Re: Plasma 5.18 release post-mortem
Some of the features or large changes that were added to 5.18 but caused regressions made it in close to the cut-off, but many were committed months ago. In particular 417424, 416695, 416358 were not the result of rush jobs, but they got broken anyway. I don't necessarily disagree though, but if we want to take a firm stance on this, I think we need to branch much earlier. "Soft" feature freezes don't cut it. We should maybe branch two months before the release rather than one. Also, we can't branch Frameworks due to their inherently rolling nature. Some of the regressions were caused by issues in Frameworks (417351, 417127, 417511). What are we going to do about those? It's not feasible to ask people to stop committing potentially risky changes to frameworks near a Plasma release because Plasma isn't the only customer of Frameworks. Nate On 2020-02-14 01:58, Marco Martin wrote: On Thu, Feb 13, 2020 at 5:40 PM Nate Graham wrote: Plasma 5.18 was a pretty buggy release, and I'd like to start a discussion about how we think it happened and what we can do better next time. Here are some of the top bugs that our users are reporting: I think we wanted to put too much in this release: it's pretty buggy but also the one that came in with more new features since quite a while. which in retrospect wasn't that good for an lts -- Marco Martin
Re: Plasma 5.18 release post-mortem
On Thu, Feb 13, 2020 at 5:40 PM Nate Graham wrote: > > Plasma 5.18 was a pretty buggy release, and I'd like to start a > discussion about how we think it happened and what we can do better next > time. Here are some of the top bugs that our users are reporting: I think we wanted to put too much in this release: it's pretty buggy but also the one that came in with more new features since quite a while. which in retrospect wasn't that good for an lts -- Marco Martin
Re: Plasma 5.18 release post-mortem
On 2020-02-13 11:55, Vlad Zahorodnii wrote: On 2/13/20 8:11 PM, David Edmundson wrote: I'm also seeing a rising amount of pushing without review on the core repos. I would like for us all to (nicely) call that out if we see any instances. Reviews are super important, the best time to fix a bug is before it even happens. Even for small commits and "safe" commits. Yes, ideally each commit must be reviewed by somebody. But what if one doesn't get _any_ feedback from code reviewers for weeks or even worse for months? What one should in that case? Speaking for myself, I manage such cases by sending private messages asking to do a code review and I feel very bad after doing this because I know people whom I talk to are are busy with their own stuff and they don't really want to deal with "problems." I think we first need to understand why people are pushing without any code review. Is it just because of desperation? or is it just because of not caring? Personally, I only ever commit without review to fix very small UI glitches or correct bad strings, and even then only when I'm 100% certain that the commit is good because the same thing was approved elsewhere or it is self-evidently, uncontroversially correct. For example, https://phabricator.kde.org/R120:2a0b1f570b3a15f8c12267889e46d9dabacb4b6d and https://phabricator.kde.org/R236:2157e258721be0c81477b2e305e6d697cafb597d. The flip side of asking for everything to go through review is that reviewers need to be on top of their review queues, or else even trivial changes get frustratingly delayed for days or weeks for no real gain. But in general, I agree with the sentiment that everything should go through review, and especially anything anything that's remotely complicated or that's used by downstream code. Nate
Re: Plasma 5.18 release post-mortem
On Thu, Feb 13, 2020 at 6:55 PM Vlad Zahorodnii wrote: > > On 2/13/20 8:11 PM, David Edmundson wrote: > > I'm also seeing a rising amount of pushing without review on the core > > repos. I would like for us all to (nicely) call that out if we see > > any instances. Reviews are super important, the best time to fix a bug > > is before it even happens. Even for small commits and "safe" commits. > > Yes, ideally each commit must be reviewed by somebody. But what if one > doesn't get _any_ feedback from code reviewers for weeks or even worse > for months? What one should in that case? Speaking for myself, I manage > such cases by sending private messages asking to do a code review and I > feel very bad after doing this because I know people whom I talk to are > are busy with their own stuff and they don't really want to deal with > "problems." > > I think we first need to understand why people are pushing without any > code review. Is it just because of desperation? or is it just because of > not caring? There's a huge difference between pushing without a review being accepted and pushing without going via phab. Whilst your point is important, my comment was referring only to the latter. No comments doesn't mean it's not been read. I know if I make a mistake N people will comment, if I upload something that's fine the same people don't say anything, but they must be reading it. David
Re: Plasma 5.18 release post-mortem
On 2/13/20 8:11 PM, David Edmundson wrote: I'm also seeing a rising amount of pushing without review on the core repos. I would like for us all to (nicely) call that out if we see any instances. Reviews are super important, the best time to fix a bug is before it even happens. Even for small commits and "safe" commits. Yes, ideally each commit must be reviewed by somebody. But what if one doesn't get _any_ feedback from code reviewers for weeks or even worse for months? What one should in that case? Speaking for myself, I manage such cases by sending private messages asking to do a code review and I feel very bad after doing this because I know people whom I talk to are are busy with their own stuff and they don't really want to deal with "problems." I think we first need to understand why people are pushing without any code review. Is it just because of desperation? or is it just because of not caring? Cheers, Vlad
Re: Plasma 5.18 release post-mortem
The important part right now is that we get on top with fixing them. For 5.18 there was /wy/ too much feature pushing for this release. Even if the new stuff itself doesn't introduce breakages, it takes developer and review time away from what should be a month of being totally on top of bugzilla. It's better to have a minor bug that you know, than uncertainty of new code. I don't know if it's because of the LTS or because of proximity to Christmas. Our next LTS should be better as we'll be doing any big work on Plasma6 during the prep time. It can have a 4 month feature freeze. I'm also seeing a rising amount of pushing without review on the core repos. I would like for us all to (nicely) call that out if we see any instances. Reviews are super important, the best time to fix a bug is before it even happens. Even for small commits and "safe" commits. -- Other positive steps, can everyone please go to bugzilla -> preferences -> saved searches and add "Plasma5-All-Critical" to their footer. Saved searches are a good thing to make use of, so please share any other useful ones, and help keep these ones up to date with changing plasma products. David