bug: Ibis `ibis.memtable(df).to_polars()` doesn't round-trip Polars

[MarcoGorelli]

What happened?

In [1]: import polars as pl

In [2]: import ibis

In [3]: ibis.set_backend('polars')

In [4]: df = pl.DataFrame({'a': [1,1,2], 'b': ['red', 'red', 'green']}, schema_overrides={'b': pl.Enum(['red', 'green',
   ...: 'white'])})

In [5]: df
Out[5]:
shape: (3, 2)
┌─────┬───────┐
│ a   ┆ b     │
│ --- ┆ ---   │
│ i64 ┆ enum  │
╞═════╪═══════╡
│ 1   ┆ red   │
│ 1   ┆ red   │
│ 2   ┆ green │
└─────┴───────┘

In [6]: ibis.memtable(df).to_polars()
Out[6]:
shape: (3, 2)
┌─────┬───────┐
│ a   ┆ b     │
│ --- ┆ ---   │
│ i64 ┆ str   │
╞═════╪═══════╡
│ 1   ┆ red   │
│ 1   ┆ red   │
│ 2   ┆ green │
└─────┴───────┘

What version of ibis are you using?

10.5.0

What backend(s) are you using, if any?

No response

Relevant log output

Code of Conduct

• I agree to follow this project's Code of Conduct

[cpcloud]

In this case it seems to be because of enums, which we don't support as a user-facing type. Are there other types where you observe the non round-trip behavior?

[MarcoGorelli]

thanks for your response - the only other one i noticed is

In [13]: df = pl.DataFrame({'a': [1,2]}, schema={'a': pl.Int128})

In [14]: ibis.memtable(df).to_polars()
---------------------------------------------------------------------------
KeyError                                  Traceback (most recent call last)
Cell In[14], line 1
----> 1 ibis.memtable(df).to_polars()

File ~/polars-api-compat-dev/.venv/lib/python3.12/site-packages/ibis/expr/api.py:423, in memtable(data, columns, schema, name)
    419     import ibis
    421     schema = ibis.schema(schema)
--> 423 return _memtable(data, name=name, schema=schema, columns=columns)

File ~/polars-api-compat-dev/.venv/lib/python3.12/site-packages/ibis/common/dispatch.py:115, in lazy_singledispatch.<locals>.call(arg, *args, **kwargs)
    112 @functools.wraps(func)
    113 def call(arg, *args, **kwargs):
    114     impl = dispatcher.dispatch(type(arg))
--> 115     return impl(arg, *args, **kwargs)

File ~/polars-api-compat-dev/.venv/lib/python3.12/site-packages/ibis/expr/api.py:572, in _memtable_from_polars_dataframe(data, name, schema, columns)
    568     assert schema is None, "if `columns` is not `None` then `schema` must be `None`"
    569     schema = sch.Schema(dict(zip(columns, sch.infer(data).values())))
    570 return ops.InMemoryTable(
    571     name=name if name is not None else util.gen_name("polars_memtable"),
--> 572     schema=sch.infer(data) if schema is None else schema,
    573     data=PolarsDataFrameProxy(data),
    574 ).to_expr()

File ~/polars-api-compat-dev/.venv/lib/python3.12/site-packages/ibis/common/dispatch.py:115, in lazy_singledispatch.<locals>.call(arg, *args, **kwargs)
    112 @functools.wraps(func)
    113 def call(arg, *args, **kwargs):
    114     impl = dispatcher.dispatch(type(arg))
--> 115     return impl(arg, *args, **kwargs)

File ~/polars-api-compat-dev/.venv/lib/python3.12/site-packages/ibis/expr/schema.py:379, in infer_polars_dataframe(df)
    374 @infer.register("polars.DataFrame")
    375 @infer.register("polars.LazyFrame")
    376 def infer_polars_dataframe(df):
    377     from ibis.formats.polars import PolarsSchema
--> 379     return PolarsSchema.to_ibis(df.collect_schema())

File ~/polars-api-compat-dev/.venv/lib/python3.12/site-packages/ibis/formats/polars.py:128, in PolarsSchema.to_ibis(cls, schema)
    124 @classmethod
    125 def to_ibis(cls, schema: dict[str, pl.DataType]) -> Schema:
    126     """Convert a polars schema to a schema."""
    127     return Schema.from_tuples(
--> 128         [(name, PolarsType.to_ibis(typ)) for name, typ in schema.items()]
    129     )

File ~/polars-api-compat-dev/.venv/lib/python3.12/site-packages/ibis/formats/polars.py:77, in PolarsType.to_ibis(cls, typ, nullable)
     72     return dt.Struct.from_tuples(
     73         [(field.name, cls.to_ibis(field.dtype)) for field in typ.fields],
     74         nullable=nullable,
     75     )
     76 else:
---> 77     return _from_polars_types[base_type](nullable=nullable)

KeyError: Int128