Re: jit-lock timer etc.

[storm@cua.dk (Kim F. Storm)]

Richard Stallman <rms@gnu.org> writes:

> Here's my patch to make the jit-lock timer reschedule itself
> instead of waiting.  It uses current-idle-time, which I have tested
> and checked in, but I can't easily tell whether steath fontification
> is working.  So I would like someone to test this before I check it in.
>
> I have not yet tried to read the other proposed change to this code.
> I was too tired when I saw it.

I strongly prefer the other change proposed by Martin Rudalics which
avoids calling sit-for at all in the timer handler.  I have installed
Martin's patch, and I haven't seen the "cursor disappearing" bug since!

However, the approach taken by Martin can be improved by using your
new current-idle-time to install a proper idle timer to resume jit-lock.

Martin, would you like to try to update your patch ?



>
> *** jit-lock.el	17 Aug 2006 11:21:10 -0400	1.54
> --- jit-lock.el	18 Aug 2006 07:57:36 -0400	
> ***************
> *** 171,176 ****
> --- 171,180 ----
>   
>   (defvar jit-lock-stealth-timer nil
>     "Timer for stealth fontification in Just-in-time Lock mode.")
> + (defvar jit-lock-stealth-resume-timer nil
> +   "Timer for resuming stealth fontification in Just-in-time Lock mode.
> + When time stealth fontification finishes one batch, it reschedules
> + itself using this timer.")
>   (defvar jit-lock-context-timer nil
>     "Timer for context fontification in Just-in-time Lock mode.")
>   (defvar jit-lock-defer-timer nil
> ***************
> *** 225,230 ****
> --- 229,241 ----
>   		 (run-with-idle-timer jit-lock-stealth-time t
>   				      'jit-lock-stealth-fontify)))
>   
> + 	 ;; Set up an idle timer for stealth fontification to resume,
> + 	 ;; but don't activate it now.
> + 	 (when (and jit-lock-stealth-time (null jit-lock-stealth-timer))
> + 	   (setq jit-lock-stealth-resume-timer (timer-create))
> + 	   (timer-set-function jit-lock-stealth-resume-timer
> + 			       'jit-lock-stealth-fontify))
> + 
>   	 ;; Init deferred fontification timer.
>   	 (when (and jit-lock-defer-time (null jit-lock-defer-timer))
>   	   (setq jit-lock-defer-timer
> ***************
> *** 449,454 ****
> --- 460,466 ----
>   This functions is called after Emacs has been idle for
>   `jit-lock-stealth-time' seconds."
>     ;; I used to check `inhibit-read-only' here, but I can't remember why.  -stef
> +   (cancel-timer jit-lock-stealth-resume-timer)
>     (unless (or executing-kbd-macro
>   	      memory-full
>   	      (window-minibuffer-p (selected-window)))
> ***************
> *** 457,512 ****
>   	  minibuffer-auto-raise
>   	  message-log-max)
>         (with-local-quit
> ! 	(while (and buffers (not (input-pending-p)))
>   	  (with-current-buffer (pop buffers)
> ! 	    (when jit-lock-mode
> ! 	      ;; This is funny.  Calling sit-for with 3rd arg non-nil
> ! 	      ;; so that it doesn't redisplay, internally calls
> ! 	      ;; wait_reading_process_input also with a parameter
> ! 	      ;; saying "don't redisplay."  Since this function here
> ! 	      ;; is called periodically, this effectively leads to
> ! 	      ;; process output not being redisplayed at all because
> ! 	      ;; redisplay_internal is never called.  (That didn't
> ! 	      ;; work in the old redisplay either.)  So, we learn that
> ! 	      ;; we mustn't call sit-for that way here.  But then, we
> ! 	      ;; have to be cautious not to call sit-for in a widened
> ! 	      ;; buffer, since this could display hidden parts of that
> ! 	      ;; buffer.  This explains the seemingly weird use of
> ! 	      ;; save-restriction/widen here.
> ! 
> ! 	      (with-temp-message (if jit-lock-stealth-verbose
> ! 				     (concat "JIT stealth lock "
> ! 					     (buffer-name)))
> ! 
> ! 		;; In the following code, the `sit-for' calls cause a
> ! 		;; redisplay, so it's required that the
> ! 		;; buffer-modified flag of a buffer that is displayed
