Check for symbol existence when working with GADT bounds in TypeComparer

[milessabin]

Without the guard the test case crashes with an assert attempting get the owner of NoSymbol.

[smarter]

You can use .maybeOwner instead of .owner to handle NoSymbol.

[milessabin]

I don't think that's going to help here ... the .owner call is in SymDenotation#typeRef, and I'm hesitant to make this change there.

I could move the check into gadtBounds and have it return null if it's called on NoSymbol ... do you think that'd be preferable?

[smarter]

What do you think @AleksanderBG ?

[milessabin]

The previous CI failures were all timeouts, BTW.

[milessabin]

Urgh ...

We need a style guide for this or something. FWIW there are a bunch of eq/ne NoSymbol in TypeComparers ... I copied those usages.

[abgruszecki]

IIUC that the problem is with NoSymbol.typeRef exploding, I'd prefer

if (sym.exists) footprint += sym.typeRef
super.gadtBounds(sym)

instead, because 1) this matters for logging 2) it should ultimately be the business of GadtConstraint to decide what to return for unconstrained symbols.

Otherwise LGTM.

[milessabin]

That looks fine.

There's similar code in the neighbouring gadtAddLowerBound and gadtAddUpperBound ... make the same change there as well?

[milessabin]

I've gone ahead and done that. Will merge when green.