Tracing fails, if a program calls __main__.main

[AndreasGocht]

When a tensorflow example uses tf.app.run() the tracing fails.

The reason is, that tf.app.run does the following [1]:

  main = main or _sys.modules['__main__'].main

  # Call the main function, passing through any arguments
  # to the final program.
_sys.exit(main(_sys.argv[:1] + flags_passthrough))

_sys.modules['__main__'].main calls the main of the tracing module. This brakes.

Possible solution:

import the target program and forward the call to target.main()

[AndreasGocht]

I have no Idea how to call the original target.main().

I added an error message, which should help to address the problem manually. I think this is all I can do.

[AndreasGocht]

It might be that runpy can help here:
https://docs.python.org/3/library/runpy.html#runpy.run_path
This needs to be checked.