1 // Copyright 2011 the V8 project authors. All rights reserved.
2 // Redistribution and use in source and binary forms, with or without
3 // modification, are permitted provided that the following conditions are
4 // met:
5 //
6 // * Redistributions of source code must retain the above copyright
7 // notice, this list of conditions and the following disclaimer.
8 // * Redistributions in binary form must reproduce the above
9 // copyright notice, this list of conditions and the following
10 // disclaimer in the documentation and/or other materials provided
11 // with the distribution.
12 // * Neither the name of Google Inc. nor the names of its
13 // contributors may be used to endorse or promote products derived
14 // from this software without specific prior written permission.
15 //
16 // THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
17 // "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
18 // LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
19 // A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
20 // OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
21 // SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
22 // LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
23 // DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
24 // THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
25 // (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
26 // OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
27
28 #include <stdarg.h>
29 #include <limits.h>
30
31 #include "v8.h"
32
33 #include "conversions-inl.h"
34 #include "dtoa.h"
35 #include "factory.h"
36 #include "scanner-base.h"
37 #include "strtod.h"
38
39 namespace v8 {
40 namespace internal {
41
42 namespace {
43
44 // C++-style iterator adaptor for StringInputBuffer
45 // (unlike C++ iterators the end-marker has different type).
46 class StringInputBufferIterator {
47 public:
48 class EndMarker {};
49
50 explicit StringInputBufferIterator(StringInputBuffer* buffer);
51
52 int operator*() const;
53 void operator++();
operator ==(EndMarker const &) const54 bool operator==(EndMarker const&) const { return end_; }
operator !=(EndMarker const & m) const55 bool operator!=(EndMarker const& m) const { return !end_; }
56
57 private:
58 StringInputBuffer* const buffer_;
59 int current_;
60 bool end_;
61 };
62
63
StringInputBufferIterator(StringInputBuffer * buffer)64 StringInputBufferIterator::StringInputBufferIterator(
65 StringInputBuffer* buffer) : buffer_(buffer) {
66 ++(*this);
67 }
68
operator *() const69 int StringInputBufferIterator::operator*() const {
70 return current_;
71 }
72
73
operator ++()74 void StringInputBufferIterator::operator++() {
75 end_ = !buffer_->has_more();
76 if (!end_) {
77 current_ = buffer_->GetNext();
78 }
79 }
80 }
81
82
83 template <class Iterator, class EndMark>
SubStringEquals(Iterator * current,EndMark end,const char * substring)84 static bool SubStringEquals(Iterator* current,
85 EndMark end,
86 const char* substring) {
87 ASSERT(**current == *substring);
88 for (substring++; *substring != '\0'; substring++) {
89 ++*current;
90 if (*current == end || **current != *substring) return false;
91 }
92 ++*current;
93 return true;
94 }
95
96
97 // Maximum number of significant digits in decimal representation.
98 // The longest possible double in decimal representation is
99 // (2^53 - 1) * 2 ^ -1074 that is (2 ^ 53 - 1) * 5 ^ 1074 / 10 ^ 1074
100 // (768 digits). If we parse a number whose first digits are equal to a
101 // mean of 2 adjacent doubles (that could have up to 769 digits) the result
102 // must be rounded to the bigger one unless the tail consists of zeros, so
103 // we don't need to preserve all the digits.
104 const int kMaxSignificantDigits = 772;
105
106
107 static const double JUNK_STRING_VALUE = OS::nan_value();
108
109
110 // Returns true if a nonspace found and false if the end has reached.
111 template <class Iterator, class EndMark>
AdvanceToNonspace(UnicodeCache * unicode_cache,Iterator * current,EndMark end)112 static inline bool AdvanceToNonspace(UnicodeCache* unicode_cache,
113 Iterator* current,
114 EndMark end) {
115 while (*current != end) {
116 if (!unicode_cache->IsWhiteSpace(**current)) return true;
117 ++*current;
118 }
119 return false;
120 }
121
122
isDigit(int x,int radix)123 static bool isDigit(int x, int radix) {
124 return (x >= '0' && x <= '9' && x < '0' + radix)
125 || (radix > 10 && x >= 'a' && x < 'a' + radix - 10)
126 || (radix > 10 && x >= 'A' && x < 'A' + radix - 10);
127 }
128
129
SignedZero(bool negative)130 static double SignedZero(bool negative) {
131 return negative ? -0.0 : 0.0;
132 }
133
134
135 // Parsing integers with radix 2, 4, 8, 16, 32. Assumes current != end.
