ModelFormSet - objects are deleted even if commit=False

[laureline.guerin@…]

In ModelFormSet, when the deletion checkbox is checked for an existing object, and the formset is saved with commit=False:

formset.save(commit=False)

the object is deleted.

[laureline.guerin@…]

Patch for described issue

[Jacob]

This patch isn't correct, since the related objects will never get deleted by anyone. This is a tricky issue to solve.

[damien_szczyt]

I can't understand why this patch isn't correct. Can you tell us more ?
I had the bug yesterday, and the actual design for delete related objects with formsets is very dangerous. When we check the "delete", and preview the formset before submit, the related objects are deleted. There should be a lot of data lost because of this issue.

[Wedg]

I'm also not understanding why the patch won't work - do we need to document the fact that if you call formset.save(commit=False) you'd need to call delete() for each instance in the formset.deleted_objects just like you'd have to call save() for other changed/new instances? (If the patch were applied...)

From what I can tell, this is all that's happening in the BaseModelFormSet.save_existing_objects method, and all the objects that would have delete() called are stored in formset.deleted_objects.

[Wedg]

Patch against latest trunk.

[Wedg]

Regression test for issue.

[Wedg]

So... hopefully I did this right. The current trunk fails the regression test (deletes the model even though commit=False), but with the patch applied it passes. Please let me know if I need to make any corrections/whatever. :D

Oh, and I'll try to get a doc patch for the behavior too, unless someone beats me to it.

[Wedg]

Doc patch describing new/fixed behavior so people don't forget to delete.

[Aymeric Augustin]

Change UI/UX from NULL to False.

[Aymeric Augustin]

Change Easy pickings from NULL to False.

[Carl Meyer]

I don't understand the rationale jacob gave for rejecting the initial patch ("related objects will never get deleted by anyone.") Flipping back to Accepted status; even if the current patch is wrong for some reason I'm not getting, I think this is still clearly a bug that should be fixed.

[Tim Graham]

Updated ​PR. Seems straightforward to me.

[Tim Graham]

Duplicate of #17988 in which @akaariai suggests changing the current behavior would be backwards-incompatible (of course, all bug fixes are -- I guess it's whether or not you consider the current design a bug or a feature). This is what Jacob meant by "related objects will never get deleted by anyone" -- the pull request solves this by a documentation update which informs users they must handle formset.deleted_objects manually.

IMO, the justification for changing the behavior is that it's unexpected that anything DB related happens with commit=False.

[Tim Graham]

​django-dev thread hasn't received any responses in an attempt to get a design decision.

[loic84]

+1 for this PR.

As mentioned on IRC, this together with formally documenting ModelFormSet.(new|changed|deleted)_objects is IMO the cleanest solution to this problem.

[Tim Graham <timograham@…>]

In 65e03a424e82e157b4513cdebb500891f5c78363:

Fixed #10284 -- ModelFormSet.save(commit=False) no longer deletes objects

Thanks laureline.guerin@ and Wedg.

[Simon Charette <charette.s@…>]

In de1d5d5df5238136e8cd114e36065857bee1ace4:

[1.6.x] Fixed #21594 -- Added note about model formsets deleting objects.

This behavior has been fixed in 65e03a424e. refs #10284.

[Simon Charette <charette.s@…>]

In a53820b1b1747d6f9b5ee5723aa7526ededdaba7:

[1.5.x] Fixed #21594 -- Added note about model formsets deleting objects.

This behavior has been fixed in 65e03a424e. refs #10284.

Backport of de1d5d5df5 from stable/1.6.x.

[Simon Charette <charette.s@…>]

In 474e7dd6d08105fce66d7f62670a6c120c0b1198:

[1.4.x] Fixed #21594 -- Added note about model formsets deleting objects.

This behavior has been fixed in 65e03a424e. refs #10284.

Backport of de1d5d5df5 from stable/1.6.x.