Warn unused type aliases

[sanxiyn]

The interesting part (type aliases used by UFCS) is already tested by issue-23808.rs. But it only tests used type aliases are not warned, so I added a simple test to test unused type aliases are warned.

Fix#37455.

[rust-highfive]

r? @Aatch

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

[petrochenkov]

r? @eddyb
(Overlap with lazy-start)

[eddyb]

What's the motivation between the changes to how path resolutions are stored?
I have a somewhat more complex design I've been working on, which satisfies the needs of incremental and on-demand type-checking.
It'd be already up but I'm still trying to untangle the privacy pass from syntactical types.

[sanxiyn]

The motivation is to read partial resolution after it is overwritten by full resolution in def_map, so that UFCS path can keep type alias alive.

It can be achieved by simply keeping a copy of def_map as returned by resolve, but this (def_map only stores full resolution) seemed cleaner to me.

[eddyb]

@sanxiyn Ah this is completely unnecessary on my branch, which has a separate HIR node for the original resolution (i.e. the T in T::A). Do you mind waiting for that?

[sanxiyn]

Sure. It would be much better if I can get the original resolution directly from HIR node without maintaining the side table. Thanks.

[eddyb]

@sanxiyn#37676 is up, you can probably rebase on top of it if you want (although I should fix Travis first).

[brson]

Nice fix.

[bors]

☔ The latest upstream changes (presumably #37732) made this pull request unmergeable. Please resolve the merge conflicts.

[bors]

☔ The latest upstream changes (presumably #37676) made this pull request unmergeable. Please resolve the merge conflicts.

[eddyb]

Sorry for the delay but this should be straight-forward now.

[petrochenkov]

Should this be closed?

[sanxiyn]

I will close this when the reimplementation lands.