Ok I just understood your version now. You use only one array, store the hex elements from the middle and store the final from the start reusing the full array at the end. I like it very much, thanks!

This is actually really nice!

oh wow. I didn't wanted to use append to avoid multiple allocations, but the cap trick is nice! but how is yours so much faster than mine? what are those allocations in my version? I explicitly wanted to avoid those, that's whhy I made a byte slice and not string concatenation

Awesome review, thank you! Yes, check_config() is different. I didn't even thought of making config dict-like. I introduced a method config.set_param() and config.get_param() but it's not really better than setattr and getattr. Space typo :)

This is not DRY. Also it's better to define those functions as methods on the Range class instead for obvious reasons.

Wow, I did not even thought about separating the token definitions! I already separated the lexer, because at the first time I was doing parsing also, so even the parser logic was in the __init__ :) but this should clear things out even more! Thanks.