underlying Read
}
func (e *ReadError) Error() string {
return "flate: read error at offset " + strconv.FormatInt(e.Offset, 10) + ": " + e.Err.Error()
}
// A WriteError reports an error encountered while writing output.
//
// Deprecated: No longer returned.
type WriteError struct {
Offset int64 // byte offset where error occurred
Err error // error returned by underlying Write
}
func (e *WriteError) Error() string {
return "flate: write error at offset " + strconv.FormatInt(e.Offset, 10) + ": " + e.Err.Error()
}
// Resetter resets a ReadCloser returned by NewReader or NewReaderDict
// to switch to a new underlying Reader. This permits reusing a ReadCloser
// instead of allocating a new one.
type Resetter interface {
// Reset discards any buffered data and resets the Resetter as if it was
// newly initialized with the given reader.
Reset(r io.Reader, dict []byte) error
}
// The data structure for decoding Huffman tables is based on that of
// zlib. There is a lookup table of a fixed bit width (huffmanChunkBits),
// For codes smaller than the table width, there are multiple entries
// (each combination of trailing bits has the same value). For codes
// larger than the table width, the table contains a link to an overflow
// table. The width of each entry in the link table is the maximum code
// size minus the chunk width.
//
// Note that you can do a lookup in the table even without all bits
// filled. Since the extra bits are zero, and the DEFLATE Huffman codes
// have the property that shorter codes come before longer ones, the
// bit length estimate in the result is a lower bound on the actual
// number of bits.
//
// See the following:
// https://github.com/madler/zlib/raw/master/doc/algorithm.txt
// chunk & 15 is number of bits
// chunk >> 4 is value, including table link
const (
huffmanChunkBits = 9
huffmanNumChunks = 1 << huffmanChunkBits
huffmanCountMask = 15
huffmanValueShift = 4
)
type huffmanDecoder struct {
min int // the minimum code length
chunks [huffmanNumChunks]uint32 // chunks as described above
links [][]uint32 // overflow links
linkMask uint32 // mask the width of the link table
}
// Initialize Huffman decoding tables from array of code lengths.
// Following this function, h is guaranteed to be initialized into a complete
// tree (i.e., neither over-subscribed nor under-subscribed). The exception is a
// degenerate case where the tree has only a single symbol with length 1. Empty
// trees are permitted.
func (h *huffmanDecoder) init(lengths []int) bool {
// Sanity enables additional runtime tests during Huffman
// table construction. It's intended to be used during
// development to supplement the currently ad-hoc unit tests.
const sanity = false
if h.min != 0 {
*h = huffmanDecoder{}
}
// Count number of codes of each length,
// compute min and max length.
var count [maxCodeLen]int
var min, max int
for _, n := range lengths {
if n == 0 {
continue
}
if min == 0 || n < min {
min = n
}
if n > max {
max = n
}
count[n]++
}
// Empty tree. The decompressor.huffSym function will fail later if the tree
// is used. Technically, an empty tree is only valid for the HDIST tree and
// not the HCLEN and HLIT tree. However, a stream with an empty HCLEN tree
// is guaranteed to fail since it will attempt to use the tree to decode the
// codes for the HLIT and HDIST trees. Similarly, an empty HLIT tree is
// guaranteed to fail later since the compressed data section must be
// composed of at least one symbol (the end-of-block marker).
if max == 0 {
return true
}
code := 0
var nextcode [maxCodeLen]int
for i := min; i <= max; i++ {
code <<= 1
nextcode[i] = code
code += count[i]
}
// Check that the coding is complete (i.e., that we've
// assigned all 2-to-the-max possible bit sequences).
// Exception: To be compatible with zlib, we also need to
// accept degenerate single-code codings. See also
// TestDegenerateHuffmanCoding.
if code != 1<<uint(max) && !(code == 1 && max == 1) {
return false
}
h.min = min
if max > huffmanChunkBits {
numLinks := 1 << (uint(max) - huffmanChunkBits)
h.linkMask = uint32(numLinks - 1)
// create link tables
link := nextcode[huffmanChunkBits+1] >> 1
h.links = make([][]uint32, huffmanNumChunks-link)
for j := uint(link); j < huffmanNumChunks; j++ {
reverse := int(bits.Reverse16(uint16(j)))
reverse >>= uint(16 - huffmanChunkBits)
off := j - uint(link)
if sanity && h.chunks[reverse] != 0 {
panic("impossible: overwriting existing chunk")
}
h.chunks[reverse] = uint32(off<<huffmanValueShift | (huffmanChunkBits + 1))
h.links[off] = make([]uint32, numLinks)
}
}
for i, n := range lengths {
if n == 0 {
continue
}
code := nextcode[n]
nextcode[n]++
chunk := uint32(i<<huffmanValueShift | n)
reverse := int(bits.Reverse16(uint16(code)))
reverse >>= uint(16 - n)
if n <= huffmanChunkBits {
for off := reverse; off < len(h.chunks); off += 1 << uint(n) {
// We should never need to overwrite
// an existing chunk. Also, 0 is
// never a valid chunk, because the
// lower 4 "count" bits should be
// between 1 and 15.
