Re: RFC: pentium optimized debian
[seperate pentium architecture]
> I think that ultimately this will work out better. No matter what choice
> you make you -will- need another directory on the ftp site to store them,
> might as well call it binary-i686.
Either way we go, there would be a separate directory for the pentium
optimized packages. It's independent of how it's implemented.
> Remember that alot of other arch's have
> target specific binaries that make use of target specific
> instructions/optimizations (arm, alpha, sparc do at least)
I think this is best handled on a case by case basis. I'm really not
interested in coming up with a general solution that handles any type of
optimizations or other incompatabilities inside an architecture group.
> Optimization: Pentium handles the limited case of egcs 'weak' P5 optimizer
> that is backwards compatible with a i386
Backward compatable optimization is the only case I'm interested in
addressing myself.
> The dpkg problem I'm not sure. We can likely easially hack it to know that
> a list of arch's are valid. In truth on some systems target specific
> optimizations will NOT run on lesser processors. Dpkg should not install
> arch specific packages in these cases.
If you want a general solution, you need a table that lists which
arch's are valid for each other arch.
However, this requires that dpkg is able to detect, ie, that the host
machine has a pentium mmx processor installed on it and may use
(hypothetical) MMX optimized packages.
You also run into thorny problems if you, say, downgrade the processor in
your box from MMX to an older pentium after installing MMX optimized packages.
I'm not interested in worrying about such complexities.
> (I think P5 specific packages
> are silly, all packages should be P5 optimized and a handfull of crucial
> ones should be i386 optimized).
Pentium optimized packages tend to be 10% larger than i386 packages. A lot
of packages don't benefit greatly from optimization. These two facts make
this suggestion impractical.
I'm really not sure, but I'd guess if they insert NOPs for pipelining
purposes, they will also run slower on 3/486's, and I'm really not
interested in slowing down those machines for no gain.
> If you put both sources then apt will see a new version of the i386 binary
> before it is avail in P5 optimized form, it will install that and wipe off
> your P5 version. When the new i686 binary comes along it will appear to be
> the same version and will not be installed. That is why APT needs to know
> the exact arch/optimization status of all installed packages. Without that
> information any implementation is simply a hack.
Ok, fine. It sounds like if I added the Optimization: tag to status, apt
could be made smart enough to treat pentium optimized packages as newer
without requiring they be a seperate architecture.
--
see shy jo
--
To UNSUBSCRIBE, email to debian-devel-request@lists.debian.org
with a subject of "unsubscribe". Trouble? Contact listmaster@lists.debian.org
Reply to: