Another problem in current -dev

Matthew Burgess matthew at linuxfromscratch.org
Wed Sep 1 12:31:03 PDT 2010


On Wed, 1 Sep 2010 20:24:19 +0100, Ken Moffat <zarniwhoop73 at googlemail.com> wrote:
> On 1 September 2010 14:41, Matthew Burgess <matthew at linuxfromscratch.org>
> wrote:
> 
>> I lied before [0] and I do see this.  Why are we using
> 'LD_LIBRARY_PATH' instead
>> of --disable-libgomp in pass2 as we do in pass 1? (yes, I'm well aware
> that it was
>> me that committed r9254 that added this as part of the upgrade to
> GCC-4.5.0).
>> If noone has any objections, I'll change this to what we do in pass1
> once I've
>> tested it does the right thing.
>>
>> Regards,
>>
>> Matt.
> 
>  Actually, I'm not that fussed about doing this either way - for some
> reason r9254 had never landed in my gmail inbox.  Using
> LD_LIBRARY_PATH is arguably as educational as disabling libgomp,

Well, it would be if we explained what it's doing/why we're doing it.
As it it, we already use and explain --disable-libgomp in pass1 so we
may as well re-use it here.

>  I'm more concerned about perl:
> http://linuxfromscratch.org/pipermail/lfs-dev/2010-August/064258.html
> On 32-bit x86 you can include objects compiled without -fPIC into a
> shared library, on x86_64 you cannot do that.
> 
> On x86_64 -Duseshrplib does the right thing, but I suspect that
> on 32-bit x86 it might make the build slower and/or bigger.

I don't particularly care about build-time or build-size if it enables
us to build on x86 and x86_64 with the same instructions and have the
same behaviour.

Regards,

Matt.




More information about the lfs-dev mailing list