Inserting a new merge commit between a commit and all its descendants?

[kwiberg]

🖖,

I have a graph like this,

               D--E
             /
main--A--B--C--F--G
             \
               H--I

and then someone adds new commits to main,

            D--E
           /
M0--A--B--C--F--G
  \        \
   \        H--I
    \
     M1--M2--main

and I’d like to merge the new stuff in a new commit X after C, and have all C’s descendants rebased on top of the new merge:

                      D--E
                     /
M0--A--B--C---------X--F--G
  \                / \
   \              /   H--I
    \            /
     M1--M2--main

I know this can be done with something like

jj new C main
jj rebase -s D -d X
jj rebase -s F -d X
jj rebase -s H -d X

but this is a lot of manual work, and seem hard to script. jj new -A C -A main almost works, except that it also tries to rebase everything that’s currently built on main. It seems to me that jj new -A C main accurately describes what I want to do, but jj new doesn’t allow a mix of -A and non--A parents of the new commit.

Is there a good way to do this that I’m not seeing?

[jennings]

You can probably do it in fewer commands:

# Insert the new commit between C and its descendants
jj new -A C

# Now we're editing this new commit
# Prepend `main` as an additional parent.
jj rebase -s @ -d main -d all:@-

I don't know if there's a way to do this in one command.

[kwiberg]

Ooohh, jj rebase -s A -d all:A- -d B, is a fully general way to take any commit A and merge B into it? That’s very powerful, especially in combination with jj new’s various ways to squeeze in a new, empty commit.

Hmm. I guess jj rebase -s A -d ... -d ... can be thought of as simply a way to edit the set of parents of A, followed by a replay of that commit and all its descendants onto the automatic merge of the new set of parents.

[kwiberg]

I tried this out, and it appears to have had just the effect I wanted. 🙏

[martinvonz]

If I understand correctly, you could even do just jj new -A C -A main.

[kwiberg]

Thanks, but like I said, that would do what I want, plus rebase every other descendant of main onto X, which isn’t acceptable. (In fact, the error I got when I tried exactly this was something along the lines of not being allowed to rebase immutable changes—probably other remote branches that are based on main.)

The crucial thing is that the new merge commit X needs to be inserted asymmetrically in the graph—the children of C should be rebased to become children of X instead, but the children of main should remain children of main. @jennings’s suggestion looks like it does this (I haven’t actually had time to verify it), and so should jj new -A C main if it were supported.

[kwiberg]

Ah, I realize that my graphs don’t show that main has other descendants too. I guess that was so obvious that I didn’t even bother drawing it... 🤓

[martinvonz]

Yes, sorry, I just looked at the graphs and didn't read the description carefully.