In the unfixed version, the performance benchmark loads each of the four pairs of metamodels into the same resource set (within the "getModel()" method). The problem is that the metamodels have the same namespace URIs over the different scenarios, violating a global assumption of EMF: ResourceSet.getPackageRegistry().getEPackage(nsURI) returns a single package, and not a list of packages. Since my code relies on this global assumption, I submit a fix, which ensures that each pair of metamodels (together with the instances) is loaded into a separate ResourceSet.
In the unfixed version, the performance benchmark loads each of the four pairs of metamodels into the same resource set (within the "getModel()" method). The problem is that the metamodels have the same namespace URIs over the different scenarios, violating a global assumption of EMF: ResourceSet.getPackageRegistry().getEPackage(nsURI) returns a single package, and not a list of packages. Since my code relies on this global assumption, I submit a fix, which ensures that each pair of metamodels (together with the instances) is loaded into a separate ResourceSet.