2016-11-03 17:16:01 -05:00
|
|
|
// Copyright 2009 The Go Authors. All rights reserved.
|
|
|
|
// Copyright (c) 2015 Klaus Post
|
|
|
|
// 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
|
|
|
|
ConstantCompression = HuffmanOnly // compatibility alias.
|
|
|
|
|
|
|
|
logWindowSize = 15
|
|
|
|
windowSize = 1 << logWindowSize
|
|
|
|
windowMask = windowSize - 1
|
|
|
|
logMaxOffsetSize = 15 // Standard DEFLATE
|
|
|
|
minMatchLength = 4 // The smallest match that the compressor looks for
|
|
|
|
maxMatchLength = 258 // The longest match for the compressor
|
|
|
|
minOffsetSize = 1 // The shortest offset that makes any sense
|
|
|
|
|
|
|
|
// The maximum number of tokens we put into a single flat block, just too
|
|
|
|
// stop things from getting too large.
|
|
|
|
maxFlateBlockTokens = 1 << 14
|
|
|
|
maxStoreBlockSize = 65535
|
|
|
|
hashBits = 17 // After 17 performance degrades
|
|
|
|
hashSize = 1 << hashBits
|
|
|
|
hashMask = (1 << hashBits) - 1
|
|
|
|
hashShift = (hashBits + minMatchLength - 1) / minMatchLength
|
|
|
|
maxHashOffset = 1 << 24
|
|
|
|
|
|
|
|
skipNever = math.MaxInt32
|
|
|
|
)
|
|
|
|
|
|
|
|
type compressionLevel struct {
|
|
|
|
good, lazy, nice, chain, fastSkipHashing, level int
|
|
|
|
}
|
|
|
|
|
|
|
|
// Compression levels have been rebalanced from zlib deflate defaults
|
|
|
|
// to give a bigger spread in speed and compression.
|
|
|
|
// See https://blog.klauspost.com/rebalancing-deflate-compression-levels/
|
|
|
|
var levels = []compressionLevel{
|
|
|
|
{}, // 0
|
|
|
|
// Level 1-4 uses specialized algorithm - values not used
|
|
|
|
{0, 0, 0, 0, 0, 1},
|
|
|
|
{0, 0, 0, 0, 0, 2},
|
|
|
|
{0, 0, 0, 0, 0, 3},
|
|
|
|
{0, 0, 0, 0, 0, 4},
|
|
|
|
// For levels 5-6 we don't bother trying with lazy matches.
|
|
|
|
// Lazy matching is at least 30% slower, with 1.5% increase.
|
|
|
|
{6, 0, 12, 8, 12, 5},
|
|
|
|
{8, 0, 24, 16, 16, 6},
|
|
|
|
// Levels 7-9 use increasingly more lazy matching
|
|
|
|
// and increasingly stringent conditions for "good enough".
|
|
|
|
{8, 8, 24, 16, skipNever, 7},
|
|
|
|
{10, 16, 24, 64, skipNever, 8},
|
|
|
|
{32, 258, 258, 4096, skipNever, 9},
|
|
|
|
}
|
|
|
|
|
2019-11-18 00:18:33 -05:00
|
|
|
// advancedState contains state for the advanced levels, with bigger hash tables, etc.
|
|
|
|
type advancedState struct {
|
|
|
|
// deflate state
|
|
|
|
length int
|
|
|
|
offset int
|
|
|
|
hash uint32
|
|
|
|
maxInsertIndex int
|
|
|
|
ii uint16 // position of last match, intended to overflow to reset.
|
2016-11-03 17:16:01 -05:00
|
|
|
|
|
|
|
// 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]
|
2019-11-18 00:18:33 -05:00
|
|
|
index int
|
|
|
|
hashMatch [maxMatchLength + minMatchLength]uint32
|
|
|
|
}
|
|
|
|
|
|
|
|
type compressor struct {
|
|
|
|
compressionLevel
|
|
|
|
|
|
|
|
w *huffmanBitWriter
|
|
|
|
|
|
|
|
// compression algorithm
|
|
|
|
fill func(*compressor, []byte) int // copy data to window
|
|
|
|
step func(*compressor) // process window
|
|
|
|
sync bool // requesting flush
|
|
|
|
|
2016-11-03 17:16:01 -05:00
|
|
|
window []byte
|
|
|
|
windowEnd int
|
|
|
|
blockStart int // window index where current tokens start
|
|
|
|
byteAvailable bool // if true, still need to process window[index-1].
