Compiler Explorer

Source code

//! Efficient decimal integer formatting.
//!
//! # Safety
//!
//! This uses `CStr::from_bytes_with_nul_unchecked` and
//! `str::from_utf8_unchecked`on the buffer that it filled itself.
#![allow(unsafe_code)]

use std::os::fd::{AsFd, AsRawFd};
use core::ffi::CStr;
use core::any::TypeId;
use core::mem::{self, MaybeUninit};
use itoa::{Buffer, Integer};
#[cfg(all(feature = "std", unix))]
use std::os::unix::ffi::OsStrExt;
#[cfg(all(feature = "std", target_os = "wasi"))]
use std::os::wasi::ffi::OsStrExt;
#[cfg(feature = "std")]
use {core::fmt, std::ffi::OsStr, std::path::Path};

pub fn foo(i: i32) -> DecInt {
    DecInt::new(i)
}

/// Format an integer into a decimal `Path` component, without constructing a
/// temporary `PathBuf` or `String`.
///
/// This is used for opening paths such as `/proc/self/fd/<fd>` on Linux.
///
/// # Examples
///
/// ```
/// # #[cfg(any(feature = "fs", feature = "net"))]
/// use rustix::path::DecInt;
///
/// # #[cfg(any(feature = "fs", feature = "net"))]
/// assert_eq!(
///     format!("hello {}", DecInt::new(9876).as_ref().display()),
///     "hello 9876"
/// );
/// ```
#[derive(Clone)]
pub struct DecInt {
    buf: [MaybeUninit<u8>; "-9223372036854775808\0".len()],
    len: usize,
}

impl DecInt {
    /// Construct a new path component from an integer.
    #[inline]
    pub fn new<Int: Integer + 'static>(i: Int) -> Self {
        let mut buf = [MaybeUninit::uninit(); 21];

let mut str_buf = Buffer::new();
        let str_buf = str_buf.format(i);
        {
            assert!(str_buf.len() < buf.len(), "{str_buf}{} unsupported.", core::any::type_name::<Int>());

buf[..str_buf.len()].copy_from_slice(unsafe {
                // SAFETY: you can always go from init to uninit
                mem::transmute::<&[u8], &[MaybeUninit<u8>]>(str_buf.as_bytes())
            });
            buf[str_buf.len()] = MaybeUninit::new(0);
        }

Self {
            buf,
            len: str_buf.len(),
        }
    }

/// Construct a new path component from a file descriptor.
    #[inline]
    pub fn from_fd<Fd: AsFd>(fd: Fd) -> Self {
        Self::new(fd.as_fd().as_raw_fd())
    }

/// Return the raw byte buffer as a `&str`.
    #[inline]
    pub fn as_str(&self) -> &str {
        // SAFETY: `DecInt` always holds a formatted decimal number, so it's
        // always valid UTF-8.
        unsafe { core::str::from_utf8_unchecked(self.as_bytes()) }
    }

/// Return the raw byte buffer as a `&CStr`.
    #[inline]
    pub fn as_c_str(&self) -> &CStr {
        let bytes_with_nul = self.as_bytes_with_nul();
        debug_assert!(CStr::from_bytes_with_nul(bytes_with_nul).is_ok());

// SAFETY: `self.buf` holds a single decimal ASCII representation and
        // at least one extra NUL byte.
        unsafe { CStr::from_bytes_with_nul_unchecked(bytes_with_nul) }
    }

/// Return the raw byte buffer including the NUL byte.
    #[inline]
    pub fn as_bytes_with_nul(&self) -> &[u8] {
        let init = &self.buf[..=self.len];
        // SAFETY: we're guaranteed to have initialized len+1 bytes.
        unsafe { mem::transmute::<&[MaybeUninit<u8>], &[u8]>(init) }
    }

