[Suggestion] easier creation of time axis

[ryofurue]

Currently, it's a bit tedious to create a time axis for a new dataset. It would be nice if you could do something like

tax = DateTime(1990, 1, 1):Dates.Day(1):DateTime(2020,1,1)
NCDataset("newfile.nc"; cfconvention=true) do nc
  defDim(nc, "time", length(tax))
  v = defVar(nc, "time", ("time", ); like = tax) #←Asks to use `tax` to determine attributes
  v[:] = tax
  # ⇒ units = "days since yyyy-mm-dd HH:MM:SS" be automatically set.
  #      axis = "T" be automatically set.
end

Here, I expect that the NCDataset library would automatically choose an underlying data type (likely Float64) and an appropriate time origin and other standard and helpful attributes.

[Alexander-Barth]

This already works:

tax = DateTime(1990, 1, 1):Dates.Day(1):DateTime(2020,1,1)
NCDataset("newfile.nc", "c") do ds
  defVar(ds,"time",tax,("time",))
end

Essentially you give the data to defVar rather than just the type.

Dataset: newfile.nc
Group: /

Dimensions
   time = 10958

Variables
  time   (10958)
    Datatype:    DateTime (Float64)
    Dimensions:  time
    Attributes:
     units                = days since 1900-01-01 00:00:00

[ryofurue]

[Edit: a typo has been fixed.] Wow, that's nice! Very very nice!

The tutorial gives the impression that you need to define a dimension (defDim) before defining the corresponding dimension variable (defVar). That's the case for the original netCDF library.

But, NCDatasets library just internally defDims when a non-existent dimension is specified in defVar. Is that right?

So, basically you do away with defDim altogether!

xax = 0:2:358 # longitudes
tax = DateTime(1990, 1, 1):Dates.Day(1):DateTime(2020,1,1)
myvar = get_var()
NCDataset("newfile.nc", "c") do ds
  defVar(ds,"time",tax,("time",))
  defVar(ds,"lon",xax,("lon",))
  ds["lon"].attrib["units"] = "degrees_east"
  defVar(ds,"myvar", myvar, ("lon", "time"))
end

[Alexander-Barth]

Yes, this is correct.
I added this to the README.md
295a715

Note that you can also define the attributes with defVar (see the examples):
https://alexander-barth.github.io/NCDatasets.jl/stable/variables/#CommonDataModel.defVar

[ryofurue]

With this, defDim() is necessary only when you need a dimension without coordinate values. Perhaps you may want the "ensemble" dimension to be like that.