Table of Contents
Integer overflows
Ints have a fixed size.
For x86-64 Linux:
| type | bits | min | max |
|---|
| signed char | 8 | $-2^{7}$ | $2^{7}-1$ |
| unsigned char | 8 | 0 | $2^{8}-1$ |
| short | 16 | $-2^{15}$ | $2^{15}-1$ |
| unsigned short | 16 | 0 | $2^{16}-1$ |
| int | 32 | $-2^{31}$ | $2^{31}-1$ |
| unsigned int | 32 | 0 | $2^{32}-1$ |
| long | 64 | $-2^{63}$ | $2^{63}-1$ |
| unsigned long | 64 | 0 | $2^{64}-1$ |
If number doesn’t fit, overflow. CPU discards bits that don’t fit.
i.e. result is computed modulo 2ⁿ (n = number of bits).
leads to unexpected results in casts, computation, comparison:
- truncation: cast to smaller type, discarding bits
- arithmetic overflow: wrap around
- signedness: negative int interpreted as unsigned