/// Return the raw byte buffer.
    #[inline]
    pub fn as_bytes(&self) -> &[u8] {
        let bytes = self.as_bytes_with_nul();
        &bytes[..bytes.len() - 1]
    }
}

mod itoa {
    //! [![github]](https://github.com/dtolnay/itoa)&ensp;[![crates-io]](https://crates.io/crates/itoa)&ensp;[![docs-rs]](https://docs.rs/itoa)
//!
//! [github]: https://img.shields.io/badge/github-8da0cb?style=for-the-badge&labelColor=555555&logo=github
//! [crates-io]: https://img.shields.io/badge/crates.io-fc8d62?style=for-the-badge&labelColor=555555&logo=rust
//! [docs-rs]: https://img.shields.io/badge/docs.rs-66c2a5?style=for-the-badge&labelColor=555555&logo=docs.rs
//!
//! <br>
//!
//! This crate provides a fast conversion of integer primitives to decimal
//! strings. The implementation comes straight from [libcore] but avoids the
//! performance penalty of going through [`core::fmt::Formatter`].
//!
//! See also [`ryu`] for printing floating point primitives.
//!
//! [libcore]: https://github.com/rust-lang/rust/blob/b8214dc6c6fc20d0a660fb5700dca9ebf51ebe89/src/libcore/fmt/num.rs#L201-L254
//! [`core::fmt::Formatter`]: https://doc.rust-lang.org/std/fmt/struct.Formatter.html
//! [`ryu`]: https://github.com/dtolnay/ryu
//!
//! # Example
//!
//! ```
//! fn main() {
//!     let mut buffer = itoa::Buffer::new();
//!     let printed = buffer.format(128u64);
//!     assert_eq!(printed, "128");
//! }
//! ```
//!
//! # Performance (lower is better)
//!
//! ![performance](https://raw.githubusercontent.com/dtolnay/itoa/master/performance.png)

// #![doc(html_root_url = "https://docs.rs/itoa/1.0.11")]
#![no_std]
#![allow(
    clippy::cast_lossless,
    clippy::cast_possible_truncation,
    clippy::cast_possible_wrap,
    clippy::cast_sign_loss,
    clippy::expl_impl_clone_on_copy,
    clippy::must_use_candidate,
    clippy::needless_doctest_main,
    clippy::unreadable_literal
)]

use core::mem::{self, MaybeUninit};
use core::{ptr, slice, str};
#[cfg(feature = "no-panic")]
use no_panic::no_panic;

/// A correctly sized stack allocation for the formatted integer to be written
/// into.
///
/// # Example
///
/// ```
/// let mut buffer = itoa::Buffer::new();
/// let printed = buffer.format(1234);
/// assert_eq!(printed, "1234");
/// ```
pub struct Buffer {
    bytes: [MaybeUninit<u8>; I128_MAX_LEN],
}

impl Default for Buffer {
    #[inline]
    fn default() -> Buffer {
        Buffer::new()
    }
}

impl Copy for Buffer {}

impl Clone for Buffer {
    #[inline]
    #[allow(clippy::non_canonical_clone_impl)] // false positive https://github.com/rust-lang/rust-clippy/issues/11072
    fn clone(&self) -> Self {
        Buffer::new()
    }
}

impl Buffer {
    /// This is a cheap operation; you don't need to worry about reusing buffers
    /// for efficiency.
    #[inline]
    #[cfg_attr(feature = "no-panic", no_panic)]
    pub fn new() -> Buffer {
        let bytes = [MaybeUninit::<u8>::uninit(); I128_MAX_LEN];
        Buffer { bytes }
    }

/// Print an integer into this buffer and return a reference to its string
    /// representation within the buffer.
    #[cfg_attr(feature = "no-panic", no_panic)]
    pub fn format<I: Integer>(&mut self, i: I) -> &str {
        let str = i.write(unsafe {
            &mut *(&mut self.bytes as *mut [MaybeUninit<u8>; I128_MAX_LEN]
                as *mut <I as private::Sealed>::Buffer)
        });
        if str.len() > I::MAX_STR_LEN {
            unsafe { core::hint::unreachable_unchecked() }
        }
        str
    }
}

/// An integer that can be written into an [`itoa::Buffer`][Buffer].
///
/// This trait is sealed and cannot be implemented for types outside of itoa.
pub trait Integer: private::Sealed {
    /// The maximum length of the formated str for this integer.
    const MAX_STR_LEN: usize;
}

// Seal to prevent downstream implementations of the Integer trait.
mod private {
    pub trait Sealed: Copy {
        type Buffer: 'static;
        fn write(self, buf: &mut Self::Buffer) -> &str;
    }
}

const DEC_DIGITS_LUT: &[u8] = b"\
      0001020304050607080910111213141516171819\
      2021222324252627282930313233343536373839\
      4041424344454647484950515253545556575859\
      6061626364656667686970717273747576777879\
      8081828384858687888990919293949596979899";

