Inline constructors vs. Void (continued)

[Simn]

This code kills the analyzer with a stack overflow:

function main() {
	[var tmp];
}

While I definitely want to fix that in the analyzer too, the code that comes out of the inline constructor processing has a problem too:

[Function:() -> Void]
        [Block:Array<Void>]
                [Var inlarr_0<6843>(VAssigned):Void]
                [Binop:Void]
                        [Local inlarr_0(6843):Void:Void]
                        =
                        [Block:Void]

There shouldn't be an empty TBlock in a value place.

cc @basro

[basro]

I'd expect this to fail during type checking and thus never reach the constructor inliner logic. Is that not supposed to be the case?

[Simn]

It's not always possible during the initial pass because of type inference. A silly example:

function f() {}

function main() {
	var tmp = null;
	[$type(tmp())]; // Unknown<0>
	tmp = f;
	$type(tmp()); // Void
}

Here we don't know the type of tmp while typing the array declaration, and it is later bound to () -> Void. This makes it necessary to check this afterwards, and we currently do that in the analyzer, which runs after inline constructors do.

There may be better ways to handle this altogether, but at the moment it can happen like this.

[basro]

Ok, I'll give it some thought.

Maybe cancelling inlining if one of the inlined fields is found to be of type Void would suffice.