resolve: allow limited inherent impl's outside of the implementee's module.

[eddyb]

Allow inherent impl's in different modules than the one the type they are implemented for is in.
Unlike regular inherent impl's, these cannot contain static methods or associated items and are not subject to UFCS.
NOTE: this requires an RFC amending RFC 57.

[alexcrichton]

Isn't this reversing RFC 57? What is the motivation for this change? I also think that to reverse an RFC will require an RFC itself.

[eddyb]

@alexcrichton Sorry for not clarifying it from the start, I edited the description, hopefully it's good now.

Reading that RFC again, the only thing I don't handle is impl X where X is imported from a different module.
As of right now, the compiler claims that in such a case the import of X conflicts with the impl X "module".
That's harder to fix properly, but the error message could be improved.

[apoelstra]

I'd support this change. When I wrote RFC 57 I was concerned that allowing this would result in too specific a rule (and I was already fighting comments along the lines of "you shouldn't be changing the language in response to the current resolve design") for users to remember, but (a) this pattern is really useful, and (b) it is easy to provide good error messages. I'm working on the latter in #17321 --- it just needs some more unit tests and it'll be ready.

[huonw]

The description seems unclear still. Would an accurate summary be: impl X { ... } is allowed in other modules only if it only has methods (things mentioning self)?

(Closing pending an RFC update.)