WIP: Save path segments

[nrc]

DON'T LAND

The idea with this PR is to save a def for every element of a path (we currently know this in name resolution but don't save it). Then we can save it in save-analysis and in an IDE we will be able to jump from the prefix segments of a path to their definition (which is a glaring omission in the RLS right now).

Obviously this is missing the save-analysis parts.

However, I am stuck because the compiler is panicking when bootstrapping std. I'm getting path resolved multiple times from here. From backtraces this is happening when resolving imports, (finalize_import is in the backtrace). However, it looks like that is only called once, from resolve_crate, and I think that is the only time I would record the path segments (i.e., I wouldn't when calling resolve_import because record_used is false).

So, questions - am I doing anything stupid here which I shouldn't be doing at all or should be doing differently? Or do you know/can you see any way which we could be resolving the path segments more than once?

cc @petrochenkov @eddyb @jseyfried

[rust-highfive]

r? @oli-obk

(rust_highfive has picked a reviewer for you, use r? to override)

[nrc]

This should be a fresh node id, I think, but the id that is getting multiply resolved is not CRATE_NODE_ID

[oli-obk]

I'm getting path resolved multiple times

Do you have more details? Is that happening on the last segment?

[nrc]

cc rust-dev-tools/rls-analysis#109

[nrc]

I figured out what was going wrong - when resolving a nested import, we were using the same path prefix, rather than cloning, which was fine before, but now the segments would be resolved multiple times

[eddyb]

@nrc Ouch. If we don't hack around it, we could probably use it as a start of a redesign of how imports are resolved, to only resolve the prefix once, and then have the nested imports depend on the prefix import.