[Insight-developers] RE: SGI stdlib still needed?

[Miller, James V (Research)]

Brad,

I thought at one point you were hacking up the STL allocator code to make
this work.  Is that not the case anymore?

If the following is true:

1) All of Insight (as well as your test program) run properly on the SGI
using the standard SGI distribution of STL/streams without defining _PTHREADS.
The assumption here is that the SGI standard implementation of the STL/streams
has its own fast locking mechanism.

2) All other platforms (except the SUN) work properly with out the _PTHREADS
being set. That includes Linux, Mac, GCC on SGI, etc. then we can run without
the _PTHREADS defined.

3) Solaris runs fine with the _PTHREADS

Then it looks like we could avoid defining _PTHREADS on everything except
on Solaris.  I have two concerns:

A) The original problem arose when running on a multi-processor Solaris 
machine with 12 processors.  Will the problem reappear when Linux boxes 
start to have a large number of processors?

B) What if someone is using ITK as a 3rd party library to their application
and they build their application with _PTHREADS defined?

Perhaps the safest thing is to define _PTHREADS on every (unix) platform
that is not SGI.

I agree with Bill Hoffman that we should remove the stdlib distribution for
the SGI as soon as we are all comfortable that everything is working. I just
don't see that happening before the 1.1 tag.  We need the 1.1 tag to be
ready for the course at SPIE (which is in a week or so).






> -----Original Message-----
> From: Brad King [mailto:brad.king@kitware.com]
> Sent: Friday, January 31, 2003 4:06 PM
> To: Miller, James V (Research)
> Cc: Insight Developers
> Subject: SGI stdlib still needed?
> 
> 
> Jim,
> 
> I think the latest we knew about the SGI/STL/streams problem was that
> leaving _PTHREADS undefined works everywhere except sun.  We should be
> able to remove the stdlib library from ITK now by just 
> defining _PTHREADS
> on the sun only.
> 
> Thoughts?
> -Brad
>