[cs] Pointer detection broken

[Simn]

9e7a820 broke the C# pointer detection with some changes like this in the dump:

< 	[Cast:cs.PointerAccess<unit._TestCSharp.SomeStruct>] [Local addr(45777):cs.Pointer<unit._TestCSharp.SomeStruct>:cs.Pointer<unit._TestCSharp.SomeStruct>]
---
> 	[Cast:unit._TestCSharp.SomeStruct] [Cast:cs.PointerAccess<unit._TestCSharp.SomeStruct>] [Local addr(45777):cs.Pointer<unit._TestCSharp.SomeStruct>:cs.Pointer<unit._TestCSharp.SomeStruct>]

PointerAccess is a forwarding abstract:

@:forward abstract PointerAccess<T>(T) {}

The C# generator follows it away to look for the cs.Pointer type, but with the new dump the Cast:unit._TestCSharp.SomeStruct now gets in the way of that detection.

The problem here is that the new typing is more correct: We cast the abstract to the underlying type, then access its field. I'm not even sure why it wasn't like that before because the { e with etype = t } should also turn the expression into the underlying type.

[Simn]

Addressed in e2e359b. There's now @:forward.accessOnAbstract which generates the field access on the abstract instead of the underlying type. I think the way C# handles this is quite clever, so I'd like to still support it. It would be nice to have a more general way to "tag" field access, but I can't really think of anything.

[Simn]

C# is still broken, but of course only on Linux and Mac because that seems to be this month's theme.

[Simn]

< 			return ( (global::haxe.lang.StringExt.charCodeAt(global::haxe.lang.Runtime.toString(global::haxe.lang.Runtime.getField(a, "name", 1224700491, true)), 0)).@value < (global::haxe.lang.StringExt.charCodeAt(global::haxe.lang.Runtime.toString(global::haxe.lang.Runtime.getField(b, "name", 1224700491, true)), 0)).@value );
---
> 			return ( (global::haxe.lang.StringExt.charCodeAt(global::haxe.lang.Runtime.toString(global::haxe.lang.Runtime.getField(((object) (a) ), "name", 1224700491, true)), 0)).@value < (global::haxe.lang.StringExt.charCodeAt(global::haxe.lang.Runtime.toString(global::haxe.lang.Runtime.getField(((object) (b) ), "name", 1224700491, true)), 0)).@value );

So it adds ((object) (a) ) casts. While that's certainly weird, I can't immediately tell why it causes a syntax error:

/home/runner/work/haxe/haxe/tests/unit/src/unit/issues/Issue9123.hx(8,220): error CS1031: Type expected
/home/runner/work/haxe/haxe/tests/unit/src/unit/issues/Issue9123.hx(8,254): error CS1525: Unexpected symbol `)', expecting `;'

[kLabz]

Kinda looks like #7200 ?

[Simn]

I tried dodging the cast altogether if the types are equal, but there's some insanity going on with >>> generation that causes loads of Cannot implicitly convert type 'long' to 'int'. An explicit conversion exists (are you missing a cast?) errors.

[Simn]

Next problem: The compiler comes across this crap:

class Task_1<T0> {
    static public var Factory(get,never) : cs.system.threading.tasks.TaskFactory_1<T0>;
}

This fails since c5f913c. I've hacked it so that the type parameter lookup is allowed for @:libType classes, but that's a terrible workaround. A better workaround will be the removal of the C# target.

[Simn]

Well, I guess I'll close here again because everything is "fixed".