Message ID | f339d15842370b97558b704593848e318462b68d.1683878120.git.mail@cbaines.net |
---|---|
State | New |
Headers | show |
Series | [bug#63459] doc: Rewrite the branching strategy. | expand |
Christopher Baines <mail@cbaines.net> writes: > Move away from using staging and core-updates, and make the strategy > independant of branch names. > > Keep the 300 dependent threshold for changes to master, as I don't have any > specific reason to change this. > > Most importantly, require using guix-patches issues to coordinate merging of > the branches, as I think that'll address the key issues that have shown up > recently where it's been unclear which branch should be merged next. > > * doc/contributing.texi (Submitting Patches): Rewrite branching strategy. > --- > doc/contributing.texi | 58 +++++++++++++++++-------------------------- > 1 file changed, 23 insertions(+), 35 deletions(-) Following on from the discussion recently about moving away from staging and core-updates, I've sent a patch to update the branching strategy. The key thing is obviously to just remove mentions of staging and core-updates, making the guidance more generic. However, I've also added some requirements to use guix-patches issues to track the intentions to merge branches, as I think that'll help address some of the issues that came up recently with uncertainty around which branch will be merged next. I'm also hoping that these issues then can be used to automate the QA process, triggering the qa-frontpage to automatically start building the relevant branches. Thanks, Chris
Hello, Thanks for this initiative! Christopher Baines <mail@cbaines.net> skribis: > Move away from using staging and core-updates, and make the strategy > independant of branch names. > > Keep the 300 dependent threshold for changes to master, as I don't have any > specific reason to change this. > > Most importantly, require using guix-patches issues to coordinate merging of > the branches, as I think that'll address the key issues that have shown up > recently where it's been unclear which branch should be merged next. > > * doc/contributing.texi (Submitting Patches): Rewrite branching strategy. [...] > +Changes to packages with 300 dependent packages or less can be pushed to > +the @code{master} branch. > + > +Larger changes should be first pushed to a branch other than > +@code{master}. This allows for testing and for the build farms to > +process the changes prior to being pushed to the @code{master} branch. I’d be more specific: Larger changes should first be pushed to a topic branch other than @code{master}; the set of changes should be consistent---e.g., ``GNOME update'', ``NumPy update'', etc. This allows for testing: the branch will automatically show up at @indicateurl{https://qa.guix.gnu.org/branch/@var{branch}}, with an indication of its build status on various platforms. “Automatic” is a bit of an overstatement; that sentence probably needs to be tweaked. :-) But I think it’s good to link to the QA platform to make things more concrete. > +To help coordinate the merging of branches, you must create a new > +guix-patches issue each time you wish to merge a branch. These issues ^ + (@pxref{Tracking Bugs and Patches}) > +indicate the order in which the branches should be merged, so take a > +look at the open issues for merging branches and mark the issue you > +create as blocked by the issue previously at the back of the queue. s/blocked/@dfn{blocked}/ Perhaps add a footnote or paren stating how to “block” an issue in Debbugs? > +Normally branches will be merged in a ``first come, first merged'' > +manor, tracked through the guix-patches issues. If you agree a different s/manor/manner/ s/agree a/agree on a/ > +order with those involved, you can track this by updating which issues > +block which other issues. Therefore, to know which branch is at the > +front of the queue, look for the issue which isn't blocked by any other > +branch merges. > + > +Once a branch is at the front of the queue, wait until sufficient time > +has passed for the build farms to have processed the changes, and for > +the necessary testing to have happened. This is a bit technical. How can I know “which branch is at the front of the queue”? Even as a seasoned Debbugs users, I’m not sure what I’m supposed to do here. Do you think we could provide ready to use commands (debbugs.el or ‘mumi’) or at least a sequence of steps to follow? Last but not least: two spaces after end-of-sentence period please. :-) This is mostly about tweaking words; I think this is a great step forward, very much in line with what was discussed in February at the Guix Days. Thank you! Ludo’.
