Community
Participate
Working Groups
+++ This bug was initially created as a clone of Bug #403589 +++ Build ID: 20130509-1108_eclipse-classic-kepler-M7-win32.win32.x86.zip Doing a binary compare of the eclipse classic M7 package (mentioned above) against the eclipse-SDK-4.3M7-win32.zip , I noticed that binary contents of all the bundles is different. I've looked into org.eclipse.compare_3.5.400.v20130423-1526.jar and noticed the following diff: epp-classic-kepler-M7/compare/META-INF/MANIFEST.MF: Eclipse-SourceReferences: scm:git:git://git.eclipse.org/gitroot/platfo rm/eclipse.platform.team.git;path="bundles/org.eclipse.compare";tag=" I20130424-0830";commitId=2d3e21f97bf11d1bb5abb7fc1a426d2f2b377a54 platform-SDK-kepler-M7/compare/META-INF/MANIFEST.MF: Eclipse-SourceReferences: scm:git:git://git.eclipse.org/gitroot/platfo rm/eclipse.platform.team.git;path="bundles/org.eclipse.compare";tag=" I20130501-0616";commitId=688c909262be4125519f167e1ce5b5d99ddca0f2 As a result of this diff, the signing info is also different between epp and platform SDK. As you see, the "tag" and "commitId" of the epp-classic package is too old, compared to the respective bundle in platform-SDK. I've had a look at the git repo and noticed that the old tag had been valid a couple days before the final kepler M7 build of the Platform: https://git.eclipse.org/c/platform/eclipse.platform.team.git/log/ I'm not quite sure who's fault this is: - Apparently there are bundles out there in the wild, which have the same ID, Version and Qualifier but different contents. That seems like something the Platform team should not allow, and might point to a problem in the Platform team's releng infrastructure (Comparator in particular). - Apparently when the EPP project produces the package, it pulled content from an old p2 repository that has contents which does not match the Platform team's M7 delivery. This might point to a problem in the EPP definition of the repositories, OR in the Platform team providing too much or nodeterministic contents in its repoistory. - Finally it might also be a problem in Equinox/p2 being nondeterministic when multiple solutions to a provisioning exist (see also https://bugs.eclipse.org/bugs/show_bug.cgi?id=408208 ). I've filed the bug against EPP initially because I think that even with non-perfect Platform deliveries or p2 procedures, the EPP project should be able to name exactly the same build input for a milestone that the Platform also used, and thus produce exactly identical builds.
Marking P2 / critical / target RC1 since we don't have much time left getting this right for Kepler. I don't know exactly where the EPP build pulls its contents from, so I'm kindly asking Markus to have a look.
I think this is a dup of bug 407134 ... and suspect EPP did get the "correct" versions from repo. The ones in our project archives just happened to be "incorrect". :)
Marking this particular issue as RESOLVED with RC1 since I cannot observe the problem with different MANIFEST.MF contents any more. Unfortunately I still see binary diffs, but apparently different ones. So as a follow-up work on the "binary diff" saga, I've filed https://bugs.eclipse.org/bugs/show_bug.cgi?id=408944 to investigate. Feel free to "duplicate" the issue with bug 407134 if you think that's appropriate.