Closed bangerth closed 12 months ago
I think what you see is a combination of two error messages. The first is ctest failing, in this case because of the multigrid failures with deal.II dev. For example the first part of the message is not present in the passing 9.5 tester output: https://github.com/geodynamics/aspect/actions/runs/7118963444/job/19383040245?pr=5511#step:7:18).
The second error that is a real problem is the line with warning: Not a git repository.
and below. I am relatively certain that this is the issue I tried to fix in #5499. I didnt realize that generate_reference_output
already calls git diff
, so I think the change from #5499 may have to move up one line to fix this. The problem as far as I understand is that the git version on the github actions runner that checks out the repository is slightly incompatible with the git version inside our tester container, and the version inside the container does not recognize the repository as safe, which leads to it refusing to do commands like git diff
. I can prepare a PR, which will hopefully fix the issue.
In our CI steps (see for example https://github.com/geodynamics/aspect/actions/runs/7118963444/job/19383040624?pr=5511), we break running and reporting tests into several sub-steps. In the "Write test results" step, we always get this error:
This is from the
cmake/generate_reference_output.sh
script, which is called by thegenerate_reference_output
target inCMakeLists.txt
:Note that it is called in the build directory.
The script itself looks like this:
I think that we ought to be ending up in the top-level source directory, but
git diff
nevertheless says that this is a git directory in the error message. Does anyone see what is going on here? Are we only pulling from a git repository instead of creating a local clone?