Map crate numbers between compilations

[nikomatsakis]

It appears we are not checking for crate-numbers having changed. The best thing here would probably be to find the new crate-number. The easiest thing, however, would be to store -- for each crate-number we have -- the name/discriminant and then double-check that nothing has changed. I'll probably start with the easy fix for now.

[nikomatsakis]

I have a simple fix for that where, on startup, the compiler just validates that the crate numbers belong to the same crate as they used to (it doesn't try to find a new crate number if they don't match). This is suboptimal, clearly, but not terrible.

[nikomatsakis]

Removing from milestone since the current solution is good enough for the time being, I believe.

[nikomatsakis]

Tagging with E-mentor. This would be a good entry point for people looking to help out with incremental compilation. The remaining work is to edit the code in use crate-name/disambiguator in src/librustc_incremental/persist/directory.rs so that instead of calling krate_still_valid to check whether a given crate-number is still valid, it creates a mapping from the old crate numbers to the new ones.

We already save (and re-load) the crate-name and disambiguator (two bits of data that together identify a crate) for each crate-number, so this is basically a matter of re-loading them in the new state and comparing the old against the new.