Hmm, that seems like a really nice interface. I have two ideas | questions though: Have you considered automatically exploring different inputs, instead of always asking the user to provide them? A bit like what property-based testing frameworks (e.g. QuickCheck) or fuzzing tools do. You mention Flamegraphs as a potential tool for finding which places in the code have the most impact on performance. But have you heard of causal profiling?
Loved this talk. I didn't attend it live because the room was always full, but I'm glad it's on here now. Also good to see that lots of people refuse to call Twitter X😄
Definitely going to try this out. If I can place benchmarks in the same file as code, I'd be thrilled. Often times the thing I want to benchmark against isn't public to the create.
Thanks! I don't have a guide yet but you can mimic what I do in Divan, which is to feature-gate internal benchmarks and enable the feature from a separate benchmarking crate. github.com/nvzqz/divan/tree/v0.1.11/internal_benches
