Il 09/08/2013 16:51, Alex Bligh ha scritto: > > On 9 Aug 2013, at 15:36, Paolo Bonzini wrote: > >>> Would that not require change the huge number of qemu_timer_new references >>> to use this new API? That sounds less than automatic! Not in favour of >>> that one. >> >> qemu_timer_new can remain for now (only waiting for the next >> mass-rewriting script to be written). I would just prefer to have the >> new AioContext-/TimerList-aware not do any allocation. > > One nit: > > Currently struct QEMUTimer lives nicely within qemu_timer.c, hidden > from the .h, which only contains: > typedef struct QEMUTimer QEMUTimer; > > If what you are after is for people to put a QEMUTimer into their > struct, rather than a QEMUTimer * (and hopefully to remember to > put qemu_timer_init in!) then I need to move the whole of the > declaration of the struct to the .h file which is a bit yucky.
Yeah, that's fine. We're not relying too much on opaque types anyway. Paolo > Other than that, fair enough. >