[dhsysusbsjsi]

What I haven’t seen discussed here much is optimising for developer maintenance. I think the author’s solution is great for this; it’s easy to understand and each move has a bit pattern. Easy to debug. Somebody taking over in the future will understand this compression.

If on the other hand you can squeeze another 10% storage from Huffman encoded inverse tree lookup tables that only neckbeards understand, you’re limiting your pool of people able to do maintenance on this system in the future when the author is long gone.

[stevage]

Hopefully you just use a library for the Huffman tables and there's not much to maintain.