Posted by rhatcher on Aug 10, 2012; 2:06pm
URL: https://forum.jogamp.org/AWT-JOGL-Deadlock-Revisited-tp4025772.html

Fedora 12/i386
Sun/Oracle 7u5
JogAmp RC9

This past week we switched our code base from JOGL 1.1.1a to JogAmp RC9.  With a little help from perl the conversion process was painless, and things are working very well in general.

However...

After the switch we are sporadically seeing a deadlock similar to the one described in this topic from a few weeks ago:  http://forum.jogamp.org/Deadlock-during-JOGL-initialization-tc4025625.html.

I wasn't sure what the protocol was on the forum, i.e. should I start a new topic or squat on the one mentioned above.  That topic ended without apparent resolution so I thought it might be better to start over.

The presentation is that perhaps a third of the time our viewer desktop application will deadlock at startup.  Drawing never starts, menus don't work, and the event queue seems to be hung up across the board.

At the end of the topic mentioned above Sven had asked for a jstack output from the locked process.  Ours looks like this:

stack.txt

The backtraces in the relevant areas are similar to those discussed in the other topic.  The deadlock says:

Found one Java-level deadlock:
=============================
"Timer-5":
  waiting for ownable synchronizer 0xac94ad18, (a java.util.concurrent.locks.ReentrantLock$NonfairSync),
  which is held by "AWT-EventQueue-0"
"AWT-EventQueue-0":
  waiting to lock monitor 0x09e2423c (object 0xac8d1da0, a sun.awt.PostEventQueue),
  which is held by "Timer-5"

In the other topic there is no obvious association with JogAmp in the classes apparently involved in the deadlock, and that's the case here too.  The only association is indirect in that the problem started immediately after we switched, and JogAmp JOGL just happens to get involved with AWT event stuff.

This could easily be something we're doing that JOGL was tolerating better but at the moment I don't have any guesses.  I'm going to try a new hotness dev build and see if the problem remains.

Any other ideas?