Building glibc fails - needs .init_array and .fini_array support

[Jonathan Brandvein]

I've been trying to build OpenEmbedded for a few days now and I can't
get past the do_configure phase of building glibc-intermediate.

| checking for .preinit_array/.init_array/.fini_array support... no
| configure: error: Need linker with .init_array/.fini_array support.

It seems like this would be a problem with the way binutils-cross was
built. The closest thing I found to this being mentioned anywhere in
the context of open embedded was this bug:

http://projects.linuxtogo.org/pipermail/openembedded-issues/2007-January/001113.html
(At least, I think that's the correct link. I can't check as the site
is down at the moment.)

The symptom is the same but the cause is entirely different as I do
not have the MACHINE variable set in my shell environment.

I'm trying to build the Angstrom distribution for the AT91SAM9263EK. I
tried switching both the distro and machine to other variants (generic
distro and i686-generic machine), but the problem still persists. I've
been through several different branches including .dev, .stable, and
.angstrom-2007.12-stable, always making sure to update to the most
recent version as per the GettingStarted wiki page, and am amazed at
how nothing I try seems to affect this error. I'm using task-base as
the target, but that doesn't make much of a difference. I've also
tried wiping out the tree and starting over to make sure a bad
configuration didn't persist in the form of a built package
dependency.

Given that other people have apparently been able to build angstrom
for the 9263ek, and I still get this problem when I change machine
architectures and distros, I'm at a loss to explain what the problem
is. Nor do I know how to debug the OpenEmbedded build process, or even
what controls .init_array and .fini_array support in binutils, aside
from (somehow) the MACHINE variable. I'm getting that sinking feeling
that it could be a real stupid error on my part rather than a problem
in the package metadata. (Yesterday I managed to spend over an hour
"debugging" bitbake when it couldn't find bitbake.conf, before I
realized I had failed to export BBPATH from my .bashrc. That took me
down a peg or two, so I'm ready to believe this problem isn't
necessarily OE or angstrom's fault ;-) ).