This (nearly) completes the work to isolate all the test modules from
each other. This is now more important as importing models from another
module will case PendingDeprecationWarnings if those modules are not in
INSTALLED_APPS. The only remaining obvious dependencies are:
- d.c.auth depends on d.c.admin (because of the is_admin flag to some
views), but this is not so important and d.c.admin is in
always_installed_apps
- test_client_regress depends on test_client. Eventually these should
become a single module, as the split serves no useful purpose.
Python 3.5 will change the default value of convert_charrefs, so 3.4
gives warnings if it's not present. This is slightly technical as 2.7
doesn't have the kwarg. Thankfully, we already have a bunch of
workarounds for different versions.
Since the app registry is always populated before the first request is
processed, the situation described in #18251 for the old app cache
cannot happen any more.
Refs #18251, #21628.
Also ensured the transaction state is clean on Oracle while I was there.
This change cannot be backported to 1.6 because it's
backwards-incompatible for custom database backends.
When settings.DATABASES['default']['AUTOCOMMIT'] = False, the connection
wasn't in autocommit mode but Django pretended it was.
Thanks Anssi for analysing this issue.
Refs #17062.
Validating STATIC_ROOT in StaticFilesStorage.__init__ turned out to be
problematic - especially with tests - because the storage refuses to work even
if there are no actual interactions with the file system, which is backward
incompatible.
Originally the validation happened in the StaticFilesStorage.path method, but
that didn't work as expected because the call to FileSystemStorage.__init__
replaced the empty value by a valid path. The new approach is to move back the
check to the StaticFilesStorage.path method, but ensure that the location
attribute remains None after the call to super.
Refs #21581.
Originating WSGIRequests are now attached to the ``wsgi_request`` attribute of
the ``HttpResponse`` returned by the testing client.
Thanks rvdrijst for the suggestion.
Make use of `weakref.finalize` and `weakref.WeakMethod` on python 3.4.
Simplified the removal of receivers, the old function looked overly
complicated.
Many thanks go to Antoine Pitrou for helping me to debug and explain all
the failures I ran into while writing that patch.
Models are now attached to any application they're defined in. Since
not_installed was inside app_loading, these models were mistakenly
attached to app_loading. The test that used them passed accidentally
when run after EggLoadingTest because that class' tearDown() method
replaces apps.all_models['app_loading'] by a copy of itself, while
it should remain the same as apps.app_configs['app_loading'].models.
Surprisingly, this commit doesn't change any behavior at all. When a
model is defined with the same name as another model in the same app,
the definition of the first class is bound to the name of the class
regardless of the definition of the second class.
