Rational number support for integer exponentiation?

[dcnorris]

This post by @triska brought Scryer's built-in rational number support to my attention. Because of results like this ...

?- checkprob(2, 12, Prob).
% |Qfs| = 306
   Prob = 1.0000000000000009. % 8^(

... I would like to do some probability calculations exactly. This requires doing $q^n$ for $q \in \mathbb{Q}, n \in \mathbb{N}$. But both 'obvious' operators yield a floating-point result:

?- R is 1 rdiv 5, R2a is R^2, R2b is R**2.
   R = 1 rdiv 5, R2a = 0.04000000000000001, R2b = 0.04000000000000001.

Is Scryer's rational number support documented anywhere?

[triska]

It seems to me that (^)/2 better yield a rational number in this case!

As a workaround, you can perform multiplications instead for now.

[dcnorris]

Should I submit an Issue, then?

[dcnorris]

Now #2753

[dcnorris]

Ah, I see it is ^ that corresponds to (WAM?) instruction IntPow, and which should therefore be expected to yield the rational result:

scryer-prolog/src/arithmetic.rs

Lines 237 to 238 in 2856cc4

	atom!("**") => Ok(Instruction::Pow(a1, a2, t)),
	atom!("^") => Ok(Instruction::IntPow(a1, a2, t)),

[dcnorris]

Comparing implementations of IntPow and Pow:

scryer-prolog/src/machine/dispatch.rs

Lines 649 to 659 in 2856cc4

	&Instruction::IntPow(ref a1, ref a2, t) => {
	let n1 = try_or_throw!(self.machine_st, self.machine_st.get_number(a1));
	let n2 = try_or_throw!(self.machine_st, self.machine_st.get_number(a2));
	self.machine_st.interms[t - 1] = try_or_throw_gen!(
	&mut self.machine_st,
	int_pow(n1, n2, &mut self.machine_st.arena)
	);
	self.machine_st.p += 1;
	}

scryer-prolog/src/machine/dispatch.rs

Lines 671 to 679 in 2856cc4

	&Instruction::Pow(ref a1, ref a2, t) => {
	let n1 = try_or_throw!(self.machine_st, self.machine_st.get_number(a1));
	let n2 = try_or_throw!(self.machine_st, self.machine_st.get_number(a2));
	self.machine_st.interms[t - 1] =
	try_or_throw_gen!(&mut self.machine_st, pow(n1, n2, atom!("**")));
	self.machine_st.p += 1;
	}

... apart from the formatting, the only differences I see are between (Rust functions?) int_pow vs pow, and the presence of atom!("**") in the invocation of the latter.