Community
Participate
Working Groups
I generate a webstart application for bug 268733 from Eclipse 3.5M6 and with export for multi platforms. When I run it with jdk 1.5, it starts with the right os and system : !SESSION 2009-04-02 13:27:59.671 ----------------------------------------------- eclipse.buildId=unknown java.version=1.5.0_16 java.vendor=Apple Inc. BootLoader constants: OS=macosx, ARCH=x86, WS=carbon, NL=fr_FR But fails because of missing cocoa bundle : !ENTRY org.eclipse.osgi 2 0 2009-04-02 13:28:00.321 !MESSAGE The following is a complete list of bundles which are not resolved, see the prior log entry for the root cause if it exists: !SUBENTRY 1 org.eclipse.osgi 2 0 2009-04-02 13:28:00.321 !MESSAGE Bundle org.eclipse.ui.cocoa_1.0.0.I20090120-1425 [409] was not resolved. !SUBENTRY 2 org.eclipse.ui.cocoa 2 0 2009-04-02 13:28:00.321 !MESSAGE Platform filter did not match: (osgi.ws=cocoa)
In this case do you want to use carbon or cocoa? If you want to use carbon then you need to include the swt carbon fragment instead of the cocoa one.
Well, the jnlp file is generated by eclipse and requires both fragments to be downloaded. See http://nricheton.homeip.net/public/nebula/features/org.eclipse.rcp_3.5.0.v20090307-9RA-FqhFstOCxSdfkxdojLKc.jnlp (I don't even selected Linux or Solaris targets.). I forget to mention that all fragments (both cocoa and carbon) are downloaded correctly. In that case, I expect the application to run with carbon, since this is the selected windowing system. (WS=carbon) What I don't understand is that eclipse seems to look for cocoa event if ws=carbon (added complete log)
Created attachment 130693 [details] Complete log
Webstart takes care of downloading all the jars before eclipse is even touched. Webstart is not capable of filtering on ws which is why both are downloaded. By having both listed in your jnlp file, you will always get this kind of warning since one or the other will not resolve. This by itself should not be a problem since the other platform would resolve. It may be necessary to edit the jnlp files that get generated (there are a few bugs, bug 196089 in particular). You can try debugging the startup by getting org.eclipse.equinox.launcher in your workspace, set a breakpoint in WebStartMain.discoverBundles to see what jars we are getting from JNLP. Set the following environment variables in a terminal to launch webstart in debug mode (this is windows syntax, on linux I used export) set JAVAWS_TRACE_NATIVE=1 set JAVAWS_VM_ARGS=-Xdebug -Xnoagent -Xrunjdwp:transport=dt_socket, address=8888,server=y,suspend=y javaws http://server:port/descriptor.jnlp You can then attach using a remote Java application launch config in eclipse. You may need to attach twice as webstart sometimes has a restart after downloading the jars.
Ok, Same problem than bug 270421 When Eclipse generates the jnlp for RCP, it uses <resources os="Mac" arch="x86"> But the real arch is "i386". So the carbon fragment is not downloaded. Moving this to PDE ?
(In reply to comment #5) > Ok, Same problem than bug 270421 > > When Eclipse generates the jnlp for RCP, it uses > <resources os="Mac" arch="x86"> > > But the real arch is "i386". So the carbon fragment is not downloaded. > > Moving this to PDE ? > Yes, I think PDE is a good place to start.
I believe build generates the JNLP.
*** This bug has been marked as a duplicate of bug 270421 ***
