"push creates new remote head" issue

Question

I read all SO questions about this issue, and I still can't resolve it.

I am using TortoiseHg. I worked on a side-branch, and now I want to merge it back to the main branch. I pulled all changes made in both branches, updated to the main branch, and merged (and committed). But still when I try to push all this, I get the "abort:push creates new remote head" message.

I also tried (as was suggested in one of the questions in SO) to close the branch using the --close-branch option.

The only thing I did not try is to 'force' push. Any suggestions? Or is force-pushing the only option?

Solution 1

I just tried a similar setup, and I get the same warning. Apparently, although the second head you are trying to push is closed, it is seen as another head during the push. And closing both heads does not seem to be pushable either.

You can force the push, it should be ok, but you could eventually get the same issue if you keep multiple heads on your visualization branch, like you already have with changesets 14 and 20. To solve the issue once and for all, I would instead suggest to merge both changesets (14 and 20) and reclose the final head.

Solution 2

Just for everyone else that runs into this problem. What caused this problem for me were some local revisions on the default branch that I didn't push before I started to work on a new branch.

I had merged the latest revision I pulled for the default branch with my new branch, but these leaves your local changes to the default branch committed but un-pushed.

If you try to push them, it's not your new branch that is creating a remote head, it's the un-pushed revisions to the default branch that is creating a remote head.

When I stripped out those revision with hg strip -r 1234

hg push --new-branch 

went perfect.

What put me on the right track was

hg heads

With showed I had two heads that both had the name of the default branch with different revision numbers.

Solution 3

Thanks for the answers, I definitely learned some new tricks.

What I ended up doing, is cloning an early revision from the remote repository, that is, a repository that doesn't have all the commits of my merges etc. I then pulled the change-sets, merged, and committed. Then the push finally succeeded.

It was basically the same steps I tried to do before, but apparently on the first (unsuccessful) trial I broke it down to more steps than were needed, and something went wrong at some point.

Comments

[Itamar Katz]

Merging 14 and 20, that's what I did on 21, and it still gave the same message.

[Vince]

No, you did not. 21 is a merge of visualization in default, not visualization in visualization. When I meant merge 14 and 20, it wasn't for a descendant of 14 and changeset 20...

[Itamar Katz]

Ok, so merging the two heads of visualization into default still leaves me with two heads in that branch. Now it makes sense... Thanks.

[matt wilkie]

Thank you for "it's unpushed revs to default branch creating remote head"!! For me the fix was to a) open the destination repository (on a local file share) in TortoiseHg Workbench, b) Update to the latest revision there, c) Pull from my local repository 1 revision newer than the server rev, d) close server repo and finally push all changes from local repository.

[Brett Slocum]

hg strip doesn't exist anymore that I can tell. What would the new command be?