NCZarr fill_value is Nan instead of NaN on Windows

[malmans2]

In this xarray's PR I found out that I'm having issues reading with python NCZarr data created on Windows.
Looks like the issue has to do with fill_value, which is Nan instead of NaN, and therefore can't be properly parsed by python's json.

Sorry, I don't have much more details as I don't have a Windows machine to explore this a bit better. I thought I would open an issue here in case you already have some ideas about what's causing the problem. I guess it's possible that the problem is actually related to netcdf4-python as I'm using it to create the NCZarr file.

This is the Python code I'm using to create the NCZarr data

from xarray.tests.test_dataset import create_test_data
ds = create_test_data()
ds = ds.drop_vars("dim3")
ds.to_netcdf("file://test_data.zarr#mode=nczarr")

[DennisHeimbigner]

Looks like a typo on my part.
You should be able to fix as follows:

1. edit the file netcdf-c/libnczarr/zcvt.c
2. Near line 308 you should see this line: strcpy(s,"Nan"); break;
3. Change to this: strcpy(s,"NaN"); break;

I will submit a PR for this after I hear if it fixes the problem.

[malmans2]

I don't have an easy way to test that change on Windows.

However, I replaced myself all Nan in .zattrs and .zarray with NaN, and I can confirm that I'm able to correctly read the data. Therefore, I'm pretty sure that changing strcpy(s,"Nan") to strcpy(s,"NaN") will fix the problem and I suggest to go ahead with the PR.

[DennisHeimbigner]

See PR #2267