Disambiguation of target/source fields for intermediary models when using ManyToManyField

[German M. Bravo]

In cases where one has more than one foreign keys to either or both the source or the target models in a intermediary "through" model when using a ManyToManyField, django raises a validation error stating the ambiguity of the situation. To disambiguate this, I suggest adding two optional arguments to ManyToManyFields for those cases: 'source_fk_name' and 'target_fk_name'. I'm attaching a patch with an implementation of my suggestion.

[Ramiro Morales]

This needs tests and is currently causing a test error precisely in the model validation tests (modeltests.invalid_models):

FAIL: runTest (__main__.InvalidModelTestCase)
----------------------------------------------------------------------
Traceback (most recent call last):
  File "./runtests.py", line 85, in runTest
    self.assert_(not unexpected, "Unexpected Errors: " + '\n'.join(unexpected))
AssertionError: False is not True : Unexpected Errors: invalid_models.group: Intermediary model RelationshipDoubleFK has more than one foreign key to Person. Add a target_fk_name argument to the definition for tertiary.

[German M. Bravo]

I have fixed the error in the tests (it was only an error due the new messages) and improved the patch a little, cleaning the messages a bit, fixing the wording and completing the help. It also now validates the case of many-to-many relationships to itself. Perhaps it needs a test where it actually uses the source_fk_name and target_fk_name arguments... but I'm not too sure on how to go about adding such test. I'm attaching the file now.

[Akis Kesoglou]

I'm attaching an updated patch that works on master and a test project with extra tests that prove it works. I haven't written any documentation.

All tests pass (i.e. both extras and Django's existing tests).

After loic84's suggestion in IRC, I implemented this feature as a through_fields=('<source_field_name>', '<target_field_name>') keyword argument.

I'd love to have feedback and see the feature eventually added to Django.

[Akis Kesoglou]

updated patch for django@9db4271bd11ac23a5a565

[Akis Kesoglou]

test project

[loic84]

It's worth noting that I suggested through_fields=tuple() before knowing and reviewing the existing patch; before we move forward, we need to make a call on the public API.

Personally I like through_fields better, my rationale being:

• The through_ prefix makes it obvious it's an option that is only available to custom through relations.
• IMO what is "source" and "target" doesn't always make sense for M2M (i.e. anything other than relations to self with symmetrical=False). In that respect, I would document it as through_fields=('<field1>', '<field2>') and leave the concept of "source" and "target" out of the equation (with a special note for symmetrical=False).
• I prefer being fully explicit, rather than having a fallback mechanism for only one of the FK.

@dfunckt, would you mind integrating your tests directly in the Django test suite (probably in django/tests/many_to_many/tests.py) rather than in a separate test project? Also it would be helpful if you made a github feature branch against Django's master (​https://docs.djangoproject.com/en/dev/internals/contributing/writing-code/working-with-git/), that would facilitate the review process.

Edit: Reworded confusing point regarding "source" and "target".

[Akis Kesoglou]

Added tests, documentation and cleaned up the code a bit. Just made a pull request: ​https://github.com/django/django/pull/2327

[Akis Kesoglou]

For this patch to go forward, is it possible that a core dev can confirm or discuss the API?

I personally agree with loic84 regarding the API which I believe is short, simple and explicit. However, I'm all for re-implementing the patch if the suggested API is not accepted and there are other suggestions.

[Akis Kesoglou]

I'm marking as RTC as suggested by akaariai, after review.

[Akis Kesoglou]

I'm removing the RFC flag, since we're currently trying to improve the docs a bit, along with loic84.

[loic84]

Looks good to me, tested, flake8ed, has lots of docs and tests.

[Anssi Kääriäinen <akaariai@…>]

In c627da0ccc12861163f28177aa7538b420a9d310:

Fixed #14549 - Removed restriction of single FKs on intermediary tables

Thanks to Loic Bistuer for review. Minor changes to error messages
done by committer.