Tracking implicit dependencies

[jlstevens]

Originally HoloViews was only targeting matplotlib and as a result we were happy to depend on a few things that matplotlib depended on. Likewise, IPython is now Jupyter and has split into a number of repositories.

Here are some dependencies we need to worry about for people using bokeh without matplotlib:

• pyparsing is need for the %%opts magic (included if matplotlib is installed).
• Annoyingly you now need conda install nbformat for the NotebookArchive (used to be included).
• jinja2 (although this is still a dependency of notebook) so only needed if you want to run unit tests without notebook installed.

I think that is all of them.

Edit: I didn't look carefully enough - it seems nbformat is included with notebook. So I suppose the key point is that notebook is required to run the unit tests.

[jbednar]

So, do we need to have two conda packages for HoloViews, one relying only on Param (and numpy?), and the other building on that to add these non-core dependencies?

[jlstevens]

So, do we need to have two conda packages for HoloViews, one relying only on Param (and numpy?), and the other building on that to add these non-core dependencies?

I suppose so. I just wish conda supported something like extras_require!

[jbednar]

Agreed. I was just talking to the conda people about noarch support, another glaring problem, and it's clear that they are overloaded right now, so I wouldn't expect it to get better soon!