spack update to 0.23.1#1413
Conversation
Seems to me that |
|
|
Issue was that spack no longer has version entry for |
|
a key difference between conduit and ascent spack package logic: this one line change made it so that the host config wasn't copied during the host config stage, but instead only during final install. This undermined the uberenv-pkg mode, which depends on the host config being generated earlier -- since it stops after the host config stage ... I added to add an extra copy to the derived uberenv-ascent package. Hunting this down took about a day of builds and sleuthing. Here is the ascent logic for ref: remaining failures: minimal failure: MPI is off, but we are trying to build conduit relay mpi for some reason In the conduit package proper in spack, This differs from ascent (another subtle difference that we should resolve) |
|
From the spack lock file: We are getting |
|
confirming it is HDF5 that is flipping MPI_FOUND to true: |
|
hdf5+mpi this was an issue with the spack spec not getting passed correctly in our ci |
|
More fun: Bummer about this message is that is is not shown in the console, I had to read the log to see what the underlying error was. I also think this is an error during pip install and solve -- not with spack constraints -- so this slipped past the spack defenses. I am going to swing to python 3.9. |
|
No more mysteries! It was the spack concretizer, with the candlestick, in the cloud. |
No description provided.