Quid of sugared ops:: traits and GAT-ifying them

[danielhenrymantilla]

EDIT: see comment on #2

Based on the "GAT motivation" discussion over Zulip, and more precisely, around this question (archive link for those not on Zulip); people have started pondering about the idea that traits representing some of our sugary operations, such as […]-indexing, or deref-coercions, could maybe be GAT-ified in some retro-compatible fashion, so that they not be required to necessarily yield & or &mut references respectively.

Motivation

Plenty of examples in that Zulip thread; but people can also provide some to this very GH "issue".

But to summarize the main ones:

• custom (re-)borrowing (e.g., Pin<&mut _>) could be featured with a GAT-ified Deref{,Mut}), and more general borrowing (e.g., &mut pinned_box being able to coerce to pinned_box.as_mut());

• custom indexing/slicing.

  • without involving unsafe or other RefCast-like approaches;
  • when the reference type does not necessarily have the layout of a Rust reference (e.g., more embedded data (could be solved by ptr_metadata), or FFI requirements).
  • when other GAT APIs become widespread, and thus types such as <Ty as Trait>::Gat<'lt> become more common.

Questions:

• What would exactly be the semantics of a GAT-ified operator such as Index{,Mut} (I suspect the same concerns apply to Deref{,Mut})?

  indexable[idx]

  is actually sugar for a *-dereference (to a place expression) of the &Index::Output or &mut Index::Output (depending on the kind of usage of the place), so the starting point of these traits is &indexable[idx] or &mut indexable[idx].

  Wouldn't GAT-ifying this result in either:

  • reversing the relationship;
  • and/or having &indexable[idx] not yield a &T (and same for &mut).
    This last idea seems like the most likely solution? Maybe?

• How could the change be made in a retro-compatible fashion? I suspect we'll need new traits, to be the really sugared ones, and have the current ones be subsets which blanketly implement the new ones.

[danielhenrymantilla]

Whoops, I raced with #2, lemme move this there