Bug in rt.sh jet section when ACCNR is not set

[climbfuji]

Description

A recent PR (#981) updated the jet section in rt.sh to allow setting ACCNR instead of using the hardcoded h-nems. A fallback/default value h-nems was intended in this part:

ACCNR=${ACCNR:-h-nems}
PARTITION=xjet
DISKNM=/lfs4/HFIP/h-nems/emc.nemspara/RT
dprefix=${dprefix:-/lfs4/HFIP/$ACCNR/$USER}

The problem is that bash gets confused by the - in the h-nems default value, which results in

dprefix=/lfs4/HFIP/nems/$USER

and causes rt.sh abort with

+ ACCNR=nems
+ PARTITION=xjet
+ DISKNM=/lfs4/HFIP/h-nems/emc.nemspara/RT
+ dprefix=/lfs4/HFIP/nems/emc.nemspara
+ STMP=/lfs4/HFIP/nems/emc.nemspara/RT_BASELINE
+ PTMP=/lfs4/HFIP/nems/emc.nemspara/RT_RUNDIRS
+ SCHEDULER=slurm
+ cp fv3_conf/fv3_slurm.IN_jet fv3_conf/fv3_slurm.IN
+ cp fv3_conf/compile_slurm.IN_jet fv3_conf/compile_slurm.IN
+ mkdir -p /lfs4/HFIP/nems/emc.nemspara/RT_BASELINE/emc.nemspara
mkdir: cannot create directory '/lfs4/HFIP/nems': Permission denied
++ echo 'rt.sh error on line 395'

To Reproduce:

Try running rt.sh on jet without setting ACCNR. Then, oops, look at the bug.

[junwang-noaa]

I got the error too, so I was setting the ACCNR to "h-nems". I think the ACCNR was set to nems by default.sh. Maybe we need to add a line in jet section in rt.sh to reset ACCNR to "h-nems" if it's "nems".

[climbfuji]

I got the error too, so I was setting the ACCNR to "h-nems". I think the ACCNR was set to nems by default.sh. Maybe we need to add a line in jet section in rt.sh to reset ACCNR to "h-nems" if it's "nems".

Your explanation makes a lot more sense than my assumption that the syntax could be messed up by a dash in the default value. I don't think there should be an entry in default_vars.sh at all, since the ACCNR depends on the HPC.

[SamuelTrahanNOAA]

If the ACCNR default should depend on the platform, then default_vars.sh should not set it. A special case of overriding nems with h-nems (even when it is set manually to nems by the user) seem sloppy.

Also, ACCR is the wrong variable, you should set ACCNR=h-nems in your script before running rt.sh. If you set ACCR, then the error is yours. Was that a typo in your title? Or an actual error in your execution of rt.sh?

[climbfuji]

Typo in the tile, fixed it.

[SamuelTrahanNOAA]

I will test a fix for this (move the default to rt.sh) and issue a PR tomorrow-ish. It'll have to be tested on all platforms, in both baseline generation and regression test mode. That means we should merge it with a PR that changes the results.

[climbfuji]

I will test a fix for this (move the default to rt.sh) and issue a PR tomorrow-ish. It'll have to be tested on all platforms, in both baseline generation and regression test mode. That means we should merge it with a PR that changes the results.

Thanks a lot for doing this, Sam! Yes, agreed, should be part of a baseline-changing PR. And needs to be tested manually as well, because auto-rt is always setting ACCNR (that's why the error didn't show up).

[junwang-noaa]

Thanks, Sam. According to current commit queue, it will be with PR#982 planned on 1/25.

[SamuelTrahanNOAA]

Barring extraordinary events, that should be doable.

[SamuelTrahanNOAA]

The fix is in #1016 but I only did minimal testing to avoid wasting resources. I tested enough to ensure ACCNR was passed correctly. Please give me another PR to merge into #1016 as soon as possible so I can do full testing.

There may be an unwanted side-effect: If someone was setting ACCNR incorrectly before running rt.sh, that incorrect value will now be passed through to the batch system. Make sure your automated test suites are setting the right ACCNR value, or none at all!

[SamuelTrahanNOAA]

@climbfuji When PR #1009 was merged, it should have fixed your problem. Can you confirm and close the ticket?

[junwang-noaa]

The autoRT runs fine on jet now. I will close the issue.