|
2019-11-18 00:18:33 -05:00
|
|
|
err error
|
2016-11-03 17:16:01 -05:00
|
|
|
|
|
|
|
// queued output tokens
|
|
|
|
tokens tokens
|
2019-11-18 00:18:33 -05:00
|
|
|
fast fastEnc
|
|
|
|
state *advancedState
|
2016-11-03 17:16:01 -05:00
|
|
|
}
|
|
|
|
|
|
|
|
func (d *compressor) fillDeflate(b []byte) int {
|
2019-11-18 00:18:33 -05:00
|
|
|
s := d.state
|
|
|
|
if s.index >= 2*windowSize-(minMatchLength+maxMatchLength) {
|
2016-11-03 17:16:01 -05:00
|
|
|
// shift the window by windowSize
|
|
|
|
copy(d.window[:], d.window[windowSize:2*windowSize])
|
2019-11-18 00:18:33 -05:00
|
|
|
s.index -= windowSize
|
2016-11-03 17:16:01 -05:00
|
|
|
d.windowEnd -= windowSize
|
|
|
|
if d.blockStart >= windowSize {
|
|
|
|
d.blockStart -= windowSize
|
|
|
|
} else {
|
|
|
|
d.blockStart = math.MaxInt32
|
|
|
|
}
|
2019-11-18 00:18:33 -05:00
|
|
|
s.hashOffset += windowSize
|
|
|
|
if s.hashOffset > maxHashOffset {
|
|
|
|
delta := s.hashOffset - 1
|
|
|
|
s.hashOffset -= delta
|
|
|
|
s.chainHead -= delta
|
|
|
|
// Iterate over slices instead of arrays to avoid copying
|
|
|
|
// the entire table onto the stack (Issue #18625).
|
|
|
|
for i, v := range s.hashPrev[:] {
|
2016-11-03 17:16:01 -05:00
|
|
|
if int(v) > delta {
|
2019-11-18 00:18:33 -05:00
|
|
|
s.hashPrev[i] = uint32(int(v) - delta)
|
2016-11-03 17:16:01 -05:00
|
|
|
} else {
|
2019-11-18 00:18:33 -05:00
|
|
|
s.hashPrev[i] = 0
|
2016-11-03 17:16:01 -05:00
|
|
|
}
|
|
|
|
}
|
2019-11-18 00:18:33 -05:00
|
|
|
for i, v := range s.hashHead[:] {
|
2016-11-03 17:16:01 -05:00
|
|
|
if int(v) > delta {
|
2019-11-18 00:18:33 -05:00
|
|
|
s.hashHead[i] = uint32(int(v) - delta)
|
2016-11-03 17:16:01 -05:00
|
|
|
} else {
|
2019-11-18 00:18:33 -05:00
|
|
|
s.hashHead[i] = 0
|
2016-11-03 17:16:01 -05:00
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
n := copy(d.window[d.windowEnd:], b)
|
|
|
|
d.windowEnd += n
|
|
|
|
return n
|
|
|
|
}
|
|
|
|
|
2019-11-18 00:18:33 -05:00
|
|
|
func (d *compressor) writeBlock(tok *tokens, index int, eof bool) error {
|
2016-11-03 17:16:01 -05:00
|
|
|
if index > 0 || eof {
|
|
|
|
var window []byte
|
|
|
|
if d.blockStart <= index {
|
|
|
|
window = d.window[d.blockStart:index]
|
|
|
|
}
|
|
|
|
d.blockStart = index
|
2019-11-18 00:18:33 -05:00
|
|
|
d.w.writeBlock(tok, eof, window)
|
2016-11-03 17:16:01 -05:00
|
|
|
return d.w.err
|
|
|
|
}
|
|
|
|
return nil
|
|
|
|
}
|
|
|
|
|
|
|
|
// writeBlockSkip writes the current block and uses the number of tokens
|
|
|
|
// to determine if the block should be stored on no matches, or
|
|
|
|
// only huffman encoded.
|
2019-11-18 00:18:33 -05:00
|
|
|
func (d *compressor) writeBlockSkip(tok *tokens, index int, eof bool) error {
|
2016-11-03 17:16:01 -05:00
|
|
|
if index > 0 || eof {
|
|
|
|
if d.blockStart <= index {
|
|
|
|
window := d.window[d.blockStart:index]
|
|
|
|
// If we removed less than a 64th of all literals
|
|
|
|
// we huffman compress the block.
|
|
|
|
if int(tok.n) > len(window)-int(tok.n>>6) {
|
2019-11-18 00:18:33 -05:00
|
|
|
d.w.writeBlockHuff(eof, window, d.sync)
|
2016-11-03 17:16:01 -05:00
|
|
|
} else {
|
|
|
|
// Write a dynamic huffman block.