Ludovic Courtès <ludo@gnu.org> writes: > Hello, > > Thanks for this initiative! > > Christopher Baines <mail@cbaines.net> skribis: > >> Move away from using staging and core-updates, and make the strategy >> independant of branch names. >> >> Keep the 300 dependent threshold for changes to master, as I don't have any >> specific reason to change this. >> >> Most importantly, require using guix-patches issues to coordinate merging of >> the branches, as I think that'll address the key issues that have shown up >> recently where it's been unclear which branch should be merged next. >> >> * doc/contributing.texi (Submitting Patches): Rewrite branching strategy. > > [...] > >> +Changes to packages with 300 dependent packages or less can be pushed to >> +the @code{master} branch. >> + >> +Larger changes should be first pushed to a branch other than >> +@code{master}. This allows for testing and for the build farms to >> +process the changes prior to being pushed to the @code{master} branch. > > I’d be more specific: > > Larger changes should first be pushed to a topic branch other than > @code{master}; the set of changes should be consistent---e.g., ``GNOME > update'', ``NumPy update'', etc. This allows for testing: the branch > will automatically show up at > @indicateurl{https://qa.guix.gnu.org/branch/@var{branch}}, with an > indication of its build status on various platforms. > > “Automatic” is a bit of an overstatement; that sentence probably needs > to be tweaked. :-) But I think it’s good to link to the QA platform to > make things more concrete. That sounds fine to me. Everything apart from starting the builds is already automatic, and I want to automate that through the issues described here. >> +To help coordinate the merging of branches, you must create a new >> +guix-patches issue each time you wish to merge a branch. These issues > ^ > + (@pxref{Tracking Bugs and Patches}) > >> +indicate the order in which the branches should be merged, so take a >> +look at the open issues for merging branches and mark the issue you >> +create as blocked by the issue previously at the back of the queue. > > s/blocked/@dfn{blocked}/ > > Perhaps add a footnote or paren stating how to “block” an issue in > Debbugs? Yeah, I'll try and write something. >> +Normally branches will be merged in a ``first come, first merged'' >> +manor, tracked through the guix-patches issues. If you agree a different > > s/manor/manner/ > s/agree a/agree on a/ > >> +order with those involved, you can track this by updating which issues >> +block which other issues. Therefore, to know which branch is at the >> +front of the queue, look for the issue which isn't blocked by any other >> +branch merges. >> + >> +Once a branch is at the front of the queue, wait until sufficient time >> +has passed for the build farms to have processed the changes, and for >> +the necessary testing to have happened. > > This is a bit technical. How can I know “which branch is at the front > of the queue”? Even as a seasoned Debbugs users, I’m not sure what I’m > supposed to do here. Do you think we could provide ready to use > commands (debbugs.el or ‘mumi’) or at least a sequence of steps to > follow? So, I think there's two technical hurdles to overcome here. The first is identifying the issues for merging branches, maybe for that we can set out a format for the title of the bug, but I'm very open to suggestions. Any way of identifying the open issues should be usable through debbugs.el and mumi. The second hurdle is the queuing behaviour, which I think the blocking behaviour is a natural fit for. Maybe the tooling is lacking but I think that can be addressed. I want the qa-frontpage to display the queue of branches (and issues) in a clear way, as well as providing links to make changes (as it does for marking issues as moreinfo). > Last but not least: two spaces after end-of-sentence period please. :-) > > This is mostly about tweaking words; I think this is a great step > forward, very much in line with what was discussed in February at the > Guix Days. Thank you! Great, thanks for taking a look! Chris
