On Saturday, 21 May 2016 at 22:01:40 UTC, Walter Bright wrote:
Give it a try, see if it pans out.
Other possibilities are the lookback dictionary is fixed at
1023 characters. Experimenting and benchmarking can indicate if
it should be more or less. Perhaps 512 will yield just as good
compression on real world identifier strings.
Or try another scheme entirely, like huffman coding.
I'm going to try for a lookback and scan which uses quite a bit
of memory. On the other hand you can reuse the memory over and
over again, so that isn't an issue. Although it will take up a
Meg in memory. Also written entirely in D, so...
I think I'll go with a similar setup you have for the encoding
(3 bytes). Should be nearly identical in output.