Bug: Linux/Python 3.13: DataFrame/NumPy path crashes with numpy_type.cpp UNREACHABLE_CODE

[ahgh0]

Ladybug version

v0.17.1

What operating system are you using?

Ubuntu 22.04.5 LTS

What happened?

Hi LadybugDB team,
we hit a native assertion in Ladybug when running a graph validation workload on Linux with Python 3.13.

Environment:

• OS: Ubuntu 22.04.5 LTS
• Python: 3.13 and 3.11
• Ladybug: observed with 0.16.1 and 0.17.1
• NumPy: same version as a Windows environment where the workload works
• Workload: in-memory graph, loading node/rel tables and evaluating Cypher queries
  (The same python code from me (and same versions of numpy, ladybug etc) works on a Windows machine, so this is a linux-related bug in Ladybug)

Failure:
RuntimeError: Assertion failed in file "/project/scripts/pip-package/cibw-source/sdist/ladybug-source/tools/python_api/src_cpp/numpy/numpy_type.cpp" on line 86: UNREACHABLE_CODE

Initial failing code path:

df = pandas.DataFrame(data, columns=cols)
conn.execute(f"COPY {table} FROM df")

Stack excerpt:
File ".../test.py", line 1438, in copy_df
    conn.execute(f"COPY {table} FROM df")
File ".../site-packages/ladybug/connection.py", line 328, in execute
    query_result_internal = self._execute_with_pybind(query, parameters)
File ".../site-packages/ladybug/connection.py", line 258, in _execute_with_pybind
    return py_connection.execute(prepared, parameters)
RuntimeError: Assertion failed in file ".../numpy_type.cpp" on line 86: UNREACHABLE_CODE

We also saw similar failures when using query-side DataFrame input such as:
LOAD FROM ab_df

Workarounds tested:
- Avoiding pandas/DataFrame usage completely works.
- Avoiding get_as_df() and using rows_as_dict().get_all() works.
- Literal UNWIND [...] input works but is much slower.
- A hybrid approach works best for now: CSV COPY for large graph load, no pandas/DataFrame for rule/query rows, and rows_as_dict() for results.


Expected behavior:   Ladybug should either:
- support the pandas/NumPy dtype combination on Linux/Python 3.13, or
- raise a Python-level error that identifies the unsupported dtype/column instead of hitting native UNREACHABLE_CODE.

Could you please check the NumPy dtype handling around numpy_type.cpp:86, especially for pandas DataFrame inputs with mixed string/int/float/null columns on Linux/Python 3.13?


### Are there known steps to reproduce?

1. Create a Linux Python 3.13 virtual environment.
2. Install dependencies:   pip install ladybug==0.16.1 pandas numpy
3. Start Python and run:

import ladybug as lb
import pandas as pd
conn = lb.Connection(lb.Database(":memory:"))
conn.execute("CREATE NODE TABLE T(id STRING PRIMARY KEY, name STRING, n INT64, v DOUBLE)")

df = pd.DataFrame([
    {"id": "a", "name": "x", "n": 1, "v": 1.5},
    {"id": "b", "name": "y", "n": 2, "v": None},
])
conn.execute("COPY T FROM df")

4. On the affected Linux host this crashes with:

RuntimeError: Assertion failed in file ".../numpy_type.cpp" on line 86: UNREACHABLE_CODE

(the same error also appears with query-side DataFrame input:
ab_df = pd.DataFrame([...])
conn.execute("LOAD FROM ab_df RETURN *")

Workaround: avoiding pandas/DataFrame input and avoiding get_as_df() prevents the crash.)