According to the documentation, “This function is almost ten times
faster than rational, but is slightly less accurate. For 94.2% of
numbers, this function and rational give identical results, but for
the remaining 5.8%, this function loses precision around the 15th
decimal place. For 0.001% of numbers, this function will lose
precision at the 13th or 14th decimal place.”. What happens is that
for our tests, it can happen that “Attoparsec.double (show a_double)”
is quite different from “a_double”, such that we have much more than
1e-12 absolute difference.

Since our xm lists should not be too big, I think switching to
rational is better. Next patch also changes the way we compare
doubles, so maybe this patch is not really needed…

Signed-off-by: Iustin Pop <iustin@google.com>
Reviewed-by: Michele Tartara <mtartara@google.com>