I\'ve set up a remote non-bare \"main\" repo and cloned it to my computer. I made some local changes, updated my local repository, and pushed the changes back to my remote r
This was our solution for replacing master on a corporate gitHub repository while maintaining history.
push -f
to master on corporate repositories is often disabled to maintain branch history. This solution worked for us.
git fetch desiredOrigin
git checkout -b master desiredOrigin/master // get origin master
git checkout currentBranch // move to target branch
git merge -s ours master // merge using ours over master
// vim will open for the commit message
git checkout master // move to master
git merge currentBranch // merge resolved changes into master
push your branch to desiredOrigin
and create a PR
And if push --force
doesn't work you can do push --delete
. Look at 2nd line on this instance:
git reset --hard HEAD~3 # reset current branch to 3 commits ago
git push origin master --delete # do a very very bad bad thing
git push origin master # regular push
But beware...
In other words:
force
push on a public repository.pull
.reset
or rewrite
history in a repo someone might have already pulled.Of course there are exceptionally rare exceptions even to this rule, but in most cases it's not needed to do it and it will generate problems to everyone else.
And always be careful with what you push to a public repo. Reverting:
git revert -n HEAD~3..HEAD # prepare a new commit reverting last 3 commits
git commit -m "sorry - revert last 3 commits because I was not careful"
git push origin master # regular push
In effect, both origin HEADs (from the revert and from the evil reset) will contain the same files.
push --force
Another problem push --force
may bring is when someone push anything before you do, but after you've already fetched. If you push force your rebased version now you will replace work from others.
git push --force-with-lease
introduced in the git 1.8.5 (thanks to @VonC comment on the question) tries to address this specific issue. Basically, it will bring an error and not push if the remote was modified since your latest fetch.
This is good if you're really sure a push --force
is needed, but still want to prevent more problems. I'd go as far to say it should be the default push --force
behaviour. But it's still far from being an excuse to force a push
. People who fetched before your rebase will still have lots of troubles, which could be easily avoided if you had reverted instead.
And since we're talking about git --push
instances...
@linquize brought a good push force example on the comments: sensitive data. You've wrongly leaked data that shouldn't be pushed. If you're fast enough, you can "fix"*
it by forcing a push on top.
*
The data will still be on the remote unless you also do a garbage collect, or clean it somehow. There is also the obvious potential for it to be spread by others who'd fetched it already, but you get the idea.