Define vl-convert-python dependency outside of altair[all]

[apoorvkh]

What is your suggestion?

The vl-convert-python package is necessary for saving plots in PNG / SVG / PDF formats. I believe this is a rather critical function, so I'm wondering why vl-convert-python needs to be manually installed for this capability. (I see that altair[all] is an alternative, but I think vega_datasets, etc. are less critical than the save functionality.)

vl-convert-python has zero dependencies. So I think this would be a simple addition. Maybe that is the reason Is this dependency optional because the built wheels are ~30 MB? Could this size be reduced?

Thanks so much for your consideration!

Have you considered any alternative solutions?

Alternatively, could this (at least) be added to a new optional dependency group altair[save], as suggested here?

[jonmmease]

Hi @apoorvkh,
As you noticed, vl-convert-python is a native compiled wheel, and it's fairly large. Both of these are consequences of the fact that vl-convert embeds the V8 JavaScript runtime (through the Deno project) in order to run the Vega-Lite and Vega JavaScript libraries without external dependencies. There is no obvious way, that I'm aware of, to reduce the size of vl-convert without reducing functionality.

This can't be a required dependency of the altair package, because it would prevent Altair's use in architectures that aren't supported by Deno, but do run Python. A prominent example of this is Pyodide, for running Python in the browser.

I don't have a problem with adding an additional dependency group, but could you say more about why you wouldn't want to use the all group?

[apoorvkh]

Thanks a lot for the explanation!

could you say more about why you wouldn't want to use the all group?

I just want to use Altair to plot / interact with my data and save to a static file, so I can put figures in my academic papers. I simply don't need any of these (besides vl-convert-python)! So I prefer not to depend on them (if I can) to avoid potential conflicts. And I am just bringing this up, because I feel that saving plots to static files is a rather core functionality.

altair/pyproject.toml

Lines 57 to 66 in 712680b

	all = [
	"vega_datasets>=0.9.0",
	"vl-convert-python>=1.6.0",
	"pandas>=0.25.3",
	"numpy",
	"pyarrow>=11",
	"vegafusion[embed]>=1.6.6",
	"anywidget>=0.9.0",
	"altair_tiles>=0.3.0"
	]

E.g. not even pandas because I am using polars; surprisingly, not even numpy, because torch seems to cover everything I need that is ML-related.

[apoorvkh]

I made a simple PR with the change, in case this is sufficiently compelling to you. Please let me know, thanks!

#3609