On 30/08/2013, at 6:39 AM, johnrengelman <[email protected]> wrote:
> I've gone back to the drawing board and came up with a simpler solution where
> in ExternalResourceResolver
> I check for any artifact existing and then only mark the artifact as
> resolved by that repository if one is found.
We want to behave differently in this case for the Maven cache compared to
arbitrary repository. It's expected in the Maven cache for an artefact to be
missing, but it means a badly formed module everywhere else.
I think we want something like this:
1. If we find a pom and no artifact in the Maven cache
* If we can find the module in some other repository, silently ignore
the stuff in the Maven cache and use the module from the other repository.
* If we cannot find the module in some other repository, fail with an
'artefact missing' exception.
2. If we find a pom and no artefact in some repository, fail with an 'artefact
missing' exception.
Or, perhaps it might be nicer in #2 if we warn instead of fail when we can find
the module in another repository.
Regardless of exactly what we do here, it's not the resolvers responsibility to
decide whether to fail or continue the search. Instead, it should just
communicate what it found, so that `UserResolverChain`, which is coordinating
the search, can take care of stopping or continuing or whatever. To do this, we
might add some new states to `BuildableModuleVersionMetaDataResolveResult`, so
that the resolver can let the search algorithm know that it found a badly
formed or partial module.
>
> The problem I'm having is handling POM dependencies. There doesn't seem to
> be enough information passed in the Module Descriptor to know if the
> dependency that is currently being resolved has POM packaging.
>
> I tried one work around by looking at the ResolvedArtifact for the module
> descriptor and if it ends in ".pom" using the PomReader to parse it and get
> the packaging. That worked, but then I ran into issues with some integration
> tests where the the artifact packaging and classifiers didn't match
> (ArtifactDependenciesIntegrationTests.canUseClassifiersCombinedWithArtifactWithNonStandardPackaging).
> It appears that it was relying on the fact that the artifacts aren't checked
> before attempting resolution.
I would change the `ModuleDescriptorParser` implementations to return a
`MutableModuleVersionMetaData` instead of `ModuleDescriptor` (this is the plan
anyway). Then, the `ModuleVersionMetaData` can describe whether the module must
have artefacts or not. The pom parser can set this to false when packaging is
pom.
--
Adam Murdoch
Gradle Co-founder
http://www.gradle.org
VP of Engineering, Gradleware Inc. - Gradle Training, Support, Consulting
http://www.gradleware.com
Join us at the Gradle eXchange 2013, Oct 28th in London, UK:
http://skillsmatter.com/event/java-jee/gradle-exchange-2013
