fail to load H5 objects with Pandas 3 and pyarrow NOT installed

[gdementen]

Also on Windows/local tests:

• larray/tests/test_session.py::test_init_session_hdf
• larray/tests/test_checked_session.py::test_init_checkedsession_hdf

Specific to CI:

• larray/tests/test_checked_session.py::test_h5_io_cs - TypeError:
  ['a0', 'a1']
  Length: 2, dtype: str has an invalid type (StringArray) for a key
• larray/tests/test_group.py::test_h5_io_lgroup - TypeError:
  ['P01', 'P03', 'P05']
  Length: 3, dtype: str has an invalid type (StringArray) for a key
• larray/tests/test_session.py::test_h5_io: AssertionError: caught 2 matching warning(s) but expected 4 instead

They seem related to string columns from Pandas 3 (#1166), but I don't know why they do not trigger locally?!

[gdementen]

On CI, I have

numpy                      2.4.3            py313h08c6c3d_0
pandas                     3.0.1            py313h86c3e14_0
pytables                   3.11.1           py313h4db2897_0

while locally (dev313) I have:

numpy                      2.3.5            py313h050da96_0
pytables                   3.10.2           py313h0217527_2
pandas                     3.0.0            py313hf6ff4c0_0

[gdementen]

Well. I updated --all local and got the same versions than CI but I still only have 2 failures. This is going to be hard to debug...

[gdementen]

Looking at the detailed tracebacks, at least it seems like it is the same issue. I still don't know why it does not trigger locally.

larray/tests/test_group.py:227: in test_h5_io_lgroup
    named2 = read_hdf(fpath, key=named_axis_not_in_file.name)
             ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
larray/inout/hdf.py:128: in read_hdf
    res = LGroup(key=key, name=name, axis=axis)
          ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
larray/core/group.py:1582: in __init__
    key = _to_key(key)
          ^^^^^^^^^^^^
larray/core/group.py:597: in _to_key
    raise TypeError(f"{v} has an invalid type ({type(v).__name__}) for a key")
E   TypeError: <StringArray>
E   ['P01', 'P03', 'P05']
E   Length: 3, dtype: str has an invalid type (StringArray) for a key

larray/tests/test_checked_session.py:558: in test_h5_io_cs
    _test_io_cs(tmp_path, meta, engine='pandas_hdf', ext='h5')
larray/tests/test_checked_session.py:361: in _test_io_cs
    cs.load(fpath, engine=engine)
larray/core/session.py:443: in load
    metadata, objects = handler.read(names, display=display, **kwargs)
                        ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
larray/inout/common.py:149: in read
    res[key] = self._read_item(key, type_, *args, **kwargs)
               ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
larray/inout/hdf.py:169: in _read_item
    return read_hdf(self.handle, hdf_key, *args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
larray/inout/hdf.py:128: in read_hdf
    res = LGroup(key=key, name=name, axis=axis)
          ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
larray/core/group.py:1582: in __init__
    key = _to_key(key)
          ^^^^^^^^^^^^
larray/core/group.py:597: in _to_key
    raise TypeError(f"{v} has an invalid type ({type(v).__name__}) for a key")
E   TypeError: <StringArray>
E   ['a0', 'a1']
E   Length: 2, dtype: str has an invalid type (StringArray) for a key

larray/tests/test_session.py:287: in test_h5_io
    _test_io(tmp_path, session, meta, engine='pandas_hdf', ext='h5')
larray/tests/test_session.py:223: in _test_io
    s.load(fpath, engine=engine)
larray/core/session.py:443: in load
    metadata, objects = handler.read(names, display=display, **kwargs)
                        ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
larray/inout/common.py:149: in read
    res[key] = self._read_item(key, type_, *args, **kwargs)
               ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
larray/inout/hdf.py:169: in _read_item
    return read_hdf(self.handle, hdf_key, *args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
larray/inout/hdf.py:128: in read_hdf
    res = LGroup(key=key, name=name, axis=axis)
          ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
larray/core/group.py:1582: in __init__
    key = _to_key(key)
          ^^^^^^^^^^^^
larray/core/group.py:597: in _to_key
    raise TypeError(f"{v} has an invalid type ({type(v).__name__}) for a key")
E   TypeError: <StringArray>
E   ['a0', 'a1']
E   Length: 2, dtype: str has an invalid type (StringArray) for a key

[gdementen]

Ok, I can reproduce the failures now. It turns out I have pyarrow installed in all the local dev environments I tested but not in CI which changes the type returned by Pandas. Should be an easy fix now.