| 1 | /*! |
| 2 | |
| 3 | Floating-point number to decimal conversion routines. |
| 4 | |
| 5 | # Problem statement |
| 6 | |
| 7 | We are given the floating-point number `v = f * 2^e` with an integer `f`, |
| 8 | and its bounds `minus` and `plus` such that any number between `v - minus` and |
| 9 | `v + plus` will be rounded to `v`. For the simplicity we assume that |
| 10 | this range is exclusive. Then we would like to get the unique decimal |
| 11 | representation `V = 0.d[0..n-1] * 10^k` such that: |
| 12 | |
| 13 | - `d[0]` is non-zero. |
| 14 | |
| 15 | - It's correctly rounded when parsed back: `v - minus < V < v + plus`. |
| 16 | Furthermore it is shortest such one, i.e., there is no representation |
| 17 | with less than `n` digits that is correctly rounded. |
| 18 | |
| 19 | - It's closest to the original value: `abs(V - v) <= 10^(k-n) / 2`. Note that |
| 20 | there might be two representations satisfying this uniqueness requirement, |
| 21 | in which case some tie-breaking mechanism is used. |
| 22 | |
| 23 | We will call this mode of operation as to the *shortest* mode. This mode is used |
| 24 | when there is no additional constraint, and can be thought as a "natural" mode |
| 25 | as it matches the ordinary intuition (it at least prints `0.1f32` as "0.1"). |
| 26 | |
| 27 | We have two more modes of operation closely related to each other. In these modes |
| 28 | we are given either the number of significant digits `n` or the last-digit |
| 29 | limitation `limit` (which determines the actual `n`), and we would like to get |
| 30 | the representation `V = 0.d[0..n-1] * 10^k` such that: |
| 31 | |
| 32 | - `d[0]` is non-zero, unless `n` was zero in which case only `k` is returned. |
| 33 | |
| 34 | - It's closest to the original value: `abs(V - v) <= 10^(k-n) / 2`. Again, |
| 35 | there might be some tie-breaking mechanism. |
| 36 | |
| 37 | When `limit` is given but not `n`, we set `n` such that `k - n = limit` |
| 38 | so that the last digit `d[n-1]` is scaled by `10^(k-n) = 10^limit`. |
| 39 | If such `n` is negative, we clip it to zero so that we will only get `k`. |
| 40 | We are also limited by the supplied buffer. This limitation is used to print |
| 41 | the number up to given number of fractional digits without knowing |
| 42 | the correct `k` beforehand. |
| 43 | |
| 44 | We will call the mode of operation requiring `n` as to the *exact* mode, |
| 45 | and one requiring `limit` as to the *fixed* mode. The exact mode is a subset of |
| 46 | the fixed mode: the sufficiently large last-digit limitation will eventually fill |
| 47 | the supplied buffer and let the algorithm to return. |
| 48 | |
| 49 | # Implementation overview |
| 50 | |
| 51 | It is easy to get the floating point printing correct but slow (Russ Cox has |
| 52 | [demonstrated](https://research.swtch.com/ftoa) how it's easy), or incorrect but |
| 53 | fast (naïve division and modulo). But it is surprisingly hard to print |
| 54 | floating point numbers correctly *and* efficiently. |
| 55 | |
| 56 | There are two classes of algorithms widely known to be correct. |
| 57 | |
| 58 | - The "Dragon" family of algorithm is first described by Guy L. Steele Jr. and |
| 59 | Jon L. White. They rely on the fixed-size big integer for their correctness. |
| 60 | A slight improvement was found later, which is posthumously described by |
| 61 | Robert G. Burger and R. Kent Dybvig. David Gay's `dtoa.c` routine is |
| 62 | a popular implementation of this strategy. |
| 63 | |
| 64 | - The "Grisu" family of algorithm is first described by Florian Loitsch. |
| 65 | They use very cheap integer-only procedure to determine the close-to-correct |
| 66 | representation which is at least guaranteed to be shortest. The variant, |
| 67 | Grisu3, actively detects if the resulting representation is incorrect. |
| 68 | |
| 69 | We implement both algorithms with necessary tweaks to suit our requirements. |
