Thread: How can I check if a integer variable is overflowed?

Code:

if( valuetoadd > maximumvalue - currentvalue )
    ...boo
else
    ...yay

Quzah.

Originally Posted by itCbitC

test if current value < last value

Odd as it may seem... this is probably the only reliable test you have... see if it rolled over the top.

An integer will never be > INT_MAX ... as itCbitC points out it will roll back over to 0 or -1 first.

Assuming 32 bits...

Code:

int a =  2147483645;
a = a + 3;
printf("%d",a);  // - 1

unsigned int b = 4294967293;
b = b + 3;
printf("%d",b); // 1

Originally Posted by CommonTater

Odd as it may seem... this is probably the only reliable test you have... see if it rolled over the top.

For an unsigned integer type, this is true (they work with modulo arithmetic, so are required to "roll over the top".

This is not true for signed integer types. That is why quzah's test (test before doing an operation) is needed.

Originally Posted by Subsonics

But then add this:

Code:

if( ((uint64_t)val + value_to_add) < INT_MAX ) val += value_to_add;

Which only works if val and value_to_add are 32 bits.

Back in the good old Pascal days everyone was wondering how TP detected overflows at run time when in-code solutions like these failed so miserably. The answer turned out to be quite simple... The overflow flag in the CPU... a little bit of assembler at the end of each math function would check the OF flag, if it was lit up, the program tossed an exception.

http://en.wikipedia.org/wiki/FLAGS_register_(computing)

A similar process could probably be implemented in C, but it would most likely mean rewriting very large parts of everything to make it work...

You might try...

Code:

if ((INT_MAX - current_value) > value_to_add)  
   current_value += value_to_add
else
   DoTheMassivelyAnnoyingErrorThing();

Originally Posted by CommonTater

Which only works if val and value_to_add are 32 bits.

Back in the good old Pascal days everyone was wondering how TP detected overflows at run time when in-code solutions like these failed so miserably. The answer turned out to be quite simple... The overflow flag in the CPU... a little bit of assembler at the end of each math function would check the OF flag, if it was lit up, the program tossed an exception.

Of course but the question was about ints, otherwise you would not use INT_MAX, but really quazah's method works no matter what. And checking the overflow flag is a bit overkill I think, it's only set after the fact, so you would need to create a test function specific for addition and ints, something like this.

Code:

int overflow_add(int a, int b)
{
    int test = a + b;
    int ofl = 0;

    __asm__(
        "seto %b0;"
        : "=g"(ofl)
        :
        : "%bh"
    );

    return ofl;
}

Originally Posted by CommonTater

You might try...

Code:

if ((INT_MAX - current_value) > value_to_add)  
   current_value += value_to_add
else
   DoTheMassivelyAnnoyingErrorThing();

Obviously, it's already been posted and agreed upon.

Originally Posted by itCbitC

As is quzah's code won't work for signed and neither will current_value < last_value.

Of course mine will, it will also work for self-imposed limits:

Code:

if( 300(add) > 1000(max) - 800(current) )
    boo
else
    yay

Unless you are talking about negative numbers, and going lower instead of going higher as you check for limits, but then all you do is change 'max' to 'min', and > to < and + to - ... which while easy enough to do, isn't what he asked for.

Don't just tell me it doesn't work, show me how it doesn't work, because I'm not seeing it.


Quzah.