Don't inspect the generated existential type items
r? @nikomatsakis
My debugging led me to the `hir::ItemExistential(..)` checks, which are entirely unnecessary because we never use the items directly. The issue was that items were iterated over in a random order (due to hashmaps), so if you checked the `ItemExistential` before the function that has the actual return `impl Trait`, you'd run into those ICEs you encountered.
Our implementation ends up changing the `PatKind::Range` variant in the
AST to take a `Spanned<RangeEnd>` instead of just a `RangeEnd`, because
the alternative would be to try to infer the span of the range operator
from the spans of the start and end subexpressions, which is both
hideous and nontrivial to get right (whereas getting the change to the
AST right was a simple game of type tennis).
This is concerning #51043.
`Self` in where clauses may not be object safe
Needs crater, virtually certain to cause regressions.
In #50781 it was discovered that our object safety rules are not sound because we allow `Self` in where clauses without restrain. This PR is a direct fix to the rules so that we disallow methods with unsound where clauses.
This currently uses hard error to measure impact, but we will want to downgrade it to a future compat error.
Part of #50781.
r? @nikomatsakis
three diagnostics upgrades
* reword `...` expression syntax error to not imply that you should use it in patterns either (#51043) and make it a structured suggestion
* shorten the top-line message for the trivial-casts lint by tucking the advisory sentence into a help note
* structured suggestion for pattern-named-the-same-as-variant warning
r? @oli-obk
This is virtually certain to cause regressions, needs crater.
In #50781 it was discovered that our object safety rules are not sound because we allow `Self` in where clauses without restrain. This PR is a direct fix to the rules so that we disallow methods with unsound where clauses.
This currently uses hard error to measure impact, but we will want to downgrade it to a future compat error.
Fixes#50781.
r? @nikomatsakis
Prohibit `global_allocator` in submodules
Background: #44113 is caused by weird interactions with hygiene. Hygiene is hard. After a lot of playing around, we decided that the best path forward would be to prohibit `global_allocator`s from being in submodules for now. When somebody gets it working, we can re-enable it.
This PR contains the following
- Some hygiene "fixes" -- things I suspect are the correct thing to do that will make life easier in the future. This includes using call_site hygiene for the generated module and passing the correct crate name to the expansion config.
- Comments and minor formatting fixes
- Some debugging code
- Code to prohibit `global_allocator` in submodules
- Test checking that the proper error occurs.
cc #44113#49320#51241
r? @alexcrichton
- we need to figure out hygiene first
- change the test to check that the prohibition works with a good error
msg
- leaves some comments and debugging code
- leaves some of our supposed fixes
The top level message shouldn't be too long; the
replaced-by-coercion/temporary-variable advice can live in a note. Also,
don't mention type ascription when it's not actually available as a real
thing. (The current state of discussion on the type ascription tracking
issue #23416 makes one rather suspect it will never be a stable thing in
its current form, but that's not for us to adjudicate in this commit.)
While we're here, yank out the differentiating parts of the
numeric/other conditional and only have one codepath emitting the
diagnostic.
Now that `..=` inclusive ranges are stabilized, people probably
shouldn't be using `...` even in patterns, even if it's still legal
there (see #51043). To avoid drawing attention to `...` being a real
thing, let's reword this message to just say "unexpected token" rather
"cannot be used in expressions".
async/await
This PR implements `async`/`await` syntax for `async fn` in Rust 2015 and `async` closures and `async` blocks in Rust 2018 (tracking issue: https://github.com/rust-lang/rust/issues/50547). Limitations: non-`move` async closures with arguments are currently not supported, nor are `async fn` with multiple different input lifetimes. These limitations are not fundamental and will be removed in the future, however I'd like to go ahead and get this PR merged so we can start experimenting with this in combination with futures 0.3.
Based on https://github.com/rust-lang/rust/pull/51414.
cc @petrochenkov for parsing changes.
r? @eddyb
Add label to lint for lifetimes used once
```
error: lifetime parameter `'a` only used once
--> $DIR/fn-types.rs:19:10
|
LL | a: for<'a> fn(&'a u32), //~ ERROR `'a` only used once
| ^^ -- ...is used only here
| |
| this lifetime...
```
Support future deprecation for rustc_deprecated
Follow-up to #49179 to allow `since` parameters to be set to future versions of Rust and correspondingly to not be treated as deprecated until that version. This is required for #30459 to be completed (though we'll need to wait until this hits beta).
yet another "old borrowck" bug around match default bindings
We were getting the type of the parameter from its pattern, but that didn't include adjustments. I did a `ripgrep` around and this seemed to be the only affected case.
The reason this didn't show up as an ICE earlier is that mem-categorization is lenient with respect to weird discrepancies. I am going to add more delay-span-bug calls shortly around that (I'll push onto the PR).
This example is an ICE, but I presume that there is a way to make a soundness example out of this -- it basically ignores borrows occuring inside match-default-bindings in a closure, though only if the implicit deref is at the top-level. It happens though that this occurs frequently in iterators, which often give a `&T` parameter.
Fixes#51415Fixes#49534
r? @eddyb
