Ruff 0.17.5
Ruff 0.17.5 shipped on November 12, 2026, three weeks after 0.17.4. Call it a maintenance release with ambition: five new preview rules, five autofix repairs, one rule behavior change that will alter existing findings, and a large batch of performance work that the changelog does not bother to quantify.
If your CI runs Ruff with default settings, expect nothing to move. If you enable preview rules, or if you rely on B008 to flag function calls in argument defaults, read on.
Preview rules add real checks
Preview rules stay behind the preview flag, so this is opt-in surface area rather than a breaking change:
pytest-raises-exception-group(ASYNC401) arrives inflake8-async(PR #28772).no-method-argument(PLE0211) lands inpylint(PR #29748).renamed-rule(RUF202) is new (PR #30079), andredirected-noqa(RUF101) now expands to other suppression comments (PR #30078).- Renamed rules are now redirected automatically (PR #30081), which is the quiet quality-of-life win in this release: old
noqacodes for renamed rules stop silently doing nothing. G004inflake8-logging-formatis now marked as an unsafe fix (PR #30087), andrefurbdeclines to offer a fix whenwriteis nested in an expression (FURB103, PR #29789).
The rule change that will show up in diffs
flake8-bugbear now ignores frozen dataclass instantiations for B008 (PR #29231). That is a deliberate narrowing: constructing an immutable default was never the bug the rule was written to catch, and plenty of codebases carry noqa: B008 comments that exist only because of it. Some of those suppressions can go. For anyone who has B008 enabled and relies on it to flag dataclass defaults, findings will disappear and that is intended.
Autofix repairs
Five fixes addressed cases where the tool was confidently wrong, which is the category that costs the most time:
- Fixes that add or remove a keyword argument now handle a parenthesized value correctly (PR #30189).
ICN001inflake8-import-conventionsno longer shadows existing names when rewriting import aliases (PR #30035).B010autofix handles parenthesizedsetattrcalls (PR #30168).PYI034autofix handles parenthesizedselfannotations (PR #30173).UP031no longer duplicates closing parentheses around the format string (PR #30172).
There is also a crash fix worth noting for anyone piping Ruff output into a short-lived process: logging to a closed stderr pipe no longer takes the process down (PR #30209). That scenario is common when a parent process reads a few lines and exits, and it used to surface as a confusing panic rather than a broken pipe.
Performance, mostly unmeasured
The changelog lists eighteen pull requests under performance (PRs #30375, #30211, #30368, #30259, #30354, #30115, #30133, #30132, #30358, #30135, #30369, #30351, #30352, #30347, #30379, #30328, #30329 and #29446) without numbers. Given that Ruff already advertises editor-grade latency, the practical effect is small on a laptop and more visible in large monorepos where the linter is invoked repeatedly in a pre-commit hook. I would treat the claim as plausible rather than dramatic; the release notes give no benchmarks, and eighteen small PRs rarely add up to a step change.
Platform fix and upgrade
FreeBSD source builds failed against libc 0.2.190 and now compile again (PR #30176). Everything else is unchanged: bump the version with uv tool upgrade ruff, pip install --upgrade ruff, or your pre-commit hook revision. Twelve contributors are credited, including the usual core maintainers and a handful of first-time names.
For a point release, this is a reasonable payload. The caution is the same as always with Ruff: preview rules are preview for a reason, and the fixes are only as good as the tests around them, so read the diff before you commit an autofix run across a large repository.