[enhancement] support empty array var[ [] ]

[ryofurue]

For a regular vector, v[idx] "works" even if idx == Int64[]. This is a useful property to gracefully handle edge cases. For a netCDF variable, however, v[ [] ] fails with

ERROR: LoadError: MethodError: reducing over an empty collection is not allowed; consider supplying `init` to the reducer

It would be nice if an empty array is supported. The following is a self-contained example where it would be useful.

using Dates
using NCDatasets
using Plots

t0 = DateTime(2015,5,27)
t1 = DateTime(2015,5,28)
t2 = DateTime(2015,5,29)
t3 = DateTime(2015,5,30)

function makeplots(tax, v)
  ls = findall(t -> (t1 <= t <= t2), tax)
  @show ls
  plot(tax[ls], v[ls]; xlim=(t1,t3))
  sleep(3)
  ls = findall(t -> (t2 <= t <= t3), tax)
  @show ls
  display(plot!(tax[ls], v[ls]))
end

tax  = DateTime(2015,5,28):Dates.Hour(1):DateTime(2015,5,28,23)
myvar = rand(Float64, length(tax)) # fake values

# works
makeplots(tax,myvar) # works

# create a sample
NCDataset("tmp.nc", "c") do ds
  defVar(ds, "time", tax, ("time",))
  defVar(ds, "myvar", myvar, ("time",))
end

# fails
NCDataset("tmp.nc", "r") do ds
  makeplots(ds["time"], ds["myvar"])
end

[Alexander-Barth]

Thank you for preparing the reproducer; it helps a lot to find the error quickly!
I fixed this in the master version and added a test case of it.

d79a646

[ryofurue]

it helps a lot to find the error quickly

I didn't know that was an "error"! I meant to suggest that it would be nice if the empty-array behavior is emulated as an enhancement in the future because it's sometimes useful. (If I thought it was an error, I would have shortened my example, just to reproduce the error. I wanted to explain in what kind of situations the empty-array behavior may be useful.)

Anyway, thank you for your fast response!

[Alexander-Barth]

In fact, the old behavior was inconsistent (and empty range 1:0 would have worked, but not Int[]). The error message was just triggered in the checking assert code.

Thanks again!