FR: jj backout should apply working copy commit as descendant of the revert commit

[fleimgruber]

This originated from a conversation on discord

Is your feature request related to a problem? Please describe.
As a user, if I jj backout -r A from an empty working copy revision, the "revert" commit A' is made as a descendant of the working copy revision. I expected the working copy revision to be the descendant of the revert commit A'.

As a user, currently I need to jj rebase -r @ -d A' to achieve the expected result.

Describe the solution you'd like
The working copy revision is the descendant of the revert commit after a jj backout.

Describe alternatives you've considered

Additional context
From:

A-B-@(empty)

I expected

A-B-A'-@(empty)

but got

A-B-@(empty)-A'

[yuja]

fwiw, rebase avoids this problem by requiring -d/--destination flag. I agree the current default isn't useful, but I don't know if it's good idea to add special case for the empty @.

#1188 is somewhat related, which suggests --insert-before.

[ilyagr]

I wouldn't mind making --destination mandatory as Yuya suggested. I use backout very rarely, so having it be verbose is probably fine.

I can think of a few other options:

1. We could have jj backout -r A result in A-B-A'-@ even if @ is not empty. This could be generalized to Yuya's second suggestion above: we could introduce jj backout --insert-before and make the default jj backout --insert-before @. Or, we could just have it be the default behavior without creating a new flag.

2. We could change the default of -d to @- instead of @. Then, you'd get:

A-B-A'
   \-@

This would be weird if @ is a merge commit. I'd be fine with jj backout erroring out in this case, suggesting an explicit -d.

3. We could change the default of -d to @- and add a --checkout or --no-checkout option to jj backout (not sure what would be the default; probably --checkout). Then, jj backout --no-checkout would work as in point 2. above. jj backout --checkout would result in:

A-B-A'-@(empty)
   \-X

where X is the old working-copy, if it was non-empty.

[ilyagr]

In fact, jj backout --insert-before @ is no better than jj backout --destination @- --checkout if @ is a merge commit. So, I think I prefer either making -d mandatory or one of the last two options in my previous comment. -d would still be required if @ is a merge commit.

We can also make it mandatory now and then consider implementing something better later.