manipulateConfig.R does not warn if a supplied key is missing

[mikapfl]

Hi,

I'm not quite sure if this is a bug or intended behaviour: If I call manipulateConfig on an empty file, it will just be happy and not report any error or warning. This is a problem if there are typos in the config to be manipulated (or in my new config keys), or if I introduced a new config switch - both will just happily do nothing and report nothing for the affected keys.

Probably, we would get quite a lot of false positives if we added a warning at the moment because not all settings actually end up in main.gms in normal operation, but I think we should try to find a solution, silently doing the wrong thing / nothing is a bad response to a typo or missing definition.

[tscheypidi]

Currently it is indeed the case that the function is applied on files where it is expected that not all settings actually show up. Hence, it would be quite complicated to introduce a note or warning here.
Perhaps the function could return a list of setting which were found and changed in the corresponding file? This could then be used to trigger such warnings for the not found settings, if the application allows for it. What do you think?

[mikapfl]

Either that, or add an extra argument unchangedOkKeys or so, and if unchangedOkKeys is not Null and anything is not changed, and not in unchangedOkKeys, raise the warning.
Then the warning-raising would be a service of manipulateConfig and it we would have less duplication of code in the applications (just need to figure out a list of keys where it is okay not to change them, not also do the warning-raising).

[orichters]

Here is an example where this actually lead to wrong model behavior:

In REMIND, @LaviniaBaumstark and I found that main.gms defined

$setGlobal cm_EsubGrowth  low

but the default.cfg defined cfg$gms$cm_esubGrowth with a small e. The substitution done by manipulateConfig resp. manipulateFile just dropped this setting without any notification or so, so the settings were simply not applied.

See mismatch here:

> cd /p/tmp/oliverr/NGFS_v3/2022-05-22/remind/output/C_h_ndc_bIT-rem-5
> grep cm_[eE]subGrowth cfg.txt full.gms
cfg.txt:  cm_esubGrowth: middle
full.gms:$setGlobal cm_EsubGrowth         low  !! def = low

Looking at the structure of the code, I don't see an easy option to fix that and issue warnings etc., because manipulateConfig simply defines lots of manipulations with the intent that lots of them will never be useful, so how could manipulateFile where the manipulations are applied know which of them are problematic. But maybe @dklein-pik or @tscheypidi have an idea? Or may it be simply be useful to do the replacement not case-sensitive, given that GAMS is case insensitive anyway.

Still, the problem should be much reduced in REMIND simply because we don't store much in default.cfg anymore and mismatches between main.gms and default.cfg should be substantially reduced.

[orichters]

I had a suspicion, and indeed. Since the switches were moved to main.gms in REMIND, there is indeed a warning if you run ./start.R -it config/scenario_config_NGFS_v3.csv:

Automated checks did not find counterparts in default.cfg for these config file columns:
  cm_esubGrowth

So it is indeed only relevant for mismatches between default.cfg and main.gms (which affects only a few things in REMIND, maybe more in MAgPIE)