136 template <int radix_log_2, class Iterator, class EndMark>
InternalStringToIntDouble(UnicodeCache * unicode_cache,Iterator current,EndMark end,bool negative,bool allow_trailing_junk)137 static double InternalStringToIntDouble(UnicodeCache* unicode_cache,
138 Iterator current,
139 EndMark end,
140 bool negative,
141 bool allow_trailing_junk) {
142 ASSERT(current != end);
143
144 // Skip leading 0s.
145 while (*current == '0') {
146 ++current;
147 if (current == end) return SignedZero(negative);
148 }
149
150 int64_t number = 0;
151 int exponent = 0;
152 const int radix = (1 << radix_log_2);
153
154 do {
155 int digit;
156 if (*current >= '0' && *current <= '9' && *current < '0' + radix) {
157 digit = static_cast<char>(*current) - '0';
158 } else if (radix > 10 && *current >= 'a' && *current < 'a' + radix - 10) {
159 digit = static_cast<char>(*current) - 'a' + 10;
160 } else if (radix > 10 && *current >= 'A' && *current < 'A' + radix - 10) {
161 digit = static_cast<char>(*current) - 'A' + 10;
162 } else {
163 if (allow_trailing_junk ||
164 !AdvanceToNonspace(unicode_cache, ¤t, end)) {
165 break;
166 } else {
167 return JUNK_STRING_VALUE;
168 }
169 }
170
171 number = number * radix + digit;
172 int overflow = static_cast<int>(number >> 53);
173 if (overflow != 0) {
174 // Overflow occurred. Need to determine which direction to round the
175 // result.
176 int overflow_bits_count = 1;
177 while (overflow > 1) {
178 overflow_bits_count++;
179 overflow >>= 1;
180 }
181
182 int dropped_bits_mask = ((1 << overflow_bits_count) - 1);
183 int dropped_bits = static_cast<int>(number) & dropped_bits_mask;
184 number >>= overflow_bits_count;
185 exponent = overflow_bits_count;
186
187 bool zero_tail = true;
188 while (true) {
189 ++current;
190 if (current == end || !isDigit(*current, radix)) break;
191 zero_tail = zero_tail && *current == '0';
192 exponent += radix_log_2;
193 }
194
195 if (!allow_trailing_junk &&
196 AdvanceToNonspace(unicode_cache, ¤t, end)) {
197 return JUNK_STRING_VALUE;
198 }
199
200 int middle_value = (1 << (overflow_bits_count - 1));
201 if (dropped_bits > middle_value) {
202 number++; // Rounding up.
203 } else if (dropped_bits == middle_value) {
204 // Rounding to even to consistency with decimals: half-way case rounds
205 // up if significant part is odd and down otherwise.
206 if ((number & 1) != 0 || !zero_tail) {
207 number++; // Rounding up.
208 }
209 }
210
211 // Rounding up may cause overflow.
212 if ((number & ((int64_t)1 << 53)) != 0) {
213 exponent++;
214 number >>= 1;
215 }
216 break;
217 }
218 ++current;
219 } while (current != end);
220
221 ASSERT(number < ((int64_t)1 << 53));
222 ASSERT(static_cast<int64_t>(static_cast<double>(number)) == number);
223
224 if (exponent == 0) {
225 if (negative) {
226 if (number == 0) return -0.0;
227 number = -number;
228 }
229 return static_cast<double>(number);
230 }
231
232 ASSERT(number != 0);
233 // The double could be constructed faster from number (mantissa), exponent
234 // and sign. Assuming it's a rare case more simple code is used.
235 return static_cast<double>(negative ? -number : number) * pow(2.0, exponent);
236 }
237
238
239 template <class Iterator, class EndMark>
InternalStringToInt(UnicodeCache * unicode_cache,Iterator current,EndMark end,int radix)240 static double InternalStringToInt(UnicodeCache* unicode_cache,
241 Iterator current,
242 EndMark end,
243 int radix) {
244 const bool allow_trailing_junk = true;
245 const double empty_string_val = JUNK_STRING_VALUE;
246
247 if (!AdvanceToNonspace(unicode_cache, ¤t, end)) {
248 return empty_string_val;
249 }
250
251 bool negative = false;
252 bool leading_zero = false;
253
254 if (*current == '+') {
255 // Ignore leading sign; skip following spaces.
