In the spirit of testing the "there are no dumb questions" theory, why does Github's squash-and-merge strategy update the committer after merging a pull request?
That is, suppose I author a pull request in my fork of a project. Suppose it has two commits in it. Suppose I ask the project maintainer to now merge my pull request to master of the upstream project.
Suppose she chooses the squash-and-merge strategy so that there will be one commit at the end.
Suppose further that the merge can be accomplished via a fast-forward, i.e. as simple as possible a case as I can think of. (Fast-forward merges should just update the branch pointer, which seems to me like it would leave the commit completely unchanged.)
Why is it, then, that the latest commit on master after the squash and merge operation features me as the author (I understand that part) but the maintainer as the committer? Isn't that an alteration of a commit, and doesn't squashing and fast-forward merging ensure that commits are not altered?
Is Github silently doing the equivalent of something like git amend under the covers?
Isn't that an alteration of a commit
More than an alteration: a creation of a new commit, which reflect the changes of your two commits.
And your two commits are no longer referenced by the main repo: it references only the squashed commit.
That new commit has still you as an author, but is committed by the maintainer.
Related
I'm working on feature branches that can have over a hundred commits over the course of a week or two. I've expected to get daily marks on my contribution graph for these daily commits once my pull request is accepted, but when that pull request gets "Squashed and merged", it looks like it just shows up as a single commit on the contribution graph.
Is this the intended behavior, or are my commits missing for a different reason?
Yes, this is the intended behavior. You only got one commit merged. That's the point of squash and merge.
I do not recommend squash merges. It takes all the carefully considered commit history about how the branch was developed and mashes it into a convenient but homogeneous paste.
I'm working on feature branches that can have over a hundred commits over the course of a week or two.
I would say this is the real problem. Feature branches should not regularly have over a hundred commits. Your features are too
big, or you're making a lot of commits to fix previous commits, or you have a lot of incidental merge commits from updating your branch, or all of the above.
Update your branch using rebase instead of merge. git rebase main rewrites your branch as if it were written on top of the latest version of main all along. This removes merge commits which say nothing but "I updated my branch" and makes it easier to see what work has been done on the branch. You can change git pull to do a rebase instead of a merge with git pull --rebase and by default by setting pull.rebase. See git-config and Git Branching - Rebasing.
Break up large features into a series of smaller features. This is a skill beyond the scope of this answer, but often a big change can be broken down into a series of smaller refactorings.
Eliminate incidental fixup commits. If you need to fix something in the previous commit, instead of making a new commit "amend" your changes to the previous one with git commit --amend.
For a general cleanup, do an interactive rebase and squash your incidental commits individually. Use fixup.
$ git rebase -i master
pick abcd1234 feature: do that thing
fixup 1234dcba whoops, had a typo in that thing
fixup b33fdead docs: document that thing
fixup afdcefff test: test that thing
fixup deadb33f fix that thing
pick efga1389 feature: did another thing
You had six commits, but you only really did two things. Now you will have just two commits for doing two things.
Using these techniques you can bring your PRs down to a handful of well thought out commits, each conveying important information to the reviewer and future developers wondering why the code was written that way.
See Rewriting History for more.
Yes, they show up as a single commit. That's because GitHub only counts commits that end up on the default branch, and when you squash a PR, you end up with one commit on the default branch.
I personally don't worry about my contribution graph, so this doesn't matter to me. However, I also don't use squash and merge, since I write nice, bisectable commits with good commit messages, and squash and merge destroys that. If it's important to you not to reduce the work that a contributor has made down to a single commit, then you would need a merge strategy whose goal isn't explicitly to destroy history.
Our team is deciding on what merge strategy to use in our repo. We're looking at squash merges for the ability to have a clean commit history. However, we want to be able to review intermediate changes if we want more detail on a particular commit - in particular, the intermediate commits on the PR that were squashed.
