Right now we return smallvec, but I'm not sure that it's any faster than vec (it could potentially be slower, since vec doesn't have to move it's backing store).
We should benchmark it both ways, and remove if it doesn't provide any significant speed improvement.
Right now we return smallvec, but I'm not sure that it's any faster than vec (it could potentially be slower, since vec doesn't have to move it's backing store).
We should benchmark it both ways, and remove if it doesn't provide any significant speed improvement.