Preprocessing routines fail if there is a dot in the config "keys"

[earosenberg]

I found that preprocessing routines fail when there is a dot in one of the "keys" of the preprocessing config.

Example: I was trying to run preproc_or_load_group. One of my entries in the config looked like

name: "atomic_info"
process:
    noise_ratio_Q.rmean: rmeanQ

This caused an error on line 1118

sotodlib/sotodlib/preprocess/preprocess_util.py

Line 1118 in ff83638

proc_aman.save(outputs_proc['temp_file'], outputs_proc['db_data']['dataset'], overwrite)

It seems the problem is actually in get_pcfg_check_aman which is trying to save noise_ratio_Q.rmean as an AxisManager with the single entry rmeanQ. It successfully does this but later when you try to save the aman, it thinks noise_ratio_Q.rmean is looking for a child of another aman noise_ratio_Q and throws an error.

I've done a workaround (swapping the key/value in my pair the order didn't matter) but thought best to flag.

[msilvafe]

What is the "atomic_info" process? I'm not sure we want to allow the keys to have dots in them. I suppose if we have restrictions like this we should have a checker that rejects a config file if names violate our rules.

[earosenberg]

atomic_info adds things to a particular aman for later merging into the atomic db, added in #1406 (but likely to change as you requested there).
Having such a restriction seems fine but agree that a checker would be good as the current error is pretty opaque. Might (also?) make sense to have a check at the AxisManager level to throw an error when you create a sub-aman with a dot name in the first place

[msilvafe]

Just checking on this, do you know if this error arises at the yaml loading level or when passed in as a field to aman or specifically when it tries to do the pcfg check? I can try to implement a fix for this but if I could start with a simple notebook or script from you to trigger the error that'd be helpful. Thanks.