fix #44735, raise exception on Vararg-arguments in subtype rather than abort

[aviatesk]

No description provided.

[aviatesk]

@vtjnash does this seem sensible to you?

[vtjnash]

I am not precisely sure how Vararg works now, but seems probably safe. Seems like something that should be handled internally to type-intersection though, where other conditions are checked?

[aviatesk]

yeah, if jl_type_intersection does really only expect Type-objects, I think it might be better to change the signature to typeintersect(@nospecialize(a::Type), @nospecialize(b::Type)), and remove the vararg assertion from jl_type_intersection?

EDIT: typeintersect should accept other arbitrary values also -- so nvm.

[aviatesk]

ok, now the exception would be raised within subtype.c.

[Keno]

IIRC @JeffBezanson felt pretty strongly that passing non-type object here is a bug and should assert, but maybe I'm misremembering. Regardless, we should check with him when he's back from vacation.

[JeffBezanson]

Right, I don't think subtype/intersection should need to throw any errors. Currently there is only one (circular type parameter constraint), and it's basically a misplaced type-definition-time check.

I think we wanted this assert after the TypeofVararg change to make sure we caught all the cases that treated it like a type. Now that it's been working for a while, I think we could just remove the assert and let it fall through to egal like all other random values. If we later want to add checks that the arguments to typeintersect at the julia level are Types, that might be ok too.

[aviatesk]

Thanks @JeffBezanson for your comment. I removed the jl_is_vararg assertions from subtype.c, and also added Julia level assertions that checks if input arguments are Types.
I will try the inference benchmark: I will remove the Julia level error checks if benchmark result shows it can lead to some performance problem.

@nanosoldier runbenchmarks("inference", vs=":master")

[nanosoldier]

Your benchmark job has completed - possible performance regressions were detected. A full report can be found here.

[aviatesk]

@nanosoldier runbenchmarks("inference", vs=":master")

[nanosoldier]

Your benchmark job has completed - possible performance regressions were detected. A full report can be found here.

[aviatesk]

@nanosoldier runbenchmarks("inference", vs=":master")

[nanosoldier]

Your benchmark job has completed - no performance regressions were detected. A full report can be found here.

[aviatesk]

Seems good to go?