On Sun, Jun 20, 2010 at 06:27:00PM -0400, Mike Frysinger wrote:
> On Sunday, June 20, 2010 09:55:39 Arfrever Frehtes Taifersar Arahesis wrote:
> > 2010-06-19 22:53:37 Mike Frysinger napisał(a):
> > > On Thursday, June 10, 2010 16:45:29 Arfrever Frehtes Taifersar wrote:
> > > > 2010-06-10 22:20:44 Nirbheek Chauhan napisał(a):
> > > > > On Fri, Jun 11, 2010 at 1:30 AM, Arfrever Frehtes Taifersar wrote:
> > > > > > 2010-06-10 21:27:40 Jeremy Olexa napisał(a):
> > > > > >> I see no reason to *not* add a ChangeLog entry here.
> > > > > > 
> > > > > > ChangeLog entries are not required for trivial changes.
> > > > > 
> > > > > A "trivial" change is fixing a typo, or a manifest problem, a missing
> > > > > quotation mark, etc. Anything else is not "trivial".
> > > > > 
> > > > > Anything that changes how an ebuild functions, what it does, or the
> > > > > installed files (and/or their contents) is NOT a trivial change.
> > > > 
> > > > This commit only removed some compiler warnings.
> > > 
> > > mucking with CFLAGS without documentation is wrong.  compiler warnings
> > > come and go, so a flag that was relevant one day could be completely
> > > extraneous the next.
> > > 
> > > however, especially with strict aliasing, you arent "just fixing
> > > warnings", you're changing optimization behavior of gcc to workaround
> > > broken C code. this obviously does not fall anywhere near the "trivial"
> > > mark.
> > > 
> > > i see you still havent fixed this, so get on it already.  a bug needs to
> > > be opened somewhere to get the package properly *fixed*
> > 
> > This problem is probably caused by bugs in Python 2, which have been fixed
> > in Python 3.
> 
> the new information you've provided here only reinforces the fact the current 
> code is wrong.  properly document the append-flags and properly restrict it to 
> when python-2 is being used.
> 
> considering python-2 is going to be in our lives for a while, how hard is it 
> to backport the changes in question to the headers ?  presumably it's 
> struct/union/cast shuffling in the headers.

This is upstream python bug 969718 offhand; basically if cflags exists 
for distutils consumers, base cflags don't make it fully through- 
meaning no -fno-strict-aliasing as is generally needed for building 
python extensions.

As for py3k, looks of it py3.1 still suffers it.


Either way, this is the wrong thing to fix- python's distutils 
needs fixing, not consumers.  In snakeoil, we detect and fix it on the 
fly to provide a fixed version of distutils, but obviously not many 
pkgs consume that...

~harring