[Nottingham] Debian devotion [was: OE Reply Fixer]

[Robert Davies]

On Wednesday 05 March 2003 08:17, you wrote:
> On Wed, Mar 05, 2003 at 07:54:51AM +0000, Robert Davies wrote:
> > For dialup, surely downloading source patches and applying then
> > re-compiling, linking and installing automatically, will be the
> > fastest way.
>
> With 1meg CM or whatever you've got then grabbing source patches will
> indeed be quick however you then have to patch and compile.  I know
> people who run gentoo and all their computers seem to *do* is be
> compiling something.
>
> Depends what you want to do really.

Well my Debian dialup, all it seemed to do was download something, if you do 
not have flat rate package it is a royal pain.

I'm going to try Gentoo out, once it's installed then have a working KDE and 
X environment, then I'd try to keep that stable, as indeed source changes 
there will result in a lot of re-compilation.

There may be a way to keep objects about, and let make sort out to re-compile 
only the bits it needs.

All depends on what trade offs you want, if you have a lot of CPU and 
diskspace, but poor bandwidth then using source makes sense.  Even the patch 
rpm's which are available now for SuSE, on something like kernel result in 
12MB downloads compared to 15 or so.  The source patches are rarely bigger 
than 100KB, and often trivially small.

> Is the i386 vs optimisations really that big a step?  Are there any
> decent benchmarks on it in normal use for instance?

It's become more important with P4 and Athlon, gcc-3.2 is getting cleverer 
to.  The P4 has some annoying braindamage, which Intel excepts Software 
Developers to work round, according to their optimisation guides.

In past my experience was it wrung out 5-10% for architecture, with using -O3 
sensibly to, gave up to 30% increased performance on CPU intensive 
applications, so gzip & bzip, crypto libraries, ssh, glibc and gcc itsell 
would qualify for special attention.  There's some news out about someone 
re-compiling gcc itself with optomisation (I always used to do it on Sun 
anyway) and finding a 20% increase in compile speed.

Rob