coerce_path() in SchemaGenerator will not get correct pk field if get_queryset() method is used on view

[mhamid3d]

The coerce_path method is trying to grab the queryset attribute from the view. If you are using get_queryset on the view, it will not satisfy the if model and then default to id.

def coerce_path(self, path, method, view):
    """
    Coerce {pk} path arguments into the name of the model field,
    where possible. This is cleaner for an external representation.
    (Ie. "this is an identifier", not "this is a database primary key")
    """
    if not self.coerce_path_pk or '{pk}' not in path:
        return path
    model = getattr(getattr(view, 'queryset', None), 'model', None)
    if model:
        field_name = get_pk_name(model)
    else:
        field_name = 'id'
    return path.replace('{pk}', '{%s}' % field_name)

[tomchristie]

Can you provide a simple-as-possible example that demonstrates the issue?

[mhamid3d]

For example, if I was using uuid instead of id as my primary key, and using get_queryset on my view.

app/models.py

class Product(models.Model):
    uuid = models.UUIDField(primary_key=True)

app/views.py

class ProductView(viewsets.ModelViewSet):

    def get_queryset(self):
        qs = models.Product.objects.all()
        # do stuff
        return qs

Then generate a schema:

from rest_framework.schemas.openapi import SchemaGenerator

print(SchemaGenerator().get_schema())

Resulting schema will have paths such as:

app/product/{id}

instead of

app/product/{uuid}

If it doesn't make a difference to add a queryset = ... in addition to the get_queryset method, then maybe it should be suggested or disclaimer in the schema docs.

[tomchristie]

Seems reasonable to me.

If you'd like to try a pull request for this, then we can see if it introduces any unexpected effects vs. the test suite. If not then I think this would go in, yes.