Segfault when sorting with nulls_last on all-null no-category lexical categorical sicne 0.20.22

[wenjuno]

Checks

• I have checked that this issue has not already been reported.
• I have confirmed this bug exists on the latest version of Polars.

Reproducible example

import polars as pl
df1 = pl.DataFrame({'a': [1]})
df2 = pl.DataFrame({'a': [-1], 'b': pl.Series(['A'], dtype=pl.Categorical('lexical'))})
df3 = df1.join(df2, on='a', how='left')
print(f"{df3=}")
print(f"{df3.dtypes=}")
print(f"{df3['b'].cat.get_categories()=}")
print(f"{df3.sort(['a', 'b'], nulls_last=True)=}")
df4 = df3.cast({'b': pl.Categorical('lexical')})
print(f"{df4=}")
print(f"{df4.dtypes=}")
print(f"{df4['b'].cat.get_categories()=}")
print(f"{df4.sort(['a', 'b'], nulls_last=True)=}")

Log output

df3=shape: (1, 2)
┌─────┬──────┐
│ a   ┆ b    │
│ --- ┆ ---  │
│ i64 ┆ cat  │
╞═════╪══════╡
│ 1   ┆ null │
└─────┴──────┘
df3.dtypes=[Int64, Categorical(ordering='lexical')]
df3['b'].cat.get_categories()=shape: (1,)
Series: 'b' [str]
[
        "A"
]
df3.sort(['a', 'b'], nulls_last=True)=shape: (1, 2)
┌─────┬──────┐
│ a   ┆ b    │
│ --- ┆ ---  │
│ i64 ┆ cat  │
╞═════╪══════╡
│ 1   ┆ null │
└─────┴──────┘
df4=shape: (1, 2)
┌─────┬──────┐
│ a   ┆ b    │
│ --- ┆ ---  │
│ i64 ┆ cat  │
╞═════╪══════╡
│ 1   ┆ null │
└─────┴──────┘
df4.dtypes=[Int64, Categorical(ordering='lexical')]
df4['b'].cat.get_categories()=shape: (0,)
Series: 'b' [str]
[
]
zsh: segmentation fault (core dumped)

Issue description

To be clear, the bug seems to be triggered under these conditions since 0.20.22:

1. Sort by multiple columns with nulls_last=True AND
2. One or more columns are lexical ordering Categorical with no categories and all-null values.

It works fine with 0.20.21.

Note that in the example the df2["b"] doesn't have to be a lexical Categorical, it could very well be string type and sort on df4 will still segfault. I make df2["b"] a Categorical to show that having an empty category list seems to be essential to trigger the problem. The cast seems to recompute the categories even when the type is not actually changing.

Expected behavior

Sort the dataset without crashing.

Installed versions

--------Version info---------
Polars:               0.20.23
Index type:           UInt32
Platform:             Linux-6.8.7-arch1-1-x86_64-with-glibc2.39
Python:               3.11.9 | packaged by conda-forge | (main, Apr 19 2024, 18:36:13) [GCC 12.3.0]

----Optional dependencies----
adbc_driver_manager:  <not installed>
cloudpickle:          <not installed>
connectorx:           <not installed>
deltalake:            <not installed>
fastexcel:            <not installed>
fsspec:               <not installed>
gevent:               <not installed>
hvplot:               <not installed>
matplotlib:           <not installed>
nest_asyncio:         <not installed>
numpy:                <not installed>
openpyxl:             <not installed>
pandas:               <not installed>
pyarrow:              <not installed>
pydantic:             <not installed>
pyiceberg:            <not installed>
pyxlsb:               <not installed>
sqlalchemy:           <not installed>
xlsx2csv:             <not installed>
xlsxwriter:           <not installed

[orlp]

Can reproduce.

[ritchie46]

Ah, contention. Nevermind