This seems really nice, very ambitious to do it as a "public" state machine that you sink data into and poll compressed bytes out. I can imagine that makes it really nice if you're trying to decompress something in a real-time environment, to cap the execution time. Cool.

Another related algorithm is LZJB, which I've successfully used in embedded environments in production. I have a pure Python implementation [1] that can be useful for "offline" tool development, i.e. when building images for consumption by some embedded target, just thought I'd mention that.

[1]: https://github.com/unwind/python-lzjb