🐛 Fix list convertor with an empty list default factory

[svlandeg]

Fixing the regression that happened after #1018, recently released as v0.19.1.

In v0.19.0, this was the original code for the internal_converor:

    def internal_convertor(value: Sequence[Any]) -> Optional[List[Any]]:
        if default_value is None and len(value) == 0:
            return None
        return [convertor(v) if convertor else v for v in value]

The author in #1018 changed that to:

    def internal_convertor(value: Sequence[Any]) -> Optional[List[Any]]:
        if default_value is None and (value is None or len(value) == 0):
            return None
        return [convertor(v) if convertor else v for v in value]

which made the (new) tests succeed, but had some typing issues. value should optionally be None (the fact that we didn't account for that was the reason for the bug that #1018 was solving).

Then mypy (understandably) needed a guarantee that value wasn't None by the final statement. Because of this, I rewrote the fix in #1018 to

    def internal_convertor(value: Optional[Sequence[Any]]) -> Optional[List[Any]]:
        if value is None or len(value) == 0:
            return default_value
        return [convertor(v) if convertor else v for v in value]

but this was a mistake and has caused a new bug as reported in #1349. I've written a new unit test to cover that bug in this PR, and now propose to revert this code to a solution in the middle:

    def internal_convertor(value: Optional[Sequence[Any]]) -> Optional[List[Any]]:
        if (value is None) or (default_value is None and len(value) == 0):
            return None
        return [convertor(v) if convertor else v for v in value]

comparing this to the original code from v0.19.0, the only thing that changes is the (value is None) or part. This seems a much cleaner and much less error-prone suggestion than what we eventually had in #1018.

[YuriiMotov]

LGTM!

[tiangolo]

Nice, thank you! I'll release this in 0.19.2 in a bit 🚀