Incorrect "Alternative patterns bind different names" with mapping capture statements

[A5rocks]

Can someone reopen this bug? There's a valid repro.
@cdce8p you're correct about str, but can you reopen this bug? There’s another repro of this "Alternative patterns bind different names" bug that seems to be new in 1.18 (before it was silently Any).

d1: dict[str, tuple[str, ...]] = {}
d2: dict[str, str] = {}

match (d1, d2):
    case ({"Symbol": (sym,)}, _) | (_, {"Symbol": sym}):
        reveal_type(sym)

playground

# main.py:5: error: Alternative patterns bind different names [misc]

I added a breakpoint to the source.
This looks like an issue in checkpattern.visit_or_pattern.
Debugging shows one alternative is given Never instead of a tuple type::

mypy/mypy/checkpattern.py(187)visit_or_pattern()
-> self.msg.fail(message_registry.OR_PATTERN_ALTERNATIVE_NAMES, o.patterns[i])
(Pdb) p pattern_types[0]
PatternType(type=tuple[builtins.dict[builtins.str, tuple[builtins.str]], builtins.dict[builtins.str, builtins.str]], rest_type=tuple[builtins.dict[builtins.str, tuple[builtins.str]], Never], captures={<mypy.nodes.NameExpr object at 0x7f50c2c2b9a0>: builtins.str})
(Pdb) p pattern_types[1]
PatternType(type=Never, rest_type=tuple[Never, Never], captures={})

So mypy concludes the alternatives bind differently, even though both capture sym.

Originally posted by @Hnasar in #13950

cc @cdce8p

[neopexrl]

Hi, can I work on this issue?

[A5rocks]

Actually this is already being fixed in #19993, that PR doesn't reference this yet because I just made this issue for it.