Over the years we've had multiple instances of hit and misses when
emitting warnings: either setting the wrong stacklevel or not setting
it at all.
This work adds assertions for the existing warnings that were declaring
the correct stacklevel, but were lacking tests for it.
Refs #34429: Following the implementation allowing the setting of
unusable passwords via the admin site, the `BaseUserCreationForm` and
`UserCreationForm` were extended to include a new field for choosing
whether password-based authentication for the new user should be enabled
or disabled at creation time.
Given that these forms are designed to be extended when implementing
custom user models, this branch ensures that this new field is moved to
a new, admin-dedicated, user creation form `AdminUserCreationForm`.
Regression in e626716c28.
Thanks Simon Willison for the report, Fabian Braun and Sarah Boyce for
the review.
This work also allows to subclass BaseUserCreationFormTest to reuse the
tests and assertions for testing forms that extend BaseUserCreationForm,
which is now used for UserCreationFormTest, increasing its coverage.
This also caused un-ordered sliced prefetches to crash as they rely on Window.
Regression in e16d0c176e that made OrderByList
piggy-back ExpressionList without porting the empty handling that the latter
provided.
Supporting explicit empty ordering on Window functions and slicing is arguably
a foot-gun design due to how backends will return undeterministic results but
this is a problem that requires a larger discussion.
Refs #35064.
Thanks Andrew Backer for the report and Mariusz for the review.
Aggregation optimization didn't account for not referenced set-returning annotations on Postgres.
Co-authored-by: Simon Charette <charette.s@gmail.com>
This is a follow up for the fix of CVE-2024-39329
(5d86458579) where the timing of
verify_password() was standardized when checking unusable passwords.
Regression in b0ad41198b.
Refs #34013. The initial logic did not consider that annotation aliases
can include lookup or transform separators.
Thanks Gert Van Gool for the report and Mariusz Felisiak for the review.
LocaleMiddleware didn't handle the ValueError raised by
get_supported_language_variant() when language codes were
over 500 characters.
Regression in 9e9792228a.
This is the exact same issue as refs #30408 but for creating a model with a
constraint containing % escapes instead of column addition. All of these issues
stem from a lack of SQL and parameters separation from the BaseConstraint DDL
generating methods preventing them from being mixed with other parts of the
schema alteration logic that do make use of parametrization on some backends
(e.g. Postgres, MySQL for DEFAULT).
Prior to the addition of Field.db_default and GeneratedField in 5.0
parametrization of DDL was never exercised on model creation so this is
effectively a bug with db_default as the GeneratedField case was addressed by
refs #35336.
Thanks Julien Chaumont for the report and Mariusz Felisiak for the review.
Partially reverts 0b33a3abc2.
Storage.exists(name) was documented to "return False if
the name is available for a new file." but return True if
the file exists. This is ambiguous in the overwrite file
case. It will now always return whether the file exists.
Thank you to Natalia Bidart and Josh Schneier for the
review.
While refs #34125 focused on the SQL correctness of slicing of union of
potentially empty queries it missed an optimization opportunity to avoid
performing a query at all when all queries are empty.
Thanks Lucidiot for the report.
The original tests required the creation of a model that is no longer necessary
and were exercising Model.full_clean(validate_constraints) which has nothing
to do with the nulls_distinct feature.
The logic allowing UniqueConstraint(fields).validate to preserve backward
compatiblity with Model.unique_error_message failed to account for cases where
the constraint might not be attached to a model which is a common pattern
during testing.
This changes allows for arbitrary UniqueConstraint(fields) to be tested in
isolation without requiring actual models backing them up.
Co-authored-by: Mark G <mark.gensler@protonmail.com>
Fixed a regression which would cause multiple To, Cc, and
Reply-To headers in the result of EmailMessage.message() if
values were supplied for both to/cc/reply_to and the
corresponding extra_headers fields.
Updated related tests to check the generated message() has
exactly one of each expected header using get_all().
Regression in b03d500295.
Previously the order was always extra_fields + model_fields + annotations with
respective local ordering inferred from the insertion order of *selected.
This commits introduces a new `Query.selected` propery that keeps tracks of the
global select order as specified by on values assignment. This is crucial
feature to allow the combination of queries mixing annotations and table
references.
It also allows the removal of the re-ordering shenanigans perform by
ValuesListIterable in order to re-map the tuples returned from the database
backend to the order specified by values_list() as they'll be in the right
order at query compilation time.
Refs #28553 as the initially reported issue that was only partially fixed
for annotations by d6b6e5d0fd.
Thanks Mariusz Felisiak and Sarah Boyce for review.
The transitional logic added to deprecate the usage of *args for
Model.save()/asave() introduced two issues that this branch fixes:
* Passing extra positional arguments no longer raised TypeError.
* Passing a positional but empty update_fields would save all fields.
Co-authored-by: Natalia <124304+nessita@users.noreply.github.com>
This avoids passing "blank=False" and "srid=4326" to field classes,
which are the default values, and removes special treatment for the
"blank" parameter.
Applied the ProhibitNullCharactersValidator to ModelChoiceField and ModelMultipleChoiceField.
Co-authored-by: Viktor Paripás <viktor.paripas@gmail.com>
Co-authored-by: Vasyl Dizhak <vasyl@dizhak.com>
Co-authored-by: Arthur Vasconcelos <vasconcelos.arthur@gmail.com>
The auth forms using SetPasswordMixin were incorrectly including the
'This field is required.' error when additional validations (e.g.,
overriding `clean_password1`) were performed and failed.
