[mypyc] fix: isinstance_native fast path conflict w/ interp. subclasses [1/1]

[BobTheBuidler]

There is a bug in isinstance_native when one or more of the classes allows interpreted subclasses.

When interpreted subclasses are allowed, we cannot make any assumptions about the number of subclasses at compile time.

I'm not sure if this fix is better suited in all_concrete_classes

[p-sawicki]

I'm not sure if this fix is better suited in all_concrete_classes

i think it would make sense for all_concrete_classes to return None when any class in the hierarchy allows interpreted subclasses because that means that we can't find all the concrete classes.

Edit: though looking at the internals of all_concrete_classes this should already be the case.

i see there's one more use of all_concrete_classes to generate a type check that i think is affected by the same bug you're fixing.

could you also add an IR build test with code that hits this bug?

[BobTheBuidler]

This is preventing me from updating one of my projects to 1.18, what would you say about cherry-picking this and the crash fix in #19960 to a 1.18.3?

[p-sawicki]

not sure if there's going to be a 1.18.3 since #19764 is closed but maybe you could suggest it there.