Memory leaks with Python 3.14?

[yamatveev]

In our project (https://gitlab.com/tango-controls/pytango/-/issues/744) we presumably see memory leaks with Python 3.14. It is reported by valgrind and I also see RES memory usage grows. Interestingly, with all older Pythons I do not see it.

E.g. one of the leaks is for the line:

    // Create a capsule to manage the lifetime of value_ptr
    py::capsule base(value_ptr, dev_var_attribute_array_deleter<tangoTypeConst>);

    // Create the numpy array without copying the data, and associate the base object
    read_value = py::array(py::dtype::of<TangoScalarType>(),
                           dims,
                           strides,
                           buffer,
                           base // Associate the capsule as the base object
    );

This is valgrind report:

  Leak 1:  1,616 (864 direct, 752 indirect) bytes in 9 blocks are definitely lost in loss record 22,818 of 23,253
    at  0x483976B: malloc (.../valgrind/coregrind/m_replacemalloc/vg_replace_malloc.c:442 in .../local/libexec/valgrind/vgpreload_memcheck-amd64-linux.so)
    by   0x2BD062: UnknownInlinedFun (.../conda/python-3.14.3/Objects/obmalloc.c:1478 in .../envs/tangodev_14/bin/python3.14)
    by   0x2BD062: UnknownInlinedFun (.../python-3.14.3/Include/internal/pycore_object_alloc.h:46 in .../envs/tangodev_14/bin/python3.14)
    by   0x2BD062: UnknownInlinedFun (.../conda/python-3.14.3/Objects/typeobject.c:2365 in .../envs/tangodev_14/bin/python3.14)
    by   0x2BD062: PyType_GenericAlloc (.../conda/python-3.14.3/Objects/typeobject.c:2396 in .../envs/tangodev_14/bin/python3.14)
    by  0xE2C5897: PyArray_NewFromDescr_int (...:0 in .../site-packages/numpy/_core/_multiarray_umath.cpython-314-x86_64-linux-gnu.so)
    by  0xE2C640F: PyArray_NewFromDescr (...:0 in .../site-packages/numpy/_core/_multiarray_umath.cpython-314-x86_64-linux-gnu.so)
    by  0x7F02A7F: pybind11::array::array(pybind11::dtype c...er<long>, void const*, pybind11::handle) (.../pybind11/include/pybind11/numpy.h:1070 in .../python3.14/site-packages/tango/_tango.cpython-314-x86_64-linux-gnu.so)
    by  0x7EF56A6: void PyDeviceAttribute::array_value_from...l, pybind11::object&, pybind11::object&) (.../ext/convertors/attributes/array_cpp_to_python.h:216 in .../python3.14/site-packages/tango/_tango.cpython-314-x86_64-linux-gnu.so)

Since this code works with Python <=3.13 and only leaks with 3.14 - is there anything known about memory management/leaks in 3.14?

Update:

When I added printouts to the dev_var_attribute_array_deleter I see, that it is not called in Pyhton 3.14. So array really not deleted

[yamatveev]

See #5998