Joe Eli McIlvain

This can be solved by: during union flattening, check the current frame and see if we are in a constraint before invoking `is_subtype` - if we are we should invoke...

Careful caveat though - we don't want to use `is_subtype_constraint` when the union is within the type arguments of a constraint type. For example, in `[A: Array[(B | C)]`, we'd...

It compiles correctly if the `UnsignedInteger` trait is not used as a constraint.

Might be a recursion issue, then. :disappointed:

I wonder if this is related to #2808...

Thanks for reporting! This looks like a nasty one, so I've added the "triggers release" label.

Interestingly enough, I think I've found a type system bug that was hiding a capability safety issue in `filter_stateful`. I'll try to figure out if it is related.

Sorry about this. It turned out that this gencap didn't actually solve my problem (5 years ago), and neither I nor anyone else has had a pressing reason to add...

Given the assert location, this regression was probably introduced in #3304

Note that if it's like the M1 issue, then the problem goes away if you link the program against the runtime bitcode (which I still think should be the default...