#22288 closed Bug (fixed)
F() expression not compatible with __range field look up
Reported by: | Owned by: | Matthew Wilkes | |
---|---|---|---|
Component: | Database layer (models, ORM) | Version: | dev |
Severity: | Normal | Keywords: | |
Cc: | github@…, josh.smeaton@…, matthew@… | Triage Stage: | Accepted |
Has patch: | yes | Needs documentation: | no |
Needs tests: | no | Patch needs improvement: | yes |
Easy pickings: | no | UI/UX: | no |
Description (last modified by )
class TestModel(models.Model): a = models.SmallIntegerField() b = models.SmallIntegerField() TestModel.objects.filter(a__range=(F('b')-1, F('b')+1) TypeError: int() argument must be a string or a number, not 'ExpressionNode'
Change History (16)
comment:2 by , 11 years ago
Triage Stage: | Unreviewed → Accepted |
---|---|
Version: | 1.5 → master |
comment:3 by , 11 years ago
Has patch: | set |
---|
I've had a go at fixing this, the commit is here: https://github.com/timmartin/django/commit/9ea32634efda8f4a74ef9ee9a742a1437bc27ca0
I'd like feedback on this fix, since I'm not sure I fully understand the potential wider impact of the changes I've made (though it passes all regression tests). It also looks like the changes in the pull request for #14030 will interact with this.
comment:4 by , 10 years ago
Patch needs improvement: | set |
---|
I checked the patch. It seems we need changes to too many places just to support something__range=(F('foo'), F('bar'))
. The main issue is that we have many rounds of value preparation.
The main issue here is that we just don't handle the case here value is an iterable correctly for value preparation. While the above mentioned commit touches the area around range lookups, we should likely fix all the lookups to lists of F() objects properly. Even if the only other example in core is somefield__in=[F()]
.
It might be better to wait for #14030 before applying this one.
In addition the tests of the above patch aren't correct, the code checks generated SQL, but that SQL differs between backends (quote character is not always "
, and some backends IIRC use upper case table names).
Sorry for delay in commenting this issue.
comment:6 by , 9 years ago
I think a side effect of https://github.com/django/django/pull/5090/ is that range and in lookups will now understand expressions properly. I'm not certain of this though, it just looks like similar code paths.
https://github.com/django/django/pull/5090/files#diff-b6b218ec29b7fb6a7d89868a94bfc73eR61
In fact, I'll incorporate the test case above into this patch to check.
comment:8 by , 9 years ago
Description: | modified (diff) |
---|
comment:9 by , 9 years ago
I have had the same problem, albeit with date objects. I've begun working on a patch, WIP is at https://github.com/django/django/compare/master...MatthewWilkes:22288-iterables-and-F-objects
Current state is 1 failing test on sqlite (not tested on other backend) so certainly not ready for review at the moment. Posting here for visibility, as I've spoken to many of the people involved on IRC.
While working on this I've started to think that checking for expected parameter types in lookups being done by comparing their names is a bad idea, given the work done on custom lookups. Perhaps it would make sense for some of the value preparation to be offloaded onto the lookup, or at least the parsing of value data structures, to allow for richer lookups to use expressions too.
EDIT: On a practical note, I didn't intentionally unset the docs and tests parameters, they were unset on load and checking them changed the message to say that I'd changed it to checked. I guess they are newish fields?
comment:10 by , 9 years ago
Cc: | added |
---|
comment:11 by , 9 years ago
Cc: | added |
---|
I've added a WIP pull request at https://github.com/django/django/pull/6216
Any feedback on what additional testing would be required as well as on my general approach would be very much appreciated.
comment:13 by , 8 years ago
I have updated the PR at https://github.com/django/django/pull/6216 - I believe this is in a good state but will need re-reviewing by ORM experts.
Is this something that should be against 1.10 or 1.11 at this stage?
comment:14 by , 8 years ago
Owner: | changed from | to
---|---|
Patch needs improvement: | unset |
Status: | new → assigned |
Triage Stage: | Accepted → Ready for checkin |
The PR is against 1.11 and ready for final review. Setting to ready for check-in following jarshwah's instructions on GitHub.
comment:15 by , 8 years ago
Patch needs improvement: | set |
---|---|
Triage Stage: | Ready for checkin → Accepted |
I left a few comments for improvement.
Hi,
I can reproduce this issue on latest master as well as on 1.6 so it doesn't appear to be related to the new custom lookup feature.
Regarding your formatting issues, you can make code blocks on trac using
{{{ code goes here }}}
, as explained on the WikiFormatting page.Thanks.