Agreed!

In fact, I think this makes it so all print() calls by default write to stderr:

  >>> import sys, functools, builtins
  >>> real_print = print
  >>> builtins.print = functools.partial(real_print, file=sys.stderr)
  >>> print("Hello")
  Hello

Or use logging module as you should, which prints to stderr by default.

    import logging
    logging.basicConfig()

    logging.info('something happened')

The problem for me with logging is that it's slow for production code:

Python 2.7.15 Linux (logging is 60x slower):

  [root@hbtest ~]# py -m timeit -s 'import logging' "logging.info('something happened')"
  1000000 loops, best of 3: 1.31 usec per loop

  [root@hbtest ~]# py -m timeit -s 'import sys; debug = False' "if debug: print >>sys.stderr, 'something happened'"
  10000000 loops, best of 3: 0.0216 usec per loop

Python 2.7.15 OSX (logging is 71x slower):

  $ py -m timeit -s 'import logging' "logging.info('something happened')"
  1000000 loops, best of 3: 0.925 usec per loop

  [jim@mbp hbrel]$ py -m timeit -s 'import sys; debug = False' "if debug: print >>sys.stderr, 'something happened'"
  100000000 loops, best of 3: 0.013 usec per loop

Python 3.6.8 Linux (logging is 99x slower):

  [root@hbtest ~]# python3 -m timeit -s 'import logging' "logging.info('something happened')"
  1000000 loops, best of 3: 1.62 usec per loop

  [root@hbtest ~]# python3 -m timeit -s 'import sys; debug = False' "if debug: print >>sys.stderr, 'something happened'"
  100000000 loops, best of 3: 0.0163 usec per loop

And, as usual, Python 3 is 25% slower than Python 2 for the logging case (but 2x faster for the noop if debug case). I hope the recent efforts to increase Python 3 performance are successful.

Microseconds are not important unless in a very tight loop. Avoid there and/or make a shortcut beforehand.

And we don't know why tus666 wants to print to stderr, nor what the issue was with Python, so this digression seems built on sand.