| 70 | In particular, published literatures are short of the actual implementation |
| 71 | difficulties like how to avoid arithmetic overflows. Each implementation, |
| 72 | available in `strategy::dragon` and `strategy::grisu` respectively, |
| 73 | extensively describes all necessary justifications and many proofs for them. |
| 74 | (It is still difficult to follow though. You have been warned.) |
| 75 | |
| 76 | Both implementations expose two public functions: |
| 77 | |
| 78 | - `format_shortest(decoded, buf)`, which always needs at least |
| 79 | `MAX_SIG_DIGITS` digits of buffer. Implements the shortest mode. |
| 80 | |
| 81 | - `format_exact(decoded, buf, limit)`, which accepts as small as |
| 82 | one digit of buffer. Implements exact and fixed modes. |
| 83 | |
| 84 | They try to fill the `u8` buffer with digits and returns the number of digits |
| 85 | written and the exponent `k`. They are total for all finite `f32` and `f64` |
| 86 | inputs (Grisu internally falls back to Dragon if necessary). |
| 87 | |
| 88 | The rendered digits are formatted into the actual string form with |
| 89 | four functions: |
| 90 | |
| 91 | - `to_shortest_str` prints the shortest representation, which can be padded by |
| 92 | zeroes to make *at least* given number of fractional digits. |
| 93 | |
| 94 | - `to_shortest_exp_str` prints the shortest representation, which can be |
| 95 | padded by zeroes when its exponent is in the specified ranges, |
| 96 | or can be printed in the exponential form such as `1.23e45`. |
| 97 | |
| 98 | - `to_exact_exp_str` prints the exact representation with given number of |
| 99 | digits in the exponential form. |
| 100 | |
| 101 | - `to_exact_fixed_str` prints the fixed representation with *exactly* |
| 102 | given number of fractional digits. |
| 103 | |
| 104 | They all return a slice of preallocated `Part` array, which corresponds to |
| 105 | the individual part of strings: a fixed string, a part of rendered digits, |
| 106 | a number of zeroes or a small (`u16`) number. The caller is expected to |
| 107 | provide a large enough buffer and `Part` array, and to assemble the final |
| 108 | string from resulting `Part`s itself. |
| 109 | |
| 110 | All algorithms and formatting functions are accompanied by extensive tests |
| 111 | in `coretests::num::flt2dec` module. It also shows how to use individual |
| 112 | functions. |
| 113 | |
| 114 | */ |
| 115 | |
| 116 | // while this is extensively documented, this is in principle private which is |
| 117 | // only made public for testing. do not expose us. |
| 118 | #![doc (hidden)] |
| 119 | #![unstable ( |
| 120 | feature = "flt2dec" , |
| 121 | reason = "internal routines only exposed for testing" , |
| 122 | issue = "none" |
| 123 | )] |
| 124 | |
| 125 | pub use self::decoder::{DecodableFloat, Decoded, FullDecoded, decode}; |
| 126 | use super::fmt::{Formatted, Part}; |
| 127 | use crate::mem::MaybeUninit; |
| 128 | |
| 129 | pub mod decoder; |
| 130 | pub mod estimator; |
| 131 | |
| 132 | /// Digit-generation algorithms. |
| 133 | pub mod strategy { |
| 134 | pub mod dragon; |
| 135 | pub mod grisu; |
| 136 | } |
| 137 | |
| 138 | /// The minimum size of buffer necessary for the shortest mode. |
| 139 | /// |
| 140 | /// It is a bit non-trivial to derive, but this is one plus the maximal number of |
| 141 | /// significant decimal digits from formatting algorithms with the shortest result. |
| 142 | /// The exact formula is `ceil(# bits in mantissa * log_10 2 + 1)`. |
| 143 | pub const MAX_SIG_DIGITS: usize = 17; |
| 144 | |
| 145 | /// When `d` contains decimal digits, increase the last digit and propagate carry. |
| 146 | /// Returns a next digit when it causes the length to change. |
| 147 | #[doc (hidden)] |
| 148 | pub fn round_up(d: &mut [u8]) -> Option<u8> { |
| 149 | match d.iter().rposition(|&c: u8| c != b'9' ) { |
| 150 | Some(i: usize) => { |
| 151 | // d[i+1..n] is all nines |
| 152 | d[i] += 1; |
| 153 | for j: usize in i + 1..d.len() { |
