[issue28797] Modifying class __dict__ inside __set_name__

Changes by Donald Stufft : -- pull_requests: +941 ___ Python tracker ___ ___

[issue28797] Modifying class __dict__ inside __set_name__

Changes by Ned Deily : -- priority: release blocker -> ___ Python tracker ___ ___

[issue28797] Modifying class __dict__ inside __set_name__

Nick Coghlan added the comment: Regarding the docs suggestion above, I need to make some other docs changes for issue 23722, so I'll roll that update in with those. Given that, I think we can mark this issue as resolved. -- resolution: -> fixed stage: patch review -> resolved status:

[issue28797] Modifying class __dict__ inside __set_name__

Roundup Robot added the comment: New changeset 6b8f7d1e2ba4 by Serhiy Storchaka in branch '3.6': Issue #28797: Modifying the class __dict__ inside the __set_name__ method of https://hg.python.org/cpython/rev/6b8f7d1e2ba4 New changeset 18ed518d2eef by Serhiy Storchaka in branch 'default': Issue

[issue28797] Modifying class __dict__ inside __set_name__

Nick Coghlan added the comment: (Plus, as Martin noted, the tuple creation overhead could easily make the more complex patch slower in many cases) -- ___ Python tracker

[issue28797] Modifying class __dict__ inside __set_name__

Nick Coghlan added the comment: I'd be OK with starting with the simpler patch, and only looking at the more complex one if the benchmark suite shows a significant slowdown (I'd be surprised if it did, as it should mainly impact start-up, and class dicts generally aren't that big) --

[issue28797] Modifying class __dict__ inside __set_name__

Martin Teichmann added the comment: I personally prefer the first patch, which iterates over a copy of __dict__. Making a copy of a dict is actually a pretty fast operation, I would even expect that it is faster than the proposed alternative, creating tuples. Even if the second approach

[issue28797] Modifying class __dict__ inside __set_name__

Serhiy Storchaka added the comment: Here is much more complex patch that implements Nick's suggestion. -- Added file: http://bugs.python.org/file45681/set_name-filtered-copy.patch ___ Python tracker

[issue28797] Modifying class __dict__ inside __set_name__

Nick Coghlan added the comment: Ah, I'd missed that. In that case, I think my suggestion to change the name of the local variable is still valid, while the specific approach just needs a comment saying it's handled as a full namespace snapshot so the descriptor type name can be reported in

[issue28797] Modifying class __dict__ inside __set_name__

Serhiy Storchaka added the comment: > Have I missed something that would prevent that from working? Error message contains value->ob_type->tp_name. -- ___ Python tracker

[issue28797] Modifying class __dict__ inside __set_name__

Nick Coghlan added the comment: I was thinking that the notification dict could consist of mappings from attribute names to already bound __set_name__ methods, so the double lookup would be avoided that way (but hadn't actually sent the review where I suggested that). That is, on the second

[issue28797] Modifying class __dict__ inside __set_name__

Serhiy Storchaka added the comment: I considered this option, but if save only descriptors with the __set_name__ attribute, this would need either double looking up the __set_name__ attribute or packing it in a tuple with a value. All this too cumbersome. I chose copying all dictionary as the

[issue28797] Modifying class __dict__ inside __set_name__

Nick Coghlan added the comment: Rather than taking a snapshot of the whole class namespace, I think it will make more sense to make a new empty dictionary of descriptors to notify, and then split the current loop into two loops: - the first loop adds descriptors with __set_name__ attributes

[issue28797] Modifying class __dict__ inside __set_name__

Ned Deily added the comment: Nick, as the original shepherd of PEP 487 (Issue27366), can you review Serhiy's proposed patch for 3.6.0rc1? Thanks! -- nosy: +ncoghlan ___ Python tracker

[issue28797] Modifying class __dict__ inside __set_name__

Serhiy Storchaka added the comment: Proposed patch iterates a copy of __dict__. -- keywords: +patch stage: needs patch -> patch review versions: +Python 3.7 Added file: http://bugs.python.org/file45635/set_name-dict-copy.patch ___ Python tracker

[issue28797] Modifying class __dict__ inside __set_name__

Changes by Emanuel Barry : -- nosy: +Martin.Teichmann ___ Python tracker ___ ___

[issue28797] Modifying class __dict__ inside __set_name__

Emanuel Barry added the comment: Ow! I can confirm the bug still happens on latest trunk. Nice finding! -- nosy: +ebarry, ned.deily, rhettinger, serhiy.storchaka priority: normal -> release blocker stage: -> needs patch ___ Python tracker

[issue28797] Modifying class __dict__ inside __set_name__

New submission from Wombatz: This behavior occurs at least in python-3.6.0b3 and python-3.6.0b4. Modifying the class __dict__ inside the __set_name__ method of a descriptor that is used inside that class can lead to a bug that possibly prevents other descriptors from being initialized (the