Problems using methodmap as a default argument

[Fyren]

methodmap MM {
        public MM(int i) { return view_as<MM>(i); }
}

void f(MM mm = MM(0)) {}

Given this code, both 1.12 and master error, saying the argument default isn't a constant. I guess this is the expected behavior, but maybe could actually work? As a nitpick, the location given by master is off, pointing at the opening paren:

     5 | void f(MM mm = MM(0)) {}
---------------^

However, if we flip around the declarations like so and actually use f:

void f(MM mm = MM(0)) {}

methodmap MM {
        public MM(int i) { return view_as<MM>(i); }
}

public void main() { f(); }

Now 1.12 errors saying f's parameter doesn't have a default value. If we just don't use f at all, then it compiles. I didn't look into what ends up in the binary.

However, master segfaults here:

sourcepawn/compiler/semantics.cpp

Line 1934 in d23897e

if (fun->return_type()->isArray() || fun->return_type()->isEnumStruct()) {

The FunctionDecl for MM is probably wrong as fun->decl_.type.type is null, which is what return_type() returns. I didn't look into what 1.12 is actually doing.

[dvander]

I'm actually kind of surprised this parses. I think for now I'll just address the segfault, and figure out how to get more advanced default arguments working in 1.13.

[dvander]

In this case, when binding the default argument, we don't have a return type for "MM(0)" because we haven't bound "MM" yet. This is surprisingly difficult to fix and is part of a class of bugs that I'm hoping to resolve more completely, eventually. I think we'll need to recursively resolve name bindings (something the experimental compiler could do), but I'm a ways off from getting there.

I'll leave the bug open in case a simpler fix pops in mind. I think trying to harden all of Semantics against partial ASTs is a losing battle though.