Normally, squashing and deleting the branch will remove all references to those intermediate commits in the reflog and git garbage collection could clean those up. However, in looking at other repos, it seems like GitHub still retains those intermediate commits.
What we're looking for is confirmation that GitHub is retaining those commit IDs in the reflog and that we're guaranteed they'll stick around. Alternatively, information telling us that's not guaranteed is also good.
Apparently it is guaranteed behavior. This document shows that you can check out pull requests locally. This will have the branch's commit history.
I made a pull request which was merged to a project. It so happens that the code needs a minor modification. How can I possibly change this with no access to the master project whatsoever? Am I out of luck and I have to do a new PR?
Thanks.
Yes, you should submit another pull request.
When using any of Git's "rewriting" commands like commit --amend or rebase it is important that you avoid changing shared commits.
In its Rebasing chapter, the Git Book says:
The Perils of Rebasing
Ahh, but the bliss of rebasing isn’t without its drawbacks, which can be summed up in a single line:
Do not rebase commits that exist outside your repository.
If you follow that guideline, you’ll be fine. If you don’t, people will hate you, and you’ll be scorned by friends and family.
When you rebase stuff, you’re abandoning existing commits and creating new ones that are similar but different. If you push commits somewhere and others pull them down and base work on them, and then you rewrite those commits with git rebase and push them up again, your collaborators will have to re-merge their work and things will get messy when you try to pull their work back into yours.
So, I'm new to Mercurial, and on my first try I screwed up.
Before pushing my data, I tried to merge with one branch (which I think was an old branch)
Instead of pushing it properly to change from "draft" to "public" I manually changed it from Tortoise by using the change phase from "draft" to "public"
I thought it meant that it would be placed in the repository, but when I went to the actual repository (on Bit Bucket) I couldn't find it.
So what happened, is I think I never made a push.
This is causing me an issue to make new changes, because when I tried committing other files, merging it with other branches and pushing it to the repository, i get the error:
searching for changes
remote has heads on branch 'branch one' that are not known locally: 62bf86f93d4f
abort: push creates new remote head 7edef5d2e2e5 on branch 'branch one'!
hint: pull and merge or see "hg help push" for details about pushing new heads
[command returned code 255 Thu May 27 14:43:44 2015]
So I think the solution would be to merge with the original draft (which is now public)
Thus my question: Is it possible to still merge and push a revision that was already deemed "public", even if I can't find it in the repository on bitbucket?
The phase of your commit (public vs. draft vs. secret) has no relevance for whether the commit can be merged. The difference between draft and public only affects whether the history involving the commit can be altered (e.g. via hg commit --amend or hg rebase). Public revisions are considered immutable, draft revisions can still be altered.
You can still merge and push. In theory, if you are positive that you didn't push the commit, you can also safely change the phase back to draft. However, that would be pointless [1], because pushing the merge will immediately make it public again [2].
[1] It can be useful if you use hg pull --rebase, but if you don't know what that does, you probably shouldn't use it.
[2] Unless you push to a non-publishing repository, but that's unlikely unless your group is using the evolve extension. If you don't know what a non-publishing repository or the evolve extension is, you don't need to know.
I forked a repo from GitHub and make a bunch of changes. Then I found one of my modification in one file can be a patch to an issue of the original repo, but the author don't want to merge my other modifications, so I don't want to send a pull request directly. And I think forking it again and just modify that file to make a patch and then send pull request seems not so elegant. Are there any "standard" way to do that?
In my opinion, you should make a new branch with the same root, then use cherry-pick to add every commit you made except those who are not accepted by the author.
Then send a pull request on this branch.
Moreover, if you want to regroup all your modifications into one simple commit, you may use a squash rebase on a local branch before pushing it online.
Write down the SHA-1 hash of the desired commit. Switch back to the original project's master branch, create a new branch off of it and cherry-pick that one commit onto the new branch.
You can then push the new branch to GitHub and send a PR with its changes.
If you have modified multiple files within one commit, you'll have to rewrite the commit somehow.