[3.6] bpo-26153: fix abort in _PyErr_WarnUnawaitedCoroutine during shutdown (GH-5337)#6536
Conversation
…utdown (pythonGH-5337) When an unawaited coroutine is collected very late in shutdown -- like, during the final GC at the end of PyImport_Cleanup -- then it was triggering an interpreter abort, because we'd try to look up the "warnings" module and not only was it missing (we were prepared for that), but the entire module system was missing (which we were not prepared for). I've tried to fix this at the source, by making the utility function get_warnings_attr robust against this in general. Note that it already has the convention that it can return NULL without setting an error, which is how it signals that the attribute it was asked to fetch is missing, and that all callers already check for NULL returns. There's a similar check for being late in shutdown at the top of warn_explicit, which might be unnecessary after this fix, but I'm not sure so I'm going to leave it.. (cherry picked from commit dba976b) Co-authored-by: Nathaniel J. Smith <njs@pobox.com>
Yhg1s
changed the title
|
the original commit message doesn't make much sense (the problem exists in 3.6, but the trigger in coroutine cleanup doesn't) -- but I've left it in place because it's a backport. I've pointed to the right issue for the non-coroutine crash, though. |
|
@njsmith Nathaniel, can you take a look into this? |
|
Looks fine to me. I'd hit merge, but I'm not sure what the commit message is supposed to look like, so I'll leave that for someone else (maybe @Yhg1s?) |
Hm, this actually looks like your commit, Nathaniel, backported to 3.6. So feel free to hit the button :) |
|
It's my commit, but fixing an entirely different issue than the one described in my message, plus there's the extra backport metadata, and then there's the double-GH-links that I forget how you're supposed to handle. So yeah, I'm not confident about how best to do it. |
|
I'm just going to merge it as-is then, to get this in 3.6.6. The commit message may be a little misleading but I don't think that's a big deal. |
|
@Yhg1s: Please replace |
5 participants
When an unawaited coroutine is collected very late in shutdown --
like, during the final GC at the end of PyImport_Cleanup -- then it
was triggering an interpreter abort, because we'd try to look up the
"warnings" module and not only was it missing (we were prepared for
that), but the entire module system was missing (which we were not
prepared for).
I've tried to fix this at the source, by making the utility function
get_warnings_attr robust against this in general. Note that it already
has the convention that it can return NULL without setting an error,
which is how it signals that the attribute it was asked to fetch is
missing, and that all callers already check for NULL returns.
There's a similar check for being late in shutdown at the top of
warn_explicit, which might be unnecessary after this fix, but I'm not
sure so I'm going to leave it..
(cherry picked from commit dba976b)
Co-authored-by: Nathaniel J. Smith njs@pobox.com
https://bugs.python.org/issue32591