atof.go 11.1 KB
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544
// Copyright 2009 The Go Authors. All rights reserved.
// Use of this source code is governed by a BSD-style
// license that can be found in the LICENSE file.

package strconv

// decimal to binary floating point conversion.
// Algorithm:
//   1) Store input in multiprecision decimal.
//   2) Multiply/divide decimal by powers of two until in range [0.5, 1)
//   3) Multiply by 2^precision and round to get mantissa.

import "math"
import "runtime"

var optimize = true // can change for testing

func equalIgnoreCase(s1, s2 string) bool {
	if len(s1) != len(s2) {
		return false
	}
	for i := 0; i < len(s1); i++ {
		c1 := s1[i]
		if 'A' <= c1 && c1 <= 'Z' {
			c1 += 'a' - 'A'
		}
		c2 := s2[i]
		if 'A' <= c2 && c2 <= 'Z' {
			c2 += 'a' - 'A'
		}
		if c1 != c2 {
			return false
		}
	}
	return true
}

func special(s string) (f float64, ok bool) {
	if len(s) == 0 {
		return
	}
	switch s[0] {
	default:
		return
	case '+':
		if equalIgnoreCase(s, "+inf") || equalIgnoreCase(s, "+infinity") {
			return math.Inf(1), true
		}
	case '-':
		if equalIgnoreCase(s, "-inf") || equalIgnoreCase(s, "-infinity") {
			return math.Inf(-1), true
		}
	case 'n', 'N':
		if equalIgnoreCase(s, "nan") {
			return math.NaN(), true
		}
	case 'i', 'I':
		if equalIgnoreCase(s, "inf") || equalIgnoreCase(s, "infinity") {
			return math.Inf(1), true
		}
	}
	return
}

func (b *decimal) set(s string) (ok bool) {
	i := 0
	b.neg = false
	b.trunc = false

	// optional sign
	if i >= len(s) {
		return
	}
	switch {
	case s[i] == '+':
		i++
	case s[i] == '-':
		b.neg = true
		i++
	}

	// digits
	sawdot := false
	sawdigits := false
	for ; i < len(s); i++ {
		switch {
		case s[i] == '.':
			if sawdot {
				return
			}
			sawdot = true
			b.dp = b.nd
			continue

		case '0' <= s[i] && s[i] <= '9':
			sawdigits = true
			if s[i] == '0' && b.nd == 0 { // ignore leading zeros
				b.dp--
				continue
			}
			if b.nd < len(b.d) {
				b.d[b.nd] = s[i]
				b.nd++
			} else if s[i] != '0' {
				b.trunc = true
			}
			continue
		}
		break
	}
	if !sawdigits {
		return
	}
	if !sawdot {
		b.dp = b.nd
	}

	// optional exponent moves decimal point.
	// if we read a very large, very long number,
	// just be sure to move the decimal point by
	// a lot (say, 100000).  it doesn't matter if it's
	// not the exact number.
	if i < len(s) && (s[i] == 'e' || s[i] == 'E') {
		i++
		if i >= len(s) {
			return
		}
		esign := 1
		if s[i] == '+' {
			i++
		} else if s[i] == '-' {
			i++
			esign = -1
		}
		if i >= len(s) || s[i] < '0' || s[i] > '9' {
			return
		}
		e := 0
		for ; i < len(s) && '0' <= s[i] && s[i] <= '9'; i++ {
			if e < 10000 {
				e = e*10 + int(s[i]) - '0'
			}
		}
		b.dp += e * esign
	}

	if i != len(s) {
		return
	}

	ok = true
	return
}

// readFloat reads a decimal mantissa and exponent from a float
// string representation. It sets ok to false if the number could
// not fit return types or is invalid.
func readFloat(s string) (mantissa uint64, exp int, neg, trunc, ok bool) {
	const uint64digits = 19
	i := 0

	// optional sign
	if i >= len(s) {
		return
	}
	switch {
	case s[i] == '+':
		i++
	case s[i] == '-':
		neg = true
		i++
	}

	// digits
	sawdot := false
	sawdigits := false
	nd := 0
	ndMant := 0
	dp := 0
	for ; i < len(s); i++ {
		switch c := s[i]; true {
		case c == '.':
			if sawdot {
				return
			}
			sawdot = true
			dp = nd
			continue

		case '0' <= c && c <= '9':
			sawdigits = true
			if c == '0' && nd == 0 { // ignore leading zeros
				dp--
				continue
			}
			nd++
			if ndMant < uint64digits {
				mantissa *= 10
				mantissa += uint64(c - '0')
				ndMant++
			} else if s[i] != '0' {
				trunc = true
			}
			continue
		}
		break
	}
	if !sawdigits {
		return
	}
	if !sawdot {
		dp = nd
	}

