Equivalent command for git reset --soft with destinations other than @-?

[lgarron]

I occasionally find that I forgot to run jj new and edited a mutable commit instead. ¹

https://github.com/jj-vcs/jj/blob/main/docs/git-command-table.md documents that git reset --soft HEAD~ corresponds to jj squash --from @-, which is unfortunately insufficient for this.

What I'd like to do is:

• Create a new change.
• Set the parent change to a given commit — usually a different commit of the same changeset, but not always.
• Keep the same exact state of the checkout, with any changes (relative with the new parent commit) becoming part of @.

This is trivial in git:

git reset --soft (new parent branch or commit)

I've tried to look for existing issues/discussions/documentation on this, but I find myself using very complicated workarounds. jj duplicate --destination seems to be close but doesn't seem to produce the right results for me.

Is there a simple equivalent in jj that works with destinations other than @-?

Footnotes

1. Lazy "new" could help with this, but of course doesn't exist yet. ↩

[jennings]

See this FAQ: I accidentally changed files in the wrong commit, how do I move the recent changes into another commit?

I do think this answer could be improved. It uses jj restore to restore the contents of the original commit, but it updates the timestamp so the change doesn't end up pointing to the same commit it did before. This might be important if there's a remote bookmark on the change.

[lgarron]

This might be important if there's a remote bookmark on the change.

Yeah, this is an important detail for branches that I'm collaborating with other people on.

[joyously]

There's a huge discussion about split, but it seems like it could be similar to this case: #3419 What should jj split do with bookmarks?

[lgarron]

Okay, I'm sure this breaks in a lot of ways, but it seems to work specifically for the situation where I've modified a change that I previously pushed as a bookmark:

# fish
function jj_soft_reset
  set TARGET_PARENT_REVISION $argv[1]
  if [ (count $argv) -gt 1 ]
    set CHANGESET_TO_SOFT_RESET $argv[2]
  else 
    set CHANGESET_TO_SOFT_RESET "@"
  end
  echo "CHANGESET_TO_SOFT_RESET: $CHANGESET_TO_SOFT_RESET"
  # TODO: is there a way to do this without checking out the changeset? (`jj duplicate` doesn't seem to have a way to specify the revision/changeset?)
  jj edit --ignore-immutable $CHANGESET_TO_SOFT_RESET
  set JJ_DUPLICATE_OUTPUT (jj duplicate 2>&1)
  echo $JJ_DUPLICATE_OUTPUT

  set FROM_COMMIT_ID (echo $JJ_DUPLICATE_OUTPUT | grep -o "^Duplicated [0-9a-f]\+ as [k-z]\+ " | awk "{print \$2}")
  echo "FROM_COMMIT_ID: $FROM_COMMIT_ID"
  set DUPLICATED_CHANGESET_ID (echo $JJ_DUPLICATE_OUTPUT | grep -o "^Duplicated [0-9a-f]\+ as [k-z]\+ " | awk "{print \$4}")
  echo "DUPLICATED_CHANGESET_ID: $DUPLICATED_CHANGESET_ID"

  jj edit $DUPLICATED_CHANGESET_ID
  jj describe --message "(Changes from a soft reset.)"
  # TODO: this doesn't work for revs that are more than 1 ancestor ago. Do the semantics `jj restore` work for this?
  # TODO: produces false positives for conflicts depending on the contents of `$TARGET_PARENT_REVISION`. Can we automatically resolve conflicts to "ours"?
  jj rebase --destination $TARGET_PARENT_REVISION
  jj abandon $FROM_COMMIT_ID
end

This is… much more complicated than I would like. 😳

[joyously]

Did you see the discussion on #6000 jj squash --restore-descendants ?

[lgarron]

Not yet! I'm having trouble following how that thread is helpful, though. Isn't git reset --soft about restoring ancestors rather than descendants?

[joyously]

It explains the FAQ referenced already, and shows how that option is still a bit unsettled.