| 154 | d[j] = b'0' ; |
| 155 | } |
| 156 | None |
| 157 | } |
| 158 | None if d.len() > 0 => { |
| 159 | // 999..999 rounds to 1000..000 with an increased exponent |
| 160 | d[0] = b'1' ; |
| 161 | for j: usize in 1..d.len() { |
| 162 | d[j] = b'0' ; |
| 163 | } |
| 164 | Some(b'0' ) |
| 165 | } |
| 166 | None => { |
| 167 | // an empty buffer rounds up (a bit strange but reasonable) |
| 168 | Some(b'1' ) |
| 169 | } |
| 170 | } |
| 171 | } |
| 172 | |
| 173 | /// Formats given decimal digits `0.<...buf...> * 10^exp` into the decimal form |
| 174 | /// with at least given number of fractional digits. The result is stored to |
| 175 | /// the supplied parts array and a slice of written parts is returned. |
| 176 | /// |
| 177 | /// `frac_digits` can be less than the number of actual fractional digits in `buf`; |
| 178 | /// it will be ignored and full digits will be printed. It is only used to print |
| 179 | /// additional zeroes after rendered digits. Thus `frac_digits` of 0 means that |
| 180 | /// it will only print given digits and nothing else. |
| 181 | fn digits_to_dec_str<'a>( |
| 182 | buf: &'a [u8], |
| 183 | exp: i16, |
| 184 | frac_digits: usize, |
| 185 | parts: &'a mut [MaybeUninit<Part<'a>>], |
| 186 | ) -> &'a [Part<'a>] { |
| 187 | assert!(!buf.is_empty()); |
| 188 | assert!(buf[0] > b'0' ); |
| 189 | assert!(parts.len() >= 4); |
| 190 | |
| 191 | // if there is the restriction on the last digit position, `buf` is assumed to be |
| 192 | // left-padded with the virtual zeroes. the number of virtual zeroes, `nzeroes`, |
| 193 | // equals to `max(0, exp + frac_digits - buf.len())`, so that the position of |
| 194 | // the last digit `exp - buf.len() - nzeroes` is no more than `-frac_digits`: |
| 195 | // |
| 196 | // |<-virtual->| |
| 197 | // |<---- buf ---->| zeroes | exp |
| 198 | // 0. 1 2 3 4 5 6 7 8 9 _ _ _ _ _ _ x 10 |
| 199 | // | | | |
| 200 | // 10^exp 10^(exp-buf.len()) 10^(exp-buf.len()-nzeroes) |
| 201 | // |
| 202 | // `nzeroes` is individually calculated for each case in order to avoid overflow. |
| 203 | |
| 204 | if exp <= 0 { |
| 205 | // the decimal point is before rendered digits: [0.][000...000][1234][____] |
| 206 | let minus_exp = -(exp as i32) as usize; |
| 207 | parts[0] = MaybeUninit::new(Part::Copy(b"0." )); |
| 208 | parts[1] = MaybeUninit::new(Part::Zero(minus_exp)); |
| 209 | parts[2] = MaybeUninit::new(Part::Copy(buf)); |
| 210 | if frac_digits > buf.len() && frac_digits - buf.len() > minus_exp { |
| 211 | parts[3] = MaybeUninit::new(Part::Zero((frac_digits - buf.len()) - minus_exp)); |
| 212 | // SAFETY: we just initialized the elements `..4`. |
| 213 | unsafe { parts[..4].assume_init_ref() } |
| 214 | } else { |
| 215 | // SAFETY: we just initialized the elements `..3`. |
| 216 | unsafe { parts[..3].assume_init_ref() } |
| 217 | } |
| 218 | } else { |
| 219 | let exp = exp as usize; |
| 220 | if exp < buf.len() { |
| 221 | // the decimal point is inside rendered digits: [12][.][34][____] |
| 222 | parts[0] = MaybeUninit::new(Part::Copy(&buf[..exp])); |
| 223 | parts[1] = MaybeUninit::new(Part::Copy(b"." )); |
| 224 | parts[2] = MaybeUninit::new(Part::Copy(&buf[exp..])); |
| 225 | if frac_digits > buf.len() - exp { |
| 226 | parts[3] = MaybeUninit::new(Part::Zero(frac_digits - (buf.len() - exp))); |
| 227 | // SAFETY: we just initialized the elements `..4`. |
| 228 | unsafe { parts[..4].assume_init_ref() } |
| 229 | } else { |
| 230 | // SAFETY: we just initialized the elements `..3`. |
| 231 | unsafe { parts[..3].assume_init_ref() } |
| 232 | } |
| 233 | } else { |
| 234 | // the decimal point is after rendered digits: [1234][____0000] or [1234][__][.][__]. |
| 235 | parts[0] = MaybeUninit::new(Part::Copy(buf)); |
| 236 | parts[1] = MaybeUninit::new(Part::Zero(exp - buf.len())); |
| 237 | if frac_digits > 0 { |
| 238 | parts[2] = MaybeUninit::new(Part::Copy(b"." )); |
| 239 | parts[3] = MaybeUninit::new(Part::Zero(frac_digits)); |
