gh-100403: Collect GC statistics via -X option

[pablogsal]

• Issue: Improving GC collections: dynamic thresholds, single generation gc and time barriers #100403

[pablogsal]

Example plot this tool does for ./python -m black --check Lib (black over the whole stdlib):

[nascheme]

For testing with Python 3.11.x, this is a version ported on top of v3.11.1 .

The pystats.h header is not really needed so I removed the references to it. I also added a check for the PYTHON_GC_STATS env var. That keeps stderr cleaner when you are building stuff and not actually wanting stats. Rather than hard-coding /tmp/py_stats you could get that path from an env var too. An alternative idea would be to have a new -X flag that enables the stats.

I'm planning to collect stats on some applications but haven't got too far on that yet. Initial indication is that the plots look quite different depending on application.

[pablogsal]

That keeps stderr cleaner when you are building stuff and not actually wanting stats. Rather than hard-coding /tmp/py_stats you could get that path from an env var too. An alternative idea would be to have a new -X flag that enables the stats.

For this PR I opted to mirror the workings and contracts of the other interpreter stats collection that were added in the specializing interpreter to not end with many different ways to activate and retrieve statistics.

[nascheme]

I pushed some changes here: https://github.com/nascheme/cpython/commits/gc_stats in case you want to cherry-pick or merge.

[pablogsal]

@nascheme I have implemented the stats via a -X option where the filename is provided there. This should be a first baseline for the stats collection and we can add public interfaces in the future as I am keeping the containers around. Let me know if you think we should do further modifications.

[bedevere-bot]

Pull request does not meet quality requirements.

[bedevere-bot]

Spurious edit.

[arhadthedev]

Sir Bedevere obtained new powers?

[bedevere-bot]

Spurious edit.