//@no-rustfix #![feature(repr128)] #![allow(incomplete_features)] #![warn( clippy::cast_precision_loss, clippy::cast_possible_truncation, clippy::cast_sign_loss, clippy::cast_possible_wrap )] #![allow(clippy::cast_abs_to_unsigned, clippy::no_effect, clippy::unnecessary_operation)] fn main() { // Test clippy::cast_precision_loss let x0 = 1i32; x0 as f32; //~^ ERROR: casting `i32` to `f32` causes a loss of precision (`i32` is 32 bits wide, //~| NOTE: `-D clippy::cast-precision-loss` implied by `-D warnings` let x1 = 1i64; x1 as f32; //~^ ERROR: casting `i64` to `f32` causes a loss of precision (`i64` is 64 bits wide, x1 as f64; //~^ ERROR: casting `i64` to `f64` causes a loss of precision (`i64` is 64 bits wide, let x2 = 1u32; x2 as f32; //~^ ERROR: casting `u32` to `f32` causes a loss of precision (`u32` is 32 bits wide, let x3 = 1u64; x3 as f32; //~^ ERROR: casting `u64` to `f32` causes a loss of precision (`u64` is 64 bits wide, x3 as f64; //~^ ERROR: casting `u64` to `f64` causes a loss of precision (`u64` is 64 bits wide, // Test clippy::cast_possible_truncation 1f32 as i32; //~^ ERROR: casting `f32` to `i32` may truncate the value 1f32 as u32; //~^ ERROR: casting `f32` to `u32` may truncate the value //~| ERROR: casting `f32` to `u32` may lose the sign of the value //~| NOTE: `-D clippy::cast-sign-loss` implied by `-D warnings` 1f64 as f32; //~^ ERROR: casting `f64` to `f32` may truncate the value 1i32 as i8; //~^ ERROR: casting `i32` to `i8` may truncate the value 1i32 as u8; //~^ ERROR: casting `i32` to `u8` may truncate the value 1f64 as isize; //~^ ERROR: casting `f64` to `isize` may truncate the value 1f64 as usize; //~^ ERROR: casting `f64` to `usize` may truncate the value //~| ERROR: casting `f64` to `usize` may lose the sign of the value 1f32 as u32 as u16; //~^ ERROR: casting `u32` to `u16` may truncate the value //~| ERROR: casting `f32` to `u32` may truncate the value //~| ERROR: casting `f32` to `u32` may lose the sign of the value { let _x: i8 = 1i32 as _; //~^ ERROR: casting `i32` to `i8` may truncate the value 1f32 as i32; //~^ ERROR: casting `f32` to `i32` may truncate the value 1f64 as i32; //~^ ERROR: casting `f64` to `i32` may truncate the value 1f32 as u8; //~^ ERROR: casting `f32` to `u8` may truncate the value //~| ERROR: casting `f32` to `u8` may lose the sign of the value } // Test clippy::cast_possible_wrap 1u8 as i8; //~^ ERROR: casting `u8` to `i8` may wrap around the value //~| NOTE: `-D clippy::cast-possible-wrap` implied by `-D warnings` 1u16 as i16; //~^ ERROR: casting `u16` to `i16` may wrap around the value 1u32 as i32; //~^ ERROR: casting `u32` to `i32` may wrap around the value 1u64 as i64; //~^ ERROR: casting `u64` to `i64` may wrap around the value 1usize as isize; //~^ ERROR: casting `usize` to `isize` may wrap around the value // should not wrap, usize is never 8 bits 1usize as i8; //~^ ERROR: casting `usize` to `i8` may truncate the value // wraps on 16 bit ptr size 1usize as i16; //~^ ERROR: casting `usize` to `i16` may truncate the value //~| ERROR: casting `usize` to `i16` may wrap around the value on targets with 16-bit //~| NOTE: `usize` and `isize` may be as small as 16 bits on some platforms // wraps on 32 bit ptr size 1usize as i32; //~^ ERROR: casting `usize` to `i32` may truncate the value on targets with 64-bit wid //~| ERROR: casting `usize` to `i32` may wrap around the value on targets with 32-bit // wraps on 64 bit ptr size 1usize as i64; //~^ ERROR: casting `usize` to `i64` may wrap around the value on targets with 64-bit // should not wrap, isize is never 8 bits 1u8 as isize; // wraps on 16 bit ptr size 1u16 as isize; //~^ ERROR: casting `u16` to `isize` may wrap around the value on targets with 16-bit //~| NOTE: `usize` and `isize` may be as small as 16 bits on some platforms // wraps on 32 bit ptr size 1u32 as isize; //~^ ERROR: casting `u32` to `isize` may wrap around the value on targets with 32-bit // wraps on 64 bit ptr size 1u64 as isize; //~^ ERROR: casting `u64` to `isize` may truncate the value on targets with 32-bit wid //~| ERROR: casting `u64` to `isize` may wrap around the value on targets with 64-bit // Test clippy::cast_sign_loss 1i32 as u32; -1i32 as u32; //~^ ERROR: casting `i32` to `u32` may lose the sign of the value 1isize as usize; -1isize as usize; //~^ ERROR: casting `isize` to `usize` may lose the sign of the value 0i8 as u8; i8::MAX as u8; i16::MAX as u16; i32::MAX as u32; i64::MAX as u64; i128::MAX as u128; (-1i8).abs() as u8; (-1i16).abs() as u16; (-1i32).abs() as u32; (-1i64).abs() as u64; (-1isize).abs() as usize; (-1i8).checked_abs().unwrap() as u8; (-1i16).checked_abs().unwrap() as u16; (-1i32).checked_abs().unwrap() as u32; (-1i64).checked_abs().unwrap() as u64; (-1isize).checked_abs().unwrap() as usize; (-1i8).rem_euclid(1i8) as u8; (-1i8).rem_euclid(1i8) as u16; (-1i16).rem_euclid(1i16) as u16; (-1i16).rem_euclid(1i16) as u32; (-1i32).rem_euclid(1i32) as u32; (-1i32).rem_euclid(1i32) as u64; (-1i64).rem_euclid(1i64) as u64; (-1i64).rem_euclid(1i64) as u128; (-1isize).rem_euclid(1isize) as usize; (1i8).rem_euclid(-1i8) as u8; (1i8).rem_euclid(-1i8) as u16; (1i16).rem_euclid(-1i16) as u16; (1i16).rem_euclid(-1i16) as u32; (1i32).rem_euclid(-1i32) as u32; (1i32).rem_euclid(-1i32) as u64; (1i64).rem_euclid(-1i64) as u64; (1i64).rem_euclid(-1i64) as u128; (1isize).rem_euclid(-1isize) as usize; (-1i8).checked_rem_euclid(1i8).unwrap() as u8; (-1i8).checked_rem_euclid(1i8).unwrap() as u16; (-1i16).checked_rem_euclid(1i16).unwrap() as u16; (-1i16).checked_rem_euclid(1i16).unwrap() as u32; (-1i32).checked_rem_euclid(1i32).unwrap() as u32; (-1i32).checked_rem_euclid(1i32).unwrap() as u64; (-1i64).checked_rem_euclid(1i64).unwrap() as u64; (-1i64).checked_rem_euclid(1i64).unwrap() as u128; (-1isize).checked_rem_euclid(1isize).unwrap() as usize; (1i8).checked_rem_euclid(-1i8).unwrap() as u8; (1i8).checked_rem_euclid(-1i8).unwrap() as u16; (1i16).checked_rem_euclid(-1i16).unwrap() as u16; (1i16).checked_rem_euclid(-1i16).unwrap() as u32; (1i32).checked_rem_euclid(-1i32).unwrap() as u32; (1i32).checked_rem_euclid(-1i32).unwrap() as u64; (1i64).checked_rem_euclid(-1i64).unwrap() as u64; (1i64).checked_rem_euclid(-1i64).unwrap() as u128; (1isize).checked_rem_euclid(-1isize).unwrap() as usize; // no lint for `cast_possible_truncation` // with `signum` method call (see issue #5395) let x: i64 = 5; let _ = x.signum() as i32; let s = x.signum(); let _ = s as i32; // Test for signed min // should be linted because signed (-99999999999i64).min(1) as i8; //~^ ERROR: casting `i64` to `i8` may truncate the value // Test for various operations that remove enough bits for the result to fit (999999u64 & 1) as u8; (999999u64 % 15) as u8; (999999u64 / 0x1_0000_0000_0000) as u16; ({ 999999u64 >> 56 }) as u8; ({ let x = 999999u64; x.min(1) }) as u8; 999999u64.clamp(0, 255) as u8; // should still be linted 999999u64.clamp(0, 256) as u8; //~^ ERROR: casting `u64` to `u8` may truncate the value #[derive(Clone, Copy)] enum E1 { A, B, C, } impl E1 { fn test(self) { // Don't lint. `0..=2` fits in u8 let _ = self as u8; } } #[derive(Clone, Copy)] enum E2 { A = 255, B, } impl E2 { fn test(self) { let _ = self as u8; //~^ ERROR: casting `main::E2` to `u8` may truncate the value let _ = Self::B as u8; //~^ ERROR: casting `main::E2::B` to `u8` will truncate the value //~| NOTE: `-D clippy::cast-enum-truncation` implied by `-D warnings` // Don't lint. `255..