bpo-46006: Fix _PyUnicode_EqualToASCIIId() for interned strings

[vstinner]

Fix string comparisons for subinterpreters: no longer make the assumption
that two interned strings are not equal if they have different different
memory addresses. This assumption is no longer true since interned
strings have been made per interpreter in Python 3.10.

Fix _PyUnicode_EqualToASCIIId() and type update_slot() functions.

The change makes update_slot() a little bit less efficient:
update_slot() now compares the strings contents, rather than
comparing strings memory addresses.

https://bugs.python.org/issue46006

[vstinner]

cc @methane @serhiy-storchaka @corona10

We will be able to add again this micro-optimization once Python C extensions and third-party C extensions will no longer define and use static types and will stop sharing objects between two interpreters. Sadly, this long term goal may take a few years...

[markshannon]

It is meaningless to compare strings from one fully isolated interpreter with those in another.
So the assumption that id(o1) != id(o2) implies o1 != o2 is still true for a valid comparison of interned strings.

Feel free to add an assert, but please avoid this sort of pessimization for a hypothetical future which may well not happen.

[hroncok]

Suggestion: add the reproducer from https://bugs.python.org/issue46034 as a test?

[vstinner]

Feel free to add an assert, but please avoid this sort of pessimization for a hypothetical future which may well not happen.

@markshannon: I'm not sure what you are talking about. https://bugs.python.org/issue46006 is affecting multiple Python 3.10 users right now. It's not an hypothetical bug. At least, jep and mod_wsgi projects are known to be affected by the bug:

• SubIntepreters don't work with DontWriteBytecodeFlag in Python 3.10 ninia/jep#358
• Decorators on classes ignored GrahamDumpleton/mod_wsgi#729
• (Py3.10) Error on loading: TypeError: TypeChecker() takes no arguments  poljar/weechat-matrix#293
• https://bugzilla.redhat.com/show_bug.cgi?id=2030621
• etc.

The bug may also explain https://bugs.python.org/issue46070 which affects more projects.

[vstinner]

It is meaningless to compare strings from one fully isolated interpreter with those in another.

@markshannon: I added long comments as part of the change to explain the issue. Did you see my comments? It's about interned strings of static types which are shared by all interpreters.

[markshannon]

I'll comment on the issue.

[markshannon]

I agree that a fix is needed. But this is not the needed fix.

[vstinner]

I merged GH-20085 instead.