[NOMRG] Try build and tests on ffmpeg6

[NicolasHug]

Trying this now that #8096 is merged

[pytorch-bot]

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/vision/8211

• 📄 Preview Python docs built from this PR

Note: Links to docs will display an error until the docs builds have been completed.

✅ You can merge normally! (6 Unrelated Failures)

As of commit b33e843 with merge base a00a72b ():

FLAKY - The following jobs failed but were likely due to flakiness present on trunk:

• CMake / windows (windows.4xlarge, cpu) / windows-job (gh)
  The process 'C:\Program Files\Git\cmd\git.exe' failed with exit code 128
• Tests / unittests-extended / linux-job (gh)
  The process '/usr/bin/git' failed with exit code 128
• Tests / unittests-windows (3.10, windows.4xlarge, cpu) / windows-job (gh)
  The process 'C:\Program Files\Git\cmd\git.exe' failed with exit code 128
• Tests / unittests-windows (3.11, windows.4xlarge, cpu) / windows-job (gh)
  The process 'C:\Program Files\Git\cmd\git.exe' failed with exit code 128
• Tests / unittests-windows (3.8, windows.4xlarge, cpu) / windows-job (gh)
  The process 'C:\Program Files\Git\cmd\git.exe' failed with exit code 128
• Tests / unittests-windows (3.9, windows.4xlarge, cpu) / windows-job (gh)
  The process 'C:\Program Files\Git\cmd\git.exe' failed with exit code 128

This comment was automatically generated by Dr. CI and updates every 15 minutes.

[NicolasHug]

Dataset tests are segfaulting, probably for the same (unknkown) reasons as in #7296. I assume it comes from the video datasets. I can't reproduce the segfault locally but with ffmpeg6 and pyav installed I'm getting errors like

__________________________________________ UCF101TestCase.test_num_examples ___________________________________________
test/datasets_utils.py:555: in test_num_examples
    with self.create_dataset(config) as (dataset, info):
../.miniconda3/envs/ffmpeg6/lib/python3.10/contextlib.py:135: in __enter__
    return next(self.gen)
test/datasets_utils.py:378: in create_dataset
    info = self._inject_fake_data(tmpdir, complete_config) if inject_fake_data else None
test/datasets_utils.py:482: in _inject_fake_data
    info = self.inject_fake_data(tmpdir, config)
test/test_datasets.py:871: in inject_fake_data
    video_files = self._create_videos(video_folder)
test/test_datasets.py:883: in _create_videos
    video_files = [
test/test_datasets.py:884: in <listcomp>
    datasets_utils.create_video_folder(root, cls, lambda idx: file_name_fn(cls, idx), num_examples_per_class)
test/datasets_utils.py:123: in inner_wrapper
    return fn(*args, **kwargs)
test/datasets_utils.py:940: in create_video_folder
    return [
test/datasets_utils.py:941: in <listcomp>
    create_video_file(root, file_name_fn(idx), size=size(idx) if callable(size) else size, **kwargs)
test/datasets_utils.py:123: in inner_wrapper
    return fn(*args, **kwargs)
test/datasets_utils.py:890: in create_video_file
    torchvision.io.write_video(str(file), video.permute(0, 2, 3, 1), fps, **kwargs)
torchvision/io/video.py:91: in write_video
    stream = container.add_stream(video_codec, rate=fps)
av/container/output.pyx:62: in av.container.output.OutputContainer.add_stream
    ???
av/codec/codec.pyx:179: in av.codec.codec.Codec.__cinit__
    ???
av/codec/codec.pyx:187: in av.codec.codec.Codec._init
    ???
E   av.codec.codec.UnknownCodecError: libx264

When pyav is not installed, all those tests are skipped as per the skip mark.

[NicolasHug]

After a0ff08b which ignores the dataset tests, there are still segfaults, this time in test_decode_jpeg_error() which seems completely unrelated 🤷

[NicolasHug]

Tried to avoid the segfault by only running the video tests. They're all being skipped because the extension cannot be loaded

vision/torchvision/io/_video_opt.py

Lines 11 to 15 in 1de7a74

	try:
	_load_library("video_reader")
	_HAS_VIDEO_OPT = True
	except (ImportError, OSError):
	_HAS_VIDEO_OPT = False

Printing the exception gives:

/lib64/libstdc++.so.6: version `GLIBCXX_3.4.20' not found (required by /opt/conda/envs/ci/lib/./libx265.so.199)

(torchvision is still built with ffmpeg support)

[NicolasHug]

Won't have time to dig this further for now.