Avoids reports of bulk_update() sending Cast expressions
to JSONField.get_prep_value().
Co-authored-by: Simon Charette <charette.s@gmail.com>
Backport of 0bf412111b from main.
While the top-level objects fed to serialization are bound to the test
database being created nothing prevents code invoked during
serialization from performing queries against other connections entries
that haven't been swapped yet.
The reported example of that is a database router directing all reads to
a test mirror for a set of models involving auto-created many-to-many
fields. It might be tempting to address the many-to-many field case but
this a symptom of a larger problem where the test framework yields the
flow execution to user code that could interact with non-test databases
in unexpected ways.
Deferring test database fixture serialization until the point where all
connections entries have been swapped for their test equivalent ensures
that no code triggered during serialization can interact with non-test
databases.
Thanks Jake Howard for the report and Jacob Walls for the initial
investigation.
Backport of dc69a63f84 from main
Adding proper support for subquery right-hand-sides to TupleIn made it
obsolete.
Backport of d386405e04 from main
Co-authored-by: Simon Charette <charette.s@gmail.com>
When Expression.__init__() overrides make use of *args, **kwargs
captures their argument values are respectively bound as a tuple and
dict instances. These composite values might themselves contain values
that require special identity treatments such as Concat(output_field)
as it's a Field instance.
Refs #30628 which introduced bound Field differentiation but lacked
argument captures handling.
Thanks erchenstein for the report.
Backport of df2c4952df from main
xgettext only extracts comment blocks if there is no program code between
the comment and the string that gets extracted. For details, see:
https://www.gnu.org/software/gettext/manual/html_node/xgettext-Invocation.html#Operation-mode
Black formatting has been turned off in some places to ensure the
comments are not moved, which previously resulted in them being removed
from the po files when scripts/manage_translations.py was run.
Backport of 6fcd0440aa from main.
Non-tuple exact and in lookups have specialized logic for subqueries that can
be adapted to properly assign select mask if unspecified and ensure the number
of involved members are matching on both side of the operator.
Backport of 41239fe34d from main.
It's necessary to specify a repository for `.pypirc` user configurations
with multiple per-project PyPI tokens.
Follow up to 26aedbbc08.
Backport of 0dc61495b2 from main.
This should allow third-party backends to define Tuple.as_vendor()
overrides that are taken into consideration which calling as_sql()
directly prevents.
Backport of c326cfe3b1 from main
Now that In.get_source_expression() includes its right-hand-side when it
contains expressions (refs #36025) it no longer requires a specialized
get_refs() method.
Backport of 0bac41fc7e from main.
In order for Expression.relabeled_clone to work appropriately its
get_source_expressions method must return all resolvable which wasn't the case
for Lookup when its right-hand-side is "direct" (not a compilable).
While refs #22288 added support for non-literals iterable right-hand-side
lookups it predated the subclassing of Lookup(Expression) refs #27021 which
could have been an opportunity to ensure right-hand-sides are always resolvable
(ValueList and ExpressionList).
Addressing all edge case with non-resolvable right-hand-sides would require
a significant refactor and deprecation of some parts of the Lookup interface so
this patch only focuses on FieldGetDbPrepValueIterableMixin (In and Range
lookups) by making sure that a right-hand-side containing resolvables are dealt
with appropriately during the resolving phase.
Thanks Aashay Amballi for the report.
Backport of 089deb82b9 from main.
The original queryset._next_is_sticky() call never had the intended effect as
no further filtering was applied internally after the pk__in lookup making it
a noop.
In order to be coherent with how related filters are applied when retrieving
objects from a related manager the effects of what calling _next_is_sticky()
prior to applying annotations and filters to the queryset provided for
prefetching are emulated by allowing the reuse of all pre-existing JOINs.
Thanks David Glenck and Thiago Bellini Ribeiro for the detailed reports and
tests.
Backport of 2598b371a9 from main.
The get_(local|foreign)_related_value methods of GenericRelation must be
reversed because it defines (from|to)_fields and associated related_fields
in the reversed order as it's effectively a reverse GenericForeignKey
itself.
The related value methods must also account for the fact that referenced
primary key values might be stored as a string on the model defining the
GenericForeignKey but as integer on the model defining the GenericRelation.
This is achieved by calling the to_python method of the involved content type
in get_foreign_related_value just like GenericRelatedObjectManager does.
Lastly reverse many-to-one manager's prefetch_related_querysets should use
set_cached_value instead of direct attribute assignment as direct assignment
might are disallowed on ReverseManyToOneDescriptor descriptors. This is likely
something that was missed in f5233dc (refs #32511) when the is_cached guard
was added.
Thanks 1xinghuan for the report.
Backport of 198b30168d from main.
The `make black` target in the docs directory used Linux-specific
syntax for its `find` command. Changed to syntax that also works
on macOS and other BSD Unix variants.
Backport of 248d8457cb from main.
