src/libcore/fmt/num.rs

   1 //! Integer and floating-point number formatting
   2
   3 use crate::fmt;
   4 use crate::mem::MaybeUninit;
   5 use crate::num::flt2dec;
   6 use crate::ops::{Div, Rem, Sub};
   7 use crate::ptr;
   8 use crate::slice;
   9 use crate::str;
  10
  11 #[doc(hidden)]
  12 trait Int:
  13     PartialEq + PartialOrd + Div<Output = Self> + Rem<Output = Self> + Sub<Output = Self> + Copy
  14 {
  15     fn zero() -> Self;
  16     fn from_u8(u: u8) -> Self;
  17     fn to_u8(&self) -> u8;
  18     fn to_u16(&self) -> u16;
  19     fn to_u32(&self) -> u32;
  20     fn to_u64(&self) -> u64;
  21     fn to_u128(&self) -> u128;
  22 }
  23
  24 macro_rules! doit {
  25     ($($t:ident)*) => ($(impl Int for $t {
  26         fn zero() -> Self { 0 }
  27         fn from_u8(u: u8) -> Self { u as Self }
  28         fn to_u8(&self) -> u8 { *self as u8 }
  29         fn to_u16(&self) -> u16 { *self as u16 }
  30         fn to_u32(&self) -> u32 { *self as u32 }
  31         fn to_u64(&self) -> u64 { *self as u64 }
  32         fn to_u128(&self) -> u128 { *self as u128 }
  33     })*)
  34 }
  35 doit! { i8 i16 i32 i64 i128 isize u8 u16 u32 u64 u128 usize }
  36
  37 /// A type that represents a specific radix
  38 #[doc(hidden)]
  39 trait GenericRadix {
  40     /// The number of digits.
  41     const BASE: u8;
  42
  43     /// A radix-specific prefix string.
  44     const PREFIX: &'static str;
  45
  46     /// Converts an integer to corresponding radix digit.
  47     fn digit(x: u8) -> u8;
  48
  49     /// Format an integer using the radix using a formatter.
  50     fn fmt_int<T: Int>(&self, mut x: T, f: &mut fmt::Formatter<'_>) -> fmt::Result {
  51         // The radix can be as low as 2, so we need a buffer of at least 128
  52         // characters for a base 2 number.
  53         let zero = T::zero();
  54         let is_nonnegative = x >= zero;
  55         let mut buf = [MaybeUninit::<u8>::uninit(); 128];
  56         let mut curr = buf.len();
  57         let base = T::from_u8(Self::BASE);
  58         if is_nonnegative {
  59             // Accumulate each digit of the number from the least significant
  60             // to the most significant figure.
  61             for byte in buf.iter_mut().rev() {
  62                 let n = x % base; // Get the current place value.
  63                 x = x / base; // Deaccumulate the number.
  64                 byte.write(Self::digit(n.to_u8())); // Store the digit in the buffer.
  65                 curr -= 1;
  66                 if x == zero {
  67                     // No more digits left to accumulate.
  68                     break;
  69                 };
  70             }
  71         } else {
  72             // Do the same as above, but accounting for two's complement.
  73             for byte in buf.iter_mut().rev() {
  74                 let n = zero - (x % base); // Get the current place value.
  75                 x = x / base; // Deaccumulate the number.
  76                 byte.write(Self::digit(n.to_u8())); // Store the digit in the buffer.
  77                 curr -= 1;
  78                 if x == zero {
  79                     // No more digits left to accumulate.
