r10547 broke testing with Oracle

[jbronn]

More than typos were fixed in r10547, as I can no longer run any part of the test suite. I haven't figured out why (nor have I looked too deeply) -- if a fix can't be found I propose just reverting back to r10032.

C:\django\trunk\tests>python runtests.py --settings=settings --verbosity=2 aggregation
Importing model aggregation
Creating test database...
Creating test database...
_create_test_db(): dbname = test_orcl
CREATE TABLESPACE test_orcl
               DATAFILE 'test_orcl.dbf' SIZE 20M
               REUSE AUTOEXTEND ON NEXT 10M MAXSIZE 100M

CREATE TEMPORARY TABLESPACE test_orcl_temp
               TEMPFILE 'test_orcl_temp.dbf' SIZE 20M
               REUSE AUTOEXTEND ON NEXT 10M MAXSIZE 100M

Creating test user...
_create_test_user(): username = test_geo
CREATE USER test_geo
               IDENTIFIED BY Im_a_lumberjack
               DEFAULT TABLESPACE test_orcl
               TEMPORARY TABLESPACE test_orcl_temp

GRANT CONNECT, RESOURCE TO test_geo
Traceback (most recent call last):
  File "runtests.py", line 191, in <module>
    django_tests(int(options.verbosity), options.interactive, args)
  File "runtests.py", line 161, in django_tests
    failures = test_runner(test_labels, verbosity=verbosity, interactive=interactive, extra_tests=extra_tests)
  File "C:\django\trunk\django\test\simple.py", line 191, in run_tests
    connection.creation.create_test_db(verbosity, autoclobber=not interactive)
  File "C:\django\trunk\django\db\backends\creation.py", line 326, in create_test_db
    can_rollback = self._rollback_works()
  File "C:\django\trunk\django\db\backends\creation.py", line 383, in _rollback_works
    cursor = self.connection.cursor()
  File "C:\django\trunk\django\db\backends\__init__.py", line 81, in cursor
    cursor = self._cursor()
  File "C:\django\trunk\django\db\backends\oracle\base.py", line 307, in _cursor
    self.connection = Database.connect(conn_string, **self.settings_dict['DATABASE_OPTIONS'])
cx_Oracle.DatabaseError: ORA-12154: TNS:could not resolve the connect identifier specified

[Alex Gaynor]

I'm not an expert, but why is the Oracle backend modifying the global settings, we specifically changed all the backends *not* to do that.

[Erin Kelly]

Replying to Alex:

I'm not an expert, but why is the Oracle backend modifying the global settings, we specifically changed all the backends *not* to do that.

I don't recall off-hand whether it's necessary or not, but the Oracle backend still modifies the global settings when running the test suite because that's what all the other backends do. See ​http://code.djangoproject.com/browser/django/trunk/django/db/backends/creation.py#L324

[Matt Boersma]

It was commit [10547] that broke the tests for Oracle. I agree the creation.py code is messy, but there was no actual reproducible bug described, no test case, and evidently it was checked in without actually running the tests against Oracle. We're a bit past due on Django 1.1, so let's save the code beautification projects for later and just revert it.

[Alex Gaynor]

I understand the use of settings.TEST_* settings, what I don't understand is why [10547] would assign to settings.DATABASE_*. The settings.TEST_* is something that's going to have to be dealt with as a part of multidb(my inclination is to move it to DATABASE_OPTIONS), but for now we should be keeping our hands off of settings.DATABASE_* as that's just a regression.

[Ramiro Morales]

Two things to remember, if the decision to revert the changes is finally made:

• [10548] implemented the same changes to the 1.0.x branch
• Maybe the fix for this obvious typo bug can be left out of the patch revert?: ​http://code.djangoproject.com/browser/django/trunk/django/db/backends/oracle/creation.py?rev=10546#L247

[Erin Kelly]

I'm agreed with Matt. The change that was made in [10547] looks like it's just fixing some omissions, but it actually significantly changes the way testing works in Oracle (admittedly, oracle/creation.py is rather confusing in terms of where it pulls parameters from and how it applies them). This should be reverted, and the changes made should only be considered for 1.2.

[Karen Tracey]

We've also got #10716 re-opened for this problem, which is where I put a patch as I hadn't seen this one yet. Can anyone with more Oracle knowledge than me comment on the patch I added to #10716? It does revert the database name change, but instead of reverting the ettings->settings typo fix it tries to fix the problem that that typo fix revealed.

[Malcolm Tredinnick]

If it needs to be reverted, then revert it. I checked that patch in because it was fixing one very obvious bug and the other changes looked reasonable and probably necessary. By checking it in, it got tested and a problem was found, which was the whole point! So now can somebody who can test on Oracle please bring this to a close.

[Jacob]

Milestone 1.1 deleted