Move some math functions into plugins.
This commit is contained in:
parent
631c4be064
commit
b956639c2a
@ -1,5 +1,7 @@
|
||||
use crate::def_package;
|
||||
use crate::module::Module;
|
||||
use crate::parser::INT;
|
||||
use crate::plugin::*;
|
||||
|
||||
#[cfg(not(feature = "no_float"))]
|
||||
use crate::parser::FLOAT;
|
||||
@ -22,35 +24,105 @@ pub const MAX_INT: INT = i32::MAX;
|
||||
pub const MAX_INT: INT = i64::MAX;
|
||||
|
||||
def_package!(crate:BasicMathPackage:"Basic mathematic functions.", lib, {
|
||||
init_module(lib);
|
||||
});
|
||||
|
||||
#[cfg(not(feature = "no_float"))]
|
||||
#[export_module]
|
||||
mod trig {
|
||||
use crate::parser::FLOAT;
|
||||
|
||||
pub fn sin(x: FLOAT) -> FLOAT {
|
||||
x.to_radians().sin()
|
||||
}
|
||||
pub fn cos(x: FLOAT) -> FLOAT {
|
||||
x.to_radians().cos()
|
||||
}
|
||||
pub fn tan(x: FLOAT) -> FLOAT {
|
||||
x.to_radians().tan()
|
||||
}
|
||||
pub fn sinh(x: FLOAT) -> FLOAT {
|
||||
x.to_radians().sinh()
|
||||
}
|
||||
pub fn cosh(x: FLOAT) -> FLOAT {
|
||||
x.to_radians().cosh()
|
||||
}
|
||||
pub fn tanh(x: FLOAT) -> FLOAT {
|
||||
x.to_radians().tanh()
|
||||
}
|
||||
pub fn asin(x: FLOAT) -> FLOAT {
|
||||
x.asin().to_degrees()
|
||||
}
|
||||
pub fn acos(x: FLOAT) -> FLOAT {
|
||||
x.acos().to_degrees()
|
||||
}
|
||||
pub fn atan(x: FLOAT) -> FLOAT {
|
||||
x.atan().to_degrees()
|
||||
}
|
||||
pub fn asinh(x: FLOAT) -> FLOAT {
|
||||
x.asinh().to_degrees()
|
||||
}
|
||||
pub fn acosh(x: FLOAT) -> FLOAT {
|
||||
x.acosh().to_degrees()
|
||||
}
|
||||
pub fn atanh(x: FLOAT) -> FLOAT {
|
||||
x.atanh().to_degrees()
|
||||
}
|
||||
}
|
||||
|
||||
#[cfg(not(feature = "no_float"))]
|
||||
#[export_module]
|
||||
mod float {
|
||||
use crate::parser::FLOAT;
|
||||
|
||||
pub fn sqrt(x: FLOAT) -> FLOAT {
|
||||
x.sqrt()
|
||||
}
|
||||
pub fn exp(x: FLOAT) -> FLOAT {
|
||||
x.exp()
|
||||
}
|
||||
pub fn ln(x: FLOAT) -> FLOAT {
|
||||
x.ln()
|
||||
}
|
||||
pub fn log(x: FLOAT, base: FLOAT) -> FLOAT {
|
||||
x.log(base)
|
||||
}
|
||||
pub fn log10(x: FLOAT) -> FLOAT {
|
||||
x.log10()
|
||||
}
|
||||
pub fn floor(x: FLOAT) -> FLOAT {
|
||||
x.floor()
|
||||
}
|
||||
pub fn ceiling(x: FLOAT) -> FLOAT {
|
||||
x.ceil()
|
||||
}
|
||||
pub fn round(x: FLOAT) -> FLOAT {
|
||||
x.ceil()
|
||||
}
|
||||
pub fn int(x: FLOAT) -> FLOAT {
|
||||
x.trunc()
|
||||
}
|
||||
pub fn fraction(x: FLOAT) -> FLOAT {
|
||||
x.fract()
|
||||
}
|
||||
pub fn is_nan(x: FLOAT) -> bool {
|
||||
x.is_nan()
|
||||
}
|
||||
pub fn is_finite(x: FLOAT) -> bool {
|
||||
x.is_finite()
|
||||
}
|
||||
pub fn is_infinite(x: FLOAT) -> bool {
|
||||
x.is_infinite()
|
||||
}
|
||||
}
|
||||
|
||||
fn init_module(lib: &mut Module) {
|
||||
#[cfg(not(feature = "no_float"))]
|
||||
{
|
||||
// Advanced math functions
|
||||
lib.set_fn_1("sin", |x: FLOAT| Ok(x.to_radians().sin()));
|
||||
lib.set_fn_1("cos", |x: FLOAT| Ok(x.to_radians().cos()));
|
||||
lib.set_fn_1("tan", |x: FLOAT| Ok(x.to_radians().tan()));
|
||||
lib.set_fn_1("sinh", |x: FLOAT| Ok(x.to_radians().sinh()));
|
||||
lib.set_fn_1("cosh", |x: FLOAT| Ok(x.to_radians().cosh()));
|
||||
lib.set_fn_1("tanh", |x: FLOAT| Ok(x.to_radians().tanh()));
|
||||
lib.set_fn_1("asin", |x: FLOAT| Ok(x.asin().to_degrees()));
|
||||
lib.set_fn_1("acos", |x: FLOAT| Ok(x.acos().to_degrees()));
|
||||
lib.set_fn_1("atan", |x: FLOAT| Ok(x.atan().to_degrees()));
|
||||
lib.set_fn_1("asinh", |x: FLOAT| Ok(x.asinh().to_degrees()));
|
||||
lib.set_fn_1("acosh", |x: FLOAT| Ok(x.acosh().to_degrees()));
