The change e42e2bd47168 introduced a regression, test_os started to fail on FreeBSD. What is the rationale of the change? For example, test_os pass in the previous build on the same buildbot.

I don't know if it makes sense to pass None to headers and/or trailers.

Please revert the change or fix the test. ((But we must continue to test headers=None and trailers=None, even if these parameters now raise a TypeError).

======================================================================
ERROR: test_keywords (test.test_os.TestSendfile)
----------------------------------------------------------------------
Traceback (most recent call last):
  File "/usr/home/buildbot/python/3.x.koobs-freebsd10/build/Lib/test/test_os.py", line 2334, in test_keywords
    headers=None, trailers=None, flags=0)
TypeError: sendfile() headers must be a sequence or None

I believe revision e42e2bd47168 is actually the cure, and test failure is actually caused by a new test I added in revision fdb5d84f9948. So hopefully things are fixed now, but I will keep an eye on the buildbots.

In fdb5d84f9948, I originally added the test_keywords() test to ensure the correct parameter name of “count” to sendfile().

Neither headers=None nor trailers=None were ever supported. It is just that the documentation and error messages which mislead me to write that test.

Victor, the BSD and OS X buildbots seem to be passing test_os now. E.g.:

http://buildbot.python.org/all/buildslaves/koobs-freebsd10
http://buildbot.python.org/all/buildslaves/intel-yosemite-icc

Do you agree we can close this?

Sure, close the issue if the regression is fixed.