gtk2reactor consumes loads of CPU on Ubuntu Hardy after SIGCHLD

[niemeyer]

The simple attached script will consume a lot of CPU for as long as it runs in the Ubuntu Hardy environment. I've tested it both with the Twisted version currently in Hardy (2.5.0-2build2), and also with the current trunk (revision 22957).

The issue is only reproducible with the glib2 reactor, and is not reproducible in Ubuntu Gutsy (even with Twisted trunk), which hints about behavior change in a dependency. It was also possible to execute the script a few times without getting the bad behavior, which might indicate a race somewhere.

[niemeyer]

Test case

[niemeyer]

It turns out that the actual reason for the process to spin is the fact that it receives a SIGCHLD or not. I'm attaching a new test case which reproduces the problem consistently.

[niemeyer]

Updated test case

[radix]

Apparently this is only being seen after a particular upgrade of glib2/pygtk:

Broken are:

• libglib2.0-0_2.16.1-1ubuntu3_i386.deb
• python-gobject_2.14.1-2_i386.deb

Working are:

• libglib2.0-0_2.14.1-1ubuntu1_i386.deb
• python-gobject_2.14.0-0ubuntu1_i386.deb

It may be that Twisted needs to be updated for an API change, or it may be a newly introduced bug. We should investigate exactly what changed between those releases.

[niemeyer]

Sebastien Bacher found out that the bug is introduced by the modification in pygtk's gobject discussed in the following ticket:

 http://bugzilla.gnome.org/show_bug.cgi?id=481569

[glyph]

Hmm. I wonder if "wontfix" is really the right thing to say for tickets like this.

Is there a chance that we're going to need to work around this revision of gobject "in the wild"? Should we track the issue being dealt with in bugzilla?

(Grumble, grumble, I really need to write that intertracker protocol specification that I keep promising the launchpad guys I'll write one day...)

[exarkun]

#3981 appears to have been a duplicate of this.