Exploit pointer equality in Eq (memcmp does not seem to do this)

[andreasabel]

Eq for Data.Text is implemented in terms of Data.Text.Array.equal, which call the C memcmp function.

int _hs_text_memcmp(const void *a, size_t aoff, const void *b, size_t boff,
size_t n)
{
  return memcmp(a + (aoff<<1), b + (boff<<1), n<<1);
}

However, the C implementation does not seem to check for pointer equality first. Would it not make sense to check equality of a + (aoff<<1) and b + (boff<<1) first? That would bring the equality test down from O(n) to O(1) for identical texts!

[bos]

You are correct that this is possible, but I would be surprised if it's common enough to offset the small performance hit of the added compare-and-branch. I am open to data proving otherwise, of course.

[andreasabel]

Fair enough. At least I could not notice a different on my (large) test case.