Instead of creating a job script which uses bash, instead internalise it in Python. The runcirrus script exists and is the same both on TGX and LSF cluster, so we can reuse it.
- User writes
runcirrus [..]
- Runcirrus executes
bsub -- runcirrus --job-dispatch [..]
- Runcirrus sees
--job-dispatch, and does what the job script does today (ie, execute mpirun).