BoundaryML/baml · error

__class_getitem__

Error message

__class_getitem__

What it means

During pythonize_strict, BAML wraps values that have checks in a Python `Checked[...]` generic type by calling `__class_getitem__` on the Checked class from the baml runtime module. The `.expect("__class_getitem__")` panics when that call fails — typically because the Checked generic or the typing parameters could not be constructed in the embedded Python interpreter.

Solutions

1. Reinstall matching baml Python runtime (pip install -U baml) so the Checked generic supports parameterization
2. Verify the Python interpreter used by BAML imports baml.types correctly (python -c "from baml_py import ...")
3. Upgrade the BAML CLI and language client to the same version
4. Report to BAML maintainers if versions match — this expect() should be a PyErr

Example fix

null

import baml_py
assert hasattr(baml_py, 'Checked')
from baml_py import Checked
Checked[(int, __import__('typing').Literal['a'])]  # parameterization works

def checked_param_ok(cls):
    try:
        cls[(int, __import__('typing').Literal['x'])]
        return True
    except TypeError:
        return False

try:
    parsed = client.cast_to(result, MyType)
except Exception as e:
    logger.exception('BAML checked-type conversion failed: %s', e)

When it happens

Trigger: Calling parse_llm_response or cast_to on a result whose value carries check annotations (checked blocks/assertions), when the Python `Checked` class does not support `__class_getitem__` or the type-parameters tuple contains invalid items.

Common situations: Mismatch between the Rust FFI and the installed baml-py Python package (stale or mixed versions after an upgrade), or a broken embedded Python environment where `typing.Literal` parameterization fails.

Understand the failure class

Background: "This is a bug, please report it": internal invariant violations, unreachable panics, and SNH errors explained — this error's family across 47 libraries.

Related errors

• __class_getitem__ for streaming
• getattr(StreamState)
• Internal error
• Internal error
• nil success in InvocationResponse

AI-assisted analysis of BoundaryML/baml@bd85ce9dee (2026-09-12). Data as JSON: /api/errors/2aefbc01abf84a29. Report an issue: GitHub.

Appendix: source

Thrown at engine/language_client_python/src/types/function_results.rs:335

        // Collect check names as &str and turn them into a Python tuple
        let check_names: Vec<&str> = checks.iter().map(|check| check.name.as_str()).collect();
        let literal_args = PyTuple::new(py, check_names)?;

        // Call Literal[...] dynamically
        let literal_check_names = literal.get_item(literal_args).expect("get_item");

        let class_checked_type_constructor =
            cls_module.getattr("Checked").expect("getattr(Checked)");

        // Prepare type parameters for Checked[...]
        let type_parameters_tuple =
            PyTuple::new(py, [value_type.as_ref(), &literal_check_names]).expect("PyTuple::new");

        // Create the Checked type using __class_getitem__
        let class_checked_type: Bound<'_, PyAny> = class_checked_type_constructor
            .call_method1("__class_getitem__", (type_parameters_tuple,))
            .expect("__class_getitem__");

        // Prepare the properties dictionary
        let properties_dict = pyo3::types::PyDict::new(py);
        properties_dict.set_item("value", py_value_without_constraints)?;
        if !checks.is_empty() {
            properties_dict.set_item("checks", python_checks)?;
        }

        // Validate the model with the constructed type
        let checked_instance = class_checked_type
            .call_method(model_validate_method, (properties_dict.clone(),), None)
            .expect(model_validate_method);

        Ok::<Py<PyAny>, PyErr>(checked_instance.into())
    } else {
        Ok(py_value_without_constraints)
    }?;

View on GitHub (pinned to bd85ce9dee)