Ensure the gc module is available during interpreter exit #4166
Conversation
There was a problem hiding this comment.
To clarify, this is only in this method since it also is registered as an atexit function. Otherwise it wouldn't be necessary here.
There was a problem hiding this comment.
I suppose it would also be worth adding a comment that this is here deliberately (since gc is already imported at the module level any delinter is going to complain about this...)
There was a problem hiding this comment.
the # noqa flag works on most linters?
|
Seems fine -- and pretty harmless in the 99.9% of non-corner cases where this is a non-issue. |
|
👍 from me modulo a comment / linter protection flag. |
|
Merged locally to add comment + linter flage |
|
@tacaswell Where did you merge locally? GitHub just shows that you merged my PR as-is. |
|
Or to be more specific, I'm wondering how you got GitHub to recognize this as merged, while also including your own changes. It would be lovely if I knew how to do that. |
|
Merged as 13a8f31 GH keeps track of 'mergestate' based on if the commits on the merge-branch are accessible from the target branch so all you have to do is merge locally and push to the target branch and GH is happy. |
This is directly related to #4165. Another side effect of
run_setupis that any modules that are imported during therun_setupcall are later removed fromsys.modules. If there are no other references to that module they will be garbage collected and any previous references to globals in that module will be replaced withNone.Because
destroy_allis registered as anatexitfunction it will run even after all the module itself has been garbage collected (there are other contexts where this might happen as well). This seems to otherwise work since it was changed to aclassmethod, so now theGcfclass itself is still hanging around. But it may be necessary to re-import thegcmodule if it's to be made use of.