Dear fellow emacsers.

How can we make undo work on the scratch buffer? I’m sure it is a simple configuration issue, but I’m not getting the appropriate keyword to find the relevant documentation.

Thanks a lot in advance for your help and attention…
:) /\

  • Sunoc@sh.itjust.works
    1·
    1 year ago

    Hi, That’s a surprising issue, especially if the scratch buffer is the only text buffer in which the undo doesn’t work. Here is what I’m thinking right now:

    • Just to be sure, isn’t your scratch buffer in read-only ? You can toggle that with C-x C-q.
    • Is the keybinding you are using to undo set properly ? You can check it with C-h k C-/ (for example, for checking C-/).
    • If it’s none of these, I’d try to run the undo function directly with M-x undo to see if you can get some meaningful error messages.

    I confirmed with both my personal config that uses undo-tree (great tool btw) and also in a default configuration of Emacs; both cases work fine to undo in scratch.

    Hope any of that can help!

    Cheers

    • EduMercoOPEnglish
      2·
      1 year ago

      Hello Sunoc.

      • the buffer is not in read-only, but it always start as so.
      • the keybinding is ok since it works in all the rest of the buffers.
      • Upon calling undo, emacs says “no undo information in this buffer” (searching for this didn’t seem to give relevant information except something from 2006 and then something about specific of a table-spreadsheet that I don’t use).

      From this and @ouigol@beehaw.org (I will relpy after this one) it looks that it may be my configuration, since I’m using vanilla emacs with my own configuration.

      Anyway, thanks a lot for your time! :)

      • Sunoc@sh.itjust.worksEnglish
        1·
        1 year ago

        After checking the source, it looks like this error is triggered when the buffer-undo-list variable is equal to t, which means the buffer is unmodified, or at least that this variable is not modified from the scratch buffer.

        There is something else in the documentation of this variable: <If the value of the variable is t, undo information is not recorded.> Maybe you can try to fix your issue by forcing the buffer-undo-list variable to nil for the elisp mode, as follow:

        (add-hook 'elisp-mode (lambda () (setq-local buffer-undo-list nil)))

        • EduMercoOPEnglish
          1·
          1 year ago

          Maybe you can try to fix your issue by forcing the buffer-undo-list variable to nil for the elisp mode, as follow: (add-hook 'elisp-mode (lambda () (setq-local buffer-undo-list nil)))

          I tried it but still no joy…

          Looking at this issue I noticed 3 things that may help to pinpoint the origin…:

          • I have 2 variables set that affect the scratch buffer in my config (without yours, that it’s the 3rd):
                            (setq initial-major-mode 'org-mode)
                  (setq-default major-mode 'org-mode)
          • Before this reply, (setq-default major-mode 'org-mode) was (setq default-major-mode 'org-mode). Both expressions seems indistinguishable in the resulted behavior.
          • The scratch buffer starts in org-roam mode (and read-only mode), which is very strange, but can obviously be related and/or part of the problem. The org-roam section (my config is literate in org-mode) comes before this variables being set.

          What does this info may imply?

          • Sunoc@sh.itjust.worksEnglish
            1·
            1 year ago

            Okay, I could reproduce your situation by evaluating

            (setq initial-major-mode 'org-mode)
(setq-default major-mode 'org-mode)

            This setup forces the scratch buffer to be in org-mode, apparently breaking most of the available keybindings for it, including the undo.

            The value for the initial-major-mode should be kept as lisp-interaction-mode, if your configuration requires to set it.

            Otherwise I would remove both of these lines all together and see if the default behavior brings back the scratch buffer to be in the expected lisp-interaction-mode again. The additional hook I provided in my previous post shouldn’t be needed either.