GeezerVison is showing github.com — Remove reference cycle when writing tarfiles · Issue #115256 · python/cpython
Remove reference cycle when writing tarfiles #115256DescriptionFeature or enhancement
Proposal:
The following code keeps a file handle for eternity and even gc.collect does not help.
>>> import tarfile
>>> tarfile.TarFile("archive.tar.gz","w").add("somefile.py")The culprit is a line that itself says that it is not needed. Indeed, it is never used anywhere. https://github.com/python/cpython/blob/main/Lib/tarfile.py#L2033
Now the example might be a rather bad style, but even a typical use case where a TarFile variable is defined and closed afterwards will also keep some stuff in memory until gc.collect happens due to this line. Removing the line also adds a ResourceWarning to the bad style example which currently does not appear.
Has this already been discussed elsewhere?
This is a minor feature, which does not need previous discussion elsewhere
Links to previous discussion of this feature:
No response
Linked PRs
Activityfurkanonder commented on Feb 11, 2024 encukou commented on Feb 12, 2024 How can we know that no one is using this member?
If it needs to be removed, it should be deprecated according to PEP-387.
@furkanonder Do you have a particular question? I'm not sure why you mentioned me here. Reacted by Serhiy Storchaka serhiy-storchaka commented on Feb 12, 2024 When and why this field was added? How it was used in earlier versions? When and why the comment was added? pan324 commented on Feb 12, 2024 ContributorAuthor More actions It was not used even when it was introduced. Several other classes at the time also define self.tarfile and use it in methods but TarInfo does not inherit from them. Here's what things looked like 17 years ago:
The comment was added by @vadmium 8 years ago when he extended some of the docstrings. f817a48
The only public git repo I can find that accesses info.tarfile uses it to find replace symlink infos by their resolved infos: https://github.com/jlevy/instaclone/blob/master/instaclone/archives.py#L65-L75
It appears that this is not just about lingering handles but may actually corrupt archives as reported here: #78843 vadmium commented on Feb 12, 2024 I don't remember why I added the comment. I was probably studying how the module used TarInfo objects, and noticed it setting the tarfile attribute without any documentation or usage of that attribute.
The attribute itself seems comes from implementing Pax format (c64e402). furkanonder commented on Feb 15, 2024
How can we know that no one is using this member? If it needs to be removed, it should be deprecated according to PEP-387.
@furkanonder Do you have a particular question? I'm not sure why you mentioned me here.
I've seen you fixing issues with the tarfile module. I tagged you in case you might have an idea about the issue raised here. Sorry if I've disturbed you. hugovk commented on Jan 29, 2025 Triage: looks like this can be closed, please re-open if there more to do. Thanks!
MetadataMetadataStatusShow more project fields
DevelopmentNo branches or pull requestsIssue actions
|