Notify significant projects that still refer to python-igraph instead of igraph

[szhorvat]

On PyPI, python-igraph was renamed to igraph in October 2021. The standard way to install igraph is now:

pip install igraph

The python-igraph name was kept functional, with the intention to terminate support in September 2022. However, as of today (July 2023), over a third of downloads are still through python-igraph:

• https://pypistats.org/packages/igraph
• https://pypistats.org/packages/python-igraph

Here's the the fraction of python-igraph downloads based on a 2-week moving average:

This issue is to notify downstream igraph users (packages, educational resources, etc.) to start using the new name.

[szhorvat]

• py4cytoscape, see https://github.com/search?q=repo%3Acytoscape%2Fpy4cytoscape+python-igraph&type=code
• scanpy
• pygraphistry, fix: change python-igraph package name to igraph in setup.py graphistry/pygraphistry#494
• scvelo, python-igraph is renamed to igraph theislab/scvelo#1099
• plotly, python-igraph was renamed to igraph plotly/plotly.py#4285
• https://github.com/cctbx/cctbx_project/, see https://github.com/cctbx/cctbx_project/blob/master/mmtbx/domains_from_pae.py#L199
• starcal, change python-igraph to igraph ilius/starcal#72
• scikit-multilearn, refer to igraph instead of python-igraph in docs scikit-multilearn/scikit-multilearn#299

[szhorvat]

Update on the fraction of python-igraph downloads. The following show the fraction based on a 2-week moving average, as well as daily data.

There is a steady improvement, and I assume the situation will continue to improve as more of the above PRs get merged and as more downstream packages release new versions. Closing this issue now.

[szhorvat]

Updated plots as of today:

It's strange to see the mass of python-igraph downloads in March. Notice that the "fraction" value goes above 1 here. The fraction is calculated as python-igraph downloads divided by igraph downloads, with the assumption that each python-igraph download also triggers an igraph downloads due to the dependency between the two. Since this ratio went above 1, there must have been lots of python-igraph-only downloads for some reason. Otherwise we'll still hovering just below 30%.

[ntamas]

Caching may screw the stats up. pip has a local cache of downloaded wheels on each machine, on a per-user basis. If the user has already downloaded igraph==x.y.z due to some project and another project that depends on python-igraph==x.y.z is installed, this will trigger a download for python-igraph==x.y.z but not igraph==x.y.z since that is already in the cache. Institutions may also use local institution-wise PyPI caches (like devpi).