Coveralls (and other coverage sites e.g. Jenkins) are great. Knowing which code is untested helps find bugs and unused code.
miniblog.
Related Posts
I find test coverage a useful thing to track, but I can't find a good metric.
"Coverage decreased by 10%" sounds bad. If you're deleting pointless code that has tests, it's good!
Perhaps total untested lines of code is better?
Given coverage data, how would you build a tool to decide which untested parts of a codebase most need tests?
The best I can think of is using profile data to ensure hotspots are tested. Not sure if it's ideal though: top level code and well-exercised logic would be highlighted.