[AccessBase] FindAccessBaseVisitor visits nested accesses.

[nate-chandler]

Implemented visitNestedAccess in FindAccessBaseVisitor to visit the value found by the superclass. Doing so is required by AccessPhiVisitor.

rdar://115033825

[atrick]

Thanks!

The initial thinking was that we could never see a begin_access after following a phi. That's not actually enforceable and we should end up being conservative anyway if we care about the access scope. So, this fix is correct.

But... I stared at this source for a long time to try to prove that it's complete. I mean, a very long time.

It all rests on the fact that all of the other implementations of FindAccessVisitorImpl happen to use NestedAccessType::IgnoreAccessBegin.

I think we need to leave some breadcrumb about what we've learned. For example

Why doesn't FindAccessVisitorImpl::visitNestedAccess always do this itself in the StopAtAccessBegin case`?

Or if that doesn't work, at least comment in the other visitors why they don't need to overried visitNestedAccess.

Or some other idea?

[nate-chandler]

Yeah, it's incomplete. Like you say, the other subclasses would need to be updated or the superclass would. I've come around to updating the superclass to implement visitNestedAccess as in this patch except instead of calling SuperTy::visitNestedAccess it either directly looks at the two cases (what's nestedAccessTy) or calls a renaming of the current impl of visitNestedAccess.

[nate-chandler]

@swift-ci please test

[atrick]

Yeah, it's incomplete. Like you say, the other subclasses would need to be updated or the superclass would. I've come around to updating the superclass to implement visitNestedAccess as in this patch except instead of calling SuperTy::visitNestedAccess it either directly looks at the two cases (what's nestedAccessTy) or calls a renaming of the current impl of visitNestedAccess.

Nice! I think that's better.