This fix ensures accurate error reporting for password fields.
Co-authored-by: Natalia <124304+nessita@users.noreply.github.com>
This work should not generate any change of functionality, and
`urlsplit` is approximately 6x faster.
Most use cases of `urlparse` didn't touch the path, so they can be
converted to `urlsplit` without any issue. Most of those which do use
`.path`, simply parse the URL, mutate the querystring, then put them
back together, which is also fine (so long as urlunsplit is used).
Sleep call leads to a hard to trace error in CI. Using an Event is
more deterministic, and should be less prone to environment
variations.
Bug in 11393ab131.
Previously, `-1` was converted to `"-1th"`. This has been updated to
return negative numbers "as is", so that for example `-1` is
converted to `"-1"`. This is now explicit in the docs.
Co-authored-by: Martin Jonson <artin.onson@gmail.com>
Co-authored-by: Adam Johnson <me@adamj.eu>
Co-authored-by: Mehmet İnce <mehmet@mehmetince.net>
Co-authored-by: Sarah Boyce <42296566+sarahboyce@users.noreply.github.com>
This work improves the accessibility of the add and change pages in the
admin site by adding <details> and <summary> elements to the collapsible
fieldsets. This has the nice side effect of no longer requiring custom
JavaScript helpers to implement the fieldsets' show/hide capabilities.
Thanks to James Scholes for the accessibility advice, and to Sarah Boyce
and Tom Carrick for reviews.
Co-authored-by: Natalia <124304+nessita@users.noreply.github.com>
Co-authored-by: Sarah Boyce <42296566+sarahboyce@users.noreply.github.com>
Before this change, HTML <fieldset> elements in the admin site did not
have an associated label to describe them. This commit defines a unique
HTML id for the heading labeling a fieldset, and sets its
aria-labelledby property to link the heading with the fieldset.
FieldCacheMixin is used by related fields to track their cached values.
This work migrates get_cache_name() to be a cached property to optimize
performance by reducing unnecessary function calls when working with
related fields, given that its value remains constant.
Co-authored-by: Simon Charette <charette.s@gmail.com>
Co-authored-by: Sarah Boyce <42296566+sarahboyce@users.noreply.github.com>
Co-authored-by: Natalia <124304+nessita@users.noreply.github.com>
This also introduces Expression.constraint_validation_compatible that
allows specifying that expression should be ignored during a constraint
validation.
Hidden elements are not visible for both accessibility tools and browsers presentation layer. This change therefore only reduces the size of the generated HTML.
Updated OrderableAggMixin.as_sql() to separate the order_by parameters
from the filter parameters. Previously, the parameters and SQL were
calculated by the Aggregate parent class, resulting in a mixture of
order_by and filter parameters.
Thanks Simon Charette for the review.
Refactored the filter and order_by expressions in the Aggregate class to
return a list of Expression (or None) values, ensuring that the list
item is always available and represents the filter expression.
For the PostgreSQL OrderableAggMixin, the returned list will always
include the filter and the order_by value as the last two elements.
Lastly, emtpy Q objects passed directly into aggregate objects using
Aggregate.filter in admin facets are filtered out when resolving the
expression to avoid errors in get_refs().
Thanks Simon Charette for the review.
Keep consistent behaviour of slice() filter between python 3.12 and prior
versions in the case of a dict passed to the filter (catch the new to python
3.12 KeyError exception).
While refs #34612 surfaced issues with reverse one-to-one fields
deferrals, it missed that switching to storing remote fields would break
self-referential relationships.
This change switches to storing related objects in the select mask
instead of remote fields to prevent collisions when dealing with
self-referential relationships that might have a different directional
mask.
Despite fixing #21204 introduced a crash under some self-referential
deferral conditions, it was simply not working even before that as it
aggregated the sets of deferred fields by model.
Thanks Joshua van Besouw for the report and Mariusz Felisiak for the
review.
Generated fields have to be excluded from the INSERT query against the remade
table including the index.
Thanks Moshe Dicker for the report, David Sanders and Mariusz Felisiak for the
review.
Postgres >= 12.18, 13.14, 14.11, 15.6, 16.2 changed the way the immutability
of generated and default expressions is detected in postgres/postgres@743ddaf.
The adjusted test semantic is presereved by switching from __icontains to
__contains as both make use of a `%` literal which requires proper escaping.
Refs #35336.
Thanks bcail for the report.
Following the ASGI HTTP Connection Scope docs[0], the provided `path`
is already the correct value that Django requires.
In combination with `root_path`, from which `script_name` is derived,
the `path_info` variable is set. It's then redundant to
re-calculate `path` from `script_name` and `path_info`.
See also, a clarifying discussion on the ASGIref repo[1].
[0]: https://asgi.readthedocs.io/en/latest/specs/www.html#http-connection-scope
[1]: https://github.com/django/asgiref/issues/424
A longer term solution is likely to have a better separation of parametrized
DDL altogether to handle checks, constraints, defaults, and generated fields
but such a change would require a significant refactor that isn't suitable
for a backport.
Thanks Adrian Garcia for the report.
We already use implementation_name and python_version markers
so it's more consistent. Moreover, marker values can be specified
via pip's command line options.
