flate: read data with partial flush

[vajexal]

problem: data written with zlib's Z_PARTIAL_FLUSH can't be read because decompressor returns only after Z_SYNC_FLUSH block or data end (not taking into account internal buffer flush). Maybe flush mode parameter could be added to decompressor? Using this parameter we could control finishBlock method to return after each block. Something like

func (f *decompressor) finishBlock() {
	if f.final {
		if f.dict.availRead() > 0 {
			f.toRead = f.dict.readFlush()
		}
		
		f.err = io.EOF
	} else if f.flushMode == PARTIALPARTIAL_FLUSH && f.dict.availRead() > 0 {
		f.toRead = f.dict.readFlush()
	}

	f.step = nextBlock
}

New constructor (something like NewPartialBlockReader) will be needed

code to test

package main

import (
	"bytes"
	"encoding/base64"
	"fmt"

	"github.com/klauspost/compress/flate"
)

func main() {
	data, err := base64.StdEncoding.DecodeString("ykjNycnXUSjPL8pJAQg=")
	if err != nil {
		panic(err)
	}

	r := flate.NewReader(bytes.NewReader(data))
	rb := make([]byte, 32)
	n, err := r.Read(rb)
	if err != nil {
		panic(err)
	}

	expected := "hello, world"
	actual := string(rb[:n])
	if expected != actual {
		panic(fmt.Sprintf("expected: %v, got: %v", expected, actual))
	}
}

[klauspost]

@vajexal Ah, so it slaps on an empty block with predefined table. But yeah, we will not have all the 10 bits, so it wouldn't be something we could do automatically.

Actually adding the option is a bit more annoying, since they are returning an interface. I am not a fan of "yet another New only for this", and adding an interface seems annoying to use. I think a func NewReaderOpts(w, ReaderOptions{}) is the best I can come up with.

[vajexal]

if you ok with adding new constructor (NewReaderOpts), i can make a pr

[klauspost]

@vajexal Sure. A Dict []byte is the only other property that comes to mind. In the future a decode window max could maybe be relevant.