It seems less likely that one would want to apply common attrs to a var in several files and then rename it.

Agreed.

Regarding renaming vars and dims, did you consider defining that in meta.yaml instead?

It feels like a natural match. The command line options are mostly switches, or involved with the processing of specific files whereas the meta.yaml files are reusable elements and are amenable to more complex data structures.

One implementation option is to add a rename key, e.g.

global:
    Tiddly: "winks"
    Publisher: "A long impressive sounding name"
    unlikelytobeoverwritten:
    notactualattribute:
rename:
    variables:
        time_bounds: "time_bnds"
    dimensions:
        nv: "bnds"
variables:
    temp:
        Tiddly: "winks"
        units: "Kelvin"
        _FillValue: 

Obviously it would be possible to just have the rename key and reuse that where appropriate

rename:
    dimensions:
        nv: "bnds"
    variables:
        time_bounds: "time_bnds"

Regarding renaming vars and dims, did you consider defining that in meta.yaml instead?

Great idea!

Putting these in the .yamls will be easier to find and maintain I think.

Perhaps we should allow renaming to not crash if old name doesn't exist? That would allow us to have a single time.yaml file for all of the timely files in esm1.6:

rename:
  variables:
    time_bounds: time_bnds

variables:
  time:
    axis: "T"
    bounds: "time_bnds"
    calendar: "proleptic_gregorian"
    calendar_type:
    long_name: "time"
    standard_name: "time"
  time_bnds:
    axis:
    calendar:
    long_name:
    standard_name:
    units:
    missing_value:
    _FillValue:

time_bounds won't exist for atmosphere and ocean files. addmeta could just print a warning and move on, in it's current state it would crash out.

Perhaps we should allow renaming to not crash if old name doesn't exist? That would allow us to have a single time.yaml file for all of the timely files in esm1.6

Good idea right back at ya!

Agreed, sounds good, and is consistent with how the rest of the tool works.

I think it is good for addmeta to be as fault tolerant as possible, as we would rather have slightly incorrect metadata than crash a climate run. Ensuring correctness should be a separate step, with a considered decision about how to remedy any problems.