rustdoc: Don't try to load source files from external crates
Local items defined in external macros shouldn't generate rendered source files and should link to the external crate's docs instead.
Part of #70757
r? @GuillaumeGomez
cc @eddyb
add basic support of OsStrExt for HermitCore
- this patch increases the compatibility to other operating systems
- in principle `ffi.rs` is derived from `src/libstd/sys/unix/ext/ffi.rs`
Downgrade implicit_hasher to pedantic
From the [documentation](https://rust-lang.github.io/rust-clippy/master/index.html#implicit_hasher), this lint is intended to suggest:
```diff
- pub fn foo(map: &mut HashMap<i32, i32>) { }
+ pub fn foo<S: BuildHasher>(map: &mut HashMap<i32, i32, S>) { }
```
I think this is pedantic. I get that this lint can benefit core libraries like serde, but that's exactly the use case for pedantic lints; a library like serde will [enable clippy_pedantic](fd6741f4b0/src/lib.rs (L304)) and take the time to go through everything possible. Similar for libraries doing a libz blitz style checkup before committing to a 1.0 release; it would make sense to run through all the available pedantic lints then.
But otherwise, for most codebases and certainly for industrial codebases, the above suggested change just makes the codebase more obtuse for questionable benefit.
changelog: Remove implicit_hasher from default set of enabled lints
Check fn header along with decl when suggesting to implement trait
When checking for functions that are potential candidates for trait
implementations check the function header to make sure modifiers like
asyncness, constness and safety match before triggering the lint.
Fixes#5413, #4290
changelog: check fn header along with decl for should_implement_trait
Move cognitive_complexity to nursery
As discussed in https://github.com/rust-lang/rust-clippy/issues/5418#issuecomment-610054361; Clippy's current understanding of cognitive complexity is not good enough yet at analyzing code for understandability to have this lint be enabled by default.
changelog: Remove cognitive_complexity from default set of enabled lints
Remove unnecessary TypeFlags::NOMINAL_FLAGS
This was a relic from when we had "nominal flags" and "cached
properties." The latter no longer exists, so nominal flags are no
longer necessary. In fact, every flag is considered a nominal flag. I
went ahead and removed all references to NOMINAL_FLAGS.
Fixes rust-lang#70836
Corrects a typo in rustdoc documentation.
Fixes rust-lang#70856
Sorry, I am just learning git and I think I may have accidentally done some wacky stuff that closed my last PR. Here it is again.
wf: refactor `compute_trait_ref`
moves `extend_cause_with_original_assoc_item_obligation` out of `compute_trait_ref` and
changes `trait_assoc_items` to an iterator. This saves us from building an unnecessary `Vec<_>`
r? @eddyb i guess
ci: move /var/lib/docker to /mnt on GHA
There are some builders that are running out of disk space while building the Docker images, such as arm-android. This moves and symlinks `/var/lib/docker` to the `/mnt` partition on Linux GHA.
Example of a build failing because of this: https://github.com/rust-lang-ci/rust/runs/564628621
remove false positives of unused_braces
fixes#70717
We could potentially be more aggressive when linting let bindings by checking if there are any explicit `ref`s.
I have been unable to create a snippet which compiles when using braces but has a borrowck error
without them. The closes I've gotten is [the following (playground)](https://play.rust-lang.org/?version=stable&mode=debug&edition=2018&gist=4a1552ebe9648cb13fcb8dd969189a6c).
r? @eddyb
When checking for functions that are potential candidates for trait
implementations check the function header to make sure modifiers like
asyncness, constness and safety match before triggering the lint.
Fixes#5413, #4290
Downgrade unreadable_literal to pedantic
As motivated by #5418. This is the top most commonly suppressed Clippy style lint, which indicates that the community has decided they don't share Clippy's opinion on the best style of this.
I've left the lint in as pedantic, though it could be that "restriction" would be better -- I can see this lint being useful as an opt-in restriction in some codebases.
changelog: Remove unreadable_literal from default set of enabled lints
Add new lint for `Result<T, E>.map_or(None, Some(T))`
Fixes#5414
PR Checklist
---
- [x] Followed lint naming conventions (the name is a bit awkward, but it seems to conform)
- [x] Added passing UI tests (including committed .stderr file)
- [x] cargo test passes locally
- [x] Executed cargo dev update_lints
- [x] Added lint documentation
- [x] Run cargo dev fmt
`Result<T, E>` has an [`ok()`](https://doc.rust-lang.org/std/result/enum.Result.html#method.ok) method that adapts a `Result<T,E>` into an `Option<T>`.
