On 16/11/2020 14.31, Peter Maydell wrote: > On Mon, 16 Nov 2020 at 13:28, Philippe Mathieu-Daudé <f4...@amsat.org> wrote: >> Well, this is not an usable device but a part of a bigger device, >> so here we want the opposite: not list this device in any category. >> >> Maybe we could add a DEVICE_CATEGORY_COMPOSITE for all such QOM >> types so management apps can filter them out? (And so we are sure >> all QOM is classified). >> >> Thomas, you already dealt with categorizing devices in the past, >> what do you think about this? Who else could help? Maybe add >> someone from libvirt in the thread? > > If we could get to the point where we can assert() that > dc->categories is non-zero in class init, we would be able > to avoid further "forgot to categorize device" bugs getting > into the tree in future, which seems like an argument for > having some way of marking "really just an implementation > detail" devices I guess?
IMHO we need: assert(dc->user_creatable == false || categories != 0) then we don't need something like DEVICE_CATEGORY_COMPOSITE. Thomas
