RFC 401 should account for `Deref` coercions

[nrc]

RFC 401 specifies coercions in Rust. However, it doesn't deal with the case of coercions caused by a type implementing the Deref trait (specified in RFC 241). In particular, should coercions due to Deref be included in the transitivity case for implicit coercions? Currently the compiler does not (I'm not even sure of the precise behaviour when combining Deref coercions with other coercions.

cc rust-lang/rust#33220

[nrc]

Nominating for discussion in the lang meeting (cc @rust-lang/lang). Assigning P-high so hopefully this issue doesn't get lost.

[nikomatsakis]

We discussed this in the @rust-lang/lang meeting. We decided that while we would probably prefer for coercions to be transitive, we are not entirely comfortable with the implications of trying to make "deref" and "unsizing" coercions transitive with respect to one another. If nothing else, this could be very expensive from a compilation time point of view.

Consider coercing from &Box<T> to &SomeTrait, where T: SomeTrait. This would in principle require first checking whether Box<T>: SomeTrait (slash the unsizing rules etc), then deref to &T, then check again for T: SomeTrait. Maybe not a problem, but certainly not cheap.

Since extending to fully transitive coercions would be backwards compatible, we're not going to pursue it at this time, but maybe later.

[nikomatsakis]

(To be clear, it'd still be good to update the RFC to reflect the current semantics.)

[brson]

@nikomatsakis Is this supposed to be P-high. It sounds like you are saying we just need to fix the RFC. Can you assign someone to this or demote it, or close it?