Some examples might require additional dependencies

[HealthyPear]

Bug report

Bug summary

I noticed this while trying the external package yt-idv, but I thin it might be a bug here, as it seems related to
the loading of sample datasets from yt itself.

It might also be a shortcoming in the documentation in this or the other project.

Feel free to move the issue there if this is not the case.

Code for reproduction

import yt

import yt_idv

ds = yt.load_sample("IsolatedGalaxy")

rc = yt_idv.render_context(height=800, width=800, gui=True)
sg = rc.add_scene(ds, "density", no_ghost=True)
rc.run()

Actual outcome

ModuleNotFoundError: No module named 'pooch'
Something went wrong while trying to lazy-import pooch. Please make sure that pooch is properly installed.
If the problem persists, please file an issue at https://github.com/yt-project/yt/issues/new

and then in series the same for pandas and h5py respectively.

Expected outcome

The example should work immediately (and it does, when those 3 dependencies are installed).

Version Information

• Operating System: macos 14.7
• Python Version: 3.12.7
• yt version: 4.3.1 via conda
• Other Libraries (if applicable): yt-idv 0.4.1 via PyPI

[welcome]

Hi, and welcome to yt! Thanks for opening your first issue. We have an issue template that helps us to gather relevant information to help diagnosing and fixing the issue.

[neutrinoceros]

yes, pooch and pandas are both requires for yt.load_sample, and h5py is needed for "IsolatedGalaxy", which is an enzo data sample. Is the error message unclear ?

I can see a couple ways we could maybe improve this experience:

• clarify the error message (how ?)
• stop requiring pandas in load_sample (I'm convinced it's doable, but I never managed to do it)
• implement a mechanism to report all missing dependencies at once instead of letting the user discover them iteratively (which I agree is bad UX). This is a very hard problem in the general case but maybe we could do something specifically for load_sample, which is now very commonly used in documentation.

[HealthyPear]

The error message is clear enough, the problem is that it is unexpected.

What I would do is this:

• add an extra set of options into pyproject.toml called examples
• add this set of optional dependencies to the next conda recipe by default
• update the docs so that one can either install with conda or pip install yt[examples] if the examples need to be run

[neutrinoceros]

It seems I already worked on this problem in the past. Here's what I found

• UX: better error messages for frontends with missing bindings #3409 (and its dupe, UX: add hints to YTUnidentifiedDataType #4274)
• UX: add information about missing optional dependencies as warnings and error messages in yt.load #4275

Though I agree there's still room for improvement ! I like your solution of an adding a new set of extras. In fact, your suggestion resonates a lot with an on-going discussion that's happening over at astropy (see for instance conda-forge/astropy-feedstock#137). This even lead to a (draft) PEP: astrofrog/peps#1 (currently open for feedback !).

I think I would prefer naming the new extra "recommended" to mirror astropy's similar set of "optional but strongly recommended" dependencies. This would also provide a nice solution to #4690, where we couldn't agree on whether h5py should be a required dependency or not.
Regarding conda, I'd tend to agree with you that it's better to include more than just hard-required dependencies to compensate for the lack of extras, as supported by pip/PyPI.

Updating the docs is an obvious yes, though we should first see if the rest of your proposal makes sense to other devs.
ping @matthewturk, @chrishavlin, @cphyc, @brittonsmith, @jzuhone, @chummels, @Xarthisius