| 240 | // SAFETY: we just initialized the elements `..4`. |
| 241 | unsafe { parts[..4].assume_init_ref() } |
| 242 | } else { |
| 243 | // SAFETY: we just initialized the elements `..2`. |
| 244 | unsafe { parts[..2].assume_init_ref() } |
| 245 | } |
| 246 | } |
| 247 | } |
| 248 | } |
| 249 | |
| 250 | /// Formats the given decimal digits `0.<...buf...> * 10^exp` into the exponential |
| 251 | /// form with at least the given number of significant digits. When `upper` is `true`, |
| 252 | /// the exponent will be prefixed by `E`; otherwise that's `e`. The result is |
| 253 | /// stored to the supplied parts array and a slice of written parts is returned. |
| 254 | /// |
| 255 | /// `min_digits` can be less than the number of actual significant digits in `buf`; |
| 256 | /// it will be ignored and full digits will be printed. It is only used to print |
| 257 | /// additional zeroes after rendered digits. Thus, `min_digits == 0` means that |
| 258 | /// it will only print the given digits and nothing else. |
| 259 | fn digits_to_exp_str<'a>( |
| 260 | buf: &'a [u8], |
| 261 | exp: i16, |
| 262 | min_ndigits: usize, |
| 263 | upper: bool, |
| 264 | parts: &'a mut [MaybeUninit<Part<'a>>], |
| 265 | ) -> &'a [Part<'a>] { |
| 266 | assert!(!buf.is_empty()); |
| 267 | assert!(buf[0] > b'0' ); |
| 268 | assert!(parts.len() >= 6); |
| 269 | |
| 270 | let mut n = 0; |
| 271 | |
| 272 | parts[n] = MaybeUninit::new(Part::Copy(&buf[..1])); |
| 273 | n += 1; |
| 274 | |
| 275 | if buf.len() > 1 || min_ndigits > 1 { |
| 276 | parts[n] = MaybeUninit::new(Part::Copy(b"." )); |
| 277 | parts[n + 1] = MaybeUninit::new(Part::Copy(&buf[1..])); |
| 278 | n += 2; |
| 279 | if min_ndigits > buf.len() { |
| 280 | parts[n] = MaybeUninit::new(Part::Zero(min_ndigits - buf.len())); |
| 281 | n += 1; |
| 282 | } |
| 283 | } |
| 284 | |
| 285 | // 0.1234 x 10^exp = 1.234 x 10^(exp-1) |
| 286 | let exp = exp as i32 - 1; // avoid underflow when exp is i16::MIN |
| 287 | if exp < 0 { |
| 288 | parts[n] = MaybeUninit::new(Part::Copy(if upper { b"E-" } else { b"e-" })); |
| 289 | parts[n + 1] = MaybeUninit::new(Part::Num(-exp as u16)); |
| 290 | } else { |
| 291 | parts[n] = MaybeUninit::new(Part::Copy(if upper { b"E" } else { b"e" })); |
| 292 | parts[n + 1] = MaybeUninit::new(Part::Num(exp as u16)); |
| 293 | } |
| 294 | // SAFETY: we just initialized the elements `..n + 2`. |
| 295 | unsafe { parts[..n + 2].assume_init_ref() } |
| 296 | } |
| 297 | |
| 298 | /// Sign formatting options. |
| 299 | #[derive (Copy, Clone, PartialEq, Eq, Debug)] |
| 300 | pub enum Sign { |
| 301 | /// Prints `-` for any negative value. |
| 302 | Minus, // -inf -1 -0 0 1 inf nan |
| 303 | /// Prints `-` for any negative value, or `+` otherwise. |
| 304 | MinusPlus, // -inf -1 -0 +0 +1 +inf nan |
| 305 | } |
| 306 | |
| 307 | /// Returns the static byte string corresponding to the sign to be formatted. |
| 308 | /// It can be either `""`, `"+"` or `"-"`. |
| 309 | fn determine_sign(sign: Sign, decoded: &FullDecoded, negative: bool) -> &'static str { |
| 310 | match (*decoded, sign) { |
| 311 | (FullDecoded::Nan, _) => "" , |
| 312 | (_, Sign::Minus) => { |
| 313 | if negative { |
| 314 | "-" |
| 315 | } else { |
| 316 | "" |
| 317 | } |
| 318 | } |
| 319 | (_, Sign::MinusPlus) => { |
| 320 | if negative { |
| 321 | "-" |
| 322 | } else { |
| 323 | "+" |
| 324 | } |
| 325 | } |
| 326 | } |
| 327 | } |
| 328 | |
| 329 | /// Formats the given floating point number into the decimal form with at least |
| 330 | /// given number of fractional digits. The result is stored to the supplied parts |
| 331 | /// array while utilizing given byte buffer as a scratch. `upper` is currently |
| 332 | /// unused but left for the future decision to change the case of non-finite values, |
| 333 | /// i.e., `inf` and `nan`. The first part to be rendered is always a `Part::Sign` |
| 334 | /// (which can be an empty string if no sign is rendered). |
| 335 | /// |
