We have observed some recent build failures (e.g., here on PR #251) that appear to be due to a failure of the CI scripts to clean up after themselves. Specifically, the offending build gave the following link error:
/usr/bin/ld: ./.libs/libtps.so: undefined reference to `GaussianInterpExtData::GaussianInterpExtData(mfem::ParMesh*, LoMachOptions*, TPS::Tps*)'
which is nonsensical b/c the GaussianInterExtData class was not introduced until PR #253.
But, these PRs did kick of the automated CI on marvin in such a way that if the CI from PR #253 ran and didn't clean up sufficiently well, a stale object from that build could have caused a problem for the next CI run on PR #251.
So, that is the current hypothesis. If correct, the fix is to make sure each build cleans up after itself.
We have observed some recent build failures (e.g., here on PR #251) that appear to be due to a failure of the CI scripts to clean up after themselves. Specifically, the offending build gave the following link error:
which is nonsensical b/c the
GaussianInterExtDataclass was not introduced until PR #253.But, these PRs did kick of the automated CI on
marvinin such a way that if the CI from PR #253 ran and didn't clean up sufficiently well, a stale object from that build could have caused a problem for the next CI run on PR #251.So, that is the current hypothesis. If correct, the fix is to make sure each build cleans up after itself.