Improve error string compression

[samblenny]

By my reading of py/makecompresseddata.py, it looks like error string compression only works for pure-ASCII strings. If that's true, builds for languages that give us trouble on SAMD21 M0 boards are probably using uncompressed strings.

Implementing better compression might be an easy win for freeing up code space. A possible fix: 1) write a python script to generate a zlib preset dictionary over all the error strings for a translation, then 2) modify the decompression to use zlib with inflateSetDictionary().

[samblenny]

After more research, this looks a bit more complicated: 1) The full zlib API isn't available in CircuitPython core. Instead it has the more limited uzlib. 2) zlib and uzlib both use LZ77/DEFLATE with a sliding window (not LZ78 or LZW), so building a good shared dictionary to use with random access decompression of individual strings might be a little tricky. Plausibly doable, but tricky.

[tannewt]

I don't think that is the right file for compression. (Sorry) I think that is what MicroPython uses and they only support English.

We use https://github.com/adafruit/circuitpython/blob/main/py/maketranslationdata.py which does handle Unicode. That was one of the main requirements of the compression.

[samblenny]

So, in that case, it looks like everything is just Huffman coded? It seems plausible there might still be benefit in switching to LZ77/DEFLATE using uzlib with a shared preset dictionary? In theory, that might save more bytes for redundant substrings across error messages? Does that sound interesting? I'm worried that the USB error handling stuff I want to add may not fit if I end up needing to upstream stuff to TinyUSB rather than adding it to usb_host (or whatever the module might be for RP2350-specific usb code, if that's a thing).

[tannewt]

It's more complicated that you think. It isn't just huffman. I wouldn't spend time on it.

We can worry about fitting the USB stuff after you've added it. We can always put things behind #ifdef.

[samblenny]

okay. This sounds like a dead end then.