Why are there two implementations of Maven dependency analysis?

[ascheman]

In the current code we find two implementations of Maven dependency analysis:

• The normal dependency analysis scans the output of an execution of mvn dependency:tree,
• if this fails (either by error, or empty result), an internal implementation analyses the pom.xml files directly (which is also not error-free).

My personal opinion (for discussion) is: There should be one and only one implementation for the analysis. If it fails for whatever reason this should be treated as a bug, and the bug should be fixed. Maintaining two implementations is to much overhead and confuses contributors.

Confusion can be seen by errors like #390, which only can happen in the second case, and #392, which happen by incomplete implementation of Maven output parsing (among other problems like #393). I would even ask what makes the first one the default implemenation and the second operation the fallback? Why isn't it the other way round? Why aren't both combined to a working solution?

[ascheman]

How a root problem in the analysis leads to second problem is documented here.

[jmle]

We might not need the fallback anymore, but I am not sure about why it was added in the first place. @pranavgaikwad do we still need it? I don't think we will, especially with the new plugin