postprocessing_utils.to_glb function does not finish its execution. [stdout.read()]

In example_text.py script, the postprocessing_utils.to_glb function doesn't terminate its execution. The code runs until

# mesh postprocess
    vertices, faces = postprocess_mesh(
        vertices, faces,
        simplify=simplify > 0,
        simplify_ratio=simplify,
        fill_holes=fill_holes,
        fill_holes_max_hole_size=fill_holes_max_size,
        fill_holes_max_hole_nbe=int(250 * np.sqrt(1-simplify)),
        fill_holes_resolution=1024,
        fill_holes_num_views=1000,
        debug=debug,
        verbose=verbose,
    )

Going into the definition of postprocess_mesh function in postprocessing_utils.py, the function is executed until

vertices, faces = _fill_holes(
            vertices, faces,
            max_hole_size=fill_holes_max_hole_size,
            max_hole_nbe=32, # max_hole_nbe=fill_holes_max_hole_nbe,
            resolution=32, #resolution=fill_holes_resolution,
            num_views=10,#num_views=fill_holes_num_views,
            debug=debug,
            verbose=verbose,
        )

Going into the definition of _fill_holes function in postprocessing_utils.py, the function is executed until
rastctx = utils3d.torch.RastContext(backend='cuda')

Going into the definition of RastContext constructor in utils3d.torch.rasterization.py, the function is executed until
self.nvd_ctx = dr.RasterizeCudaContext(device=device)

Going into the definition of RasterizeCudaContext constructor in nvidiffrast.torch.ops.py, the function is executed until
self.cpp_wrapper = _get_plugin().RasterizeCRStateWrapper(cuda_device_idx).
_get_plugin() function doesn't terminate.

Going into the definition of _get_plugin() function in nvidiffrast.torch.ops.py, the function is executed until
torch.utils.cpp_extension.load(name=plugin_name, sources=source_paths, extra_cflags=common_opts+cc_opts, extra_cuda_cflags=common_opts+['-lineinfo'], extra_ldflags=ldflags, with_cuda=True, verbose=False)

Going into the definition of load function in torch.utils.cpp_extension.py, the code executes to the end. The return statement calls the function _jit_compile.

Going into the definition of _jit_compile in torch.utils.cpp_extension.py, the function is executed until

_write_ninja_file_and_build_library(
                        name=name,
                        sources=sources,
                        extra_cflags=extra_cflags or [],
                        extra_cuda_cflags=extra_cuda_cflags or [],
                        extra_ldflags=extra_ldflags or [],
                        extra_include_paths=extra_include_paths or [],
                        build_directory=build_directory,
                        verbose=verbose,
                        with_cuda=with_cuda,
                        is_standalone=is_standalone)

Going into the definition of _write_ninja_file_and_build_library function in torch.utils.cpp_extension.py, the function runs 3 subprocesses (completed successfully), and then execution continues until

subprocess.run(
            command,
            stdout=stdout_fileno if verbose else subprocess.PIPE,
            stderr=subprocess.STDOUT,
            cwd=build_directory,
            check=True,
            env=env)

Going into the definition of run function in python3.10.subprocess.py, the function is executed until
tdout, stderr = process.communicate(input, timeout=timeout).

Going into the definition of communicate function in the same .py script, the function is executed until
stdout = self.stdout.read().
Unlike the other 3 subprocesses, this subprocess blocks on the read statement and I don't know what is causing it and what to do.

Do you have any idea what this subprocess does and how to fix it?

Start with example_text.py and trace postprocessing_utils.py through utils3d.torch.rasterization.py and nvidiffrast.torch.ops.py, then inspect the torch.utils.cpp_extension.load build invoked by _get_plugin(). Reproduce the to_glb call and determine why the compilation subprocess remains blocked; done means the plugin build completes and to_glb returns.