Christopher Baines <mail@cbaines.net> writes: > Ludovic Courtès <ludo@gnu.org> writes: > >> Hello, >> >> Thanks for this initiative! >> >> Christopher Baines <mail@cbaines.net> skribis: >> >>> +order with those involved, you can track this by updating which issues >>> +block which other issues. Therefore, to know which branch is at the >>> +front of the queue, look for the issue which isn't blocked by any other >>> +branch merges. >>> + >>> +Once a branch is at the front of the queue, wait until sufficient time >>> +has passed for the build farms to have processed the changes, and for >>> +the necessary testing to have happened. >> >> This is a bit technical. How can I know “which branch is at the front >> of the queue”? Even as a seasoned Debbugs users, I’m not sure what I’m >> supposed to do here. Do you think we could provide ready to use >> commands (debbugs.el or ‘mumi’) or at least a sequence of steps to >> follow? > > So, I think there's two technical hurdles to overcome here. The first is > identifying the issues for merging branches, maybe for that we can set > out a format for the title of the bug, but I'm very open to > suggestions. Any way of identifying the open issues should be usable > through debbugs.el and mumi. > > The second hurdle is the queuing behaviour, which I think the blocking > behaviour is a natural fit for. Maybe the tooling is lacking but I think > that can be addressed. > > I want the qa-frontpage to display the queue of branches (and issues) in > a clear way, as well as providing links to make changes (as it does for > marking issues as moreinfo). I've sent a v2 now which makes more changes, most importantly it pulls the content out from the "Submitting Patches" section to it's own section, and also moves content from the Commit Policy in and references it. I've also made some progress with the qa-frontpage, it now shows a list of branches with the corresponding issues on the homepage.
Hey! The changes in #63459 have strayed now in to touching the commit policy [1]. My intent was to simplify the guidance by grouping it better, but I think the significant change here is that the commit policy now references the entire branching strategy, rather than just talking about sending patches for review. 1: https://guix.gnu.org/manual/devel/en/html_node/Commit-Access.html#Commit-Policy That new branching strategy makes some "should" requirements on sending patches for review and pushing to topic branches for larger changes. It also makes a "must" requirement on opening guix-patches issues to track and manage merging branches. I'd like to merge these changes next week since they've been up for a few weeks, so do comment if you have any thoughts or if you'd like more time to review them. Thanks, Chris
Hello Chris, thanks for taking up this issue! I agreed with Ludovic's comments, so things look good now for me. A very minor point: In the section on "trivial" changes, I would drop this sentence (which was already there before): "This is subject to being adjusted, allowing individuals to commit directly on non-controversial changes on parts they’re familiar with." The sentence is meaningless, as everything is all the time subject to being adjusted; and we do not have immediate plans to adjust it. Looking forward to the merge since it clarifies things and removes the staging and core-updates branches not only from our minds, but also the texts. Andreas
Andreas Enge <andreas@enge.fr> writes: > thanks for taking up this issue! I agreed with Ludovic's comments, so > things look good now for me. A very minor point: In the section on > "trivial" changes, I would drop this sentence (which was already there > before): > "This is subject to being adjusted, allowing individuals to commit directly > on non-controversial changes on parts they’re familiar with." > The sentence is meaningless, as everything is all the time subject to being > adjusted; and we do not have immediate plans to adjust it. My reading of this line is that "adjusted" is probably not the right word to use, but I think the intent here is to talk about how currently it's accepted that people can and will push non-controversial changes on parts they’re familiar with directly to master. I'm not sure if others read this similarly.
