Overflow after an implicit cast

Question

When I try to make an implicit cast from a double to an unsigned long, I have an overflow warning : "warning: overflow in implicit constant conversion [-Woverflow]".

Here is the instruction :

unsigned long ulongMax = pow(2.0, 64.0) - 1;

But when I explicitly cast this like below, it's ok !

unsigned long ulongMax = (unsigned long) (pow(2.0, 64.0) - 1);

I don't understand why i have a warning, the result (18446744073709551615) is the same than ULONG_MAX from the header "limits.h".

Your question should include the exact warning message, and it'be good to confirm the size of ulong, e.g. `printf("%lu\n", ULONG_MAX);` — M.M, Nov 26 '15 at 23:57
I've included the warning message. On my system (Arch linux 64bits), ULONG_MAX value is 18446744073709551615 as i said. — Burrich, Nov 27 '15 at 10:11

score 3 · Accepted Answer · edited May 23 '17 at 10:27

pow(2.0, 64.0) returns a double.

However (assuming a normal IEEE754 system), the values pow(2.0, 64.0) and pow(2.0, 64.0) - 1 are actually equal. This is because we are outside the range where adjacent integers are exactly representible in double. (Of course, a 64-bit double cannot represent all 64-bit integers).

Now, out-of-range casts from floating-point to integer type causes undefined behaviour, with no diagnostic required.

Your compiler is trying to be helpful by warning you about this undefined behaviour in the first case, but (presumably) it treats the addition of the cast as a message from you saying "I don't want to hear this warning".

In my mind, a double was represented like an int, but yes you're right. So the maximum value of an integer inside a double should be 2^52 -1. Thank you. — Burrich, Nov 27 '15 at 09:59

score 2 · Answer 2 · answered Nov 26 '15 at 23:41

the warning means that it potentialy could loss precision if pow(2.0, 64.0) - 1 were too big or with fractional part (say 1.7*10^308, which is the max for doubles) or 0.9, which would be truncated to 0).

The reason you don't get the warning when using explicit cast (unsigned long) (pow(2.0, 64.0) - 1) is that being explicit you say "I actually want to get a unsigned long from this double (whatever the nasty consequences of it)"

score 2 · Answer 3 · answered Nov 26 '15 at 23:45

The most likely reason why you get a warning is that unsigned long on your platform is 32-bit, not 64-bit, in size.

If you switch to a 64-bit unsigned, the warning would go away:

unsigned long long ulongMax = pow(2.0, 64.0) - 1; 
printf("%llu", ulongMax);

Demo.

Explicit cast eliminates the warning because essentially it tells the compiler that you are aware of what is going on, and you want it to be quiet and do as you say.

Overflow after an implicit cast

3 Answers3