| 336 | /// `format_shortest` should be the underlying digit-generation function. |
| 337 | /// It should return the part of the buffer that it initialized. |
| 338 | /// You probably would want `strategy::grisu::format_shortest` for this. |
| 339 | /// |
| 340 | /// `frac_digits` can be less than the number of actual fractional digits in `v`; |
| 341 | /// it will be ignored and full digits will be printed. It is only used to print |
| 342 | /// additional zeroes after rendered digits. Thus `frac_digits` of 0 means that |
| 343 | /// it will only print given digits and nothing else. |
| 344 | /// |
| 345 | /// The byte buffer should be at least `MAX_SIG_DIGITS` bytes long. |
| 346 | /// There should be at least 4 parts available, due to the worst case like |
| 347 | /// `[+][0.][0000][2][0000]` with `frac_digits = 10`. |
| 348 | pub fn to_shortest_str<'a, T, F>( |
| 349 | mut format_shortest: F, |
| 350 | v: T, |
| 351 | sign: Sign, |
| 352 | frac_digits: usize, |
| 353 | buf: &'a mut [MaybeUninit<u8>], |
| 354 | parts: &'a mut [MaybeUninit<Part<'a>>], |
| 355 | ) -> Formatted<'a> |
| 356 | where |
| 357 | T: DecodableFloat, |
| 358 | F: FnMut(&Decoded, &'a mut [MaybeUninit<u8>]) -> (&'a [u8], i16), |
| 359 | { |
| 360 | assert!(parts.len() >= 4); |
| 361 | assert!(buf.len() >= MAX_SIG_DIGITS); |
| 362 | |
| 363 | let (negative, full_decoded) = decode(v); |
| 364 | let sign = determine_sign(sign, &full_decoded, negative); |
| 365 | match full_decoded { |
| 366 | FullDecoded::Nan => { |
| 367 | parts[0] = MaybeUninit::new(Part::Copy(b"NaN" )); |
| 368 | // SAFETY: we just initialized the elements `..1`. |
| 369 | Formatted { sign, parts: unsafe { parts[..1].assume_init_ref() } } |
| 370 | } |
| 371 | FullDecoded::Infinite => { |
| 372 | parts[0] = MaybeUninit::new(Part::Copy(b"inf" )); |
| 373 | // SAFETY: we just initialized the elements `..1`. |
| 374 | Formatted { sign, parts: unsafe { parts[..1].assume_init_ref() } } |
| 375 | } |
| 376 | FullDecoded::Zero => { |
| 377 | if frac_digits > 0 { |
| 378 | // [0.][0000] |
| 379 | parts[0] = MaybeUninit::new(Part::Copy(b"0." )); |
| 380 | parts[1] = MaybeUninit::new(Part::Zero(frac_digits)); |
| 381 | Formatted { |
| 382 | sign, |
| 383 | // SAFETY: we just initialized the elements `..2`. |
| 384 | parts: unsafe { parts[..2].assume_init_ref() }, |
| 385 | } |
| 386 | } else { |
| 387 | parts[0] = MaybeUninit::new(Part::Copy(b"0" )); |
| 388 | Formatted { |
| 389 | sign, |
| 390 | // SAFETY: we just initialized the elements `..1`. |
| 391 | parts: unsafe { parts[..1].assume_init_ref() }, |
| 392 | } |
| 393 | } |
| 394 | } |
| 395 | FullDecoded::Finite(ref decoded) => { |
| 396 | let (buf, exp) = format_shortest(decoded, buf); |
| 397 | Formatted { sign, parts: digits_to_dec_str(buf, exp, frac_digits, parts) } |
| 398 | } |
| 399 | } |
| 400 | } |
| 401 | |
| 402 | /// Formats the given floating point number into the decimal form or |
| 403 | /// the exponential form, depending on the resulting exponent. The result is |
| 404 | /// stored to the supplied parts array while utilizing given byte buffer |
| 405 | /// as a scratch. `upper` is used to determine the case of non-finite values |
| 406 | /// (`inf` and `nan`) or the case of the exponent prefix (`e` or `E`). |
| 407 | /// The first part to be rendered is always a `Part::Sign` (which can be |
| 408 | /// an empty string if no sign is rendered). |
| 409 | /// |
| 410 | /// `format_shortest` should be the underlying digit-generation function. |
| 411 | /// It should return the part of the buffer that it initialized. |
| 412 | /// You probably would want `strategy::grisu::format_shortest` for this. |
| 413 | /// |
| 414 | /// The `dec_bounds` is a tuple `(lo, hi)` such that the number is formatted |
| 415 | /// as decimal only when `10^lo <= V < 10^hi`. Note that this is the *apparent* `V` |
| 416 | /// instead of the actual `v`! Thus any printed exponent in the exponential form |