It's possible to get around this adapter by writing `Result<T,E>.map_or(None, Some)`.
This lint is implemented as a new variant of the existing [`option_map_none` lint](https://github.com/rust-lang/rust-clippy/pull/2128)
Downgrade inefficient_to_string to pedantic
From the [documentation](https://rust-lang.github.io/rust-clippy/master/index.html#inefficient_to_string):
> ```diff
> - ["foo", "bar"].iter().map(|s| s.to_string());
>
> + ["foo", "bar"].iter().map(|&s| s.to_string());
> ```
I feel like saving 10 nanoseconds from the formatting machinery isn't worth asking the programmer to insert extra `&` / `*` noise in the *vast* majority of cases. This is a pedantic lint.
changelog: Remove inefficient_to_string from default set of enabled lints
Downgrade trivially_copy_pass_by_ref to pedantic
The rationale for this lint is documented as:
> In many calling conventions instances of structs will be passed through registers if they fit into two or less general purpose registers.
I think the purported performance benefits of clippy's recommendation are overstated. This isn't worth asking people to sprinkle code with more `*``*``&``*``&` to chase the alleged performance.
This should be a pedantic lint that is disabled by default and opted in if some specific performance sensitive codebase determines that it is worthwhile.
As a reminder, a typical place that a reference to a primitive would come up is if the function is used as a filter. Triggering a performance-oriented lint on this type of code is the definition of pedantic.
```rust
fn filter(_n: &i32) -> bool {
true
}
fn main() {
let v = vec![1, 2, 3];
v.iter().copied().filter(filter).for_each(drop);
}
```
```console
warning: this argument (4 byte) is passed by reference, but would be more efficient if passed by value (limit: 8 byte)
--> src/main.rs:1:15
|
1 | fn filter(_n: &i32) -> bool {
| ^^^^ help: consider passing by value instead: `i32`
```
changelog: Remove trivially_copy_pass_by_ref from default set of enabled lints
Downgrade let_unit_value to pedantic
Given that the false positive in #1502 is marked E-hard and I don't have much hope of it getting fixed, I think it would be wise to disable this lint by default. I have had to suppress this lint in every substantial codebase (\>100k line) I have worked in. Any time this lint is being triggered, it's always the false positive case.
The motivation for this lint is documented as:
> A unit value cannot usefully be used anywhere. So binding one is kind of pointless.
with this example:
> ```rust
> let x = {
> 1;
> };
> ```
Sure, but the author would find this out via an unused_variable warning or from `x` not being the type that they need further down. If there ends up being a type error on `x`, clippy's advice isn't going to help get the code compiling because it can only run if the code already compiles.
changelog: Remove let_unit_value from default set of enabled lints
Fix update_lints
This fixes a bug in update_lints, where `internal` lints were not registered properly. This also cleans up some code. For example: The code generation functions no longer filter the lints the are given. This is now the task of the caller. This way, it is more obvious in the `replace_in_file` calls which lints will be included in which part of a file.
This also turns the lint modules private. There is no need for them to be public, since shared code should be in the utils module anyway.
And last but not least, this fixes the `register_lints` code generation, so also internal lints get registered.
changelog: none
Use assoc int and float consts instead of module level ones
changelog: Recommend primitive type associated constants instead of module level constants
In Rust 1.43 integer and float primitive types will have a number of new associated constants. For example `MAX`, `MIN` and a number of constants related to the machine representation of floats. https://github.com/rust-lang/rust/pull/68952
These new constants are preferred over the module level constants in `{core,std}::{f*, u*, i*}`. I have in the last few days made sure that the documentation in the main rust repository uses the new constants in every place I could find (https://github.com/rust-lang/rust/pull/69860, https://github.com/rust-lang/rust/pull/70782). So the next step is naturally to make the linter recommend the new constants as well.
This PR only changes two lints. There are more. But I did not want the PR to be too big. And since I have not contributed to clippy before it felt saner to start with a small PR so I see if there are any quirks. More will come later.
