Tweaks to decode interface

This commit is contained in:
Caleb Gardner
2022-12-17 16:47:33 -06:00
parent 1b5078c7bd
commit 4f8f5f6928
11 changed files with 87 additions and 75 deletions
+1 -3
View File
@@ -17,6 +17,4 @@ Thanks also to [distri's squashfs library](https://github.com/distr1/distri/tree
## Performance ## Performance
This library, decompressing the Firefox AppImage and using go tests, takes about twice as long as `unsquashfs` on my quad core laptop. (~1 second with the library and about half a second with `unsquashfs`). Testing on a zstd compressed file, my library is anywhere from 5x ~ 7x slower then `unsquashfs`
**My recents tests have shown the Firefox AppImage might be an outlier and this library might be considerably slower (4x ~ 6x time slower then `unsquashfs`)**
+31 -13
View File
@@ -49,16 +49,16 @@ func (r FullReader) process(index int, offset int64, od *outDat, out chan *outDa
od.data = make([]byte, r.blockSize) od.data = make([]byte, r.blockSize)
return return
} }
// rdr := io.LimitReader(toreader.NewReader(r.r, offset), int64(size))
if size == r.sizes[index] { if size == r.sizes[index] {
//Special workaround for zstd for increased performancce. if dec, ok := r.d.(decompress.Decoder); ok {
if zstd, ok := r.d.(*decompress.Zstd); ok { dat := make([]byte, size)
od.data = make([]byte, size) _, od.err = r.r.ReadAt(dat, offset)
_, od.err = r.r.ReadAt(od.data, offset) if od.err != nil {
if od.err == nil { return
od.data, od.err = zstd.Decode(od.data) }
od.data, od.err = dec.Decode(dat, int(r.blockSize))
return
} }
} else {
var rdr io.ReadCloser var rdr io.ReadCloser
rdr, od.err = r.d.Reader(io.LimitReader(toreader.NewReader(r.r, offset), int64(size))) rdr, od.err = r.d.Reader(io.LimitReader(toreader.NewReader(r.r, offset), int64(size)))
if od.err != nil { if od.err != nil {
@@ -69,7 +69,6 @@ func (r FullReader) process(index int, offset int64, od *outDat, out chan *outDa
read, od.err = rdr.Read(od.data) read, od.err = rdr.Read(od.data)
od.data = od.data[:read] od.data = od.data[:read]
rdr.Close() rdr.Close()
}
} else { } else {
od.data = make([]byte, size) od.data = make([]byte, size)
_, od.err = r.r.ReadAt(od.data, offset) _, od.err = r.r.ReadAt(od.data, offset)
@@ -206,23 +205,25 @@ func (r FullReader) WriteTo(w io.Writer) (n int64, err error) {
go r.process(i, int64(offset), od, out) go r.process(i, int64(offset), od, out)
offset += uint64(realSize(r.sizes[i])) offset += uint64(realSize(r.sizes[i]))
} }
wt, ok := w.(io.WriterAt)
if !ok {
var cur int var cur int
cache := make(map[int]outDat) cache := make(map[int]outDat)
var tmpN int var tmpN int
for dat := range out { var dat *outDat
for cur < len(r.sizes) {
dat = <-out
defer pol.Put(dat)
if dat.err != nil { if dat.err != nil {
err = dat.err err = dat.err
pol.Put(dat)
return return
} }
if dat.i != cur { if dat.i != cur {
cache[dat.i] = *dat cache[dat.i] = *dat
pol.Put(dat)
continue continue
} }
tmpN, err = w.Write(dat.data) tmpN, err = w.Write(dat.data)
n += int64(tmpN) n += int64(tmpN)
pol.Put(dat)
if err != nil { if err != nil {
return return
} }
@@ -242,5 +243,22 @@ func (r FullReader) WriteTo(w io.Writer) (n int64, err error) {
cur++ cur++
} }
} }
} else {
var done int
var dat *outDat
for done < len(r.sizes) {
dat = <-out
defer pol.Put(dat)
if dat.err != nil {
err = dat.err
return
}
_, err = wt.WriteAt(dat.data, int64(dat.i*int(r.blockSize)))
if err != nil {
return
}
done++
}
}
return return
} }
+2 -2
View File
@@ -53,14 +53,14 @@ func (r *Reader) advance() (err error) {
} else { } else {
r.cur = io.LimitReader(r.master, int64(size)) r.cur = io.LimitReader(r.master, int64(size))
if size == r.blockSizes[0] { if size == r.blockSizes[0] {
if r.d.Resetable() { if rs, ok := r.d.(decompress.Resetable); ok {
if r.comRdr == nil { if r.comRdr == nil {
r.cur, err = r.d.Reader(r.cur) r.cur, err = r.d.Reader(r.cur)
if err != nil { if err != nil {
return return
} }
} else { } else {
err = r.d.Reset(r.comRdr, r.cur) err = rs.Reset(r.comRdr, r.cur)
r.cur = r.comRdr r.cur = r.comRdr
} }
} else { } else {
-2
View File
@@ -12,8 +12,6 @@ func (g GZip) Reader(src io.Reader) (io.ReadCloser, error) {
return zlib.NewReader(src) return zlib.NewReader(src)
} }
func (g GZip) Resetable() bool { return true }
func (g GZip) Reset(old, src io.Reader) error { func (g GZip) Reset(old, src io.Reader) error {
return old.(zlib.Resetter).Reset(src, nil) return old.(zlib.Resetter).Reset(src, nil)
} }
+8 -5
View File
@@ -1,19 +1,22 @@
package decompress package decompress
import ( import (
"errors"
"io" "io"
) )
var ErrNotResetable = errors.New("decompressor not resetable")
type Decompressor interface { type Decompressor interface {
//Creates a new decompressor reading from src. //Creates a new decompressor reading from src.