	// optional exponent moves decimal point.
	// if we read a very large, very long number,
	// just be sure to move the decimal point by
	// a lot (say, 100000).  it doesn't matter if it's
	// not the exact number.
	if i < len(s) && (s[i] == 'e' || s[i] == 'E') {
		i++
		if i >= len(s) {
			return
		}
		esign := 1
		if s[i] == '+' {
			i++
		} else if s[i] == '-' {
			i++
			esign = -1
		}
		if i >= len(s) || s[i] < '0' || s[i] > '9' {
			return
		}
		e := 0
		for ; i < len(s) && '0' <= s[i] && s[i] <= '9'; i++ {
			if e < 10000 {
				e = e*10 + int(s[i]) - '0'
			}
		}
		dp += e * esign
	}

	if i != len(s) {
		return
	}

	exp = dp - ndMant
	ok = true
	return

}

// decimal power of ten to binary power of two.
var powtab = []int{1, 3, 6, 9, 13, 16, 19, 23, 26}

func (d *decimal) floatBits(flt *floatInfo) (b uint64, overflow bool) {
	var exp int
	var mant uint64

	// Zero is always a special case.
	if d.nd == 0 {
		mant = 0
		exp = flt.bias
		goto out
	}

	// Obvious overflow/underflow.
	// These bounds are for 64-bit floats.
	// Will have to change if we want to support 80-bit floats in the future.
	if d.dp > 310 {
		goto overflow
	}
	if d.dp < -330 {
		// zero
		mant = 0
		exp = flt.bias
		goto out
	}

	// Scale by powers of two until in range [0.5, 1.0)
	exp = 0
	for d.dp > 0 {
		var n int
		if d.dp >= len(powtab) {
			n = 27
		} else {
			n = powtab[d.dp]
		}
		d.Shift(-n)
		exp += n
	}
	for d.dp < 0 || d.dp == 0 && d.d[0] < '5' {
		var n int
		if -d.dp >= len(powtab) {
			n = 27
		} else {
			n = powtab[-d.dp]
		}
		d.Shift(n)
		exp -= n
	}

	// Our range is [0.5,1) but floating point range is [1,2).
	exp--

	// Minimum representable exponent is flt.bias+1.
	// If the exponent is smaller, move it up and
	// adjust d accordingly.
	if exp < flt.bias+1 {
		n := flt.bias + 1 - exp
		d.Shift(-n)
		exp += n
	}

	if exp-flt.bias >= 1<<flt.expbits-1 {
		goto overflow
	}

	// Extract 1+flt.mantbits bits.
	d.Shift(int(1 + flt.mantbits))
	mant = d.RoundedInteger()

	// Rounding might have added a bit; shift down.
	if mant == 2<<flt.mantbits {
		mant >>= 1
		exp++
		if exp-flt.bias >= 1<<flt.expbits-1 {
			goto overflow
		}
	}

	// Denormalized?
	if mant&(1<<flt.mantbits) == 0 {
		exp = flt.bias
	}
	goto out

overflow:
	// ±Inf
	mant = 0
	exp = 1<<flt.expbits - 1 + flt.bias
	overflow = true

out:
	// Assemble bits.
	bits := mant & (uint64(1)<<flt.mantbits - 1)
	bits |= uint64((exp-flt.bias)&(1<<flt.expbits-1)) << flt.mantbits
	if d.neg {
		bits |= 1 << flt.mantbits << flt.expbits
	}
	return bits, overflow
}

// Exact powers of 10.
var float64pow10 = []float64{
	1e0, 1e1, 1e2, 1e3, 1e4, 1e5, 1e6, 1e7, 1e8, 1e9,
	1e10, 1e11, 1e12, 1e13, 1e14, 1e15, 1e16, 1e17, 1e18, 1e19,
	1e20, 1e21, 1e22,
}
var float32pow10 = []float32{1e0, 1e1, 1e2, 1e3, 1e4, 1e5, 1e6, 1e7, 1e8, 1e9, 1e10}

// If possible to convert decimal representation to 64-bit float f exactly,
// entirely in floating-point math, do so, avoiding the expense of decimalToFloatBits.
// Three common cases:
//	value is exact integer
//	value is exact integer * exact power of ten
//	value is exact integer / exact power of ten
// These all produce potentially inexact but correctly rounded answers.
func atof64exact(mantissa uint64, exp int, neg bool) (f float64, ok bool) {
	if mantissa>>float64info.mantbits != 0 {
		return
	}
	// gccgo gets this wrong on 32-bit i386 when not using -msse.
	// See TestRoundTrip in atof_test.go for a test case.
	if runtime.GOARCH == "386" {
		return
	}
	f = float64(mantissa)
	if neg {
		f = -f
	}
	switch {
	case exp == 0:
		// an integer.
		return f, true
	// Exact integers are <= 10^15.
	// Exact powers of ten are <= 10^22.
	case exp > 0 && exp <= 15+22: // int * 10^k
		// If exponent is big but number of digits is not,
		// can move a few zeros into the integer part.
		if exp > 22 {
			f *= float64pow10[exp-22]
			exp = 22
		}
		if f > 1e15 || f < -1e15 {
			// the exponent was really too large.
			return
		}
		return f * float64pow10[exp], true
	case exp < 0 && exp >= -22: // int / 10^k
		return f / float64pow10[-exp], true
	}
	return
}