| 417 | /// cannot be in this range, avoiding any confusion. |
| 418 | /// |
| 419 | /// The byte buffer should be at least `MAX_SIG_DIGITS` bytes long. |
| 420 | /// There should be at least 6 parts available, due to the worst case like |
| 421 | /// `[+][1][.][2345][e][-][6]`. |
| 422 | pub fn to_shortest_exp_str<'a, T, F>( |
| 423 | mut format_shortest: F, |
| 424 | v: T, |
| 425 | sign: Sign, |
| 426 | dec_bounds: (i16, i16), |
| 427 | upper: bool, |
| 428 | buf: &'a mut [MaybeUninit<u8>], |
| 429 | parts: &'a mut [MaybeUninit<Part<'a>>], |
| 430 | ) -> Formatted<'a> |
| 431 | where |
| 432 | T: DecodableFloat, |
| 433 | F: FnMut(&Decoded, &'a mut [MaybeUninit<u8>]) -> (&'a [u8], i16), |
| 434 | { |
| 435 | assert!(parts.len() >= 6); |
| 436 | assert!(buf.len() >= MAX_SIG_DIGITS); |
| 437 | assert!(dec_bounds.0 <= dec_bounds.1); |
| 438 | |
| 439 | let (negative, full_decoded) = decode(v); |
| 440 | let sign = determine_sign(sign, &full_decoded, negative); |
| 441 | match full_decoded { |
| 442 | FullDecoded::Nan => { |
| 443 | parts[0] = MaybeUninit::new(Part::Copy(b"NaN" )); |
| 444 | // SAFETY: we just initialized the elements `..1`. |
| 445 | Formatted { sign, parts: unsafe { parts[..1].assume_init_ref() } } |
| 446 | } |
| 447 | FullDecoded::Infinite => { |
| 448 | parts[0] = MaybeUninit::new(Part::Copy(b"inf" )); |
| 449 | // SAFETY: we just initialized the elements `..1`. |
| 450 | Formatted { sign, parts: unsafe { parts[..1].assume_init_ref() } } |
| 451 | } |
| 452 | FullDecoded::Zero => { |
| 453 | parts[0] = if dec_bounds.0 <= 0 && 0 < dec_bounds.1 { |
| 454 | MaybeUninit::new(Part::Copy(b"0" )) |
| 455 | } else { |
| 456 | MaybeUninit::new(Part::Copy(if upper { b"0E0" } else { b"0e0" })) |
| 457 | }; |
| 458 | // SAFETY: we just initialized the elements `..1`. |
| 459 | Formatted { sign, parts: unsafe { parts[..1].assume_init_ref() } } |
| 460 | } |
| 461 | FullDecoded::Finite(ref decoded) => { |
| 462 | let (buf, exp) = format_shortest(decoded, buf); |
| 463 | let vis_exp = exp as i32 - 1; |
| 464 | let parts = if dec_bounds.0 as i32 <= vis_exp && vis_exp < dec_bounds.1 as i32 { |
| 465 | digits_to_dec_str(buf, exp, 0, parts) |
| 466 | } else { |
| 467 | digits_to_exp_str(buf, exp, 0, upper, parts) |
| 468 | }; |
| 469 | Formatted { sign, parts } |
| 470 | } |
| 471 | } |
| 472 | } |
| 473 | |
| 474 | /// Returns a rather crude approximation (upper bound) for the maximum buffer size |
| 475 | /// calculated from the given decoded exponent. |
| 476 | /// |
| 477 | /// The exact limit is: |
| 478 | /// |
| 479 | /// - when `exp < 0`, the maximum length is `ceil(log_10 (5^-exp * (2^64 - 1)))`. |
| 480 | /// - when `exp >= 0`, the maximum length is `ceil(log_10 (2^exp * (2^64 - 1)))`. |
| 481 | /// |
| 482 | /// `ceil(log_10 (x^exp * (2^64 - 1)))` is less than `ceil(log_10 (2^64 - 1)) + |
| 483 | /// ceil(exp * log_10 x)`, which is in turn less than `20 + (1 + exp * log_10 x)`. |
| 484 | /// We use the facts that `log_10 2 < 5/16` and `log_10 5 < 12/16`, which is |
| 485 | /// enough for our purposes. |
| 486 | /// |
| 487 | /// Why do we need this? `format_exact` functions will fill the entire buffer |
| 488 | /// unless limited by the last digit restriction, but it is possible that |
| 489 | /// the number of digits requested is ridiculously large (say, 30,000 digits). |
| 490 | /// The vast majority of buffer will be filled with zeroes, so we don't want to |
| 491 | /// allocate all the buffer beforehand. Consequently, for any given arguments, |
| 492 | /// 826 bytes of buffer should be sufficient for `f64`. Compare this with |
| 493 | /// the actual number for the worst case: 770 bytes (when `exp = -1074`). |
| 494 | fn estimate_max_buf_len(exp: i16) -> usize { |
| 495 | 21 + ((if exp < 0 { -12 } else { 5 } * exp as i32) as usize >> 4) |
| 496 | } |
| 497 | |
| 498 | /// Formats given floating point number into the exponential form with |
| 499 | /// exactly given number of significant digits. The result is stored to |
