Advice about /usr/ports/math/gmp

[Baptiste Daroussin]

On Fri, Jan 03, 2014 at 08:23:06PM +0100, Tijl Coosemans wrote:
> On Fri, 3 Jan 2014 19:37:48 +0100 Baptiste Daroussin wrote:
> > On Fri, Jan 03, 2014 at 07:27:29PM +0100, Dimitry Andric wrote:
> >> On 03 Jan 2014, at 17:53, Torbjorn Granlund <tg at gmplib.org> wrote:
> >>> We are about to release GMP 5.2.
> >>> 
> >>> We have been forced to add three FreeBSD-related items to the releases
> >>> notes:
> >>> 
> >>>  * This release will not work on FreeBSD/amd64 7.x, 8.x or 9 series
> >>>    before 9.3 with a Haswell CPU or any other CPU which supports the
> >>>    BMI2 instructions.  The reason is that the FreeBSD m4 command is not
> >>>    correctly implemented.  (Workaround: Use an older GMP release, or
> >>>    install GNU m4 from /usr/ports and tell GMP to use it.)
> >>> 
> >>>  * This release will not work on FreeBSD/amd64 before version 10 using
> >>>    the 32-bit ABI.  The reason is broken limits.h and broken dynamic
> >>>    linking.  (Workaround: Use an older GMP release if using the 32-bit
> >>>    ABI on these FreeBSD releases is important.)
> >>> 
> >>>  * This release will not work on FreeBSD/amd64 10.0 using the 32-bit
> >>>    ABI.  The reason is bugs in the compiler 'clang'.  (Workaround:
> >>>    Compiling gcc from /usr/ports might work, except that gcc depends on
> >>>    GMP; we have not been able to test that workaround since
> >>>    FreeBSD/i386 10.0 does not work for us under KVM or Xen.)
> >>> 
> >>> The first item is a show-stopper.  It would be possible to implement a
> >>> workaround in GMP.  We choose not to do that since (1) we adviced the
> >>> FreeBSD project two years ago the m4 bug, and FreeBSD chose to make 4
> >>> releases without fixing m4, and (2) the fix is ugly, and (3) our use of
> >>> m4 which triggers the bug is actually part of a workaround for a broken
> >>> assembler (to much complexity to maintain workarounds for workarounds).
> >>> 
> >>> The second item should not affect /usr/ports builds since they would use
> >>> the default 64-bit ABI on amd64 machines.
> >>> 
> >>> The third item is a show-stopper until clang is fixed.  We have not been
> >>> able to isolate this problem due to lack of time and due to a deeply
> >>> malfunctioning filesystem of FreeBSD/i386 under KVM and Xen+NetBSD.  We
> >>> don't have any more information about these bugs.
> >>> 
> >>> 
> >>> We do not plan to implement workarounds for the above bugs for GMP 5.2.x
> >>> for any x.  I would advice that you stick with GMP 5.1.3.
> >> 
> >> Uhm, if you provide approximately zero information about these supposed
> >> "bugs", how do you expect anyone to help fixing them?
> > 
> > In particular concerning m4, I'd like to hear what is buggy and how, our m4 do
> > respect the m4 specs:
> > http://pubs.opengroup.org/onlinepubs/9699919799/utilities/m4.html plus it has
> > support for some (all?) of the gnu extension via the -g option.
> > 
> > If anything is buggy I would like to hear about it and fix.
> 
> The first item is http://www.freebsd.org/cgi/query-pr.cgi?pr=166994

Oh great I see, I forgot to mfc it. Thanks for having merge it.

> The second item is about compiling with -m32.  We only fully implemented
> that in FreeBSD 10.
> The third item I don't know anything about.

regards,
Bapt
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 196 bytes
Desc: not available
URL: <http://lists.freebsd.org/pipermail/freebsd-ports/attachments/20140103/0fdfa3c4/attachment.sig>