256 ++current;
257 if (!AdvanceToNonspace(unicode_cache, ¤t, end)) {
258 return JUNK_STRING_VALUE;
259 }
260 } else if (*current == '-') {
261 ++current;
262 if (!AdvanceToNonspace(unicode_cache, ¤t, end)) {
263 return JUNK_STRING_VALUE;
264 }
265 negative = true;
266 }
267
268 if (radix == 0) {
269 // Radix detection.
270 if (*current == '0') {
271 ++current;
272 if (current == end) return SignedZero(negative);
273 if (*current == 'x' || *current == 'X') {
274 radix = 16;
275 ++current;
276 if (current == end) return JUNK_STRING_VALUE;
277 } else {
278 radix = 8;
279 leading_zero = true;
280 }
281 } else {
282 radix = 10;
283 }
284 } else if (radix == 16) {
285 if (*current == '0') {
286 // Allow "0x" prefix.
287 ++current;
288 if (current == end) return SignedZero(negative);
289 if (*current == 'x' || *current == 'X') {
290 ++current;
291 if (current == end) return JUNK_STRING_VALUE;
292 } else {
293 leading_zero = true;
294 }
295 }
296 }
297
298 if (radix < 2 || radix > 36) return JUNK_STRING_VALUE;
299
300 // Skip leading zeros.
301 while (*current == '0') {
302 leading_zero = true;
303 ++current;
304 if (current == end) return SignedZero(negative);
305 }
306
307 if (!leading_zero && !isDigit(*current, radix)) {
308 return JUNK_STRING_VALUE;
309 }
310
311 if (IsPowerOf2(radix)) {
312 switch (radix) {
313 case 2:
314 return InternalStringToIntDouble<1>(
315 unicode_cache, current, end, negative, allow_trailing_junk);
316 case 4:
317 return InternalStringToIntDouble<2>(
318 unicode_cache, current, end, negative, allow_trailing_junk);
319 case 8:
320 return InternalStringToIntDouble<3>(
321 unicode_cache, current, end, negative, allow_trailing_junk);
322
323 case 16:
324 return InternalStringToIntDouble<4>(
325 unicode_cache, current, end, negative, allow_trailing_junk);
326
327 case 32:
328 return InternalStringToIntDouble<5>(
329 unicode_cache, current, end, negative, allow_trailing_junk);
330 default:
331 UNREACHABLE();
332 }
333 }
334
335 if (radix == 10) {
336 // Parsing with strtod.
337 const int kMaxSignificantDigits = 309; // Doubles are less than 1.8e308.
338 // The buffer may contain up to kMaxSignificantDigits + 1 digits and a zero
339 // end.
340 const int kBufferSize = kMaxSignificantDigits + 2;
341 char buffer[kBufferSize];
342 int buffer_pos = 0;
343 while (*current >= '0' && *current <= '9') {
344 if (buffer_pos <= kMaxSignificantDigits) {
345 // If the number has more than kMaxSignificantDigits it will be parsed
346 // as infinity.
347 ASSERT(buffer_pos < kBufferSize);
348 buffer[buffer_pos++] = static_cast<char>(*current);
349 }
350 ++current;
351 if (current == end) break;
352 }
353
354 if (!allow_trailing_junk &&
355 AdvanceToNonspace(unicode_cache, ¤t, end)) {
356 return JUNK_STRING_VALUE;
357 }
358
359 ASSERT(buffer_pos < kBufferSize);
360 buffer[buffer_pos] = '\0';
361 Vector<const char> buffer_vector(buffer, buffer_pos);
362 return negative ? -Strtod(buffer_vector, 0) : Strtod(buffer_vector, 0);
363 }
364
365 // The following code causes accumulating rounding error for numbers greater
366 // than ~2^56. It's explicitly allowed in the spec: "if R is not 2, 4, 8, 10,
367 // 16, or 32, then mathInt may be an implementation-dependent approximation to
368 // the mathematical integer value" (15.1.2.2).
369
370 int lim_0 = '0' + (radix < 10 ? radix : 10);
371 int lim_a = 'a' + (radix - 10);
372 int lim_A = 'A' + (radix - 10);
373
374 // NOTE: The code for computing the value may seem a bit complex at
375 // first glance. It is structured to use 32-bit multiply-and-add
376 // loops as long as possible to avoid loosing precision.
377
378 double v = 0.0;
379 bool done = false;
380 do {
381 // Parse the longest part of the string starting at index j
382 // possible while keeping the multiplier, and thus the part
383 // itself, within 32 bits.
