4:59 p.m.
Hi everyone,
If you're not on cross-distro and are interested in this topic, please
do sign up to the list. I will share the dialin info here anyway.
Jon.
-------- Original Message --------
Subject: Re: Armhf dynamic linker path
Date: Tue, 10 Apr 2012 17:56:38 -0400
From: Jon Masters <jcm(a)redhat.com>
Organization: Red Hat, Inc.
To: Jeff Law <law(a)redhat.com>
CC: Mike Frysinger <vapier(a)gentoo.org>, cross-distro(a)lists.linaro.org,
Adam Conrad <adconrad(a)debian.org>, linaro-toolchain(a)lists.linaro.org
On 04/10/2012 12:38 AM, Jeff Law wrote:
On 04/09/2012 10:33 PM, Mike Frysinger wrote:
> his point is you can't install multiple architectures into the same root.
> alpha, arm oabi, and m32r for example have ldso set to /lib/ld-linux.so.2. a
> quick grep of GLIBC_DYNAMIC_LINKER in gcc's config/ tree shows other
> collisions.
Hmmm, why would anyone want to install distinct architectures in the
same root... I must admit I don't recall that being discussed at Plumbers.
So with multi-arch, you can do system emulation on the fly or whatever
else you feel like doing today. It's technically cute, but of no
interest to Fedora. However, here we're solely concerned with the path
to the linker, which we do incidentally need to fix (and the proposal
that we adopt the same one that happens to be used by multi-arch is
entirely reasonable on technical grounds).
Here's my viewpoint:
1). We do need a single linker path[0] so that someone can build a
binary on Ubuntu or Fedora and have it work on the other distro.
2). The linker path should ideally be identical (the one compiled into
the ELF header - whether that is the original path, or a symlink to a
symlink...) so that we don't get a situation wherein binaries built for
one distro won't run on the other, or the inverse. Selfishly, I care
that everyone uses Fedora to build stuff and then it "just works" on
Debian or Ubuntu (if they feel so inclined), not the other way around.
That's my motive right there to care that we all agree on this :)
3). There exists no technical reason to oppose using the multi-arch path
that we had discussed previously. I think we would be best served
clearly separating a single file (as Adam has note on many occasions)
from the bigger issue of multi-arch that is "not loved" everywhere. If
we can do that, I think we can then go to the step of advocating to do
this upstream, and pull it into Fedora's ARM port. I'd go as far as to
say that this is an ARM specific issue on the Fedora end, and thus we
don't really need the entire Fedora community to change one path.
4). Frankly, whether or not we like the Ubuntu approach - and I'm going
to be careful here so I'm not quoted out of context - they are shipping
a 5 year LTS release of Ubuntu on ARM. Since they are the first to do an
ARM release of that nature, they are setting a platform precedent in
their actions that those of us with 6 or 12 month cycles don't have. We
are in an easier position to get with the program than they are by now.
I hope we've all learned a lesson for AArch64, live and learn.
Anyway. My proposal is that we all join the call that Steve is putting
together on Thursday pm/eve, using my Red Hat bridge. He'll send out the
information. I would very much appreciate it if you (Jeff) could join
because you are widely known as a calming voice of reason and sanity (if
only you scaled to the point we could have one of you on every call). I
would also like it very much if anyone in the Fedora community who wants
to share their input could also dial into that meeting. Let's have this
out, get it finally nailed down, and move on.
Thanks everyone,
Jon.
[0] If we're even going to fool ourselves even remotely into believing
in cross-distribution binary compatibility. I'm a true believer in
having binaries that just work, but it's not the new shiny any more.
Everyone loves to hate LSB so much but they never try to improve it.