[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: amd64 and sarge

> > On Thu, Jul 29, 2004 at 12:34:33PM -0400, Stephen Frost wrote:
> > > That's what pure64 *is*.  The number of people 'upgrading' from 32bit
> > > systems is probably around 1 (that being you), the rest of us have moved
> > > on to pure64 already, and did so a long ass time ago.

> * Raul Miller (moth@debian.org) wrote:
> > If this logic were correct, no one would need to install amd64 in
> > the future.

On Thu, Jul 29, 2004 at 02:13:00PM -0400, Stephen Frost wrote:
> This statement just doesn't make any sense.

Alternatively: you're predicting the future, not providing any logic.

> > Maybe you can't imagine that people would replace a motherboard and keep
> > the same hard drive, but that doesn't mean nobody works that way.
> People might do that, fewer people do that today than did it in the
> past but even so, people are likely to expect to need to reinstall in
> that case.  Certainly if they do their research and they want to take
> advantage of the new motherboard/processor they've got they're likely to
> *want* to do a new install.

I agree about the expectations part -- everyone distribution besides
Debian insists on a reinstall for most every upgrade, so people tend
not to expect any better than that.

That doesn't make it a good thing.

> > > This is just blatently false.  There certainly is gain in making every
> > > package supported on both architectures.  It gives our users *options*.
> > > For the amd64 side, it allows programs (*all* of them) to use more than 
> > > 2G of memory if they have a need to, it makes *most* of them run faster 
> > > and more effeciently.  We need the i386 stuff anyway since there are
> > > i386-only systems out there today.  Perhaps some day we will be able to
> > > remove i386, but I don't expect that to happen anytime soon.
> > 
> > You're talking about optimization.  If you're really concerned about
> > optimization, you'd be talking about building and installing packages
> > from source.  That offers far more in the way of choices and tailoring.
> Uh, being able to access > 2G of memory isn't what I'd consider an
> 'optimization'.


You were talking about the advantage of installing i386 packages on an
amd64 system.  Those i386 packages won't be able to access > 2G memory.

> Being able to access many more registers is closer to
> an optimization but it can be done for all cases, and will work on all
> amd64 platforms, and is much more likely to speed things up than slow
> them down.  Regardless, however, I want the programs I install to be
> fast, effecient, and in some cases they'll need to address > 2G of
> memory.  That doesn't mean I want to compile them all myself, I sure as
> hell don't, that's why I use Debian, so I don't have to.

I thought your point was that biarch was half-assed.  I don't see how
what you're saying here has anything to do with your point, whatsoever.

Did you forget what you were talking about, or is there really some
relevant point here?

> > My biarch proposal doesn't address how to make sure amd64 packages don't
> > replace an i386 packages across upgrades of those packages, but that's
> > because I don't care about that issue -- not because it can't be done.
> Uh, I'm not interested in making sure amd64 packages don't replace i386
> packages.  I've really got nfc where theis comment came from.

You seemed to have raised the issue of how biarch was half assed because
people might want to install i386 packages on an amd64 system even though
amd64 packages were available.  Maybe I misunderstood you?

> > > The current pure64 port has gone far beyond the half-ass biarch you're 
> > > referring to.  Unfortunately, you can't manage to see that.
> > 
> > False.
> > 
> > First, in a very literal sense, the pure64 port is incorporated in the
> > biarch I'm referring to.
> From what I saw, maybe a few bits and pieces of it here and there.

It's based on pure64 with simple changes to only a very small number of
packages (gcc toolchain, dpkg tools, and maybe a handful of others).

> > Second, the changes I've proposed have obviously not been incorporated
> > into pure64.
> That's because they're not necessary or useful.

To the current amd64 porting team, sure.  That doesn't mean that the
proposal is short-sighted.

> > More fundamentally, "half-assed" is purely pejorative, and most of what
> > you're saying is more about belittling than conveying useful information.
> It's half-assed because it's addressing only perhaps 1% of the packages
> in Debian.  This is so that you can claim how 'easy' it is while making
> the assumption that no one will care that they're running i386 on their
> Opteron.

I don't know why you're saying this.

Yeah, the proposal would only touch a small number of packages, and
the volume of packages it's relevant to is small, but..  after a person
upgraded an i386 system to the proposed biarch, they'd be using amd64
code for everything which has been ported.

My guess is that you're objecting to something unrelated.


Reply to: