On 10/2/24 9:13 AM, Christoph Hellwig wrote: > On Wed, Oct 02, 2024 at 09:03:22AM -0600, Jens Axboe wrote: >>> The previous stream separation approach made total sense, but just >>> needed a fair amount of work. But it closely matches how things work >>> at the hardware and file system level, so it was the right approach. >> >> What am I missing that makes this effort that different from streams? >> Both are a lifetime hint. > > A stream has a lot less strings attached. But hey, don't make me > argue for streams - you pushed the API in despite reservations back > then, and we've learned a lot since. Hah, I remember arguing for it back then! It was just a pain to use, but thankfully we could kill it when devices didn't materialize with a quality of implementation that made them useful. And then we just yanked it. Nothing is forever, and particularly hints is something that can always be ignored. Same for this. >>> Suddenly dropping that and ignoring all comments really feels like >>> someone urgently needs to pull a marketing stunt here. >> >> I think someone is just trying to finally get this feature in, so it >> can get used by customers, after many months of fairly unproductive >> discussion. > > Well, he finally started on the right approach and gave it up after the > first round of feedback. That's just a bit weird. I'm a big fan of keep-it-simple-stupid, and then you can improve it down the line. I think the simple approach stands just fine alone. -- Jens Axboe
