On Thu, May 15, 2008 at 09:44:35AM -0700, Stephen Hahn wrote:
> * Venky <[EMAIL PROTECTED]> [2008-05-15 04:36]:
> > On Wed, May 14, 2008 at 09:55:02AM -0700, Stephen Hahn wrote:
> > > > Let me try and re-state my point in an unambiguous manner -- if
> > > > a project is open source, the only format relevant to a distribution
> > > > maintainer would be source code.  If there is an exception to this
> > > > rule, I am not aware of it.
> > > 
> > >   (It's somewhat generous to call a tautology a rule.)  
> > 
> > Not really.  There are reasons why distribution maintainers prefer
> > building from source, and that is to ensure quality and also because
> > each distribution's requirements differ.  We do that with the SFW
> > consolidation.  The default install prefix of most GNU applications
> > is /usr/local and we switch that to /usr or /usr/gnu.
>  
>   I was being polite.  Let me go further and state that a solution that
>   would somehow prefer software for which the source code was available is
>   certainly incomplete or probably also incorrect.  

That's totally different from what I was trying to say.  I know of
no package manager which cannot handle closed source applications
right now (including Gentoo's predominantly source-based portage
system), so that would not even be a tautology, let alone a rule.

My point was it *if* there is source code available for a package,
that will be the preferred medium from a distribution maintainer's
perspective.  A package manager which prefers software for which
source code is available is incomplete, true.  But every package
manager around prefers building its own packages from source code to
using pre-built binaries, given a choice.

> > >   As has been
> > >   stated before, pkg(5) does not contain a build system by design--as a
> > >   result, the practical problem of absorbing components from package
> > >   publishers using both legacy packages and proto areas has been very
> > >   straightforward.
> > 
> > True, though I'd suspect a lot of it is because we have been
> > depending on the standardized build procedures currently used to
> > build the SVR4 packages.  If we had to build from source all the
> > packages we distribute currently via pkg.opensolaris.org, the lack
> > of a standard build system would be felt more acutely.
> 
>   If you believe that we could convince all of the current System V
>   package providers to just switch over directly, that's... interesting.

Again, totally different from what I was trying to say.  I do not
believe that current System V package providers need to switch to
anything else.  In fact, my point was that I like the status quo,
and the fact that we have (mostly) standardized build procedures for
producing System V packages, and that I believe the transition from
SVR4 to IPS has been smooth precisely because we have System
V package providers with standardized, repeatable build procedures.

Disparate build systems delivering packages without support from
pkg(5) work right now because they are all coming from within Sun
and we can be reasonably assured of the quality.  That is the reason
why we are not forced to re-build the packages ourselves.

Assuming we plan to start allowing people outside Sun to publish
packages eventually, we will need some quality control measures in
place.  And this will, IMO, boil down to having trusted package
maintainers who will (assuming the source is available) build the
packages and upload them.  These maintainers will have to set up
build environments specific to each package -- which is fine --
somebody has to do it.  The problem is that anybody who wants to
replicate this package build will have to go through the process of
figuring out the build environment all over again because IPS does
not have a way currently of recording the steps required to setup
and run the build.  (pkgbuild-pkg(5) would go a long way towards
solving this, by the way.  I agree that there is no real reason why
build *has* to be part of the packaging system.)

Having a package manager support build systems does not mean it will
dictate the build procedure to the package.  I know of no package
manager that does that.  The packages continue being built the way
they were designed by the developer -- be it autotools, ant, imake,
whatever.  All that the package manager does is make it easy for
someone else to replicate the build by recording these steps.

The primary consumer of a packaging system (not counting end users
who use it to install packages) are distribution maintainers and not
the developers.  And IPS in its current state is more
developer-friendly than maintainer-friendly.

>   Of course, since we know that we don't have unencumbered access to
>   source for many of them, the scenario is irrelevant.

True, and I expect any package management system to be able to
handle closed source packages too, so that was not something I was
contesting at any point of time.

>   I will let you have the last word, of course.

:)  Anyway, this thread is rapidly approaching the point where it
stops being of any value whatsoever, so I guess I'll shut up now.

Venky.

-- 
One hundred thousand lemmings can't be wrong.
_______________________________________________
pkg-discuss mailing list
[email protected]
http://mail.opensolaris.org/mailman/listinfo/pkg-discuss

Reply via email to