[bedroge]

Currently site installations seem to fail when building for a new CPU target, see EESSI/dev.eessi.io-riscv#136. This will hopefully solve it by creating the installation prefix, which should ensure that the EESSI module will not error out.

[bedroge]

Tested in EESSI/dev.eessi.io-riscv#136. Initially it already worked, but still tried to create some wrong paths as well:

mkdir: created directory '/cvmfs/dev.eessi.io/riscv/versions/2025.06-001/software/linux/riscv64/spacemit'
mkdir: created directory '/cvmfs/dev.eessi.io/riscv/versions/2025.06-001/software/linux/riscv64/spacemit/x60'
mkdir: created directory '/cvmfs/dev.eessi.io/riscv/versions/2025.06-001/software/linux/riscv64/spacemit/x60/modules'
mkdir: created directory '/cvmfs/dev.eessi.io/riscv/versions/2025.06-001/software/linux/riscv64/spacemit/x60/software'
mkdir: created directory '/cvmfs/dev.eessi.io/versions'
mkdir: created directory '/cvmfs/dev.eessi.io/versions/2025.06-001'
mkdir: created directory '/cvmfs/dev.eessi.io/versions/2025.06-001/software'
mkdir: created directory '/cvmfs/dev.eessi.io/versions/2025.06-001/software/linux'
mkdir: created directory '/cvmfs/dev.eessi.io/versions/2025.06-001/software/linux/riscv64'
mkdir: created directory '/cvmfs/dev.eessi.io/versions/2025.06-001/software/linux/riscv64/spacemit'
mkdir: created directory '/cvmfs/dev.eessi.io/versions/2025.06-001/software/linux/riscv64/spacemit/x60'
mkdir: created directory '/cvmfs/dev.eessi.io/versions/2025.06-001/software/linux/riscv64/spacemit/x60//modules'
mkdir: created directory '/cvmfs/dev.eessi.io/versions/2025.06-001/software/linux/riscv64/spacemit/x60//modules/all'

Fixed in 6a4f0ec by adding square brackets to that check. Confirmed by doing another build (EESSI/dev.eessi.io-riscv#136 (comment)) and that showed:

mkdir: created directory '/cvmfs/dev.eessi.io/riscv/versions/2025.06-001/software/linux/riscv64/spacemit'
mkdir: created directory '/cvmfs/dev.eessi.io/riscv/versions/2025.06-001/software/linux/riscv64/spacemit/x60'
mkdir: created directory '/cvmfs/dev.eessi.io/riscv/versions/2025.06-001/software/linux/riscv64/spacemit/x60/modules'
mkdir: created directory '/cvmfs/dev.eessi.io/riscv/versions/2025.06-001/software/linux/riscv64/spacemit/x60/software'

Also used EESSI_SITE_SOFTWARE_PREFIX instead of EESSI_CVMFS_REPO_OVERRIDE (in 0a53860) for constructing the software installation prefix, to solve:

Lmod file '/cvmfs/dev.eessi.io/versions/2025.06-001/software/linux/riscv64/spacemit/x60/.lmod/lmodrc.lua' does not exist yet; creating it...
python3 is /cvmfs/software.eessi.io/versions/2025.06/compat/linux/riscv64/usr/bin/python3
ERROR: Prefix directory /cvmfs/dev.eessi.io/versions/2025.06-001/software/linux/riscv64/spacemit/x60 does not exist!

Finally, tested with a regular build in EESSI/software-layer#1903 (comment), which still worked fine.

[casparvl]

Use of EESSI_SITE_SOFTWARE_PREFIX_BACKUP isn't super elegant, but I guess we have no other choice. Also, the test build at EESSI/software-layer#1903 (comment) doesn't use the latest commit in this PR. However, I checked and the subsequent two commits only affect things inside if-statements directly specific to site builds, so to speak with @ocaisa : I'll close my eyes to this and approve the change. Agreed with @bedroge that he'll buy everyone a beer if this breaks it.

On the whole, I think this is a good fix to something we overlooked when first implementing the site-building functionality. Thanks @bedroge