Update Git branches from master
git rebase master
is the proper way to do this. Merging would mean a commit would be created for the merge, while rebasing would not.
You have two options:
The first is a merge, but this creates an extra commit for the merge.
Checkout each branch:
git checkout b1
Then merge:
git merge origin/master
Then push:
git push origin b1
Alternatively, you can do a rebase:
git fetch
git rebase origin/master
You have basically two options:
You merge. That is actually quite simple, and a perfectly local operation:
git checkout b1 git merge master # repeat for b2 and b3
This leaves the history exactly as it happened: You forked from master, you made changes to all branches, and finally you incorporated the changes from master into all three branches.
git
can handle this situation really well, it is designed for merges happening in all directions, at the same time. You can trust it be able to get all threads together correctly. It simply does not care whether branchb1
mergesmaster
, ormaster
mergesb1
, the merge commit looks all the same to git. The only difference is, which branch ends up pointing to this merge commit.You rebase. People with an SVN, or similar background find this more intuitive. The commands are analogue to the merge case:
git checkout b1 git rebase master # repeat for b2 and b3
People like this approach because it retains a linear history in all branches. However, this linear history is a lie, and you should be aware that it is. Consider this commit graph:
A --- B --- C --- D <-- master \ \-- E --- F --- G <-- b1
The merge results in the true history:
A --- B --- C --- D <-- master \ \ \-- E --- F --- G +-- H <-- b1
The rebase, however, gives you this history:
A --- B --- C --- D <-- master \ \-- E' --- F' --- G' <-- b1
The point is, that the commits
E'
,F'
, andG'
never truly existed, and have likely never been tested. They may not even compile. It is actually quite easy to create nonsensical commits via a rebase, especially when the changes inmaster
are important to the development inb1
.The consequence of this may be, that you can't distinguish which of the three commits
E
,F
, andG
actually introduced a regression, diminishing the value ofgit bisect
.I am not saying that you shouldn't use
git rebase
. It has its uses. But whenever you do use it, you need to be aware of the fact that you are lying about history. And you should at least compile test the new commits.