|
2019-11-18 00:18:33 -05:00
|
|
|
d.w.writeBlockDynamic(tok, eof, window, d.sync)
|
2016-11-03 17:16:01 -05:00
|
|
|
}
|
|
|
|
} else {
|
2019-11-18 00:18:33 -05:00
|
|
|
d.w.writeBlock(tok, eof, nil)
|
2016-11-03 17:16:01 -05:00
|
|
|
}
|
|
|
|
d.blockStart = index
|
|
|
|
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 start/reset.
|
|
|
|
func (d *compressor) fillWindow(b []byte) {
|
|
|
|
// Do not fill window if we are in store-only mode,
|
|
|
|
// use constant or Snappy compression.
|
2019-11-18 00:18:33 -05:00
|
|
|
if d.level == 0 {
|
|
|
|
return
|
|
|
|
}
|
|
|
|
if d.fast != nil {
|
|
|
|
// encode the last data, but discard the result
|
|
|
|
if len(b) > maxMatchOffset {
|
|
|
|
b = b[len(b)-maxMatchOffset:]
|
|
|
|
}
|
|
|
|
d.fast.Encode(&d.tokens, b)
|
|
|
|
d.tokens.Reset()
|
2016-11-03 17:16:01 -05:00
|
|
|
return
|
|
|
|
}
|
2019-11-18 00:18:33 -05:00
|
|
|
s := d.state
|
2016-11-03 17:16:01 -05:00
|
|
|
// 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[d.windowEnd:], b)
|
|
|
|
|
|
|
|
// Calculate 256 hashes at the time (more L1 cache hits)
|
|
|
|
loops := (n + 256 - minMatchLength) / 256
|
|
|
|
for j := 0; j < loops; j++ {
|
|
|
|
startindex := j * 256
|
|
|
|
end := startindex + 256 + minMatchLength - 1
|
|
|
|
if end > n {
|
|
|
|
end = n
|
|
|
|
}
|
|
|
|
tocheck := d.window[startindex:end]
|
|
|
|
dstSize := len(tocheck) - minMatchLength + 1
|
|
|
|
|
|
|
|
if dstSize <= 0 {
|
|
|
|
continue
|
|
|
|
}
|
|
|
|
|
2019-11-18 00:18:33 -05:00
|
|
|
dst := s.hashMatch[:dstSize]
|
|
|
|
bulkHash4(tocheck, dst)
|
2016-11-03 17:16:01 -05:00
|
|
|
var newH uint32
|
|
|
|
for i, val := range dst {
|
|
|
|
di := i + startindex
|
|
|
|
newH = val & hashMask
|
|
|
|
// Get previous value with the same hash.
|
|
|
|
// Our chain should point to the previous value.
|
2019-11-18 00:18:33 -05:00
|
|
|
s.hashPrev[di&windowMask] = s.hashHead[newH]
|
2016-11-03 17:16:01 -05:00
|
|
|
// Set the head of the hash chain to us.
|
2019-11-18 00:18:33 -05:00
|
|
|
s.hashHead[newH] = uint32(di + s.hashOffset)
|
2016-11-03 17:16:01 -05:00
|
|
|
}
|
2019-11-18 00:18:33 -05:00
|
|
|
s.hash = newH
|
2016-11-03 17:16:01 -05:00
|
|
|
}
|
|
|
|
// Update window information.
|
|
|
|
d.windowEnd += n
|
2019-11-18 00:18:33 -05:00
|
|
|
s.index = n
|
2016-11-03 17:16:01 -05:00
|
|
|
}
|
|
|
|
|
|
|
|
// Try to find a match starting at index whose length is greater than prevSize.
|
|
|
|
// We only look at chainCount possibilities before giving up.
