[oe] Where do I override DISTRO_FEATURES?

[Thilo Fromm]

Hello Denys,

>> >> I'm trying to remove an element from DISTRO_FEATURES (specifically
>> >> "ld-is-gold"), but I am unable to find the right spot to do it. The
>> >> distribution we're using (Angstrom, as it happens) sets this in
>> >> conf/distro/angstrom-v2012.x.conf. My machine configuration is the
>> >> wrong spot to modify DISTRO_FEATURES, since it will be processed
>> >> first. local.conf is not working, too.
>>
>> It's a backwards compatibility thing: we're bound to gcc-4.5 (we have
>> kernel build issues with newer gcc versions; we're nailed to
>> linux-2.6.37 thanks to lack of support by Texas Instruments) which
>> fails to detect the correct version of binutils GOLD (LD would work)
>> and therefore misses critical features which break the build later on.
>> I won't, however, try to convince the Angstrom guys to further support
>> ancient toolchains and make gold optional again just because of TI's
>> laziness.
>
> I think you are severely misinformed! Above issues (binutils-2.20/gcc-4.5
> requirement and gold ld problem with Thumb) were already fixed. I've been
> testing the builds with gcc-4.6 and binutils-2.22 for few weeks and even
> gcc-4.7 with gold linker since last week.

You're right, gcc-4.7 works well in the current setup. gcc-4.5,
however, does not detect the right binutils version with the GOLD
linker. GOLD version output is:

GNU gold (GNU Binutils 2.22) 1.11

The configure logic of gcc-4.5 would extract "1.11", not "2.22", for
the version number. Since Binutils 1.11 is ancient, some features of
gcc-initial are not set up correctly (gcc thinks the linker does not
support the "hidden" attribute). This breaks the build of
eglibc-initial later on.

I overslept the whole thing and I guess I'll just provide a patch to
gcc-4.5's configure so the GOLD linker version is detected correctly
instead of setting up my own distribution config.

> I don't know what layer you are
> using, but you may need to update, if you are using meta-ti.

meta-ti does not provide any toolchain anywhere. I have a toolchain
issue. The issue is with gcc-4.5 from meta-oe/toolchain-layer in
combination with binutils-2.22 from
openembedded-core/meta/recipes-devtools/.

Just in case you're curious, our layer structure is documented here:
<https://github.com/DFE/HidaV/wiki/Hidav-oe-layers>.

> Otherwise, if your platform is not directly supported by meta-ti, please send
> the bug report with the error message to meta-ti mailing list and I'll try to
> address it.

Which meta-ti mailing list are you talking about? There seem to be
many: this one, the arago lists, yocto-ti, ... And then again, it's a
toolchain issue (see above); meta-ti doesn't have toolchains.

> As of linux-2.6.37 for your specific platform - this one I can't help with,
> you would need to talk to your TI rep from the division that makes the
> platform and convince them to provide support for newer kernels.

I know, it's OK. We talked to TI a lot for quite a while, but our
realistic assessment from our TI experiences is that they'll never do
it. There are still a lot of issues with their old 2.6.37 kernel, TI
continually keeps breaking things. They don't have proper issue
trackers or mailing lists for communication with their developers.
>From the reps and FAEs we'd get the occasional empty promise, but
that's it. Personally I think we went wrong with buying TI in the
first place.

Regards,
Thilo

-- 
Dipl.-Ing (FH) Thilo Fromm, MSc., Embedded Systems Architect
DResearch Fahrzeugelektronik GmbH
Otto-Schmirgal-Str. 3, D-10319 Berlin, Germany
Tel: +49 (30) 515 932 228   mailto:fromm at dresearch-fe.de
Fax: +49 (30) 515 932 77    http://www.dresearch.de
Amtsgericht: Berlin Charlottenburg, HRB 130120 B
Ust.-IDNr. DE273952058
Geschäftsführer: Dr. M. Weber, W. Mögle