[geeks] Solaris 9 changes

[Scott Howard]

On Tue, Jul 27, 2004 at 10:46:55PM -0500, Bill Bradford wrote:
> Solaris 9 9/04 is out, along with a list of changes:
> 
> http://www.sun.com/bigadmin/content/eof.html
> 
> Some of these were listed in earlier end-of-software-support lists.
> 
> Notable things removed as of this release:

NO! We don't remove things in update releases.  All of these things
disappeared in the original Solaris 9, not in the releases after that.

Update releases will NEVER remove something, and as a rule will NEVER
change the default behavour of something - although that one can be
gotten around with a lot of work. One example is that 9/04 has UFS
logging on by default where previous releases didn't.

> Current support that may be removed in a future release:
> - fork() (?)

This needs clarification as it's very badly put in the release notes.
The fork() system call, fairly obviously, isn't going away. What is
changing is it's behaviour when called in multi-threaded program.
Historically Solaris has had two fork calls - fork() and fork1(). The
difference between the two is their behaviour in threaded programs as to
which threads get forked - all of them or just the current thread.

Depending on how your program was linked, what libs it dynamically links
on and a few other factors which probably included the day of the week
fork() can be changed on the fly to act like fork1(), even if that's not
what you want. This is, at best messy.

So what they have done is to remove the current fork() completely, alias
fork() to fork1() (so it will only fork the current thread), and added a
new forkall() which forks all thread.

The end result to users should be completely transparent...

> I'm surprised that sun4m support has lasted this long, as well as 32-bit
> UltraSPARC. 

As a rule they won't drop support for a product until after the hardware
itself has passed EOSL (End of Support Life). In the case of sun4m this
didn't happen until September 2003 (SS5 was last shipped 9/98, and
supported for 5 years after that).
For UltraSPARC I it's a little more tricky. The Ultra-1 is long gone
(EOSL was July 2003), however there are still E3000's and Ultra 2's out
there that are not yet EOSL (and will not be at the time Sol10 ships)
which have US-1's in them - although they are few and far between.

At the end of the day I think they just decided that it was easier to
drop support for these few chips which haven't in themselves been sold
for many years and thus be able to drop all 32 bit (on SPARC) kernel
support.

> a bad taste in my mouth.  I heard that NIS+ was implemented for one or two
> big customers, and Sun themselves continued/continues to use NIS.

A mixture of NIS, NIS+, LDAP and even a few other things, depending on
the area/country/etc. The problem with NIS+ is that most people expected
it to be as easy to configure/maintain as NIS (after all, it's just NIS,
plus!) where it isn't...

  Scott.