1
0
Fork 0
forgejo/vendor/github.com/ulikunitz/xz/lzma/encoder.go

269 lines
6.6 KiB
Go
Raw Normal View History

2021-02-28 18:08:33 -05:00
// Copyright 2014-2021 Ulrich Kunitz. All rights reserved.
// Use of this source code is governed by a BSD-style
// license that can be found in the LICENSE file.
package lzma
import (
"fmt"
"io"
)
// opLenMargin provides the upper limit of the number of bytes required
// to encode a single operation.
const opLenMargin = 16
// compressFlags control the compression process.
type compressFlags uint32
// Values for compressFlags.
const (
// all data should be compressed, even if compression is not
// optimal.
all compressFlags = 1 << iota
)
// encoderFlags provide the flags for an encoder.
type encoderFlags uint32
// Flags for the encoder.
const (
// eosMarker requests an EOS marker to be written.
eosMarker encoderFlags = 1 << iota
)
// Encoder compresses data buffered in the encoder dictionary and writes
// it into a byte writer.
type encoder struct {
dict *encoderDict
state *state
re *rangeEncoder
start int64
// generate eos marker
marker bool
limit bool
margin int
}
// newEncoder creates a new encoder. If the byte writer must be
// limited use LimitedByteWriter provided by this package. The flags
// argument supports the eosMarker flag, controlling whether a
// terminating end-of-stream marker must be written.
func newEncoder(bw io.ByteWriter, state *state, dict *encoderDict,
flags encoderFlags) (e *encoder, err error) {
re, err := newRangeEncoder(bw)
if err != nil {
return nil, err
}
e = &encoder{
dict: dict,
state: state,
re: re,
marker: flags&eosMarker != 0,
start: dict.Pos(),
margin: opLenMargin,
}
if e.marker {
e.margin += 5
}
return e, nil
}
// Write writes the bytes from p into the dictionary. If not enough
// space is available the data in the dictionary buffer will be
// compressed to make additional space available. If the limit of the
// underlying writer has been reached ErrLimit will be returned.
func (e *encoder) Write(p []byte) (n int, err error) {
for {
k, err := e.dict.Write(p[n:])
n += k
if err == ErrNoSpace {
if err = e.compress(0); err != nil {
return n, err
}
continue
}
return n, err
}
}
// Reopen reopens the encoder with a new byte writer.
func (e *encoder) Reopen(bw io.ByteWriter) error {
var err error
if e.re, err = newRangeEncoder(bw); err != nil {
return err
}
e.start = e.dict.Pos()
e.limit = false
return nil
}
// writeLiteral writes a literal into the LZMA stream
func (e *encoder) writeLiteral(l lit) error {
var err error
state, state2, _ := e.state.states(e.dict.Pos())
if err = e.state.isMatch[state2].Encode(e.re, 0); err != nil {
return err
}
litState := e.state.litState(e.dict.ByteAt(1), e.dict.Pos())
match := e.dict.ByteAt(int(e.state.rep[0]) + 1)
err = e.state.litCodec.Encode(e.re, l.b, state, match, litState)
if err != nil {
return err
}
e.state.updateStateLiteral()
return nil
}
// iverson implements the Iverson operator as proposed by Donald Knuth in his
// book Concrete Mathematics.
func iverson(ok bool) uint32 {
if ok {
return 1
}
return 0
}
// writeMatch writes a repetition operation into the operation stream
func (e *encoder) writeMatch(m match) error {
var err error
if !(minDistance <= m.distance && m.distance <= maxDistance) {
panic(fmt.Errorf("match distance %d out of range", m.distance))
}
dist := uint32(m.distance - minDistance)
if !(minMatchLen <= m.n && m.n <= maxMatchLen) &&
!(dist == e.state.rep[0] && m.n == 1) {
panic(fmt.Errorf(
"match length %d out of range; dist %d rep[0] %d",
m.n, dist, e.state.rep[0]))
}
state, state2, posState := e.state.states(e.dict.Pos())
if err = e.state.isMatch[state2].Encode(e.re, 1); err != nil {
return err
}
g := 0
for ; g < 4; g++ {
if e.state.rep[g] == dist {
break
}
}
b := iverson(g < 4)
if err = e.state.isRep[state].Encode(e.re, b); err != nil {
return err
}
n := uint32(m.n - minMatchLen)
if b == 0 {
// simple match
e.state.rep[3], e.state.rep[2], e.state.rep[1], e.state.rep[0] =
e.state.rep[2], e.state.rep[1], e.state.rep[0], dist
e.state.updateStateMatch()
if err = e.state.lenCodec.Encode(e.re, n, posState); err != nil {
return err
}
return e.state.distCodec.Encode(e.re, dist, n)
}
b = iverson(g != 0)
if err = e.state.isRepG0[state].Encode(e.re, b); err != nil {
return err
}
if b == 0 {
// g == 0
b = iverson(m.n != 1)
if err = e.state.isRepG0Long[state2].Encode(e.re, b); err != nil {
return err
}
if b == 0 {
e.state.updateStateShortRep()
return nil
}
} else {
// g in {1,2,3}
b = iverson(g != 1)
if err = e.state.isRepG1[state].Encode(e.re, b); err != nil {
return err
}
if b == 1 {
// g in {2,3}
b = iverson(g != 2)
err = e.state.isRepG2[state].Encode(e.re, b)
if err != nil {
return err
}
if b == 1 {
e.state.rep[3] = e.state.rep[2]
}
e.state.rep[2] = e.state.rep[1]
}
e.state.rep[1] = e.state.rep[0]
e.state.rep[0] = dist
}
e.state.updateStateRep()
return e.state.repLenCodec.Encode(e.re, n, posState)
}
// writeOp writes a single operation to the range encoder. The function
// checks whether there is enough space available to close the LZMA
// stream.
func (e *encoder) writeOp(op operation) error {
if e.re.Available() < int64(e.margin) {
return ErrLimit
}
switch x := op.(type) {
case lit:
return e.writeLiteral(x)
case match:
return e.writeMatch(x)
default:
panic("unexpected operation")
}
}
// compress compressed data from the dictionary buffer. If the flag all
// is set, all data in the dictionary buffer will be compressed. The
// function returns ErrLimit if the underlying writer has reached its
// limit.
func (e *encoder) compress(flags compressFlags) error {
n := 0
if flags&all == 0 {
n = maxMatchLen - 1
}
d := e.dict
m := d.m
for d.Buffered() > n {
op := m.NextOp(e.state.rep)
if err := e.writeOp(op); err != nil {
return err
}
d.Discard(op.Len())
}
return nil
}
// eosMatch is a pseudo operation that indicates the end of the stream.
var eosMatch = match{distance: maxDistance, n: minMatchLen}
// Close terminates the LZMA stream. If requested the end-of-stream
// marker will be written. If the byte writer limit has been or will be
// reached during compression of the remaining data in the buffer the
// LZMA stream will be closed and data will remain in the buffer.
func (e *encoder) Close() error {
err := e.compress(all)
if err != nil && err != ErrLimit {
return err
}
if e.marker {
if err := e.writeMatch(eosMatch); err != nil {
return err
}
}
err = e.re.Close()
return err
}
// Compressed returns the number bytes of the input data that been
// compressed.
func (e *encoder) Compressed() int64 {
return e.dict.Pos() - e.start
}