Question

When calling hash on the built in NSData class in the foundation framework — what implementation is used to return the hash value? (CRC32, something else?)

Answer 1

Something else. Actually it’s an implementation detail, which does not need to use a fixed algorithm in different versions.

You can check the implementation in the open source version of Core Foundation. Note that NSData is toll-free bridged to CFDataRef. From http://opensource.apple.com/source/CF/CF-635.21/CFData.c:

static CFHashCode __CFDataHash(CFTypeRef cf) {
    CFDataRef data = (CFDataRef)cf;
    return CFHashBytes((uint8_t *)CFDataGetBytePtr(data), __CFMin(__CFDataLength(data), 80));
}

we see that the first 80 bytes are used to compute the hash. The function CFHashBytes is implemented as using the ELF hash algorithm:

#define ELF_STEP(B) T1 = (H << 4) + B; T2 = T1 & 0xF0000000; if (T2) T1 ^= (T2 >> 24); T1 &= (~T2); H = T1;

CFHashCode CFHashBytes(uint8_t *bytes, CFIndex length) {
    /* The ELF hash algorithm, used in the ELF object file format */
    UInt32 H = 0, T1, T2;
    SInt32 rem = length;
    while (3 < rem) {
    ELF_STEP(bytes[length - rem]);
    ELF_STEP(bytes[length - rem + 1]);
    ELF_STEP(bytes[length - rem + 2]);
    ELF_STEP(bytes[length - rem + 3]);
    rem -= 4;
    }
    switch (rem) {
    case 3:  ELF_STEP(bytes[length - 3]);
    case 2:  ELF_STEP(bytes[length - 2]);
    case 1:  ELF_STEP(bytes[length - 1]);
    case 0:  ;
    }
    return H;
}

#undef ELF_STEP