How best to abandon an entire branch?

[sersorrel]

Yes, jj doesn't have branches – but git does, and if I have some history like this:

Z <-- foo
|
Y  C <-- bar
|  |
X  B
| /
A <-- main

it can happen that I am no longer interested in the entire branch that bar points to the tip of, i.e. I don't just wish to abandon C, but also B (but not A). For such a simple example I can just jj abandon B C, but if there are e.g. 80+ changes on the bar branch then that's more cumbersome.

So far what I came up with was jj abandon '(visible_heads() ~ bar)..bar (which I think is approximately "all commits that are reachable exclusively from the tip that bar points to"), and that seems to work ok. It's a bit annoying that I can't easily make an alias for it (without jj util exec) though.

After playing around a bit more, I realised that the above revset misbehaves in the following scenario:

Z <-- foo
|
Y  C <-- bar
|  |
X  B <-- baz
| /
A <-- main

Here, I'd hope to only abandon C, but B gets dropped too (since baz doesn't point to a head). A version that avoids this is jj abandon '((visible_heads() | bookmarks()) ~ bar)..bar', but we are really getting into the very-unwieldy-to-type zone now.

Are there any situations in which this updated revset might not work well? Is there a "standard" way of doing this that I haven't found?

[arialdomartini]

Possibly related: #5872

I guess you could simply do

• jj abandon B::C (both B and C - and any other revsets in between - would be included)
• or jj abandon A..C (A would not be included)

[ilyagr]

For real-world usage, I'd recommend jj abandon foo..C and @arialdomartini 's suggestions.

For the problem as stated, I'm not sure I can do much better than '((visible_heads() | bookmarks()) ~ bar)..bar', which you could make a revset alias (with a parameter).

This revset is not perfect. It's unclear to me whether it's behavior if there was another bookmark at the same place as bar is good, but then it's also unclear what you'd want to happen in this situation.

Others might have suggestions, but I think that for it to really work, we'd need some form of #4496 (which we aren't sure how to do right).

[sersorrel]

A revset alias with parameter is a good suggestion, thank you!

Yeah, I have been thinking that what I really want is some way to group together a contiguous (or even linear) set of commits, which is currently approximated by a bookmark pointing at one end of the set, and that sounds like roughly what the conversation around topics is also about.