Invalid identifiers from build macros make it to native compilation

[Gama11]

#if macro
import haxe.macro.Context;
import haxe.macro.Expr;
#else
@:build(Macro.invalidIdentifier())
#end
class Main {
    static function main() {}
}

class Macro 
{
    #if macro
    macro public static function invalidIdentifier():Array<Field>
    {
        var fields = Context.getBuildFields();
        fields.push({
            name: "0",
            doc: null,
            access: [Access.APublic, Access.AStatic],
            kind: FieldType.FVar(macro:String, null),
            pos: Context.currentPos()
        });
        return fields;
    }
    #end
}

Fails during cpp compilation:

include\Main.h(37): error C2059: syntax error: '-'
include\Main.h(37): error C2238: unexpected token(s) preceding ';'

Cs compilation:

src\Main.cs(30,23): error CS1519: Ung�ltiges Token '-' in Klasse, Struktur oder Schnittstellenmemberdeklaration.
Compilation error
Native compilation failed
Error: Build failed
Build halted with errors.

...and probably every other target with a native compilation step.

Not a regression.

[Simn]

I'm not sure if we should detect that, you could cause the same problems using @:native native. I'm inclined to file this under "Don't do stupid things".

[Gama11]

It's not like there's a valid use case for this though, right? It seems strange that macro-generated identifiers wouldn't go through the same checks as regular ones.

Not checking this leads to a lot of target-specific, cryptic errors that could be detected much earlier.

[Simn]

I still think this should be filed under "don't do stupid things". Otherwise we would have to add "is valid field name" functions for every target, and then we would probably need the same for type names, var names and whatnot. So yeah, don't do stupid things. :)

[Gama11]

Meh, I really dislike that you can get failures during native compilation (or even at runtime on some targets) because of this...

we would have to add "is valid field name" functions for every target

Huh? Why not just require it to be a valid Haxe identifier as usual- how is that target-specific?

[Gama11]

Does it make sense to do this check during decoding in macroApi?

[Simn]

I'm still not convinced this should be checked at all, but if anything this should be handled during typing. Maybe all the places where we check for that leading $.

[Gama11]

Well.. to me it seems like a pretty big hole in the type safety of the macro API. In a way it's almost as if identifiers are dynamically typed. Effectively this makes macro-generated code less type safe than regular Haxe code, which is a strange disparity.

Edit: maybe "type" safe is the wrong term here.. anyway you get the point. :)