Re: [AD] timer_mutex again

[Eric Botcazou <ebotcazou@xxxxxxxxxx>]

> Ah, cheers!  Attached is a patch for dialog_message() instead, with a
> verbose explanation.

The only potential nasty side-effect I can think of is that dialog_message() 
will acquire/release the screen for every object.  But given that 
scare_mouse() is also invoked for every object, that may go unnoticed in 
practice.

> >I should have documented it in timer.c.
>
> Patch attached; what do you think?

First of all, thanks. :-)

However,

 static void *timer_mutex = NULL;          /* global timer mutex */
+   /* Why we need timer_mutex:
+    *
+    * 1. So _timer_queue[] is not changed in the middle of
+    * _handle_timer_tick's loop, possibly causing a null pointer
+    * dereference or corruption;
+    *
+    * 2. So that after "remove_int(my_callback);" returns, my_callback
+    * is guaranteed not to be in the _timer_queue and will not be
+    * called from the timer thread.
+    */

After remove_int(my_callback), my_callback is guaranteed not be in the timer 
queue anymore (unless install_int, but that's #1).  The problem is that 
my_callback can have been fired but has not completed yet; context switch; 
the main thread returns from remove_int and set my_pointer to 0; context 
switch; the callback dereferences my_pointer.  Ergo remove_int must wait for 
all callbacks to have completed before returning.

-- 
Eric Botcazou