384 unsigned int part = 0, multiplier = 1;
385 while (true) {
386 int d;
387 if (*current >= '0' && *current < lim_0) {
388 d = *current - '0';
389 } else if (*current >= 'a' && *current < lim_a) {
390 d = *current - 'a' + 10;
391 } else if (*current >= 'A' && *current < lim_A) {
392 d = *current - 'A' + 10;
393 } else {
394 done = true;
395 break;
396 }
397
398 // Update the value of the part as long as the multiplier fits
399 // in 32 bits. When we can't guarantee that the next iteration
400 // will not overflow the multiplier, we stop parsing the part
401 // by leaving the loop.
402 const unsigned int kMaximumMultiplier = 0xffffffffU / 36;
403 uint32_t m = multiplier * radix;
404 if (m > kMaximumMultiplier) break;
405 part = part * radix + d;
406 multiplier = m;
407 ASSERT(multiplier > part);
408
409 ++current;
410 if (current == end) {
411 done = true;
412 break;
413 }
414 }
415
416 // Update the value and skip the part in the string.
417 v = v * multiplier + part;
418 } while (!done);
419
420 if (!allow_trailing_junk &&
421 AdvanceToNonspace(unicode_cache, ¤t, end)) {
422 return JUNK_STRING_VALUE;
423 }
424
425 return negative ? -v : v;
426 }
427
428
429 // Converts a string to a double value. Assumes the Iterator supports
430 // the following operations:
431 // 1. current == end (other ops are not allowed), current != end.
432 // 2. *current - gets the current character in the sequence.
433 // 3. ++current (advances the position).
434 template <class Iterator, class EndMark>
InternalStringToDouble(UnicodeCache * unicode_cache,Iterator current,EndMark end,int flags,double empty_string_val)435 static double InternalStringToDouble(UnicodeCache* unicode_cache,
436 Iterator current,
437 EndMark end,
438 int flags,
439 double empty_string_val) {
440 // To make sure that iterator dereferencing is valid the following
441 // convention is used:
442 // 1. Each '++current' statement is followed by check for equality to 'end'.
443 // 2. If AdvanceToNonspace returned false then current == end.
444 // 3. If 'current' becomes be equal to 'end' the function returns or goes to
445 // 'parsing_done'.
446 // 4. 'current' is not dereferenced after the 'parsing_done' label.
447 // 5. Code before 'parsing_done' may rely on 'current != end'.
448 if (!AdvanceToNonspace(unicode_cache, ¤t, end)) {
449 return empty_string_val;
450 }
451
452 const bool allow_trailing_junk = (flags & ALLOW_TRAILING_JUNK) != 0;
453
454 // The longest form of simplified number is: "-<significant digits>'.1eXXX\0".
455 const int kBufferSize = kMaxSignificantDigits + 10;
456 char buffer[kBufferSize]; // NOLINT: size is known at compile time.
457 int buffer_pos = 0;
458
459 // Exponent will be adjusted if insignificant digits of the integer part
460 // or insignificant leading zeros of the fractional part are dropped.
461 int exponent = 0;
462 int significant_digits = 0;
463 int insignificant_digits = 0;
464 bool nonzero_digit_dropped = false;
465 bool fractional_part = false;
466
467 bool negative = false;
468
469 if (*current == '+') {
470 // Ignore leading sign.
471 ++current;
472 if (current == end) return JUNK_STRING_VALUE;
473 } else if (*current == '-') {
474 ++current;
475 if (current == end) return JUNK_STRING_VALUE;
476 negative = true;
477 }
478
479 static const char kInfinitySymbol[] = "Infinity";
480 if (*current == kInfinitySymbol[0]) {
481 if (!SubStringEquals(¤t, end, kInfinitySymbol)) {
482 return JUNK_STRING_VALUE;
483 }
484
485 if (!allow_trailing_junk &&
486 AdvanceToNonspace(unicode_cache, ¤t, end)) {
487 return JUNK_STRING_VALUE;
488 }
489
490 ASSERT(buffer_pos == 0);
491 return negative ? -V8_INFINITY : V8_INFINITY;
492 }
493
494 bool leading_zero = false;
495 if (*current == '0') {
496 ++current;
497 if (current == end) return SignedZero(negative);
498
499 leading_zero = true;
500
501 // It could be hexadecimal value.
502 if ((flags & ALLOW_HEX) && (*current == 'x' || *current == 'X')) {
503 ++current;
504 if (current == end || !isDigit(*current, 16)) {
505 return JUNK_STRING_VALUE; // "0x".