if sanity && h.chunks[off] != 0 {
panic("impossible: overwriting existing chunk")
}
h.chunks[off] = chunk
}
} else {
j := reverse & (huffmanNumChunks - 1)
if sanity && h.chunks[j]&huffmanCountMask != huffmanChunkBits+1 {
// Longer codes should have been
// associated with a link table above.
panic("impossible: not an indirect chunk")
}
value := h.chunks[j] >> huffmanValueShift
linktab := h.links[value]
reverse >>= huffmanChunkBits
for off := reverse; off < len(linktab); off += 1 << uint(n-huffmanChunkBits) {
if sanity && linktab[off] != 0 {
panic("impossible: overwriting existing chunk")
}
linktab[off] = chunk
}
}
}
if sanity {
// Above we've sanity checked that we never overwrote
// an existing entry. Here we additionally check that
// we filled the tables completely.
for i, chunk := range h.chunks {
if chunk == 0 {
// As an exception, in the degenerate
// single-code case, we allow odd
// chunks to be missing.
if code == 1 && i%2 == 1 {
continue
}
panic("impossible: missing chunk")
}
}
for _, linktab := range h.links {
for _, chunk := range linktab {
if chunk == 0 {
panic("impossible: missing chunk")
}
}
}
}
return true
}
// The actual read interface needed by NewReader.
// If the passed in io.Reader does not also have ReadByte,
// the NewReader will introduce its own buffering.
type Reader interface {
io.Reader
io.ByteReader
}
// Decompress state.
type decompressor struct {
// Input source.
r Reader
roffset int64
// Input bits, in top of b.
b uint32
nb uint
// Huffman decoders for literal/length, distance.
h1, h2 huffmanDecoder
// Length arrays used to define Huffman codes.
bits *[maxNumLit + maxNumDist]int
codebits *[numCodes]int
// Output history, buffer.
dict dictDecoder
// Temporary buffer (avoids repeated allocation).
buf [4]byte
// Next step in the decompression,
// and decompression state.
step func(*decompressor)
stepState int
final bool
err error
toRead []byte
hl, hd *huffmanDecoder
copyLen int
copyDist int
}
func (f *decompressor) nextBlock() {
for f.nb < 1+2 {
if f.err = f.moreBits(); f.err != nil {
return
}
}
f.final = f.b&1 == 1
f.b >>= 1
typ := f.b & 3
f.b >>= 2
f.nb -= 1 + 2
switch typ {
case 0:
f.dataBlock()
case 1:
// compressed, fixed Huffman tables
f.hl = &fixedHuffmanDecoder
f.hd = nil
f.huffmanBlock()
case 2:
// compressed, dynamic Huffman tables
if f.err = f.readHuffman(); f.err != nil {
break
}
f.hl = &f.h1
f.hd = &f.h2
f.huffmanBlock()
default:
// 3 is reserved.
f.err = CorruptInputError(f.roffset)
}
}
func (f *decompressor) Read(b []byte) (int, error) {
for {
if len(f.toRead) > 0 {
n := copy(b, f.toRead)
f.toRead = f.toRead[n:]
if len(f.toRead) == 0 {
return n, f.err
}
return n, nil
}
if f.err != nil {
return 0, f.err
}
f.step(f)
if f.err != nil && len(f.toRead) == 0 {
f.toRead = f.dict.readFlush() // Flush what's left in case of error
}
}
}
func (f *decompressor) Close() error {
if f.err == io.EOF {
return nil
}
return f.err
}
// RFC 1951 section 3.2.7.
// Compression with dynamic Huffman codes
var codeOrder = [...]int{16, 17, 18, 0, 8, 7, 9, 6, 10, 5, 11, 4, 12, 3, 13, 2, 14, 1, 15}
func (f *decompressor) readHuffman() error {
// HLIT[5], HDIST[5], HCLEN[4].
for f.nb < 5+5+4 {
if err := f.moreBits(); err != nil {
return err
}
}
nlit := int(f.b&0x1F) + 257
if nlit > maxNumLit {
return CorruptInputError(f.roffset)
}
f.b >>= 5
ndist := int(f.b&0x1F) + 1
if ndist > maxNumDist {
return CorruptInputError(f.roffset)
}
f.b >>= 5
nclen := int(f.b&0xF) + 4
// numCodes is 19, so nclen is always valid.
