Junio C Hamano venit, vidit, dixit 08.08.2016 19:42: > Duy Nguyen <pclouds@xxxxxxxxx> writes: > >> git-notes was mentioned in this thread back in 2015, but I think it's >> discarded because of the argument that's part of the cover letter was >> not meant to be kept permanently. > > I do not think the reason why we didn't think the notes mechanism > was a good match was mainly because the cover letter material was > about a branch as a whole, which does not have a good counter-part > in Git at the conceptual level. "A branch is just a moving pointer > that points at one commit that happens to be at the tip" is not a > perfect match to "I am holding these N patches to achieve X and I am > constantly adding, rewinding and rebuilding". The notes mechanism > gives an easy way to describe the former (i.e. annotate one commit, > and let various commands to move that notes as you rewind and > rebuild) but not the latter (i.e. "branch.description" configuration > is the best match, but that is just a check-box feature and does not > make any serious attempt to be part of a version-control system). > >> But I think we can still use it as a >> local/temporary place for cover letter instead of the empty commit at >> the topic's tip. It is a mark of the beginning of commit, it does not >> require rewriting history when you update the cover letter, and >> git-merge can be taught to pick it up when you're ready to set it in >> stone. > > That depends on what you exactly mean by "the beginning of". Do you > mean the first commit that is on the topic? Then that still requires > you to move it around when the topic is rebuilt. If you mean the > commit on the mainline the topic forks from, then of course that > would not work, as you can fork multiple topics at the same commit. Well, my idea back then was: attach notes to refs rather than commits, in this case to the branch ref. That would solve both the "branch moves" as well as the "cover letter refers to the whole branch/topic" issues. In fact, I had an implementation that I had been rebasing and using for quite some time, but it never became popular, and branch.description landed in-tree. [short version: notes attached to (virtual) refname objects (virtual blobs - not stored, but "existing" for (fsck, notes prune and the like)] The notes based approach suffered from the old notes deficiency: we don't have good simple tooling for sharing notes; really, we don't have good tooling for dealing with any remote refs besides branches (read: ref namespace reorg project is stalled), unless you set up specific refspecs. OTOH, branch.description is inherently local, too, and can't even be transported after setting up refspecs or such. Also, notes trees have a history, so you would gain a log on your cover letter edits; again, our tooling around that notes feature is sub-optimal, that is: the feature is there, the ui could improve. Michael -- To unsubscribe from this list: send the line "unsubscribe git" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html