| // Copyright 2015 The Rust Project Developers. See the COPYRIGHT |
| // file at the top-level directory of this distribution and at |
| // http://rust-lang.org/COPYRIGHT. |
| // |
| // Licensed under the Apache License, Version 2.0 <LICENSE-APACHE or |
| // http://www.apache.org/licenses/LICENSE-2.0> or the MIT license |
| // <LICENSE-MIT or http://opensource.org/licenses/MIT>, at your |
| // option. This file may not be copied, modified, or distributed |
| // except according to those terms. |
| |
| #![allow(unused_imports)] |
| |
| // Note: the relevant lint pass here runs before some of the constant |
| // evaluation below (e.g. that performed by trans and llvm), so if you |
| // change this warn to a deny, then the compiler will exit before |
| // those errors are detected. |
| |
| use std::fmt; |
| use std::{i8, i16, i32, i64, isize}; |
| use std::{u8, u16, u32, u64, usize}; |
| |
| const VALS_I8: (i8, i8, i8, i8) = |
| (-i8::MIN, |
| //~^ ERROR attempted to negate with overflow |
| i8::MIN - 1, |
| //~^ ERROR attempted to subtract with overflow |
| i8::MAX + 1, |
| //~^ ERROR attempted to add with overflow |
| i8::MIN * 2, |
| //~^ ERROR attempted to multiply with overflow |
| ); |
| |
| const VALS_I16: (i16, i16, i16, i16) = |
| (-i16::MIN, |
| //~^ ERROR attempted to negate with overflow |
| i16::MIN - 1, |
| //~^ ERROR attempted to subtract with overflow |
| i16::MAX + 1, |
| //~^ ERROR attempted to add with overflow |
| i16::MIN * 2, |
| //~^ ERROR attempted to multiply with overflow |
| ); |
| |
| const VALS_I32: (i32, i32, i32, i32) = |
| (-i32::MIN, |
| //~^ ERROR attempted to negate with overflow |
| i32::MIN - 1, |
| //~^ ERROR attempted to subtract with overflow |
| i32::MAX + 1, |
| //~^ ERROR attempted to add with overflow |
| i32::MIN * 2, |
| //~^ ERROR attempted to multiply with overflow |
| ); |
| |
| const VALS_I64: (i64, i64, i64, i64) = |
| (-i64::MIN, |
| //~^ ERROR attempted to negate with overflow |
| i64::MIN - 1, |
| //~^ ERROR attempted to subtract with overflow |
| i64::MAX + 1, |
| //~^ ERROR attempted to add with overflow |
| i64::MAX * 2, |
| //~^ ERROR attempted to multiply with overflow |
| ); |
| |
| const VALS_U8: (u8, u8, u8, u8) = |
| (-(u8::MIN as i8) as u8, |
| u8::MIN - 1, |
| //~^ ERROR attempted to subtract with overflow |
| u8::MAX + 1, |
| //~^ ERROR attempted to add with overflow |
| u8::MAX * 2, |
| //~^ ERROR attempted to multiply with overflow |
| ); |
| |
| const VALS_U16: (u16, u16, u16, u16) = |
| (-(u16::MIN as i16) as u16, |
| u16::MIN - 1, |
| //~^ ERROR attempted to subtract with overflow |
| u16::MAX + 1, |
| //~^ ERROR attempted to add with overflow |
| u16::MAX * 2, |
| //~^ ERROR attempted to multiply with overflow |
| ); |
| |
| const VALS_U32: (u32, u32, u32, u32) = |
| (-(u32::MIN as i32) as u32, |
| u32::MIN - 1, |
| //~^ ERROR attempted to subtract with overflow |
| u32::MAX + 1, |
| //~^ ERROR attempted to add with overflow |
| u32::MAX * 2, |
| //~^ ERROR attempted to multiply with overflow |
| ); |
| |
| const VALS_U64: (u64, u64, u64, u64) = |
| (-(u64::MIN as i64) as u64, |
| u64::MIN - 1, |
| //~^ ERROR attempted to subtract with overflow |
| u64::MAX + 1, |
| //~^ ERROR attempted to add with overflow |
| u64::MAX * 2, |
| //~^ ERROR attempted to multiply with overflow |
| ); |
| |
| fn main() { |
| foo(VALS_I8); |
| foo(VALS_I16); |
| foo(VALS_I32); |
| foo(VALS_I64); |
| |
| foo(VALS_U8); |
| foo(VALS_U16); |
| foo(VALS_U32); |
| foo(VALS_U64); |
| } |
| |
| fn foo<T:fmt::Debug>(x: T) { |
| println!("{:?}", x); |
| } |