factor size zero tests into variadic function

[bob-carpenter]

Summary:

There is code duplication in checking if any of the inputs to a probability funciton are size zero. These should all be replaceable by a single variadic C++11 function.

Description:

The non-variadic version for three arguments would replace

!(stan::length(y) && stan::length(alpha) && stan::length(beta))

with

size_zero(y, alpha, beta)

defined in the fixed size case by

template <typename T1, typename T2, typename T3>
bool size_zero(const T1& x1, const T2& x2, const T3& x3) {
  return !length(x1) || !length(x2) || !length(x3);
}

Distributing the negation through the conjunction leads to the simple recursive form for a variadic template:

template <typename T>
bool size_zero(T& x) {
  return !length(x);
}

template <typename T, typename ... Ts>
bool size_zero(T& x, Ts&&... xs) {
  return size_zero(x) || size_zero(std::forward<Ts>(xs)...);
}

I tested that the above works. You need the std::forward to "unpack" the parameter pack xs and std::forward with that whacky ... syntax to allow it to be passed to another argument. The const gets rolled into the T and Ts on the call.

This should be defined in the lowest typed prim/*/fun where length is available to be imported.

Current Version:

v2.16.0

[andrjohns]

Simplifying the variadic a little gives:

template <typename T>
bool size_zero(const T& x) {
  return !length(x);
}

template <typename T, typename ... Ts>
bool size_zero(const T& x, const Ts& ... xs) {
  return (size_zero(x) || size_zero(xs...));
}

Still getting my head around the stan code structure (the wiki is helping greatly), so apologies in advance for stupid questions. Did you want this defined as a new function (i.e. math/prim/scal/fun/size_zero.hpp), with unit tests?

[bob-carpenter]

Yes, I think it needs to be a new function. I'm pretty sure you need && and std::forward to pass by reference as I wrote in the issue.

[bob-carpenter]

The point isn't so much the new function as pushing it through all the probability functions.

[andrjohns]

Yep, my mistake, you do need those. I'll make the changes to the probability functions and include those in the pull request.

[syclik]

Closing. It looks like these were fixed.