On Thu, 04 May 2023 17:36:11 +0200, Oswald Buddenhagen wrote: > > On Thu, May 04, 2023 at 04:49:58PM +0200, Takashi Iwai wrote: > > Sorry, that doesn't work. The review is done upon the patch, and if > > this patch can't be reviewed easily, it's simply no-go. > > > that's a self-imposed limitation. And that's how the process works. Over decades. > it's beyond me why in 2023 anyone working on a bigger project is still > using a patch-based review process, given the existence of proper > review tools like gerrit (and crutches like github and gitlab). All those git-based work flows are based on commits, and commits consist of patches. So, reviewing each commit is nothing but reviewing a patch. IOW, if you do a PR via github, it'll hit the very same problem; when the commit is not understandable enough for reviewers, it'll be NAK'ed. > i always view patches with 10 lines of context, and regularly use the > "expand context" widgets to get even more into view. > in the small projects i maintain i apply more complex patches first > and view them with -U10 or more. > > "i don't see enough to judge this" isn't a complaint that would ever > occur to me leveling at a contributor. It's not only about contexts. It's just not clear what your patch does as partial revert and as fix. I really would like to see one change for one fix, and one change for one code refactoring. It's difficult to achieve if we have to partially revert and partially fix in a shot. > > Again, the problem is the mixture; it partially reverts to the > > original code while it modifies some part in some other way. > > > the baseline is irrelevant - it was already broken, and almost > impossible to reason about. Reverting temporarily to the original state (even if it's the broken state) is the very standard way. It's a clear way to start from the scratch and build up things again more cleanly. And, if the complain were only mine, I'd reconsider. But it's not one person's view, but multiple reviewers think so, so it's a sign of no-go. > > By reverting the whole and reapplying fixes -- although it'll need > > more steps -- we can see more clearly what change fixes which part. > > > that's not what actually happens. > this is all deeply intertwined code. Err, I can't follow; in the previous patch and this patch, you added more comments, use different terms and variable names, and use different way to calculate the hw_avail value, etc -- which are basically irrelevant from the behavior fix itself, but they are just code refactoring. Those could be separated easily. > splitting up the patch will > merely give you the *illusion* of better understanding the pieces. but > to actually make sense of it, you need to see the whole, in its end > state, because there are no fully functional intermediate states. Again, I can't follow your logic. Why splitting into small pieces can't lead to a better understanding *at all*? Why you must refuse it? Certainly one needs to take a look at big picture. But it's a different story. > the original patch was three patches at first. when i was attempting > to write proper commit messages explaining what fixes what, i found > that it's just impossible to untangle it without lying by omission. so > i squashed them and wrote a cumulative description. and you accepted > it. The acceptance of your patch was my failure, yeah. I should have rejected it. So this failure doesn't happen again. You're seeing the result now. Takashi