[x86-64] AMD Dev Center usage proposal

[marc.miller@amd.com]

Kernels:

I agree on focusing on 2.4.x as the near term priority, but as a lot of the AMD64-specific code was integrated into 2.5, that branch is worth playing with, especially if you run into any problems with 2.4.x and need to backport any piece of 2.5.

Architecture name:

Just about everything now is written to understand x86_64 as the architecture name, and AMD is encouraging amd64 to be an alias to x86_64.  For package names, it's really up to you... thing is I don't know what alien will call a .deb if it is processing something ending in .x86_64.rpm.  

Something else you should consider is that there will be cases where a package can depend on either 32-bit or 64-bit versions of the package (e.g. when a perl interpreter is needed to run a script, it can be 32-bit or 64-bit, but libraries must match the architecture of the executable).  Many other vendors have chosen to make all 64-bit packages depend on 64-bit packages, even when it's possible to substitute 32-bit packages.  For your initial build, I suggest you do the same for simplicity.  



-----Original Message-----
From: Arnd Bergmann [mailto:arnd@arndb.de]
Sent: Friday, May 23, 2003 2:11 PM
To: debian-x86-64-devel@lists.alioth.debian.org
Subject: Re: [x86-64] AMD Dev Center usage proposal


-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

On Friday 23 May 2003 21:08, Falk Hueffner wrote:

> I think we should concentrate on one kernel first. Assuming it is not
> totally broken currently, I'd prefer 2.4, since it is less of a moving
> target.

Agreed. 

> > 5) port init script tools
> >   - ability to get to a prompt with only 64-bit binaries
> >   - sysvinit, util-linux, sh, grep, etc.
> >
> > 6) port remainder of the 'base' packages
> >   - this will be the majority of the effort (IIRC 50+ packages)
>
> If we manage to install a build daemon, I suppose this won't be
> actually all that much effort. I think we should try that right after
> step 4.

The autobuilder won't be much use unless we have ported a significant
amount of libraries by hand. I have a set of changes to dpkg and
debhelper that make the migration to /lib64 easier, but many libraries
still have to be fixed beyond what can be done automatically.
IMHO, 5) should actually be 'port all base libraries'.

> Other things to do:
>
> * Choose a name for the port. I vote for amd64, since it avoids the
>   nasty - vs _ problems and is also more likely to be recognized by
>   people in the future. I suppose "x86-64" is going the way of
>   "axp"...
Agreed.

> * Hack dpkg to understand the new architecture.
Yes. For the moment, we can keep building all packages as i386 .debs,
but later, dpkg needs to know that amd64 is a superset of i386. That
will require bigger changes than just adding another identifier
to some lists.

	Arnd <><
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.2.1 (GNU/Linux)

iD8DBQE+zo5f5t5GS2LDRf4RAjpGAJ9r4xcxVq91RWdy+mAVkWPBgcugWACdFKRf
KY3p5eRKj4e2WlT6GXN5iQ8=
=ikqL
-----END PGP SIGNATURE-----


_______________________________________________
Debian-x86-64-devel mailing list
Debian-x86-64-devel@lists.alioth.debian.org
http://lists.alioth.debian.org/mailman/listinfo/debian-x86-64-devel