Community
Participate
Working Groups
I found this in the composite platform update site : S-3.7M1-201008051700/features/org.eclipse.sdk_3.7.0.v20100629-7Q7t-DPY2lXAsz0ngujjWy0i17bPiZUwAz0A8mqupM3lG.jar S-3.7M2a-201009211024/features/org.eclipse.sdk_3.7.0.v20100629-7Q7t-DPY2mYAyjigmrpYEb4Q44gPf5Y-94Wa22bALqtMJ.jar S-3.7M3-201010281441/features/org.eclipse.sdk_3.7.0.v20100629-7Q7t-DPX2naDL1jfutmm2idmoY-eH81sz-yIoQxy-Yeup.jar Why do all 3 features have the same date (20100629)? Following the alphabetical order, 3.7M2a is last, so p2 would consider it the most recent. Since all 3 milestones are defined in the same composite, projects in the release train building towards M3 that don't specify explicit dependency ranges are in fact depending on the M2a version of the platform. The same problem also affects JDT (both M2a and M3 are 3.7.0.v20100824-0800): ./S-3.7M2a-201009211024/features/org.eclipse.jdt_3.7.0.v20100824-0800-7z8dF_ZFMTfEO809v9JgQejMH155.jar ./S-3.7M3-201010281441/features/org.eclipse.jdt_3.7.0.v20100824-0800-7z8dFb7FMTfEQ4wDz0DJlHYd9H15.jar ./S-3.7M1-201008051700/features/org.eclipse.jdt_3.7.0.v20100805-1200-7z8dF_ZFMTfEMAp0kLJpz-ai9H15.jar and PDE (all three milestones have 3.7.0.v20100803): ./S-3.7M2a-201009211024/features/org.eclipse.pde_3.7.0.v20100803-7b7oFUFFEx2XmfZUk45657.jar ./S-3.7M3-201010281441/features/org.eclipse.pde_3.7.0.v20100803-7b7oFVoFEx2XmfgLsb9MXO7.jar ./S-3.7M1-201008051700/features/org.eclipse.pde_3.7.0.v20100803-7b7nFUFFEx2XmfZGeS6G751.jar and probably others I didn't check
That the date is the same is expected and OK. But the M3 string is wrong since it is less than the M2a string. PW
Looks like a bug in the qualifier generator.
We added and removed some fragments from the rcp feature in M3. It looks like this change didn't percolate up to the sdk feature, probably due to bug 208143. I've changed our Helios contribution to point to the child repo instead of just the composite. Also, I've tagged the SDK feature for the next build.
(In reply to comment #2) > Looks like a bug in the qualifier generator. And, not only that ... but isn't the "comparator test" supposed to catch it when that generator limitation occurs?
We only compare feature versions with the previous release, not the previous build.
This is not bug 208143, the symptoms of that are that small changes at the end of the qualifier get truncated leaving the version the same. Here, we actually have the SDK version decreasing, at a significant location in the suffix. This indicates some other problem with the suffix calculation. I reviewed all the other feature versions in the SDK and only org.eclipse.sdk has the problem, all the other versions incremented correctly. I'm moving this over to PDE/Build for further investigation. Since Kim has retagged the feature, the decreasing version is "fixed" there.
This is actually bug 162022, originally marked as Won't Fix as we didn't have any good solutions.
In the near term, the lesson is that we should tag all parent features when a bundle or fragment is removed from a feature. I.e., since a fragment was removed from org.eclipse.rcp, ensure that platform and SDK are also tagged. Hopefully fragment removals are rare enough that this is a manageable workaround.
The reported problem got fixed, see comment 3. For the underlying problem see bug 162022.
