The unused import check is honestly doing more harm than it does good, and we should seriously consider getting rid of it altogether.
An unused import is benign. But that check has generated an unjustifiable amount of extra work in fixup/ninja/force-psuh commits. > On 29 Sep 2025, at 14:03, Josh McKenzie <[email protected]> wrote: > >> Last I remember the check was part of the CI run. Why do we need a separate >> script now? >> I feel I am missing something here. > It's common for people to tweak something right before commit and merge based > on reviewer feedback w/out fully re-running all of CI; CI is heavy. And it's > common enough to forget to re-run check to look for unused imports. > > To reinforce my earlier point: if we're not using merge functionality on > github we have to rely on committers remembering to either a) run "ant check" > or b) check the PR for hooks to make sure ninja edits didn't break the build. > > While I appreciate any movement in this direction, I don't think this will > ultimately "solve" the problem; we'd need to either move to PR-based merges > with automated CI hooks or move to a staging branch that has to pass some > subset of CI before merging to trunk. Or remove the unused import check. > > On Sun, Sep 28, 2025, at 3:59 PM, Jordan West wrote: >> +1 >> >> On Sat, Sep 27, 2025 at 05:49 Ekaterina Dimitrova <[email protected] >> <mailto:[email protected]>> wrote: >> Last I remember the check was part of the CI run. Why do we need a separate >> script now? >> I feel I am missing something here. >> Is it because CI run may have been run not on the final rebased version >> (missing 1-2 unrelated patches) and as a final stop gap? >> >> Also, thanks for always trying to improve and simplify our pipelines, Mick! >> >> On Fri, 26 Sep 2025 at 19:06, Francisco Guerrero <[email protected] >> <mailto:[email protected]>> wrote: >> Nice! +1 >> >> On 2025/09/26 17:48:28 Mick wrote: >> > I'd like to propose the following patch: >> > https://github.com/apache/cassandra/compare/trunk...thelastpickle:cassandra:mck/gha-lint >> > >> > It's a one-liner `check-code.sh` and is light and quick to run at ~6 >> > minutes. >> > >> > We are supposed to always run `ant check` (what check-code.sh does) before >> > merging, but it's understandably an easy thing to forget now and again. >> > Such a lightweight gh action will I believe prevent most of the breakages, >> > since many of us push our dev branches to gh. I'm not in favour of making >> > it any heavier (that can be a seperate discussion for latter), nor at >> > first trying to change/improve our checkstyle rules (that too can be done >> > separately). >> > >> > >> > Example result run here: >> > https://github.com/thelastpickle/cassandra/actions/runs/17880689988/job/50847932781 >> > >> > Any objections ? (a jira ticket will be created) >> > >> > regards, >> > Mick >
