KeyError: key Date not found

[Datseris]

Describe the bug
Trying to write a dimension with values Date("2000-01-01"):Month(1):Date("2019-12-01") to an .nc file I get

 KeyError: key Date not found

which happens in accessing ncType[vtype] in the definition of defVar.

The fix is to add an entry to this "dictionary ncType that allows Date and it handles it the same as DateTime.

If I transform my data into DateTime it works.

I can't even find where in the source ncType is defined though... Where is it defined? this seems like a simple fix.

[visr]

Are you looking for this?

NCDatasets.jl/src/dataset.jl

Lines 103 to 120 in b8751a9

	# Mapping between NetCDF types and Julia types
	const jlType = Dict(
	NC_BYTE => Int8,
	NC_UBYTE => UInt8,
	NC_SHORT => Int16,
	NC_USHORT => UInt16,
	NC_INT => Int32,
	NC_UINT => UInt32,
	NC_INT64 => Int64,
	NC_UINT64 => UInt64,
	NC_FLOAT => Float32,
	NC_DOUBLE => Float64,
	NC_CHAR => Char,
	NC_STRING => String
	)
	# Inverse mapping
	const ncType = Dict(value => key for (key, value) in jlType)

[Datseris]

I guess so, but now it is not so clear to me how to modify it so that it can handle Date, because I don't see DateTime in it...

[visr]

Ah perhaps this needs to include Date?

NCDatasets.jl/src/cfvariable.jl

Lines 174 to 183 in 947db89

	# data has the type e.g. Vector{DateTime}, Array{Union{Missing,DateTime},3} or
	# Vector{DateTime360Day}
	# Data is always stored as Float64 in the NetCDF file
	function defVar(ds::NCDataset,
	name,
	data::AbstractArray{<:Union{Missing,nctype},N},
	dimnames;
	kwargs...) where nctype <: Union{DateTime,AbstractCFDateTime} where N
	_defVar(ds::NCDataset,name,data,Float64,dimnames; kwargs...)
	end

Probably with a new test case in test_timeunits.jl.

[Datseris]

YeaP, this seems like it!

[visr]

I guess for full Date support CFTime.jl would need to support Date as well as DateTime. Though there may be a reason this is not supported, as time in NetCDF is more of a continuous variable, and fit better with DateTime. Unless we decide that Date represents 00:00.

[Datseris]

For me it seems reasonable and straight-forward that assuming 00:00 should happen, and Base Julia does the same:

julia> using Dates

julia> x = Date("2000-01-01"):Month(1):Date("2019-12-01");

julia> DateTime.(x)
240-element Array{DateTime,1}:
 2000-01-01T00:00:00
 2000-02-01T00:00:00
 2000-03-01T00:00:00
 2000-04-01T00:00:00
 2000-05-01T00:00:00
 2000-06-01T00:00:00
 2000-07-01T00:00:00
 2000-08-01T00:00:00
 2000-09-01T00:00:00

[Alexander-Barth]

I agree that this would be a nice feature. I do not need this myself, but I would be happy to review a PR.