Build path parallelism and "distribution"

Steps to get from a structured software environment description (i.e. a list of spack package dependency incl. version and variant constraints) to a deployed software environment can be parallelized and distributed:

• Separation into "connected" (cache interaction and downloading of spack package sources)
• and "isolated" tasks (building, packaging (image construction))

Open questions:

• Can we make use of spack env depfile to "dispatch" individual spack install (build task) steps?
  • job-local parallelism (→ -j)
  • Handling of per-package requirements — e.g., memory per core (i.e. memory_per_core * max_jobs < total_mem)
  • unicore?
• Can we automatize the "build task category" dependency handling? (e.g., via some build or "workflow" tool like snakemake?)