> ! 		;; has the right value---otherwise the mode line of
> ! 		;; an unmodified buffer would show a `*'.
> ! 		(let (start
> ! 		      (nice (or jit-lock-stealth-nice 0))
> ! 		      (point (point-min)))
> ! 		  (while (and (setq start
> ! 				    (jit-lock-stealth-chunk-start point))
> ! 			      ;; In case sit-for runs any timers,
> ! 			      ;; give them the expected current buffer.
> ! 			      (with-current-buffer outer-buffer
> ! 				(sit-for nice)))
> ! 
> ! 		    ;; fontify a block.
> ! 		    (jit-lock-fontify-now start (+ start jit-lock-chunk-size))
> ! 		    ;; If stealth jit-locking is done backwards, this leads to
> ! 		    ;; excessive O(n^2) refontification.   -stef
> ! 		    ;; (when (>= jit-lock-context-unfontify-pos start)
> ! 		    ;;   (setq jit-lock-context-unfontify-pos end))
> ! 
> ! 		    ;; Wait a little if load is too high.
> ! 		    (when (and jit-lock-stealth-load
> ! 			       (> (car (load-average)) jit-lock-stealth-load))
> ! 		      ;; In case sit-for runs any timers,
> ! 		      ;; give them the expected current buffer.
> ! 		      (with-current-buffer outer-buffer
> ! 			(sit-for (or jit-lock-stealth-time 30))))))))))))))
>   
>   
>   \f
> --- 469,527 ----
>   	  minibuffer-auto-raise
>   	  message-log-max)
>         (with-local-quit
> ! 	(when (and jit-lock-mode buffers (not (input-pending-p)))
>   	  (with-current-buffer (pop buffers)
> ! 	    ;; This is funny.  Calling sit-for with 3rd arg non-nil
> ! 	    ;; so that it doesn't redisplay, internally calls
> ! 	    ;; wait_reading_process_input also with a parameter
> ! 	    ;; saying "don't redisplay."  Since this function here
> ! 	    ;; is called periodically, this effectively leads to
> ! 	    ;; process output not being redisplayed at all because
> ! 	    ;; redisplay_internal is never called.  (That didn't
> ! 	    ;; work in the old redisplay either.)  So, we learn that
> ! 	    ;; we mustn't call sit-for that way here.  But then, we
> ! 	    ;; have to be cautious not to call sit-for in a widened
> ! 	    ;; buffer, since this could display hidden parts of that
> ! 	    ;; buffer.  This explains the seemingly weird use of
> ! 	    ;; save-restriction/widen here.
> ! 
> ! 	    (with-temp-message (if jit-lock-stealth-verbose
> ! 				   (concat "JIT stealth lock "
> ! 					   (buffer-name)))
> ! 
> ! 	      ;; In the following code, the `sit-for' calls cause a
> ! 	      ;; redisplay, so it's required that the
> ! 	      ;; buffer-modified flag of a buffer that is displayed
> ! 	      ;; has the right value---otherwise the mode line of
> ! 	      ;; an unmodified buffer would show a `*'.
> ! 	      (let (start
> ! 		    (nice (or jit-lock-stealth-nice 0))
> ! 		    (point (point-min)))
> ! 		(while (and (setq start
> ! 				  (jit-lock-stealth-chunk-start point))
> ! 			    ;; In case sit-for runs any timers,
> ! 			    ;; give them the expected current buffer.
> ! 			    (with-current-buffer outer-buffer
> ! 			      (sit-for nice)))
> ! 
> ! 		  ;; fontify a block.
> ! 		  (jit-lock-fontify-now start (+ start jit-lock-chunk-size))
> ! 		  ;; If stealth jit-locking is done backwards, this leads to
> ! 		  ;; excessive O(n^2) refontification.   -stef
> ! 		  ;; (when (>= jit-lock-context-unfontify-pos start)
> ! 		  ;;   (setq jit-lock-context-unfontify-pos end))
> ! 
> ! 		  ;; Wait a little if load is too high.
> ! 		  (when (and jit-lock-stealth-load
> ! 			     (> (car (load-average)) jit-lock-stealth-load))
> ! 		    ;; Make this timer resume jit-lock in a few
> ! 		    ;; seconds.  Compute an idleness time suitably
> ! 		    ;; larger than the current one.
> ! 		    (timer-set-idle-time jit-lock-stealth-resume-timer
> ! 					 (current-idle-time))
> ! 		    (timer-inc-time (or jit-lock-stealth-time 30))
> ! 		    (timer-activate-when-idle
> ! 		     jit-lock-stealth-resume-timer t)))))))))))
>   
>   
>   \f

-- 
Kim F. Storm <storm@cua.dk> http://www.cua.dk