|
2019-11-18 00:18:33 -05:00
|
|
|
// pos = s.index, prevHead = s.chainHead-s.hashOffset, prevLength=minMatchLength-1, lookahead
|
2016-11-03 17:16:01 -05:00
|
|
|
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] {
|
2019-11-18 00:18:33 -05:00
|
|
|
n := matchLen(win[i:i+minMatchLook], wPos)
|
2016-11-03 17:16:01 -05:00
|
|
|
|
|
|
|
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
|
|
|
|
}
|
2019-11-18 00:18:33 -05:00
|
|
|
i = int(d.state.hashPrev[i&windowMask]) - d.state.hashOffset
|
2016-11-03 17:16:01 -05:00
|
|
|
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
|
|
|
|
}
|
|
|
|
|
|
|
|
// 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 {
|
2019-11-18 00:18:33 -05:00
|
|
|
b = b[:4]
|
|
|
|
return hash4u(uint32(b[3])|uint32(b[2])<<8|uint32(b[1])<<16|uint32(b[0])<<24, hashBits)
|
2016-11-03 17:16:01 -05:00
|
|
|
}
|
|
|
|
|
|
|
|
// bulkHash4 will compute hashes using the same
|
|
|
|
// algorithm as hash4
|
|
|
|
func bulkHash4(b []byte, dst []uint32) {
|
2019-11-18 00:18:33 -05:00
|
|
|
if len(b) < 4 {
|
2016-11-03 17:16:01 -05:00
|
|
|
return
|
|
|
|
}
|
|
|
|
hb := uint32(b[3]) | uint32(b[2])<<8 | uint32(b[1])<<16 | uint32(b[0])<<24
|
2019-11-18 00:18:33 -05:00
|
|
|
dst[0] = hash4u(hb, hashBits)
|
|
|
|
end := len(b) - 4 + 1
|
2016-11-03 17:16:01 -05:00
|
|
|
for i := 1; i < end; i++ {
|
|
|
|
hb = (hb << 8) | uint32(b[i+3])
|
2019-11-18 00:18:33 -05:00
|
|
|
dst[i] = hash4u(hb, hashBits)
|
2016-11-03 17:16:01 -05:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
func (d *compressor) initDeflate() {
|
|
|
|
d.window = make([]byte, 2*windowSize)
|
|
|
|
d.byteAvailable = false
|
2019-11-18 00:18:33 -05:00
|
|
|
d.err = nil
|
|
|
|
if d.state == nil {
|
2016-11-03 17:16:01 -05:00
|
|
|
return
|
|
|
|
}
|
2019-11-18 00:18:33 -05:00
|
|
|
s := d.state
|
|
|
|
s.index = 0
|
|
|
|
s.hashOffset = 1
|
|
|
|
s.length = minMatchLength - 1
|
|
|
|
s.offset = 0
|
|
|
|
s.hash = 0
|
|
|
|
s.chainHead = -1
|
2016-11-03 17:16:01 -05:00
|
|
|
}
|
|
|
|
|
|
|
|
// deflateLazy is the same as deflate, but with d.fastSkipHashing == skipNever,
|
|
|
|
// meaning it always has lazy matching on.
|
|
|
|
func (d *compressor) deflateLazy() {
|
2019-11-18 00:18:33 -05:00
|
|
|
s := d.state
|
2016-11-03 17:16:01 -05:00
|
|
|
// Sanity enables additional runtime tests.
|
|
|
|
// It's intended to be used during development
|
|
|
|
// to supplement the currently ad-hoc unit tests.
|
|
|
|
const sanity = false
|
|
|
|
|
2019-11-18 00:18:33 -05:00
|
|
|
if d.windowEnd-s.index < minMatchLength+maxMatchLength && !d.sync {
|
2016-11-03 17:16:01 -05:00
|
|
|
return
|
|
|
|
}
|
|
|
|
|
2019-11-18 00:18:33 -05:00
|
|
|
s.maxInsertIndex = d.windowEnd - (minMatchLength - 1)
|
|
|
|
if s.index < s.maxInsertIndex {
|
|
|
|
s.hash = hash4(d.window[s.index : s.index+minMatchLength])
|
2016-11-03 17:16:01 -05:00
|
|
|
}
|
|
|
|
|
|
|
|
for {
|
2019-11-18 00:18:33 -05:00
|
|
|
if sanity && s.index > d.windowEnd {
|
2016-11-03 17:16:01 -05:00
|
|
|
panic("index > windowEnd")
|
|
|
|
}
|
2019-11-18 00:18:33 -05:00
|
|
|
lookahead := d.windowEnd - s.index
|
2016-11-03 17:16:01 -05:00
|
|
|
if lookahead < minMatchLength+maxMatchLength {
|
|
|
|
if !d.sync {
|
|
|
|
return
|
|
|
|
}
|
