Re: GCC and as

[Jason Bacon <outpaddling%yahoo.com@localhost>]

On 05/16/18 16:41, Jason Bacon wrote:
> On 05/15/18 16:31, Jason Bacon wrote:
> > On 05/15/18 15:22, J. Lewis Muir wrote:
> > > On 05/13, Jason Bacon wrote:
> > > > The problem is actually more widespread than I realized.  I found
> > > > more failures in the logs from my 2018Q1 bulk builds, including
> > > > qt5-qtbase, which prevented several other packages from building on
> > > > CentOS 6.
> > > >
> > > > I'd like to get a patch into all the GCC packages ASAP.
> > > >
> > > > Does anyone have suggestions for improving on the patch above? After
> > > > giving it a little thought, I don't like the idea of directly
> > > > checking the version of tool chains outside pkgsrc, as they may
> > > > change or even be removed unbeknownst to pkgsrc.  I think
> > > > consistently using pkgsrc binutils when Enterprise Linux is detected
> > > > will lead to the fewest problems.
> > > Hi, Jason!
> > >
> > > Hmm, to me this doesn't seem like the best way to go.  I would be
> > > in favor of checking for the feature you need rather than the OS.
> > > Or in this case, if you need a minimum version of binutils, I think
> > > it's better to check the binutils version rather than the OS version.
> > > This seems similar to the principle of checking for the feature in an
> > > Autotools configure script rather than checking for the OS as noted in
> > > the pkgsrc guide in section 11.3.4, Patching guidelines, at:
> > >
> > > http://www.netbsd.org/docs/pkgsrc/components.html#components.patches.guidelines 
> > >
> > >
> > > Regards,
> > >
> > > Lewis
> > OK, thanks.  I'll give this some thought while I do more testing. The 
> > patch is working great for normal builds
> > but for some reason it's failing to find the linker under pbulk. I'll 
> > get back to you when I have more info.
> >
> > Regards,
> >
> >     Jason
>
> So pbulk revealed that we also need to set
>
> BUILDLINK_DEPMETHOD.binutils=   full
>
> Pbulk deinstalls build dependencies as soon as they are no longer 
> needed, so it exposed
> the risk of binutils going away after gcc5 is built.
>
> Here is my current patch:
>
> .if exists(/etc/redhat-release)
> CONFIGURE_ARGS+=        --with-gnu-as --with-as=${PREFIX}/bin/gas
> CONFIGURE_ARGS+=        --with-gnu-ld --with-ld=${PREFIX}/bin/gld
> BUILDLINK_DEPMETHOD.binutils=   full
> .  include "../../devel/binutils/buildlink3.mk"
> .endif
>
> I've hit some libdl issues that I suspect might be due to the fact 
> that my tree had other packages installed
> built with the old binutils, so I'm rebuilding a test package in a 
> pristine tree with the patched gcc5.
>
> Also doing a full pbulk build with the patch.
>
> While we're waiting for that, how exactly would you suggest changing 
> the patch above to check the
> binutils version instead of the Linux distro?

One more thing:

If we go with an OS check, I would switch to

.if ${OS_VARIANT} == "redhat"

instead of the exists() check, since this is already supported in 
bsd.prefs.mk.