Repair violations from Ruff 0.16's expanded defaults

[jaraco]

Ref jaraco/skeleton#210.

Ruff 0.16 broadened its built-in default rule set, and because ruff.toml uses extend-select, this project inherited 19 new violations. All are now repaired; ruff check . and ruff format --check . are clean and the test suite (including mypy) passes.

Changes

Mechanical / no behavior change

• git.py: startswith called once with a tuple (PIE810); '...'.split() replaced with list literals here and in migrate-config.py (SIM905); Project.tags annotated ClassVar (RUF012); Project.__new__'s first parameter renamed self → cls (PLW0211).
• compiler.py: nested with statements combined (SIM117). The blind except Exception is deliberate — the function's contract is to report any compile failure as a string — so it carries a noqa and a comment rather than a narrowed clause (BLE001).
• indent.py, migrate-config.py: unused unpacked variables renamed to _ (RUF059).

Behavior changes worth a look

• init-azure.py and macos-build-python.py pass check=True: their steps are sequential and dependent (configure → make; create project → create endpoint → create pipeline), so continuing after a failure only produced a confusing second failure (PLW1510).
• git.py's URLScheme.load and cpython-sync.py's final git commit pass check=False, since a non-zero exit is an expected outcome for both (no configured substitutions; nothing to commit). Both are commented.
• projects-run.py: the = [] typer defaults become | None = None, matching the style already in update-projects.py, with the two lists joined via itertools.chain (B006). Verified that --tag/--keyword still collect multiple values.
• github.py: Repo.get_public_key() becomes the Repo.public_key cached property (B019). functools.lru_cache on a method kept every Repo instance alive for the process and shared one key across repos; cached_property caches per instance. This is the one API break — happy to keep the method name via jaraco.functools.method_cache instead if you'd rather not spend a major version on it.

🤖 Generated with Claude Code

[jaraco]

CI note: everything passes except collateral (diffcov), which requires 100% coverage of changed lines. 19 of the 27 changed lines live in modules that had no coverage to begin with (init-azure, macos-build-python, migrate-config, compiler, indent), so touching them is enough to fail the gate. diffcov is also failing on main right now for the same reason, so this isn't a regression introduced here.

Also worth noting: .pre-commit-config.yaml still pins ruff-pre-commit at v0.12.0, so CI never saw these violations — they only appear with a local Ruff 0.16. I left the pin alone since it's skeleton-managed and jaraco/skeleton#210 declines to take on the bumping burden.

[jaraco]

I'm not loving this change. The prior implementation had a simpler signature and usage. Is there a way we can continue to use degenerate sequences instead of a separate None type?

[jaraco]

Agreed — reverted to the original signature and tag + keyword body in 8aece41, with a # noqa: B006 on each default.

I did try to keep it lint-clean without the suppression, but typer boxes you in: = () works at runtime (typer replaces it with a list either way, verified) yet mypy rejects a tuple default against a list[...] annotation, and typer can't consume the annotations that would fix that — Sequence[str] raises RuntimeError: Type not yet supported, and tuple[str, ...] raises Type not yet supported: Ellipsis. So list[X] = [] is the only spelling that satisfies both typer and mypy.

The suppression is honest here: typer always supplies the option value, so the default is never bound, let alone mutated — the same reason the skeleton already ignores B008 globally for = typer.Option(...). If you'd rather not carry inline noqas, B006 could join B008 in the skeleton's ignore list on the same rationale; happy to make that change instead.