2019-11-18 00:18:33 -05:00
|
|
|
if sanity && s.index > d.windowEnd {
|
2016-11-03 17:16:01 -05:00
|
|
|
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
|
2019-11-18 00:18:33 -05:00
|
|
|
d.tokens.AddLiteral(d.window[s.index-1])
|
2016-11-03 17:16:01 -05:00
|
|
|
d.byteAvailable = false
|
|
|
|
}
|
|
|
|
if d.tokens.n > 0 {
|
2019-11-18 00:18:33 -05:00
|
|
|
if d.err = d.writeBlock(&d.tokens, s.index, false); d.err != nil {
|
2016-11-03 17:16:01 -05:00
|
|
|
return
|
|
|
|
}
|
2019-11-18 00:18:33 -05:00
|
|
|
d.tokens.Reset()
|
2016-11-03 17:16:01 -05:00
|
|
|
}
|
|
|
|
return
|
|
|
|
}
|
|
|
|
}
|
2019-11-18 00:18:33 -05:00
|
|
|
if s.index < s.maxInsertIndex {
|
2016-11-03 17:16:01 -05:00
|
|
|
// Update the hash
|
2019-11-18 00:18:33 -05:00
|
|
|
s.hash = hash4(d.window[s.index : s.index+minMatchLength])
|
|
|
|
ch := s.hashHead[s.hash&hashMask]
|
|
|
|
s.chainHead = int(ch)
|
|
|
|
s.hashPrev[s.index&windowMask] = ch
|
|
|
|
s.hashHead[s.hash&hashMask] = uint32(s.index + s.hashOffset)
|
|
|
|
}
|
|
|
|
prevLength := s.length
|
|
|
|
prevOffset := s.offset
|
|
|
|
s.length = minMatchLength - 1
|
|
|
|
s.offset = 0
|
|
|
|
minIndex := s.index - windowSize
|
2016-11-03 17:16:01 -05:00
|
|
|
if minIndex < 0 {
|
|
|
|
minIndex = 0
|
|
|
|
}
|
|
|
|
|
2019-11-18 00:18:33 -05:00
|
|
|
if s.chainHead-s.hashOffset >= minIndex && lookahead > prevLength && prevLength < d.lazy {
|
|
|
|
if newLength, newOffset, ok := d.findMatch(s.index, s.chainHead-s.hashOffset, minMatchLength-1, lookahead); ok {
|
|
|
|
s.length = newLength
|
|
|
|
s.offset = newOffset
|
2016-11-03 17:16:01 -05:00
|
|
|
}
|
|
|
|
}
|
2019-11-18 00:18:33 -05:00
|
|
|
if prevLength >= minMatchLength && s.length <= prevLength {
|
2016-11-03 17:16:01 -05:00
|
|
|
// There was a match at the previous step, and the current match is
|
|
|
|
// not better. Output the previous match.
|
2019-11-18 00:18:33 -05:00
|
|
|
d.tokens.AddMatch(uint32(prevLength-3), uint32(prevOffset-minOffsetSize))
|
2016-11-03 17:16:01 -05:00
|
|
|
|
|
|
|
// 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.
|
|
|
|
var newIndex int
|
2019-11-18 00:18:33 -05:00
|
|
|
newIndex = s.index + prevLength - 1
|
2016-11-03 17:16:01 -05:00
|
|
|
// Calculate missing hashes
|
|
|
|
end := newIndex
|
2019-11-18 00:18:33 -05:00
|
|
|
if end > s.maxInsertIndex {
|
|
|
|
end = s.maxInsertIndex
|
2016-11-03 17:16:01 -05:00
|
|
|
}
|
|
|
|
end += minMatchLength - 1
|
2019-11-18 00:18:33 -05:00
|
|
|
startindex := s.index + 1
|
|
|
|
if startindex > s.maxInsertIndex {
|
|
|
|
startindex = s.maxInsertIndex
|
2016-11-03 17:16:01 -05:00
|
|
|
}
|
|
|
|
tocheck := d.window[startindex:end]
|
|
|
|
dstSize := len(tocheck) - minMatchLength + 1
|
|
|
|
if dstSize > 0 {
|
2019-11-18 00:18:33 -05:00
|
|
|
dst := s.hashMatch[:dstSize]
|
2016-11-03 17:16:01 -05:00
|
|
|
bulkHash4(tocheck, dst)
|
|
|
|
var newH uint32
|
|
|
|
for i, val := range dst {
|
|
|
|
di := i + startindex
|
|
|
|
newH = val & hashMask
|
|
|
|
// Get previous value with the same hash.
|
|
|
|
// Our chain should point to the previous value.
|
2019-11-18 00:18:33 -05:00
|
|
|
s.hashPrev[di&windowMask] = s.hashHead[newH]
|
2016-11-03 17:16:01 -05:00
|
|
|
// Set the head of the hash chain to us.
