fix(python): raise if DF is input to context (with_columns, select, etc)

[deanm0000]

addresses #15588

[MarcoGorelli]

nice - fancy adding a little test too?

[deanm0000]

A test to verify that it errors?

[MarcoGorelli]

Yup, you can use the pytest.raises context manager

[deanm0000]

well that was a good callout b/c there was a way that is snuck by that is now fixed.

[codecov]

Codecov Report

Attention: Patch coverage is 62.50000% with 3 lines in your changes are missing coverage. Please review.

Project coverage is 81.20%. Comparing base (31df06d) to head (c42184c).
Report is 61 commits behind head on main.

Files	Patch %	Lines
py-polars/polars/functions/lit.py	62.50%	2 Missing and 1 partial ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main   #15598      +/-   ##
==========================================
+ Coverage   81.14%   81.20%   +0.06%     
==========================================
  Files        1362     1370       +8     
  Lines      174951   175696     +745     
  Branches     2533     2533              
==========================================
+ Hits       141960   142677     +717     
- Misses      32508    32543      +35     
+ Partials      483      476       -7     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[stinodego]

Maybe this should be parsed as a struct column instead of raising?

[deanm0000]

Maybe this should be parsed as a struct column instead of raising?

If they want a struct they can just df.to_struct(). I have three reasons for thinking it's not a great idea to do this for people behind the scenes:

1. If they're putting one df as the input to another df's with_columns/select then they're doing an anti-pattern and they should go ask for help so the error is the best thing for them
2. People are likely to be more befuddled by getting a struct where they once got individual columns so it'll feel more like changing behavior without notice than fixing the bug that let them accidentally do the thing.
3. It's a very pandas thing to do to try to anticipate and fix the way people do things incorrectly.

I was scratching my head on suggestions to include in the error message. I was thinking of "hey maybe you need a join or an hstack or to just refactor generally" but there are too many possibilities of what they're doing wrong that I just went with the short one that you see.

[stinodego]

Thanks for the PR, but after thinking on this for a bit, this should not raise at all. The existing behavior is correct. Please see the linked issue.