Trouble creating conda envs with geoviews=1.8.1

Since the recent upgrade to Holoviews=1.13.1, Geoviews=1.8…1 and Bokeh=2.0.0, I have frequent trouble to create and pinning envs with these packages.

The following somehow works:

conda install -c pyviz -c conda-forge geopandas fiona jupyterlab geoviews holoviews nbconvert descartes numpy mapclassify bokeh jupyter_contrib_nbextensions

but it will install Geoviews 1.6.6 and pyproj=2.6.0.

This also doesn’t work:

conda install -c pyviz -c conda-forge geopandas fiona jupyterlab "geoviews=1.8.1" holoviews nbconvert descartes numpy mapclassify bokeh jupyter_contrib_nbextensions

It will install pyproj=1.9.6, it also falls back to flexible solve and takes really long.

I was wondering if there’s an explanation for this behavior? How can I pin geoviews to 1.8.1 from the pyviz channel?

When using -c pyviz/label/dev, conda also falls back to flexible solve. The results seem to change pretty much every day.

I’ve tried with fewer packages, but still - solving the environment takes very long and it almost always falls back to flexible solve (my env is set to strict).

Here’s the complete order of steps to reproduce:

conda create -n geoviews_env python=3.7 -c pyviz -c conda-forge
conda activate geoviews_env 
conda config --env --set channel_priority strict
conda config --show channel_priority # verify
conda install -c pyviz -c conda-forge geopandas fiona jupyterlab geoviews holoviews nbconvert descartes numpy mapclassify bokeh jupyter_contrib_nbextensions

The documentation is very short when it comes to installation instructions, but I often encounter problems like these. Ideally, I want to provide a guide to the readers of my notebooks so that they can re-create an env that will be guaranteed compatible to the code in the Notebook.

• running conda 4.8.3 in Linux (WSL)

[philippjfr]

You should be able to drop pyviz entirely.

By the way, I never experienced any problems with proj4>6 (e.g. proj_create_from database: Cannot find proj.db) with conda and WSL (Ubuntu)

That’s not the only problem, certain projection operations return downright wrong results when combined with cartopy=0.17.

[philippjfr]

@Sieboldianus Try inverting the channels in that command, the conda-forge version does not pin proj in the same way as the pyviz channel version as I indicated above.