f.b >>= 4
f.nb -= 5 + 5 + 4
// (HCLEN+4)*3 bits: code lengths in the magic codeOrder order.
for i := 0; i < nclen; i++ {
for f.nb < 3 {
if err := f.moreBits(); err != nil {
return err
}
}
f.codebits[codeOrder[i]] = int(f.b & 0x7)
f.b >>= 3
f.nb -= 3
}
for i := nclen; i < len(codeOrder); i++ {
f.codebits[codeOrder[i]] = 0
}
if !f.h1.init(f.codebits[0:]) {
return CorruptInputError(f.roffset)
}
// HLIT + 257 code lengths, HDIST + 1 code lengths,
// using the code length Huffman code.
for i, n := 0, nlit+ndist; i < n; {
x, err := f.huffSym(&f.h1)
if err != nil {
return err
}
if x < 16 {
// Actual length.
f.bits[i] = x
i++
continue
}
// Repeat previous length or zero.
var rep int
var nb uint
var b int
switch x {
default:
return InternalError("unexpected length code")
case 16:
rep = 3
nb = 2
if i == 0 {
return CorruptInputError(f.roffset)
}
b = f.bits[i-1]
case 17:
rep = 3
nb = 3
b = 0
case 18:
rep = 11
nb = 7
b = 0
}
for f.nb < nb {
if err := f.moreBits(); err != nil {
return err
}
}
rep += int(f.b & uint32(1<<nb-1))
f.b >>= nb
f.nb -= nb
if i+rep > n {
return CorruptInputError(f.roffset)
}
for j := 0; j < rep; j++ {
f.bits[i] = b
i++
}
}
if !f.h1.init(f.bits[0:nlit]) || !f.h2.init(f.bits[nlit:nlit+ndist]) {
return CorruptInputError(f.roffset)
}
// As an optimization, we can initialize the min bits to read at a time
// for the HLIT tree to the length of the EOB marker since we know that
// every block must terminate with one. This preserves the property that
// we never read any extra bytes after the end of the DEFLATE stream.
if f.h1.min < f.bits[endBlockMarker] {
f.h1.min = f.bits[endBlockMarker]
}
return nil
}
// Decode a single Huffman block from f.
// hl and hd are the Huffman states for the lit/length values
// and the distance values, respectively. If hd == nil, using the
// fixed distance encoding associated with fixed Huffman blocks.
func (f *decompressor) huffmanBlock() {
const (
stateInit = iota // Zero value must be stateInit
stateDict
)
switch f.stepState {
case stateInit:
goto readLiteral
case stateDict:
goto copyHistory
}
readLiteral:
// Read literal and/or (length, distance) according to RFC section 3.2.3.
{
v, err := f.huffSym(f.hl)
if err != nil {
f.err = err
return
}
var n uint // number of bits extra
var length int
switch {
case v < 256:
f.dict.writeByte(byte(v))
if f.dict.availWrite() == 0 {
f.toRead = f.dict.readFlush()
f.step = (*decompressor).huffmanBlock
f.stepState = stateInit
return
}
goto readLiteral
case v == 256:
f.finishBlock()
return
// otherwise, reference to older data
case v < 265:
length = v - (257 - 3)
n = 0
case v < 269:
length = v*2 - (265*2 - 11)
n = 1
case v < 273:
length = v*4 - (269*4 - 19)
n = 2
case v < 277:
length = v*8 - (273*8 - 35)
n = 3
case v < 281:
length = v*16 - (277*16 - 67)
n = 4
case v < 285:
length = v*32 - (281*32 - 131)
n = 5
case v < maxNumLit:
length = 258
n = 0
default:
f.err = CorruptInputError(f.roffset)
return
}
if n > 0 {
for f.nb < n {
if err = f.moreBits(); err != nil {
f.err = err
return
}
}
length += int(f.b & uint32(1<<n-1))
f.b >>= n
f.nb -= n
}
var dist int
if f.hd == nil {
for f.nb < 5 {
if err = f.moreBits(); err != nil {
f.err = err
return
}
}
dist = int(bits.Reverse8(uint8(f.b & 0x1F << 3)))
f.b >>= 5
f.nb -= 5
} else {
if dist, err = f.huffSym(f.hd); err != nil {
f.err = err
return
}
}
switch {
case dist < 4:
dist++
case dist < maxNumDist:
nb := uint(dist-2) >> 1
// have 1 bit in bottom of dist, need nb more.
extra := (dist & 1) << nb
for f.nb < nb {
if err = f.moreBits(); err != nil {
f.err = err
return
}
}
extra |= int(f.b & uint32(1<<nb-1))
f.b >>= nb
f.nb -= nb
dist = 1<<(nb+1) + 1 + extra
default:
f.err = CorruptInputError(f.roffset)
return
}
// No check on length; encoding can be prescient.
if dist > f.dict.histSize() {
f.err = CorruptInputError(f.roffset)
return
}
f.copyLen, f.copyDist = length, dist
goto copyHistory
}
copyHistory:
// Perform a backwards copy according to RFC section 3.2.3.
