Community
Participate
Working Groups
org.eclipse.mylar.context.core.ContextCorePlugin.start can deadlock! This seems to be a problem with sun java VMs. The problem comes from the fact that ContextCorePlugin.start tries to create a executable extension. For details (including stack trace) see bug 188943. I filed this bug against equinox and they marked it as fixed: > Actually there is a more general "best practice" that says "do as few as > possible in the activator" and since creating an executable extension is not a > trivial operation it falls under the typical things that you should not do. I think this is a major bug, because the only way to get out of the deadlock is to kill eclipse. It happend to me several times at startup. I use org.eclipse.mylar.ide_feature_2.0.0.v20070403 with eclipse 3.3 M7
Yes, we are trying to do as little as we can in the activator, but have additional requirements there that other plug-ins don't (e.g. see comment#10 on bug 187202). That' said, we haven't had a deadlock reported for months, since we made these improvements, so I need to figure out what could have caused it. Thanks for the pointer to the thread dump.
Btw, does this happen every time, frequently, or intermittently?
I upgraded to mylar 2.0.0.v20070514 and it still happened. I was not able to start one of my workspaces without deadlock. Then I decided to uninstall mylar, which made me really suffer, because it's hard to work without mylar. After 2 days of suffering without mylar, I reinstalled it and the problem is gone. It is very strange that it happened 5-6 times in a row and I had to kill eclipse every time and I could not get into my workspace. Now it has not happened for 2 days or so.
The context store is now extension point based and we should longer do any extension point reading on start(..).
Created attachment 72089 [details] mylyn/context/zip
cool :-)
