adding warnings for __isnull (#3033)

[DeXtAr47-oss]

This PR adds the feature to generate warnings while incorrectly using __isnull feature,

• Detect lookups ending with __isnull
• Check whether the provided value is Literal[True]
• Remove the __isnull suffix to obtain the field lookup path.
• Resolve the lookup using django_context.resolve_lookup_into_field.
• Inspect the resolved Django Field.null attribute.
• If the field has null=False, emit a mypy error.

Related issues

• Closes Feature request : warning when incorrectly using __isnull filters #3033

[sobolevn]

Please, add tests

[ngnpope]

A few quick drive-by comments...

[ngnpope]

Why does this need Exception?

[DeXtAr47-oss]

to skip for invalid fields or unsupported lookup chains instead of raising an error

[UnknownPlatypus]

We should use more narrow exception types then, this catches too many things

[ngnpope]

This checks that it's a literal type, but not that it's True.

[DeXtAr47-oss]

thanks for pointing that out I would make the changes by checking the provided_type.value is True

[ngnpope]

Instead of the fragility of slicing 8 characters from the end of the string, this could be made more legible, e.g.

        lookup_path, sep, lookup_name = lookup_kwarg.rpartition(LOOKUP_SEP)
        is_true_literal = isinstance(provided_type, LiteralType) and provided_type.value is True

        if lookup_name == "isnull" and is_true_literal:
            try:
                field, _ = django_context.resolve_lookup_into_field(model_cls, lookup_path)
            except LookupsAreUnsupported:
                pass
            else:
                if field is not None and field.null is False:
                    ctx.api.fail(
                        f'Field "{field.name}" does not allow NULL;'
                        f'using "__isnull=True" will always return an empty queryset.',
                        ctx.context,
                    )

LOOKUP_SEP can be imported via from django.db.models.constants import LOOKUP_SEP.

[DeXtAr47-oss]

ok sure I will do that

[UnknownPlatypus]

@DeXtAr47-oss Thanks for the work on this!
There are a lot of ci failure, could you assess them to see if they are true or false positives ?