On Thu, Aug 02, 2012 at 10:59:20AM +0200, Paolo Bonzini wrote: > Il 02/08/2012 09:57, Wenchao Xia ha scritto: > >>> +int qbi_init(struct QBlockInfo *info) > >> > >> Should return void. > >> > > yes, it is useless now, but maybe in future it may fail. Returning > > int for every API results in only one more "eax store", I hope I can > > keep all API returns indicate success/fail, while add comments to say > > which API would not fail now. > > The question is *how* can it fail? It's a memset, do you envision it > becoming something else? > > It's not about the performance of the processor, it's about the > performance of whoever has to read the documentation and will wonder > what the return code is for. > > (Though I proposed getting rid of this function too).
Yes, please. Drop _init()/_uninit() and do this in _open()/_close(). I'm also not sure why qb_create() needs a QBlockState* argument. Can it be dropped? Stefan
