Bug 328990

Description Scott Lewis 2010-10-28 18:43:53 EDT

When the following repo is installed into a fresh copy of Eclipse 3.6.1 sdk on windows

https://ecf2.osuosl.org/hudson/job/R-Release_3_4-sdk.feature/6/

the wrong version of the org.eclipse.ecf.examples.remoteservices.hello bundle gets installed.  It *should* install the built version (same qualifier as other bundles included in the repo...i.e. 3.0.0.v20101028-0432), but instead installs the version from the Helios repository (i.e. 2.0.0.v20100913-1814).  When the hello example code is run on the install it fails with a class not found exception for HelloMessage (which was indeed added in between 2.0.0 and 3.0.0).

Same effect see for the master/HEAD continuous build...i.e.

https://ecf2.osuosl.org/hudson/job/C-HEAD-sdk.feature/1360

This may very well be a problem in p2, and/or Hucky. 

In an attempt to deal with this as quickly as possible in anticipation of ECF's 3.4 release (10/29/2010), I've added package version for the export of org.eclipse.ecf.examples.remoteservices.hello, and the consumer import where the new HelloMessage class is accessed.  I've then initiated a C-HEAD build, resulting in the repo from this build:

https://ecf2.osuosl.org/hudson/job/C-HEAD-sdk.feature/1361/

When this new repo (with the package version set to 3.0.0) is used to install into 3.6.1, it results in *both* the 3.0.0.v20101028-2059 *and* the 2.0.0.v20100913-1814 bundles being installed into Eclipse.  This is weird...but as a workaround, this does work (and the HelloMessage class is loaded, and , although it's non-optimal as it should not install the 2.0.0.v20100913-1814 bundle. 

So...as a workaround for 3.4 release I would be willing to go with the changes I've already made to master (i.e. adding the package version info)...and if you/Markus agree, I suggest that the changes to master be applied to Release_3_4 branch (I have not done that, but I think it could just be a full master-to-Release_3_4 merge as nothing else has changed on master).  

If you/Markus have any insight into what might be causing this problem, and can identify a short term fix, then we can do that.  I don't immediately understand this behavior, though, and it seems like it may be a fairly subtle problem in p2, the p2 publisher, or perhaps Buckminster.