Re: [4.2 PATCH]: backport solaris11 c99-math fixincludes patch

["Richard Guenther" <richard dot guenther at gmail dot com>]

On Sat, Mar 1, 2008 at 5:05 PM, Kaveh R. GHAZI <ghazi@caip.rutgers.edu> wrote:
>
> On Wed, 27 Feb 2008, Richard Guenther wrote:
>
>  > On Wed, Feb 27, 2008 at 8:23 PM, Andreas Tobler <andreast-list@fgznet.ch> wrote:
>  > > Hi Kaveh,
>  > >
>  > >
>  > >  Kaveh R. GHAZI wrote:
>  > >
>  > >  >> 2008-02-27  Kaveh R. Ghazi  <ghazi@caip.rutgers.edu>
>  > >  >>          Andreas Tobler  <a.tobler@schweiz.org>
>  > >  >>
>  > >  >>      * inclhack.def (solaris_math_10): New.
>  > >  >>      * tests/base/iso/math_c99.h: Update.
>  > >  >>      * fixincl.x: Regenerate.
>  > >  >>
>  > >  >
>  > >  > Hi Andreas - I believe this also applies to 4.1.  Bruce generally likes
>  > >  > fixincludes patches to be applied to all active branches.  See:
>  > >  > http://gcc.gnu.org/ml/gcc-patches/2008-02/msg01334.html
>  > >  >
>  > >
>  > >  I know, I just wanted to highlight that I have the patch already
>  > >  finished for 4.2. For 4.1 I need to co the branch and test.
>  >
>  > The 4.1 branch is in deep maintainance mode and should receive only
>  > serious wrong-code bug fixes and fixes for regressions on the branch itself.
>  > It also won't get any more releases due to GPL transition issues.
>  > Richard.
>
>  Richard,
>
>  I believe this bugfix qualifies for gcc-4.1.
>
>  It's a wrong code bug.  It's a regression (if you count that moving from
>  solaris10 to solaris11 you'll see a new bug).  The restriction
>  "regressions on the branch itself" is not followed *at all by anyone*.
>  Just look at the 4.1 ChangeLog, so please let's not apply that
>  qualification to just this one patch.
>
>  I agree it's a corner case, but whether it's "serious" to a user depends
>  on whether their code relies on the correct behavior of this feature.  So
>  IMHO we should balance the relative frequency of this case vs. the
>  invasiveness of the fix.
>
>  Regarding that, the patch is small & simple, done by fixincludes (i.e. not
>  a behavior change to cc1), it's limited to solaris11 in just one header
>  file on that OS.
>
>  And the fixincludes maintainer has expressed his policy regarding
>  backports is that "all active branches" should get all fixes.
>
> http://gcc.gnu.org/ml/gcc-patches/2008-02/msg01334.html
>  We generally give some leeway to maintainers within their area of
>  expertise.
>
>  Weighing all this together, I'd like you to please reconsider.

I am trying to set new "policies" here, in that with branch age there should
be a focus to fix regressions on the branch only, as after all for example
4.2 is available for other bugfixes.  The idea is that patches to an old branch
should have zero chance of causing regressions on that branch.

So I'm going to make that point more often if I notice such changes (I agree,
that there is no consistency here).

Richard.