[OpenAFS] Question: discontinuing support for Solaris < 2.6

Marcus Watts mdw@umich.edu
Tue, 16 Jan 2007 15:34:20 -0500


Russ Allbery <rra@stanford.edu> sent but probably didn't write:
> > I am pretty sure it doesnt kill a lot of messy ifdefs, except maybe some
> > of the threaded code where you can use pthread instead of sun thread
> > support. IIRC pthread wasn't fully implemented until solaris 8..
> > but maybe they backported it to the older versions. I don't really know.
...

I believe pthreads works fine in 2.6 -- I've heard people claim it
worked better there than in later versions.  (2.6 being the "last"
32-bit only system gave it a certain advantage in terms of stability...)
I don't know of any reason to avoid it.

The only solaris thread stuff I can find (in 1.5.12) is in
lwp.c -- conditioned by the manifest USE_SOLARIS_THREADS which
is apparently only turned on in the apparently defunct
src/lwp/NTMake9x .

I think you can pull all that old USE_SOLARIS_THREADS and not
have any problems.

There may be one other thing that matters - maybe.
In a copy of "openafs 1.2.10" for solaris 8 that I seem
to have just had sitting around, there's one solaris threads
call left -- "thr_yield()" calls.  rx_pthread.h defined
pthread_yield to be various things on different systems - this
appears to be one area where pthread isn't really a standard,
and thr_yield happened to be the winner on solaris.  This is
apparently still true today.  I think this can be chalked up
as a generic "known OS issue with pthreads", not a specific
problem with ancient solaris.

				-Marcus