Schema migration: test_add_field_temp_default_boolean

[Maximiliano Robaina]

Currently, in the schema test app, the test_add_field_temp_default_boolean method does an exception for mysql engine, which return IntegerField for boolean fields, else it returns the BooleanField type. In my case, firebird, a SmallIntegerField is returned, so that the test fails.
It is necessary to support a better approach.

[Tim Graham]

See also #23073 for a problem on some versions of Oracle.

[Shai Berger <shai@…>]

In 56252e7f46afce36fd112971c28188a3fd509a43:

Fixed schema test for Oracle 11.2.0.1 which is used in Django Project's CI.

Refs #23073 Workaround.

Refs #22738 Repeats the mysql "offense". When the issue is solved, the
Oracle special case should be made to play with the solution (that is,
Oracle should be fixed the same way that mysql and the 3rd-party backneds
are).

[Shai Berger <shai@…>]

In 588f66d18235762f8e0f96200c88bc4ba6c1c579:

[1.7.x] Fixed schema test for Oracle 11.2.0.1 which is used in Django Project's CI.

Refs #23073 Workaround.

Refs #22738 Repeats the mysql "offense". When the issue is solved, the
Oracle special case should be made to play with the solution (that is,
Oracle should be fixed the same way that mysql and the 3rd-party backneds
are).

Backport of 56252e7 from master

[Claude Paroz]

Use can_introspect_boolean_field flag

[Claude Paroz]

I've attached my proposal. Shai, can you see what Oracle backend thinks about it?

[Shai Berger]

Oracle won't like it, because it can_introspect_boolean_field even in the buggy version; the problem is specifically with fields that have defaults. This is explained in #23073.

[Claude Paroz]

Take Oracle bug into account

[Claude Paroz]

I see. Is the new patch better at this regard?

[Shai Berger]

Yes, and it passes as well.

I had hoped for a more "surgical" solution, where each backend gets to specify exactly what is expected here; the current suggestion is a little blunt in this regard, and Oracle stays special-cased in the test, which I'd rather avoid; but the suggestion does solve the problem in an acceptable way, IMO.

[Claude Paroz]

Shai, I'll try to outperform your expectations :-)
​https://github.com/django/django/pull/3278

[Claude Paroz <claude@…>]

In dbdae3a75512bddbf0b75ea6354fb3fc4bdb53cf:

Fixed #22738 -- Abstracted boolean field type introspection

Thanks maxi for the report, Shai Berger for his help with the patch
and Tim Graham for the review.

[Shai Berger]

The new patch is, indeed, a significant improvement compared to the previous approach, except for one small thing -- it fails on Oracle :).

I'll fix it.

[Shai Berger]

...so, let's see what Jenkins thinks of ​https://github.com/django/django/pull/3283

If I read the earlier failures right -- this will fail as well, because, apparently, the buggy behavior only kicks in if the column is added by ALTER TABLE, not if it is already present in the CREATE TABLE. If that turns out to be the case, I'll add another argument to the DatabaseFeatures method.

[Shai Berger <shai@…>]

In c1ae0621bab6b013025ec9024691ce7ad556409e:

Fixed #22738 -- made finer distinctions for when Boolean is not detected on Oracle

Thanks Claude Paroz for partial fix and Simon Charrette for review

[Shai Berger <shai@…>]

In a54adcecff891374341adaad6a8187d416595b70:

Fixed git blunder, refs #22738

[Claude Paroz]

Thanks for fixing this. But frankly, I think this is too much code to just workaround a bug in a proprietary product. I don't think Django deserves that. But now it's in, it's in :-)