bpo-1635741: Port gc module to multiphase initialization

[tiran]

Signed-off-by: Christian Heimes christian@python.org

https://bugs.python.org/issue1635741

Automerge-Triggered-By: GH:tiran

[bedevere-bot]

🤖 New build scheduled with the buildbot fleet by @pablogsal for commit 668780e30978e11065113c438850074f3fefab5b 🤖

If you want to schedule another build, you need to add the ":hammer: test-with-buildbots" label again.

[pablogsal]

LGTM, but I added the buildbot label because the gc module has some special handling on deallocation

[tiran]

refleak bots are complaining about a new reference like in test_ast.

[tiran]

refleak bots are complaining about a new reference like in test_ast.

Found it! gcstate->garbage is initialized much earlier at the beginning of an interpreter lifecycle.

[bedevere-bot]

🤖 New build scheduled with the buildbot fleet by @tiran for commit d1d2cbcb6b14b2f2d8464cb56095b44a4e118ac6 🤖

If you want to schedule another build, you need to add the ":hammer: test-with-buildbots" label again.

[vstinner]

This looks wrong to me. If you create multiple instances of the GC modul, callbacks member will be overriden by a new list at each call. It must be initialized exactly once in _PyGC_Init().

By the way, if (gcstate->garbage == NULL) { test in _PyGC_Init() can be removed, it's useless. This function is called exactly once per interpreter.

[tiran]

Makes sense, I'm moving gcstate->callbacks = PyList_New(0) to _PyGC_Init().

[bedevere-bot]

🤖 New build scheduled with the buildbot fleet by @tiran for commit 7ad98d9 🤖

If you want to schedule another build, you need to add the ":hammer: test-with-buildbots" label again.

[vstinner]

LGTM.

[miss-islington]

@tiran: Status check is done, and it's a pending ❌ .

[miss-islington]

@tiran: Status check is done, and it's a success ✅ .