Opened 10 years ago
Closed 10 years ago
#25579 closed Bug (fixed)
Lack of type adaptation in ArrayField querying/lookups
| Reported by: | Matt C | Owned by: | Matt C |
|---|---|---|---|
| Component: | contrib.postgres | Version: | 1.8 |
| Severity: | Normal | Keywords: | ArrayField query lookup |
| Cc: | Triage Stage: | Accepted | |
| Has patch: | yes | Needs documentation: | no |
| Needs tests: | no | Patch needs improvement: | no |
| Easy pickings: | no | UI/UX: | no |
Description (last modified by )
This commit: https://github.com/django/django/commit/39d95fb6ada99c59d47fa0eae6d3128abafe2d58 replaces the get_prep_value method with get_db_prep_value and I believe this causes a regression with regards to querying on an ArrayField.
get_db_prep_value is never called during querying/lookup, hence the array items/values are passed directly to the low-level database Python wrappers without first being transformed (from Python objects to SQL-friendly parameters).
See this question: http://stackoverflow.com/questions/33250371/arrayfield-class-missing-query-lookup-methods
Could both the get_prep_value and get_db_prep_value methods be overridden with duplicate/shared logic to overcome this issue?
Patch pull request: https://github.com/django/django/pull/6278
Change History (13)
comment:1 by , 10 years ago
comment:2 by , 10 years ago
#25143 is to do with converting DB query result values to Python values, whilst this is to do with converting Python values to DB values, for the purpose of lookups/querying.
I'll change the title to remove the word regression.
comment:3 by , 10 years ago
| Summary: | ArrayField query/lookup regression → Lack of type adaptation in ArrayField querying/lookups |
|---|
comment:4 by , 10 years ago
| Triage Stage: | Unreviewed → Accepted |
|---|
I can reproduce "can't adapt type 'Tag'" by adding list(OtherTypesArrayModel.objects.filter(tags=Tag(1))) to the PR from #25143.
comment:5 by , 10 years ago
I've created a pull request for this, but I'm not sure if I need to change any of the configuration (e.g. "Has patch" and whether I should take ownership) for this ticket, or if that's done by the core devs.
comment:6 by , 10 years ago
Please assign this ticket to yourself to prevent parallel work, check Has Patch and preferably link to your PR.
The only thing you can't do is mark your own patch as Ready for checkin.
comment:7 by , 10 years ago
| Has patch: | set |
|---|---|
| Owner: | set to |
| Status: | new → assigned |
Thanks @charettes.
I wasn't sure if linking to the pull request was necessary, because it seemed to happen automatically in the details box at the top, but here is the link anyway:
comment:8 by , 10 years ago
| Description: | modified (diff) |
|---|
comment:9 by , 10 years ago
| Patch needs improvement: | set |
|---|
Left comments for improvement on the PR. Please uncheck "patch needs improvement" when you update it. Thanks!
comment:10 by , 10 years ago
With the introduction of this commit: https://github.com/django/django/commit/2495023a4cae28f494d0a6172abfac3a47a0b816
...I realised that the solution may lie at a higher level.
The tests.postgres_tests.models.TagField class has overridden the get_db_prep_value() method, with the only change being that it ignores the prepared keyword-arg and always calls get_prep_value() on the input value.
This was introduced because the author of that test code would have encountered the type adaptation issue which is what this ticket is about.
So I can't use the TagField class to write tests for a solution to this ticket, because the problem is solved by that class.
I.e. Take away the TagField's overriding implementation of get_db_prep_value() and the problem of this ticket arises.
comment:12 by , 10 years ago
| Patch needs improvement: | unset |
|---|
Deleted TagField.get_db_prep_value(), but didn't add any new tests, because one of the tests from https://github.com/django/django/commit/2495023a4cae28f494d0a6172abfac3a47a0b816 failed after deleting that method (as I said it would).
I also took a different approach to solving it, re-using existing code as much as possible.
Looks similar to #25143 but I think it's a separate issue. I'm not sure about the characterization of the issue as a regression since
ArrayFieldwas never present in a stable release without the commit you pointed out, but it might be okay to backport anyway under the "bug in a new feature" rationale.