Re: Detecting hard float/version in pkgsrc

[coypu%SDF.ORG@localhost]

On Sun, Jul 24, 2016 at 03:17:59PM +0200, Manuel Bouyer wrote:
> On Sun, Jul 24, 2016 at 12:48:47PM +0000, coypu%SDF.ORG@localhost wrote:
> > A good number of packages seem to default a worse case of hard float as
> > opposed to checking.
> > 
> > e.g.:
> > - lang/go/version.mk, assumes VFPv1 (GOARM=6 default). it could use VFPv3
> > for newer hardware, and it would probably not work for softfloat.
> > 
> > they will likely transition to GOARM=7 default at some point, and in the
> > absence of detection, it might be broken for older hardware.
> > 
> > - x11/pixman/Makefile is an example of all i386 being penalized for most
> > machines not having SSE2.
> > 
> > Others example of hacks exist, e.g. -mstackrealign for Firefox on i386,
> > but it is probably only needed for non-SSE2 machines. Or ffmpeg.
> > 
> > It might be nicer to have fancier float checks than just using
> > MACHINE_ARCH and assuming the worst case scenario when such hacks are
> > needed.
> > 
> > Opinions?
> 
> You seems to assume you'll be running the packages on the machine which
> built them. For me it's almost never the case (either because I'm
> using prebuilt packages, or because I have a dedicated build machine).
> 
> -- 
> Manuel Bouyer <bouyer%antioche.eu.org@localhost>
>      NetBSD: 26 ans d'experience feront toujours la difference

We can keep on defaulting for the worse case.
But it would be more technically correct to have if HAS_SSE2 rather than
poorly matching MACHINE_ARCH.

I imagine these string comparisons to i386 are not too valid for
non-NetBSD...

We could theoretically let users use ./bootstrap -native for those, put
in devel/cpuflags, too...