Remark about hashing using json: isn't it a lot faster to convert to a frozendict and call hash on that? Or (named)tuples?

Remark about hashing using json: isn't it a lot faster to convert to a frozendict and call hash on that? Or (named)tuples?

I tried. Yes it's faster, going from 12us to 3us for the mesh in gltf_unlit.py. However, the mesh shader has one template var which is a dict (used_uv), so the code will have to check for these.

Implemented a variant where dicts and lists are tuple-ized, and with 9 us it's still faster. Will open a new pr, because this is rather orthogonal.

Implemented a variant where dicts and lists are tuple-ized, and with 9 us it's still faster.

To do this well, checking for instances quickly adds up, bringing it to about the same performance as the json implementation. Also we can calculate 100 hashes in 1 ms, so I'm not looking into this further right now.

The only way, I think, that this hashing can be made faster is if you can traverse the object tree and calculate the hash on the fly, rather than producing a big tuple of tuples and hashing that. Updating a hash is easy with the tools in hashlib, but this these operate on bytes, not Python objects. Maybe something for another day.