Profiling infrastructure in Python is fragmented, often hard to use, suffers from lack of maintainer time and resources (py-spy hasn't had a release for quite a while), and is sometimes effectively impossible (Numba had no profiling until I wrote Profila, and that is Linux-only and is likely to stay that way). It also is often less focused on the needs of scientific or more broadly batch computing (e.g. flamegraphs are almost useless at visualizing parallelism).
lucascolley
commented
1 week ago
Profiling infrastructure in Python is fragmented, often hard to use, suffers from lack of maintainer time and resources (py-spy hasn't had a release for quite a while), and is sometimes effectively impossible (Numba had no profiling until I wrote Profila, and that is Linux-only and is likely to stay that way). It also is often less focused on the needs of scientific or more broadly batch computing (e.g. flamegraphs are almost useless at visualizing parallelism).