  80                     break;
  81                 };
  82             }
  83         }
  84         let buf = &buf[curr..];
  85         // SAFETY: The only chars in `buf` are created by `Self::digit` which are assumed to be
  86         // valid UTF-8
  87         let buf = unsafe {
  88             str::from_utf8_unchecked(slice::from_raw_parts(MaybeUninit::first_ptr(buf), buf.len()))
  89         };
  90         f.pad_integral(is_nonnegative, Self::PREFIX, buf)
  91     }
  92 }
  93
  94 /// A binary (base 2) radix
  95 #[derive(Clone, PartialEq)]
  96 struct Binary;
  97
  98 /// An octal (base 8) radix
  99 #[derive(Clone, PartialEq)]
 100 struct Octal;
 101
 102 /// A hexadecimal (base 16) radix, formatted with lower-case characters
 103 #[derive(Clone, PartialEq)]
 104 struct LowerHex;
 105
 106 /// A hexadecimal (base 16) radix, formatted with upper-case characters
 107 #[derive(Clone, PartialEq)]
 108 struct UpperHex;
 109
 110 macro_rules! radix {
 111     ($T:ident, $base:expr, $prefix:expr, $($x:pat => $conv:expr),+) => {
 112         impl GenericRadix for $T {
 113             const BASE: u8 = $base;
 114             const PREFIX: &'static str = $prefix;
 115             fn digit(x: u8) -> u8 {
 116                 match x {
 117                     $($x => $conv,)+
 118                     x => panic!("number not in the range 0..={}: {}", Self::BASE - 1, x),
 119                 }
 120             }
 121         }
 122     }
 123 }
 124
 125 radix! { Binary,    2, "0b", x @  0 ..=  1 => b'0' + x }
 126 radix! { Octal,     8, "0o", x @  0 ..=  7 => b'0' + x }
 127 radix! { LowerHex, 16, "0x", x @  0 ..=  9 => b'0' + x,
 128 x @ 10 ..= 15 => b'a' + (x - 10) }
 129 radix! { UpperHex, 16, "0x", x @  0 ..=  9 => b'0' + x,
 130 x @ 10 ..= 15 => b'A' + (x - 10) }
 131
 132 macro_rules! int_base {
 133     ($Trait:ident for $T:ident as $U:ident -> $Radix:ident) => {
 134         #[stable(feature = "rust1", since = "1.0.0")]
 135         impl fmt::$Trait for $T {
 136             fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
 137                 $Radix.fmt_int(*self as $U, f)
 138             }
 139         }
 140     };
 141 }
 142
 143 macro_rules! debug {
 144     ($T:ident) => {
 145         #[stable(feature = "rust1", since = "1.0.0")]
 146         impl fmt::Debug for $T {
 147             #[inline]
 148             fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
 149                 if f.debug_lower_hex() {
 150                     fmt::LowerHex::fmt(self, f)
 151                 } else if f.debug_upper_hex() {
 152                     fmt::UpperHex::fmt(self, f)
 153                 } else {
 154                     fmt::Display::fmt(self, f)
 155                 }
 156             }
 157         }
 158     };
 159 }
 160
 161 macro_rules! integer {
 162     ($Int:ident, $Uint:ident) => {
 163         int_base! { Binary   for $Int as $Uint  -> Binary }
 164         int_base! { Octal    for $Int as $Uint  -> Octal }
 165         int_base! { LowerHex for $Int as $Uint  -> LowerHex }
 166         int_base! { UpperHex for $Int as $Uint  -> UpperHex }
 167         debug! { $Int }
 168
 169         int_base! { Binary   for $Uint as $Uint -> Binary }
 170         int_base! { Octal    for $Uint as $Uint -> Octal }
 171         int_base! { LowerHex for $Uint as $Uint -> LowerHex }
 172         int_base! { UpperHex for $Uint as $Uint -> UpperHex }
 173         debug! { $Uint }
 174     };
 175 }
 176 integer! { isize, usize }
 177 integer! { i8, u8 }
 178 integer! { i16, u16 }
 179 integer! { i32, u32 }
 180 integer! { i64, u64 }
 181 integer! { i128, u128 }
 182
 183 static DEC_DIGITS_LUT: &[u8; 200] = b"0001020304050607080910111213141516171819\
 184       2021222324252627282930313233343536373839\
 185       4041424344454647484950515253545556575859\
 186       6061626364656667686970717273747576777879\
 187       8081828384858687888990919293949596979899";
 188
 189 macro_rules! impl_Display {
 190     ($($t:ident),* as $u:ident via $conv_fn:ident named $name:ident) => {
 191         fn $name(mut n: $u, is_nonnegative: bool, f: &mut fmt::Formatter<'_>) -> fmt::Result {
 192             // 2^128 is about 3*10^38, so 39 gives an extra byte of space
 193             let mut buf = [MaybeUninit::<u8>::uninit(); 39];
 194             let mut curr = buf.len() as isize;
 195             let buf_ptr = MaybeUninit::first_ptr_mut(&mut buf);
 196             let lut_ptr = DEC_DIGITS_LUT.as_ptr();
 197
 198             // SAFETY: Since `d1` and `d2` are always less than or equal to `198`, we
 199             // can copy from `lut_ptr[d1..d1 + 1]` and `lut_ptr[d2..d2 + 1]`. To show
 200             // that it's OK to copy into `buf_ptr`, notice that at the beginning
 201             // `curr == buf.len() == 39 > log(n)` since `n < 2^128 < 10^39`, and at
 202             // each step this is kept the same as `n` is divided. Since `n` is always
 203             // non-negative, this means that `curr > 0` so `buf_ptr[curr..curr + 1]`
 204             // is safe to access.
