Warn about overlapping patterns

[lkuper]

We already check pattern matches for non-exhaustiveness. It would be nice to also produce a warning for overlapping match patterns ("over-exhaustiveness"). GHC, for example, does this.

[nikomatsakis]

We do report useless patterns (those that cannot match because a prior
pattern must have matched it). Did you mean something else?

[lkuper]

The unreachable pattern warning is good, but I'm imagining something that would warn any time the sets of things matched by each pattern were not disjoint. The point of the warning would be to say, "Be advised: the order in which these match arms are written is relevant!"

On the other hand, maybe this isn't right for Rust because we often write a catch-all _ => ... pattern at the end of matches, and this warning would always warn about that.

[nikomatsakis]

Ah, I see! Well, you could make it default to off. Or perhaps we could somehow treat _ as a special case.

[lkuper]

Having it default to off sounds like the right thing.

[catamorphism]

far-future

[msullivan]

I don't think that GHC warns about the order of match arms being relevant? Certainly ML compilers don't.

[lkuper]

What GHC does is warn about a set of patterns being overlapping -- -fwarn-overlapping-patterns is on by default. And this is a little more strict than merely warning about unreachable patterns, which Rust does already.

The idea is that if you have a set of patterns such that in one order you'd get an unreachable pattern warning in Rust, but in a different order you wouldn't, then that code is not robust to reordering. Sometimes this is fine, but other times you might want to at least get a warning that the code is fragile.

[huonw]

Triage: no progess that I know of.

[huonw]

previous_comment.clone()

[steveklabnik]

previous_comment.clone()

[Manishearth]

This would have to be a lint, since it's not strictly undesirable behavior. Perhaps this should be closed in favor of an issue on clippy since we're not adding new lints?