Bugzilla – Full Text Bug Listing
| Summary: | [prov] Cannot install 20080204-0010 | ||||||
|---|---|---|---|---|---|---|---|
| Product: | [Eclipse Project] Equinox | Reporter: | Susan McCourt <susan> | ||||
| Component: | Incubator | Assignee: | Pascal Rapicault <pascal> | ||||
| Status: | RESOLVED FIXED | QA Contact: | |||||
| Severity: | critical | ||||||
| Priority: | P3 | CC: | dj.houghton, jeffmcaffer, pascal, simon_kaegi | ||||
| Version: | 3.4 | ||||||
| Target Milestone: | 3.4 M5 | ||||||
| Hardware: | PC | ||||||
| OS: | Windows XP | ||||||
| Whiteboard: | |||||||
| Attachments: |
|
||||||
|
Description
Susan McCourt
I am running ProvAdminUI from HEAD, synched this morning. My guess is this is another case of being broken by a feature that changed, but whose version identifier didn't change due to rounding errors. I can install 20080204-0800 fine, but get the same error as Susan with 20080204-0010. Scratch the above - I compared the feature versions between 0010 and 0800, and they appear correct. The older versions that are in conflict here seem to be coming from a build that occurred on 20080201-1639 - there wasn't a scheduled build at that time so I don't know how it got in there. Also, I don't see how they would be dragged in by installed the SDK from 20080204-0010. This smells like perhaps an optional pre-requisite on a non-singleton bundle that is causing IUs from older builds to be dragged in. The build with the strange timestamp is probably one that we did locally and pushed up to the eclipse.org servers. Currently I am able to download the agent from 20080204-0800 and use it to install and run a provisioned SDK with no problems. I am using the repo at http://download.eclipse.org/eclipse/testUpdates. Same here - 20080204-0800 installs fine. I can install SDK, user ui, then install releng.tools from user ui. I'm self-hosting on this now. Source lookup is broken, and the install roots are missing so I can't upgrade, but it runs. it is working for me now too. you should be ableto see the roots by turning off the "show only roots" option I'm seeing the same problem installing either I20080205-0010 or I20080204-1800. It seems only I20070204-0800 can be installed from our test repository. Anybody else have success? I'm also having the same problems with I20080205-0010 or I20080204-1800. I'm wondering if this could be caused by the fact that feature groups are not marked as singletons. The precise four part version ranges shown in the planner exception only appear in feature groups, and root IUs. Root IUs are marked as singletons so we should never get multiple copies of a root IU resolving together. This core of the problem is that the root IU requires multiple category IUs (whereas it should require none). In turns each category IU requires each feature thus causing the resolver to do the right thing and trying to resolve a system where everything should be satisfied which fails. John is investigating a fix to the metadata generator. Mid-air collision, but similar conclusion here. Leaving it to John. --- Looking closely at the error for I20080205-0010 I noticed something suspicious: Unsatisfied dependency: [sdk 3.4.0.I20080205-0010] requiredCapability: org.eclipse.equinox.p2.iunamespace/I20080204-1800/[0.0.0,0.0.0] e.g. The SDK seems to be requiring the previous category IU. I double-checked that the categoryIU require is really in the I20080205-0010 SDK IU although I'm not sure why it's there. If we bring in this older category IU we will end up requiring the older agentFeatureGroup which will cause all sorts of problems. Created attachment 88898 [details]
A repo for testing
In addition of the IUCategory pb, we have been hit by bug #208143. I have released a fix to the generator so that category IUs are not required by the root id. This required some refactoring in the generator because we passed the "resultantIUs" set in most methods. This is now factored to make it much easier to omit other IUs from the root if we decide it makes sense. |