When EMMA coverage data is generated for classes that are not compiled with debug enabled, the coverage data collected is slight different from the data collected with classes containing debug information. Specifically, no line coverage statistics are collected and the source file level is omitted from the coverage report xml when the classes do not contain debug information. In these cases, no coverage statistics for the class scope are being loaded by the EMMA report parser, because it assumes there will always be a source file level in the xml hierarchy. This also presents an issue for selecting the preferred code coverage granularity for EMMA coverage data. The preferred code coverage granularity (line, block, branch, class, etc.) is used to select the code coverage statistics to use when all granularities cannot be taken into account. For example, when calculating the code coverage delta by user, it would be too confusing to graph this for all users for all granularities (5 authors with line, block, class, and method stats would produce a graph with 20 lines), so the data graphed needs to be limited to one granularity. Line coverage stats had been selected as the preferred granularity for EMMA, primarily because I expected this to be the most widely understood and utilized granularity. For the purposes of resolving this issue, I think changing the preferred granularity to block is the prudent choice, since this is the closest to line coverage and is the type of code coverage stat showcased by EMMA. Ideally, the preferred granularity would be a user-configurable setting.