Summary: Arrays in Phobos shold be pre-allocated.
Looking at the sources to Phobos, I see several places, including
std.algorithm.map(), and the array-to-array conversion function of std.conv,
where the ~= operator is used, even though the length of the array returned by
these functions is known at the beginning of the function. This severely hurts
the performance of these functions, given how slow ~= is. Furthermore, even if
~= were efficient, reallocating an array instead of preallocating is still too
inefficient when it can be trivially optimized out.