invalid assignment through `untyped` causes assertion fail

[Antriel]

class Test {
  static function main() {
    var foo = new Foo();
    untyped foo.bar = function() trace('bar');
    foo.bar();
  }
}
class Foo { 
  public function new() { } 
  public inline function bar() trace('foo'); 
}

Results in assertion fail at

haxe/src/typing/typer.ml

Lines 514 to 515 in f2f4dc1

	| AKInline _ | AKMacro _ ->
	assert false)

Without untyped it properly gives error message. Should this assertion be changed into error or should it be handled sooner?

[back2dos]

Silly question: why should it even fail at compile time? I think it should try to perform the assignment at runtime and thus behavior according to whatever the runtime semantics are.

[nadako]

What will happen if you do (cast foo).bar = function ...?

[Antriel]

@back2dos because it's inlined, but I guess that doesn't meant it shouldn't be able to do that. But it would be even more obscure, as it would compile, but not work. Hmm.
@nadako it compiles, but since it's inlined, it still traces out 'foo'.

[Simn]

IMO this is wontfix. There are probably many ways to cause assertion failures through untyped.

[nadako]

FWIW it would be nice to attach some sensible messages to our assertion failures :D

[RealyUniqueName]

I agree, assertions which rich end user should be fixed or replaced with proper error messages.

[kLabz]

Since 4.2, there is no error anymore.

untyped foo.bar = assignment isn't replacing foo.bar inlined values, which I think is fine.
I'm not sure what to test though, since the assertion failure (nor a replacement "proper error") doesn't seem to be reachable anymore 🤔

[kLabz]

Well.. changing target to one where this cannot work, I can trigger (another) assertion failure in genhl: https://try.haxe.org/#5628fd59

And runtime error on eval (which is expected?)