Build python wheel fails

[vlasov01]

Describe the issue

Building the wheel from the source (1.20.0) fails on Ubuntu 22 for armv7l.
It is similar to #21145

Linux pynq 5.15.19-xilinx-v2022.1 #1 SMP PREEMPT Mon Apr 11 17:52:14 UTC 2022 armv7l armv7l armv7l GNU/Linux
GNU ld (GNU Binutils for Ubuntu) 2.38
cmake version 3.31.20241016-gee2eadb
numpy 1.26.4

Urgency
No response

Target platform
armv7l

Build script

cd onnxruntime
./build.sh --config Release --parallel --skip_submodule_sync --skip_tests --use_xnnpack --build_wheel --cmake_extra_defines CMAKE_CXX_FLAGS="-Wno-dev -Wall -Wextra -O3 -g3 -mfloat-abi=hard -mfpu=neon-vfpv3 -mcpu=cortex-a9 -mfloat-abi=hard" CMAKE_C_FLAGS="-Wno-dev -Wall -Wextra -O3 -g3 -mfloat-abi=hard -mfpu=neon-vfpv3 -mcpu=cortex-a9 -mfloat-abi=hard" onnxruntime_BUILD_UNIT_TESTS=OFF

Error / output

/home/xilinx/onnxruntime/onnxruntime/python/onnxruntime_pybind_state.cc: In function ‘pybind11::array onnxruntime::python::PrimitiveTensorToNumpyFromDevice(const OrtValu                                                                                                                                                                            e&, const DataTransferAlternative&)’:
/home/xilinx/onnxruntime/onnxruntime/python/onnxruntime_pybind_state.cc:232:46: error: no matching function for call to ‘pybind11::dtype::dtype(const int&)’
  232 |   pybind11::array result(py::dtype(numpy_type), tensor.Shape().GetDims());
      |                                              ^
In file included from /usr/local/include/pybind11/cast.h:13,
                 from /usr/local/include/pybind11/attr.h:13,
                 from /usr/local/include/pybind11/pybind11.h:45,
                 from /home/xilinx/onnxruntime/onnxruntime/python/onnxruntime_pybind.h:18,
                 from /home/xilinx/onnxruntime/onnxruntime/python/onnxruntime_pybind_exceptions.h:6,
                 from /home/xilinx/onnxruntime/onnxruntime/python/onnxruntime_pybind_state.cc:5:
/usr/local/include/pybind11/numpy.h:464:5: note: candidate: ‘template<class Policy_> pybind11::dtype::dtype(const pybind11::detail::accessor<Policy_>&)’
  464 |     PYBIND11_OBJECT_DEFAULT(dtype, object, detail::npy_api::get().PyArrayDescr_Check_);
      |     ^~~~~~~~~~~~~~~~~~~~~~~
/usr/local/include/pybind11/numpy.h:464:5: note:   template argument deduction/substitution failed:
/home/xilinx/onnxruntime/onnxruntime/python/onnxruntime_pybind_state.cc:232:46: note:   mismatched types ‘const pybind11::detail::accessor<Policy_>’ and ‘const int’
  232 |   pybind11::array result(py::dtype(numpy_type), tensor.Shape().GetDims());
      |                                              ^
In file included from /home/xilinx/onnxruntime/onnxruntime/python/onnxruntime_pybind_mlvalue.h:9,
                 from /home/xilinx/onnxruntime/onnxruntime/python/onnxruntime_pybind_state.cc:6:
/usr/local/include/pybind11/numpy.h:478:5: note: candidate: ‘pybind11::dtype::dtype(pybind11::list, pybind11::list, pybind11::list, pybind11::ssize_t)’
  478 |     dtype(list names, list formats, list offsets, ssize_t itemsize) {
      |     ^~~~~
/usr/local/include/pybind11/numpy.h:478:5: note:   candidate expects 4 arguments, 1 provided
/usr/local/include/pybind11/numpy.h:476:5: note: candidate: ‘pybind11::dtype::dtype(const char*)’ (near match)
  476 |     dtype(const char *format) : dtype(std::string(format)) { }
      |     ^~~~~
/usr/local/include/pybind11/numpy.h:476:5: note:   conversion of argument 1 would be ill-formed:
/home/xilinx/onnxruntime/onnxruntime/python/onnxruntime_pybind_state.cc:232:36: error: invalid conversion from ‘int’ to ‘const char*’ [-fpermissive]
  232 |   pybind11::array result(py::dtype(numpy_type), tensor.Shape().GetDims());
      |                                    ^~~~~~~~~~
      |                                    |
      |                                    int

[snnn]

Your pybind version is too low. Delete /usr/local/include/pybind11 , or upgrade it, or do the build in a clean docker environment.