Re: gcc version selection

[Jason Bacon <jwbacon%tds.net@localhost>]

On 10/03/14 07:29, Jonathan Perkin wrote:
> * On 2014-10-03 at 13:21 BST, Patrick Welche wrote:
>
> > On Fri, Oct 03, 2014 at 12:34:20PM +0100, Jonathan Perkin wrote:
> > > * On 2014-10-03 at 12:14 BST, Patrick Welche wrote:
> > >
> > > > I'm still confused by gcc version selection - how is it meant to work?
> > > >
> > > > On NetBSD 5/i386 (i.e., a system with gcc 4.1.3):
> > > >
> > > > # cd /usr/pkgsrc/graphics/openexr
> > > > # make show-var VARNAME=GCC_REQD
> > > > 4.3 2.8.0
> > > > # make show-depends
> > > > {gcc49,gcc49-libs}>=4.9:../../lang/gcc49-libs
> > > > ilmbase>=2.2.0nb1:../../graphics/ilmbase
> > > > #
> > > >
> > > > So, how did we get from 4.3 to 4.9?
> > >
> > > Likely because pkgsrc has no GCC 4.3, and the current logic is that if
> > > none of the patterns match we use the latest, i.e. 4.9.
> >
> > Actually, as rules go, that doesn't seem like a bad one. The only problem
> > is that -current has 4.8, so I saw trouble in ImageMagick that doesn't seem
> > to be happy being compiled with 4.9 (still to be looked at)
> >
> > > Yes, we should probably fix this.
> >
> > 1) teach mk/compiler/gcc.mk to use 4.4 for 4.3?
> > 2) change GCC_REQD in Makefiles from 4.3 to 4.4?
> > 3) something else?
>
> #1 in a general case, i.e. if there is no exact match then use the
> next higher version that we offer rather than simply jumping straight
> to the highest.
>
> We shouldn't do #2 as it would penalise users of native GCC 4.3.

Case in point: boost currently requires GCC45 for all but NetBSD.

(meta-pkgs/boost/Makefile.common)

This requirement prevents boost from building on CentOS 6.5, which uses 
GCC 4.4 as the base compiler.

Boost builds and works fine if the requirement is changed to GCC44.

--
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  Jason W. Bacon
  jwbacon%tds.net@localhost

  Circumstances don't make a man:
  They reveal him.
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~