Avoided introducing a new regex-based SQL splitter in the migrations
framework, before we're bound by backwards compatibility.
Adapted this change to the legacy "initial SQL data" feature, even
though it's already deprecated, in order to facilitate the transition
to migrations.
sqlparse becomes mandatory for RunSQL on some databases (all but
PostgreSQL). There's no API to provide a single statement and tell
Django not to attempt splitting. Since we have a more robust splitting
implementation, that seems like a good tradeoff. It's easier to add a
new keyword argument later if necessary than to remove one.
Many people contributed to both tickets, thank you all, and especially
Claude for the review.
Refs #22401.
Changed the migration autodetector to remove models last so that FK
and M2M fields will not be left as dangling references. Added a check
in the migration state renderer to error out in the presence of
dangling references instead of leaving them as strings. Fixed a bug
in the sqlite backend to handle the deletion of M2M fields with
"through" models properly (i.e., do nothing successfully).
Thanks to melinath for report, loic for tests and andrewgodwin and
charettes for assistance with architecture.
Changed the migration autodetector to remove models last so that FK
and M2M fields will not be left as dangling references. Added a check
in the migration state renderer to error out in the presence of
dangling references instead of leaving them as strings. Fixed a bug
in the sqlite backend to handle the deletion of M2M fields with
"through" models properly (i.e., do nothing successfully).
Thanks to melinath for report, loic for tests and andrewgodwin and
charettes for assistance with architecture.
This commit reverts 69d4b1c and tackle the issue from a different angle.
Models remain present in the project state, but are now ignored by the
autodetector.
Added reversible property to RunPython so that migrations will not
refuse to reverse migrations including RunPython operations, so long as
reverse_code is set in the RunPython constructor. Included tests to
check the reversible property on RunPython and the similar RunSQL.
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()'.
This is the result of Christopher Medrela's 2013 Summer of Code project.
Thanks also to Preston Holmes, Tim Graham, Anssi Kääriäinen, Florian
Apolloner, and Alex Gaynor for review notes along the way.
Also: Fixes#8579, fixes#3055, fixes#19844.
Adjusted several tests that used it to add apps to the app cache and
then attempted to remove them by manipulating attributes directly.
Also renamed invalid_models to invalid_models_tests to avoid clashing
application labels between the outer and the inner invalid_models
applications.
Got rid of AppConfig._stub. As a side effect, app_cache.app_configs now
only contains entries for applications that are in INSTALLED_APPS, which
is a good thing and will allow dramatic simplifications (which I will
perform in the next commit). That required adjusting all methods that
iterate on app_configs without checking the "installed" flag, hence the
large changes in get_model[s].
Introduced AppCache.all_models to store models:
- while the app cache is being populated and a suitable app config
object to register models isn't available yet;
- for applications that aren't in INSTALLED_APPS since they don't have
an app config any longer.
Replaced get_model(seed_cache=False) by registered_model() which can be
kept simple and safe to call at any time, and removed the seed_cache
argument to get_model[s]. There's no replacement for that private API.
Allowed non-master app caches to go through populate() as it is now
safe to do so. They were introduced in 1.7 so backwards compatibility
isn't a concern as long as the migrations framework keeps working.
Improved Andrew's hack to create temporary app caches to handle
migrations. Now the main app cache has a "master" flag set to True
(which is a non-default keyword argument, thus unlikely to be used by
mistake). Other app cache instances have "master" set to False.
The only sanctioned way to access the app cache is by importing
django.core.apps.app_cache.
If you were instanciating an app cache and relying on the Borg pattern,
you'll have to refactor your code.
Added comments in the three empty models.py files that are still needed.
Adjusted the test runner to add applications corresponding to test
labels to INSTALLED_APPS even when they don't have a models module.
This commit is a refactoring with no change of functionality, according
to the following invariants:
- An app_label that was in app_configs and app_models stays in
app_config and has its 'installed' attribute set to True.
- An app_label that was in app_models but not in app_configs is added to
app_configs and has its 'installed' attribute set to True.
As a consequence, all the code that iterated on app_configs is modified
to check for the 'installed' attribute. Code that iterated on app_models
is rewritten in terms of app_configs.
Many tests that stored and restored the state of the app cache were
updated.
In the long term, we should reconsider the usefulness of allowing
importing models from non-installed applications. This doesn't sound
particularly useful, can be a trap in some circumstances, and causes
significant complexity in sensitive areas of Django.
Since the original ones in django.db.models.loading were kept only for
backwards compatibility, there's no need to recreate them. However, many
internals of Django still relied on them.
They were also imported in django.db.models. They never appear in the
documentation, except a quick mention of get_models and get_app in the
1.2 release notes to document an edge case in GIS. I don't think that
makes them a public API.
This commit doesn't change the overall amount of global state but
clarifies that it's tied to the app_cache object instead of hiding it
behind half a dozen functions.
We need to make sure content read from the file is decoded from UTF-8
right from the start so Python doesn't try to use another encoding
(read: ASCII/CP1252 under Windows.)