 205             unsafe {
 206                 // need at least 16 bits for the 4-characters-at-a-time to work.
 207                 assert!(crate::mem::size_of::<$u>() >= 2);
 208
 209                 // eagerly decode 4 characters at a time
 210                 while n >= 10000 {
 211                     let rem = (n % 10000) as isize;
 212                     n /= 10000;
 213
 214                     let d1 = (rem / 100) << 1;
 215                     let d2 = (rem % 100) << 1;
 216                     curr -= 4;
 217
 218                     // We are allowed to copy to `buf_ptr[curr..curr + 3]` here since
 219                     // otherwise `curr < 0`. But then `n` was originally at least `10000^10`
 220                     // which is `10^40 > 2^128 > n`.
 221                     ptr::copy_nonoverlapping(lut_ptr.offset(d1), buf_ptr.offset(curr), 2);
 222                     ptr::copy_nonoverlapping(lut_ptr.offset(d2), buf_ptr.offset(curr + 2), 2);
 223                 }
 224
 225                 // if we reach here numbers are <= 9999, so at most 4 chars long
 226                 let mut n = n as isize; // possibly reduce 64bit math
 227
 228                 // decode 2 more chars, if > 2 chars
 229                 if n >= 100 {
 230                     let d1 = (n % 100) << 1;
 231                     n /= 100;
 232                     curr -= 2;
 233                     ptr::copy_nonoverlapping(lut_ptr.offset(d1), buf_ptr.offset(curr), 2);
 234                 }
 235
 236                 // decode last 1 or 2 chars
 237                 if n < 10 {
 238                     curr -= 1;
 239                     *buf_ptr.offset(curr) = (n as u8) + b'0';
 240                 } else {
 241                     let d1 = n << 1;
 242                     curr -= 2;
 243                     ptr::copy_nonoverlapping(lut_ptr.offset(d1), buf_ptr.offset(curr), 2);
 244                 }
 245             }
 246
 247             // SAFETY: `curr` > 0 (since we made `buf` large enough), and all the chars are valid
 248             // UTF-8 since `DEC_DIGITS_LUT` is
 249             let buf_slice = unsafe {
 250                 str::from_utf8_unchecked(
 251                     slice::from_raw_parts(buf_ptr.offset(curr), buf.len() - curr as usize))
 252             };
 253             f.pad_integral(is_nonnegative, "", buf_slice)
 254         }
 255
 256         $(
 257             #[stable(feature = "rust1", since = "1.0.0")]
 258             impl fmt::Display for $t {
 259                 #[allow(unused_comparisons)]
 260                 fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
 261                     let is_nonnegative = *self >= 0;
 262                     let n = if is_nonnegative {
 263                         self.$conv_fn()
 264                     } else {
 265                         // convert the negative num to positive by summing 1 to it's 2 complement
 266                         (!self.$conv_fn()).wrapping_add(1)
 267                     };
 268                     $name(n, is_nonnegative, f)
 269                 }
 270             })*
 271     };
 272 }
 273
 274 macro_rules! impl_Exp {
 275     ($($t:ident),* as $u:ident via $conv_fn:ident named $name:ident) => {
 276         fn $name(
 277             mut n: $u,
 278             is_nonnegative: bool,
 279             upper: bool,
 280             f: &mut fmt::Formatter<'_>
 281         ) -> fmt::Result {
 282             let (mut n, mut exponent, trailing_zeros, added_precision) = {
 283                 let mut exponent = 0;
 284                 // count and remove trailing decimal zeroes
 285                 while n % 10 == 0 && n >= 10 {
 286                     n /= 10;
 287                     exponent += 1;
 288                 }
 289                 let trailing_zeros = exponent;
 290
