use super::{reg_binary, reg_unary}; use crate::def_package; use crate::fn_register::{map_dynamic as map, map_result as result}; use crate::parser::INT; use crate::result::EvalAltResult; use crate::token::Position; #[cfg(not(feature = "no_float"))] use crate::parser::FLOAT; use num_traits::{ identities::Zero, CheckedAdd, CheckedDiv, CheckedMul, CheckedNeg, CheckedRem, CheckedShl, CheckedShr, CheckedSub, }; use crate::stdlib::{ fmt::Display, format, ops::{Add, BitAnd, BitOr, BitXor, Div, Mul, Neg, Rem, Shl, Shr, Sub}, {i32, i64, u32}, }; // Checked add fn add(x: T, y: T) -> Result { x.checked_add(&y).ok_or_else(|| { EvalAltResult::ErrorArithmetic( format!("Addition overflow: {} + {}", x, y), Position::none(), ) }) } // Checked subtract fn sub(x: T, y: T) -> Result { x.checked_sub(&y).ok_or_else(|| { EvalAltResult::ErrorArithmetic( format!("Subtraction underflow: {} - {}", x, y), Position::none(), ) }) } // Checked multiply fn mul(x: T, y: T) -> Result { x.checked_mul(&y).ok_or_else(|| { EvalAltResult::ErrorArithmetic( format!("Multiplication overflow: {} * {}", x, y), Position::none(), ) }) } // Checked divide fn div(x: T, y: T) -> Result where T: Display + CheckedDiv + PartialEq + Zero, { // Detect division by zero if y == T::zero() { return Err(EvalAltResult::ErrorArithmetic( format!("Division by zero: {} / {}", x, y), Position::none(), )); } x.checked_div(&y).ok_or_else(|| { EvalAltResult::ErrorArithmetic( format!("Division overflow: {} / {}", x, y), Position::none(), ) }) } // Checked negative - e.g. -(i32::MIN) will overflow i32::MAX fn neg(x: T) -> Result { x.checked_neg().ok_or_else(|| { EvalAltResult::ErrorArithmetic(format!("Negation overflow: -{}", x), Position::none()) }) } // Checked absolute fn abs(x: T) -> Result { // FIX - We don't use Signed::abs() here because, contrary to documentation, it panics // when the number is ::MIN instead of returning ::MIN itself. if x >= ::zero() { Ok(x) } else { x.checked_neg().ok_or_else(|| { EvalAltResult::ErrorArithmetic(format!("Negation overflow: -{}", x), Position::none()) }) } } // Unchecked add - may panic on overflow fn add_u(x: T, y: T) -> ::Output { x + y } // Unchecked subtract - may panic on underflow fn sub_u(x: T, y: T) -> ::Output { x - y } // Unchecked multiply - may panic on overflow fn mul_u(x: T, y: T) -> ::Output { x * y } // Unchecked divide - may panic when dividing by zero fn div_u(x: T, y: T) -> ::Output { x / y } // Unchecked negative - may panic on overflow fn neg_u(x: T) -> ::Output { -x } // Unchecked absolute - may panic on overflow fn abs_u(x: T) -> ::Output where T: Neg + PartialOrd + Default + Into<::Output>, { // Numbers should default to zero if x < Default::default() { -x } else { x.into() } } // Bit operators fn binary_and(x: T, y: T) -> ::Output { x & y } fn binary_or(x: T, y: T) -> ::Output { x | y } fn binary_xor(x: T, y: T) -> ::Output { x ^ y } // Checked left-shift fn shl(x: T, y: INT) -> Result { // Cannot shift by a negative number of bits if y < 0 { return Err(EvalAltResult::ErrorArithmetic( format!("Left-shift by a negative number: {} << {}", x, y), Position::none(), )); } CheckedShl::checked_shl(&x, y as u32).ok_or_else(|| { EvalAltResult::ErrorArithmetic( format!("Left-shift by too many bits: {} << {}", x, y), Position::none(), ) }) } // Checked right-shift fn shr(x: T, y: INT) -> Result { // Cannot shift by a negative number of bits if y < 0 { return Err(EvalAltResult::ErrorArithmetic( format!