In isort, this is called `add-imports`, but I prefer the declarative
name.
The idea is that by adding the following to your `pyproject.toml`, you
can ensure that the import is included in all files:
```toml
[tool.ruff.isort]
required-imports = ["from __future__ import annotations"]
```
I mostly reverse-engineered isort's logic for making decisions, though I
made some slight tweaks that I think are preferable. A few comments:
- Like isort, we don't enforce this on empty files (like empty
`__init__.py`).
- Like isort, we require that the import is at the top-level.
- isort will skip any docstrings, and any comments on the first three
lines (I think, based on testing). Ruff places the import after the last
docstring or comment in the file preamble (that is: after the last
docstring or comment that comes before the _first_ non-docstring and
non-comment).
Resolves#1700.
We don't have any doctests, but `cargo test --all` spends more than half
the time on doctests? A little confusing, but this brings the test time
from > 4s to < 2s on my machine.
This commit is a first attempt at addressing issue #1003.
The default `isort` behavior is `force-sort-within-sections = false`,
which places `from X import Y` statements after `import X` statements.
When `force-sort-within-sections = true` all imports are sorted by
module name.
When module names are equivalent, the `import` statement comes before
the `from` statement.
I ran the following code in Python 3.10 to automatically generate a list
of enums.
```python
import unittest
print(
",\n".join(
sorted(
m.removeprefix("assert") if m != "assert_" else "Underscore"
for m in dir(unittest.TestCase)
if m.startswith("assert")
)
)
)
```
The changes in this commit were generated by running:
for f in $(find src -name '*.rs'); do sed -Ei 's/use crate::registry::.*;/\0use crate::violations;/g' $f; done
for f in $(find src -name '*.rs'); do sed -Ei 's/CheckKind::([A-Z])/violations::\1/g' $f; done
git checkout src/registry.rs src/lib.rs src/lib_wasm.rs src/violations.rs
cargo +nightly fmt