zlender is a small and fast data compression and decompression library. It is basically a LZW like algorithm optimized for small data blocks about 8 KB.
Unlike other implementations I am aware of, the compressed data consists of a stream of 16bit words, where the first word is the length of the compressed data and the second word is the length of the original data.
If the compressed data would be larger than the original data, the original data is returned. Because of this extra payload, this implementation is not usable for small strings.
If the expansions fails due to corrupted data, null is returned and the error reason is in
Compress.lastError.
Usage:
var s = "ababaabaabababa1234567abababbbababa";
var bs = haxe.io.Bytes.ofString(s);
var c = zlender.Compress.compress(bs);
trace(bs.length);
trace(c.data.length);
var ed = c.expand();
trace(bs.compare(ed));
timings where made on my notebook with a core i7-5700HQ
with random data:
higher numbers are better
| target | compression speed | decompression speed | zip | unzip | inflate |
|---|---|---|---|---|---|
| node js | 11.10 MB/sec | 17.84 MB/sec | |||
| neko 32bit | 12.20 MB/sec | 27.06 MB/sec | 28.33 | 113.55 | 3.13 |
| luajit | 12.73 MB/sec | 45.88 MB/sec | |||
| cpp (vcc 32) | 108.66 MB/sec | 186.27 MB/sec | 86.92 | 98.81 | 30.50 |
| cpp (vcc 64) | 121.38 MB/sec | 190.59 MB/sec | 95.88 | 123.50 | 38.12 |
| cppia | 6.07 MB/sec | 11.67 MB/sec | |||
| cppia -jit | 44.11 MB/sec | fails | |||
| cs | 124.75 MB/sec | 360.17 MB/sec | |||
| java | 128.07 MB/sec | 293.70 MB/sec | 79.55 | 119.09 | 49.50 |
| hl interp | 46.31 MB/sec | 116.60 MB/sec | |||
| hl (gcc 32) | 105.59 MB/sec | 312.00 MB/sec |
with const data:
higher numbers are better
| target | compression speed | decompression speed | zip | unzip | inflate |
|---|---|---|---|---|---|
| node js | 16.66 MB/sec | 53.80 MB/sec | |||
| neko 32bit | 15.73 MB/sec | 74.61 MB/sec | 28.33 | 113.55 | |
| luajit | 23.14 MB/sec | 71.34 MB/sec | |||
| cpp (vcc 32) | 134.80 MB/sec | 284.02 MB/sec | 192.88 | 125.09 | 59.37 |
| cpp (vcc 64) | 162.85 MB/sec | 254.76 MB/sec | 221.50 | 145.69 | |
| cppia | 7.14 MB/sec | 18.29 MB/sec | |||
| cppia -jit | 53.63 MB/sec | fails | |||
| cs | 166.68 MB/sec | 625.51 MB/sec | |||
| java | 178.77 MB/sec | 554.07 MB/sec | 76.97 | 42.36 | 121.87 |
| hl interp | 57.95 MB/sec | 208.33 MB/sec | |||
| hl (gcc 32) | 168.28 MB/sec | 694.11 MB/sec |
zlender is very light weight compression/decompression algorithm written in pure Haxe without any outside dependencies. It is available on all tested targets and even on interpreted targets its speed is decent. On native targets it can outperform zip.