Build breaks in video.swift

[jjedele]

When trying to build the current master, it fails for me in video.swift with following error:

ERROR: /Users/jeff/repos/io/tools/build/swift/BUILD:17:1: Compiling Swift module video failed (Exit 1)
tensorflow_io/core/swift/video.swift:40:12: error: value of type 'CMSampleBuffer' has no member 'numSamples'
        if sampleBuffer.numSamples != 1 {
           ^~~~~~~~~~~~ ~~~~~~~~~~
tensorflow_io/core/swift/video.swift:41:40: error: value of type 'CMSampleBuffer' has no member 'numSamples'
            print("number of samples \(sampleBuffer.numSamples) is not supported")
                                       ^~~~~~~~~~~~ ~~~~~~~~~~
tensorflow_io/core/swift/video.swift:93:9: warning: initialization of immutable value 'deviceName' was never used; consider replacing with assignment to '_' or removing it
    let deviceName = String(cString: devname)
    ~~~~^~~~~~~~~~
    _
tensorflow_io/core/swift/video.swift:138:16: warning: comparing non-optional value of type 'UnsafeMutablePointer<VideoContext>' (aka 'UnsafeMutablePointer<(session: AVCaptureSession, semaphore_in: DispatchSemaphore, semaphore_out: DispatchSemaphore, delegate: VideoDataOutputSampleBufferDelegate)>') to 'nil' always returns true
    if context != nil {
       ~~~~~~~ ^  ~~~
tensorflow_io/core/swift/video.swift:158:16: warning: comparing non-optional value of type 'UnsafeMutablePointer<VideoContext>' (aka 'UnsafeMutablePointer<(session: AVCaptureSession, semaphore_in: DispatchSemaphore, semaphore_out: DispatchSemaphore, delegate: VideoDataOutputSampleBufferDelegate)>') to 'nil' always returns true
    if context != nil {
       ~~~~~~~ ^  ~~~

I did expunge and rebuild 2 times, but it didn't help. Also when I look at the CMSampleBuffer API Docs, there is only a sampleCount and no numSamples attribute in there.

Any ideas what the problem could be?

[yongtang]

@jjedele Do you know the Xcode version you are using? Maybe it is the version mismatch as the numSamples is in the docs:
https://developer.apple.com/documentation/coremedia/cmsamplebuffer

[jjedele]

@yongtang 10.3. I will update it and see write here if it works helps.

[yongtang]

@jjedele My Xcode is 11.3.1, let me know if upgrade works. Otherwise I can try to see if I can update the code to support both Xcode 10 and Xcode 11.

[jjedele]

@yongtang I updated Xcode and now it builds, but when I'm trying to load the library I get following error:

Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/Users/jeff/anaconda3/lib/python3.7/site-packages/tensorflow_core/python/framework/load_library.py", line 152, in load_library
    py_tf.TF_LoadLibrary(lib)
tensorflow.python.framework.errors_impl.NotFoundError: dlopen(bazel-bin/tensorflow_io/core/python/ops/libtensorflow_io.so, 6): Symbol not found: _$sSo17CMSampleBufferRefa9CoreMediaE10numSamplesSivg
  Referenced from: bazel-bin/tensorflow_io/core/python/ops/libtensorflow_io.so (which was built for Mac OS X 10.15)
  Expected in: /usr/lib/swift/libswiftCoreMedia.dylib
 in bazel-bin/tensorflow_io/core/python/ops/libtensorflow_io.so

Seems I have to update something else. Hope it's not the whole OS. Looking into it.

[jjedele]

Xcode now also 11.3.1, OSX 10.14.6.

[yongtang]

@jjedele It looks like it is due to the Swift 5 Runtime? Though I am not so sure. In old versions of macOS the Runtime needs to be installed separately, but 10.14.6 could be new enough already.

[jjedele]

@yongtang Haven't been able to figure this out yet unfortunately. Apparently Swift 5 is in XCode 11.3 which I have installed. Is there an way to deactivate compiling the Swift parts selectively?

[yongtang]

Is there an way to deactivate compiling the Swift parts selectively?

@jjedele It may not be straightforward though you could:

1. In tensorflow_io/core/BUILD remove line 216 "//tools/build/swift:audio_swift",
2. In tensorflow_io/core/BUILD remove line 236 "//tools/build/swift:video_swift",
3. In tensorflow_io/core/kernels/audio_mp4_kernels.cc remove line 23-33 so that you have dump functions: https://github.com/tensorflow/io/blob/master/tensorflow_io/core/kernels/audio_mp4_kernels.cc#L23-L33
4. In tensorflow_io/core/kernels/video_kernels.cc remove line 20-24 so that you have dummy functions: https://github.com/tensorflow/io/blob/master/tensorflow_io/core/kernels/video_kernels.cc#L20-L24

After that you can selectively build .so only:

bazel build -s --verbose_failures //tensorflow_io/core:python/ops/libtensorflow_io.so

[jjedele]

@yongtang Thx, that kind of worked. Unfortunately I'm now back to (like before the Kafka update)

tensorflow.python.framework.errors_impl.NotFoundError: dlopen(bazel-bin/tensorflow_io/core/python/ops/libtensorflow_io.so, 6): Symbol not found: _timespec_get
  Referenced from: bazel-bin/tensorflow_io/core/python/ops/libtensorflow_io.so (which was built for Mac OS X 10.15)
  Expected in: /usr/lib/libSystem.B.dylib
 in bazel-bin/tensorflow_io/core/python/ops/libtensorflow_io.so

I'm getting the feeling that something's messed up with my environment, it somehow also says the library is built for OSX 10.15 despite I'm on 10.14.

[yongtang]

Ah, I think you may need to pass

            --copt -mmacosx-version-min=10.13 \
            --linkopt -mmacosx-version-min=10.13 

to bazel, so:

bazel build -s \
    --copt -mmacosx-version-min=10.13 \
    --linkopt -mmacosx-version-min=10.13 \
 --verbose_failures //tensorflow_io......

We actually passes those two options in our CI:
https://github.com/tensorflow/io/blob/master/.github/workflows/build.yml#L84-L106

So it was not getting noticed. Probably should pass those two options in .bazelrc (created from https://github.com/tensorflow/io/blob/master/tools/build/configure.py#L104-L106 ), so that it is not needed to pass that two long options in command line.

[yongtang]

@jjedele The issue should have been fixed by #953. All usage of 10.15 APIs in swift have been removed. You still need to pass `--copt="-mmacosx-version-min=10.13" and --linkopt="-mmacosx-verson-min=10.13" to bazel build, then build will be successful.

I will close this issue for now, but feel free to re-open if the issue persists.