Make RPITITs inherit the `assumed_wf_types` of their parent method
... and then move the RPITIT well-formedness check to just use the regular logic of wfchecking an associated type.
---
We need to inherit the `assumed_wf_types` of the RPITIT's parent function in order for the given code to be considered well-formed:
```rust
trait Foo {
fn bar<'a, T>(_: &'a T) -> impl Iterator<Output = &'a T>;
}
```
Since for `&'a T` to be WF, we need `T: 'a`.
In order for this to work for late-bound lifetimes, we need to do some additional mapping of any late-bound lifetimes captured by these assumed wf types. This is because within the body of the function (and thus in the `assumed_wf_types`), they're represented as `ReFree` variants of the original late-bound lifetimes declared in the function's generics, but in the RPITIT's GAT, they're represented as "reified" `ReEarlyBound` vars (duplicated during opaque type lowering). Luckily, the mapping between these two is already [stored in the opaque](https://doc.rust-lang.org/nightly/nightly-rustc/rustc_hir/hir/struct.OpaqueTy.html#structfield.lifetime_mapping).
Fixes#113796
Problem
Language in the Vec->Indexing documentation sounds stilted due to
incorrect word ordering: "... type allows to access values by index."
Solution
Reorder words in the Vec->Indexing documentation to flow better:
"... type allows access to values by index." The phrase "allows access to"
also matches other existing documentation.
TB: Redefine trigger condition for protectors
The Coq formalization revealed that as currently implemented, read accesses did not always commute.
Indeed starting from a lazily initialized `Active` protected tag, applying a foreign read then a child read produces `Frozen`, but child read then foreign read triggers UB (because the child read initializes _before_ the `Active -> Frozen`).
This reformulation of when protectors trigger fixes that issue:
- instead of `Active + foreign read -> Frozen` and `Active -> Frozen` when protected is UB
- we do `Active + foreign read -> if protected { Disabled } else { Frozen }`
There is already precedent for transitions being dependent on the presence of a protector (`Reserved + foreign read -> if protected { Frozen } else { Reserved }`), and this has the nice side-effect of simplifying the protector trigger condition to just an equality check against `Disabled` since now there is protector UB iff a protected tag becomes `Disabled`.
In order not to introduce an extra `if`, it was decided that `Disabled -> Disabled` would be UB when protected, which was not the case previously. This is merely a theoretical for now because a protected `Disabled` is unreachable in the first place.
The extra test is not directly related to this modification, but also checks things related to protectors and lazy initialization.
Gracefully handle ternary operator
Fixes#112578
~~May not be the best way to do this as it doesn't check for a single `:`, so it could perhaps appear even when the actual issue is just a missing semicolon. May not be the biggest deal, though?~~
Nevermind, got it working properly now ^^
Update the minimum external LLVM to 15
With this change, we'll have stable support for LLVM 15 through 17 (pending release).
For reference, the previous increase to LLVM 14 was #107573.
Refactor + improve diagnostics for `&mut T`/`T` mismatch inside Option/Result
Follow up to #114052. This also makes the diagnostics structured + translatable.
r? `@WaffleLapkin`
Change LLVM BOLT flags
I talked to the BOLT maintainers about the binary size effect of BOLT. Currently, BOLTing LLVM increases its binary size from ~120 MiB to ~170 MiB, which is not ideal. Now we can track both runtime performance and (rustc, LLVM, ...) artifact sizes in perf.RLO, so I'd like to try experimenting with changing the flags to reduce `libLLVM.so` size without regressing the performance gains of BOLT too much.
r? `@ghost`
Rename and allow `cast_ref_to_mut` lint
This PR is a small subset of https://github.com/rust-lang/rust/pull/112431, that is the renaming of the lint (`cast_ref_to_mut` -> `invalid_reference_casting`).
BUT also temporarily change the default level of the lint from deny-by-default to allow-by-default until https://github.com/rust-lang/rust/pull/112431 is merged.
r? `@Nilstrieb`
fix(resolve): update the ambiguity glob binding as warning recursively
Fixes#47525Fixes#56593, but `issue-56593-2.rs` is not fixed to ensure backward compatibility.