 291                 let (added_precision, subtracted_precision) = match f.precision() {
 292                     Some(fmt_prec) => {
 293                         // number of decimal digits minus 1
 294                         let mut tmp = n;
 295                         let mut prec = 0;
 296                         while tmp >= 10 {
 297                             tmp /= 10;
 298                             prec += 1;
 299                         }
 300                         (fmt_prec.saturating_sub(prec), prec.saturating_sub(fmt_prec))
 301                     }
 302                     None => (0,0)
 303                 };
 304                 for _ in 1..subtracted_precision {
 305                     n/=10;
 306                     exponent += 1;
 307                 }
 308                 if subtracted_precision != 0 {
 309                     let rem = n % 10;
 310                     n /= 10;
 311                     exponent += 1;
 312                     // round up last digit
 313                     if rem >= 5 {
 314                         n += 1;
 315                     }
 316                 }
 317                 (n, exponent, trailing_zeros, added_precision)
 318             };
 319
 320             // 39 digits (worst case u128) + . = 40
 321             // Since `curr` always decreases by the number of digits copied, this means
 322             // that `curr >= 0`.
 323             let mut buf = [MaybeUninit::<u8>::uninit(); 40];
 324             let mut curr = buf.len() as isize; //index for buf
 325             let buf_ptr = MaybeUninit::first_ptr_mut(&mut buf);
 326             let lut_ptr = DEC_DIGITS_LUT.as_ptr();
 327
 328             // decode 2 chars at a time
 329             while n >= 100 {
 330                 let d1 = ((n % 100) as isize) << 1;
 331                 curr -= 2;
 332                 // SAFETY: `d1 <= 198`, so we can copy from `lut_ptr[d1..d1 + 2]` since
 333                 // `DEC_DIGITS_LUT` has a length of 200.
 334                 unsafe {
 335                     ptr::copy_nonoverlapping(lut_ptr.offset(d1), buf_ptr.offset(curr), 2);
 336                 }
 337                 n /= 100;
 338                 exponent += 2;
 339             }
 340             // n is <= 99, so at most 2 chars long
 341             let mut n = n as isize; // possibly reduce 64bit math
 342             // decode second-to-last character
 343             if n >= 10 {
 344                 curr -= 1;
 345                 // SAFETY: Safe since `40 > curr >= 0` (see comment)
 346                 unsafe {
 347                     *buf_ptr.offset(curr) = (n as u8 % 10_u8) + b'0';
 348                 }
 349                 n /= 10;
 350                 exponent += 1;
 351             }
 352             // add decimal point iff >1 mantissa digit will be printed
 353             if exponent != trailing_zeros || added_precision != 0 {
 354                 curr -= 1;
 355                 // SAFETY: Safe since `40 > curr >= 0`
 356                 unsafe {
 357                     *buf_ptr.offset(curr) = b'.';
 358                 }
 359             }
 360
 361             // SAFETY: Safe since `40 > curr >= 0`
 362             let buf_slice = unsafe {
 363                 // decode last character
 364                 curr -= 1;
 365                 *buf_ptr.offset(curr) = (n as u8) + b'0';
 366
 367                 let len = buf.len() - curr as usize;
 368                 slice::from_raw_parts(buf_ptr.offset(curr), len)
 369             };
 370
 371             // stores 'e' (or 'E') and the up to 2-digit exponent
 372             let mut exp_buf = [MaybeUninit::<u8>::uninit(); 3];
 373             let exp_ptr = MaybeUninit::first_ptr_mut(&mut exp_buf);
 374             // SAFETY: In either case, `exp_buf` is written within bounds and `exp_ptr[..len]`
 375             // is contained within `exp_buf` since `len <= 3`.
