Further tweak lifetime errors involving `dyn Trait` and `impl Trait` in return position #72804

estebank · 2020-05-31T03:19:46Z

Suggest substituting 'static lifetime in impl/dyn Trait + 'static instead of Trait + 'static + '_
When 'static is explicit, also suggest constraining argument with it
Reduce verbosity of suggestion message and mention lifetime in label
Tweak output for overlapping required/captured spans
Give these errors an error code

Follow up to #72543.

estebank · 2020-05-31T03:20:19Z

Public feedback given a small number of people in the community in replies to https://twitter.com/ekuber/status/1266565884379725824

estebank · 2020-05-31T17:49:54Z

src/test/ui/issues/issue-16922.stderr

+error[E0758]: cannot infer an appropriate lifetime
  --> $DIR/issue-16922.rs:4:14
   |
 LL | fn foo<T: Any>(value: &T) -> Box<dyn Any> {
-   |                       -- data with this lifetime...
+   |                       -- this data with an anonymous lifetime `'_`...
 LL |     Box::new(value) as Box<dyn Any>
-   |     ---------^^^^^-
-   |     |        |
-   |     |        ...and is captured here
-   |     ...is required to be `'static` by this...
+   |              ^^^^^ ...is captured here requiring it to live as long as `'static`
   |
-help: to permit non-static references in a `dyn Trait` value, you can add an explicit bound for the anonymous lifetime #1 defined on the function body at 3:1
+help: to permit non-static references in a trait object value, you can add an explicit bound for an anonymous lifetime `'_`
   |
 LL | fn foo<T: Any>(value: &T) -> Box<dyn Any + '_> {
   |                                          ^^^^


The suggestion here isn't entirely correct. We should be suggesting fn foo<'a, T: Any + 'a>(value: &'a T) -> Box<dyn Any + 'a> instead.

estebank · 2020-06-01T17:09:26Z

When adding documentation for the new error code, I noticed that we don't address the implicit 'static lifetime in impl Trait and dyn Trait anywhere in our documentation, like we do do for type parameters. Should I open a ticket for the book to explicitly address this and explain how to get deal with it?

src/test/ui/impl-trait/must_outlive_least_region_or_bound.stderr

varkor · 2020-06-01T18:05:25Z

src/test/ui/impl-trait/must_outlive_least_region_or_bound.stderr

+LL | fn elided4(x: &i32) -> Box<dyn Debug + 'static> { Box::new(x) }
+   |               ----                                         ^ ...is captured here requiring it to live as long as `'static`
+   |               |
+   |               this data with an anonymous lifetime `'_`...


When we have sentences like this, can we make sure that the first fragment is always above the second? I think users will probably read left-to-right over top-to-bottom, but it wouldn't hurt to make it unambiguous.

This is what this code would look like, with more realistic formatting:

I see: it's not so much of a problem in general, then. Would it be difficult to alter the behaviour with regards to which labels are pushed down? If we can improve the situation even for more uncommon cases, it seems worth doing, unless it's significant work.

src/test/ui/impl-trait/static-return-lifetime-infered.stderr

src/test/ui/async-await/issues/issue-62097.stderr

src/test/ui/impl-trait/must_outlive_least_region_or_bound.nll.stderr

varkor · 2020-06-01T19:30:50Z

src/test/ui/impl-trait/must_outlive_least_region_or_bound.nll.stderr

+LL | fn foo<'a>(x: &i32) -> impl Copy + 'a { x }
+   |               ----                      ^ lifetime `'a` required
+   |               |
+   |               help: add explicit lifetime `'a` to the type of `x`: `&'a i32`


I think something as simple as:

Suggested change

| help: add explicit lifetime `'a` to the type of `x`: `&'a i32`

| help: add an explicit lifetime: `&'a i32`

would do; it's easier to read.

The explicit lifetime already exists, so saying an would imply the creation of a new lifetime (that's how it'd read to me). Removing the lifetime name and the description of x's type removes some jargon, but I think there's a case to be made about this being a good moment to teach that jargon.

