Remove more redundant casts

[rwbarton]

No description provided.

[rwbarton]

The cast in typeanal.py is redundant for a fairly non-obvious reason; it's redundant only because each of the branches of the preceding 65-line if/else-chain exits with a return and the last condition is not isinstance(sym.node, TypeInfo). I don't know what a good way to handle this is though. Writing a # type: TypeInfo comment doesn't quite work because there is a previous definition of info (though that one could just be renamed).

[gnprice]

Looks good -- thanks!

In the typeanal.py case, I think in fact that previous info definition:

             elif fullname == 'typing.Tuple':
                 if len(t.args) == 2 and isinstance(t.args[1], EllipsisType):
                     # Tuple[T, ...] (uniform, variable-length tuple)
                     node = self.lookup_fqn_func('builtins.tuple')
                     info = cast(TypeInfo, node.node)
                     return Instance(info, [t.args[0].accept(self)], t.line)

would probably be better named something like tuple_info in any case. Generally the other locals with names like t, sym, fullname, tvar_expr, etc., that are just named after a kind of thing refer to that kind of thing as derived in a canonical way from the argument t that we're visiting and the symbol bound to it. This one is not derived from t, but rather we went and got builtins.tuple off the shelf because t turned out to refer to a particular kind of tuple type.

I think this overall chain of conditional logic could also be made clearer -- in particular, I get uneasy reading it because it's not made totally manifest that all the conditions are mutually exclusive, and if they aren't there's no indication of why the order they're in should be the right one. For example, two conditions check sym.kind, one after all the fullname checks and the other before. I suspect the conditions actually are mutually exclusive, but it'd be very reassuring as a reader for that to be explicit.

[rwbarton]

In the typeanal.py case, I think in fact that previous info definition:

         elif fullname == 'typing.Tuple':
             if len(t.args) == 2 and isinstance(t.args[1], EllipsisType):
                 # Tuple[T, ...] (uniform, variable-length tuple)
                 node = self.lookup_fqn_func('builtins.tuple')
                 info = cast(TypeInfo, node.node)
                 return Instance(info, [t.args[0].accept(self)], t.line)

would probably be better named something like tuple_info in any case.

That was my feeling too, so I renamed it and added the type annotation to info below.

[rwbarton]

BTW, in this case the timing was a coincidence, but my preference would be to do this sort of cleanup immediately after each release (if applicable). My workflow involves using mypy mypy a lot and I install each new release of mypy so it helps me if mypy HEAD at least mostly continues to type check using the latest released version of mypy.