svn commit: r258672 - in head: . share/mk

[Konstantin Belousov]

On Tue, Nov 26, 2013 at 10:34:30PM -0800, Peter Wemm wrote:
> On 11/26/13, 9:03 PM, Glen Barber wrote:
> > On Wed, Nov 27, 2013 at 04:54:24AM +0000, Peter Wemm wrote:
> >> Author: peter
> >> Date: Wed Nov 27 04:54:23 2013
> >> New Revision: 258672
> >> URL: http://svnweb.freebsd.org/changeset/base/258672
> >>
> >> Log:
> >>   At great personal risk, change the default for LIB32 from yes to no.  As
> >>   mentioned in UPDATING, you can even do it as an as-needed operation after
> >>   doing a buildworld/installworld.  You can set WITH_LIB32=yes in make.conf
> >>   or src.conf.
> >>
> > 
> > Thank you.  Long overdue, IMHO.
> > 
> > Glen
> > 
> 
> A slightly longer explanation of what I was thinking:
> 
> - There's a new round of 'make -j' problems lurking in there.  We are
> missing chunks of the ordering glue that cause libraries to be built in the
> right order when they depend on each other.
> - It's a waste of cpu time for the usual case, particularly for the 11.x
> cycle for the next 1-2 years.
Why ?

> - We don't build them properly - we invent cpu flags etc.
What do you mean there ?  Do you reference the fact that lib32 build on
amd64 assumes sse2 and all previous coprocessor extensions ?

> 
> The usual use case for 32 bit binaries seems to be:
> - running a 32 bit chroot or jail - this is unaffected.
> - running old binaries, usually from 4.x or 6.x when the 64 bit port was
> really green - WITH_LIB32 doesn't actually help much with this because most
> of the libraries are missing.
> 
> It seems more likely we can do a better job with packages.  With some
> massaging, we should be able to use the compat-6.x/i386 libraries as-is, and
> solve the "old 4.x/6.x binary" issue in one go.
> 
> However, ld-elf32.so.1 does require special handling.  I have something in
> mind that might make this moot though.
> 
> I suspect I've made the powerpc folks angry though...

I disagree with the change. It was not discussed, and the motivation
presented ('the build has bugs') is not valid for removing a useful
feature.

All other OSes I am aware of implement multi-arch fully. For 10/11, we
have quite good compat32 layer for non-managing interfaces, and have
user-mode compilation environment. I think that the route to go forward
is to have multi-arch for ports, or at least, enable to have 32bit ports
installation on 64bit host.

I think that this is step backward.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 834 bytes
Desc: not available
URL: <http://lists.freebsd.org/pipermail/svn-src-all/attachments/20131127/66b39689/attachment.sig>