Self parameters & C-variadics should be syntactically legal in parenthesized generic argument lists

[fmease]

Self parameters shouldn't be syntactically forbidden in this position (only semantically) since that diverges from every other place that can contain "function parameters" IINM.

We intentionally allow self parameters in e.g., function pointer types (CC #68764, rust-lang/reference#2240 (comment)) and free function items. As it stands, it means we reject X(self), X(&self), X(&mut self) but accept X(&&self), X(&&mut self) for example (the latter are rightfully legal as they're interpreted as X(_: &&self) & X(_: &&mut self) essentially¹).

Similarly, for C-variadics. They should be syntactically allowed, too, IMO.

Unifying this would once again simplify the grammar; we shouldn't fight the fact that self parameters & C-variadics are (meant to be) syntactically valid fn params always.

Originally posted by @fmease in #158517 (comment)

Footnotes

1. They're however always semantically ill-formed as self is a module not a type. ↩

[fmease]

RFC 3955 doesn't mention self parameters / C-variadics at all AFAICT. They did cross my mind when the RFC was still open but I didn't feel like discussing it given I was already submitting so many comments (arguably I should've though). However, when implementing this feature in my parser I thought about this topic again.

I did label this with T-lang but their input isn't necessary until the stabilization of this feature. Of course, this change should be noted down in the tracking issue as a divergence from the RFC, so it doesn't get lost.