Thanks Zach Smith for the patch.

Thanks keturn for the reporti, Chris Medrela for details and
Tim Graham for the review.
Refs #19915.

Overriding the error messages now works for both unique fields, unique_together
and unique_for_date.

This patch changed the overriding logic to allow customizing NON_FIELD_ERRORS
since previously only fields' errors were customizable.

Refs #20199.

Thanks leahculver for the suggestion.

Thanks Aymeric Augustin and Simon Charette for reviews.

MySQL does allow primary key with value 0. It only forbids autoincrement
primary key with value 0.

Thanks Claude Paroz for the report.

Thanks jordan at aace.org for the suggestion.

Refs #20625.

This reverts commit 2ee447fb.

That commit introduced a regression (#21882) and didn't really
do what it was supposed to: while it did delay the evaluation
of lazy objects passed to mark_safe(), they weren't actually
marked as such so they could end up being escaped twice.

Refs #21882.

Refs #21952

Thanks Michael Manfre for the report.

Refs #21911. Now that we have a more traditional login form, we
don't need any more a special field telling us we are dealing with
the login form.

Historically, the Django admin used to pass through the request
from an unauthorized access to the login view directly. Now we
are using a proper redirection, which is also preventing
inadvertantly changing data when POSTing login data to an admin
view when user is already authorized.
Thanks Marc Tamlyn and Tim Graham for the reviews.

Refs #21911.

Made sure Django treats case .filter(NOT (a AND b)) the same way as
.filter((NOT a OR NOT b)) for join promotion.

Thanks Siecje.

Fix regress added to migrations.test_operations.test_alter_field_pk_fk

Fixed regression introduced by fabc678f. Thanks Michael Manfre
for the report.

Fixed #21751 -- Explicitly closed cursors.

This commit touchs various parts of the code base and test framework. Any
found usage of opening a cursor for the sake of initializing a connection
has been replaced with 'ensure_connection()'.

Updated SQLUpdateCompiler.execute_sql to match the behavior described in
the docstring; the 'first non-empty query' will now include all queries,
not just the main and first related update.

Added CURSOR and NO_RESULTS result_type constants to make the usages more
self documenting and allow execute_sql to explicitly close the cursor when
it is no longer needed.

Thanks Russell for the advice.