[WIP] Resolve TypeInfo variables after semantic

[yebblies]

Only resolve the type during semantic, and leave all expressions as typeid(xxx). Theoretically this should allow a clean way to ban access to typeinfo at runtime, while still allowing usage during ctfe.

@jpf91

[jpf91]

Great idea, thanks! I'll update D-Programming-GDC/gdc#100 later today.

eg alias X = typeof(typeid(x).name); should work even on a system that has no typeinfo at runtime.

Can you give an example why/how this is necessary? Minimal runtimes right now implement TypeInfo as empty classes. For this to work we really need valid, full TypeInfo declarations. It could also be kinda strange if code like this doesn't work:

static if(is(typeof(typeid(a).name)))
{
    auto b = typeid(a).name;
}

[yebblies]

Ok, I guess it's more that there's existing code in druntime that depends on finding the members, so the type must be valid if it's used. Otherwise you'll get 'declaration not found' errors instead of 'cannot use typeinfo' errors. Maybe this is acceptable.

[jpf91]

Yes I think that's acceptable. Looks like we don't have to do anything special about this: If the runtime doesn't provide the declarations and somebody uses typeof(typeid(x).name); they'll get the declaration not found error. If the runtime provides the declarations that code should (and will) work with -fno-rtti.

[yebblies]

Passing now.

[9rnsr]

Maybe this is unnecessary.

[yebblies]

It's for isType.

[9rnsr]

Ah, thanks.

[timotheecour]

this caused regression 1418:
see digger output here: https://issues.dlang.org/show_bug.cgi?id=14814