On 13/06/2012 5:40 PM, Christopher Faylor wrote:
On Wed, Jun 13, 2012 at 05:07:17PM -0400, Ken Brown wrote:
On 6/13/2012 2:40 PM, Ken Brown wrote:
On 6/10/2012 8:45 PM, Ken Brown wrote:
The bisection shows that the first problematic commit is this one:
http://git.gnome.org/browse/glib/commit/?h=glib-2-32&id=7eae486179e2799c369ed9ffcea663bf9161ce79
Author: Ryan Lortie<[email protected]>
Date: Wed Aug 31 22:07:02 2011 -0400
GMain: simplify logic for g_wakeup_acknowledge()
Instead of messing around with context->poll_waiting, just look at the
GPollFD to see if the GWakeup needs to be acknowledged.
I think this commit contains a typo: "events" should be "revents".
(context->wake_up_rec.events is always nonzero at this point in the
code, so it makes no sense to test that.) As a result,
g_wakeup_acknowledge() is being called much more often than necessary. I
think this could easily explain the performance problems that have been
reported, but I won't have a chance to test this on my (slow) XP system
for a while, and possibly not until tomorrow.
I've confirmed that fixing the typo solves the problem on my XP system.
I suspect that this issue is not specific to XP after all and has more
to do with the slowness of the machine, so that the performance problem
is more noticeable. In any case, it's clearly a glib bug.
I've filed a bug report upstream:
https://bugzilla.gnome.org/show_bug.cgi?id=678052
What incredible perseverence you've shown in tracking this down. Bravo!
++
Amazing. This is one of the nastier kinds of bugs I've seen anybody
track down... bug not where anybody thinks it is, and in library code
generally assumed to be robust, bisecting the monster that is glibc, bug
hunter knows (or willing to learn) glibc well enough to tell what
revents and events do (and that revents even exists!), but not so well
that the eyes glossed over the typo.
Ironically, the 'r' key on my laptop has been acting up lately, so I can
totally relate to the poor fellow who tried to type 'revents' and didn't
notice when 'events' compiled and checked in instead...
Ryan
--
Problem reports: http://cygwin.com/problems.html
FAQ: http://cygwin.com/faq/
Documentation: http://cygwin.com/docs.html
Unsubscribe info: http://cygwin.com/ml/#unsubscribe-simple