=256` fits in i16 let _ = self as i16; // Don't lint. let _ = Self::A as u8; } } #[derive(Clone, Copy)] enum E3 { A = -1, B, C = 50, } impl E3 { fn test(self) { // Don't lint. `-1..=50` fits in i8 let _ = self as i8; } } #[derive(Clone, Copy)] enum E4 { A = -128, B, } impl E4 { fn test(self) { // Don't lint. `-128..=-127` fits in i8 let _ = self as i8; } } #[derive(Clone, Copy)] enum E5 { A = -129, B = 127, } impl E5 { fn test(self) { let _ = self as i8; //~^ ERROR: casting `main::E5` to `i8` may truncate the value let _ = Self::A as i8; //~^ ERROR: casting `main::E5::A` to `i8` will truncate the value // Don't lint. `-129..=127` fits in i16 let _ = self as i16; // Don't lint. let _ = Self::B as u8; } } #[derive(Clone, Copy)] #[repr(u32)] enum E6 { A = u16::MAX as u32, B, } impl E6 { fn test(self) { let _ = self as i16; //~^ ERROR: casting `main::E6` to `i16` may truncate the value // Don't lint. `2^16-1` fits in u16 let _ = Self::A as u16; // Don't lint. `2^16-1..=2^16` fits in u32 let _ = self as u32; // Don't lint. let _ = Self::A as u16; } } #[derive(Clone, Copy)] #[repr(u64)] enum E7 { A = u32::MAX as u64, B, } impl E7 { fn test(self) { let _ = self as usize; //~^ ERROR: casting `main::E7` to `usize` may truncate the value on targets wi // Don't lint. let _ = Self::A as usize; // Don't lint. `2^32-1..=2^32` fits in u64 let _ = self as u64; } } #[derive(Clone, Copy)] #[repr(i128)] enum E8 { A = i128::MIN, B, C = 0, D = i128::MAX, } impl E8 { fn test(self) { // Don't lint. `-(2^127)..=2^127-1` fits it i128 let _ = self as i128; } } #[derive(Clone, Copy)] #[repr(u128)] enum E9 { A, B = u128::MAX, } impl E9 { fn test(self) { // Don't lint. let _ = Self::A as u8; // Don't lint. `0..=2^128-1` fits in u128 let _ = self as u128; } } #[derive(Clone, Copy)] #[repr(usize)] enum E10 { A, B = u32::MAX as usize, } impl E10 { fn test(self) { let _ = self as u16; //~^ ERROR: casting `main::E10` to `u16` may truncate the value // Don't lint. let _ = Self::B as u32; // Don't lint. let _ = self as u64; } } } fn avoid_subtract_overflow(q: u32) { let c = (q >> 16) as u8; //~^ ERROR: casting `u32` to `u8` may truncate the value c as usize; let c = (q / 1000) as u8; //~^ ERROR: casting `u32` to `u8` may truncate the value c as usize; } fn issue11426() { (&42u8 >> 0xa9008fb6c9d81e42_0e25730562a601c8_u128) as usize; } fn issue11642() { fn square(x: i16) -> u32 { let x = x as i32; (x * x) as u32; x.pow(2) as u32; (-2_i32).pow(2) as u32 } let _a = |x: i32| -> u32 { (x * x * x * x) as u32 }; (-2_i32).pow(3) as u32; //~^ ERROR: casting `i32` to `u32` may lose the sign of the value let x: i32 = 10; (x * x) as u32; (x * x * x) as u32; //~^ ERROR: casting `i32` to `u32` may lose the sign of the value let y: i16 = -2; (y * y * y * y * -2) as u16; //~^ ERROR: casting `i16` to `u16` may lose the sign of the value (y * y * y * y * 2) as u16; (y * y * y * 2) as u16; //~^ ERROR: casting `i16` to `u16` may lose the sign of the value (y * y * y * -2) as u16; //~^ ERROR: casting `i16` to `u16` may lose the sign of the value fn foo(a: i32, b: i32, c: i32) -> u32 { (a * a * b * b * c * c) as u32; (a * b * c) as u32; //~^ ERROR: casting `i32` to `u32` may lose the sign of the value (a * -b * c) as u32; //~^ ERROR: casting `i32` to `u32` may lose the sign of the value (a * b * c * c) as u32; (a * -2) as u32; //~^ ERROR: casting `i32` to `u32` may lose the sign of the value (a * b * c * -2) as u32; //~^ ERROR: casting `i32` to `u32` may lose the sign of the value (a / b) as u32; (a / b * c) as u32; //~^ ERROR: casting `i32` to `u32` may lose the sign of the value (a / b + b * c) as u32; //~^ ERROR: casting `i32` to `u32` may lose the sign of the value a.pow(3) as u32; //~^ ERROR: casting `i32` to `u32` may lose the sign of the value (a.abs() * b.pow(2) / c.abs()) as u32 } }