Please note: this is just a quick experiment to see how it would work, it is not production quality, and has not been properly tested.
Here are some results on the Silesia compression corpus:
I did not include smallz4 because it does not create output in the legacy format, so the results are not directly comparable on files larger than 4MiB.
blz4 uses Meson to generate build systems. To create one for the tools on your platform, and build blz4, use something along the lines of:
mkdir build cd build meson .. ninja
You can also simply compile and link the source files.
blz4 includes the leparse and ssparse algorithms from BriefLZ, which gives
-9 and the very slow
- blz4 is finding the closest match of each length, but since match offsets are coded using two bytes regardless of distance, it is enough to find the longest match at each position. If we replaced the match finding with a suffix array or -tree, that should speed up ssparse.
- LZ4 appears to do flexible parsing, is very close to optimal, and much faster.
This projected is licensed under the zlib License (Zlib).