So after spending so much time implementing and reverse engineering decompression code, I decided to see what I could do with the knowledge I gained from that, and created a compression algorithm named Wilt.
It was specifically designed to be very simple to implement. The decompressor is about a page of C code, but it still manages to beat Deflate most of the time. It lies somewhere between Deflate and bzip2 on average in terms of compression ratio.
> The decompressor is about a page of C code, but it still manages to beat Deflate most of the time. It lies somewhere between Deflate and bzip2 on average in terms of compression ratio.
how does it compare to deflate with the 7-zip deflate encoder? i've found that it beats bzip2 in a lot of cases.
If you can find some data on how that does on the Canterbury Corpus, you can compare with http://code.google.com/p/wilt-compressor/wiki/WiltBenchmarks.
The compressor I wrote for Wilt doesn't use the kind of trickery that the 7-zip Deflate implementation does, though, so it probably has some room to improve too.
well, after i didn't find any data on how it does on the calgary corpus i downloaded the corpus and did the benchmarks myself...
here are my results from using advdef -z4:
text fax Csrc Excl SPRC tech poem html list man play Average
2.73 0.77 2.19 1.37 2.56 2.59 3.07 2.52 2.61 3.25 2.99 2.42
7-zip deflate beats bzip2 on fax, SPRC, list, and man. it beats wilt on text, fax, Csrc, html, list, man, and play.