jj absorb is not squashing my file-level changes as expected

[zx8]

Answered by @martinvonz @ #5710 (comment).

I hadn't realised that jj absorb looks for changes at line-level, rather than file-level.

Keeping this open to see if there is appetite for a flag along the lines of --granularity=file, with --granularity=line being the default behaviour. Or perhaps jj absorb could simply fall back to absorbing into the same file if there are no line-level matches?

I appreciate I can do this manually with jj squash, but jj absorb is a lot more convenient!

---

Original Post

Apologies if PEBCAK and not a bug, but I assumed this would work based on the description of jj absorb...

I can't share the exact commits and repository as they're private, but I'll describe the state of things.

I'm working on a feature, based off my colleague's branch, as follows:

@  psswxrxt [email protected] 2025-02-14 19:23:23 4fbbb337
│  (no description set)
○  wootuxwm [email protected] 2025-02-14 19:04:11 example/my-branch git_head() 7592656d
│  add brand new code in new directory
○  vzrmlxxs [email protected] 2025-02-14 19:04:11 7f69a590
│  update colleague's changes in directory `foo/`
○  nlznywlw [email protected] 2025-02-11 17:37:02 example/colleague-branch 753008a0
│  add some more changes in directory `foo/`
○  qyrunlwq [email protected] 2025-02-11 17:35:27 38bc805d
│  add some changes
◆  ztxrpqqt [email protected] 2025-02-10 19:39:01 main df397142
│  init
~

I've made some changes in @ that modify files in foo/ (most recent change modifying these files was in vzrmlxxs). Now, both my colleague and myself have made changes in these files.

If I run jj absorb with no arguments, the output is:

Absorbed changes into these revisions:
  qyrunlwq 3c3cf48f add some changes
Rebased 3 descendant commits.
Working copy now at: onsmswty 2740eb3e (empty) (no description set)
Parent commit      : wootuxwm 7592656d example/my-branch* | add brand new code in new directory

This is what I would expect.

However, I don't want to modify my colleague's changes. Instead, I want to absorb into my own previous changes, so I run jj absorb --into='mutable() & mine()', and the output is:

Nothing changed.

Why is jj not absorbing the changes into my previous change (vzrmlxxs) when I limit the revisions in this way, but is happy to modify my colleague's changes?

[martinvonz]

jj absorb will absorb into the closest commit that modified the same line. Maybe your commit didn't modify the same line? What does jj file annotate say about it?

[zx8]

Aha, that's the part I missed - modifying the same line! I didn't realise it was that fine-grained. I modified elsewhere in the file.

It would be neat if there was a flag/config option (e.g. --granularity=line|file) to operate at the file level, rather than the line level, even if the default is line.

Is this a feature you might consider adding? If so, would you like me to raise a new issue for it, or shall we track it here?

[zx8]

@yuja As the original author of jj absorb, what are your thoughts here? I'm currently working around this by either jj squash'ing manually, or adding arbitrary whitespace changes around lines I plan to modify, which isn't ideal.