`significant_drop_in_scrutinee`: Trigger lint only if lifetime allows early significant drop
I want to argue that the following code snippet should not trigger `significant_drop_in_scrutinee` (https://github.com/rust-lang/rust-clippy/issues/8987). The iterator holds a reference to the locked data, so it is expected that the mutex guard must be alive until the entire loop is finished.
```rust
use std::sync::Mutex;
fn main() {
let mutex_vec = Mutex::new(vec![1, 2, 3]);
for number in mutex_vec.lock().unwrap().iter() {
dbg!(number);
}
}
```
However, the lint should be triggered when we clone the vector. In this case, the iterator does not hold any reference to the locked data.
```diff
- for number in mutex_vec.lock().unwrap().iter() {
+ for number in mutex_vec.lock().unwrap().clone().iter() {
```
Unfortunately, it seems that regions on the types of local variables are mostly erased (`ReErased`) in the late lint pass. So it is hard to tell if the final expression has a lifetime relevant to the value with a significant drop.
In this PR, I try to make a best-effort guess based on the function signatures. To avoid false positives, no lint is issued if the result is uncertain. I'm not sure if this is acceptable or not, so any comments are welcome.
Fixes https://github.com/rust-lang/rust-clippy/issues/8987
changelog: [`significant_drop_in_scrutinee`]: Trigger lint only if lifetime allows early significant drop.
r? `@flip1995`
Rollup of 8 pull requests
Successful merges:
- #125271 (use posix_memalign on almost all Unix targets)
- #125451 (Fail relating constants of different types)
- #125478 (Bump bootstrap compiler to the latest beta compiler)
- #125498 (Stop using the avx512er and avx512pf x86 target features)
- #125510 (remove proof tree formatting, make em shallow)
- #125513 (Don't eagerly monomorphize drop for types that are impossible to instantiate)
- #125514 (Structurally resolve before `builtin_index` in EUV)
- #125527 (Add manual Sync impl for ReentrantLockGuard)
r? `@ghost`
`@rustbot` modify labels: rollup
Add manual Sync impl for ReentrantLockGuard
Fixes: #125526
Tracking Issue: #121440
this impl is even shown in the summary in the tracking issue, but apparently was forgotten in the actual implementation
remove proof tree formatting, make em shallow
Debugging via tracing `RUSTC_LOG=rustc_trait_selection::solve=debug` is now imo slightly more readable then the actual proof tree formatter. Removing everything that's not needed for the `analyse` visitor allows us to remove a bunch of code.
I personally believe that we should continue to use tracing over proof trees for debugging:
- it eagerly prints, allowing us to debug ICEs
- the proof tree builder ends up going out of sync with the actual runtime behavior, which is confusing
- using shallow proof trees is a lot more performant as we frequently do not recurse into all nested goals when using an analyse visitor
- this allows us to clean up the implementation and remove some code
r? ```@compiler-errors```
Bump bootstrap compiler to the latest beta compiler
This PR updates the bootstrap compiler, aka stage0 to the latest beta version, since it contains rust-lang/cargo#13925.
It removes those unconditional Cargo warnings:
```
warning: [...]/rust/library/core/Cargo.toml: unused manifest key: lints.rust.unexpected_cfgs.check-cfg
warning: [...]/rust/library/std/Cargo.toml: unused manifest key: lints.rust.unexpected_cfgs.check-cfg
warning: [...]/rust/library/alloc/Cargo.toml: unused manifest key: lints.rust.unexpected_cfgs.check-cfg
```
for all contributors/users of this repository (including CI).
I don't know if that's something we do, or if it's even advisable, feel free to close.
r? `@Mark-Simulacrum`
Fail relating constants of different types
fixes#121585fixes#121858fixes#124151
I gave this several attempts before, but we lost too many important diagnostics until I managed to make compilation never bail out early. We have reached this point, so now we can finally fix all those ICEs by bubbling up an error instead of continueing when we encounter a bug.
Update cargo
7 commits in 84dc5dc11a9007a08f27170454da6097265e510e..a8d72c675ee52dd57f0d8f2bae6655913c15b2fb
2024-05-20 18:57:08 +0000 to 2024-05-24 03:34:17 +0000
- Improve error description when deserializing partial field struct (rust-lang/cargo#13956)
- fix: remove symlink dir on Windows (rust-lang/cargo#13910)
- Fix wrong type of rustc-flags in documentation (rust-lang/cargo#13957)
- Add more high level traces (rust-lang/cargo#13951)
- upgrade gix from 0.62 to 0.63 (rust-lang/cargo#13948)
- Use `i32` rather than `usize` as "default integer" in library template (rust-lang/cargo#13939)
- fetch specific commits even if the github fast path fails (rust-lang/cargo#13946)
r? ghost
The body of these benchmarks is close to empty but not literally empty.
