rustc_resolve: only prepend CrateRoot to a non-keyword segment.

[eddyb]

Fixes #53770 by treating use paths as absolute in a finer-grained manner, specifically:

use {a, crate::b, self::c, super::d};

Used to be interpreted as if it were (when uniform_paths is not enabled):

use ::{a, crate::b, self::c, super::d};

With this PR, the CrateRoot pseudo-keyword indicating an absolute path is only inserted when the first path segment is found (if it's not a keyword), i.e. the example behaves like:

use {::a, crate::b, self::c, super::d};

This should (finally) make use {path}; fully equivalent to use path;.

r? @petrochenkov cc @cramertj @joshtriplett @nikomatsakis

[petrochenkov]

FWIW, the RFC explicitly specified that the :: prefix is always added, so that we add it once and never add anything into braces at arbitrary and possibly different depth levels.

It also means you can always prepends a single crate:: to the import to migrate from 2015 to 2018.

[petrochenkov]

Without looking at code, how use {{a::b::c}}; behaves now? Can root segment be injected into deeper levels?

[petrochenkov]

I see, root segments can indeed be injected at arbitrary different levels inside braces.
Could you add a test for this case?

[petrochenkov]

I have mixed feelings and not ready to approve this single-handedly.
@eddyb, could you start a quick @rust-lang/lang review?

---

In non-uniform_paths schemes in both 2015 and 2018 editions use paths are absolute by default, it means implicit :: is added to them.

Should this implicit :: be added once at the start (current rules)

use {a, {self::b}, {c, {super::d}}};
=>
use ::{a, {self::b}, {c, {super::d}}}; // `::self::b` and `::super::d` are errors

or "adaptively", multiple times inside braces when appropriate (this PR)

use {a, {b}, {c, {d}}};
=>
use {::a, {self::b}, {::c, {super::d}}}; // OK

?

[Centril]

I would expect the following to hold:

(1)

use {$p_0};
=>
use $p_0;

(2)

use {$p_0, $p_1, $p_2};
=>
use $p_0;
use $p_1;
use $p_2;

(3)

use {$p_0, {$p_1, {$p_2}}};
=>
use $p_0;
use $p_1;
use $p_2;

This applies however arbitrarily deep you nest it.
(My understanding is that this is what @eddyb's PR achieves)

I think that prefixing :: to each use item here works poorly particularly with uniform_paths but also with anchored_paths. In the latter case, it works poorly because use {self::x}; becomes use ::self::x; which is always an error. This also holds for crate and super. For uniform_paths it works even worse since local paths can be referred to instead of referring to an extern crate. Thus, you would likely get surprising behavior from use {...}.

[bors]

☔ The latest upstream changes (presumably #54005) made this pull request unmergeable. Please resolve the merge conflicts.

[eddyb]

For uniform_paths it works even worse

uniform_paths already never injects CrateRoot aka prefix :: (if you look at the diff, logic specific that injection being done on non-uniform_paths, was moved to happen later/further-in).

So with #![feature(uniform_paths)], everything already worked like @Centril's and my intuition.

[eddyb]

@petrochenkov You mean :: not crate::, right? As in, only in Rust 2015 is use {a}; is equivalent to use crate::a;, in Rust 2018 it looks for a crate named a instead.

[petrochenkov]

@eddyb
Yes, I've updated the comment.

[bors]

☔ The latest upstream changes (presumably #54021) made this pull request unmergeable. Please resolve the merge conflicts.

[aturon]

I definitely agree with @Centril's intuition as well -- a nested use statement should always be equivalent to its "flattening" into a list of single-item use statements.

[eddyb]

@aturon @Centril Do we need some sort of process here (FCP merge, I'd guess?), or are you saying that @petrochenkov can already go ahead and r+ this PR?

[Centril]

Since we're under a bit of a time pressure, let's instead do:

@rfcbot poll lang Can we go ahead and r+ this PR?

[rfcbot]

Team member @Centril has asked teams: T-lang, for consensus on:

Can we go ahead and r+ this PR?

• @Centril
• @aturon
• @cramertj
• @eddyb
• @joshtriplett
• @nikomatsakis
• @nrc
• @pnkfelix
• @scottmcm
• @withoutboats

[petrochenkov]

Ok, @bors r+
The remaining people rarely pay attention to ticky boxes anyway.

[bors]

📌 Commit e0e7cf3 has been approved by petrochenkov

[bors]

⌛ Testing commit e0e7cf3 with merge 3d2fc45...

[bors]

☀️ Test successful - status-appveyor, status-travis
Approved by: petrochenkov
Pushing 3d2fc45 to master...