Diff commands jj diff, jj diffedit, jj interdiff with --from argument are confusing in some cases

[zummenix]

I don't look diffs for a range of revisions too often but when I do I always think twice what revision to pass in --from argument. I think the confusion comes from the fact that --from doesn't include the revision in the diff. For example: jj diff --from u --to t shows the diff between u and t not including u. But intuitively I'd like to see the diff including u. On the other hand, when I look at a branch diff using for example jj diff --from main I don't want to see changes of a revision that main branch is set to and this is intuitively what I want and how it currently behaves.

At least, I think there should be a clarification in the cli docs that --from doesn't include the revision in the diff for jj diff, jj diffedit, jj interdiff commands. Alternatively, I'd like to suggest ditching the --from argument completely or changing its semantics.

[martinvonz]

Note that the command works with any two commits, not only where u is an ancestor of t. Let's say you have this history:

B C
|/
A

If you now do jj diff --from B --to C, I would not expect that to show the diff from B's parent (i.e. A) to C.

[zummenix]

If you now do jj diff --from B --to C, I would not expect that to show the diff from B's parent (i.e. A) to C.

Yes, I have the same expectations.

So, the intuition behind --from and --to arguments can be the following: what changes need to be made starting from a revision (but not including) in --from that would lead to a state in --to revision. Is this a correct mental model in general that I can use?

[martinvonz]

Yes, that's correct.

[arxanas]

I think this matches the git diff semantics. It may be gitbrainedness for me, but checking the contents after u personally seems logical.

With your suggestion, it becomes ambiguous on how to include the changes "before" u when u is a merge commit and has multiple parents — which one would you use?

[zummenix]

I guess, my problem is that I have always thought about ranges of commits in context of a diff command. And arguments like --from and --to throw me away. If I try to think about them as "base" and "destination" (sort of) it becomes more clear for me.