("Right-shift by a negative number: {} >> {}", x, y), Position::none(), )); } CheckedShr::checked_shr(&x, y as u32).ok_or_else(|| { EvalAltResult::ErrorArithmetic( format!("Right-shift by too many bits: {} % {}", x, y), Position::none(), ) }) } // Unchecked left-shift - may panic if shifting by a negative number of bits fn shl_u>(x: T, y: T) -> >::Output { x.shl(y) } // Unchecked right-shift - may panic if shifting by a negative number of bits fn shr_u>(x: T, y: T) -> >::Output { x.shr(y) } // Checked modulo fn modulo(x: T, y: T) -> Result { x.checked_rem(&y).ok_or_else(|| { EvalAltResult::ErrorArithmetic( format!("Modulo division by zero or overflow: {} % {}", x, y), Position::none(), ) }) } // Unchecked modulo - may panic if dividing by zero fn modulo_u(x: T, y: T) -> ::Output { x % y } // Checked power fn pow_i_i(x: INT, y: INT) -> Result { #[cfg(not(feature = "only_i32"))] { if y > (u32::MAX as INT) { Err(EvalAltResult::ErrorArithmetic( format!("Integer raised to too large an index: {} ~ {}", x, y), Position::none(), )) } else if y < 0 { Err(EvalAltResult::ErrorArithmetic( format!("Integer raised to a negative index: {} ~ {}", x, y), Position::none(), )) } else { x.checked_pow(y as u32).ok_or_else(|| { EvalAltResult::ErrorArithmetic( format!("Power overflow: {} ~ {}", x, y), Position::none(), ) }) } } #[cfg(feature = "only_i32")] { if y < 0 { Err(EvalAltResult::ErrorArithmetic( format!("Integer raised to a negative index: {} ~ {}", x, y), Position::none(), )) } else { x.checked_pow(y as u32).ok_or_else(|| { EvalAltResult::ErrorArithmetic( format!("Power overflow: {} ~ {}", x, y), Position::none(), ) }) } } } // Unchecked integer power - may panic on overflow or if the power index is too high (> u32::MAX) fn pow_i_i_u(x: INT, y: INT) -> INT { x.pow(y as u32) } // Floating-point power - always well-defined #[cfg(not(feature = "no_float"))] fn pow_f_f(x: FLOAT, y: FLOAT) -> FLOAT { x.powf(y) } // Checked power #[cfg(not(feature = "no_float"))] fn pow_f_i(x: FLOAT, y: INT) -> Result { // Raise to power that is larger than an i32 if y > (i32::MAX as INT) { return Err(EvalAltResult::ErrorArithmetic( format!("Number raised to too large an index: {} ~ {}", x, y), Position::none(), )); } Ok(x.powi(y as i32)) } // Unchecked power - may be incorrect if the power index is too high (> i32::MAX) #[cfg(feature = "unchecked")] #[cfg(not(feature = "no_float"))] fn pow_f_i_u(x: FLOAT, y: INT) -> FLOAT { x.powi(y as i32) } macro_rules! reg_unary_x { ($lib:expr, $op:expr, $func:ident, $($par:ty),*) => { $(reg_unary($lib, $op, $func::<$par>, result);)* }; } macro_rules! reg_unary { ($lib:expr, $op:expr, $func:ident, $($par:ty),*) => { $(reg_unary($lib, $op, $func::<$par>, map);)* }; } macro_rules! reg_op_x { ($lib:expr, $op:expr, $func:ident, $($par:ty),*) => { $(reg_binary($lib, $op, $func::<$par>, result);)* }; } macro_rules! reg_op { ($lib:expr, $op:expr, $func:ident, $($par:ty),*) => { $(reg_binary($lib, $op, $func::<$par>, map);)* }; } def_package!(ArithmeticPackage:"Basic arithmetic", lib, { // Checked basic arithmetic #[cfg(not(feature = "unchecked"))] { reg_op_x!(lib, "+", add, INT); reg_op_x!(lib, "-", sub, INT); reg_op_x!(lib, "*", mul, INT); reg_op_x!(lib, "/", div, INT); #[cfg(not(feature = "only_i32"))] #[cfg(not(feature = "only_i64"))] { reg_op_x!(lib, "+", add, i8, u8, i16, u16, i32, i64, u32, u64, i128, u128); reg_op_x!(lib, "-", sub, i8, u8, i16, u16, i32, i64, u32, u64, i128, u128); reg_op_x!(lib, "*", mul, i8, u8, i16, u16, i32, i64, u32, u64, i128, u128); reg_op_x!