[gst-devel] Re: [gst-cvs] wingo gstreamer: gstreamer/ gstreamer/docs/design/ gstreamer/gst/ gstreamer/gst/base/

[Andy Wingo]

Hey Dave,

On Sun, 2005-10-02 at 18:59 -0700, David Schleef wrote:
> On Sun, Oct 02, 2005 at 11:57:19AM -0700, Andy Wingo wrote:
> > CVS Root:       /cvs/gstreamer
> > Module:         gstreamer
> > Changes by:     wingo
> > Date:           Sun Oct 02 2005  11:57:19 PDT
> > 
> > Log message:
> > 2005-10-02  Andy Wingo  <wingo at pobox.com>
> > 	* gst/gstobject.h (GST_OBJECT_REFCOUNT_VALUE): Just use the int.
> > 	It is volatile, after all.
> 
> This doesn't work on processors that require a memory barrier to
> serialize loads across multiple CPUs.

I was getting compilation errors on ppc, iirc, and I thought "well glib
uses this idiom in their debugging code too...", i.e. all the
g_return_if_fail stuff in gobject.c with the new atomic int stuff. As
this macro is only for debugging, e.g. in the test suite, I thought this
would be fine. Given that it's only for debugging is it still incorrect?
Presumably you would only debug refcounts at points in a program where
it is well defined.

Ignorantly yours,
-- 
Andy Wingo
http://wingolog.org/