API docs are being generated and will be shortly available at: https://godot-rust.github.io/docs/gdext/pr-1328

Thanks for your contribution.

You started originally with "the to_string in custom callables is too slow", but now this turned into a +/- 400 LoC diff refactoring the whole CallContext mechanism. I know there was a TODO but that doesn't mean we shouldn't be careful about it 😉 What we need to consider is that the logic gets quite a bit more complex with lazy initialization, since you have to pass through closures and introduce new generic parameters in many places.

What is the end goal -- performance? If yes, could we maybe consider on having a cheap CallContext construction in Release mode, with less information?

Like I wrote on Discord:

Since this is a performance optimization, it would be good to know if the change actually improves things, and by how much. We have the #[bench] infrastructure to run isolated benchmarks: https://github.com/godot-rust/gdext/blob/master/itest/rust/src/benchmarks/mod.rs

It would be nice if you could set up a small benchmark for custom callable and compare it between master and your branch, so we see the impact of to_string(). They are very easy to use, you can look at some existing examples :slight_smile:

TLDR: I think it's good to cut down on useless string conversions, but we need to approach it with a clear scope 🙂

Yeahhh hahah this got way out of hand. The irony is that my original take on this was to just wrap the name in a debug_assertions check, but I wanted to see if there was a solution that didn't have that limitation. I really thought I was on the right track, but I guess that's the thing about UB. On the plus side, I have a much deeper understanding of the library after this deep dive.

Long story short, I wound up right back at the debug_assertions limitation in a seemingly futile attempt to make it safe, but much more circuitously. I'm just going to close this and put up a PR that's like... Maybe 6 lines of code not counting a benchmark 😅