[3.12] gh-116510: Fix a Crash Due to Shared Immortal Interned Strings (gh-124865)

[miss-islington]

Fix a crash caused by immortal interned strings being shared between
sub-interpreters that use basic single-phase init. In that case, the string
can be used by an interpreter that outlives the interpreter that created and
interned it. For interpreters that share obmalloc state, also share the
interned dict with the main interpreter.

This is an un-revert of gh-124646 that then addresses the Py_TRACE_REFS
failures identified by gh-124785.
(cherry picked from commit f2cb399)

Co-authored-by: Eric Snow [email protected]

• Issue: CPython 3.12 embedded in WeeChat causes segfault on subsequent calls to Py_EndInterpreter #116510

[ericsnowcurrently]

@nascheme I thought the backport wouldn't apply cleanly, but here we are. Do you recall what was different about the 3.12 fix?

[vstinner]

Please don't merge until the TraceRefs regression is fixed: #124865 (comment)

[nascheme]

I think this can be merged now with GH-125709 applied on top or squashed into it.

[encukou]

@ericsnowcurrently, is this the way you'd do it in 3.12?

[bedevere-bot]

🤖 New build scheduled with the buildbot fleet by @encukou for commit e9591b2 🤖

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

[encukou]

The failed buildbots aren't applicable to 3.12, or fail for unrelated & pre-existing reasons.

[ericsnowcurrently]

@ericsnowcurrently, is this the way you'd do it in 3.12?

This should be good to go. It is backport of gh-124865 + gh-125709. There shouldn't be anything particular to 3.12 that would need to be addressed differently.