[willvarfar]

I remember the early runs of Coverity on the Symbian kernel and user-space.

Because Symbian's C++ is not very conventional, and is very careful in its own conventions, Coverity found basically nothing except false alarms.

[ernstsson]

True, Coverity needs to be configured correctly for the target system to get it working for you (I assume that especially Symbian C++ can be tricky). Similar here, the tolerance values for complexity might be set wrong giving us a lower score than we should have. The image itself is valid though.

[willvarfar]

Isn't the very low quality score setting off alarm bells and making you question if you're measuring the right things?

[ernstsson]

Yes, definitely. I think (or hope?) the main reason for the score is that the directory structure and the actual architecture of Linux isn't fully in sync. Most of the score comes from the top directory tangles and complexity. When I compare it to the commonly used kernel map (http://makelinux.homelinux.com/kernel_map/LKM3_512.png) I really don't see it matching as well as Arqua requires for a good score. Note also that the score doesn't say that the software doesn't work, just that the structure of the source in relation to the call graph of the functions is complex.