3 ms·

> Some people get it in their heads that their commit history needs to be "clean". It doesn't... If by "clean" you mean "readable" (in the literate-programming

by derefr 2y ago

> Some people get it in their heads that their commit history needs to be "clean". It doesn't... If by "clean" you mean "readable" (in the literate-programming sense), then I'd be a strong advocate of that. I would like to one day see a repo that is maintained in such a way that, if you just dump out the commits in chronological order (e.g. `git log --patch --reverse --pretty=format:"%s"`), the result is essentially a jupyter notebook that explains the system being built as it builds it. Such a thing would require a lot of rebasing, of course. --- But of course, actually looking at giimly's commit history (https://github.com/gliimly/gliimly/commits/main/ https://github.com/gliimly/gliimly/commits/main/), you don't mean "readable." You mean "says nothing." I think this repo is clearly the result of "development in private", but I don't think the commit messages here were "cleaned up." I think this is just this person's way of using git in a single-user context — they just want to incrementally save WIPs, so they number them. (Maybe they even use an editor plugin that does this for them on save, or through a keyboard-shortcut.) That being said, I do sometimes also see this in the cases of private repos that were so messy and fraught, with long-lived now unmergeable feature branches, that the best way to integrate everything was to just convert each long-lived branch into a set of .patch files; manually prune and edit those .patch files in a text editor; and then `git am` each patchset into a synthetic commit. Those commits don't really have coherent purposes — they're just incremental steps of a single long integration — so you may as well just number them. (Usually, though, people rebase-squash these all once they're done.)

Alupis 2y ago
> If by "clean" you mean "readable" (in the literate-programming sense), then I'd be a strong advocate of that I don't disagree. However the commit messages should have some meaning. Looking at the repo closer, it appears the commit messages are release numbers: Commit 91: https://github.com/gliimly/gliimly/commit/956f8001890d85a4d39dda932ea95c01f220c531 https://github.com/gliimly/gliimly/commit/956f8001890d85a4d3... Release 91: https://gliimly.blogspot.com/2024/11/gliimly-91-released.html https://gliimly.blogspot.com/2024/11/gliimly-91-released.htm... Which, if the case, is even more awful. Now to understand the change history, I need to consult/cross-reference some documentation page that might 404 in the future. Additionally, the releases have many un-related changes grouped together (as one would expect for a release, not a commit). Lately I've become fond of the squash-merge. All development is in a branch, and to merge you open a Pull Request (or equivalent on whatever platform you use). The PR gets squashed into one commit that references the original PR, which retains all of the commit history and discussion/review comments. Best of all worlds in my opinion, and frees developers to commit freely without consequence or care.
- aeontech 2y ago
  Just to comment on your last paragraph - I don't really understand people's obsession with having squashed commits on main. I may not have a local copy of the branch that was merged months ago, so I can't get the un-squashed history without jumping through hoops. It also makes tools like git-bisect much less useful. With my team, once you are done with your PR, you do a quick rebase pass over your branch and squash any remaining partial/in progress/fixup commits into a final set of meaningful logical commits - this set gets merged into mainline wholesale, without any further squashing. Each commit has the ticket id in the commit message, so you can find all related changes easily if ever needed. This gives us history that's both granular enough to be useful, and meaningful. (Granted, if your team is disciplined enough to only have very small and granular PRs that only deliver a small logical unit of change at a time, the distinction between approaches becomes moot... but in my experience most PRs tend to be a little bigger that that, and contain at least a couple of related but distinct sets of changes) And yeah, the commit history in this project is just hilarious.
  - lolinder 2y ago
    > With my team, once you are done with your PR, you do a quick rebase pass over your branch and squash any remaining partial/in progress/fixup commits into a final set of meaningful logical commits - this set gets merged ... A lack of this simple discipline is what causes people to push for squashing. If you squash every PR then your commits may be too coarse, but you don't end up with random "fix tests" commits everywhere. If it's not part of your company culture to do the cleanup then it's harder to get that rolling than to just impose a repo rule that only squash merges are allowed in.

by derefr 2y ago

Alupis 2y ago

> If by "clean" you mean "readable" (in the literate-programming sense), then I'd be a strong advocate of that I don't disagree. However the commit messages should have some meaning. Looking at the repo closer, it appears the commit messages are release numbers: Commit 91: https://github.com/gliimly/gliimly/commit/956f8001890d85a4d39dda932ea95c01f220c531 https://github.com/gliimly/gliimly/commit/956f8001890d85a4d3... Release 91: https://gliimly.blogspot.com/2024/11/gliimly-91-released.html https://gliimly.blogspot.com/2024/11/gliimly-91-released.htm... Which, if the case, is even more awful. Now to understand the change history, I need to consult/cross-reference some documentation page that might 404 in the future. Additionally, the releases have many un-related changes grouped together (as one would expect for a release, not a commit). Lately I've become fond of the squash-merge. All development is in a branch, and to merge you open a Pull Request (or equivalent on whatever platform you use). The PR gets squashed into one commit that references the original PR, which retains all of the commit history and discussion/review comments. Best of all worlds in my opinion, and frees developers to commit freely without consequence or care.

aeontech 2y ago
Just to comment on your last paragraph - I don't really understand people's obsession with having squashed commits on main. I may not have a local copy of the branch that was merged months ago, so I can't get the un-squashed history without jumping through hoops. It also makes tools like git-bisect much less useful. With my team, once you are done with your PR, you do a quick rebase pass over your branch and squash any remaining partial/in progress/fixup commits into a final set of meaningful logical commits - this set gets merged into mainline wholesale, without any further squashing. Each commit has the ticket id in the commit message, so you can find all related changes easily if ever needed. This gives us history that's both granular enough to be useful, and meaningful. (Granted, if your team is disciplined enough to only have very small and granular PRs that only deliver a small logical unit of change at a time, the distinction between approaches becomes moot... but in my experience most PRs tend to be a little bigger that that, and contain at least a couple of related but distinct sets of changes) And yeah, the commit history in this project is just hilarious.
- lolinder 2y ago
  > With my team, once you are done with your PR, you do a quick rebase pass over your branch and squash any remaining partial/in progress/fixup commits into a final set of meaningful logical commits - this set gets merged ... A lack of this simple discipline is what causes people to push for squashing. If you squash every PR then your commits may be too coarse, but you don't end up with random "fix tests" commits everywhere. If it's not part of your company culture to do the cleanup then it's harder to get that rolling than to just impose a repo rule that only squash merges are allowed in.