Add --empty-parent and --empty-child to jj split

[samueltardieu]

Following the discussion in #1155, I tried an implementation of --empty-parent (-P) and --empty-child (-C) with jj split. They do not prompt for a description: the non-empty change keeps the original description and change id. The empty change gets an empty description and a new change id.

The fact that the non-empty commit keeps the change id when using --empty-parent differs from what is done with jj split nonexistent, but it seems more logical to me. I've done this in a separate commit as to be able to revert it easily if needed.

Checklist

If applicable:

• I have updated CHANGELOG.md
• I have updated the documentation (README.md, docs/, demos/)
• I have added tests to cover my changes

[ilyagr]

This approach (adding options to jj split) looks fine to me, but we should wait for @martinvonz and/or @yuja to comment.

I think there's an annoying change needed to which commit gets which change id. There's no reason to fix that unless we officially settle on this approach. Other than that, this looks good to me; I just have some minor take-it-or-leave-it comments.

[martinvonz]

My main concern with --empty-parent and --empty-child is that split is probably not the command I would reach for to insert a new commit in a stack. OTOH, if the user actually did look at jj help split and wanted to create an empty child or parent, it does seem better to have flags for that than to have them figure out the jj split . and jj split nonexistent tricks. So I'm not sure what's best. I also left a comment on #1155.

[yuja]

My main concern with --empty-parent and --empty-child is that split is probably not the command I would reach for to insert a new commit in a stack.

I feel the same. Functionality wise, these jj split options work for me. But maybe successors/predecessors chain would be different from what I expect for "new" commit insertion?

[ilyagr]

I'm approving this to get rid of the annoying red "changes requested" marker. Please don't merge it without an OK from others.

[ilyagr]

This needs a comma and has too many "after"s. "an empty parent before, or an empty child after..."

[samueltardieu]

Done

[samueltardieu]

This PR is suspended at the moment as it is less powerful than #1172 although less intimidating. Let's wait until #1172 is settled.

[samueltardieu]

I'll close this one, for the time being jj new -B and jj new -A do a perfect job for me.