Installing pyobsplot fails in jupyterlite xeus-python kernel

[martenrichter]

I have tried to install pyobsplot with jupyter lite and the xeus kernel, and it fails because of the watchfiles dependency. Is it possible to make this dependency optional?

See:

The following packages are incompatible
└─ pyobsplot is not installable because there are no viable options
   ├─ pyobsplot [0.3.7|0.3.8|0.4.0] would require
   │  └─ polars >=0.16.0 , which does not exist (perhaps a missing channel);
   ├─ pyobsplot 0.4.1 would require
   │  └─ pyarrow >=13.0.0,<14.0.0 , which does not exist (perhaps a missing channel);
   ├─ pyobsplot 0.4.2 would require
   │  └─ polars >=0.20.0 , which does not exist (perhaps a missing channel);
   └─ pyobsplot [0.5.0|0.5.1] would require
      └─ watchfiles >=0.18.0 , which does not exist (perhaps a missing channel).

[juba]

I'm not really familiar with jupyter lite and the xeus kernel. Could you tell me how you are intalling pyobsplot ? Using an environment.yml file ?

[martenrichter]

Yes with envirnonment.yml : https://github.com/fails-components/jupyterfails/blob/master/environment.yml (if I add you package there it fails). Jupyter lite has only support for a limited number of binary packages. And it is super handy for setups in universities.

[juba]

Unfortunately I think that the main problem will be the absence of the polars package. If I add it to environment.yml it seems that it is not available.

It may have been possible to avoid the watchfiles dependency, but polars is really needed by pyobsplot...

[martenrichter]

OK, so it complains only about one random missing package, and Polars is still needed? Polars is a binary package? If so, then it is not as easy as I thought. Sorry...

[juba]

That's my understanding. It complains about watchfiles for pyobsplot 0.5.x, but I think it would complain about polars if watchfiles was available.

Polars is a package with an interface to a rust codebase, so it must indeed be a binary package.

[koaning]

To my knowledge polars now does support WASM. There are still some hiccups with pyarrow, but polars seems to run fine in many of my WASM marimo notebooks.

[mscolnick]

The typist package causes issues for wasm as well. I created a hack to get around that, and have a happy path rendering in wasm: https://marimo.app/l/hgstox

[juba]

pyobsplot 0.5.3.2 (just released) should now be installable in jupyterlite or marimo.