The tests from this patch fail on Linux.

-----
First: There is no trailing % test on Linux, and glibc's strftime() happily ignores a trailing %, so no ValueError is raised.

Python should do either one or the other of the following:

1) Python should enforce no trailing % in the strftime format string,
   or

2) the test suite shouldn't assume that a trailing % in the strftime
   value string raises a ValueError.

I can live with either of these, not sure what the right decision is.

-----
Second: The test from the patch assumes that strftime('%#') will raise a ValueError.  Again, strftime in Linux glibc happily accepts "%#" as a format string and thus no ValueError is raised.

Python is agnostic about native format units in the strftime() format string.  Therefore I strongly assert that Python must not assume that "%#" is an illegal format string.  Therefore the tests must not assume that "%#" raises ValueError.

Given that the code used to crash, I do want the code path exercised in the test suite.  So I propose that the test attempt time.strftime('%#') and accept either success or ValueError.


-----

Given that I've accepted this patch into 3.5.0, and it's now blocking my release, it is implicitly a "release blocker".  I need to resolve this tonight before I can tag 3.5.0rc3.  I'm going to dinner, and maybe we can have a quick discussion and come to a decision in the next hour or two.


p.s. The checkin also flunked PEP 7.  *sigh*