Properly resolve relation type with nested source

[bashlyss]

Fixes #912

Description of the Change

I'm not sure if this specifically fixes the entirety of the first issue linked, I haven't had any problems with nested relationships that aren't M2M. However, this solves nesting for M2M on non declared fields. (Requires workaround for DRF in ModelSerializer.build_field to handle dotted field sources. Separate bug in that repo)

Slightly contrived example documenting equivalent structure to my local issue

class CourseSectionSerializer(ModelSerializer):
    class Meta:
        model = Student
        fields = ("students",)  # M2M on "course" related field
        extra_kwargs = {
            "students": {"source": "course.students"}
        }

Currently, this crashes on the line highlighted in the issue linked. It appears that somewhere along the line DRF updated how many relationships work and the previous elif above the code I specified doesn't lookup the current attributes properly. I'm not sure if there is a version/edge case that still can flow through it or not so I left it as is.

It's my first time adding something to this repo, so please direct me for where an appropriate unit test should live or any docs to update. Right now this is a quick in browser edit to fix a bug I'm having in development.

Checklist

• PR only contains one change (considered splitting up PR)
• unit-test added
• documentation updated
• CHANGELOG.md updated (only for user relevant changes)
• author name in AUTHORS

[sliverc]

Thanks for working on this. I am not sure if this is the right fix though. We have to play around with it a little. So best is to add a test case first and then see what works. Best add a new test case for nested M2M source in the file tests/test_utils.py. There are already some tests at the bottom of the file starting with test_get_related_resource_type.

And then what might work and simplify the code is replacing the following code bit:

        if parent_model is not None:
            if relation.source:
                if relation.source != "*":
                    parent_model_relation = getattr(parent_model, relation.source)
                else:
                    parent_model_relation = getattr(parent_model, relation.field_name)
            else:
                parent_model_relation = getattr(
                    parent_model, parent_serializer.field_name
                )

with something like this

 if parent_model is not None:
    parent_model_relation = relation.get_attribute(parent_model)

Needs to be tested whether this works.

[bashlyss]

I've tried that and it doesn't work as far as I can tell, at least for the ManyRelatedField relation that I'm currently working with. When I have time to play around with it I will get the test in and see. Like I said, it may actually be a bit of a different (but similar) bug.

In my environment this code resolves the ManyRelatedField issue (unsure if there is a larger global one in the current version). That is using against 4.2.1 rather than 5.0.0 and DRF 3.12.4.

[bashlyss]

@sliverc either I didn't fully understand what you intended there (get_attribute isn't a method as it happens) or it doesn't work. Adjusted my code slightly and got the tests in.

Since this is an internal, I assume no changelog/docs changes are necessary?

[sliverc]

Thanks for your update. I want to have a closer look at this. This will take some time till I have a free slot, just wanted to update you that I haven't forgotten about it.

[sliverc]

I had a deep dive into this issue again. The get_related_resource_type is very complex and therefore error-prone. Simpler would be to get the actual value of a field and resolve the resource type from the value. It would mean though to rewrite the renderer quite a bit, which is risky and best done in a complete new renderer class.

This PR is a simple bugfix though, so let us merge it as you proposed it and rethink the renderer issue at a later point. As I was looking over the code, I did some small modifications and added a changelog entry (as this actually fixes a bug).

Thanks for your work on it. Very much appreciated.