[1085] in SIPB_Linux_Development
Re: ELF migration
daemon@ATHENA.MIT.EDU (Erik Nygren)
Fri Oct 6 13:14:29 1995
To: jered@MIT.EDU
Cc: linux-dev@MIT.EDU, linux-afs-bugs@MIT.EDU
In-Reply-To: Your message of "Fri, 06 Oct 1995 11:01:18 EDT."
<199510061501.LAA28818@vorlon.mit.edu>
Date: Fri, 06 Oct 1995 13:13:43 -0400
From: Erik Nygren <nygren@MIT.EDU>
> It's all well and good that we agree that a script should be written
> to help people migrate to ELF...but we knew that already. The more
> pressing issue is how to install ELF binaries on Athena, and I have
> heard no comments on that. Would people please give their opinions of
> i386_linuxelf1 (or i386_linux1elf, or i385_linux1e)?
First of all, i385_* would be a bad idea... :-)
Seriously, I think it would be far too confusing to have this mix. As
was pointed out earlier, Linux already has the longest @sys name of
any platform so we should shorten it if anything.
I think the best solution is to just mix the binaries in the same
bindir. If people want to run elf binaries then they just install the
elf library upgrade package. (Under the condition that they don't
have to change their existing libraries or compiler, just ld.so). We
then start moving applications over slowly, thus forcing people to run
mixed elf/a.out systems for awhile. The switchover should be as
painless as possible with the exception of forcing people to install
the elf libraries. If people have to start compiling every binary for
elf and a.out or if we have to get locker maintainers to deal with
this then we'll really lose and people will have a very good reason to
switch over to NetBSD.
On a related topic is the issue of Linux-AFS 1.3.x support. Derek
mentioned that he'd need to switch the @sys value to do this because
of a recent kernel change. Would it be possible to keep the value the
same and just distribute two linux-afs packages with conditional
compilation in the source tree? This will come down to being Derek's
decision since he's doing this all on his free time but it might be
appropriate to discuss it here? I don't think it would be a good idea
to change the @sys value for any reason unless it's ABSOLUTLEY
necessary because of major, unavoidable binary incompatability. The
primary reason in my mind is that we don't have control of many
of the lockers that linux binaries are in (such as course lockers)
and we have the symlinks or binaries out of the goodness of the
maintainers and it may get difficult/annoying for them to have
to install more symlinks than are already necessary for Linux support.
--- Erik
