fixed bug in Item_func_div::val_int() that broke all functions that

do val_int() on their arguments before starting the computation. 
Similar fixes are need for +-* and probably several other but I want
to make sure Monty is fine with my fix approach before changing a lot
of code. 

Amazingly,
this bug is not as critical as you would expect since very few functions do val_int()
on their arguments ( from_unixtime(), sec_to_time()), and those not
very frequently perform a computation on their floating point arguments.
which is probably why no one has yet reported this bug. Another 
possibility is that the result is usually wrong by no more than 5%,
which makes it hard to catch it. I found it when trying to compute mile
splits for 30:47 10K - it told me 5:07, and I knew it was wrong because
5:00 mile gives you 31:08. However, if I had not run as many 10K races,
I would have easily believed that 30:47 10K is a 5:07 mile pace and 
would not have noticed the bug.

mysql-test/r/func_time.result

@@ -6,8 +6,8 @@ now()-now()	weekday(curdate())-weekday(now())	unix_timestamp()-unix_timestamp(no
 0	0	0
 from_unixtime(unix_timestamp("1994-03-02 10:11:12"))	from_unixtime(unix_timestamp("1994-03-02 10:11:12"),"%Y-%m-%d %h:%i:%s")	from_unixtime(unix_timestamp("1994-03-02 10:11:12"))+0
 1994-03-02 10:11:12	1994-03-02 10:11:12	19940302101112
-sec_to_time(9001)	sec_to_time(9001)+0	time_to_sec("15:12:22")
-02:30:01	23001	54742
+sec_to_time(9001)	sec_to_time(9001)+0	time_to_sec("15:12:22")	sec_to_time(time_to_sec("0:30:47")/6.21)
+02:30:01	23001	54742	00:04:57
 now()-curdate()*1000000-curtime()
 0
 strcmp(current_timestamp(),concat(current_date()," ",current_time()))

mysql-test/t/func_time.test

@@ -7,7 +7,8 @@ select from_days(to_days("960101")),to_days(960201)-to_days("19960101"),to_days(
 select period_add("9602",-12),period_diff(199505,"9404") ;
 select now()-now(),weekday(curdate())-weekday(now()),unix_timestamp()-unix_timestamp(now());
 select from_unixtime(unix_timestamp("1994-03-02 10:11:12")),from_unixtime(unix_timestamp("1994-03-02 10:11:12"),"%Y-%m-%d %h:%i:%s"),from_unixtime(unix_timestamp("1994-03-02 10:11:12"))+0;
-select sec_to_time(9001),sec_to_time(9001)+0,time_to_sec("15:12:22");
+select sec_to_time(9001),sec_to_time(9001)+0,time_to_sec("15:12:22"),
+  sec_to_time(time_to_sec("0:30:47")/6.21);
 select now()-curdate()*1000000-curtime();
 select strcmp(current_timestamp(),concat(current_date()," ",current_time()));
 select date_format("1997-01-02 03:04:05", "%M %W %D %Y %y %m %d %h %i %s %w");

sql/item_func.cc

@@ -309,11 +309,14 @@ double Item_func_div::val()
 
 longlong Item_func_div::val_int()
 {
-  longlong value=args[0]->val_int();
-  longlong val2=args[1]->val_int();
-  if ((null_value= val2 == 0 || args[0]->null_value || args[1]->null_value))
-    return 0;
-  return value/val2;
+  // the integer result of division of two arguments needs to be computed
+  // as a type-cast division of val(), not as diviion of val_int() of each
+  // argument. For example, val_int(41.5/3.4) = val_int(12.206) = 12, but
+  // if you do val_int(41.5)/val_int(3.4), as in the old code, we get 42/3=
+  // 14, which is wrong. This would break sec_to_time(a/b),
+  // from_unixtime(a/b), and
+  // all functions that do val_int() on their arguments
+  return (longlong)val();
 }
 
 void Item_func_div::fix_length_and_dec()