Why is there a param_type sub class in random distributions?

[thijs]

Can someone motivate the design choice that led to random distributions having their parameters stored in the param_type sub class?

E.g. the uniform integer distribution has two parameter a and b that specify the bounds.

uniform_int_distribution<> dice(1,6);

dice.a(); // gives a
dice.b(); // gives b

However, all distributions also have a param() member funtion and a nested class param_type that captures all the distribution parameters.

dice.param().a() // give also a
dice.param().b() // give also b


Why has it been decided to move the parameters of the distribution to a nested class param_type?? Why not simply store them in the distribution class directly?

 

[Stefan Ram]

Why has it been decided to move the parameters of the
distribution to a nested class param_type?? Why not simply
store them in the distribution class directly?

A param_type object can also be passed for temporary use in
a call of the distribution. It is not only used as a nested
type. So it has a certain autonomy. When the parameters are
bundled as an object, it might be more easy to safe or
restore them together as a unit with a single expression.