{
cnt := f.dict.tryWriteCopy(f.copyDist, f.copyLen)
if cnt == 0 {
cnt = f.dict.writeCopy(f.copyDist, f.copyLen)
}
f.copyLen -= cnt
if f.dict.availWrite() == 0 || f.copyLen > 0 {
f.toRead = f.dict.readFlush()
f.step = (*decompressor).huffmanBlock // We need to continue this work
f.stepState = stateDict
return
}
goto readLiteral
}
}
// Copy a single uncompressed data block from input to output.
func (f *decompressor) dataBlock() {
// Uncompressed.
// Discard current half-byte.
f.nb = 0
f.b = 0
// Length then ones-complement of length.
nr, err := io.ReadFull(f.r, f.buf[0:4])
f.roffset += int64(nr)
if err != nil {
f.err = noEOF(err)
return
}
n := int(f.buf[0]) | int(f.buf[1])<<8
nn := int(f.buf[2]) | int(f.buf[3])<<8
if uint16(nn) != uint16(^n) {
f.err = CorruptInputError(f.roffset)
return
}
if n == 0 {
f.toRead = f.dict.readFlush()
f.finishBlock()
return
}
f.copyLen = n
f.copyData()
}
// copyData copies f.copyLen bytes from the underlying reader into f.hist.
// It pauses for reads when f.hist is full.
func (f *decompressor) copyData() {
buf := f.dict.writeSlice()
if len(buf) > f.copyLen {
buf = buf[:f.copyLen]
}
cnt, err := io.ReadFull(f.r, buf)
f.roffset += int64(cnt)
f.copyLen -= cnt
f.dict.writeMark(cnt)
if err != nil {
f.err = noEOF(err)
return
}
if f.dict.availWrite() == 0 || f.copyLen > 0 {
f.toRead = f.dict.readFlush()
f.step = (*decompressor).copyData
return
}
f.finishBlock()
}
func (f *decompressor) finishBlock() {
if f.final {
if f.dict.availRead() > 0 {
f.toRead = f.dict.readFlush()
}
f.err = io.EOF
}
f.step = (*decompressor).nextBlock
}
// noEOF returns err, unless err == io.EOF, in which case it returns io.ErrUnexpectedEOF.
func noEOF(e error) error {
if e == io.EOF {
return io.ErrUnexpectedEOF
}
return e
}
func (f *decompressor) moreBits() error {
c, err := f.r.ReadByte()
if err != nil {
return noEOF(err)
}
f.roffset++
f.b |= uint32(c) << f.nb
f.nb += 8
return nil
}
// Read the next Huffman-encoded symbol from f according to h.
func (f *decompressor) huffSym(h *huffmanDecoder) (int, error) {
// Since a huffmanDecoder can be empty or be composed of a degenerate tree
// with single element, huffSym must error on these two edge cases. In both
// cases, the chunks slice will be 0 for the invalid sequence, leading it
// satisfy the n == 0 check below.
n := uint(h.min)
// Optimization. Compiler isn't smart enough to keep f.b,f.nb in registers,
// but is smart enough to keep local variables in registers, so use nb and b,
// inline call to moreBits and reassign b,nb back to f on return.
nb, b := f.nb, f.b
for {
for nb < n {
c, err := f.r.ReadByte()
if err != nil {
f.b = b
f.nb = nb
return 0, noEOF(err)
}
f.roffset++
b |= uint32(c) << (nb & 31)
nb += 8
}
chunk := h.chunks[b&(huffmanNumChunks-1)]
n = uint(chunk & huffmanCountMask)
if n > huffmanChunkBits {
chunk = h.links[chunk>>huffmanValueShift][(b>>huffmanChunkBits)&h.linkMask]
n = uint(chunk & huffmanCountMask)
}
if n <= nb {
if n == 0 {
f.b = b
f.nb = nb
f.err = CorruptInputError(f.roffset)
return 0, f.err
}
f.b = b >> (n & 31)
f.nb = nb - n
return int(chunk >> huffmanValueShift), nil
}
}
}
func makeReader(r io.Reader) Reader {
if rr, ok := r.(Reader); ok {
return rr
}
return bufio.NewReader(r)
}
func fixedHuffmanDecoderInit() {
fixedOnce.Do(func() {
// These come from the RFC section 3.2.6.
