String slices as the keys of sendable maps bring about pain

[pcwalton]

As far as I can see it, the region and slice systems make it impossible to use a send_map<&str,V>, unless you're only comparing static strings. The problem, as far as I can see it, is that find() wants to take its argument by reference, so the region checker requires invariance somehow and borrow check errors happen.

@nikomatsakis, thoughts? Maybe we need a separate StringMap type.

[pcwalton]

The other possibility is that we could have a ComparableTo<K> trait and a method find_related<Q:ComparableTo<K>>(&const self, +k: &Q), which gets rid of the need for a StringMap. That signature makes me very sad though.

The basic problem is that the query string isn't the same type as the strings inside the map...

[nikomatsakis]

Currently all type parameters are invariant. I have considered that a distinct StringMap could be useful. However, we could also apply the region variance inference to type parameters without big changes, which would be cool.

[emberian]

@pcwalton as mentioned in #6095, find_equiv and Equiv present a solution to this.

Can be closed?

[thestinger]

@cmr: I think so, the problem is solved for HashMap since it uses equality