506 }
507
508 return InternalStringToIntDouble<4>(unicode_cache,
509 current,
510 end,
511 negative,
512 allow_trailing_junk);
513 }
514
515 // Ignore leading zeros in the integer part.
516 while (*current == '0') {
517 ++current;
518 if (current == end) return SignedZero(negative);
519 }
520 }
521
522 bool octal = leading_zero && (flags & ALLOW_OCTALS) != 0;
523
524 // Copy significant digits of the integer part (if any) to the buffer.
525 while (*current >= '0' && *current <= '9') {
526 if (significant_digits < kMaxSignificantDigits) {
527 ASSERT(buffer_pos < kBufferSize);
528 buffer[buffer_pos++] = static_cast<char>(*current);
529 significant_digits++;
530 // Will later check if it's an octal in the buffer.
531 } else {
532 insignificant_digits++; // Move the digit into the exponential part.
533 nonzero_digit_dropped = nonzero_digit_dropped || *current != '0';
534 }
535 octal = octal && *current < '8';
536 ++current;
537 if (current == end) goto parsing_done;
538 }
539
540 if (significant_digits == 0) {
541 octal = false;
542 }
543
544 if (*current == '.') {
545 if (octal && !allow_trailing_junk) return JUNK_STRING_VALUE;
546 if (octal) goto parsing_done;
547
548 ++current;
549 if (current == end) {
550 if (significant_digits == 0 && !leading_zero) {
551 return JUNK_STRING_VALUE;
552 } else {
553 goto parsing_done;
554 }
555 }
556
557 if (significant_digits == 0) {
558 // octal = false;
559 // Integer part consists of 0 or is absent. Significant digits start after
560 // leading zeros (if any).
561 while (*current == '0') {
562 ++current;
563 if (current == end) return SignedZero(negative);
564 exponent--; // Move this 0 into the exponent.
565 }
566 }
567
568 // We don't emit a '.', but adjust the exponent instead.
569 fractional_part = true;
570
571 // There is a fractional part.
572 while (*current >= '0' && *current <= '9') {
573 if (significant_digits < kMaxSignificantDigits) {
574 ASSERT(buffer_pos < kBufferSize);
575 buffer[buffer_pos++] = static_cast<char>(*current);
576 significant_digits++;
577 exponent--;
578 } else {
579 // Ignore insignificant digits in the fractional part.
580 nonzero_digit_dropped = nonzero_digit_dropped || *current != '0';
581 }
582 ++current;
583 if (current == end) goto parsing_done;
584 }
585 }
586
587 if (!leading_zero && exponent == 0 && significant_digits == 0) {
588 // If leading_zeros is true then the string contains zeros.
589 // If exponent < 0 then string was [+-]\.0*...
590 // If significant_digits != 0 the string is not equal to 0.
591 // Otherwise there are no digits in the string.
592 return JUNK_STRING_VALUE;
593 }
594
595 // Parse exponential part.
596 if (*current == 'e' || *current == 'E') {
597 if (octal) return JUNK_STRING_VALUE;
598 ++current;
599 if (current == end) {
600 if (allow_trailing_junk) {
601 goto parsing_done;
602 } else {
603 return JUNK_STRING_VALUE;
604 }
605 }
606 char sign = '+';
607 if (*current == '+' || *current == '-') {
608 sign = static_cast<char>(*current);
609 ++current;
610 if (current == end) {
611 if (allow_trailing_junk) {
612 goto parsing_done;
613 } else {
614 return JUNK_STRING_VALUE;
615 }
616 }
617 }
618
619 if (current == end || *current < '0' || *current > '9') {
620 if (allow_trailing_junk) {
621 goto parsing_done;
622 } else {
623 return JUNK_STRING_VALUE;
624 }
625 }
626
627 const int max_exponent = INT_MAX / 2;
628 ASSERT(-max_exponent / 2 <= exponent && exponent <= max_exponent / 2);
629 int num = 0;
630 do {
631 // Check overflow.
