This PR sets up an "ecosystem" check as an optional part of the CI step for pull requests. The primary piece of this is a new script in `scripts/check_ecosystem.py` which takes two ruff binaries as input and compares their outputs against a corpus of open-source code in parallel. I used ruff's `text` reporting format and stdlib's `difflib` (rather than JSON output and jsondiffs) to avoid adding another dependency. There is a new ecosystem-comment workflow to add a comment to the PR (see [this link](https://securitylab.github.com/research/github-actions-preventing-pwn-requests/) which explains why it needs to be done as a new workflow for security reasons).
## Summary
This PR moves `Diagnostic`, `DiagnosticKind`, and `Fix` into their own crate, which will enable us to further split up Ruff, since sub-linter crates (which need to implement functions that return `Diagnostic`) can now depend on `ruff_diagnostics` rather than Ruff.
Implement PYI006 "bad version info comparison"
## What it does
Ensures that you only `<` and `>=` for version info comparisons with
`sys.version_info` in `.pyi` files. All other comparisons such as
`<`, `<=` and `==` are banned.
## Why is this bad?
```python
>>> import sys
>>> print(sys.version_info)
sys.version_info(major=3, minor=8, micro=10, releaselevel='final', serial=0)
>>> print(sys.version_info > (3, 8))
True
>>> print(sys.version_info == (3, 8))
False
>>> print(sys.version_info <= (3, 8))
False
>>> print(sys.version_info in (3, 8))
False
```
Co-authored-by: Charlie Marsh <charlie.r.marsh@gmail.com>
I worked on #2993 and ran into issues that the formatter tests are failing on Windows because `writeln!` emits `\n` as line terminator on all platforms, but `git` on Windows converted the line endings in the snapshots to `\r\n`.
I then tried to replicate the issue on my Windows machine and was surprised that all linter snapshot tests are failing on my machine. I figured out after some time that it is due to my global git config keeping the input line endings rather than converting to `\r\n`.
Luckily, I've been made aware of #2033 which introduced an "override" for the `assert_yaml_snapshot` macro that normalizes new lines, by splitting the formatted string using the platform-specific newline character. This is a clever approach and gives nice diffs for multiline fixes but makes assumptions about the setup contributors use and requires special care whenever we use line endings inside of tests.
I recommend that we remove the special new line handling and use `.gitattributes` to enforce the use of `LF` on all platforms [guide](https://docs.github.com/en/get-started/getting-started-with-git/configuring-git-to-handle-line-endings). This gives us platform agnostic tests without having to worry about line endings in our tests or different git configurations.
## Note
It may be necessary for Windows contributors to run the following command to update the line endings of their files
```bash
git rm --cached -r .
git reset --hard
```
Currently the define_rule_mapping! macro generates both the Rule enum as
well as the RuleCodePrefix enum and the mapping between the two. After
this commit series the macro will only generate the Rule enum and the
RuleCodePrefix enum and the mapping will be generated by a new map_codes
proc macro, so we rename the macro now to fit its new purpose.
```console
❯ cargo run rule B017
Finished dev [unoptimized + debuginfo] target(s) in 0.13s
Running `target/debug/ruff rule B017`
no-assert-raises-exception
Code: B017 (flake8-bugbear)
### What it does
Checks for `self.assertRaises(Exception)`.
## Why is this bad?
`assertRaises(Exception)` can lead to your test passing even if the
code being tested is never executed due to a typo.
Either assert for a more specific exception (builtin or custom), use
`assertRaisesRegex` or the context manager form of `assertRaises`.
```
To enable ruff_dev to automatically generate the rule Markdown tables in
the README the ruff library contained the following function:
Linter::codes() -> Vec<RuleSelector>
which was slightly changed to `fn prefixes(&self) -> Prefixes` in
9dc66b5a65 to enable ruff_dev to split
up the Markdown tables for linters that have multiple prefixes
(pycodestyle has E & W, Pylint has PLC, PLE, PLR & PLW).
The definition of this method was however largely redundant with the
#[prefix] macro attributes in the Linter enum, which are used to
derive the Linter::parse_code function, used by the --explain command.
This commit removes the redundant Linter::prefixes by introducing a
same-named method with a different signature to the RuleNamespace trait:
fn prefixes(&self) -> &'static [&'static str];
As well as implementing IntoIterator<Rule> for &Linter. We extend the
extisting RuleNamespace proc macro to automatically derive both
implementations from the Linter enum definition.
To support the previously mentioned Markdown table splitting we
introduce a very simple hand-written method to the Linter impl:
fn categories(&self) -> Option<&'static [LinterCategory]>;
- optional `prefix` argument for `add_plugin.py`
- rules directory instead of `rules.rs`
- pathlib syntax
- fix test case where code was added instead of name
Example:
```
python scripts/add_plugin.py --url https://pypi.org/project/example/1.0.0/ example --prefix EXA
python scripts/add_rule.py --name SecondRule --code EXA002 --linter example
python scripts/add_rule.py --name FirstRule --code EXA001 --linter example
python scripts/add_rule.py --name ThirdRule --code EXA003 --linter example
```
Note that it breaks compatibility with 'old style' plugins (generation works fine, but namespaces need to be changed):
```
python scripts/add_rule.py --name DoTheThing --code PLC999 --linter pylint
```
543865c96b introduced
RuleCode::origin() -> RuleOrigin generation via a macro, while that
signature now has been renamed to Rule::origin() -> Linter we actually
want to get rid of it since rules and linters shouldn't be this tightly
coupled (since one rule can exist in multiple linters).
Another disadvantage of the previous approach was that the prefixes
had to be defined in ruff_macros/src/prefixes.rs, which was easy to
miss when defining new linters in src/*, case in point
INP001 => violations::ImplicitNamespacePackage has in the meantime been
added without ruff_macros/src/prefixes.rs being updated accordingly
which resulted in `ruff --explain INP001` mistakenly reporting that the
rule belongs to isort (since INP001 starts with the isort prefix "I").
The derive proc macro introduced in this commit requires every variant
to have at least one #[prefix = "..."], eliminating such mistakes.
More accurate since the enum also encompasses:
* ALL (which isn't a prefix at all)
* fully-qualified rule codes (which aren't prefixes unless you say
they're a prefix to the empty string but that's not intuitive)
"origin" was accurate since ruff rules are currently always modeled
after one origin (except the Ruff-specific rules).
Since we however want to introduce a many-to-many mapping between codes
and rules, the term "origin" no longer makes much sense. Rules usually
don't have multiple origins but one linter implements a rule first and
then others implement it later (often inspired from another linter).
But we don't actually care much about where a rule originates from when
mapping multiple rule codes to one rule implementation, so renaming
RuleOrigin to Linter is less confusing with the many-to-many system.
