On Wed, Apr 07, 2021 at 01:21:28PM +0200, Thorsten Leemhuis wrote: > > On 07.04.21 12:00, Greg KH wrote: > > On Wed, Apr 07, 2021 at 11:21:56AM +0200, Thorsten Leemhuis wrote: > >> Make people CC the recently created mailing list dedicated to Linux > >> kernel regressions when reporting one. Some paragraphs had to be > >> reshuffled and slightly rewritten during the process, as the text > >> otherwise would have gotten unnecessarily hard to follow. > >> > >> The new text also makes reporters include a line useful for automatic > >> regression tracking solution which does not exist yet, but is planned. > >> The term "#regzb" (short for regression bot) is inspired by the "#syz" > >> which can be used to communicate with syszbot (see > >> https://github.com/google/syzkaller/blob/master/docs/syzbot.md). > > > > While I understand the wish to automate things like this, the #syz > > marking will actually cause something to go off and do some work, and is > > only relevant for a very small number of developers, all of whom know to > > look up the instructions before doing so. But the #regzb marking will > > be requested to be added by random users who never have submitted a > > problem report before, OR from long-time kernel developers who are lucky > > to ever remember to read the documentation as they "know" how to do > > this. > > > > So this increased workload by people on the two ends of experience is > > going to be rough, I predict a very low rate of adoption :( > > Yup, I'm aware of that. And also well aware that I will need to keep an > eye on things and jump in and reply with mails to add such tags every > time they are missing. > But I think that direction it the best shot, as tying putting all the > burden on one person (me) is likely to fail, as our history with > regression tracking showed. And I think such tags with some bot in the > background > (as outlined roughly in > https://linux-regtracking.leemhuis.info/post/hello-world/ ) have at > least the best chance, as things are not out-of-band like tracking them > in bugzilla would be – or do you think that would be a better approach > together with its email-interface? Ok, we can try it, I'm not going to say it's not going to work, just that it's going to be a marketing effort to tell people how to do this :) > > What is the tag going to be good for? The reports will need to be > > handled by a person anyway and classified and tracked out-of-band from > > the list somehow. Will a tag do all that much here? > > I think is has, as then a good regression report will make the > still-to-be-written regression-bot create and entry that links to the > report and send a reply with a unique ID; then that ID needs to end up > in the commit that fixes the regression later (similar to how the IDs > for issues found by syzbot are mentioned there, which afaics works quite > well for people) and the regression-bot can close the entry automatically. Ah, ok, that makes more sense, nevermind, no objection from me, sorry for the noise. thanks, greg k-h
