[OE-core] psuedo-native strangeness on 64 bit build machine

[Steve Sakoman]

On Wed, Feb 29, 2012 at 12:45 AM, Petr Štetiar <ynezz at true.cz> wrote:
>> On 2/28/12 10:35 PM, Steve Sakoman wrote:
>>> I did a clean build on an Ubuntu 64 bit build machine a couple of days
>>> ago with no issues.
>
> I can confirm same situation here.
>
>>> After a pull of today I did an image build for an OMAP3 machine and
>>> encountered an issue with psuedo-native.
>
> Just pseudo? What's your setup, I mean layers etc. Are you using Angstrom's
> setup-scripts?

For a simple image, just pseudo-native.  As to layers, meta,
meta-yocto, meta-oe and meta-gnome from meta-openembedded, and my own
bsp layer.

I am not using Angstrom or the Angstrom setup scripts.

>>> Has anyone else been successful building on 64 bit machines?
>
> What version of Ubuntu is that? I'm on 12.04.

I am on 10.04 64 bit server edition with ia32-libs package installed.
The build machine is quad core.

>> I'm building pseudo on 64-bit machines right now, no problem.
>
> I think, that it's some problem with parallel make. If I clean the stamps and
> tmp dir of pseudo-native manually (-c cleanall doesn't work actually, because
> pseudo-native is missing, chicken-egg problem? :)) and set PARALLEL_MAKE=1
> NUM_THREADS=-j1, then it builds just fine.

If so, it is only a parallel make issue on 64 bit -- my build on an 8
core desktop 10.04 setup works flawlessly.

> There must have been something introduced in the bitbake or the layers
> recently, since I was able to build Angstrom/systemd-image/beagleboard from
> scratch 3 weeks ago with the same settings (PARALLEL_MAKE=4 NUM_THREADS=-j4").
> Now it's impossible to finish the build, the most noticeable failure is gcc,
> which fails for me in patching task due to the parallel build settings.
> Changing the parallel settings back to 1 helps here also.

I just tried building gcc and can confirm that it fails for me on the
64 bit build machine.

Steve