Re: [Boost-build] Patch execunix.c for Darwin PPC hangs

From: K. Noel Belcourt (kbelco_at_[hidden])
Date: 2008-02-09 00:43:24

Hi Beman,

I've committed the patch to execunix.c to fix the PPC darwin vfork  
problem in the trunk and I'm satisfied it's working okay.  I'd really  
like to get this one patch into 1.35 before it's released.  It is a  
bug fix, not a new capability.

Is there any way to get the patched execunix.c into the 1.35 release  
or is it too late?

-- Noel

On Feb 7, 2008, at 1:03 PM, K. Noel Belcourt wrote:

> Hi,
>
> I've been having occasional problems running parallel (-j2) darwin  
> trunk regression tests on ppc G5 hardware under Mac OS 10.4.x.  It  
> appears that a race condition, or other unsavory interaction,  
> randomly occurs between vfork and execvp.  I've tried using execve,  
> as documented in the vfork man page, but the race condition still  
> occurs.   Simply replacing vfork with fork fixes the occasional  
> hangs I've been seeing.  While I am concerned about the increased  
> overhead of fork over vfork, I believe a robust parallel bjam is  
> well worth the increase in overhead.
>
> Attached is a patch that replaces vfork with fork, only on ppc  
> darwin, along with a couple of minor patches.  I've used the  
> __APPLE__ and __ppc__ macros to select fork over vfork.  I'm open  
> to suggestions if there's a better set of macros to use.
>
> I've tested this patch to ensure that (1) it fixes the darwin ppc  
> trunk tests and (2) doesn't break any other unix trunk regression  
> tests (Aix, Sun, Intel Darwin, Linux flavors).  If there's no  
> objection, I'd like to commit this to the trunk.
>
> Thanks.
>
> -- Noel
>
>
> <execunix.c.diff>
>
> _______________________________________________
> Unsubscribe & other changes: http://lists.boost.org/mailman/ 
> listinfo.cgi/boost-build