 376             let exp_slice = unsafe {
 377                 *exp_ptr.offset(0) = if upper {b'E'} else {b'e'};
 378                 let len = if exponent < 10 {
 379                     *exp_ptr.offset(1) = (exponent as u8) + b'0';
 380                     2
 381                 } else {
 382                     let off = exponent << 1;
 383                     ptr::copy_nonoverlapping(lut_ptr.offset(off), exp_ptr.offset(1), 2);
 384                     3
 385                 };
 386                 slice::from_raw_parts(exp_ptr, len)
 387             };
 388
 389             let parts = &[
 390                 flt2dec::Part::Copy(buf_slice),
 391                 flt2dec::Part::Zero(added_precision),
 392                 flt2dec::Part::Copy(exp_slice)
 393             ];
 394             let sign = if !is_nonnegative {
 395                 "-"
 396             } else if f.sign_plus() {
 397                 "+"
 398             } else {
 399                 ""
 400             };
 401             let formatted = flt2dec::Formatted{sign, parts};
 402             f.pad_formatted_parts(&formatted)
 403         }
 404
 405         $(
 406             #[stable(feature = "integer_exp_format", since = "1.42.0")]
 407             impl fmt::LowerExp for $t {
 408                 #[allow(unused_comparisons)]
 409                 fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
 410                     let is_nonnegative = *self >= 0;
 411                     let n = if is_nonnegative {
 412                         self.$conv_fn()
 413                     } else {
 414                         // convert the negative num to positive by summing 1 to it's 2 complement
 415                         (!self.$conv_fn()).wrapping_add(1)
 416                     };
 417                     $name(n, is_nonnegative, false, f)
 418                 }
 419             })*
 420         $(
 421             #[stable(feature = "integer_exp_format", since = "1.42.0")]
 422             impl fmt::UpperExp for $t {
 423                 #[allow(unused_comparisons)]
 424                 fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
 425                     let is_nonnegative = *self >= 0;
 426                     let n = if is_nonnegative {
 427                         self.$conv_fn()
 428                     } else {
 429                         // convert the negative num to positive by summing 1 to it's 2 complement
 430                         (!self.$conv_fn()).wrapping_add(1)
 431                     };
 432                     $name(n, is_nonnegative, true, f)
 433                 }
 434             })*
 435     };
 436 }
 437
 438 // Include wasm32 in here since it doesn't reflect the native pointer size, and
 439 // often cares strongly about getting a smaller code size.
 440 #[cfg(any(target_pointer_width = "64", target_arch = "wasm32"))]
 441 mod imp {
 442     use super::*;
 443     impl_Display!(
 444         i8, u8, i16, u16, i32, u32, i64, u64, usize, isize
 445             as u64 via to_u64 named fmt_u64
 446     );
 447     impl_Exp!(
 448         i8, u8, i16, u16, i32, u32, i64, u64, usize, isize
 449             as u64 via to_u64 named exp_u64
 450     );
 451 }
 452
 453 #[cfg(not(any(target_pointer_width = "64", target_arch = "wasm32")))]
 454 mod imp {
 455     use super::*;
 456     impl_Display!(i8, u8, i16, u16, i32, u32, isize, usize as u32 via to_u32 named fmt_u32);
 457     impl_Display!(i64, u64 as u64 via to_u64 named fmt_u64);
 458     impl_Exp!(i8, u8, i16, u16, i32, u32, isize, usize as u32 via to_u32 named exp_u32);
 459     impl_Exp!(i64, u64 as u64 via to_u64 named exp_u64);
 460 }
 461
 462 impl_Display!(i128, u128 as u128 via to_u128 named fmt_u128);
 463 impl_Exp!(i128, u128 as u128 via to_u128 named exp_u128);