var bits [288]int
for i := 0; i < 144; i++ {
bits[i] = 8
}
for i := 144; i < 256; i++ {
bits[i] = 9
}
for i := 256; i < 280; i++ {
bits[i] = 7
}
for i := 280; i < 288; i++ {
bits[i] = 8
}
fixedHuffmanDecoder.init(bits[:])
})
}
func (f *decompressor) Reset(r io.Reader, dict []byte) error {
*f = decompressor{
r: makeReader(r),
bits: f.bits,
codebits: f.codebits,
dict: f.dict,
step: (*decompressor).nextBlock,
}
f.dict.init(maxMatchOffset, dict)
return nil
}
// NewReader returns a new ReadCloser that can be used
// to read the uncompressed version of r.
// If r does not also implement io.ByteReader,
// the decompressor may read more data than necessary from r.
// It is the caller's responsibility to call Close on the ReadCloser
// when finished reading.
//
// The ReadCloser returned by NewReader also implements Resetter.
func NewReader(r io.Reader) io.ReadCloser {
fixedHuffmanDecoderInit()
var f decompressor
f.r = makeReader(r)
f.bits = new([maxNumLit + maxNumDist]int)
f.codebits = new([numCodes]int)
f.step = (*decompressor).nextBlock
f.dict.init(maxMatchOffset, nil)
return &f
}
// NewReaderDict is like NewReader but initializes the reader
// with a preset dictionary. The returned Reader behaves as if
// the uncompressed data stream started with the given dictionary,
// which has already been read. NewReaderDict is typically used
// to read data compressed by NewWriterDict.
//
// The ReadCloser returned by NewReader also implements Resetter.
func NewReaderDict(r io.Reader, dict []byte) io.ReadCloser {
fixedHuffmanDecoderInit()
var f decompressor
f.r = makeReader(r)
f.bits = new([maxNumLit + maxNumDist]int)
f.codebits = new([numCodes]int)
f.step = (*decompressor).nextBlock
f.dict.init(maxMatchOffset, dict)
return &f
}
image/internal/compress/flate/inflate_test.go
+97
View File
@@ -0,0 +1,97 @@
// Copyright 2014 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.
package flate
import (
"bytes"
"io"
"strings"
"testing"
)
func TestReset(t *testing.T) {
ss := []string{
"lorem ipsum izzle fo rizzle",
"the quick brown fox jumped over",
}
deflated := make([]bytes.Buffer, 2)
for i, s := range ss {
w, _ := NewWriter(&deflated[i], 1)
w.Write([]byte(s))
w.Close()
}
inflated := make([]bytes.Buffer, 2)
f := NewReader(&deflated[0])
io.Copy(&inflated[0], f)
f.(Resetter).Reset(&deflated[1], nil)
io.Copy(&inflated[1], f)
f.Close()
for i, s := range ss {
if s != inflated[i].String() {
t.Errorf("inflated[%d]:\ngot %q\nwant %q", i, inflated[i], s)
}
}
}
func TestReaderTruncated(t *testing.T) {
vectors := []struct{ input, output string }{
{"\x00", ""},
{"\x00\f", ""},
{"\x00\f\x00", ""},
{"\x00\f\x00\xf3\xff", ""},
{"\x00\f\x00\xf3\xffhello", "hello"},
{"\x00\f\x00\xf3\xffhello, world", "hello, world"},
{"\x02", ""},
{"\xf2H\xcd", "He"},
{"\xf2H͙0a\u0084\t", "Hel\x90\x90\x90\x90\x90"},
{"\xf2H͙0a\u0084\t\x00", "Hel\x90\x90\x90\x90\x90"},
}
for i, v := range vectors {
r := strings.NewReader(v.input)
zr := NewReader(r)
b, err := io.ReadAll(zr)
if err != io.ErrUnexpectedEOF {
t.Errorf("test %d, error mismatch: got %v, want io.ErrUnexpectedEOF", i, err)
}
if string(b) != v.output {
t.Errorf("test %d, output mismatch: got %q, want %q", i, b, v.output)
}
}
}
func TestResetDict(t *testing.T) {
dict := []byte("the lorem fox")
ss := []string{
"lorem ipsum izzle fo rizzle",
"the quick brown fox jumped over",
}
deflated := make([]bytes.Buffer, len(ss))
for i, s := range ss {
w, _ := NewWriterDict(&deflated[i], DefaultCompression, dict)
w.Write([]byte(s))
w.Close()
}
inflated := make([]bytes.Buffer, len(ss))
f := NewReader(nil)
for i := range inflated {
f.(Resetter).Reset(&deflated[i], dict)
io.Copy(&inflated[i], f)
}
f.Close()
for i, s := range ss {
if s != inflated[i].String() {
t.Errorf("inflated[%d]:\ngot %q\nwant %q", i, inflated[i], s)
}
}
}
image/internal/compress/flate/reader_test.go
+98
View File
@@ -0,0 +1,98 @@
// Copyright 2012 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.
