Re: ports broken on clang archs

On Thu, Sep 28, 2017 at 05:06:39PM +0200, Jeremie Courreges-Anglas wrote:
> On Thu, Sep 28 2017, Stuart Henderson <stu@spacehopper.org> wrote:
> > On 2017/09/27 20:00, Jeremie Courreges-Anglas wrote:
> >> On Wed, Sep 27 2017, Christian Weisgerber <naddy@mips.inka.de> wrote:
> >> > We're going to lock the ports tree for the release any day now.
> >> >
> >> > Please, no more imports or regular updates.
> >> >
> >> > Issues I'd like to see taken care of before the release:
> >> > * Fix the remaining instances of nested functions in autoconf checks.
> >> > * Pick up Firefox 56.
> >>
> >> > * Mark as BROKEN the remaining ports that have failed to build since
> >> > the clang switch on amd64/i386.
> >>
> >> Should probably use NOT_FOR_ARCHS = ${CLANG_ARCHS}.
> >
> > I would prefer "COMPILER = base-gcc" if you don't mind, the COMPILER
> > mechanism automatically handles ONLY_FOR_ARCHS.
>
> That would disable said ports on eg amd64, like what we have now
> (NOT_FOR_ARCHS).
>
> The first compiler that matches criteria will be chosen. On
> clang-based architectures, even though gcc is still compiled in
> base, ???base-gcc??? never matches.
>
> I understand that relying on gcc-4.2.1 is not a long-term solution, but
> I wonder why 'base-gcc' couldn't just set up a port to use
> /usr/bin/g(cc|++).
>
> After all, we still need gcc-4.2.1 in base for things like
> sysutils/firmware/vmm:
>
> # XXX errors with with clang
> # ./src/string.c:24:16: error: invalid output constraint '=Qi' in asm
> CC = /usr/bin/gcc
>

Not changing that this close to 6.2.

We need a long term solution past 6.2