|
||||
lib.set_fn_1("atanh", |x: FLOAT| Ok(x.atanh().to_degrees()));
|
||||
lib.set_fn_1("sqrt", |x: FLOAT| Ok(x.sqrt()));
|
||||
lib.set_fn_1("exp", |x: FLOAT| Ok(x.exp()));
|
||||
lib.set_fn_1("ln", |x: FLOAT| Ok(x.ln()));
|
||||
lib.set_fn_2("log", |x: FLOAT, base: FLOAT| Ok(x.log(base)));
|
||||
lib.set_fn_1("log10", |x: FLOAT| Ok(x.log10()));
|
||||
lib.set_fn_1("floor", |x: FLOAT| Ok(x.floor()));
|
||||
lib.set_fn_1("ceiling", |x: FLOAT| Ok(x.ceil()));
|
||||
lib.set_fn_1("round", |x: FLOAT| Ok(x.ceil()));
|
||||
lib.set_fn_1("int", |x: FLOAT| Ok(x.trunc()));
|
||||
lib.set_fn_1("fraction", |x: FLOAT| Ok(x.fract()));
|
||||
lib.set_fn_1("is_nan", |x: FLOAT| Ok(x.is_nan()));
|
||||
lib.set_fn_1("is_finite", |x: FLOAT| Ok(x.is_finite()));
|
||||
lib.set_fn_1("is_infinite", |x: FLOAT| Ok(x.is_infinite()));
|
||||
// Floating point functions
|
||||
lib.merge(&exported_module!(float));
|
||||
|
||||
// Floating point properties
|
||||
#[cfg(not(feature = "no_object"))]
|
||||
{
|
||||
lib.set_getter_fn("floor", |x: &mut FLOAT| Ok(x.floor()));
|
||||
@ -63,6 +135,9 @@ def_package!(crate:BasicMathPackage:"Basic mathematic functions.", lib, {
|
||||
lib.set_getter_fn("is_infinite", |x: &mut FLOAT| Ok(x.is_infinite()));
|
||||
}
|
||||
|
||||
// Trig functions
|
||||
lib.merge(&exported_module!(trig));
|
||||
|
||||
// Register conversion functions
|
||||
lib.set_fn_1("to_float", |x: INT| Ok(x as FLOAT));
|
||||
lib.set_fn_1("to_float", |x: f32| Ok(x as FLOAT));
|
||||
@ -105,32 +180,28 @@ def_package!(crate:BasicMathPackage:"Basic mathematic functions.", lib, {
|
||||
#[cfg(not(feature = "no_float"))]
|
||||
{
|
||||
if cfg!(not(feature = "unchecked")) {
|
||||
lib.set_fn_1(
|
||||
"to_int",
|
||||
|x: f32| {
|
||||
lib.set_fn_1("to_int", |x: f32| {
|
||||
if x > (MAX_INT as f32) {
|
||||
return EvalAltResult::ErrorArithmetic(
|
||||
format!("Integer overflow: to_int({})", x),
|
||||
Position::none(),
|
||||
).into();
|
||||
)
|
||||
.into();
|
||||
}
|
||||
|
||||
Ok(x.trunc() as INT)
|
||||
},
|
||||
);
|
||||
lib.set_fn_1(
|
||||
"to_int",
|
||||
|x: FLOAT| {
|
||||
});
|
||||
lib.set_fn_1("to_int", |x: FLOAT| {
|
||||
if x > (MAX_INT as FLOAT) {
|
||||
return EvalAltResult::ErrorArithmetic(
|
||||
format!("Integer overflow: to_int({})", x),
|
||||
Position::none(),
|
||||
).into();
|
||||
)
|
||||
.into();
|
||||
}
|
||||
|
||||
Ok(x.trunc() as INT)
|
||||
},
|
||||
);
|
||||
});
|
||||
}
|
||||
|
||||
if cfg!(feature = "unchecked") {
|
||||
@ -138,4 +209,4 @@ def_package!(crate:BasicMathPackage:"Basic mathematic functions.", lib, {
|
||||
lib.set_fn_1("to_int", |x: f64| Ok(x as INT));
|
||||
}
|
||||
}
|
||||
});
|
||||
}
|
||||
|
@ -1,5 +1,8 @@
|
||||
use rhai::{Engine, EvalAltResult, INT};
|
||||
|
||||
#[cfg(not(feature = "no_float"))]
|
||||
use rhai::FLOAT;
|
||||
|
||||
#[test]
|
||||
fn test_math() -> Result<(), Box<EvalAltResult>> {
|
||||
let engine = Engine::new();
|
||||
@ -10,6 +13,12 @@ fn test_math() -> Result<(), Box<EvalAltResult>> {
|
||||
assert_eq!(engine.eval::<INT>("1 / 2")?, 0);
|
||||
assert_eq!(engine.eval::<INT>("3 % 2")?, 1);
|
||||
|
||||
#[cfg(not(feature = "no_float"))]
|
||||
assert!((engine.eval::<FLOAT>("sin(30.0)")? - 0.5).abs() < 0.001);
|
||||
|
||||
#[cfg(not(feature = "no_float"))]
|
||||
assert!(engine.eval::<FLOAT>("cos(90.0)")?.abs() < 0.001);
|
||||
|
||||
#[cfg(not(feature = "only_i32"))]
|
||||
assert_eq!(
|
||||
engine.eval::<INT>("abs(-9223372036854775807)")?,
|
||||
|
Loading…
Reference in New Issue
Block a user