package flate
import (
"bytes"
"io"
"os"
"runtime"
"strings"
"testing"
)
func TestNlitOutOfRange(t *testing.T) {
// Trying to decode this bogus flate data, which has a Huffman table
// with nlit=288, should not panic.
io.Copy(io.Discard, NewReader(strings.NewReader(
"\xfc\xfe\x36\xe7\x5e\x1c\xef\xb3\x55\x58\x77\xb6\x56\xb5\x43\xf4"+
"\x6f\xf2\xd2\xe6\x3d\x99\xa0\x85\x8c\x48\xeb\xf8\xda\x83\x04\x2a"+
"\x75\xc4\xf8\x0f\x12\x11\xb9\xb4\x4b\x09\xa0\xbe\x8b\x91\x4c")))
}
var suites = []struct{ name, file string }{
// Digits is the digits of the irrational number e. Its decimal representation
// does not repeat, but there are only 10 possible digits, so it should be
// reasonably compressible.
{"Digits", "../testdata/e.txt"},
// Newton is Isaac Newtons's educational text on Opticks.
{"Newton", "../../testdata/Isaac.Newton-Opticks.txt"},
}
func BenchmarkDecode(b *testing.B) {
doBench(b, func(b *testing.B, buf0 []byte, level, n int) {
b.ReportAllocs()
b.StopTimer()
b.SetBytes(int64(n))
compressed := new(bytes.Buffer)
w, err := NewWriter(compressed, level)
if err != nil {
b.Fatal(err)
}
for i := 0; i < n; i += len(buf0) {
if len(buf0) > n-i {
buf0 = buf0[:n-i]
}
io.Copy(w, bytes.NewReader(buf0))
}
w.Close()
buf1 := compressed.Bytes()
buf0, compressed, w = nil, nil, nil
runtime.GC()
b.StartTimer()
for i := 0; i < b.N; i++ {
io.Copy(io.Discard, NewReader(bytes.NewReader(buf1)))
}
})
}
var levelTests = []struct {
name string
level int
}{
{"Huffman", HuffmanOnly},
{"Speed", BestSpeed},
{"Default", DefaultCompression},
{"Compression", BestCompression},
}
var sizes = []struct {
name string
n int
}{
{"1e4", 1e4},
{"1e5", 1e5},
{"1e6", 1e6},
}
func doBench(b *testing.B, f func(b *testing.B, buf []byte, level, n int)) {
for _, suite := range suites {
buf, err := os.ReadFile(suite.file)
if err != nil {
b.Fatal(err)
}
if len(buf) == 0 {
b.Fatalf("test file %q has no data", suite.file)
}
for _, l := range levelTests {
for _, s := range sizes {
b.Run(suite.name+"/"+l.name+"/"+s.name, func(b *testing.B) {
f(b, buf, l.level, s.n)
})
}
}
}
}
image/internal/compress/flate/token.go
+97
View File
@@ -0,0 +1,97 @@
// Copyright 2009 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.
