resulting `ItemsView` isn't narrowed after `isinstance(..., Mapping)` check (and other oddities)

[posita]

Summary

Code sample in pyright playground

def count_the_things(
    *sources: Mapping[T, SupportsInt] | Iterable[tuple[T, SupportsInt]],
    use_keys_instead_of_items_for_maps: bool = False,
) -> Counter[T]:
    c: Counter[T] = Counter()
    for source in sources:
        if isinstance(source, Mapping):
            if use_keys_instead_of_items_for_maps:
                for outcome in source:
                    count = source[outcome]
                    reveal_type(outcome)  # should be T, but is T@count_the_things | tuple[T@count_the_things, SupportsInt]
                    reveal_type(count)  # should be SupportsInt, but is SupportsInt | Unknown
                    c[outcome] += int(count)
            else:
                for outcome, count in source.items():
                    reveal_type(outcome)  # should be T, but is T@count_the_things | tuple[T@count_the_things, SupportsInt]
                    reveal_type(count)  # should be SupportsInt, but is SupportsInt | Unknown
                    c[outcome] += int(count)
        else:
            for outcome, count in source:
                reveal_type(outcome)
                reveal_type(count)
                c[outcome] += int(count)
    return c

To be fair, pyrefly and ty both choke on this, so I probably screwed something up, but I'm pretty sure these should be narrower than they are. mypy passes (possibly related due to fixing python/mypy#11685).

Related

• resulting ItemsView isn't narrowed after isinstance(..., Mapping) check (and other oddities) facebook/pyrefly#3106
• resulting ItemsView isn't narrowed after isinstance(..., Mapping) check (and other oddities) astral-sh/ty#3249

Version

pyright 1.1.408

[posita]

Probably not technically a bug, even though it's counterintuitive. Explanation: astral-sh/ty#3249 (comment)