Fallback `{float}` to `f32` when `f32: From<{float}>` and add `impl From<f16> for f32`
Currently, the following code compiles:
```rust
fn foo<T: Into<f32>>(_: T) {}
fn main() {
foo(1.0);
}
```
This is because the only `From<{float}>` impl for `f32` is currently `From<f32>`. However, once `impl From<f16> for f32` is added this is no longer the case. This would cause the float literal to fallback to `f64`, subsequently causing a type error as `f32` does not implement `From<f64>`. While this kind of change to type inference isn't technically a breaking change according to Rust's breaking change policy, the previous attempt to add `impl From<f16> for f32` was removed rust-lang/rust#123830 due to the large number of crates affected (by my count, there were root regressions in 42 crates and 52 GitHub repos, not including duplicates). This PR solves this problem by using `f32` as the fallback type for `{float}` when there is a trait predicate of `f32: From<{float}>`. This allows adding `impl From<f16> for f32` without affecting the code that currently compiles (such as the example above; this PR shouldn't affect what is possible on stable).
This PR also allows adding a future-incompatibility warning for the fallback to `f32` (currently implemented in the third commit) if the lang team wants one (allowing the `f32` fallback to be removed in the future); alternatively this could be expanded in the future into something more general like @tgross35 suggested in https://github.com/rust-lang/rust/issues/123831#issuecomment-2064728053. I think it would be also possible to disallow the `f32` fallback in a future edition.
As expected, a crater check showed [no non-spurious regressions](https://github.com/rust-lang/rust/pull/139087#issuecomment-2764732580).
For reference, I've based the implementation loosely on the existing `calculate_diverging_fallback`. This first commit adds the `f32` fallback, the second adds `impl From<f16> for f32`, and the third adds a FCW lint for the `f32` fallback. I think this falls under the types team, so
r? types
Fixes: rust-lang/rust#123831
Tracking issue: rust-lang/rust#116909
@rustbot label +T-lang +T-types +T-libs-api +F-f16_and_f128
To decide on whether a future-incompatibility warning is desired or otherwise (see above):
@rustbot label +I-lang-nominated
cc https://github.com/rust-lang/rust/issues/154024https://github.com/rust-lang/rust/issues/154005
Reformat `top_level_options!` and `options!` macro declarations
These macros are already tricky, and having weird formatting doesn't help. Using a more regular style makes it easier to see where nesting begins and ends.
Extracted from https://github.com/rust-lang/rust/pull/154501 after the changes in https://github.com/rust-lang/rust/pull/149357 made rebasing very difficult.
There should be no change to compiler behaviour.
compiler: update hashbrown to 0.17
See library's rust-lang/rust#155154 for the bug fixes this brings.
This PR also updates `indexmap` in the compiler as a direct dependent.
Remove dead diagnostic structs.
One of these has a "FIXME(autodiff): I should get used somewhere" comment, but I figure YAGNI applies and it's so small that reinstating it if necessary would be trivial.
r? @Kivooeo
remove PointeeParser
this parser does nothing currently, as the current scope of `rustc_attr_parsing` only includes builtin attributes, and not derive macros
it isn't defined in `compiler/rustc_feature/src/builtin_attrs.rs`
all the actual parsing for `#[pointee]` is actually handled in `compiler/rustc_builtin_macros/src/deriving/coerce_pointee.rs`
r? @JonathanBrouwer
Delete unused `rustc_trait_selection` errors.
The first two of these are duplicated elsewhere in some form or another, apparently they became orphaned during various moves and refactors. The third is just never used.
Tests for precise-capture through RPIT and TAIT
- Tests for https://github.com/rust-lang/rust/issues/155151.
These tests succeed under `-Znext-solver`, but incorrectly fail under the old trait solver.
---
The bug can be triggered via return-position `impl Trait` on stable, but requires some rather contrived code. When using type-alias `impl Trait`, it's easier to imagine the issue being triggered by real code.
Test/lexer unicode pattern white space
This PR adds a test for the Rust lexer to verify it correctly accepts vertical tab (`\x0B`) as valid whitespace between tokens. Vertical tab is part of Unicode Pattern_White_Space, which the Rust language specification uses to define whitespace.
Related: Outreachy tracking [Pattern_White_Space](https://www.unicode.org/reports/tr31/#R3a)
Clarify ascii whitespace exclusion of vertical tab in the doc
This especially means that for `c: char`, `c.is_ascii() && c.is_whitespace()` does **not** imply `c.is_ascii_whitespace()`, which can cause bug and is highly counterintuitive.
docs: clarify path search behavior in std::process::Command::new
the existing docs mentioned that `PATH` is searched in an "os defined way" and that it could be controlled by setting PATH on the command but never explained which `PATH` is actually used.
on unix the key thing to understand is that when you modify the childs environment (via `env()`, `env_remove()`, or `env_clear()`), the `PATH` search uses whatever `PATH` ends up in the child's environment not the parents. so if you call `env_clear()` and forget to set `PATH`, you don't get the parents `PATH` as a fallback; you get the OS default (typically `/bin:/usr/bin`) which often won't find what you need.
the three cases are now documented:
- unmodified env: child inherits parents `PATH`, that gets searched
- `PATH` explicitly set `via env()`: the new value is searched
- `PATH` removed (`env_clear` or `env_remove`): falls back to OS default, not the parents `PATH`
on windows rust resolves the executable before spawning rather than letting `CreateProcessW` do it. the search order is: childs `PATH` (if explicitly set) first then system-defined directories then the parents `PATH`. the existing note about issue rust-lang/rust#37519 is preserved.
limitations in this doc:
- the unix fallback path behavior ("typically `/bin:/usr/bin`") is verified for linux/glibc. behavior on macOS, BSD, and musl is similar in practice but not fully confirmed here.
