On 16 Sep 2021 at 08:06, Darrick J. Wong wrote: > Hi again, > > Now that 5.15-rc1 is past, I would like to say something: > > I've been reflecting on my (sharply higher) stress levels in 2021, and > realized that I don't enjoy our development process anymore. One of the > nice things about being a co-maintainer is that I can take advantage of > the fact that suggesting improvements == leadership, though that comes > with the responsibility that leadership == actually making it happen. > > The thing that has been bothering me these past few months is how we > decide what's going into the next merge window. People send patchsets > at various points between the second week of the merge window and the > week after -rc6, and then ... they wait to see if anyone will actually > read them. I or one of the maintainers will get to them eventually, but > as a developer it's hard to know if nobody's responding because they > don't like the patchset? Or they're quietly on leave? Or they're > drowning trying to get their own patchsets out to the list? Or they > have too many bugs to triage and distro kernel backports? Or they're > afraid of me? > > Regardless, I've had the experience that it's stressful as the > maintainer looking at all the stuff needing review; it's stressful as a > developer trying to figure out if I'm /really/ going to make any > progress this cycle; and as a reviewer it's stressful keeping up with > both of these dynamics. I've also heard similar sentiments from > everyone else. > > The other problem I sense we're having is implied sole ownership of > patchesets being developed. Reviewers make comments, but then it seems > like it's totally on the developer (as the applicant) to make all those > changes. It's ... frustrating to watch new code stall because reviewers > ask for cleanups and code restructuring that are outside of the original > scope of the series as a condition for adding a Reviewed-by tag... but > then they don't work on those cleanups. > > At that point, what's a developer to do? Try to get someone else's > attention and start the review process all over again? Try to get > another maintainer's attention and have them do it? This last thing is > hard if you're already a maintainer, because doing that slows /everyone/ > down. > > (And yes, I've been growing our XFS team at Oracle this year so that > this doesn't seem so one-sided with RedHat.) > > I've also heard from a few of you who find it offputting when patches > show up with verbiage that could be interpreted as "I know best and > won't take any further suggestions". I agree with your feelings when > I hear complaints coming in, because my own thoughts had usually already > been "hmm, this sounds preemptively defensive, why?" > > Ok, so, things I /do/ like: > > During the 5.15 development cycle I really enjoyed going back and forth > with Dave over my deferred inode inactivation series and the log > speedups that we were both proposing. I learned about percpu lists, and > I hope he found it useful to remember how that part of the inode cache > works again. This dialectic was what drew me to XFS back in 2014 when I > started working on reflink, and I've been missing it, especially since > the pandemic started. > > So the question I have is: Can we do community sprints? Let's get > together (on the lists, or irc, wherever) the week after -rc2 drops to > figure out who thinks they're close to submitting patchsets, decide > which one or two big patchsets we as a group want to try to land this > cycle, and then let's /all/ collaborate on making it happen. If you > think a cleanup would be a big help for someone else's patchset, write > those changes and make that part happen. > > There's never been a prohibition on us working like that, but I'd like > it if we were more intentional about working like a coordinated team to > get things done. What do you all think? > > (Small changes and bug fixes can be sent any time and I'll take a look > at them; I'm not proposing any changes to that part of the process.) > Apart from patchsets associated with implementing new features, small changes and bug fixes for an upcoming kernel release, may be we should also allow developers to post RFC patchsets to obtain an initial high-level feedback for a design that implements a fairly complicated new feature? -- chandan
