bug#11669: 24.1; Build failure on amd64 with x32 ABI

[Ulrich Mueller <ulm@gentoo.org>]

>>>>> On Thu, 21 Jun 2012, Glenn Morris wrote:

>> I've now applied a somewhat different solution to the Gentoo
>> package, namely patching the m/amdx86-64.h file, see attached
>> patch.

> The two patches don't (?) seem to be equivalent (wrt DATA_SEG_BITS
> and ULIMIT_BREAK_VALUE).

There's an #undef DATA_SEG_BITS in m/amdx86-64.h, but DATA_SEG_BITS
isn't defined at that point, in the first place. (It's only being
defined in some machine and system files like HPUX and IRIX that are
not relevant for our case.)

Concerning ULIMIT_BREAK_VALUE: This is used only in vm-limit.c.
However, the branch of the code where it is used isn't reached for
modern GNU/Linux systems, because for these both HAVE_GETRLIMIT and
RLIMIT_AS should be defined.

(And I wonder if is even possible to run a recent Emacs with an
address space limited to 32 MiB, as the ULIMIT_BREAK_VALUE of
32*1024*1024 suggests? This looks like a relic from the early 1990s
to me.)

> I'd prefer to install the configure one, since it is more consistent
> with previous treatments. Unless there is some reason to prefer the
> other one? Accidental merging to trunk is not an issue.

*shrug* I don't have a strong preference. Patching the machine file
looked slightly more systematic to me, because x32 really isn't
intel386.