Python: Writing EXR header fails with float32-ndarray attributes of length >=3

[vjankijevic]

Trying to copy header data over from one EXR file to another, I encountered the following error:

RuntimeError: unrecognized type of attribute 'com.arri.camera.LookCdlOffset': type=ndarray value=[0. 0. 0.] dtype=float32

Below is a minimal example of code producing the issue, with example files.zip attached.

import OpenEXR as exr

# Copy header from file_two to file_one and save new file as file_three.
with exr.File("/tmp/file_one.exr") as file_one:
    output_header = file_one.header()

    with exr.File("/tmp/file_two.exr") as file_two:

        for key, value in file_two.header().items():
            if key not in (
                "displayWindow",
                "dataWindow",
                "pixelAspectRatio",
                "screenWindowCenter",
                "screenWindowWidth",
                "lineOrder",
                "compression",
                "channels",
                "tiles",
                "view",
                "name",
                "type",
                "version",
                "chunkCount",
                "maxSamplesPerPixel",
            ):
                output_header[key] = value

    with exr.File(output_header, file_one.channels()) as output_file:
        output_file.write("/tmp/file_three.exr")

The workaround is to convert the array to a list with value.tolist().

Tested with Python 3.10.14, OpenEXR v3.3.3 and numpy 2.0.0 on macos and linux/rocky9.4

Thank you,
Vladimir