Comment 29 for bug 1112907

This second version is a bolder step, but I think it's the right thing to do. It's a little more risky than the first one, but I (and the test suite) are fairly convinced that it's correct.

Pro: addresses this kind of problem for all device types under Linux, and avoids D-BUS signalling device removals/additions from upower where the device itself did not really go away, just the parental USB/bluetooth tree structure changed
Con: More intrusive, could theoretically cause regressions (but I have no idea how, as upower already only uses the device name for building object names)

Both patches have detailled commit messages which explain their rationale.

@Richard: I recommend the second patch, but I posted both to get you a chance to compare.