- the windows search order is summarized as "system-defined directories" which actually includes the application directory (the directory of the calling process's executable) as a distinct step before the system dirs that detail is omitted for brevity.
- `posix_spawnp` `PATH` source (calling process environ vs envp argument) is ambiguous in the `POSIX` spec; the behavior here is inferred from the guard in `unix.rs` that bypasses `posix_spawn` when `PATH` has been modified.
closesrust-lang/rust#137286 (hopefully)
`BorrowedBuf`: Update outdated safety comments in `set_init` users.
These comments appear to have been written before `BorrowedBuf`'s init tracking was simplified in
https://github.com/rust-lang/rust/pull/150129. The `BufWriter` comment of the usage within `BufWriter` will be handled separately.
CC rust-lang/rust#78485, rust-lang/rust#117693.
explicit-tail-calls: disable two tests on LoongArch
A [recent LLVM change](https://github.com/llvm/llvm-project/pull/191508) broke these on LLVM 23.
I suspect these will eventually be fixed, so maybe it'd be okay/better to just leave this pending so it applies to our CI without merging it? I'm open to opinions.
Fix ICE when Self is used in enum discriminant of a generic enum
Fixesrust-lang/rust#153756
Let discriminant AnonConst inherit parent generics via Node::Variant in generics_of, and emit a proper error instead of span_bug! for the TooGeneric case in wfcheck.
One of these has a "FIXME(autodiff): I should get used somewhere"
comment, but I figure YAGNI applies and it's so small that reinstating
it if necessary would be trivial.
These comments appear to have been written before `BorrowedBuf`'s
init tracking was simplified in
https://github.com/rust-lang/rust/pull/150129. The `BufWriter` comment
of the usage within `BufWriter` will be handled separately.
Previously the lint would be reported at the right span, but could not be `allow`ed or `expect`ed in that location, because the lint was actually emitted using `CRATE_HIR_ID`
Rollup of 7 pull requests
Successful merges:
- rust-lang/rust#154049 (delegation: Track more precise spans for glob delegations)
- rust-lang/rust#155134 (Replace custom trim_ascii_start with the standard library method)
- rust-lang/rust#155235 (add the `fma4` x86 target feature)
- rust-lang/rust#155218 (coroutines: Skip the closure signature annotation check for tainted bodies)
- rust-lang/rust#155274 (limit duplicate-profiler-builtins test to targets that can do dynamic linking)
- rust-lang/rust#155276 (`#[rustc_must_match_exhaustively]` detect let else)
- rust-lang/rust#155281 (Revert "allow `windows-gnu` targets to embed gdb visualizer scripts")
* Fix ICE when Self is used in enum discriminant of a generic enum
Move the validation into the existing `check_param_uses_if_mcg` machinery
in HIR ty lowering instead of adding a new check in wfcheck. After the
`AnonConstKind` refactoring, `ForbidMCGParamUsesFolder` was only gated on
`AnonConstKind::MCG`, causing discriminant anon consts (`NonTypeSystem`) to
bypass it entirely.
Add `anon_const_forbids_generic_params()` which returns the appropriate
`ForbidParamContext` for both MCG and enum discriminant contexts. Wire it
into `check_param_uses_if_mcg` so that `Self` aliasing a generic type is
caught before reaching `const_eval_poly`. Convert the `TooGeneric` span_bug
into a proper diagnostic as a fallback for anything slipping through
type-dependent path resolution.
* Address review comments
- Rename `ForbidMCGParamUsesFolder` to `ForbidParamUsesFolder`
- Rename `MinConstGenerics` variant to `ConstArgument` with updated doc
- Simplify doc comment on `anon_const_forbids_generic_params`
- Make match on `AnonConstKind` exhaustive
- Move `anon_const_def_id` inside the `if let` in `check_param_uses_if_mcg`
- Remove now-unreachable `TooGeneric` span_err in wfcheck
* Revert TooGeneric arm back to span_bug! as requested by reviewer
* Use generics_of to determine if NonTypeSystem anon consts allow generic params
* Also check InlineConst and Closure defs nested in enum discriminants
* Simplify logic for determining anonymous constant parent in generic contexts
* add test
limit duplicate-profiler-builtins test to targets that can do dynamic linking
this is the error I got for an example of such a target
```
=== STDERR ===
error: extern location for dylib_a does not exist: libdylib_a.so
--> main.rs:2:5
|
2 | dylib_a::something();
| ^^^^^^^
error: extern location for dylib_b does not exist: libdylib_b.so
--> main.rs:3:5
|
3 | dylib_b::something_else();
| ^^^^^^^
coroutines: Skip the closure signature annotation check for tainted bodies
When a coroutine has too many parameters, `check_match` fails and `construct_error` builds a MIR body with only the coroutine's computed arguments (env + resume type). The user-provided signature, however, still reflects all the parameters the user wrote. `check_signature_annotation` then tries to `zip_eq` these two mismatched iterators, causing a panic. Checking `tainted_by_errors` and bailing early avoids this, since `construct_error` bodies cannot meaningfully be compared against user annotations.
Example currently ICEing:
```rust
fn main() {
|(1, 42), ()| yield;
}
```
Closesrust-lang/rust#139570.
Scott McMurray