Template context processors cache needs to be restored after it is cleared

[Ryan Kaskel]

I contributed the portion of the patch for #16366 where context._standard_context_processors is set to None`.

Unfortunatley, I neglected to restore the cache as it was at the end of the test.

The effect of this is that cache is now wrong for the rest of my project (and will be for others) and other tests fail.

This patch fixes this.

[Claude Paroz]

I think it would be easier to just empty the cache again at the end of the tests.

[Ryan Kaskel]

You're right. That's much simpler.

[Ryan Kaskel]

just clear the cache again after the test

[Preston Holmes]

Just ran tests and they pass - this looks like a sane approach

[Claude Paroz]

Use the signal_changed approach

[Claude Paroz]

I'd just like some other committer to judge if the signal_setting approach might be a little cleaner. I also changed TEMPLATE_CONTEXT_PROCESSORS overriding in other tests to be consistent.

[Aymeric Augustin]

I'm not comfortable with large changes this close to 1.4 final. I think we could commit the first version now, and implement the signal_setting-based technique after 1.4.

[Claude Paroz]

In [17736]:

Reset template context processors cache after tests. Refs #17848.

[Claude Paroz]

Severity back to normal. Patch using setting_changed signal will be applied later.

[Claude Paroz]

Update patch for current trunk

[Claude Paroz]

Note that currenlty, the test messages.SessionTest.test_middleware_disabled_fail_silently does not pass when it is run individually, which is a symptom of the context processors cache issue (fixed in the latest patch).

[Claude Paroz]

In [17885]:

Fixed #17848 -- Added setting_changed signal for cases when TEMPLATE_CONTEXT_PROCESSORS is overriden in tests.