Encountered 'multiple artifacts retrieved to same file' error when module does not have multiple
artifacts
----------------------------------------------------------------------------------------------------------
Key: IVY-1148
URL: https://issues.apache.org/jira/browse/IVY-1148
Project: Ivy
Issue Type: Bug
Components: Ant
Affects Versions: 2.1.0, trunk
Environment: Windows or Linux, i386
Reporter: Carlton Brown
Fix For: trunk
Recently we started experiencing intermittent problems with the error 'multiple artifacts
retrieved to same module' for artifacts that did not have multiple artifacts.
I have not been able to create a simple contrived error, though I can generally reproduce
it in our complex set of environments and artifacts.
It started happening when we started using the branch attribute in our modules. Usually
the problem happens when a module has a both a direct and an indirect dependency on another
module. We were working around it by excluding the module from being resolved indirectly.
During debugging, I traced it to the fact that a certain identical module revision was not
being handled as identical. Specifically, in line 335 of RetrieveEngine, the artifact was
being added to the conflictsReports HashSet. The ArtifactDownloadReport.equals() was determining
them to be equal, so I did not expect them to be inserted. But the ArtifactDownloadReport.hashCode()
was coming up with different integers.
The cause of this is that one of the artifacts had an additional qualified attributed called
'merged' which was not different, even though this was the same version of the same artifact.
The value of the attribute looked like:
orgA#moduleA#trunk;build1245 -> orgB#moduleB#trunk;build833
Where module Z is the module being resolved, and it has a direct dependency on module A and
moduleB, and moduleB has a direct dependency on moduleA.
So because of the different qualified attribute, an artifact that represented the same file
was returning a different hash code.
I'm not sure what this extra 'merged' information represents. It seems to represent something
about how the artifact was resolved. There is no possible retrieve pattern that could (or
should) differentiate artifacts that differ only in the 'merged' attribute, so I think this
is a little too strict.
My strategy, for which I am attaching a patch that passes existing unit tests, is to use a
hashCode() and equals() method that represents the minimum necessary to determine whether
an artifact maps to a unique file.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
|