Curious if AST allocations/layout is a bottleneck?

[osa1]

I was looking at some high-performance language tools to see how they represent parsed ASTs and I found #15657.

The issue is a bit more than a year old, and the flamegraph in the issue is unreadable at least on my system. I was wondering if AST allocations or representation (with malloc'd nodes) is still an issue today. Or maybe you have other bottlenecks/overheads now and AST representation is not an big issue anymore? (or maybe it was never a big issue?)

[MichaReiser]

You might find #12419 interesting. I haven't done any work since.

I overall suspect that AST allocations are less of a bottleneck in ty, simply because ty has to do so much more. But it can be a bottleneck in tools that do less involved work. Reducing allocations (not just in the AST) is generally good for performance and something we pay attention to.