[Stack Switching] GUFA: Fix closed world analysis of continuation functions

[kripken]

As #7841 but for GUFA.

[tlively]

Alternatively we could collect funcref contents used to allocate each continuation type (using a heap type location?), then only mark them as called when we see a resume on their continuation type.

[kripken]

Yes, this could definitely be improved. The TODOs are for that. What makes this kind of annoying to do is that cont.bind changes the types... so that will need some care.

[tlively]

This should never happen, unless I'm missing something special about GUFA?

[kripken]

This is similar to the logic for call_indirect/ref:

binaryen/src/ir/possible-contents.cpp

Lines 815 to 821 in 920ea76

	template<typename T> void handleIndirectCall(T* curr, HeapType targetType) {
	// If the heap type is not a signature, which is the case for a bottom type
	// (null) then nothing can be called.
	if (!targetType.isSignature()) {
	assert(targetType.isBottom());
	return;
	}

It handles the case of (cont.new (ref.null)). Given we've ruled out unreachable earlier, all that is left is either a proper Signature type, or a null/bottom.

[tlively]

Oh I see, we're looking at the operand's function type rather than getting the function type from the continuation type.