contrib.admin.RelatedFieldWidgetWrapper.__deepcopy__() should copy() the widget attrs

[James Bennett]

Otherwise it ends up with a shallow copy which reuses the same attrs dict across separate widget instances, playing merry hell with form classes which want to change widget attrs on a per-(form)-instance basis.

[Luke Plant]

I think we need some more analysis of why this patch is correct/needed. Here are some simplified extracts of current code:

class RelatedFieldWidgetWrapper(forms.Widget):
    def __init__(self, widget, rel, admin_site):
        self.attrs = widget.attrs
        self.widget = widget
        # <snip>

    def __deepcopy__(self, memo):
        obj = copy.copy(self)
        obj.widget = copy.deepcopy(self.widget, memo)
        obj.attrs = self.widget.attrs
        memo[id(self)] = obj
        return obj

This means:

wrapper1 = RelatedFieldWidgetWrapper(some_widget, some_rel, some_admin_site)
wrapper2 = copy.deepcopy(wrapper1) # as happens when widget is copied via BaseForm.__init__()

assert wrapper1.attrs is wrapper1.widget.attrs
assert wrapper2.attrs is wrapper2.widget.attrs
assert wrapper1.widget is not wrapper2.widget

That all seems to be correct. The attached patch would actually break the second assert, which I'm pretty sure will break the ability of the class instance to work as a wrapper.

[Aymeric Augustin]

Change UI/UX from NULL to False.

[Aymeric Augustin]

Change Easy pickings from NULL to False.

[Tim Graham]

I think #26213 gives a demonstration of a behavior which this ticket may fix.

[Garvit Sharma]

Claiming — patch ready, PR to follow.

[Garvit Sharma]

PR: ​https://github.com/django/django/pull/21877

[Garvit Sharma]

Yes, I’ve addressed the requested changes. I’ve now changed the triage state to “Ready for checkin” and set “Needs improvement” to false.

[Clifford Gama]

Hi Garvit, please refer to the ​Triage Workflow. RFC is a state marked by a reviewer, not the PR author. Thanks!

[Garvit Sharma]

Replying to Clifford Gama:

Hi Garvit, please refer to the ​Triage Workflow. RFC is a state marked by a reviewer, not the PR author. Thanks!

Thanks Clifford, understood. I'll leave the triage stage for a reviewer.