// If possible to compute mantissa*10^exp to 32-bit float f exactly,
// entirely in floating-point math, do so, avoiding the machinery above.
func atof32exact(mantissa uint64, exp int, neg bool) (f float32, ok bool) {
	if mantissa>>float32info.mantbits != 0 {
		return
	}
	f = float32(mantissa)
	if neg {
		f = -f
	}
	switch {
	case exp == 0:
		return f, true
	// Exact integers are <= 10^7.
	// Exact powers of ten are <= 10^10.
	case exp > 0 && exp <= 7+10: // int * 10^k
		// If exponent is big but number of digits is not,
		// can move a few zeros into the integer part.
		if exp > 10 {
			f *= float32pow10[exp-10]
			exp = 10
		}
		if f > 1e7 || f < -1e7 {
			// the exponent was really too large.
			return
		}
		return f * float32pow10[exp], true
	case exp < 0 && exp >= -10: // int / 10^k
		return f / float32pow10[-exp], true
	}
	return
}

const fnParseFloat = "ParseFloat"

func atof32(s string) (f float32, err error) {
	if val, ok := special(s); ok {
		return float32(val), nil
	}

	if optimize {
		// Parse mantissa and exponent.
		mantissa, exp, neg, trunc, ok := readFloat(s)
		if ok {
			// Try pure floating-point arithmetic conversion.
			if !trunc {
				if f, ok := atof32exact(mantissa, exp, neg); ok {
					return f, nil
				}
			}
			// Try another fast path.
			ext := new(extFloat)
			if ok := ext.AssignDecimal(mantissa, exp, neg, trunc, &float32info); ok {
				b, ovf := ext.floatBits(&float32info)
				f = math.Float32frombits(uint32(b))
				if ovf {
					err = rangeError(fnParseFloat, s)
				}
				return f, err
			}
		}
	}
	var d decimal
	if !d.set(s) {
		return 0, syntaxError(fnParseFloat, s)
	}
	b, ovf := d.floatBits(&float32info)
	f = math.Float32frombits(uint32(b))
	if ovf {
		err = rangeError(fnParseFloat, s)
	}
	return f, err
}

func atof64(s string) (f float64, err error) {
	if val, ok := special(s); ok {
		return val, nil
	}

	if optimize {
		// Parse mantissa and exponent.
		mantissa, exp, neg, trunc, ok := readFloat(s)
		if ok {
			// Try pure floating-point arithmetic conversion.
			if !trunc {
				if f, ok := atof64exact(mantissa, exp, neg); ok {
					return f, nil
				}
			}
			// Try another fast path.
			ext := new(extFloat)
			if ok := ext.AssignDecimal(mantissa, exp, neg, trunc, &float64info); ok {
				b, ovf := ext.floatBits(&float64info)
				f = math.Float64frombits(b)
				if ovf {
					err = rangeError(fnParseFloat, s)
				}
				return f, err
			}
		}
	}
	var d decimal
	if !d.set(s) {
		return 0, syntaxError(fnParseFloat, s)
	}
	b, ovf := d.floatBits(&float64info)
	f = math.Float64frombits(b)
	if ovf {
		err = rangeError(fnParseFloat, s)
	}
	return f, err
}

// ParseFloat converts the string s to a floating-point number
// with the precision specified by bitSize: 32 for float32, or 64 for float64.
// When bitSize=32, the result still has type float64, but it will be
// convertible to float32 without changing its value.
//
// If s is well-formed and near a valid floating point number,
// ParseFloat returns the nearest floating point number rounded
// using IEEE754 unbiased rounding.
//
// The errors that ParseFloat returns have concrete type *NumError
// and include err.Num = s.
//
// If s is not syntactically well-formed, ParseFloat returns err.Err = ErrSyntax.
//
// If s is syntactically well-formed but is more than 1/2 ULP
// away from the largest floating point number of the given size,
// ParseFloat returns f = ±Inf, err.Err = ErrRange.
func ParseFloat(s string, bitSize int) (f float64, err error) {
	if bitSize == 32 {
		f1, err1 := atof32(s)
		return float64(f1), err1
	}
	f1, err1 := atof64(s)
	return f1, err1
}