(lib, "/", div, i8, u8, i16, u16, i32, i64, u32, u64, i128, u128); } } // Unchecked basic arithmetic #[cfg(feature = "unchecked")] { reg_op!(lib, "+", add_u, INT); reg_op!(lib, "-", sub_u, INT); reg_op!(lib, "*", mul_u, INT); reg_op!(lib, "/", div_u, INT); #[cfg(not(feature = "only_i32"))] #[cfg(not(feature = "only_i64"))] { reg_op!(lib, "+", add_u, i8, u8, i16, u16, i32, i64, u32, u64, i128, u128); reg_op!(lib, "-", sub_u, i8, u8, i16, u16, i32, i64, u32, u64, i128, u128); reg_op!(lib, "*", mul_u, i8, u8, i16, u16, i32, i64, u32, u64, i128, u128); reg_op!(lib, "/", div_u, i8, u8, i16, u16, i32, i64, u32, u64, i128, u128); } } // Basic arithmetic for floating-point - no need to check #[cfg(not(feature = "no_float"))] { reg_op!(lib, "+", add_u, f32, f64); reg_op!(lib, "-", sub_u, f32, f64); reg_op!(lib, "*", mul_u, f32, f64); reg_op!(lib, "/", div_u, f32, f64); } // Bit operations reg_op!(lib, "|", binary_or, INT); reg_op!(lib, "&", binary_and, INT); reg_op!(lib, "^", binary_xor, INT); #[cfg(not(feature = "only_i32"))] #[cfg(not(feature = "only_i64"))] { reg_op!(lib, "|", binary_or, i8, u8, i16, u16, i32, i64, u32, u64, i128, u128); reg_op!(lib, "&", binary_and, i8, u8, i16, u16, i32, i64, u32, u64, i128, u128); reg_op!(lib, "^", binary_xor, i8, u8, i16, u16, i32, i64, u32, u64, i128, u128); } // Checked bit shifts #[cfg(not(feature = "unchecked"))] { reg_op_x!(lib, "<<", shl, INT); reg_op_x!(lib, ">>", shr, INT); reg_op_x!(lib, "%", modulo, INT); #[cfg(not(feature = "only_i32"))] #[cfg(not(feature = "only_i64"))] { reg_op_x!(lib, "<<", shl, i8, u8, i16, u16, i32, i64, u32, u64, i128, u128); reg_op_x!(lib, ">>", shr, i8, u8, i16, u16, i32, i64, u32, u64, i128, u128); reg_op_x!(lib, "%", modulo, i8, u8, i16, u16, i32, i64, u32, u64, i128, u128); } } // Unchecked bit shifts #[cfg(feature = "unchecked")] { reg_op!(lib, "<<", shl_u, INT, INT); reg_op!(lib, ">>", shr_u, INT, INT); reg_op!(lib, "%", modulo_u, INT); #[cfg(not(feature = "only_i32"))] #[cfg(not(feature = "only_i64"))] { reg_op!(lib, "<<", shl_u, i64, i8, u8, i16, u16, i32, i64, u32, u64, i128, u128); reg_op!(lib, ">>", shr_u, i64, i8, u8, i16, u16, i32, i64, u32, u64, i128, u128); reg_op!(lib, "%", modulo_u, i8, u8, i16, u16, i32, i64, u32, u64, i128, u128); } } // Checked power #[cfg(not(feature = "unchecked"))] { reg_binary(lib, "~", pow_i_i, result); #[cfg(not(feature = "no_float"))] reg_binary(lib, "~", pow_f_i, result); } // Unchecked power #[cfg(feature = "unchecked")] { reg_binary(lib, "~", pow_i_i_u, map); #[cfg(not(feature = "no_float"))] reg_binary(lib, "~", pow_f_i_u, map); } // Floating-point modulo and power #[cfg(not(feature = "no_float"))] { reg_op!(lib, "%", modulo_u, f32, f64); reg_binary(lib, "~", pow_f_f, map); } // Checked unary #[cfg(not(feature = "unchecked"))] { reg_unary_x!(lib, "-", neg, INT); reg_unary_x!(lib, "abs", abs, INT); #[cfg(not(feature = "only_i32"))] #[cfg(not(feature = "only_i64"))] { reg_unary_x!(lib, "-", neg, i8, i16, i32, i64, i128); reg_unary_x!(lib, "abs", abs, i8, i16, i32, i64, i128); } } // Unchecked unary #[cfg(feature = "unchecked")] { reg_unary!(lib, "-", neg_u, INT); reg_unary!(lib, "abs", abs_u, INT); #[cfg(not(feature = "only_i32"))] #[cfg(not(feature = "only_i64"))] { reg_unary!(lib, "-", neg_u, i8, i16, i32, i64, i128); reg_unary!(lib, "abs", abs_u, i8, i16, i32, i64, i128); } } // Floating-point unary #[cfg(not(feature = "no_float"))] { reg_unary!(lib, "-", neg_u, f32, f64); reg_unary!(lib, "abs", abs_u, f32, f64); } });