JBoss/ WildFly seem more like the market leading JEE appserver, rather than a solution which has been “rejected by the market”.
I wouldn’t discount OSGi too quickly either, given that it underpins Eclipse which is still the leading Java IDE.
Maven is also the leading Java build tool, while we’re on this kind of area.
From my surface experience (JBoss works well) I tend to assume that JBoss Modules, which underpins JBoss and allows different modules & disparate versions to operate in harmony within the one JVM, is likely to be good technology.
My personal opinion is that this is a “requirements problem”. I believe that competent architecture of a Java module system should establish some solid foundations for existing solutions to live within it, rather than drawing a boundary so small that it does not address compatibility & instead creates incompatibility.
Since this is a requirements error, the best approach is to rollback & revisit the requirements to include establishing a basis for compatibility with key existing technologies — JBoss Modules, OSGi and Maven.
Heck, maybe I’ll email Mark Reinhold directly.