Noble testing #391
Noble testing #391
Conversation
Let's see what breaks
* remove CsrMatrix workaround that breaks with latest scipy * undo python 2->3 workaround in python-c interface in libspud/python/libspud.c (NOTE:needs backporting into spud) * numpy.float -> numpy.float64 * loosen tolerance in mms_tracer_cv_3d_adv_excludemass * don't use "from scipy import *" to import numpy things * fladapt appeared to call python_init twice which breaks things
(was intended to be in last commit)
Most of these are not actually relevant for fluidity but cleaner to keep exact copy. Note recent changes in libspud/python/libspud.c will reversely be committed to spud.
Might break with vtk 7?
These are only in diamond, so not actually used in fluidity build
Switching back to SOR smoother nearly gives old behaviour. Probably need old MIS clustering (which isn't available in 3.19) to get exactly same convergence count as before. Also remove petsc memory usage tests - this relies on a specific output format in -log_view that's a bit hard to maintain between different versions.
Fortran fieldsplit interface appears broken in 3.23
PETSc should filter these out, but for these rows sometimes the column indices are not initialised and this produces lots of error messages when petsc is built with debugging.
Not sure this is appropriate
Can't get these to work on petsc 3.15 (noble) due to GAMG shenanigans Tests pass on petsc 3.24 NOTE: this commit should be reverted as soon as we have a newer version of petsc in CI
for more information, see https://pre-commit.ci
It appears Vec_Ignore_Negative_Entries does not get copied in duplicated vectors of type VecSeq
These should also be skipped by petsc in MatSetValues, but if the column indices are wrong it still complains.
This is a.o. to ensure h5cc can provide path to hdf5 libs in configure.
Test appears to be flaky: range of different values upon rerunning...
This is to avoid linking to both the packaged hdf5 through the packaged netcdf, and the hdf5 build by petsc.
Patol75
left a comment
Patol75
left a comment
There was a problem hiding this comment.
That is an impressive effort to keep updating Fluidity and making it available on newer environments! Sorry for not finding the time to contribute any fixes here. To try making up for it, I reviewed the changed files, and everything seems very reasonable to me. I was stunned that only a limited number of tests required amendment of their diagnostic values. That is awesome! Perhaps I just had one question about why the GAMG tests have been removed from the Stokes_may_ tests.
While I agree we should try to target Ubuntu packages (and surely that will be the case in 4 months for 26.04 LTS), I also find it very useful to include building instructions for VTK and PETSc in the Dockerfile!
Finally, if I read the latest Actions run correctly, there is only rotating_channel left failing. Again, super impressive job!
Let's see what breaks