aboutsummaryrefslogtreecommitdiff
diff options
context:
space:
mode:
Diffstat
-rw-r--r--internal/flatex/decompress_bytes.go13
-rw-r--r--internal/flatex/decompress_test.go6
-rw-r--r--internal/flatex/inflate.go577
3 files changed, 9 insertions, 587 deletions
diff --git a/internal/flatex/decompress_bytes.go b/internal/flatex/decompress_bytes.go
index 9f1ffcfd..45a8b834 100644
--- a/internal/flatex/decompress_bytes.go
+++ b/internal/flatex/decompress_bytes.go
@@ -14,17 +14,16 @@ type bufferDecompressor struct {
var bufferDecompressorPool = sync.Pool{
New: func() any {
fixedHuffmanDecoderInit()
- d := &bufferDecompressor{}
- d.inflater.bits = new([maxNumLit + maxNumDist]int)
- d.inflater.codebits = new([numCodes]int)
+ d := &bufferDecompressor{
+ inflater: sliceInflater{
+ bits: new([maxNumLit + maxNumDist]int),
+ codebits: new([numCodes]int),
+ },
+ }
return d
},
}
-func Decompress(src []byte) (bufpool.Buffer, int, error) {
- return DecompressSized(src, 0)
-}
-
func DecompressSized(src []byte, sizeHint int) (bufpool.Buffer, int, error) {
d := bufferDecompressorPool.Get().(*bufferDecompressor)
defer bufferDecompressorPool.Put(d)
diff --git a/internal/flatex/decompress_test.go b/internal/flatex/decompress_test.go
index c991ea74..e53a6581 100644
--- a/internal/flatex/decompress_test.go
+++ b/internal/flatex/decompress_test.go
@@ -22,13 +22,13 @@ func compressDeflate(t *testing.T, payload []byte) []byte {
return buf.Bytes()
}
-func TestDecompress(t *testing.T) {
+func TestDecompressSized(t *testing.T) {
payload := bytes.Repeat([]byte("golang"), 32)
compressed := compressDeflate(t, payload)
- out, _, err := Decompress(compressed)
+ out, _, err := DecompressSized(compressed, 0)
if err != nil {
- t.Fatalf("Decompress: %v", err)
+ t.Fatalf("DecompressSized: %v", err)
}
defer out.Release()
diff --git a/internal/flatex/inflate.go b/internal/flatex/inflate.go
index 1d32e8cd..a556dc42 100644
--- a/internal/flatex/inflate.go
+++ b/internal/flatex/inflate.go
@@ -8,8 +8,6 @@
package flatex
import (
- "bufio"
- "io"
"math/bits"
"strconv"
"sync"
@@ -41,44 +39,6 @@ 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]
-// 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) 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
@@ -260,514 +220,10 @@ func (h *huffmanDecoder) init(lengths []int) bool {
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
- rBuf *bufio.Reader // created if provided io.Reader does not implement io.ByteReader
- 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.
- window windowDecoder
-
- // 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.window.readFlush()
- }
- }
-}
-
-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.window.writeByte(byte(v))
- if f.window.availWrite() == 0 {
- f.toRead = f.window.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.window.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.window.tryWriteCopy(f.copyDist, f.copyLen)
- if cnt == 0 {
- cnt = f.window.writeCopy(f.copyDist, f.copyLen)
- }
- f.copyLen -= cnt
-
- if f.window.availWrite() == 0 || f.copyLen > 0 {
- f.toRead = f.window.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.window.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.window.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.window.writeMark(cnt)
- if err != nil {
- f.err = noEOF(err)
- return
- }
-
- if f.window.availWrite() == 0 || f.copyLen > 0 {
- f.toRead = f.window.readFlush()
- f.step = (*decompressor).copyData
- return
- }
- f.finishBlock()
-}
-
-func (f *decompressor) finishBlock() {
- if f.final {
- if f.window.availRead() > 0 {
- f.toRead = f.window.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 (f *decompressor) makeReader(r io.Reader) {
- if rr, ok := r.(Reader); ok {
- f.rBuf = nil
- f.r = rr
- return
- }
- // Reuse rBuf if possible. Invariant: rBuf is always created (and owned) by decompressor.
- if f.rBuf != nil {
- f.rBuf.Reset(r)
- } else {
- // bufio.NewReader will not return r, as r does not implement flate.Reader, so it is not bufio.Reader.
- f.rBuf = bufio.NewReader(r)
- }
- f.r = f.rBuf
-}
-
func fixedHuffmanDecoderInit() {
fixedOnce.Do(func() {
// These come from the RFC section 3.2.6.
@@ -787,36 +243,3 @@ func fixedHuffmanDecoderInit() {
fixedHuffmanDecoder.init(bits[:])
})
}
-
-func (f *decompressor) Reset(r io.Reader) error {
- *f = decompressor{
- rBuf: f.rBuf,
- bits: f.bits,
- codebits: f.codebits,
- window: f.window,
- step: (*decompressor).nextBlock,
- }
- f.makeReader(r)
- f.window.init(maxMatchOffset)
- 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.
-// The reader returns [io.EOF] after the final block in the DEFLATE stream has
-// been encountered. Any trailing data after the final block is ignored.
-//
-// The [io.ReadCloser] returned by NewReader also implements [Resetter].
-func NewReader(r io.Reader) io.ReadCloser {
- fixedHuffmanDecoderInit()
-
- var f decompressor
- f.makeReader(r)
- f.bits = new([maxNumLit + maxNumDist]int)
- f.codebits = new([numCodes]int)
- f.step = (*decompressor).nextBlock
- f.window.init(maxMatchOffset)
- return &f
-}
generated by cgit v1.3.1-10-gc9f91 (git 2.54.0) at 2026-05-25 21:16:30 +0000