BIND 10 #710: Compilation error on Python 3.2 systems

[BIND 10 Development]

#710: Compilation error on Python 3.2 systems
-------------------------------------+-------------------------------------
                   Reporter:  shane  |                 Owner:  jinmei
                       Type:         |                Status:  reviewing
  defect                             |             Milestone:
                   Priority:  minor  |  Sprint-20110628
                  Component:  build  |            Resolution:
  system                             |             Sensitive:  0
                   Keywords:         |           Sub-Project:  DNS
            Defect Severity:  N/A    |  Estimated Difficulty:  2.0
Feature Depending on Ticket:         |           Total Hours:  0
        Add Hours to Ticket:  0      |
                  Internal?:  0      |
-------------------------------------+-------------------------------------

Comment (by jinmei):

 Replying to [comment:12 jelte]:
 > Three commits, the most important one is the first; I've added a
 configure check that checks if we need -Wno-unused-parameter to include
 Python.h. If so, it is added to PYTHON_INCLUDES (not the very best place
 to put it, but this way we do know it'll only be used for things that
 actually use Python.h, and we don't need to add any workarounds to the
 various Makefiles).

 Unfortunately this doesn't (always) work.  (and even if it does, I'd
 probably object to abusing PYTHON_INCLUDES for this purpose:-).  In my
 system -Wno-something must be placed after the offending warning flag
 (normally -Wall or -Wextra).  I thought it's quite common, so it's
 actually surprising this worked for you:-)

 I committed a counter proposal to the branch: 97131f9.  I also moved
 the check within a g++ (and its variants) specific block in order to
 not disturb other compilers that don't understand -Wxxx.

 > The second fix is for a failing test (and presumably for failing code);
 apparently one of the APIs changed (or rather, a default parameter in a
 call), so I added an explicit str() around it.

 I suspect the revised code (with Python 3.2) doesn't produce the
 expected result.  The problem is that (as you pointed out) in Python
 3.2 xml.etree.ElementTree.tostring() always returns bytes, not a
 string (it seems to be an intentional change, see
 http://bugs.python.org/issue10942).  But simply wrapping a bytes
 object with str() results in a string in the form of "b'original
 string'", so it would result in "b'<foo>bar</foo>'" etc, instead of
 "<foo>bar</foo>".

 Assuming my concern is correct, I made a counter proposal and
 committed it (ed6ec07).

 And, this makes me think it indicates tests for stats_httpd are not
 sufficient.  We should have a test to see if StatsHttpd.xml_handler()
 produces expected results.  Then (again assuming my concern is
 correct) you should have been able to notice the first fix wasn't
 sufficient.  This is not a problem of this branch per se, but for the
 original code of stats.  So I suggest we create a separate ticket for
 that.

 > And lastly, python 3.2 now stores .pyc files in {{{__pycache__}}}
 directories, which need to be removed (for distcheck, mainly). This does
 update a lot of makefiles. I could not find a way to completely remove
 directories through a magic automake variable, so I added CLEANDIRS and a
 clean-local target in all directories containing python code.

 This one looks okay.  It would be nicer if we have a common template
 for python Makefile.am's including this one so that we won't have to
 copy-paste this block every time we create a new Makefile.am for a new
 python lib/program, but for now I'm okay with it.

 Finally, I think we should have at least one buildbot that uses Python
 3.2.  I suggest we create a separate ticket for that.

 If you are okay with my changes, the only remaining point is the
 missing changelog.

-- 
Ticket URL: <https://bind10.isc.org/ticket/710#comment:15>
BIND 10 Development <http://bind10.isc.org>
BIND 10 Development