User-agent: Mozilla Thunderbird 0.9 (Macintosh/20041103)
Aaron Jackson wrote:
On Mar 17, 2005, at 4:55 PM, Matt Watson wrote:
Sorry for jumping to the wrong conclusion. The problem is that I am
doing some numerical calculations (solving nonlinear equations with a
nonlinear constraint) and the algorithm I am using sometimes wanders off
into nowhere (I am still debugging things). I haven't done any c
programming in quite a while, and got used to the crashing behavior when
I first learned and thought that was how things operated.
Since this is undefined, I would like for the program to throw a SIGFPE,
Undefined means 'undefined'. The compiler doesn't have to catch the
exception, and (from Andrew Pinski's reports) the chip *doesn't* catch the
exception.
In order to do what you're requesting, the compiler would have to write
pre-checking code to note that it is about to do a division by 0 and
manually synthesise an exception. Personally, I'd rather my compiler *not*
add an extra test-compare into every division operation I program.
It's undefined because _it should never happen_, and because their are a
variety of different ways the *hardware* might choose to deal with this
error. The C standard tries to only mandate particular behaviour if it is
necessary for correct program execution. By definition, undefined areas
are *not* necessary for correct program execution, because they will never
be executed by a correct program.
If your code is allowing it to happen, add a test-compare before each
division operation in your code. See which one triggers.
int
my_checked_int_divide (x,y)
{
if (y)
{
return x / y;
}
else
{
// Suitable error handling goes here
}
}
Drop a breakpoint inside that else and you get an instant stack trace
whenever you attempt to divide by zero. Once you've fixed the bug, just
change the divide to:
#define MY_INT_DIV(x,y) ((x)/(y))
--
Andrew White
--------------------------------------------------------------------------
This email and any attachments are confidential. They may contain legally
privileged information or copyright material. You should not read, copy,
use or disclose them without authorisation. If you are not an intended
recipient, please contact us at once by return email and then delete both
messages. We do not accept liability in connection with computer virus,
data corruption, delay, interruption, unauthorised access or unauthorised
amendment. This notice should not be removed.
_______________________________________________
Do not post admin requests to the list. They will be ignored.
Darwin-dev mailing list (email@hidden)
Help/Unsubscribe/Update your Subscription:
http://lists.apple.com/mailman/options/darwin-dev/email@hidden
