Upgrade ruff to 0.16.0

[davegaeddert]

Upgrades ruff 0.12.0 → 0.16.0 (lockfile), and raises the plain.code floor from ruff>=0.1.0 to ruff>=0.16.0.

Ruff 0.16 changes two defaults that plain.code inherits. Because ruff_defaults.toml ships to every Plain project and the old floor was >=0.1.0, both would have reached users through a transitive ruff resolve without them changing anything — so most of this PR is pinning the config so the shipped rule set stops moving with ruff's.

Default rule set: extend-select → select

Ruff 0.16 grew its built-in default selection from 59 rules to 413. ruff_defaults.toml used extend-select, which layers on top of that default, so plain code check went from 155 rules to 445 and reported 1474 errors across this repo (RUF012 ×427, DTZ001 ×96, BLE001 ×94, …).

Spelling out the old base (E4, E7, E9, F) alongside the existing category selectors makes the shipped set version-independent. The category selectors stay whole — "UP" still means every stable UP rule, not the subset ruff now defaults to — so upgrades continue to pick up new rules inside a category we've opted into.

Net rule change is what a version bump should be:

Rule	Change
UP038	removed upstream
UP042, UP050, ISC004	added

Markdown

Ruff 0.16 lints and formats Python code blocks inside Markdown, on by default. Left alone it reformats historical CHANGELOGs and doc examples that are deliberately partial. ruff_defaults.toml now sets extend-exclude = ["*.md"].

This also keeps a single owner for Markdown in this repo — scripts/fix already runs prettier with --embedded-language-formatting off. Documented as an FAQ in the plain.code README, since users reading ruff's release notes may expect the new behavior.

isort and the plain namespace

Ruff 0.16 resolves namespace packages per-submodule. In a checkout with part of plain on disk (this monorepo — plain/ sits at the project root), bare plain sorted as first-party while plain.postgres stayed third-party, splitting one namespace across two import blocks.

known-third-party = ["plain"] makes grouping identical everywhere. It's a no-op for projects that install Plain from PyPI, where ruff already resolves plain as third-party. This is what regroups imports across 230+ files here — mechanical, and the result now matches what a Plain app sees.

Violations fixed

• ISC004 — wrapped implicit string concatenations inside collection literals (plain-postgres health runner, tools/server-worker-test.py).
• UP042 — suppressed on TextChoices. It can't become an enum.StrEnum without losing its deliberately overridden __str__ and _generate_next_value_.
• One formatter improvement in plain-postgres/plain/postgres/expressions.py (lambda + conditional parenthesization).

Verification

• ruff check / ruff format --check / ty check / annotations — all pass.
• Full test suite passes (~2,000 tests, 1 pre-existing skip).
• oxlint/oxfmt were not exercised — the binary download fails TLS in the sandbox. Untouched by this change.

Follow-up worth considering separately

Ruff's new defaults are a curated set aimed at severe issues, and a lot of it looks worth having (RUF012 mutable class defaults, DTZ naive datetimes, BLE001 blind excepts). Adopting any of it means fixing hundreds of real violations, so it should be a deliberate pass rather than a side effect of a version bump.

---

Generated by Claude Code