// Adaptation of the original implementation at
// https://github.com/rust-lang/rust/blob/b8214dc6c6fc20d0a660fb5700dca9ebf51ebe89/src/libcore/fmt/num.rs#L188-L266
macro_rules! impl_Integer {
    ($($max_len:expr => $t:ident),* as $conv_fn:ident) => {$(
        impl Integer for $t {
            const MAX_STR_LEN: usize = $max_len;
        }

impl private::Sealed for $t {
            type Buffer = [MaybeUninit<u8>; $max_len];

#[allow(unused_comparisons)]
            #[inline]
            #[cfg_attr(feature = "no-panic", no_panic)]
            fn write(self, buf: &mut [MaybeUninit<u8>; $max_len]) -> &str {
                let is_nonnegative = self >= 0;
                let mut n = if is_nonnegative {
                    self as $conv_fn
                } else {
                    // Convert negative number to positive by summing 1 to its two's complement.
                    (!(self as $conv_fn)).wrapping_add(1)
                };
                let mut curr = buf.len() as isize;
                let buf_ptr = buf.as_mut_ptr() as *mut u8;
                let lut_ptr = DEC_DIGITS_LUT.as_ptr();

// Need at least 16 bits for the 4-digits-at-a-time to work.
                if mem::size_of::<$t>() >= 2 {
                    // Eagerly decode 4 digits at a time.
                    while n >= 10000 {
                        let rem = (n % 10000) as isize;
                        n /= 10000;

let d1 = (rem / 100) << 1;
                        let d2 = (rem % 100) << 1;
                        curr -= 4;
                        unsafe {
                            ptr::copy_nonoverlapping(lut_ptr.offset(d1), buf_ptr.offset(curr), 2);
                            ptr::copy_nonoverlapping(lut_ptr.offset(d2), buf_ptr.offset(curr + 2), 2);
                        }
                    }
                }

// If we reach here, numbers are <=9999 so at most 4 digits long.
                let mut n = n as isize; // Possibly reduce 64-bit math.

// Decode 2 more digits, if >2 digits.
                if n >= 100 {
                    let d1 = (n % 100) << 1;
                    n /= 100;
                    curr -= 2;
                    unsafe {
                        ptr::copy_nonoverlapping(lut_ptr.offset(d1), buf_ptr.offset(curr), 2);
                    }
                }

// Decode last 1 or 2 digits.
                if n < 10 {
                    curr -= 1;
                    unsafe {
                        *buf_ptr.offset(curr) = (n as u8) + b'0';
                    }
                } else {
                    let d1 = n << 1;
                    curr -= 2;
                    unsafe {
                        ptr::copy_nonoverlapping(lut_ptr.offset(d1), buf_ptr.offset(curr), 2);
                    }
                }

if !is_nonnegative {
                    curr -= 1;
                    unsafe {
                        *buf_ptr.offset(curr) = b'-';
                    }
                }

let len = buf.len() - curr as usize;
                let bytes = unsafe { slice::from_raw_parts(buf_ptr.offset(curr), len) };
                unsafe { str::from_utf8_unchecked(bytes) }
            }
        }
    )*};
}

const I8_MAX_LEN: usize = 4;
const U8_MAX_LEN: usize = 3;
const I16_MAX_LEN: usize = 6;
const U16_MAX_LEN: usize = 5;
const I32_MAX_LEN: usize = 11;
const U32_MAX_LEN: usize = 10;
const I64_MAX_LEN: usize = 20;
const U64_MAX_LEN: usize = 20;

impl_Integer!(
    I8_MAX_LEN => i8,
    U8_MAX_LEN => u8,
    I16_MAX_LEN => i16,
    U16_MAX_LEN => u16,
    I32_MAX_LEN => i32,
    U32_MAX_LEN => u32
    as u32);

impl_Integer!(I64_MAX_LEN => i64, U64_MAX_LEN => u64 as u64);

#[cfg(target_pointer_width = "16")]
impl_Integer!(I16_MAX_LEN => isize, U16_MAX_LEN => usize as u16);

#[cfg(target_pointer_width = "32")]
impl_Integer!(I32_MAX_LEN => isize, U32_MAX_LEN => usize as u32);

#[cfg(target_pointer_width = "64")]
impl_Integer!(I64_MAX_LEN => isize, U64_MAX_LEN => usize as u64);

macro_rules! impl_Integer128 {
    ($($max_len:expr => $t:ident),*) => {$(
        impl Integer for $t {
            const MAX_STR_LEN: usize = $max_len;
        }

impl private::Sealed for $t {
            type Buffer = [MaybeUninit<u8>; $max_len];