This was making the runtime of the benchmarks (which are compiled
without optimizations!) flicker between 9 ns and 10 ns runtime, which
changes the padding and breaks the test. Recent changes to the standard
library have pushed the runtime closer to 10 ns when unoptimized, which
is why we haven't seen such failures before in CI.
Contributors can also induce such failures before this PR by running the
run-make tests while the system is under heavy load.
Warn (or error) when `Self` ctor from outer item is referenced in inner nested item
This implements a warning `SELF_CONSTRUCTOR_FROM_OUTER_ITEM` when a self constructor from an outer impl is referenced in an inner nested item. This is a proper fix mentioned https://github.com/rust-lang/rust/pull/117246#discussion_r1374648388.
This warning is additionally bumped to a hard error when the self type references generic parameters, since it's almost always going to ICE, and is basically *never* correct to do.
This also reverts part of https://github.com/rust-lang/rust/pull/117246, since I believe this is the proper fix and we shouldn't need the helper functions (`opt_param_at`/`opt_type_param`) any longer, since they shouldn't really ever be used in cases where we don't have this problem.
feat: `Quick Fix` for `bare URLs`
closes#12835.
*Please write a short comment explaining your change (or "none" for internal only changes)*
changelog: [`clippy::doc_markdown`]: `Quick Fix` for `bare URLs`
Rollup of 7 pull requests
Successful merges:
- #125467 (Only suppress binop error in favor of semicolon suggestion if we're in an assignment statement)
- #125483 (compiler: validate.rs belongs next to what it validates)
- #125485 (Migrate `run-make/rustdoc-with-output-dir-option` to `rmake.rs`)
- #125497 (Fix some SIMD intrinsics documentation)
- #125501 (Resolve anon const's parent predicates to direct parent instead of opaque's parent)
- #125503 (rustdoc-json: Add test for keywords with `--document-private-items`)
- #125519 (tag more stuff with `WG-trait-system-refactor`)
r? `@ghost`
`@rustbot` modify labels: rollup
Resolve anon const's parent predicates to direct parent instead of opaque's parent
When an anon const is inside of an opaque, #99801 added a hack to resolve the anon const's parent predicates *not* to the opaque's predicates, but to the opaque's *parent's* predicates. This is insufficient when considering nested opaques.
This means that the `predicates_of` an anon const might reference duplicated lifetimes (installed by `compute_bidirectional_outlives_predicates`) when computing known outlives in MIR borrowck, leading to these ICEs:
Fixes#121574Fixes#118403
~~Instead, we should be using the `OpaqueTypeOrigin` to acquire the owner item (fn/type alias/etc) of the opaque, whose predicates we're fine to mention.~~
~~I think it's a bit sketchy that we're doing this at all, tbh; I think it *should* be fine for the anon const to inherit the predicates of the opaque it's located inside. However, that would also mean that we need to make sure the `generics_of` that anon const line up in the same way.~~
~~None of this is important to solve right now; I just want to fix these ICEs so we can land #125468, which accidentally fixes these issues in a different and unrelated way.~~
edit: We don't need this special case anyways because we install the right parent item in `generics_of` anyways:
https://github.com/rust-lang/rust/blob/213ad10c8f0fc275648552366275dc4e07f97462/compiler/rustc_hir_analysis/src/collect/generics_of.rs#L150
r? `@BoxyUwU`
compiler: validate.rs belongs next to what it validates
It's hard to find code that is deeply nested and far away from its callsites, so let's move `rustc_const_eval::transform::validate` into `rustc_mir_transform`, where all of its callers are. As `rustc_mir_transform` already depends on `rustc_const_eval`, the added visible dependency edge doesn't mean the dependency tree got any worse.
This also lets us unnest the `check_consts` module.
I did look into moving everything inside `rustc_const_eval::transform` into `rustc_mir_transform`. It turned out to be a much more complex operation, with more concerns and real edges into the `const_eval` crate, whereas this was both faster and more obvious.
Only suppress binop error in favor of semicolon suggestion if we're in an assignment statement
Similar to #123722, we are currently too aggressive when delaying a binop error with the expectation that we'll emit another error elsewhere. This adjusts that heuristic to be more accurate, at the cost of some possibly poorer suggestions.
Fixes#125458
Jonas Rinke