Re: Are we losing users to Gentoo?
Chris Lawrence <lawrencc@debian.org> writes:
> > We'd need perhaps three different m68k varieties (two more than now),
> > one more Sparc, one more alpha, no more powerpc IIUC, no more arm, one
> > more mips, one more HPPA (or two?), no more ia64 or s390. So that's
> > nine more varities of 386 to consider, and maybe six for the other
> > architectures. So the total would be fifteen more copies of every
> > package, and it's 6GB per copy, so the total storage requirement is
> > about 90GB.
>
> Bah, don't forget that you'd want multiple PPC (603e, 604e, G3, G4,
> 790...) You could get away with 3 m68k (68020+MMU/030, 040, 060),
> although you might want more (040 w soft-FPU emulation for the LC040).
So I didn't UC for ppc. I said three different 68k, which is really
right.
> Of course, this would also entail separate CD images for each possible
> permutation. ("What do you mean, I have to install these crizappy
> i386 packages and then upgrade; I want ones optimized for my six-way
> Athlon MP setup out of the box! And don't give me Athlon XP packages,
> it's *just not the same*; the timings on this one instruction give a
> .0002% speedup, which dang-it I NEED!")
But the cost of CD images is even smaller, and there is no extra
bandwidth, since each person still only downloads the same.
> Not to mention build daemons for each possible permutation; some could
> conceivably be hosted on existing boxes, but they'd need extra RAM or
> disk space to keep up (and more people to keep an eye on them).
No, this isn't necessary. GNU tools (which we use) make it really
easy to set up cross-compilation environments. The only wrinkle is
being able to execute binaries midway through building, which is not
supposed to be needed for GNU packages, but others require that. This
is no trouble, however, provided you build on a machine that can
execute all the instructions you are compiling to (that is, normally
just whatever is the highest end target).
> Oh, by the way, hurd-i386 and the *BSDs will want all of these
> optimizations too. Better double your estimate :-)
And yet, it's still amazingly cheap.
The habits (and I have them too) of thinking that disk space is costly
are really old habits that it's time to break.
Reply to: