Only one variant of mumps-mpi is created about mumps-feedstock HOT 5 CLOSED

@minrk, any ideas how to fix this?

from mumps-feedstock.

That's strange. Looking into it now. The two builds have different host dependencies, so they shouldn't have the same hash.

I need to figure out how conda_build computes the hash, since it's getting the wrong inputs.

from mumps-feedstock.

I've found the first culprit: inspecting the hash computation shows that mumps-mpi is actually using the contents of mumps-seq to compute hash inputs, so it excludes mpi from used_vars. Not sure exactly why that's happening.

from mumps-feedstock.

I tracked it down. conda_build gets super confused when the existance of outputs is gated by conditionals, which this recipe does. Replacing those conditionals with build.skip = True seems to fix it. PR on its way.

from mumps-feedstock.

Thanks @minrk

from mumps-feedstock.