fix(rust, python): allow nonstrict cast of categorical/enum to enum

[mcrumiller]

Resolves #14900.

@c-peters the downside of this PR is that it no longer prints your nice error messages when a value falls outside of the target Enum's categories, and instead assigns it a null value. Strict (normal) casts will still properly fail the casts because of the new null values introduced, but non-strict casts now convert values outside to null. On the plus side, this reduces the code a bit and removes two levels of option/result indirection. Here is the example from the issue after this PR:

>>> import polars as pl
>>> pl.Series(["a", "b"]).cast(pl.Enum(["a"]), strict=False)
shape: (2,)
Series: '' [enum]
[
        "a"
        null
]
>>> pl.Series(["a", "b"], dtype=pl.Categorical).cast(pl.Enum(["a"]), strict=False)
shape: (2,)
Series: '' [enum]
[
        "a"
        null
]
>>> pl.Series(["a", "b"], dtype=pl.Enum(["a", "b"])).cast(pl.Enum(["a"]), strict=False)
shape: (2,)
Series: '' [enum]
[
        "a"
        null
]

strict (regular) casting still fails:

>>> import polars as pl
>>> pl.Series(["a", "b"], dtype=pl.Categorical).cast(pl.Enum(["a"]))
polars.exceptions.ComputeError: conversion from `cat` to `enum` failed in column '' for 1 out of 2 values: ["b"]

[deanm0000]

I'm not sure how strictness works. It seems like typically during a cast you just return None when it fails to cast for any particular value but then I'm guessing (b/c I don't see it here) that elsewhere it compares the input and output for nulls and if there are nulls in the output that weren't null in the input then that's when the strictness flag comes into play. In this case the error was baked into the casting so that's why it didn't respect strict?

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 81.01%. Comparing base (d9dead6) to head (239e184).
Report is 3 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main   #14910      +/-   ##
==========================================
- Coverage   81.01%   81.01%   -0.01%     
==========================================
  Files        1332     1332              
  Lines      172870   172843      -27     
  Branches     2458     2458              
==========================================
- Hits       140054   140031      -23     
+ Misses      32348    32344       -4     
  Partials      468      468              

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

[mcrumiller]

In this case the error was baked into the casting so that's why it didn't respect strict?

Yes. strict=False means if the value cannot be casted, a null value is returned:

>>> s = pl.Series(["a"])
>>> s.cast(pl.Date, strict=False)
shape: (1,)
Series: '' [date]
[
        null
]
>>> s.cast(pl.Int32, strict=False)
shape: (1,)
Series: '' [i32]
[
        null
]

There are some situations where strict=False doesn't return null:

>>> pl.Series([[1]]).cast(pl.Utf8, strict=False)
polars.exceptions.ComputeError: cannot cast List type (inner: 'Int64', to: 'String')

However, I feel it's reasonable to expecte that a non-strict cast from one enum to another should be treated the same way as a cast of string to enum--values outside of the categories are converted to null.

[deanm0000]

Sorry I know what strict is supposed to do. I meant I don't know about the implementation of it in rust.

[mcrumiller]

Ahh, sorry. Yes, here is the Series.strict_cast, which calls first calls regular cast (which is fallible and returns a Result, but usually can succeed). Then it checks if strict is True and, if there are more nulls after the cast, it means something failed, and it errors.

[mcrumiller]

Failures are related chrono update and should be fixed by #14928.

[ritchie46]

Yeap, good one. Thank you @mcrumiller .