[emacs] work around org source block indentation issue with evil

happens due to (setq org-src-tab-acts-natively t)

another less-elegant workaround is to unbind TAB:

(define-key evil-motion-state-map (kbd "TAB") nil)

---

excerpt of conversation with wasamasa on #evil-mode about the issue:

│12:15:00    aminb | wasamasa: mannaged to pinpoint the cause:
│12:15:05    aminb | i had (setq org-src-tab-acts-natively t) in my config
│12:15:22 wasamasa | what does that do?
│12:15:23    aminb | which, for some reason, messes with indentation, and it only happens when evil is enabled
│12:15:26    aminb | give it a shot
│12:15:43    aminb | Documentation:
│12:15:43    aminb | If non-nil, the effect of TAB in a code block is as if it were
│12:15:43    aminb | issued in the language major mode buffer.
│12:15:57 wasamasa | sounds magic
│12:16:11    aminb | i can't notice any difference, at least not with emacs-lisp blocks
│12:16:23    aminb | except that it fucks with evil somehow
│12:19:12 wasamasa | ah, now I get fun behavior
│12:19:19 wasamasa | unless I unbind C-i :P
│12:20:56    aminb | wasamasa: :p so if you unbind C-i, indentation works fine even with that var set?
│12:21:28 wasamasa | I suspect that's why it works normally in my customized session, as opposed to the minimal one (where the block is indented with two spaces for each line)
│12:21:57    aminb | hmm
│12:22:11 wasamasa | ok, it's a bit more complicated
│12:22:35 wasamasa | org looks up the command for the pressed key (evil-jump-forward), executes that in the buffer, then does reformatting
│12:22:56 wasamasa | evil-jump-forward does no indentation, but the reformatting adds indentation unless you customized that
│12:23:10 wasamasa | which I did
│12:23:16    aminb | ha
│12:23:20    aminb | is there anyway to 'fix' this?
│12:23:22 wasamasa | I also unbound TAB and bound C-i separately from that
│12:23:24 wasamasa | well, not really
│12:23:49    aminb | hm okay
│12:23:49 wasamasa | org tries to do the smart thing, but didn't anticipate there's evil where keybindings resolve to something different depending on the state
│12:24:14    aminb | right
│12:26:02 wasamasa | try (define-key evil-motion-state-map (kbd "TAB") nil)
│12:26:29 wasamasa | but as noted, that will kill C-i
│12:27:47    aminb | wasamasa: yup, that seems to do it
│12:27:50 wasamasa | looking at the source of org-babel-do-in-edit-buffer you could do some hack so that the edit buffer comes up in insert state
│12:28:04 wasamasa | then TAB would be looked up correctly
│12:29:10    aminb | i see
│12:42:41 wasamasa | well indeed, try (add-hook 'org-src-mode-hook 'evil-insert-state)

(excerpt included with permission)

diff --git a/init.org b/init.org
index c7181f4..2374cac 100644 (file)
--- a/init.org
+++ b/init.org
@@ -639,7 +639,8 @@ customizing it.
 #+begin_src emacs-lisp
 (use-package evil
   :demand t
-  :hook (view-mode . evil-motion-state)
+  :hook ((view-mode    . evil-motion-state)
+         (org-src-mode . evil-insert-state))
   :init
   (setq evil-want-integration nil)
   :config (evil-mode 1)