[Question] Python 3.13t (free-threaded) compatibility

[szalpal]

Hello!

Is there any plan of supporting Python3.13t (i.e. free-threaded Python)?

[Avnsx]

Not sure if this is a problem with opencv or with numpy, but I've had a great waste of time for multiple hours trying to make both of these libraries even just import on 3.13.2 thread unlocked...

So I'll just drop this here numpy/numpy#28275

[nrodlin]

What is the current status on the compatibility? I tried to install opencv using pip and python 3.13.2t resulting in a failure, I checked #1051 with a potential solution for Linux systems to build the Python3.13t wheels, but it seems that the PR failed to complete several tests for other OS.

Does anyone know a way to try opencv with python 3.13t at this moment? There is current support for numpy, scipy, pytorch and other relevant modules.

[EdikIakovlev]

Hello. I also can't install the opencv package for python 3.13.3t. When will python 3.13.3t be supported?
I use Windows 10 64.

[FrsECM]

Hello !
Just tested on Ubuntu22.04.
I face an issue with python 3.14t as well.

I wanted to build the headless version and i thought it will work until 99%...

      [ 99%] Built target gen_opencv_python_source
      [ 99%] Building CXX object modules/python3/CMakeFiles/opencv_python3.dir/__/src2/cv2.cpp.o

      [stderr]
      CMake Error: CMake was unable to find a build program corresponding to "Ninja".  CMAKE_MAKE_PROGRAM is not set.  You probably need
      to select a different build tool.
      CMake Warning at cmake/OpenCVDownload.cmake:271 (message):
        IPPICV: Hash mismatch: 2c9a9c3f9ed1aa657b9f689b47f1e35b
      Call Stack (most recent call first):
        3rdparty/ippicv/ippicv.cmake:38 (ocv_download)
        cmake/OpenCVFindIPP.cmake:272 (download_ippicv)
        cmake/OpenCVFindLibsPerf.cmake:12 (include)
        CMakeLists.txt:800 (include)


      CMake Warning at cmake/OpenCVDownload.cmake:271 (message):
        ADE: Hash mismatch: a8b72d0b22ca0210ba8f02b9764811d9
      Call Stack (most recent call first):
        modules/gapi/cmake/DownloadADE.cmake:5 (ocv_download)
        modules/gapi/cmake/init.cmake:20 (include)
        cmake/OpenCVModule.cmake:298 (include)
        cmake/OpenCVModule.cmake:361 (_add_modules_1)
        cmake/OpenCVModule.cmake:408 (ocv_glob_modules)
        CMakeLists.txt:1068 (ocv_register_modules)


      In file included from
      /etc/uv/.cache/sdists-v9/index/1a0ab4555af1b3df/opencv-python-headless/4.11.0.86/JkwPAOq2WUSFD-kyMq0ts/src/opencv/modules/python/src2/cv2.hpp:23,
                       from
      /etc/uv/.cache/sdists-v9/index/1a0ab4555af1b3df/opencv-python-headless/4.11.0.86/JkwPAOq2WUSFD-kyMq0ts/src/opencv/modules/python/src2/cv2.cpp:5:
      /home/f296849/.local/share/uv/python/cpython-3.14.0rc3+freethreaded-linux-x86_64-gnu/include/python3.14t/Python.h:51:4: error:
      #error "The limited API is not currently supported in the free-threaded build"
         51 | #  error "The limited API is not currently supported in the free-threaded build"

But very interested by the topic in case it move forward.

[mhawash]

Looking forward to see opencv supported on the free-threaded version of Python, this is now a crucial requirement for community to move towards a more capable programming pattern on modern hardware and which is why the threaded version of Python was made!

[rgommers]

opencv/opencv#27933 (comment) says that free-threading support is being worked on at the moment, which is great to see.

For building from source, the one-line change in gh-1051 should unblock anyone who wants to test locally. Or even just install opencv-python as part of a larger application/environment under a free-threaded interpreter. Without use of Python-level threading (e.g., via the concurrent.futures or threading stdlib modules) things will work just fine.

[Aweptimum]

Without use of Python-level threading (e.g., via the concurrent.futures or threading stdlib modules) things will work just fine

To clarify @rgommers - does this mean "not using threading whatsoever when cv2 is imported" or "only using cv2 in the mainthread"?

[rgommers]

I meant using cv2 only in a single thread. Calling cv2 functionality in parallel from multiple threads will need testing and probably some thread-safety fixes.