That sounds reasonable. What do you think about altering slightly to help: add the explicit lifetime […]? That reads slightly better to me, but is a little longer.

src/test/ui/impl-trait/must_outlive_least_region_or_bound.stderr

varkor · 2020-06-01T19:35:07Z

As always, these diagnostics improvements look great! I just have a few nit-picky comments :)

nikomatsakis

Initial comments based on reading the diffs -- I do think this PR is better than status quo, so feel free to push back on me some :)

nikomatsakis · 2020-06-01T21:19:04Z

src/test/ui/self/arbitrary_self_types_pin_lifetime_impl_trait.stderr

   |
-help: to permit non-static references in an `impl Trait` value, you can add an explicit bound for the anonymous lifetime #1 defined on the method body at 6:5
+help: to permit non-static references in an `impl Trait` value, you can add an explicit bound for an anonymous lifetime `'_`


I feel like it'd be good if we could take this opportunity to explain lifetime elision a bit better. How hard would it be to say

help: to declare that you capture borrowed data from self, you can add an explicit '_ bound

Current output:

nikomatsakis · 2020-06-01T21:19:48Z

src/test/ui/regions/regions-proc-bound-capture.stderr

+LL |     Box::new(move || { *x })
+   |              ^^^^^^^^^^^^^^ ...is captured here requiring it to live as long as `'static`
+   |
+help: consider changing the trait object's explicit `'static` bound to an anonymous lifetime `'_`


another example. It feels like it's not just "an anonymous lifetime", it's the lifetime from the argument x.

help: consider changing the trait object's explicit 'static bound to '_, to declare that you capture data from the argument x

Current output:

src/test/ui/regions/regions-close-object-into-object-4.stderr

estebank · 2020-06-01T23:21:23Z

Please check the new wording.

nikomatsakis

r=me post rebase

nikomatsakis · 2020-06-05T20:14:41Z

src/test/ui/regions/region-object-lifetime-in-coercion.stderr

   |
-help: to permit non-static references in a `dyn Trait` value, you can add an explicit bound for the anonymous lifetime #1 defined on the function body at 17:1
+help: to declare that the trait object captures data from argument `v`, you can add an explicit `'_` lifetime bound


this is really good

estebank · 2020-06-05T21:35:31Z

@bors r=nikomatsakis

bors · 2020-06-05T21:35:31Z

📌 Commit 56b1ca4f60df0ead090759b60d14200f731a08a1 has been approved by nikomatsakis

bors · 2020-06-09T03:59:53Z

☔ The latest upstream changes (presumably #73147) made this pull request unmergeable. Please resolve the merge conflicts.

bors · 2020-06-15T18:57:51Z

📌 Commit bfe1434 has been approved by nikomatsakis

@nikomatsakis

…, r=nikomatsakis Further tweak lifetime errors involving `dyn Trait` and `impl Trait` in return position * Suggest substituting `'static` lifetime in impl/dyn `Trait + 'static` instead of `Trait + 'static + '_` * When `'static` is explicit, also suggest constraining argument with it * Reduce verbosity of suggestion message and mention lifetime in label * Tweak output for overlapping required/captured spans * Give these errors an error code Follow up to rust-lang#72543. r? @nikomatsakis

@nikomatsakis

…, r=nikomatsakis Further tweak lifetime errors involving `dyn Trait` and `impl Trait` in return position * Suggest substituting `'static` lifetime in impl/dyn `Trait + 'static` instead of `Trait + 'static + '_` * When `'static` is explicit, also suggest constraining argument with it * Reduce verbosity of suggestion message and mention lifetime in label * Tweak output for overlapping required/captured spans * Give these errors an error code Follow up to rust-lang#72543. r? @nikomatsakis

@nikomatsakis

…, r=nikomatsakis Further tweak lifetime errors involving `dyn Trait` and `impl Trait` in return position * Suggest substituting `'static` lifetime in impl/dyn `Trait + 'static` instead of `Trait + 'static + '_` * When `'static` is explicit, also suggest constraining argument with it * Reduce verbosity of suggestion message and mention lifetime in label * Tweak output for overlapping required/captured spans * Give these errors an error code Follow up to rust-lang#72543. r? @nikomatsakis

