Wrong diff sent to LLM for commit message for partial commit

[neic]

When using ellama-generate-commit-message, I am always getting a generated commit message for the diff for the unstaged changes even though I am committing a partial commit. Since #146 I would expect only the staged changes to be in the diff.

I am not a lips hacker, but I fumbled around a bit with the debugger. It showed that the (vc-git-diff-switches "--cached") binding is not picked up by the vc-git-internal call:

Debugger entered--returning value: ("--histogram")
  #<subr vc-switches>(git diff)
* apply(#<subr vc-switches> (git diff))
* vc-switches(git diff)
  vc-git-diff(nil nil nil #<buffer  *temp*> nil)
  vc-call-backend(Git diff nil nil nil #<buffer  *temp*> nil)
  vc-diff-internal(nil (Git nil) nil nil nil #<buffer  *temp*>)
  ellama--diff-cached()
  ellama-generate-commit-message()
  funcall-interactively(ellama-generate-commit-message)
  command-execute(ellama-generate-commit-message)

Found by running M-x debug-on-entry vc-switches and stepping through with d.

I fumbled around a bit more and got it to work by moving the binding around the with-temp-buffer:

(defun ellama--diff-cached ()
  "Diff staged."
  (let* ((default-directory
      (if (string= ".git"
               (car (reverse
                 (cl-remove
                  ""
                  (file-name-split default-directory)
                  :test #'string=))))
          (file-name-parent-directory default-directory)
        default-directory))
     (diff (let ((vc-git-diff-switches "--cached"))
                 (with-temp-buffer
           (vc-diff-internal
            nil (vc-deduce-fileset t) nil nil nil (current-buffer))
           (buffer-substring-no-properties (point-min) (point-max))))))
    (if (string-empty-p diff)
    nil
      diff)))

I do not know why the let works and the let* does not. (let* ((a 2) (b 3) (c a)) (+ b c)) does eval to 5.

I can whip up a PR if this solution is the way to fix it.

[s-kostyaev]

Hi @neic. Thank you for report and debugging. Please provide a PR. Have you tested your solution?

[s-kostyaev]

@neic I can't reproduce your issue. Which emacs version do you have?