Code considered by coverage tool should be independent of tested configuration

Currently we have the problem that our coverage reports will report the covered source as a percentage of the compiled source.

This gives misleading and dangerous results as it means the % will change as various features are enabled / disabled.

For example, consider that someone produced an API relating to network management with a stub backend and a backend using networkmanager. The networkmanager backend may be disabled at configure time if relevant headers are not present. This means that only the stub backend will be compiled, and only the stub backend will be considered by the coverage tool - possibly reporting coverage of 100% even though the "real" backend was not tested at all!

The coverage tool should be set up more or less so that a result of 100% means that "100% of the source code in the tree was tested" - as opposed to the current meaning of "100% of the source code which was compiled was tested".