package flate
const (
// 2 bits: type 0 = literal 1=EOF 2=Match 3=Unused
// 8 bits: xlength = length - MIN_MATCH_LENGTH
// 22 bits xoffset = offset - MIN_OFFSET_SIZE, or literal
lengthShift = 22
offsetMask = 1<<lengthShift - 1
typeMask = 3 << 30
literalType = 0 << 30
matchType = 1 << 30
)
// The length code for length X (MIN_MATCH_LENGTH <= X <= MAX_MATCH_LENGTH)
// is lengthCodes[length - MIN_MATCH_LENGTH]
var lengthCodes = [...]uint32{
0, 1, 2, 3, 4, 5, 6, 7, 8, 8,
9, 9, 10, 10, 11, 11, 12, 12, 12, 12,
13, 13, 13, 13, 14, 14, 14, 14, 15, 15,
15, 15, 16, 16, 16, 16, 16, 16, 16, 16,
17, 17, 17, 17, 17, 17, 17, 17, 18, 18,
18, 18, 18, 18, 18, 18, 19, 19, 19, 19,
19, 19, 19, 19, 20, 20, 20, 20, 20, 20,
20, 20, 20, 20, 20, 20, 20, 20, 20, 20,
21, 21, 21, 21, 21, 21, 21, 21, 21, 21,
21, 21, 21, 21, 21, 21, 22, 22, 22, 22,
22, 22, 22, 22, 22, 22, 22, 22, 22, 22,
22, 22, 23, 23, 23, 23, 23, 23, 23, 23,
23, 23, 23, 23, 23, 23, 23, 23, 24, 24,
24, 24, 24, 24, 24, 24, 24, 24, 24, 24,
24, 24, 24, 24, 24, 24, 24, 24, 24, 24,
24, 24, 24, 24, 24, 24, 24, 24, 24, 24,
25, 25, 25, 25, 25, 25, 25, 25, 25, 25,
25, 25, 25, 25, 25, 25, 25, 25, 25, 25,
25, 25, 25, 25, 25, 25, 25, 25, 25, 25,
25, 25, 26, 26, 26, 26, 26, 26, 26, 26,
26, 26, 26, 26, 26, 26, 26, 26, 26, 26,
26, 26, 26, 26, 26, 26, 26, 26, 26, 26,
26, 26, 26, 26, 27, 27, 27, 27, 27, 27,
27, 27, 27, 27, 27, 27, 27, 27, 27, 27,
27, 27, 27, 27, 27, 27, 27, 27, 27, 27,
27, 27, 27, 27, 27, 28,
}
var offsetCodes = [...]uint32{
0, 1, 2, 3, 4, 4, 5, 5, 6, 6, 6, 6, 7, 7, 7, 7,
8, 8, 8, 8, 8, 8, 8, 8, 9, 9, 9, 9, 9, 9, 9, 9,
10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10,
11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11,
12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12,
12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12,
13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13,
13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13,
14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14,
14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14,
14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14,
14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14,
15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15,
15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15,
15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15,
15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15,
}
type token uint32
// Convert a literal into a literal token.
func literalToken(literal uint32) token { return token(literalType + literal) }
// Convert a < xlength, xoffset > pair into a match token.
func matchToken(xlength uint32, xoffset uint32) token {
return token(matchType + xlength<<lengthShift + xoffset)
}
// Returns the literal of a literal token
func (t token) literal() uint32 { return uint32(t - literalType) }
// Returns the extra offset of a match token
func (t token) offset() uint32 { return uint32(t) & offsetMask }
func (t token) length() uint32 { return uint32((t - matchType) >> lengthShift) }
func lengthCode(len uint32) uint32 { return lengthCodes[len] }
// Returns the offset code corresponding to a specific offset
func offsetCode(off uint32) uint32 {
if off < uint32(len(offsetCodes)) {
return offsetCodes[off]
}
if off>>7 < uint32(len(offsetCodes)) {
return offsetCodes[off>>7] + 14
}
return offsetCodes[off>>14] + 28
}
image/internal/compress/flate/writer_test.go
+237
View File
@@ -0,0 +1,237 @@
// Copyright 2012 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.
package flate
import (
"bytes"
"fmt"
"io"
"math/rand"
"runtime"
"testing"
)
func BenchmarkEncode(b *testing.B) {
doBench(b, func(b *testing.B, buf0 []byte, level, n int) {
b.StopTimer()
b.SetBytes(int64(n))
buf1 := make([]byte, n)
for i := 0; i < n; i += len(buf0) {
if len(buf0) > n-i {
buf0 = buf0[:n-i]
}
copy(buf1[i:], buf0)
}
buf0 = nil
w, err := NewWriter(io.Discard, level)
if err != nil {
b.Fatal(err)
}
runtime.GC()
b.StartTimer()
for i := 0; i < b.N; i++ {
w.Reset(io.Discard)
w.Write(buf1)
w.Close()
}
})
}
// errorWriter is a writer that fails after N writes.
type errorWriter struct {
N int
}
func (e *errorWriter) Write(b []byte) (int, error) {
if e.N <= 0 {
return 0, io.ErrClosedPipe
}
e.N--
return len(b), nil
}
// Test if errors from the underlying writer is passed upwards.
func TestWriteError(t *testing.T) {
t.Parallel()
buf := new(bytes.Buffer)
n := 65536
if !testing.Short() {
n *= 4
}
for i := 0; i < n; i++ {
fmt.Fprintf(buf, "asdasfasf%d%dfghfgujyut%dyutyu\n", i, i, i)
}
in := buf.Bytes()
// We create our own buffer to control number of writes.
