fix: Return correct dtype for s.clear() when dtype is Object

[mcrumiller]

Resolves #15303.

I think this is a good way to do it? @stinodego will need your input here.

[mcrumiller]

Also, any reason we cannot simply do a length-0 slice and clone? That works too. Update: using slice method instead to prevent other errors.

[codecov]

Codecov Report

Attention: Patch coverage is 80.00000% with 1 lines in your changes are missing coverage. Please review.

Project coverage is 81.36%. Comparing base (3785fad) to head (c860e97).
Report is 11 commits behind head on main.

❗ Current head c860e97 differs from pull request most recent head 51345cb. Consider uploading reports for the commit 51345cb to get more accurate results

Files	Patch %	Lines
crates/polars-core/src/series/mod.rs	80.00%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main   #15315      +/-   ##
==========================================
- Coverage   81.41%   81.36%   -0.06%     
==========================================
  Files        1362     1364       +2     
  Lines      176734   176613     -121     
  Branches     2531     2526       -5     
==========================================
- Hits       143896   143696     -200     
- Misses      32354    32434      +80     
+ Partials      484      483       -1     

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

[stinodego]

Thanks, I think this is a good fix. But I don't know who wrote the original 'hack' - so I'm not sure their use case is now covered. @ritchie46 could you take a quick look?

[ritchie46]

This works. Though this is not worth doing unsafe. Can we use self.take.

I also wonder why we need to clone afterwards? 🤔

[mcrumiller]

Can do. Out of curiosity, is this path a bit simpler/faster than the standard dtype path of going through full_null with zero elements? We could just remove all the if logic and just return an empty slice for all dtypes.

[mcrumiller]

Ahh unfortunately that breaks pl.Array. I'll keep the special object logic in there.

[mcrumiller]

@ritchie46 used take_slice instead of unchecked now. Moved the empty clone check to all dtypes since it'll be a little faster and simplifies a bit.

Edit: this broke some memory alignment on Windows build for some reason.

[mcrumiller]

Ok, no more errors.

[ritchie46]

slice doesn't cut it. A slice doesn't clear memory.

[mcrumiller]

@ritchie46 this one passes now and uses take. I moved the clone up front for all dtypes since that's probably a bit faster than going through full_null when the series is already empty.