fixture loading fails silently in testcases

[John Shaffer <jshaffer2112@…>]

Recently Satchmo gained some unittests that load yaml fixtures. I didn't have PyYAML installed, so the effect was for all the tests to fail horribly. With "manage.py loaddata" the errors are shown, but I think that these errors should be reported by the testing framework.

[John Shaffer <jshaffer2112@…>]

Simply raise the verbosity level. Probably shows more messages than anyone will want.

[John Shaffer <jshaffer2112@…>]

A better solution, IMO: Show "skipping fixture" messages even when verbosity is 0. You may want to change this to use whatever Django's error conventions are, rather than a plain print statement.

[curtis.thompson@…]

+1 on the raise_verbosity patch. Without this patch, there is no way to tell if your test specific fixtures are being loaded. This is especially confusing since 'initial_data' fixture loading is displayed.

[Russell Keith-Magee]

I agree that this class of error should be caught. However, just upping the verbosity level isn't the fix. Tests shouldn't generate errors unless something goes wrong.

The dont_suppress.. patch is better. However:

1. It needs to be updated for the recent management.py refactor
2. I would argue that if you attempt to load fixture.foo, and the foo format isn't supported, it shouldn't just be a message - it should be a full error.
3. The patch doesn't address the issue of errors caused by loading fixtures with an implied format. i.e., if you specify loaddata myfixture, and you have myfixture.foo, but the foo format isn't valid, no fixtures will be loaded, but the provided patch won't catch the error.

Point 3 is a little hard to catch completely; however, you could do a reasonable job by raising an error if a fixture is named, but no objects are loaded as a result.

[Keith Bussell]

Patch for django.core.management.commands.loaddata.py

[Keith Bussell]

Patch for tests.modeltests.fixtures

[Keith Bussell]

Ready for review

[Russell Keith-Magee]

This is close, but not quite ready. In particular, the 'empty' error handling isn't quite right (or at least, is only one of a possible kind of error). The original case (3) that I gave isn't covered by your tests; I suspect all that is required is to move the 'if empty' check one level less of indentation, so that you are checking if any objects have been installed for a given fixture name, rather than checking if a specific file has no objects in it.

Some other minor nitpicks:

• could you please produce diffs from the root directory, rather than at the file level
• The test case should really be in the regression tests, not the model tests. The test doesn't serve any real documenation purpose; it is only to validate that the error conditions are correctly caught.

[Keith Bussell]

Combined patch for fix and test

[Keith Bussell]

Goes in django/tests/regressiontests/fixtures_regress/fixtures

[Keith Bussell]

Goes in django/tests/regressiontests/fixtures_regress/fixtures

[Keith Bussell]

Russell, I believe the latest version has addressed your concerns. Apologies for the long delay in getting back to this bug.

[Russell Keith-Magee]

(In [7595]) Fixed #4371 -- Improved error checking when loading fixtures. Code now catches explicitly named fixture formats that are not supported (e.g, YAML fixtures if you don't have PyYAML installed), and fixtures that are empty (which can happen due to XML tag errors). Thanks to John Shaffer for the suggestion, and Keith Bussell for his work on the fix.