copyBuffer := make([]byte, 128)
for l := 0; l < 10; l++ {
for fail := 1; fail <= 256; fail *= 2 {
// Fail after 'fail' writes
ew := &errorWriter{N: fail}
w, err := NewWriter(ew, l)
if err != nil {
t.Fatalf("NewWriter: level %d: %v", l, err)
}
n, err := io.CopyBuffer(w, struct{ io.Reader }{bytes.NewBuffer(in)}, copyBuffer)
if err == nil {
t.Fatalf("Level %d: Expected an error, writer was %#v", l, ew)
}
n2, err := w.Write([]byte{1, 2, 2, 3, 4, 5})
if n2 != 0 {
t.Fatal("Level", l, "Expected 0 length write, got", n)
}
if err == nil {
t.Fatal("Level", l, "Expected an error")
}
err = w.Flush()
if err == nil {
t.Fatal("Level", l, "Expected an error on flush")
}
err = w.Close()
if err == nil {
t.Fatal("Level", l, "Expected an error on close")
}
w.Reset(io.Discard)
n2, err = w.Write([]byte{1, 2, 3, 4, 5, 6})
if err != nil {
t.Fatal("Level", l, "Got unexpected error after reset:", err)
}
if n2 == 0 {
t.Fatal("Level", l, "Got 0 length write, expected > 0")
}
if testing.Short() {
return
}
}
}
}
// Test if two runs produce identical results
// even when writing different sizes to the Writer.
func TestDeterministic(t *testing.T) {
t.Parallel()
for i := 0; i <= 9; i++ {
t.Run(fmt.Sprint("L", i), func(t *testing.T) { testDeterministic(i, t) })
}
t.Run("LM2", func(t *testing.T) { testDeterministic(-2, t) })
}
func testDeterministic(i int, t *testing.T) {
t.Parallel()
// Test so much we cross a good number of block boundaries.
var length = maxStoreBlockSize*30 + 500
if testing.Short() {
length /= 10
}
// Create a random, but compressible stream.
rng := rand.New(rand.NewSource(1))
t1 := make([]byte, length)
for i := range t1 {
t1[i] = byte(rng.Int63() & 7)
}
// Do our first encode.
var b1 bytes.Buffer
br := bytes.NewBuffer(t1)
w, err := NewWriter(&b1, i)
if err != nil {
t.Fatal(err)
}
// Use a very small prime sized buffer.
cbuf := make([]byte, 787)
_, err = io.CopyBuffer(w, struct{ io.Reader }{br}, cbuf)
if err != nil {
t.Fatal(err)
}
w.Close()
// We choose a different buffer size,
// bigger than a maximum block, and also a prime.
var b2 bytes.Buffer
cbuf = make([]byte, 81761)
br2 := bytes.NewBuffer(t1)
w2, err := NewWriter(&b2, i)
if err != nil {
t.Fatal(err)
}
_, err = io.CopyBuffer(w2, struct{ io.Reader }{br2}, cbuf)
if err != nil {
t.Fatal(err)
}
w2.Close()
b1b := b1.Bytes()
b2b := b2.Bytes()
if !bytes.Equal(b1b, b2b) {
t.Errorf("level %d did not produce deterministic result, result mismatch, len(a) = %d, len(b) = %d", i, len(b1b), len(b2b))
}
}
// TestDeflateFast_Reset will test that encoding is consistent
// across a warparound of the table offset.
// See https://github.com/golang/go/issues/34121
func TestDeflateFast_Reset(t *testing.T) {
buf := new(bytes.Buffer)
n := 65536
for i := 0; i < n; i++ {
fmt.Fprintf(buf, "asdfasdfasdfasdf%d%dfghfgujyut%dyutyu\n", i, i, i)
}
// This is specific to level 1.
const level = 1
in := buf.Bytes()
offset := 1
if testing.Short() {
offset = 256
}
// We do an encode with a clean buffer to compare.
var want bytes.Buffer
w, err := NewWriter(&want, level)
if err != nil {
t.Fatalf("NewWriter: level %d: %v", level, err)
}
// Output written 3 times.
w.Write(in)
w.Write(in)
w.Write(in)
w.Close()
for ; offset <= 256; offset *= 2 {
w, err := NewWriter(io.Discard, level)
if err != nil {
t.Fatalf("NewWriter: level %d: %v", level, err)
}
// Reset until we are right before the wraparound.
// Each reset adds maxMatchOffset to the offset.
for i := 0; i < (bufferReset-len(in)-offset-maxMatchOffset)/maxMatchOffset; i++ {
// skip ahead to where we are close to wrap around...
w.d.reset(nil)
}
var got bytes.Buffer
w.Reset(&got)
// Write 3 times, close.
for i := 0; i < 3; i++ {
_, err = w.Write(in)
if err != nil {
t.Fatal(err)
}
}
err = w.Close()
if err != nil {
t.Fatal(err)
}
if !bytes.Equal(got.Bytes(), want.Bytes()) {
t.Fatalf("output did not match at wraparound, len(want) = %d, len(got) = %d", want.Len(), got.Len())
}
}
}