reduceEdges does not always work for direct Dependencies

[ferstl]

The reduceEdges option does not always work correctly for direct dependencies for a module. For example, GraphIntegrationTest#transitiveExludes() shows this behavior. This is the currently created graph:

com.github.ferstl:module-3:compile
+- com.github.ferstl:module-1:compile
|  +- commons-codec:commons-codec:compile
|  +- org.apache.commons:commons-lang3:compile
|  \- junit:junit:test
|     \- org.hamcrest:hamcrest-core:test
+- com.github.ferstl:module-2:compile
|  +- com.github.ferstl:module-1:compile
|  +- com.google.guava:guava:compile
|  +- org.springframework:spring-core:compile (optional)
|  \- junit:junit:test
\- com.mysema.querydsl:querydsl-core:compile
   +- com.mysema.commons:mysema-commons-lang:compile
   \- com.infradna.tool:bridge-method-annotation:compile

However, the graph should look like:

com.github.ferstl:module-3:compile
+- com.github.ferstl:module-2:compile
|  +- com.github.ferstl:module-1:compile
|  |  +- commons-codec:commons-codec:compile
|  |  +- org.apache.commons:commons-lang3:compile
|  |  \- junit:junit:test
|  |     \- org.hamcrest:hamcrest-core:test
|  +- com.google.guava:guava:compile
|  +- org.springframework:spring-core:compile (optional)
|  \- junit:junit:test
\- com.mysema.querydsl:querydsl-core:compile
   +- com.mysema.commons:mysema-commons-lang:compile
   \- com.infradna.tool:bridge-method-annotation:compile

Edges are currently omitted when traversing the graph by checking if there is already another path for a new edge. In this example, this will not work, because the graph is created in this order:

• module-1 and its dependencies to commons-codec, commons-lang3 and junit
• module-2 and its dependencies to module-1, guava and spring-core
• module-3 and its dependencies to module-1, module-2 and querydsl-core

When processing module-3 and the dependency to module-1 is visited first, the edge can not be omitted because there is no other path from module-3 to module-1. When module-2 is visited afterwards, an edge from module-3 to module-2 is created. Since module-2 does already have a path to module-1, the graph does now contain two paths to module-1.

The edge from module-2 to module-1 has to be omitted, because the path via module-2 is "older" (meaning resolved earlier). This can either be achieved after each module is completely processed or after the creation of the full graph.