@jeremylong a good step forward. But definitely needs more work as it currently creates duplicate dependencies on aggregate-scanning a large multimodule project

@jeremylong Differences in results can be explained by a failure to resolve the reactor-projects in the new code plus existing behaviour that virtual dependencies are not merge together as a single one.

If you're working towards a 6.5.2 release feel free to already promote this from draft to normal and merge it. I will try to find out why the resolution did work in the 6.5.1 case and doesn't work in the 6.5.2-SNAPSHOT case from this branch, but the difference I spotted is in my view sufficiently explainable to warrant inclusion of this change in a next fixpack in order to solve the massive slowdown.

The project I used had 10 inter-module-dependencies, which were 10 resolved dependencies in 6.5.1. In 6.5.2-SNAPSHOT they become unresolved, triggering virtual dependency creation. And as some of those inter-module dependencies occur for multiple dependencies it results in a total of 24 inter-module dependencies being listed for what actually is only 10 inter-module dependencies.

That multiplication of virtual dependencies also happens in 6.5.1 if a reactor-dependency in a project is not yet resolvable.

Not resolving the (existing) release versions of the reactor-dependencies I consider a regression of this change. Duplicating the virtual dependencies I will file a separate GH issue once I've managed to tackle this regression.
If you decide to merge this already in order to put out a 6.5.2 the regression could be handled under a new GH issue.

@jeremylong already found the pattern when it fails: when the artifact searched for uses a classifier. So think it's better to get this resolved before merging.

@jeremylong already found the pattern when it fails: when the artifact searched for uses a classifier. So think it's better to get this resolved before merging.

Minor correction, not the classifier, but a non-jar packaging (ejb)

I feel like we need to release 6.5.2. We can publish another release as soon as the regressions are solved.

I feel like we need to release 6.5.2. We can publish another release as soon as the regressions are solved.

Sorry I hadn't commented that the regression was resolved by my last commit. Could've gone into the 6.5.2. But I propose to hold back 6.5.3 until the duplication of virtual dependencies (#3944) also gets resolved as well as I intend to look at that tonight.

@jeremylong Fix for #3944 is in its final testing stage and will get its own branch and PR