Reader(src io.Reader) (io.ReadCloser, error) Reader(src io.Reader) (io.ReadCloser, error)
//Reports whether Reset will work or not. }
Resetable() bool
type Resetable interface {
//Reset attempts to re-use an old decompressor with new data. //Reset attempts to re-use an old decompressor with new data.
//Will return ErrNotResetable if not Resetable(). //Will return ErrNotResetable if not Resetable().
//Must ALWAYS be provided with a reader created with Reader. //Must ALWAYS be provided with a reader created with Reader.
Reset(old, src io.Reader) error Reset(old, src io.Reader) error
} }
type Decoder interface {
//Decodes a chunk of data all at once.
Decode(in []byte, outSize int) ([]byte, error)
}
+9 -2
View File
@@ -12,9 +12,16 @@ func (l Lz4) Reader(r io.Reader) (io.ReadCloser, error) {
return io.NopCloser(lz4.NewReader(r)), nil return io.NopCloser(lz4.NewReader(r)), nil
} }
func (l Lz4) Resetable() bool { return true }
func (l Lz4) Reset(old, src io.Reader) error { func (l Lz4) Reset(old, src io.Reader) error {
old.(*lz4.Reader).Reset(src) old.(*lz4.Reader).Reset(src)
return nil return nil
} }
func (l Lz4) Decode(in []byte, outSize int) (out []byte, err error) {
out = make([]byte, outSize)
outLen, err := lz4.UncompressBlock(in, out)
if outLen < outSize {
out = out[:outLen]
}
return
}
-4
View File
@@ -12,7 +12,3 @@ func (l Lzma) Reader(r io.Reader) (io.ReadCloser, error) {
rdr, err := lzma.NewReader(r) rdr, err := lzma.NewReader(r)
return io.NopCloser(rdr), err return io.NopCloser(rdr), err
} }
func (l Lzma) Resetable() bool { return false }
func (l Lzma) Reset(old, src io.Reader) error { return ErrNotResetable }
-4
View File
@@ -16,7 +16,3 @@ func (l Lzo) Reader(r io.Reader) (io.ReadCloser, error) {
} }
return io.NopCloser(bytes.NewReader(cache)), nil return io.NopCloser(bytes.NewReader(cache)), nil
} }
func (l Lzo) Resetable() bool { return false }
func (l Lzo) Reset(old, src io.Reader) error { return ErrNotResetable }
-2
View File
@@ -13,8 +13,6 @@ func (x Xz) Reader(r io.Reader) (io.ReadCloser, error) {
return io.NopCloser(rdr), err return io.NopCloser(rdr), err
} }
func (x Xz) Resetable() bool { return true }
func (x Xz) Reset(old, src io.Reader) error { func (x Xz) Reset(old, src io.Reader) error {
return old.(*xz.Reader).Reset(src) return old.(*xz.Reader).Reset(src)
} }
+2 -4
View File
@@ -15,15 +15,13 @@ func (z Zstd) Reader(src io.Reader) (io.ReadCloser, error) {
return r.IOReadCloser(), err return r.IOReadCloser(), err
} }
func (z Zstd) Resetable() bool { return true }
func (z Zstd) Reset(old, src io.Reader) error { func (z Zstd) Reset(old, src io.Reader) error {
return old.(*zstd.Decoder).Reset(src) return old.(*zstd.Decoder).Reset(src)
} }
func (z *Zstd) Decode(in []byte) (out []byte, err error) { func (z Zstd) Decode(in []byte, outSize int) ([]byte, error) {
if z.writeToReader == nil { if z.writeToReader == nil {
z.writeToReader, _ = zstd.NewReader(nil) z.writeToReader, _ = zstd.NewReader(nil)
} }
return z.writeToReader.DecodeAll(in, nil) return z.writeToReader.DecodeAll(in, make([]byte, outSize))
} }
+3 -3
View File
@@ -26,7 +26,7 @@ func realSize(siz uint16) uint16 {
} }
func (r *Reader) advance() (err error) { func (r *Reader) advance() (err error) {
if !r.d.Resetable() { if _, ok := r.d.(decompress.Resetable); !ok {
if clr, ok := r.cur.(io.Closer); ok { if clr, ok := r.cur.(io.Closer); ok {
clr.Close() clr.Close()
} }
@@ -39,14 +39,14 @@ func (r *Reader) advance() (err error) {
size := realSize(raw) size := realSize(raw)
r.cur = io.LimitReader(r.master, int64(size)) r.cur = io.LimitReader(r.master, int64(size))
if size == raw { if size == raw {
if r.d.Resetable() { if rs, ok := r.d.(decompress.Resetable); ok {
if r.comRdr == nil { if r.comRdr == nil {
r.cur, err = r.d.Reader(r.cur) r.cur, err = r.d.Reader(r.cur)
if err != nil { if err != nil {
return return
} }
} else { } else {
err = r.d.Reset(r.comRdr, r.cur) err = rs.Reset(r.comRdr, r.cur)
r.cur = r.comRdr r.cur = r.comRdr
} }
} else { } else {