The __isnull lookup of JSONField must special case
Value(None, JSONField()) left-hand-side in order to be coherent with
its convoluted null handling.
Since psycopg>=3 offers no way to pass a NULL::jsonb the issue is
resolved by optimizing IsNull(Value(None), True | False) to
True | False.
Regression in 5c23d9f0c32f166c81ecb6f3f01d5077a6084318.
Thanks Alexandre Collet for the report.
Special characters lose their special meaning inside sets of characters.
"-" lose its special meaning if it's placed as the first or last
character.
Follow up to 7c6b66383da5f9a67142334cd2ed2d769739e8f1.
While values(*field_excluding_pk).distinct() and
distinct(*field_excluding_pk) can reduce the number of resulting rows
in a way that makes subsequent delete() calls ambiguous standalone
.distinct() calls cannot.
Since delete() already disallows chain usages with values() the only
case that needs to be handled, as originally reported, is when
DISTINCT ON is used via distinct(*fields).
Refs #32682 which had to resort to subqueries to prevent duplicates in
the admin and caused significant performance regressions on MySQL
(refs #34639).
This partly reverts 6307c3f1a123f5975c73b231e8ac4f115fd72c0d.
This avoids reading the image size when the dimensions fields
(image_width, image_height) do not exist, as that operation may be
expensive.
Partially reverts ea53e7c09f1b8864c20c65976bbeaeab77abdaec, that dropped
the check for the dimension fields in update_dimension_fields(), because
the post_init signal was no longer registered without dimension fields.
However, another code path to that function exists: when the
ImageFileField is save()d, the name from the storage is setattr()ed on
the field, and ImageFileDescriptor calls update_dimension_fields()
because the image size might have changed. Keep bailing out early when
dimensions are unused.
Besides, computing the image dimensions causes to close() the file,
resulting in a backward-incompatible change. The test protects against
that change.
An n-ary logical XOR Q(…) ^ Q(…) ^ … ^ Q(…) should evaluate to true
when an odd number of its operands evaluate to true, not when exactly
one operand evaluates to true.
Regression in b3db6c8dcb5145f7d45eff517bcd96460475c879.
Thanks Ian Cubitt for the report.
This also corrected test_inheritance_deferred2() test which was
previously properly defined and marked as an expected failure but was
then wrongly adjusted to mask the lack of support for per-alias
deferral that was fixed by #21204.
While deferring many-to-many and GFK has no effect, the previous
implementation of QuerySet.defer() ignore them instead of crashing.
Regression in b3db6c8dcb5145f7d45eff517bcd96460475c879.
Thanks Paco Martínez for the report.
Special thanks to Hannes Ljungberg for finding multiple implementation
gaps.
Thanks also to Simon Charette, Adam Johnson, and Mariusz Felisiak for
reviews.
Proper escaping of % in string literals used when defining constaints
was attempted (a8b3f96f6) by overriding quote_value of Postgres and
Oracle schema editor. The same approach was used when adding support for
constraints to the MySQL/MariaDB backend (1fc2c70).
Later on it was discovered that this approach was not appropriate and
that a preferable one was to pass params=None when executing the
constraint creation DDL to avoid any form of interpolation in the first
place (42e8cf47).
When the second patch was applied the corrective of the first were not
removed which caused % literals to be unnecessary doubled. This flew
under the radar because the existings test were crafted in a way that
consecutive %% didn't catch regressions.
This commit introduces an extra test for __exact lookups which
highlights more adequately % doubling problems but also adjust a
previous __endswith test to cover % doubling problems (%\% -> %%\%%).
Thanks Thomas Kolar for the report.
Refs #32369, #30408, #30593.