|
2019-11-18 00:18:33 -05:00
|
|
|
s.hashHead[newH] = uint32(di + s.hashOffset)
|
2016-11-03 17:16:01 -05:00
|
|
|
}
|
2019-11-18 00:18:33 -05:00
|
|
|
s.hash = newH
|
2016-11-03 17:16:01 -05:00
|
|
|
}
|
|
|
|
|
2019-11-18 00:18:33 -05:00
|
|
|
s.index = newIndex
|
2016-11-03 17:16:01 -05:00
|
|
|
d.byteAvailable = false
|
2019-11-18 00:18:33 -05:00
|
|
|
s.length = minMatchLength - 1
|
2016-11-03 17:16:01 -05:00
|
|
|
if d.tokens.n == maxFlateBlockTokens {
|
|
|
|
// The block includes the current character
|
2019-11-18 00:18:33 -05:00
|
|
|
if d.err = d.writeBlock(&d.tokens, s.index, false); d.err != nil {
|
2016-11-03 17:16:01 -05:00
|
|
|
return
|
|
|
|
}
|
2019-11-18 00:18:33 -05:00
|
|
|
d.tokens.Reset()
|
2016-11-03 17:16:01 -05:00
|
|
|
}
|
|
|
|
} else {
|
|
|
|
// Reset, if we got a match this run.
|
2019-11-18 00:18:33 -05:00
|
|
|
if s.length >= minMatchLength {
|
|
|
|
s.ii = 0
|
2016-11-03 17:16:01 -05:00
|
|
|
}
|
|
|
|
// We have a byte waiting. Emit it.
|
|
|
|
if d.byteAvailable {
|
2019-11-18 00:18:33 -05:00
|
|
|
s.ii++
|
|
|
|
d.tokens.AddLiteral(d.window[s.index-1])
|
2016-11-03 17:16:01 -05:00
|
|
|
if d.tokens.n == maxFlateBlockTokens {
|
2019-11-18 00:18:33 -05:00
|
|
|
if d.err = d.writeBlock(&d.tokens, s.index, false); d.err != nil {
|
2016-11-03 17:16:01 -05:00
|
|
|
return
|
|
|
|
}
|
2019-11-18 00:18:33 -05:00
|
|
|
d.tokens.Reset()
|
2016-11-03 17:16:01 -05:00
|
|
|
}
|
2019-11-18 00:18:33 -05:00
|
|
|
s.index++
|
2016-11-03 17:16:01 -05:00
|
|
|
|
|
|
|
// If we have a long run of no matches, skip additional bytes
|
2019-11-18 00:18:33 -05:00
|
|
|
// Resets when s.ii overflows after 64KB.
|
|
|
|
if s.ii > 31 {
|
|
|
|
n := int(s.ii >> 5)
|
2016-11-03 17:16:01 -05:00
|
|
|
for j := 0; j < n; j++ {
|
2019-11-18 00:18:33 -05:00
|
|
|
if s.index >= d.windowEnd-1 {
|
2016-11-03 17:16:01 -05:00
|
|
|
break
|
|
|
|
}
|
|
|
|
|
2019-11-18 00:18:33 -05:00
|
|
|
d.tokens.AddLiteral(d.window[s.index-1])
|
2016-11-03 17:16:01 -05:00
|
|
|
if d.tokens.n == maxFlateBlockTokens {
|
2019-11-18 00:18:33 -05:00
|
|
|
if d.err = d.writeBlock(&d.tokens, s.index, false); d.err != nil {
|
2016-11-03 17:16:01 -05:00
|
|
|
return
|
|
|
|
}
|
2019-11-18 00:18:33 -05:00
|
|
|
d.tokens.Reset()
|
2016-11-03 17:16:01 -05:00
|
|
|
}
|
2019-11-18 00:18:33 -05:00
|
|
|
s.index++
|
2016-11-03 17:16:01 -05:00
|
|
|
}
|
|
|
|
// Flush last byte
|
2019-11-18 00:18:33 -05:00
|
|
|
d.tokens.AddLiteral(d.window[s.index-1])
|
2016-11-03 17:16:01 -05:00
|
|
|
d.byteAvailable = false
|
2019-11-18 00:18:33 -05:00
|
|
|
// s.length = minMatchLength - 1 // not needed, since s.ii is reset above, so it should never be > minMatchLength
|
2016-11-03 17:16:01 -05:00
|
|
|
if d.tokens.n == maxFlateBlockTokens {
|
2019-11-18 00:18:33 -05:00
|
|
|
if d.err = d.writeBlock(&d.tokens, s.index, false); d.err != nil {
|
2016-11-03 17:16:01 -05:00
|
|
|
return
|
|
|
|
}
|
2019-11-18 00:18:33 -05:00
|
|
|
d.tokens.Reset()
|
2016-11-03 17:16:01 -05:00
|
|
|
}
|
|
|
|
}
|
|
|
|
} else {
|
2019-11-18 00:18:33 -05:00
|
|
|
s.index++
|
2016-11-03 17:16:01 -05:00
|
|
|
d.byteAvailable = true
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
func (d *compressor) store() {
|
|
|
|
if d.windowEnd > 0 && (d.windowEnd == maxStoreBlockSize || d.sync) {
|
|
|
|
d.err = d.writeStoredBlock(d.window[:d.windowEnd])
|
|
|
|
d.windowEnd = 0
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
// fillWindow will fill the buffer with data for huffman-only compression.
