New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Fix Travis CI builds #144
Comments
@czender but you seem to have fixed that |
You are correct that your OpenMP code causes TravisCI to fail on your branches not on master. We still need you to try to get OpenMP passing TravisCI on your branch before you merge it to master. It turns out that at about the same time that TravisCI started failing on your branch (because of OpenMP as shown in my link above)
, TravisCI also started failing on master because the default Travis build distribution was upgraded (I think) to Ubuntu Xenial and in any case the libnetcdf package name changed from libnetcdfc7 to libnetcdf11. This must be set in the file |
@czender #pragma omp parallel for private(idx, thr_idx) schedule(nonmonotonic: guided,20) shared(rtree, grd_lon_typ, bDirtyRats, bSort, max_nbr_vrl, pl_cnt_dbg, tot_nan_cnt, tot_wrp_cnt, pl_typ) The reason why I added the shedule parameter to the pragma is to stop the situation whereby each thread is assigned the same number of work chunks from the for loop. This results in the weird situation whereby all of the threads except one run out of work and then wait for the last one. to finish. By reducing the chunk size this situation is avoided. You can see situation evolving by looking at the thread number for each for loop iteration. ...Henry |
Suggestion: Use the |
It is fine to keep |
I am back from vacation. Have you noticed that all Travis builds since 3106 have failed?
https://travis-ci.org/nco/nco/jobs/565403229
Apparently this is due to your OpenMP changes, which include options not supported by all compilers so you probably need to add #ifdefs similar to those in nco_rgr.c.
Please make your #1 priority to fix Travis builds.
And then we need to move to a PR-only system for your modifications to master so that I can verify they do not contain breakage.
It is not OK to allow broken builds in master for > 24 hours, and success on AppVeyor only means that Windows builds fine, AFAICT, but Windows does not exercise the high-performance aspects of the code like OpenMP which must remain working on all the supercomputers we support with a variety of compilers and versions.
Please close this issue when Travis is working again and add a written acknowledgement that you will always submit PRs to change master in the future.
The text was updated successfully, but these errors were encountered: