search

wasamasa / eyebrowse

A simple-minded way of managing window configs in emacs
GNU General Public License v3.0
437 stars 24 forks source link

switching workspaces misbehaving when a window is dedicated #36

Closed bmag closed 9 years ago

bmag commented 9 years ago

Sometimes when a window is dedicated, switching a workspace fails and messes the frame's window configuration. This causes trouble with packages such as NeoTree - see this report.

Recipe:

  1. emacs -Q -l /path/to/dash.el -l /path/to/eyebrowse.el
  2. M-x eyebrowse-mode
  3. C-x b buff1
  4. M-: (set-window-dedicated-p nil t)
  5. C-x 4 b buff2
  6. C-c C-w 2 Result: buff1's window becomes the only window in the frame and I see a message: "Window is dedicated to `buff1'". According to the mode-line, a second workspace is created, but the first workspace remains the current workspace.

I looked into it, and it seems the error is produced by window-state-put. Replace step 6 with:

  1. M-: (window-state-put (window-state-get nil 'writable) (frame-root-window)) And the result is an error traceback for "Window is dedicated to `buff1'".

I think that adding (progn (delete-other-windows) (set-window-dedicated-p nil nil)) before calling window-state-put will solve this.

P.S. Latest dash and eyebrowse from melpa, emacs version 24.5.

wasamasa commented 9 years ago

Oh right, I didn't test at all with dedicated windows (and to be fair, I believe them to be a half-assed attempt at introducing some regularity in Emacs' way of window management, simply because the comments in window.el hint at a way more useful system, but don't implement it).

Bisecting revealed the issue happened after switching to the window-state-get/window-state-put storage backend (as opposed to the older current-window-configuration/set-window-configuration storage backend). This isn't really surprising as set-window-configuration is written in C and bypasses all higher-level protections whereas window-state-put is written in elisp and is subject to the limitations of window.el (which apparently include being subject to dedication status). I'll dig a bit further to see whether it's an Emacs bug with that function or something I can work around in a better way than unsetting dedication status as you've suggested.

wasamasa commented 9 years ago

FWIW, window-state-put errors out when using set-window-buffer which is defined in C. I don't see any variable or macro to temporarily(!) inhibit the dedicated status, so I'm writing an Emacs bug report for this.

edit: Here it is

wasamasa commented 9 years ago

Thanks to this discussion master got patched, I'm adding a workaround for earlier versions as Emacs 24.3, 24.4 and 24.5 are still widely in use.

bmag commented 9 years ago

Awesome, thank you

bmag commented 9 years ago

Sorry to re-raise this issue, but the fix in https://github.com/wasamasa/eyebrowse/commit/7f92251f320ac7e49e71f370b5901f6d5fd0bea6 doesn't work in Emacs 24.5.2. I guess the bug in Emacs was fixed in 25.x but the fix was not included in the 24.5.2. Probably the condition should be (version< emacs-version "25") or something...

wasamasa commented 9 years ago

Thanks for re-raising, I've totally forgot that there would inevitably be a bump in patch version.