bug#8664: * keyboard.c (make_lispy_event): Fix problem in integer overflow.

[Paul Eggert <eggert@cs.ucla.edu>]

Here's a patch for a potential problem with integer overflow
on 64-bit hosts that I plan to install after some more testing.
The problem is a bit more severe if EMACS_INT is 64-bit on
a 32-bit host, and I found it by inspection.

* keyboard.c (make_lispy_event): Fix problem in integer overflow.
Don't assume that the difference between two unsigned long values
can fit into an integer.  At this point, we know button_down_time
<= event->timestamp, so the difference must be nonnegative, so
there's no need to cast the result if double-click-time is
nonnegative, as it should be; check that it's nonnegative, just in
case.  This bug is triggered when events are more than 2**31 ms
apart (about 25 days).
=== modified file 'src/keyboard.c'
--- src/keyboard.c	2011-04-28 19:35:20 +0000
+++ src/keyboard.c	2011-05-12 19:33:15 +0000
@@ -5556,9 +5556,9 @@
 		       && (eabs (XINT (event->y) - last_mouse_y) <= fuzz)
 		       && button_down_time != 0
 		       && (EQ (Vdouble_click_time, Qt)
-			   || (INTEGERP (Vdouble_click_time)
-			       && ((int)(event->timestamp - button_down_time)
-				   < XINT (Vdouble_click_time)))));
+			   || (NATNUMP (Vdouble_click_time)
+			       && (event->timestamp - button_down_time
+				   < XFASTINT (Vdouble_click_time)))));
 	}

 	last_mouse_button = button;
@@ -5742,9 +5742,9 @@
 		       && (eabs (XINT (event->y) - last_mouse_y) <= fuzz)
 		       && button_down_time != 0
 		       && (EQ (Vdouble_click_time, Qt)
-			   || (INTEGERP (Vdouble_click_time)
-			       && ((int)(event->timestamp - button_down_time)
-				   < XINT (Vdouble_click_time)))));
+			   || (NATNUMP (Vdouble_click_time)
+			       && (event->timestamp - button_down_time
+				   < XFASTINT (Vdouble_click_time)))));
 	  if (is_double)
 	    {
 	      double_click_count++;