#[allow(unused_comparisons)]
            #[inline]
            #[cfg_attr(feature = "no-panic", no_panic)]
            fn write(self, buf: &mut [MaybeUninit<u8>; $max_len]) -> &str {
                let is_nonnegative = self >= 0;
                let n = if is_nonnegative {
                    self as u128
                } else {
                    // Convert negative number to positive by summing 1 to its two's complement.
                    (!(self as u128)).wrapping_add(1)
                };
                let mut curr = buf.len() as isize;
                let buf_ptr = buf.as_mut_ptr() as *mut u8;

// Divide by 10^19 which is the highest power less than 2^64.
                let (n, rem) = udiv128::udivmod_1e19(n);
                let buf1 = unsafe { buf_ptr.offset(curr - U64_MAX_LEN as isize) as *mut [MaybeUninit<u8>; U64_MAX_LEN] };
                curr -= rem.write(unsafe { &mut *buf1 }).len() as isize;

if n != 0 {
                    // Memset the base10 leading zeros of rem.
                    let target = buf.len() as isize - 19;
                    unsafe {
                        ptr::write_bytes(buf_ptr.offset(target), b'0', (curr - target) as usize);
                    }
                    curr = target;

// Divide by 10^19 again.
                    let (n, rem) = udiv128::udivmod_1e19(n);
                    let buf2 = unsafe { buf_ptr.offset(curr - U64_MAX_LEN as isize) as *mut [MaybeUninit<u8>; U64_MAX_LEN] };
                    curr -= rem.write(unsafe { &mut *buf2 }).len() as isize;

if n != 0 {
                        // Memset the leading zeros.
                        let target = buf.len() as isize - 38;
                        unsafe {
                            ptr::write_bytes(buf_ptr.offset(target), b'0', (curr - target) as usize);
                        }
                        curr = target;

// There is at most one digit left
                        // because u128::MAX / 10^19 / 10^19 is 3.
                        curr -= 1;
                        unsafe {
                            *buf_ptr.offset(curr) = (n as u8) + b'0';
                        }
                    }
                }

if !is_nonnegative {
                    curr -= 1;
                    unsafe {
                        *buf_ptr.offset(curr) = b'-';
                    }
                }

const U128_MAX_LEN: usize = 39;
const I128_MAX_LEN: usize = 40;

impl_Integer128!(I128_MAX_LEN => i128, U128_MAX_LEN => u128);

mod udiv128 {
    #[cfg(feature = "no-panic")]
use no_panic::no_panic;

/// Multiply unsigned 128 bit integers, return upper 128 bits of the result
#[inline]
#[cfg_attr(feature = "no-panic", no_panic)]
fn u128_mulhi(x: u128, y: u128) -> u128 {
    let x_lo = x as u64;
    let x_hi = (x >> 64) as u64;
    let y_lo = y as u64;
    let y_hi = (y >> 64) as u64;

// handle possibility of overflow
    let carry = (x_lo as u128 * y_lo as u128) >> 64;
    let m = x_lo as u128 * y_hi as u128 + carry;
    let high1 = m >> 64;

let m_lo = m as u64;
    let high2 = (x_hi as u128 * y_lo as u128 + m_lo as u128) >> 64;

x_hi as u128 * y_hi as u128 + high1 + high2
}

/// Divide `n` by 1e19 and return quotient and remainder
///
/// Integer division algorithm is based on the following paper:
///
///   T. Granlund and P. Montgomery, “Division by Invariant Integers Using Multiplication”
///   in Proc. of the SIGPLAN94 Conference on Programming Language Design and
///   Implementation, 1994, pp. 61–72
///
#[inline]
#[cfg_attr(feature = "no-panic", no_panic)]
pub fn udivmod_1e19(n: u128) -> (u128, u64) {
    let d = 10_000_000_000_000_000_000_u64; // 10^19

let quot = if n < 1 << 83 {
        ((n >> 19) as u64 / (d >> 19)) as u128
    } else {
        u128_mulhi(n, 156927543384667019095894735580191660403) >> 62
    };

let rem = (n - quot * d as u128) as u64;
    debug_assert_eq!(quot, n / d as u128);
    debug_assert_eq!(rem as u128, n % d as u128);

(quot, rem)
}

}
}