@nikomatsakis

…, r=nikomatsakis Further tweak lifetime errors involving `dyn Trait` and `impl Trait` in return position * Suggest substituting `'static` lifetime in impl/dyn `Trait + 'static` instead of `Trait + 'static + '_` * When `'static` is explicit, also suggest constraining argument with it * Reduce verbosity of suggestion message and mention lifetime in label * Tweak output for overlapping required/captured spans * Give these errors an error code Follow up to rust-lang#72543. r? @nikomatsakis

@nikomatsakis

…, r=nikomatsakis Further tweak lifetime errors involving `dyn Trait` and `impl Trait` in return position * Suggest substituting `'static` lifetime in impl/dyn `Trait + 'static` instead of `Trait + 'static + '_` * When `'static` is explicit, also suggest constraining argument with it * Reduce verbosity of suggestion message and mention lifetime in label * Tweak output for overlapping required/captured spans * Give these errors an error code Follow up to rust-lang#72543. r? @nikomatsakis

@nikomatsakis

…, r=nikomatsakis Further tweak lifetime errors involving `dyn Trait` and `impl Trait` in return position * Suggest substituting `'static` lifetime in impl/dyn `Trait + 'static` instead of `Trait + 'static + '_` * When `'static` is explicit, also suggest constraining argument with it * Reduce verbosity of suggestion message and mention lifetime in label * Tweak output for overlapping required/captured spans * Give these errors an error code Follow up to rust-lang#72543. r? @nikomatsakis

@nikomatsakis

…, r=nikomatsakis Further tweak lifetime errors involving `dyn Trait` and `impl Trait` in return position * Suggest substituting `'static` lifetime in impl/dyn `Trait + 'static` instead of `Trait + 'static + '_` * When `'static` is explicit, also suggest constraining argument with it * Reduce verbosity of suggestion message and mention lifetime in label * Tweak output for overlapping required/captured spans * Give these errors an error code Follow up to rust-lang#72543. r? @nikomatsakis

@ghost

…arth Rollup of 17 pull requests Successful merges: - rust-lang#70551 (Make all uses of ty::Error delay a span bug) - rust-lang#71338 (Expand "recursive opaque type" diagnostic) - rust-lang#71976 (Improve diagnostics for `let x += 1`) - rust-lang#72279 (add raw_ref macros) - rust-lang#72628 (Add tests for 'impl Default for [T; N]') - rust-lang#72804 (Further tweak lifetime errors involving `dyn Trait` and `impl Trait` in return position) - rust-lang#72814 (remove visit_terminator_kind from MIR visitor) - rust-lang#72836 (Complete the std::time documentation to warn about the inconsistencies between OS) - rust-lang#72968 (Only highlight doc search results via mouseover if mouse has moved) - rust-lang#73034 (Export `#[inline]` fns with extern indicators) - rust-lang#73315 (Clean up some weird command strings) - rust-lang#73320 (Make new type param suggestion more targetted) - rust-lang#73361 (Tweak "non-primitive cast" error) - rust-lang#73425 (Mention functions pointers in the documentation) - rust-lang#73428 (Fix typo in librustc_ast docs) - rust-lang#73447 (Improve document for `Result::as_deref(_mut)` methods) - rust-lang#73476 (Added tooltip for should_panic code examples) Failed merges: r? @ghost

@nikomatsakis

…, r=nikomatsakis Account for multiple impl/dyn Trait in return type when suggesting `'_` Make `impl` and `dyn` Trait lifetime suggestions a bit more resilient. Follow up to rust-lang#72804. r? @nikomatsakis

rust-highfive assigned nikomatsakis May 31, 2020

rust-highfive added the S-waiting-on-review Status: Awaiting review from the assignee but also interested parties. label May 31, 2020

estebank commented May 31, 2020

View reviewed changes

estebank force-pushed the opaque-missing-lts-in-fn-2 branch from d961861 to 27cad49 Compare June 1, 2020 17:45