On 11/2/22 8:09 AM, Pavel Begunkov wrote: > On 11/2/22 13:45, Jens Axboe wrote: >> On 11/2/22 5:32 AM, Pavel Begunkov wrote: >>> On 10/31/22 13:41, Dylan Yudaken wrote: >>>> Split out the specific sendzc parts of struct io_sr_msg as other opcodes >>>> are going to be specialized. >>> >>> I'd suggest to put the fields into a union and not splitting the structs >>> for now, it can be done later. The reason is that the file keeps changing >>> relatively often, and this change will add conflicts complicating >>> backporting and cross-tree development (i.e. series that rely on both >>> net and io_uring trees). >> >> Not super important, but I greatly prefer having them split. That >> way the ownership is much clearer than a union, which always >> gets a bit iffy. > > I'd agree in general, but I think it's easier to do in a few releases > than now. I'd rather just deal with that pain, in reality it'll be very minor. And if lots of churn is expected there, it'll be the most trivial of rejects for this particular area. > And this one is nothing in levels of nastiness comparing to > req->cqe.fd and some cases that we had before. I agree, but that's not a reason to allow more of it, quite the contrary. -- Jens Axboe
