High-resolution timer issues

[Andy Dougherty]

On Thu, 10 Sep 2009, Christoph Otto wrote:

> James E Keenan wrote:
> > config/gen/platform/generic/hires_timer.c and
> > config/gen/platform/win32/hires_timer.c were committed to trunk on Sep 06 in
> > r41081 when the 'pluggable_runcore' branch was merged to trunk.
> > config/gen/platform/darwin/hires_timer.c was added a few hours later  in
> > r41087 in an attempt to cope with Darwin-specific problems. More and more I
> > think that there are problems with this implementation.
> > 
> > See the following two Trac tickets:
> > 
> > https://trac.parrot.org/parrot/ticket/985#comment:3 and following comments.
> > Andy Dougherty cites a number of problems or limitations of the
> > implementation.
> > 
> > https://trac.parrot.org/parrot/ticket/989#comment:5, where I report SIGNAL
> > 10 (SIGBUS) errors on Darwin/PPC subsequent to the introduction of
> > darwin/hires_timer.c.  I thought they had cleared up, but then I noticed
> > other Smolder reports with the same problem.  While these errors are
> > intermittent, a look at our Smolder reports for Darwin suggest that they are
> > happening more often than not.  I don't have any expertise in either
> > high-resolution timing or diagnosis of bus errors, so I've done as much as I
> > can on this ticket.
> > 
> > "Stability!" is supposed to be our theme of the week, especially leading
> > into Tuesday's release.  But this is clearly an area of instability. I'm
> > hoping that posting to the list because I fear that if these problems are
> > not treated by Tuesday they will then simply get swept under the rug in the
> > rush to merge in the next big branch.
> > 
> > Thank you very much.
> > kid51
> 
> 
> I'm reasonably sure the merge is a red herring.  I agree that the Configure
> system should be used to figure out the optimal high-resolution timer
> functions (and it's on my todo list), but it shouldn't have any effect on the
> results from any platform.  The functions in that file are only used by the
> profiling runcore, and that runcore is not automatically tested anywhere yet
> (also on my todo list).

I'd agree about the hires_timer.c files, but the merge touched way more 
files than that.  It's difficult to tease more information out of Smolder, 
so it's hard to be more specific without further testing on that platform.

On the broader question of "Stability", I'd say kid51's remarks are spot 
on.
 
-- 
    Andy Dougherty		doughera at lafayette.edu