On Tue, Dec 13, 2016 at 06:24:26PM +0000, Richard W.M. Jones wrote: As you may remember I did something similar last year (in Python) and this went nowhere because the information in the drivers was unreliable. I didn't check lately but I guess it still is. And I tend to think that pulling in yet another dependency in libguestfs is not really desirable; the driver packaging can sort it all out with the existing libguestfs' expectations. I mean, the existing protocol of encoding the driver attribution in the directory structure is good enough and convenient, and is easy to consume both by libguestfs and by other software, without requiring deep knowledge of the driver guts. The packaging may use whatever extra knowledge it wants to lay the drivers out in this structure, but driver consumers (libguests and others) don't need to care. Roman.