docs(python): Add security warning in LazyFrame.deserialize() docstring

[douglas-raillard-arm]

Add indication that LazyFrame.deserialize() might execute arbitrary code coming from the deserialized data.

Fixes #14623

[stinodego]

Thanks, the text looks good. One note about formatting.

I think this also applies to Expr.deserialize, right?

[stinodego]

Could you add this as a Warnings section rather than an admonition? The section should be below Parameters / above See Also. See https://numpydoc.readthedocs.io/en/latest/format.html

[ritchie46]

We should elaborate here. If we put a big fat security warning we should explain under which conditions. Polars only uses pickle if the logical plan was serialized with python UDFs. In most queries Polars won't use pickle.

[douglas-raillard-arm]

@ritchie46 is that relevant for deserialize() though ? From the security perspective, what matters is whether you trust the data. If so, it does not matter whether it's using pickle. However, if you don't trust the data, it's very unlikely you would trust a promise that this JSON does not contain the UDF. As such, if the warning applies to your case, you'd have to assume it contains a malicious bit of pickle. So unless there is a way for the user to prevent loading pickle at the loss of some feature (e.g. a deserialize(..., forbid_pickle=True) parameter), the conditions of production don't really matter.

[ritchie46]

It gives you more information on when we call pickle, I think that's important. You can evenly verify that the JSON doesn't have any UDFs as it is human readable.

[ritchie46]

And for what it's worth, I can see us adding a forbid pickle.

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 81.32%. Comparing base (8f616b8) to head (88b1a58).
Report is 75 commits behind head on main.

❗ Current head 88b1a58 differs from pull request most recent head 299f73c. Consider uploading reports for the commit 299f73c to get more accurate results

Additional details and impacted files

@@            Coverage Diff             @@
##             main   #15282      +/-   ##
==========================================
- Coverage   81.32%   81.32%   -0.01%     
==========================================
  Files        1359     1359              
  Lines      176072   176072              
  Branches     2526     2526              
==========================================
- Hits       143195   143185      -10     
- Misses      32393    32403      +10     
  Partials      484      484              

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

[douglas-raillard-arm]

Updated the PR with:

• Use of warnings section instead of Sphinx warning directive
• Added information on the circumstances leading to having pickle blobs in the serialize() JSON

[ritchie46]

Thanks!