I don't think that's enough; I think point 4 has to be abandoned altogether.  
Here's why:

* some OSS libs don't bother much about backwards compatibility at the binary 
level

* the libs that _are_ part of supported [Open]Solaris need to be stable more 
than they
  need to be the latest and greatest.  In particular, they need to satisfy the 
requirements
  of other Sun supported software (whether software that Sun controls or leads 
the
  development of, or whether software that Sun supports but isn't the lead on) 
at the
  versions corresponding to any given Solaris release.  That is, while stuff 
like JDS
  should ideally not fall incredibly behind, it probably shouldn't be on the 
bleeding edge

* however, a lot of OSS that there may be demand for as part of the companion 
software
  tends to have dependencies on the latest libs as of when it was being 
developed

It sure would be nice to push out some discipline back to the various OSS 
projects,
so as to keep the conflicting dependency requirements from exploding; and 
perhaps
that can be done in cases where Sun or other OpenSolaris participants have some 
influence.
But that probably won't cover everything.  So I don't see how duplicate libs 
can be
avoided.

I think in this regard, Blastwave may be on the right track, for the most part. 
 They
worry mostly about their own consistency, and depend on relatively few Solaris
libs, generally stuff that doesn't duplicate non-OpenSolaris OSS libs; that
decouples their update cycle from that of Solaris.

Now I am inclined to think that there is some validity to point 3, in which 
case there
would likely be new versions of at least some libs and commands for the current
version of Solaris, rather than Blastwave's approach of using libs compiled on 
the
oldest supported version of Solaris as much as possible, with newer ones for 
newer
versions only when really necessary.  I don't think that necessarily needs to 
be either/or;
probably some libs are stable enough that there wouldn't be any reason to 
rebuild them
for each new version of Solaris.

To have all the above, be reasonably current (say a few weeks slower than 
Blastwave's
best turnover), and incorporate a perhaps higher level of quality control and 
testing
(so that non-building/testing users could have high confidence in what they 
downloaded),
will definitely require a lot of work, but also a highly effective, well 
managed, disciplined,
and yet distributed overall process.

The existing freeware site's package maintainers collectively have a lot of 
experience
building various software; and they have perspectives on dependency issues and
build procedures that (hopefully) satisfy their particular objectives.  But I 
think Sun
could contribute a lot in terms of process, leadership, quality control 
practices and
techniques, and so on.

What's the real need for point 4 (regarding duplication - granted that it's 
_desirable_
where it's an acceptable tradeoff, in that it simplifies troubleshooting, makes 
more
effective use of memory, probably saves some I/O, etc)?  What other potentially
mutually exclusive issues are there?
 
 
This message posted from opensolaris.org

Reply via email to