Le Mercredi 19 Mars 2003 21:08, Gwenole Beauchesne a �crit :
> Hi,
>
> > That works only for libraries linked at compile time however, while
> > symlinks
> > works also for binaries, and for dynamically loaded libraries.
>
> Why people would dlopen() a library themself if the primary goal was
> supposed to get better performance? Again, the process I described is
> the way to go. Packages doing otherwise are to be taught the right way.
Automatic runtime discovery maybe ? I don't have a clue, i'm not a developper.
BTW, perf-junkies doesn't even use dynamic libraries, they rely on statical
binding. Just see mplayer for instance. And you don't adress binaries anyway.
> > Moreover, there are two different issues there: a packaging issue
> > (should
> > optimisable package contains optimized code) and a system issue (how to
> > install this code to make it used).
>
> The former issue is maintainer's. i.e. he should check whether it worth
> it and the actual gain doing so.
But we should make it a policy how to support it whereas he decide gain is
interesting:
- allows conditional build in spec to have either optimised or non-optimised
build, leading to distinct binary packages
- build everything twice, and release both optimised and non-optimised
packages
- any other solution
Moreover, how should those package get distinguished from standard i586
packages ?
--
No matter how many resources you have, it is never enough.
-- Murphy's Computer Laws n�1