| 500 | /// the supplied parts array while utilizing given byte buffer as a scratch. |
| 501 | /// `upper` is used to determine the case of the exponent prefix (`e` or `E`). |
| 502 | /// The first part to be rendered is always a `Part::Sign` (which can be |
| 503 | /// an empty string if no sign is rendered). |
| 504 | /// |
| 505 | /// `format_exact` should be the underlying digit-generation function. |
| 506 | /// It should return the part of the buffer that it initialized. |
| 507 | /// You probably would want `strategy::grisu::format_exact` for this. |
| 508 | /// |
| 509 | /// The byte buffer should be at least `ndigits` bytes long unless `ndigits` is |
| 510 | /// so large that only the fixed number of digits will be ever written. |
| 511 | /// (The tipping point for `f64` is about 800, so 1000 bytes should be enough.) |
| 512 | /// There should be at least 6 parts available, due to the worst case like |
| 513 | /// `[+][1][.][2345][e][-][6]`. |
| 514 | pub fn to_exact_exp_str<'a, T, F>( |
| 515 | mut format_exact: F, |
| 516 | v: T, |
| 517 | sign: Sign, |
| 518 | ndigits: usize, |
| 519 | upper: bool, |
| 520 | buf: &'a mut [MaybeUninit<u8>], |
| 521 | parts: &'a mut [MaybeUninit<Part<'a>>], |
| 522 | ) -> Formatted<'a> |
| 523 | where |
| 524 | T: DecodableFloat, |
| 525 | F: FnMut(&Decoded, &'a mut [MaybeUninit<u8>], i16) -> (&'a [u8], i16), |
| 526 | { |
| 527 | assert!(parts.len() >= 6); |
| 528 | assert!(ndigits > 0); |
| 529 | |
| 530 | let (negative, full_decoded) = decode(v); |
| 531 | let sign = determine_sign(sign, &full_decoded, negative); |
| 532 | match full_decoded { |
| 533 | FullDecoded::Nan => { |
| 534 | parts[0] = MaybeUninit::new(Part::Copy(b"NaN" )); |
| 535 | // SAFETY: we just initialized the elements `..1`. |
| 536 | Formatted { sign, parts: unsafe { parts[..1].assume_init_ref() } } |
| 537 | } |
| 538 | FullDecoded::Infinite => { |
| 539 | parts[0] = MaybeUninit::new(Part::Copy(b"inf" )); |
| 540 | // SAFETY: we just initialized the elements `..1`. |
| 541 | Formatted { sign, parts: unsafe { parts[..1].assume_init_ref() } } |
| 542 | } |
| 543 | FullDecoded::Zero => { |
| 544 | if ndigits > 1 { |
| 545 | // [0.][0000][e0] |
| 546 | parts[0] = MaybeUninit::new(Part::Copy(b"0." )); |
| 547 | parts[1] = MaybeUninit::new(Part::Zero(ndigits - 1)); |
| 548 | parts[2] = MaybeUninit::new(Part::Copy(if upper { b"E0" } else { b"e0" })); |
| 549 | Formatted { |
| 550 | sign, |
| 551 | // SAFETY: we just initialized the elements `..3`. |
| 552 | parts: unsafe { parts[..3].assume_init_ref() }, |
| 553 | } |
| 554 | } else { |
| 555 | parts[0] = MaybeUninit::new(Part::Copy(if upper { b"0E0" } else { b"0e0" })); |
| 556 | Formatted { |
| 557 | sign, |
| 558 | // SAFETY: we just initialized the elements `..1`. |
| 559 | parts: unsafe { parts[..1].assume_init_ref() }, |
| 560 | } |
| 561 | } |
| 562 | } |
| 563 | FullDecoded::Finite(ref decoded) => { |
| 564 | let maxlen = estimate_max_buf_len(decoded.exp); |
| 565 | assert!(buf.len() >= ndigits || buf.len() >= maxlen); |
| 566 | |
| 567 | let trunc = if ndigits < maxlen { ndigits } else { maxlen }; |
| 568 | let (buf, exp) = format_exact(decoded, &mut buf[..trunc], i16::MIN); |
| 569 | Formatted { sign, parts: digits_to_exp_str(buf, exp, ndigits, upper, parts) } |
| 570 | } |
| 571 | } |
| 572 | } |
| 573 | |
| 574 | /// Formats given floating point number into the decimal form with exactly |
| 575 | /// given number of fractional digits. The result is stored to the supplied parts |
| 576 | /// array while utilizing given byte buffer as a scratch. `upper` is currently |
| 577 | /// unused but left for the future decision to change the case of non-finite values, |
| 578 | /// i.e., `inf` and `nan`. The first part to be rendered is always a `Part::Sign` |
| 579 | /// (which can be an empty string if no sign is rendered). |
| 580 | /// |
| 581 | /// `format_exact` should be the underlying digit-generation function. |
