Community
Participate
Working Groups
Created attachment 172562 [details] Test case Between 3.5 and 3.6 releases there is a behaviour change in retrieving packed artifacts in repositories with publishPackFilesAsSiblings=true, but artifact.uuid set on certain artifacts. In Eclipse 3.5 we would use the UUID and retrieve the artifact from the blobstore however in 3.6 we attempt to retrieve the artifact from plugins. I've attached a test along with a repository which should demonstrate the problem, minor changes are required between the two version of Eclipse.
This problem was originally reported as https://issues.sonatype.org/browse/TYCHO-464
Matt, any patch in hand?
Created attachment 177040 [details] Patch In the patch I've reversed the check for an artifact UUID ahead of the flatButPackedEnabled check.
The change of behaviour observed here comes from the fix for bug #306056. The fix that Matt suggests ends up putting the code back to the same state than what we shipped in eclipse 3.5. This code would be consistent with the createLocation() method. Unfortunately that breaks the test written for bug #306056 (ArtifactMirrorApplicationTest#testArtifactMirrorUUID_bug306056), but in retrospect I'm questioning the validity of the fix for 306056. Indeed there was two parts quoting: "Two things: 1) the destination repo should probably clear the UUID if it isn't using the blobstore 2) when getting the artifact, we should check publishPackFilesAsSiblings before checking the UUID" The first part of the fix is definitely correct, however the second part cater for the case we have had bad repositories resulting from bug fixed by the first part of the fix. I'm not sure that we should have done this. What do you guys think?
(In reply to comment #4) > 2) when getting the artifact, we should check publishPackFilesAsSiblings before > checking the UUID" > > The first part of the fix is definitely correct, however the second part cater > for the case we have had bad repositories resulting from bug fixed by the first > part of the fix. > I'm not sure that we should have done this. > > What do you guys think? It doesn't really matter to me if this part is reverted. In this case, we really have a repository that is not internally consistent, if using the UUID gets better results from the user's end, then that is probably the way to go.
3.6.1 is pretty much done at this point. Can we defer this bug?
Discussed with Pascal. We need to get this in for the RC3 rebuild. Pascal, please post a note to the releng list about this fix we are including in the respin.
I've released Matt fix and removed the test that this change was breaking. Fixed in HEAD and the branch.
There was a problem with tagging HEAD so this fix won't appear in integration builds until the first build after 3.7 M2.
