Empy commit message being passed to git

[martinbaillie]

Thanks for the neat package. I was looking for something just like this 👍🏻

I've just had a quick play (latest version) and I'm not sure if its PEBKAC or what, but despite a COMMIT_MSG buffer looking great, complete with generated content, a typical C-CC-c to commit in Magit is resulting in "Aborting commit due to empty commit message".

This is regardless of the way I have generated the message using magit-gptcommit.

If I override magit-gptcommit-commit-accept to add (save-buffer) at the end then it works but that feels heavyweight and that it might mess up some other stuff. I naively tried things like (set-buffer-modified-p t) instead but they didn't work.

Again, feels like a me problem but if you have time I'd appreciate some pointers on what to look for.

[douo]

This sounds a bit strange, but I couldn’t reproduce it in my environment. Also, the BUFFER name is COMMIT_EDITMSG, not COMMIT_MSG in my environment.

[martinbaillie]

the BUFFER name is COMMIT_EDITMSG, not COMMIT_MSG in my environment

Ah yes, mine too. I didn't pay enough attention when writing this issue.

Sounds like a me problem then. I could only make it work with some advising:

(advice-add 'magit-gptcommit-commit-accept :after
            (lambda ()
              (when-let ((buf (magit-commit-message-buffer)))
                (with-current-buffer buf
                  (save-buffer)))))

I notice if I run the magitgpt-accept function again whilst in the COMMIT_EDITMSG buffer (so already after doing c then C from the magit-status transient menu) then it works. Would you expect the user to have to do it twice?

I can reproduce with this slimline config:

(require 'use-package)
(setq auth-sources '((:source "REDACTED")))
(use-package magit)
(use-package dash)
(use-package llm :custom (llm-warn-on-nonfree nil))
(use-package magit-gptcommit
  :demand t
  :after magit
  :init
  (require 'llm-openai)

  :custom
  (magit-gptcommit-llm-provider
   (make-llm-openai
    :key (auth-source-pick-first-password :host "api.openai.com" :user "REDACTED")
    :chat-model "gpt-4o"))

  :config
  (magit-gptcommit-status-buffer-setup)
  (magit-gptcommit-mode 1))

[osheari1]

I've been using this package for quite a while, and around the same time this was posted, I started getting the same error message.

[alexjp]

I just found out about magit-gptcommit, tried it and was having the same issue.

I noticed that the edit message buffer had to become dirty and then saved, to not have the "empty commit message" error.

The advice-add workaround from @martinbaillie worked for me.

[osheari1]

Yeah, the advice add works for me as well when using c C. However, c Q still returns an error.