ZHF will be slightly different this year due to RFC 85. Most notable changes for contributors is that branch-off will be done on May 21st; so the first two weeks of ZHF don’t require backport PRs or commits (direct pushes are now discouraged so, we likely won’t see them again). This also means that there won’t be a beta for NixOS-21.05 until 21st as well.
Is there any mechanism by which one can say, “I’m working on this?” That was my main frustration with last ZHF, working on something, opening a PR after searching for duplicates, then while my own PR is being reviewed a new PR being opened for the same thing and getting merged more quickly.
I wish I had a better answer, but not currently. If we did want some notion of “this is assigned to me”, then we would have to do something like github issues per package, and that’s not really feasible.
One idea might be to create a separate repo just for ZHF 21.05 issues, with one issue per package, and no other content – just issues. That way the github-issue tooling (“assigned to”, etc) would be available without the chaos of doing it all in the main nixpkgs repo.
This is my first ZHF and I have some questions.
Is it possible to do some more advanced filtering on the hydra jobs?
For example I’d like to see failing python packages for x86_64-linux. The hydra website doesn’t seem to accept wildcard or regex. It would also be nice if it only showed the jobs with failures in their own build, not in dependencies.
Are there in general any recommended tools to support the workflow around ZHF?
Not possible, I’m afraid. Naturally you can display all and then do whatever with the output; IIRC there’s API providing the same information.
Note: the dependencies sometimes do not have their own jobs, though that’s relatively rare I think. And in my opinion, the more failures a dependency causes, the more attention it should get.
mkdir tmp
cd tmp
../eval-report $EVAL_ID
xdg-open index.html
Should generate an html document ordered by failing builds.
EDIT: the reason to make a seperate directory is that it will cache build information in $PWD, and it will quickly clutter it with 1000s of build reports.
Eventually I would like to make GitHub - jonringer/basinix: (WIP) Nix CI pull request review website into a polished review tool. Part of that would be carrying information about build status on different branches. Once I have a database of all the information, then I’m free to aggregate and present it many ways: failures for a maintainer, order by most downstream breakages, additonal filters by branch, etc.