Cftime and netCDF4 version on windows

Package Builders conda-forge
1

When we installed netCDF4, we got old version of cftime.

Creating this environment:

conda create -n junk python=3.9 netCDF4
# Name                    Version                   Build  Channel
netcdf4                   1.5.7            py39h3de5c98_1
# Name                    Version                   Build  Channel
cftime                    1.5.1.1          py39h080aedc_0

But cftime 1.6.2 exists,
If we do ‘conda update cftime’, it won’t update it.
But, ‘conda install cftime=1.6.2’ will upgrade cftime,

The following packages will be UPDATED:

  cftime             pkgs/main::cftime-1.5.1.1-py39h080aed~ --> conda-forge::cftime-1.6.2-py39hc266a54_1
  vs2015_runtime     pkgs/main::vs2015_runtime-14.27.29016~ --> conda-forge::vs2015_runtime-14.32.31332-h1d6e394_9

, and netcdf4 and cftime all work together with the new version.
Note the combination of netcdf4 and cftime is broken.

The following line of code doesn’t work:

time_2 = cftime.num2date(7020., 'minutes since 1999-11-25 00:00:00')
print('time_2XXXXX:  ', time_2)

the error message is shown below:

Traceback (most recent call last):
  File "C:\Users\leo.geng\learning_notes\PyGNOME_error_test\test.py", line 12, in <module>
    time_2 = cftime.num2date(7020., 'minutes since 1999-11-25 00:00:00')
  File "src\cftime\_cftime.pyx", line 535, in cftime._cftime.num2date
  File "src\cftime\_cftime.pyx", line 379, in cftime._cftime.cast_to_int
TypeError: '<' not supported between instances of 'numpy.longdouble' and 'int'

This code will work if we update cftime to 1.6.2 version.

2

Note that the Channel column in the output from your conda create call is left blank.
This means the defaults channel is used. (defaults::cftime, i.e., pkgs/main::cftime is not yet updated to >=1.6)
If you want to install packages from conda-forge, you should give that channel higher priority (e.g., with the --channel=conda-forge CLI parameter) and/or only use one of conda-forge/defaults (via conda config --remove channels or conda config --set channel_priority strict or similar).
See

3

Thanks – we did notice that after posting – and channel priority was set to flexible.

If you install with -c conda-forge it does work.

However, I’m pretty sure is still didn’t work with channel priority strict – I’m not on Windows right now, so can’t check for sure.

And that’s an issue because we need to install a whole pile of stuff, most, but not quite all is available on conda-forge.

Still trying to understand the channel priority settings :frowning:

4

OK, more careful testing – if conda-forge is highest priority, and channel priority is strict, then it all works as expected.

Thanks!