| 582 | /// It should return the part of the buffer that it initialized. |
| 583 | /// You probably would want `strategy::grisu::format_exact` for this. |
| 584 | /// |
| 585 | /// The byte buffer should be enough for the output unless `frac_digits` is |
| 586 | /// so large that only the fixed number of digits will be ever written. |
| 587 | /// (The tipping point for `f64` is about 800, and 1000 bytes should be enough.) |
| 588 | /// There should be at least 4 parts available, due to the worst case like |
| 589 | /// `[+][0.][0000][2][0000]` with `frac_digits = 10`. |
| 590 | pub fn to_exact_fixed_str<'a, T, F>( |
| 591 | mut format_exact: F, |
| 592 | v: T, |
| 593 | sign: Sign, |
| 594 | frac_digits: usize, |
| 595 | buf: &'a mut [MaybeUninit<u8>], |
| 596 | parts: &'a mut [MaybeUninit<Part<'a>>], |
| 597 | ) -> Formatted<'a> |
| 598 | where |
| 599 | T: DecodableFloat, |
| 600 | F: FnMut(&Decoded, &'a mut [MaybeUninit<u8>], i16) -> (&'a [u8], i16), |
| 601 | { |
| 602 | assert!(parts.len() >= 4); |
| 603 | |
| 604 | let (negative, full_decoded) = decode(v); |
| 605 | let sign = determine_sign(sign, &full_decoded, negative); |
| 606 | match full_decoded { |
| 607 | FullDecoded::Nan => { |
| 608 | parts[0] = MaybeUninit::new(Part::Copy(b"NaN" )); |
| 609 | // SAFETY: we just initialized the elements `..1`. |
| 610 | Formatted { sign, parts: unsafe { parts[..1].assume_init_ref() } } |
| 611 | } |
| 612 | FullDecoded::Infinite => { |
| 613 | parts[0] = MaybeUninit::new(Part::Copy(b"inf" )); |
| 614 | // SAFETY: we just initialized the elements `..1`. |
| 615 | Formatted { sign, parts: unsafe { parts[..1].assume_init_ref() } } |
| 616 | } |
| 617 | FullDecoded::Zero => { |
| 618 | if frac_digits > 0 { |
| 619 | // [0.][0000] |
| 620 | parts[0] = MaybeUninit::new(Part::Copy(b"0." )); |
| 621 | parts[1] = MaybeUninit::new(Part::Zero(frac_digits)); |
| 622 | Formatted { |
| 623 | sign, |
| 624 | // SAFETY: we just initialized the elements `..2`. |
| 625 | parts: unsafe { parts[..2].assume_init_ref() }, |
| 626 | } |
| 627 | } else { |
| 628 | parts[0] = MaybeUninit::new(Part::Copy(b"0" )); |
| 629 | Formatted { |
| 630 | sign, |
| 631 | // SAFETY: we just initialized the elements `..1`. |
| 632 | parts: unsafe { parts[..1].assume_init_ref() }, |
| 633 | } |
| 634 | } |
| 635 | } |
| 636 | FullDecoded::Finite(ref decoded) => { |
| 637 | let maxlen = estimate_max_buf_len(decoded.exp); |
| 638 | assert!(buf.len() >= maxlen); |
| 639 | |
| 640 | // it *is* possible that `frac_digits` is ridiculously large. |
| 641 | // `format_exact` will end rendering digits much earlier in this case, |
| 642 | // because we are strictly limited by `maxlen`. |
| 643 | let limit = if frac_digits < 0x8000 { -(frac_digits as i16) } else { i16::MIN }; |
| 644 | let (buf, exp) = format_exact(decoded, &mut buf[..maxlen], limit); |
| 645 | if exp <= limit { |
| 646 | // the restriction couldn't been met, so this should render like zero no matter |
| 647 | // `exp` was. this does not include the case that the restriction has been met |
| 648 | // only after the final rounding-up; it's a regular case with `exp = limit + 1`. |
| 649 | debug_assert_eq!(buf.len(), 0); |
| 650 | if frac_digits > 0 { |
| 651 | // [0.][0000] |
| 652 | parts[0] = MaybeUninit::new(Part::Copy(b"0." )); |
| 653 | parts[1] = MaybeUninit::new(Part::Zero(frac_digits)); |
| 654 | Formatted { |
| 655 | sign, |
| 656 | // SAFETY: we just initialized the elements `..2`. |
| 657 | parts: unsafe { parts[..2].assume_init_ref() }, |
| 658 | } |
| 659 | } else { |
| 660 | parts[0] = MaybeUninit::new(Part::Copy(b"0" )); |
| 661 | Formatted { |
| 662 | sign, |
| 663 | // SAFETY: we just initialized the elements `..1`. |
| 664 | parts: unsafe { parts[..1].assume_init_ref() }, |
| 665 | } |
| 666 | } |
| 667 | } else { |
| 668 | Formatted { sign, parts: digits_to_dec_str(buf, exp, frac_digits, parts) } |
| 669 | } |
| 670 | } |
| 671 | } |
| 672 | } |
| 673 | |