Am Sun, Jun 11, 2023 at 10:37:14AM +0100 schrieb Christopher Baines: > My reading of this line is that "adjusted" is probably not the right > word to use, but I think the intent here is to talk about how currently > it's accepted that people can and will push non-controversial changes on > parts they’re familiar with directly to master. I read it the other way round: Right now it is not accepted, but it might be adjusted to allow non-controversial changes in the future. Actually the concept of "non-controversial commits" is probably controversial in itself... Andreas
Hi Christopher, Christopher Baines <mail@cbaines.net> writes: > Andreas Enge <andreas@enge.fr> writes: > >> thanks for taking up this issue! I agreed with Ludovic's comments, so >> things look good now for me. A very minor point: In the section on >> "trivial" changes, I would drop this sentence (which was already there >> before): >> "This is subject to being adjusted, allowing individuals to commit directly >> on non-controversial changes on parts they’re familiar with." >> The sentence is meaningless, as everything is all the time subject to being >> adjusted; and we do not have immediate plans to adjust it. > > My reading of this line is that "adjusted" is probably not the right > word to use, but I think the intent here is to talk about how currently > it's accepted that people can and will push non-controversial changes on > parts they’re familiar with directly to master. > > I'm not sure if others read this similarly. That's how I read it as well. I like the ability for people to, at times depending on the situation, choose to push directly to fix or update something instead of going through the otherwise recommended 1 week QA/review flow.
I've now pushed this to master as 0ea096ae23fa81f05ce97e5e61c15647c0a475ec. There's still lots to improve, both within the guidance and in addition to it. Top on my list is making some requirements about the issues to open when you want to merge a branch (e.g. specifying the title format so that qa.guix.gnu.org can detect them). Thanks, Chris
diff --git a/doc/contributing.texi b/doc/contributing.texi index 7bf350ee0d..c54910e34d 100644 --- a/doc/contributing.texi +++ b/doc/contributing.texi @@ -1264,41 +1264,29 @@ Submitting Patches @c See <https://lists.gnu.org/archive/html/guix-devel/2016-10/msg00933.html>. @cindex branching strategy @cindex rebuild scheduling strategy -Depending on the number of dependent packages and thus the amount of -rebuilding induced, commits go to different branches, along these lines: - -@table @asis -@item 300 dependent packages or less -@code{master} branch (non-disruptive changes). - -@item between 300 and 1,800 dependent packages -@code{staging} branch (non-disruptive changes). This branch is intended -to be merged in @code{master} every 6 weeks or so. Topical changes -(e.g., an update of the GNOME stack) can instead go to a specific branch -(say, @code{gnome-updates}). This branch is not expected to be -buildable or usable until late in its development process. - -@item more than 1,800 dependent packages -@code{core-updates} branch (may include major and potentially disruptive -changes). This branch is intended to be merged in @code{master} every -6 months or so. This branch is not expected to be buildable or usable -until late in its development process. -@end table - -All these branches are @uref{https://@value{SUBSTITUTE-SERVER-1}, -tracked by our build farm} and merged into @code{master} once -everything has been successfully built. This allows us to fix issues -before they hit users, and to reduce the window during which pre-built -binaries are not available. - -When we decide to start building the @code{staging} or -@code{core-updates} branches, they will be forked and renamed with the -suffix @code{-frozen}, at which time only bug fixes may be pushed to the -frozen branches. The @code{core-updates} and @code{staging} branches -will remain open to accept patches for the next cycle. Please ask on -the mailing list or IRC if unsure where to place a patch. -@c TODO: It would be good with badges on the website that tracks these -@c branches. Or maybe even a status page. +Changes to packages with 300 dependent packages or less can be pushed to +the @code{master} branch. + +Larger changes should be first pushed to a branch other than +@code{master}. This allows for testing and for the build farms to +process the changes prior to being pushed to the @code{master} branch. + +To help coordinate the merging of branches, you must create a new +guix-patches issue each time you wish to merge a branch. These issues +indicate the order in which the branches should be merged, so take a +look at the open issues for merging branches and mark the issue you +create as blocked by the issue previously at the back of the queue. + +Normally branches will be merged in a ``first come, first merged'' +manor, tracked through the guix-patches issues. If you agree a different +order with those involved, you can track this by updating which issues +block which other issues. Therefore, to know which branch is at the +front of the queue, look for the issue which isn't blocked by any other +branch merges. + +Once a branch is at the front of the queue, wait until sufficient time +has passed for the build farms to have processed the changes, and for +the necessary testing to have happened. @item @cindex determinism, of build processes