Brian Gupta wrote:
> Actually it depends on home you define base OS.
EXACTLY! (well, assuming home -> how)
We seem to be each moving forward with a different concept.
In bullet 5, three levels were asserted (my paraphrasing):
1) core OS
2) other Sun supported
3) other
(Interesting that we are discussing "Sun supported" in an OpenSolaris
forum. Seems like a choice entirely up to Sun. Let's just ignore this and
move on.)
Under an OpenSolaris framework, it is a syntax error to assume Sun
Solaris is the only distro or only significant distro, I recall seeing mail
asserting that this was the case today. That doesn't mean we should
design systems around that temporal situation. The playing field must
be level for OpenSolaris to succeed. It doesn't matter if anybody actually
takes to the field.
In a world of many distros, each distro gets to decide what objects
they ship (and perhaps support). If they don't, what's the point of
multiple distros?
So, I have an entirely different view of categories.
1) distro delivered
2) other
With the line between the two categories being different for each
distro. (Ubuntu has kinda clouded the line of "delivered", but
their various "universes" provide the same conceptual separation.)
I would like to see a world (spoken in the tone of the GEICO
commercial) where maintainers provided packages in the form
suitable for 2), and its up to the distro (perhaps aided by an
OpenSolaris project) to convert them to the form appropriate
for 1). There seems to be some skeptisism that the maintainers
are willing to do this, and the form 2) packages could be provided
by an OpenSolaris project. That is a process detail. 8^)
OpenSolaris can't provide the packages in form 1) as they are
potentially different for each distro. A couple of examples from
the wonderful world of Linux distros:
Red Hat EL 5 Ubuntu/Debian
/bin/rpm /usr/bin/rpm kinda makes
sense if you think about it.
/bin/sort /usr/bin/sort got me as to
why, seems like a failure in the making.
The only point is that Linux distros do relocate objects. (Maybe
this isn't a good thing -> separate discussion. But I could see a
Solaris based distribution which conformed to the FSH part of
the LSB - Sun's not going to do this for compatibility reasons.)
Of course, this breaks down at some level. The kernel, libc, grub
and many other things just need to be certain places.
So, ignoring what Sun should do (which seems to be deeply
embedded in Brian's goals), what should OpenSolaris strive
for?
My "back of the cocktail napkin" proposal is that this OpenSolaris
team divide the packages/objects into two categories:
1) If delivered, must be delivered by the distro (core).
2) Everything else.
2) is much, much larger than 1)
OpenSolaris should strive to provide a repository of packages
for everything in category 2) which does not overwrite the
distro provided objects. (This may be blastwave.) These packages
should be kept as current (to the maintainer's version) as
possible/practical. It is a choice of each distro as to when to
refresh their delivered version from these references - another
area where distros may desire to differentiate themselves.
I know this doesn't come close to what I interpret Brian's
goals to be, but his desires seem to be for Sun to make certain
business decisions. That's fine, and it would be embodied in
as to which of the grand blob of packages, Sun (the distro)
decides to deliver and support. Its even appropriate for this
OpenSolaris community to lobby Sun for certain business
choices (and some of the right people are on the line), but
OpenSolaris should not design projects assuming a certain
decision.
I know there are several levels of more detailed discussion
that need to happen. I understand that, but a clear and common
understanding of the software categories must be the first
goal.
- jek3