|
|
|
|
// The number of bytes copied is returned.
|
|
|
|
func (d *compressor) fillBlock(b []byte) int {
|
|
|
|
n := copy(d.window[d.windowEnd:], b)
|
|
|
|
d.windowEnd += n
|
|
|
|
return n
|
|
|
|
}
|
|
|
|
|
|
|
|
// storeHuff 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) storeHuff() {
|
|
|
|
if d.windowEnd < len(d.window) && !d.sync || d.windowEnd == 0 {
|
|
|
|
return
|
|
|
|
}
|
2019-11-18 00:18:33 -05:00
|
|
|
d.w.writeBlockHuff(false, d.window[:d.windowEnd], d.sync)
|
2016-11-03 17:16:01 -05:00
|
|
|
d.err = d.w.err
|
|
|
|
d.windowEnd = 0
|
|
|
|
}
|
|
|
|
|
2019-11-18 00:18:33 -05:00
|
|
|
// storeFast will compress and store the currently added data,
|
2016-11-03 17:16:01 -05:00
|
|
|
// if enough has been accumulated or we at the end of the stream.
|
|
|
|
// Any error that occurred will be in d.err
|
2019-11-18 00:18:33 -05:00
|
|
|
func (d *compressor) storeFast() {
|
2016-11-03 17:16:01 -05:00
|
|
|
// We only compress if we have maxStoreBlockSize.
|
2019-11-18 00:18:33 -05:00
|
|
|
if d.windowEnd < len(d.window) {
|
2016-11-03 17:16:01 -05:00
|
|
|
if !d.sync {
|
|
|
|
return
|
|
|
|
}
|
|
|
|
// Handle extremely small sizes.
|
|
|
|
if d.windowEnd < 128 {
|
|
|
|
if d.windowEnd == 0 {
|
|
|
|
return
|
|
|
|
}
|
|
|
|
if d.windowEnd <= 32 {
|
|
|
|
d.err = d.writeStoredBlock(d.window[:d.windowEnd])
|
|
|
|
} else {
|
2019-11-18 00:18:33 -05:00
|
|
|
d.w.writeBlockHuff(false, d.window[:d.windowEnd], true)
|
2016-11-03 17:16:01 -05:00
|
|
|
d.err = d.w.err
|
|
|
|
}
|
2019-11-18 00:18:33 -05:00
|
|
|
d.tokens.Reset()
|
2016-11-03 17:16:01 -05:00
|
|
|
d.windowEnd = 0
|
2019-11-18 00:18:33 -05:00
|
|
|
d.fast.Reset()
|
2016-11-03 17:16:01 -05:00
|
|
|
return
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2019-11-18 00:18:33 -05:00
|
|
|
d.fast.Encode(&d.tokens, d.window[:d.windowEnd])
|
2016-11-03 17:16:01 -05:00
|
|
|
// If we made zero matches, store the block as is.
|
2019-11-18 00:18:33 -05:00
|
|
|
if d.tokens.n == 0 {
|
2016-11-03 17:16:01 -05:00
|
|
|
d.err = d.writeStoredBlock(d.window[:d.windowEnd])
|
|
|
|
// If we removed less than 1/16th, huffman compress the block.
|
|
|
|
} else if int(d.tokens.n) > d.windowEnd-(d.windowEnd>>4) {
|
2019-11-18 00:18:33 -05:00
|
|
|
d.w.writeBlockHuff(false, d.window[:d.windowEnd], d.sync)
|
2016-11-03 17:16:01 -05:00
|
|
|
d.err = d.w.err
|
|
|
|
} else {
|
2019-11-18 00:18:33 -05:00
|
|
|
d.w.writeBlockDynamic(&d.tokens, false, d.window[:d.windowEnd], d.sync)
|
2016-11-03 17:16:01 -05:00
|
|
|
d.err = d.w.err
|
|
|
|
}
|
2019-11-18 00:18:33 -05:00
|
|
|
d.tokens.Reset()
|
2016-11-03 17:16:01 -05:00
|
|
|
d.windowEnd = 0
|
|
|
|
}
|
|
|
|
|
|
|
|
// write will add input byte to the stream.
|
|
|
|
// Unless an error occurs all bytes will be consumed.