Fixes#98467Fixes#105235Fixes#112713
This PR had added a field called `warn_ambiguous` in `NameBinding` which is only for back compatibly reason and used for lint.
More details: https://github.com/rust-lang/rust/pull/112743
r? `@petrochenkov`
Rollup of 7 pull requests
Successful merges:
- #113773 (Don't attempt to compute layout of type referencing error)
- #114107 (Prevent people from assigning me as a PR reviewer)
- #114124 (tests/ui/proc-macro/*: Migrate FIXMEs to check-pass)
- #114171 (Fix switch-stdout test for none unix/windows platforms)
- #114172 (Fix issue_15149 test for the SGX target)
- #114173 (btree/map.rs: remove "Basic usage" text)
- #114174 (doc: replace wrong punctuation mark)
r? `@ghost`
`@rustbot` modify labels: rollup
Fix issue_15149 test for the SGX target
PR https://github.com/rust-lang/rust/pull/112390 moved tests to std. Unfortunately, this caused the `issue_15149` test to be enabled for the SGX target. This is incorrect as the SGX target does not support the `env::current_exe()` call.
Fix switch-stdout test for none unix/windows platforms
PR #112390 moved tests to std. Unfortunately, there is a typo which causes issues on platforms other than unix and windows.
tests/ui/proc-macro/*: Migrate FIXMEs to check-pass
proc-macros are processed early in the compiler pipeline. There is no need to involve codegen. So change to check-pass.
I have also looked through each changed test and to me it is sufficiently clear that codegen is not needed for the purpose of the test.
I skipped changing `tests/ui/proc-macro/no-missing-docs.rs` in this commit because it was not clear to me that it can be changed to check-pass.
Part of #62277
Rollup of 7 pull requests
Successful merges:
- #114099 (privacy: no nominal visibility for assoc fns )
- #114128 (When flushing delayed span bugs, write to the ICE dump file even if it doesn't exist)
- #114138 (Adjust spans correctly for fn -> method suggestion)
- #114146 (Skip reporting item name when checking RPITIT GAT's associated type bounds hold)
- #114147 (Insert RPITITs that were shadowed by missing ADTs that resolve to [type error])
- #114155 (Replace a lazy `RefCell<Option<T>>` with `OnceCell<T>`)
- #114164 (Add regression test for `--cap-lints allow` and trait bounds warning)
r? `@ghost`
`@rustbot` modify labels: rollup
Add regression test for `--cap-lints allow` and trait bounds warning
Closes#43134
I have verified that the test fails if stderr begins to contain output by making sure the test fails when I add
eprintln!("some output on stderr");
to the compiler (I added it to `fn build_session()`).
Replace a lazy `RefCell<Option<T>>` with `OnceCell<T>`
This code was using `RefCell<Option<T>>` to manually implement lazy initialization. Now that we have `OnceCell` in the standard library, we can just use that instead.
In particular, this avoids a confusing doubly-nested option, because the value being lazily computed is itself an `Option<Symbol>`.
Skip reporting item name when checking RPITIT GAT's associated type bounds hold
Doesn't really make sense to label an item that has a name that users can't really mention. Fixes#114145. Also fixes#113794.
r? `@spastorino`
privacy: no nominal visibility for assoc fns
Fixes#113860.
When `staged_api` is enabled, effective visibilities are computed earlier and this can trigger an ICE in some cases.
In particular, if a impl of a trait method has a visibility then an error will be reported for that, but when privacy invariants are being checked, the effective visibility will still be greater than the nominal visbility and that will trigger a `span_bug!`.
However, this invariant - that effective visibilites are limited to nominal visibility - doesn't make sense for associated functions.
ci: update ubuntu:20.04 builders to 22.04
This is mostly just maintenance to avoid bitrotting, but 22.04 also updates to cmake 3.22, so they don't need the manual builds from #113714 anymore.
