Reasons behind std::discrete_distribution interface

According to cppreference.com std::discrete_distribution interface requires library developers to implement probabilities() and template<class Generator> result_type operator()(Generator& g, const param_type& params);. The latter is not documented on cppreference.com, but according to libc++ implementation it allows user to sample from subsequence of given sequence of weights (and use passed generator as source of entropy for another generator, but it's irrelevant now). I've read N3551 (the only googleable proposal about std::discrete_distribution) and it's doesn't provide any reasons for such an interface for std::discrete_distribution.

The problem is that such an interface allows only one reasonable implementation, known as "roulette wheel selection" (requires one call to random number generator and O(log(N)) array lookups for binary search). Another algorithm, known as "alias method" (requires two calls to random number generator and one array lookup) can't be used to implement this interface (well, probabilities() can be implemented if we will store corresponding probabilities in a separate array, but it's sort of unfair and inefficient :), because alias method can't be used if we need to sample from subsequences.