Make error messages a bit more consistent

This commit is contained in:
Laurenz 2023-04-04 14:33:43 +02:00
parent 4aa5e0a4a3
commit e77abf842d
2 changed files with 32 additions and 31 deletions

View File

@ -99,28 +99,22 @@ pub fn pow(
bail!(span, "exponent is too large")
}
Num::Float(f) if !f.is_normal() && f != 0.0 => {
bail!(span, "exponent may not be NaN, infinite, or subnormal")
bail!(span, "exponent may not be infinite, subnormal, or NaN")
}
_ => {}
};
let return_value = match (base, exp) {
(Num::Int(a), Num::Int(b)) if b >= 0 => Value::Int(a.pow(b as u32)),
(a, Num::Int(b)) => Value::Float(a.float().powi(b as i32)),
(a, b) => Value::Float(a.float().powf(b.float())),
let result = match (base, exp) {
(Num::Int(a), Num::Int(b)) if b >= 0 => Num::Int(a.pow(b as u32)),
(a, Num::Int(b)) => Num::Float(a.float().powi(b as i32)),
(a, b) => Num::Float(a.float().powf(b.float())),
};
let is_nan = match return_value {
Value::Float(f) => f.is_nan(),
Value::Int(i) => (i as f64).is_nan(),
_ => false,
};
if is_nan {
bail!(span, "the return value is not a real number")
if result.float().is_nan() {
bail!(span, "the result is not a real number")
}
return_value
result.value()
}
/// Calculate the square root of a number.
@ -243,7 +237,7 @@ pub fn asin(
) -> Value {
let val = value.v.float();
if val < -1.0 || val > 1.0 {
bail!(value.span, "arcsin must be between -1 and 1");
bail!(value.span, "value must be between -1 and 1");
}
Value::Angle(Angle::rad(val.asin()))
}
@ -266,7 +260,7 @@ pub fn acos(
) -> Value {
let val = value.v.float();
if val < -1.0 || val > 1.0 {
bail!(value.span, "arccos must be between -1 and 1");
bail!(value.span, "value must be between -1 and 1");
}
Value::Angle(Angle::rad(val.acos()))
}
@ -379,23 +373,23 @@ pub fn tanh(
/// Returns: float
#[func]
pub fn log(
/// The number whose logarithm to calculate. It must be strictly positive.
/// The number whose logarithm to calculate. Must be strictly positive.
value: Spanned<Num>,
/// The base of the logarithm. It can't be null.
/// The base of the logarithm. Defaults to `{10}` and may not be zero.
#[named]
#[default(10.0)]
base: f64,
) -> Value {
let number = value.v.float();
if number <= 0 as f64 {
bail!(value.span, "a logarithm parameter must be strictly positive")
if number <= 0.0 {
bail!(value.span, "value must be strictly positive")
}
if !base.is_normal() {
bail!(value.span, "a logarithm base should be normal (not NaN, not infinite, non-null, not subnormal)")
bail!(value.span, "base may not be zero, NaN, infinite, or subnormal")
}
let return_value = if base == 2.0 {
let result = if base == 2.0 {
number.log2()
} else if base == 10.0 {
number.log10()
@ -403,11 +397,11 @@ pub fn log(
number.log(base)
};
if return_value.is_infinite() || return_value.is_nan() {
bail!(value.span, "this logarithm doesn't return a real value")
if result.is_infinite() || result.is_nan() {
bail!(value.span, "the result is not a real number")
}
Value::Float(return_value)
Value::Float(result)
}
/// Round a number down to the nearest integer.
@ -695,6 +689,13 @@ impl Num {
Self::Float(v) => v,
}
}
fn value(self) -> Value {
match self {
Self::Int(v) => Value::Int(v),
Self::Float(v) => Value::Float(v),
}
}
}
cast_from_value! {

View File

@ -90,11 +90,11 @@
#calc.pow(2, 10000000000000000)
---
// Error: 14-36 exponent may not be NaN, infinite, or subnormal
// Error: 14-36 exponent may not be infinite, subnormal, or NaN
#calc.pow(2, calc.pow(2.0, 10000.0))
---
// Error: 15-18 the return value is not a real number
// Error: 15-18 the result is not a real number
#calc.pow(-1, 0.5)
---
@ -102,15 +102,15 @@
#calc.sqrt(-1)
---
// Error: 11-13 a logarithm parameter must be strictly positive
// Error: 11-13 value must be strictly positive
#calc.log(-1)
---
// Error: 11-12 a logarithm base should be normal (not NaN, not infinite, non-null, not subnormal)
// Error: 11-12 base may not be zero, NaN, infinite, or subnormal
#calc.log(1, base: 0)
---
// Error: 11-13 this logarithm doesn't return a real value
// Error: 11-13 the result is not a real number
#calc.log(10, base: -1)
---