pola-rs/polars · error

activate 'object' feature

Error message

activate 'object' feature

What it means

new_object constructs a polars Series from arbitrary Python objects, but that capability only exists when the polars binary was compiled with the 'object' feature. Without it, the function panics with a directive to activate the feature. Official PyPI wheels include this feature; custom/minimal builds may not.

Solutions

1. Install the official polars wheel from PyPI, which has the object feature enabled.
2. If building from source, enable the feature: cargo build/pip install with --features object (or the polars['all'] extras).
3. Avoid pl.Object: store data in supported dtypes, or keep Python objects outside the DataFrame.
4. Consider storing complex objects as serialized JSON/Binary columns instead.

Example fix

# before (custom build without object feature)
s = pl.Series('objs', [MyClass(), MyClass()], dtype=pl.Object)
# after
pip install --upgrade --force-reinstall polars
s = pl.Series('objs', [obj.to_dict() for obj in objs], dtype=pl.Struct)  # or reinstall official wheel

import polars
assert 'object' in str(polars.build_info()) or hasattr(polars, 'Object'), \
    'polars build lacks object feature'

def object_dtype_supported():
    import polars
    return 'object' in str(getattr(polars, 'build_info', lambda: '')()),

try:
    s = pl.Series('a', values, dtype=pl.Object)
except BaseException as e:
    if 'object' in str(e) and 'feature' in str(e):
        s = pl.Series('a', [json.dumps(v, default=str) for v in values])  # serialize instead
    else:
        raise

When it happens

Trigger: Creating a Series with dtype=pl.Object (e.g. pl.Series('a', [some_python_object], dtype=pl.Object)) on a polars build compiled without the object feature; also hit by certain .map_elements / apply paths that fall back to object construction.

Common situations: Building polars from source with a reduced feature set; using slim/alternative builds or embedded environments (e.g. some WASM or minimal distributions) without the object feature.

Related errors

• activate object
• activate 'binary_encoding' feature
• activate 'propagate_nans'
• should be hashable
• a non-HTTPS workspace_url was given

AI-assisted analysis of pola-rs/polars@fe841f959e (2026-09-18). Data as JSON: /api/errors/8adf97c8a1dff3e4. Report an issue: GitHub.

Appendix: source

Thrown at crates/polars-python/src/series/construction.rs:374

    #[staticmethod]
    #[pyo3(signature = (name, values, strict, dtype))]
    fn new_array(
        name: &str,
        values: &Bound<PyAny>,
        strict: bool,
        dtype: Wrap<DataType>,
    ) -> PyResult<Self> {
        Self::new_from_any_values_and_dtype(name, values, dtype, strict)
    }

    #[staticmethod]
    pub fn new_object(py: Python<'_>, name: &str, values: Vec<ObjectValue>, _strict: bool) -> Self {
        #[cfg(feature = "object")]
        {
            PySeries::from(series_from_objects(py, name.into(), values))
        }
        #[cfg(not(feature = "object"))]
        panic!("activate 'object' feature")
    }

    #[staticmethod]
    fn new_null(name: &str, values: &Bound<PyAny>, _strict: bool) -> PyResult<Self> {
        let len = values.len()?;
        Ok(Series::new_null(name.into(), len).into())
    }

    #[staticmethod]
    fn from_arrow(name: &str, array: &Bound<PyAny>) -> PyResult<Self> {
        let arr = array_to_rust(array)?;

        // Compute first. The physical conversion retains offsets so the flag remains valid.
        let fast_explode = arr
            .as_any()
            .downcast_ref::<LargeListArray>()
            .is_some_and(|a| a.offsets().as_slice().windows(2).all(|w| w[0] != w[1]));

View on GitHub (pinned to fe841f959e)