Re: cvs commit to glibc-package/debian/patches by gotom
>
> > LD_LIBRARY_PATH, and if /etc/ld.so.hwcap is not existed, then hwcap is
> > disabled. The hwcap pathes are not included to search.
>
> Why does that happen? By the way, looking at the path I wonder if it
> has a typo in it. It looks like it's disabling hwcap if the file does
> NOT exist, instead of if it does exist. That would be a bug.
IIRC, the file is actually ld.so.nohwcap
> > BTW, why is this patch so important? Can we upgrade smoothly without
> > this patch? Are there another way to be able to upgrade without it?
> > Please enlighten me...
>
> If hwcap is enabled when the version of libc6-i686 on the system
> doesn't match the version of libc6, very bad things can happen. We end
> up using the linker (/lib/ld-linux.so.2) with the optimized library
> (/lib/i686/libc.so.6) and they may not work together.
>
> At least I think that's what happened.
That's exactly the problem. And since messing with the linker is just
plain bad (IOW, we should not provide an alternative with symlinks and
such), it is the only method to handle such a case. There's currently no
mechanism to spot-disable hwcap usage, and IMO, that's an upstream bug
anyway. An environment variable may be a better idea, like
LD_DISABLE_HWCAP or something.
--
Debian - http://www.debian.org/
Linux 1394 - http://www.linux1394.org/
Subversion - http://subversion.tigris.org/
Deqo - http://www.deqo.com/
Reply to: