[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: [Freestandards-ldps] Re: LDPS 0.91 - even closer to release

On Thu, Jul 27, 2000 at 07:13:45AM -0700, Theodore Ts'o wrote (in part):
> I've always pushed for Posix threads because I figured that ISV's wanted
> it, since they had programmers already trained in using the Posix
> threading model (however broken it might be), and because they have
> already existing code which uses Posix threads (again, no matter how
> broken the standard might be, if you already have code using it, you
> don't want to have to pry it open and change it).  

You're right that POSIX threading is used by ISVs (as least the one
I work for) for the reasons you mention: it's a flawed standard, but
widely supported, and we have a large body of code written to it.

> The problem is that we haven't been able to get any of these ISV's to
> step forward, so people doubt whether they actually exist.  (In fact,
> I'm beginning to have my doubts; maybe I was wrong.)  

We've not needed to step forward to complain about the current state of
affairs or to volunteer to modify the current implementation because
we're using it as-is with good success.  A 100% compliant POSIX
thread implementation is certainly not the most important Linux goal,
and it is my impression that significant architectural changes would
been needed in the kernel and library support to get there.  If the
choice is to compromise the design of Linux or of pthreads to get a
good standards/performance balance, the decision is easy.  Besides,
are any ISVs actually saying "we would write for the Linux platform if
only the POSIX threading were fully compliant?"

Bottom line IMHO - all *nixes are "broken" from the point of view of
product porting in some way and Linux is better than most.  If porting
were too easy we ISV folks would be out of our jobs and have to spend
all our time writing free software ;-).  The Linux implementation of
pthreading is serviceable in its current state.  Caveats about relying
on its current warts are reasonable for the LDPS.  Furture improvements
would be welcome, but we have to use what is in the field now.  The coding
changes needed to handle stuff like the unusual signal semantics are
minimal anyway.

- MikeT

Michael E. Touloumtzis		mailto:toumi01@ca.com	-or-
Ingres Product Development	mailto:Michael.Touloumtzis@ca.com
Computer Associates International

Reply to: