Scalene: a high-performance, high-precision CPU, GPU, and memory profiler for Python

What's Changed

Headlines

• Adds call graphs ("flame graphs") spanning from Python to native code, organized by % of time
• Timeline view of call graphs ("flame chart"), including GIL, I/O, and GC activity
• Flame graph for memory allocations, attributing memory to call stacks
• Improved memory attribution (now tracked per-thread)
• Adds proper support for free-threaded Python

New feature work

• True free-threaded Python support: full memory + CPU profiling by @emeryberger in https://github.com/plasma-umass/scalene/pull/1026
• Bucket native-thread allocations under <native> (#857) by @emeryberger in https://github.com/plasma-umass/scalene/pull/1031
• Native (C/C++) stack collection via signal-handler unwinding by @emeryberger in https://github.com/plasma-umass/scalene/pull/1034
• Stitched Python+native stacks (combined_stacks) by @emeryberger in https://github.com/plasma-umass/scalene/pull/1035
• GUI: combined_stacks viewer (top-N stitched Python+native stacks) by @emeryberger in https://github.com/plasma-umass/scalene/pull/1037
• Timeline view for stitched Python+native stacks by @emeryberger in https://github.com/plasma-umass/scalene/pull/1040
• Memory flame chart + sync per-sample stack capture by @emeryberger in https://github.com/plasma-umass/scalene/pull/1042
• Shrink stack+timeline RAM, harden profiler self-exclusion by @emeryberger in https://github.com/plasma-umass/scalene/pull/1043
• Shrink combined_stacks JSON and enforce schema validation by @emeryberger in https://github.com/plasma-umass/scalene/pull/1044
• Timeline + memory-stacks polish: axes, gridlines, source-line tooltips by @emeryberger in https://github.com/plasma-umass/scalene/pull/1045
• Minify the GUI bundle (-56% on disk) by @emeryberger in https://github.com/plasma-umass/scalene/pull/1046
• GUI improvements: resizable charts, C++ demangling, stack filtering by @emeryberger in https://github.com/plasma-umass/scalene/pull/1047
• Add per-thread native stack sampling for worker threads by @emeryberger in https://github.com/plasma-umass/scalene/pull/1048
• Make --stacks the default; add --no-stacks opt-out by @emeryberger in https://github.com/plasma-umass/scalene/pull/1050
• Drop C-side smear from per-line memory attribution and the flame view by @emeryberger in https://github.com/plasma-umass/scalene/pull/1051

Bug fixes

• Fix root cause of resource tracker death in child processes (#1017) by @emeryberger in https://github.com/plasma-umass/scalene/pull/1024
• Fix SIGSEGV on free-threaded Python with memory profiling by @emeryberger in https://github.com/plasma-umass/scalene/pull/1025
• Fix memory profiling regressions by @emeryberger in https://github.com/plasma-umass/scalene/pull/1027
• Fix CLI line number truncation for lines > 999 by @emeryberger in https://github.com/plasma-umass/scalene/pull/1028
• Fix issue #1022: pytest-xdist + --profile-all on Linux dropped user samples by @emeryberger in https://github.com/plasma-umass/scalene/pull/1055
• Clamp core_utilization to [0, 1] (fixes #1032) by @emeryberger in https://github.com/plasma-umass/scalene/pull/1033

Full Changelog: https://github.com/plasma-umass/scalene/compare/v2.2.1...v2.3.0