|
|
|
|
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, d.err
|
|
|
|
}
|
|
|
|
|
|
|
|
func (d *compressor) syncFlush() error {
|
|
|
|
d.sync = true
|
|
|
|
if d.err != nil {
|
|
|
|
return d.err
|
|
|
|
}
|
|
|
|
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).fillBlock
|
|
|
|
d.step = (*compressor).store
|
|
|
|
case level == ConstantCompression:
|
2019-11-18 00:18:33 -05:00
|
|
|
d.w.logReusePenalty = uint(4)
|
2016-11-03 17:16:01 -05:00
|
|
|
d.window = make([]byte, maxStoreBlockSize)
|
|
|
|
d.fill = (*compressor).fillBlock
|
|
|
|
d.step = (*compressor).storeHuff
|
|
|
|
case level == DefaultCompression:
|
|
|
|
level = 5
|
|
|
|
fallthrough
|
2019-11-18 00:18:33 -05:00
|
|
|
case level >= 1 && level <= 6:
|
|
|
|
d.w.logReusePenalty = uint(level + 1)
|
|
|
|
d.fast = newFastEnc(level)
|
|
|
|
d.window = make([]byte, maxStoreBlockSize)
|
|
|
|
d.fill = (*compressor).fillBlock
|
|
|
|
d.step = (*compressor).storeFast
|
|
|
|
case 7 <= level && level <= 9:
|
|
|
|
d.w.logReusePenalty = uint(level)
|
|
|
|
d.state = &advancedState{}
|
2016-11-03 17:16:01 -05:00
|
|
|
d.compressionLevel = levels[level]
|
|
|
|
d.initDeflate()
|
|
|
|
d.fill = (*compressor).fillDeflate
|
2019-11-18 00:18:33 -05:00
|
|
|
d.step = (*compressor).deflateLazy
|
2016-11-03 17:16:01 -05:00
|
|
|
default:
|
|
|
|
return fmt.Errorf("flate: invalid compression level %d: want value in range [-2, 9]", level)
|
|
|
|
}
|
|
|
|
return nil
|
|
|
|
}
|
|
|
|
|
|
|
|
// reset the state of the compressor.
|
|
|
|
func (d *compressor) reset(w io.Writer) {
|
|
|
|
d.w.reset(w)
|
|
|
|
d.sync = false
|
|
|
|
d.err = nil
|
|
|
|
// We only need to reset a few things for Snappy.
|
2019-11-18 00:18:33 -05:00
|
|
|
if d.fast != nil {
|
|
|
|
d.fast.Reset()
|
2016-11-03 17:16:01 -05:00
|
|
|
d.windowEnd = 0
|
2019-11-18 00:18:33 -05:00
|
|
|
d.tokens.Reset()
|
2016-11-03 17:16:01 -05:00
|
|
|
return
|
|
|
|
}
|
|
|
|
switch d.compressionLevel.chain {
|
|
|
|
case 0:
|
|
|
|
// level was NoCompression or ConstantCompresssion.
|
|
|
|
d.windowEnd = 0
|
|
|
|
default:
|
2019-11-18 00:18:33 -05:00
|
|
|
s := d.state
|
|
|
|
s.chainHead = -1
|
|
|
|
for i := range s.hashHead {
|
|
|
|
s.hashHead[i] = 0
|
2016-11-03 17:16:01 -05:00
|
|
|
}
|
2019-11-18 00:18:33 -05:00
|
|
|
for i := range s.hashPrev {
|
|
|
|
s.hashPrev[i] = 0
|
2016-11-03 17:16:01 -05:00
|
|
|
}
|
2019-11-18 00:18:33 -05:00
|
|
|
s.hashOffset = 1
|
|
|
|
s.index, d.windowEnd = 0, 0
|
2016-11-03 17:16:01 -05:00
|
|
|
d.blockStart, d.byteAvailable = 0, false
|
2019-11-18 00:18:33 -05:00
|
|
|
d.tokens.Reset()
|
|
|
|
s.length = minMatchLength - 1
|
|
|
|
s.offset = 0
|
|
|
|
s.hash = 0
|
|
|
|
s.ii = 0
|
|
|
|
s.maxInsertIndex = 0
|
2016-11-03 17:16:01 -05:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
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 (ConstantCompression) 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:
|
|
|
|
// http://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)
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
// ResetDict discards the writer's state and makes it equivalent to
|
|
|
|
// the result of NewWriter or NewWriterDict called with dst
|
|
|
|
// and w's level, but sets a specific dictionary.
|
|
|
|
func (w *Writer) ResetDict(dst io.Writer, dict []byte) {
|
|
|
|
w.dict = dict
|
|
|
|
w.d.reset(dst)
|
|
|
|
w.d.fillWindow(w.dict)
|
|
|
|
}
|