PSA: I've pushed https://github.com/rust-analyzer/rust-analyzer/pull/8254 to optimize debugging perf issues. The idea is that you write a
#[test] that reproduces the perf problem exactly, with Cargo.toml and such, and then you have a fast iteration when applying changes.
I am somewhat emberrassed that it took me years to add that -- so far, I've been debugging perf by running
cargo xtask install && code ., which isn't very productive.
So, now instead of generating the pseudoproject for the completions benchmark, we can pick some nasty real crate?
I guess that's more for debugging specific issues rather than for benchmarkin
it's a replacement for
analysis stats, which works for more fine-grained operations
Kind-of -- I think no one actually used analysis-bench
I actually used it a couple of times in https://github.com/rust-analyzer/rust-analyzer/issues/5397