On 29-11-2020 05:12, kefu chai wrote:
On Sun, Nov 22, 2020 at 5:45 AM Willem Jan Withagen <wjw@digiware.nl> wrote:
Hi,
Since we are adding build targets in github,
I have a question/favour to ask.... Would it be possible to also compile the sources with Clang.
As far as I'm concerned it does not have to be a blocker to get a PR merged. But it would at least signal that there could be something wrong with the PR, and it would be appreciated if the submitter looks into the errors Clang reports.
My FreeBSD ports depend on Clang compilation and I now spent a big part of the free time during the weekend fixing errors that Clang trips over. Not leaving much time to actually do development work.
Clang is available on Linux as well, so it would not even have to be a jenkins setup running on FreeBSD. Perhaps even better so, since there are also FreeBSD incompatabilities that would be hard for non-FreeBSDers to fix. Willem, thanks for bringing this up again. IIRC, we discussed this before. our jenkins perf test jobs are already using clang-10, see
- https://github.com/ceph/ceph-build/blob/master/ceph-perf-pull-requests/confi... - https://github.com/ceph/ceph-build/blob/master/ceph-perf-pull-requests/confi...
i think what you are after is the jenkins job triggered by a PR is created / updated. Exactly what I looking for, if only to warn submitters that there could be an issue. Some of the things I fix, are c++17 things that GCC does not mind but Clang signals as error (or warning). And that block my FreeBSD jenkins builds until I fix 'm. Which is not always easy since I need to figure out what the C++17 issue is, and what the author intentions are. And not all things are trivial, as you know.
i see couple options
- CI on GNU/Linux + Clang hosted in lab: 1. needs to setup a job in ceph-build 2. needs to allocate computing resources for running this job. - CI on FreeBSD + Clang hosted in lab: 1. needs to make sure we have all necessary patches which apply to master. 2. needs to setup a job in ceph-build 3. needs to allocate computing resources for running this job. 4. needs to have time/admin help on setting up / maintaining the FreeBSD build - CI on FreeBSD + Clang hosted in somewhere else (in your own Jenkins setup, for instance) 1. needs to setup a webhook notifying your jenkins 2. the admin needs to take care of the rest on the jenkins side (computing resources planning, jenkins setup, maintenance, etc)
i think all of these options above needs the help from the community. ATM I would be happy with the first, since it'll catch the difference in interpretations between GCC and Clang. Even more if it requires a pass before a PR can be committed.
The second one would be quite invasive since then it requires the maintenance I now do on my private Jenkins runs. And I doubt that there are enough people willing to maintain FreeBSD within the current ceph community. It would also require that I push more of the less important patches, and keep a sharper eye on things so FreeBSD building is not continuously failing due to a missing patch. The third one is partly already what is going on. Every 2 hours I sample github for changes in master, octopus, nautilus, and then build what is there. There is only one jenkins instance doing the CI. I'd have to work on figuring out how to build more pipelines to start working parallel. So sometimes I build a few PRs together. Which is fine for efficiency purposes, not for feeding back the build result of a specific PR. I sort of cheat when building, since I do build in a VM and destroy the VM everytime. So I have a compile time below 20 minutes. (At least before I need to start building Boost 1.74) And the build process is already split into building and testing as separate jobs. I would have a 4 node, 2 CPU, 128Mb mem system available to help with building.
also, the answer in an AMA from Sage 5 years ago still applies. i am quoting it for your reference:
We've merged patch series fixing the FreeBSD build in the past, but since we don't have CI tests that verify the build we inevitably break it. We're more than happy to take patches or time/admin help to get that set up in the lab so we can keep things working on an ongoing basis, though! Yup, I know. That was also his answer when I first asked him I he would be willing to take my PRs... And is also the reason why I running CI in a FreeBSD jenkins instance. FWIW, we still have a pull request adding clang build to gitbuilder, see https://github.com/ceph/autobuild-ceph/pull/22. the autoobuild-ceph scripts were used for building Ceph packages before we migrated to jenkins. somehow that pull request was not merged. i guess we just don't have the resource for building / testing using Clang by then. this might have changed now. Not sure if it makes sense to me, but I'll look into it.
--WjW
In the process it just might catch a possible bug, because Clang sees life different from GCC.
Like I said: it is a favour to ask.
Thanx, --WjW _______________________________________________ Dev mailing list -- dev@ceph.io To unsubscribe send an email to dev-leave@ceph.io