rustc claims that an import is both "unresolved" and "unused" in 2018 #54390
Labels
Comments
csmoe
added
A-diagnostics
bors
added a commit
that referenced
this issue
Nov 14, 2018
[beta] resolve: Implement uniform paths 2.0 With this PR paths in imports on 2018 edition are resolved as relative in the scope in which they are written, similarly to any other paths. The previous implementation worked differently - it tried to resolve the import in the current module (`self::import`) and in the "crate universe" (`::import`), and then merge these two resolutions if possible. The difference is that the new scheme can refer to strictly larger set of names in scope - names from unnamed blocks, names from all kinds of preludes, including macros imported with `#[macro_use] extern crate`, built-in types/macros, macros introduced with `macro_rules`. This means strictly more potential ambiguities and therefore ambiguity errors, since we keep the rule that any two different candidate names in scope conflict with each other during import resolution. So this is a breaking change for 2018 edition, but it should be relatively minor. All paths that don't start with an extern crate are also gated with the `uniform_paths` feature, paths that refer to extern crates are not gated (so we effectively get something like "future-proofed anchored paths" on stable). Another difference is treatment of paths in visibilities (`pub(in path)`). Few people remember about paths in visibilities, so before this PR they still used the old 2015 rules even on 2018 edition. Namely, paths in visibilities were crate-relative, analogous to 2015 edition imports. This PR resolves paths in visibilities as uniform as well, or rather future proofs them in this direction. Paths in visibilities are restricted to parent modules, so relative paths almost never make sense there, and `pub(in a)` needs to be rewritten as `pub(in crate::a)` in the uniform scheme, de-facto cementing the discouraged status of non-`pub(crate)` and non-`pub(super)` fine-grained visibilities. This is clearly a breaking change for 2018 edition as well, but also a minor one. The in-scope resolution strategy for import paths mirrors what is currently done for macro paths on stable (on both editions), so it will continue working even if the "ambiguity always means error" restriction is relaxed in the future. This PR also significantly improves diagnostics for all kinds of resolution ambiguities, from the newly introduced import ones to pretty old "glob vs glob" conflicts. (That's probably what I've spent most of the time on.) Why beta: - This is a breaking change on 2018 edition. - This is a large PR, it's less risky to forward-port it to nightly, than back-port to beta. cc #55618 cc #53130 cc rust-lang/rfcs#1289 Closes #18084 Closes #54525 Fixes #54390 Fixes #55668 r? @ghost
bors
added a commit
that referenced
this issue
Nov 18, 2018
[beta] resolve: Implement uniform paths 2.0 With this PR paths in imports on 2018 edition are resolved as relative in the scope in which they are written, similarly to any other paths. The previous implementation worked differently - it tried to resolve the import in the current module (`self::import`) and in the "crate universe" (`::import`), and then merge these two resolutions if possible. The difference is that the new scheme can refer to strictly larger set of names in scope - names from unnamed blocks, names from all kinds of preludes, including macros imported with `#[macro_use] extern crate`, built-in types/macros, macros introduced with `macro_rules`. This means strictly more potential ambiguities and therefore ambiguity errors, since we keep the rule that any two different candidate names in scope conflict with each other during import resolution. So this is a breaking change for 2018 edition, but it should be relatively minor. All paths that don't start with an extern crate are also gated with the `uniform_paths` feature, paths that refer to extern crates are not gated (so we effectively get something like "future-proofed anchored paths" on stable). Another difference is treatment of paths in visibilities (`pub(in path)`). Few people remember about paths in visibilities, so before this PR they still used the old 2015 rules even on 2018 edition. Namely, paths in visibilities were crate-relative, analogous to 2015 edition imports. This PR resolves paths in visibilities as uniform as well, or rather future proofs them in this direction. Paths in visibilities are restricted to parent modules, so relative paths almost never make sense there, and `pub(in a)` needs to be rewritten as `pub(in crate::a)` in the uniform scheme, de-facto cementing the discouraged status of non-`pub(crate)` and non-`pub(super)` fine-grained visibilities. This is clearly a breaking change for 2018 edition as well, but also a minor one. The in-scope resolution strategy for import paths mirrors what is currently done for macro paths on stable (on both editions), so it will continue working even if the "ambiguity always means error" restriction is relaxed in the future. This PR also significantly improves diagnostics for all kinds of resolution ambiguities, from the newly introduced import ones to pretty old "glob vs glob" conflicts. (That's probably what I've spent most of the time on.) Why beta: - This is a breaking change on 2018 edition. - This is a large PR, it's less risky to forward-port it to nightly, than back-port to beta. cc #55618 cc #53130 cc rust-lang/rfcs#1289 Closes #18084 Closes #54525 Fixes #54390 Fixes #55668 r? @ghost
|
Fixed in #55884 |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
compiling this under the 2015 edition says "unused import", but under 2018 edition, it gives both "unresolved import" and "unused import"
The text was updated successfully, but these errors were encountered: