On Thu 14 Feb 2008 at 04:13PM, Danek Duvall wrote:
> On Thu, Feb 14, 2008 at 03:56:19PM -0800, Philip Brown wrote:
> 
> > blastwave does not currently have a twiki package. This comes to mind, 
> > because of my recently having to deal with a similar software installation, 
> > by hand, and thinking "wouldnt it be nice if I didnt have to do this by 
> > hand, but I just had a package to do it for me".
> 
> Do you really want the package to do it for you, or do you just want the
> thing to work by the time you get to start using the software?
> 
> If it's the latter, then why does it matter if the package does the work or
> if the software does it when it realizes that it needs to be done (either
> by being told, by looking for itself, or by starting up for the first time
> after install/upgrade)?
> 
> But certainly, the less the end-user has to do for the software to just
> work, the better.

I think there's also a real distinction between "the software on
disk" and "the software instance".  /opt/webserver might exist
and contain the binaries for 10 or 20 webserver instances.

> > I will simply say, that the more popular linux distributions, take the
> > opposite view: that the purpose of "the package", is to install,
> > configure, and make functional, the desired software, in a completely
> > automated fashion, as much as possible.
> >
> > THAT, is what makes linux popular!!
> 
> There are dozens of reasons that linux is popular.  That software delivery
> is conflated with software configuration, if it is one at all, is, I'm
> pretty sure, low on the list.

UNIX has always lacked an abstraction for "I want to add another
instance of <thing>".  I'm unconvinced that reusing the package
for that is the right path; perhaps the packaging system should
interact an "factory" subsystem to create a default instance.

I've not yet seen a compelling argument for creating a monolith of
packaging with instance creation.  Packaging should be as dumb and
boring as possible.

> > Ironically, i seem to recall that, just a few months ago, someone on
> > this very list was complaining about "the evils of layering", and
> > how tight integration was far preferable (the ZFS architecture
> > example). Seems to me, that attitude would mandate  "tight
> > integration" of things that packagers need to do, into "the
> > packaging system".

"someone" == me.  You have on two occasions mischaracterized my
position.  As I have said repeatedly, the point is that refactoring and
reconsidering the constraints of a problem can yield unique value by
arriving at new abstractions and new layerings.  As a side point, I
think we have two many layers and too many abstractions today; that's a
similar but distinct issue.

For those wishing to review my thoughts on this matter, I summarized
them in a blog entry: http://blogs.sun.com/dp/entry/a_big_mess.

Others (not me) have asserted that ZFS is a monolith.  Anyone who has
studied its design knows that this is false, and that it has a clean
and elegantly factored set of layers.

        -dp

-- 
Daniel Price - Solaris Kernel Engineering - [EMAIL PROTECTED] - blogs.sun.com/dp
_______________________________________________
pkg-discuss mailing list
[email protected]
http://mail.opensolaris.org/mailman/listinfo/pkg-discuss

Reply via email to