Admin list_filter doesn't respect null=True

[StevenPotter]

Maybe it is just me, but it seems if a models field has the properties null=True, blank=True and a list filter is applied to that field one of the filter options should be "Is Null"

[Bob Thomas]

Add an "Is null" filter for foreign keys and value fields

[Bob Thomas]

Boolean filters already have an "Unknown" option if the model field is a NullBooleanField. I added null options for RelatedFilterSpec and AllValuesFilterSpec, though. Fixes my semi-duplicate ticket #9177

[(none)]

Milestone post-1.0 deleted

[Karen Tracey]

#10665 was a dup.

[Ben Davis]

I vote that instead of "Is null" the filter says "(None)". "(None)" would seem to make more sense to the lay-person, not just because "null" is sql jargon, but also because the admin already displays "(None)" in the changelist for null foreign keys.

Also, I've attached an updated patch for rev. 11627

[marcob]

Patch for Django 1.1.1 and that uses:

 "(%s)" % _("None") 

[Ben Davis]

Just FYI: this patches cleanly against 1.2-beta.

[arnaudr]

Patches cleanly against 1.2

[marcob]

Trying tests I got:

   TemplateSyntaxError: Caught AttributeError while rendering: 'RelatedObject' object has no attribute 'null'

Are you sure about that self.field.null in filterspecs.py ?

[Simon Litchfield]

#5273 was another dupe. 4+ tickets all up. Latest patch looks good. Be great to get this simple fix in for 1.3.

[Simon Litchfield]

See also #5833

[oyvind]

Patch from #14467 with tests

[Julien Phalip]

Pointing out that oyvind's latest patch isn't applying to trunk any more since r14674. Pinging DrMeers :)

[marcob]

A newer patch I apply after the #3400 one (don't know if works with trunk, give it a try)

[Julien Phalip]

I'm working on a patch. Hopefully we can ship this one in 1.3.

[Julien Phalip]

OK, I've written a patch based on marcob's, with some small tweaks. In particular I'm using EMPTY_CHANGELIST_VALUE (instead of _('None')) for consistency with the way NULL values are displayed in readonly fields and in the changelist, and also to dissociate it from the 'All' filter which has a different purpose.

The tests are based on oyvind's patch and I've also added tests for the FK and M2M filters.

Let me know what you think. I'm particularly wondering about the condition "if hasattr(self.field, 'rel')" which I had to add to avoid breaking the existing admin views tests.

[Simon Meers]

Looks good Julien. Only minor comments:

The if hasattr(self.field, 'rel') prevents dealing with a RelatedObject, which is actually not a field but a reverse relationship. I think it makes sense not to allow isnull filtering on these?

Any reason you've cast self.lookup_val_isnull to bool in RelatedFilterSpec but not in AllValuesFilterSpec?

You need an __init__.py in tests/regressiontests/admin_filterspecs/.

It might be a good idea to include a reverse relationship in the tests (e.g. a Chapter model with a ForeignKey to Book, and then test a filter in both directions)? As mentioned above, I'm not sure that an isnull lookup on a reverse relationship makes sense, but it would be nice to explicitly ensure that nothing breaks here.

In the tests, would choices[-1] be more appropriate than choices[3] and choices[4] for testing the "last choice"?

Lines in tests.py and filterspecs.py exceed PEP8's 79-character limit.

Thanks for the great work, hopefully we can get this RFC today.

[Julien Phalip]

The genuine new patch is "8528_filterspec_null_v2.2.diff" above. Please ignore "8528_filterspec_null_v2.diff".

Thanks for your great feedback, Simon.

Replying to DrMeers:

The if hasattr(self.field, 'rel') prevents dealing with a RelatedObject, which is actually not a field but a reverse relationship. I think it makes sense not to allow isnull filtering on these?

In fact, it does make sense to allow it to work with reverse relationships. For example, in the User admin, you might want to filter all users who are not authors of any books. I have changed the condition to if isinstance(self.field, models.related.RelatedObject) and self.field.field.null or hasattr(self.field, 'rel') and self.field.null. All seems to work fine now.

Any reason you've cast self.lookup_val_isnull to bool in RelatedFilterSpec but not in AllValuesFilterSpec?

Yeah that was a bit silly. The casting should occur when yiedling the resulting dictionary. Fixed in new patch.

You need an __init__.py in tests/regressiontests/admin_filterspecs/.

Done.

It might be a good idea to include a reverse relationship in the tests (e.g. a Chapter model with a ForeignKey to Book, and then test a filter in both directions)? As mentioned above, I'm not sure that an isnull lookup on a reverse relationship makes sense, but it would be nice to explicitly ensure that nothing breaks here.

Good idea, I've added such tests extending UserAdmin.

In the tests, would choices[-1] be more appropriate than choices[3] and choices[4] for testing the "last choice"?

Done. Good suggestion.

Lines in tests.py and filterspecs.py exceed PEP8's 79-character limit.

I've PEP8'ed it up ;)

Any further feedback welcome. Thanks! ;)

[Simon Meers]

Great work, thanks Julien. The only thing that caught my eye was the implicit boolean logic grouping when checking for RelatedObjects, but I don't think this makes any functional difference. Also for some reason patch still isn't creating the admin_filterspecs/__init__.py file, but I can see it in your .diff, so I'll assume that's a problem at my end?

[Russell Keith-Magee]

Two quick notes for the benefit of posterity:

1. I've made one very small addition to this patch as provided -- I've modified the has_output() method on RelatedValueFilterSpec, to allow for the fact that "none" should be included in the count of available values if the related field allows nulls. This caused me a bit of confusion in my simple manual test case because I only defined a single related object, and the filters didn't display -- because according to the filterspec, there was only one valid value, and a filter isn't required if there's only one value.

2. I'm not going to backport this to 1.2.X. The bug obviously exists in 1.2.X, but it's the new FilterSpec framework that gives us the flexibility to fix the bug easily; a backport would be a major PITA.

[Russell Keith-Magee]

In [15649]:

Fixed #8528 -- Ensure that null values are displayed as a filtering option in the admin if a field allows nulls. Thanks to StevenPotter for the report, and oyvind, marcob, Simon Meers and Julien Phalip for the patch.

[Jacob]

Milestone 1.3 deleted