Improve "print_env.sh" experience for reporting issues

[bdice]

In #15589, @betatim had difficulty finding the print_env.sh script that is requested as a part of filing an issue.

Is there a way to provide the same information via something like import cudf; cudf.print_env()?

This is probably a viable alternative to the shell script we currently recommend. I think we could re-evaluate the contents of that script. The downside would be that users having installation troubles wouldn't be able to import the package to call such a function -- but the environment information might still be helpful for diagnosis.

Also, it seems to me that many users skip the step of reporting environment information, so either the print_env.sh script is hard to find or issue reporters don't think it is important.

xref: https://github.com/rapidsai/cudf/issues/15589#issuecomment-2083061696

[galipremsagar]

I think we should have similar api support like import pandas as pd; pd.show_versions()

[bdice]

I love that idea. We can extend that as cudf.show_versions() to include any extra information we need.

@galipremsagar What do you think cudf.pandas should do?

[vyasr]

I'm confused, isn't something like cudf.show_versions exactly what you mentioned in your original post as:

The downside would be that users having installation troubles wouldn't be able to import the package to call such a function -- but the environment information might still be helpful for diagnosis.

I'm still OK with that approach if we think that covers enough use cases, just making sure I'm not missing something here.

[bdice]

A show_versions function may be enough for most use cases. I am hopeful that other cases with installation issues can be resolved from conda solver output or the pip placeholder packages showing an error that includes environment information (which should be available in recent/future releases, from the Kitmaker team’s updates).

[betatim]

Wild guess, but presumably most people who have difficulty installing cudf are trying to install it via pip or conda. This means they don't have access to anything inside the package (script or function). So helping them will need something else.

I like the idea of having cudf.show_versions(). Easier to use as a user, easier to explain to users

[vyasr]

There is a much older issue https://github.com/rapidsai/cudf/issues/931 discussing this problem. I'm going to close that one in favor of this one since the discussion here is much more current, but some of the discussions there (particularly around ensuring that we also support non-Python users in some way) still apply.