## Tuesday, April 1, 2014

### Some Emacs hacks for GDB and other stuff

I am fighting with GDB in Emacs today. Thought I would share a few hacks that make things more livable. These get hackier as you go, so… be warned.

First, GDB in many windows mode makes a bunch of dedicated windows. This can be pretty annoying. In fact, dedicated windows in general can be pretty annoying. If I want to change the buffer of a window, don't you dare stop me. Emacs Wiki has some advice code that allows you to disable the setting of dedicated windows in GDB startup (bottom of the page). But to be clear, this is a problem with the Emacs interface, not the GDB interface. The GDB interface has a good reason to not want those buffers to change, and it is annoying to use it if they do change. The problem is when I know that I really do want these to change but Emacs makes me jump through hoops to do it. So, let's fix the underlying problem. Here is a bit of code to add to your init files that will allow you to change buffers even if they are dedicated.

(defun undedicate-window (&optional window)
(interactive)
(set-window-dedicated-p (or window (get-buffer-window)) nil))

;; Removing annoying dedicated buffer nonsense
(defun switch-to-buffer! (buffer-or-name &optional norecord force-same-window)
"Like switch-to-buffer but works for dedicated buffers \(though
(interactive
(list (read-buffer-to-switch "Switch to buffer: ") nil 'force-same-window))
(when (and (window-dedicated-p (get-buffer-window))
(yes-or-no-p "This window is dedicated, undedicate it? "))
(undedicate-window))
(switch-to-buffer buffer-or-name norecord force-same-window))


I just set a global key binding of (kbd "C-x b") to switch-to-buffer! (actually I use a global minor mode to keep track of all of my keybindings, but the result is the same). This will now act exactly like switch-to-buffer unless the window is dedicated, in which case it will ask if you want to "undedicate" the window first. Now Emacs wont reuse these buffers willy nilly, but you can still do what you think is best.

Second, dedicated windows are so convenient (now that they are convenient to undedicate) that you might find that you want to start setting the dedicated flag on random windows that you don't want Emacs to change. So here is a function for that. If you want a dedicated completion window, well, just set it on that window and you won't have to worry about it getting taken over by some other Emacs pop-up.

(defun toggle-window-dedication (&optional window)
(interactive)
(let ((window (or window (get-buffer-window))))
(set-window-dedicated-p window (not (window-dedicated-p window)))))

(global-set-key (kbd "C-x d") 'toggle-window-dedication)


Third, I really hate that performing any act in GUD tends to make the source buffer you are working with jump back to the execution point. This is a problem if you are setting several breakpoints, or printing out several values from the source file. Thus I came up with this hack (and this really is a hack) to make this problem go away.

(defun nice-gud-print (arg)
(interactive "p")
(save-excursion
(gud-print arg)
(sleep-for .1)))
(global-set-key (kbd "C-x C-a C-p") 'nice-gud-print)
(defun nice-gud-break (arg)
(interactive "p")
(save-excursion
(gud-break arg)
(sleep-for .1)))
(global-set-key (kbd "C-x C-a C-b") 'nice-gud-break)
(defun nice-gud-tbreak (arg)
(interactive "p")
(save-excursion
(gud-tbreak arg)
(sleep-for .1)))
(global-set-key (kbd "C-x C-a C-t") 'nice-gud-tbreak)


The sleep-for call is necessary to make save-excursion actually work. My guess here is that GUD queues requests and then executes them later. So without the sleep, my call to GUD returns, and with it the save-excursion environment exits, long before GUD processes the command and then resets the view to the execution point. Not pretty, but it works for me at least.

Lastly, I find that it is helpful to have a key binding for gdb-restore-windows. This way you can easily jump to a particular window, make the buffer full screen, and then return to the debugger view later. However, if you like to use a vertically split screen like I do (i.e. you like 80 column width programs), it is often even better to have a toggle between just the two center windows and the full debugger view. So, here is a function to do that:

(defun gdb-windows-toggle (&optional full-restore)
(interactive "P")
(let ((window-tree (first (window-tree))))
(cond (full-restore
(gdb-restore-windows))
((and (listp window-tree)
(= (length window-tree) 5))
(let ((buffers (if (listp (fourth window-tree))
(mapcar 'window-buffer
(rest (rest (fourth window-tree))))
(list (window-buffer (fourth window-tree)))))
(first-selected (or (windowp (fourth window-tree))
(eql (first (window-list))
(third (fourth window-tree))))))
(delete-other-windows)
(when (> (length buffers) 1)
(split-window-horizontally))
(cl-loop for buffer in buffers
for window in (window-list)
do (progn (undedicate-window window)
(set-window-buffer window buffer)))
(unless first-selected
(select-window (second (window-list))))))
((or (windowp window-tree)
(and (= (length window-tree) 4)
;; horizontal split
(not (first window-tree))
;; No further splits
(windowp (third window-tree))
(windowp (fourth window-tree))))
(let ((current-buffers
(if (windowp window-tree)
(list (window-buffer window-tree))
(mapcar 'window-buffer (rest (rest window-tree)))))
(first-selected (or (windowp window-tree)
(eql (first (window-list))
(third window-tree)))))
(gdb-restore-windows)
(let ((windows (rest (rest (fourth (first (window-tree)))))))
(when (= (length current-buffers) 1)
(delete-window (second windows)))
(cl-loop for buffer in current-buffers
for window in windows
do (progn (undedicate-window window)
(set-window-buffer window buffer)))
(if first-selected
(select-window (first windows))
(select-window (second windows))))))
(t ;; If all else fails, just restore the windows
(gdb-restore-windows)))))

(global-set-key (kbd "C-c g") 'gdb-windows-toggle)


That is long and ugly, and probably could be made much simpler, but it does the trick and does it pretty well. Maybe if these really work out I can get someone involved with Emacs to include some of these little hacks (cleaned up of course). It would be nice to deal with the fact that GDB is fundamentally broken for me right now, but… we'll see where those bug reports go.