632 int digit = *current - '0';
633 if (num >= max_exponent / 10
634 && !(num == max_exponent / 10 && digit <= max_exponent % 10)) {
635 num = max_exponent;
636 } else {
637 num = num * 10 + digit;
638 }
639 ++current;
640 } while (current != end && *current >= '0' && *current <= '9');
641
642 exponent += (sign == '-' ? -num : num);
643 }
644
645 if (!allow_trailing_junk &&
646 AdvanceToNonspace(unicode_cache, ¤t, end)) {
647 return JUNK_STRING_VALUE;
648 }
649
650 parsing_done:
651 exponent += insignificant_digits;
652
653 if (octal) {
654 return InternalStringToIntDouble<3>(unicode_cache,
655 buffer,
656 buffer + buffer_pos,
657 negative,
658 allow_trailing_junk);
659 }
660
661 if (nonzero_digit_dropped) {
662 buffer[buffer_pos++] = '1';
663 exponent--;
664 }
665
666 ASSERT(buffer_pos < kBufferSize);
667 buffer[buffer_pos] = '\0';
668
669 double converted = Strtod(Vector<const char>(buffer, buffer_pos), exponent);
670 return negative ? -converted : converted;
671 }
672
673
StringToDouble(UnicodeCache * unicode_cache,String * str,int flags,double empty_string_val)674 double StringToDouble(UnicodeCache* unicode_cache,
675 String* str, int flags, double empty_string_val) {
676 StringShape shape(str);
677 if (shape.IsSequentialAscii()) {
678 const char* begin = SeqAsciiString::cast(str)->GetChars();
679 const char* end = begin + str->length();
680 return InternalStringToDouble(unicode_cache, begin, end, flags,
681 empty_string_val);
682 } else if (shape.IsSequentialTwoByte()) {
683 const uc16* begin = SeqTwoByteString::cast(str)->GetChars();
684 const uc16* end = begin + str->length();
685 return InternalStringToDouble(unicode_cache, begin, end, flags,
686 empty_string_val);
687 } else {
688 StringInputBuffer buffer(str);
689 return InternalStringToDouble(unicode_cache,
690 StringInputBufferIterator(&buffer),
691 StringInputBufferIterator::EndMarker(),
692 flags,
693 empty_string_val);
694 }
695 }
696
697
StringToInt(UnicodeCache * unicode_cache,String * str,int radix)698 double StringToInt(UnicodeCache* unicode_cache,
699 String* str,
700 int radix) {
701 StringShape shape(str);
702 if (shape.IsSequentialAscii()) {
703 const char* begin = SeqAsciiString::cast(str)->GetChars();
704 const char* end = begin + str->length();
705 return InternalStringToInt(unicode_cache, begin, end, radix);
706 } else if (shape.IsSequentialTwoByte()) {
707 const uc16* begin = SeqTwoByteString::cast(str)->GetChars();
708 const uc16* end = begin + str->length();
709 return InternalStringToInt(unicode_cache, begin, end, radix);
710 } else {
711 StringInputBuffer buffer(str);
712 return InternalStringToInt(unicode_cache,
713 StringInputBufferIterator(&buffer),
714 StringInputBufferIterator::EndMarker(),
715 radix);
716 }
717 }
718
719
StringToDouble(UnicodeCache * unicode_cache,const char * str,int flags,double empty_string_val)720 double StringToDouble(UnicodeCache* unicode_cache,
721 const char* str, int flags, double empty_string_val) {
722 const char* end = str + StrLength(str);
723 return InternalStringToDouble(unicode_cache, str, end, flags,
724 empty_string_val);
725 }
726
727
StringToDouble(UnicodeCache * unicode_cache,Vector<const char> str,int flags,double empty_string_val)728 double StringToDouble(UnicodeCache* unicode_cache,
729 Vector<const char> str,
730 int flags,
731 double empty_string_val) {
732 const char* end = str.start() + str.length();
733 return InternalStringToDouble(unicode_cache, str.start(), end, flags,
734 empty_string_val);
735 }
736
737
DoubleToCString(double v,Vector<char> buffer)738 const char* DoubleToCString(double v, Vector<char> buffer) {
739 switch (fpclassify(v)) {
740 case FP_NAN: return "NaN";
741 case FP_INFINITE: return (v < 0.0 ? "-Infinity" : "Infinity");
742 case FP_ZERO: return "0";
743 default: {
744 StringBuilder builder(buffer.start(), buffer.length());
745 int decimal_point;
746 int sign;
747 const int kV8DtoaBufferCapacity = kBase10MaximalLength + 1;
748 char decimal_rep[kV8DtoaBufferCapacity];
749 int length;
750
751 DoubleToAscii(v, DTOA_SHORTEST, 0,
752 Vector<char>(decimal_rep, kV8DtoaBufferCapacity),
753 &sign, &length, &decimal_point);
754
755 if (sign) builder.AddCharacter('-');
756
757 if (length <= decimal_point && decimal_point <= 21) {
758 // ECMA-262 section 9.8.1 step 6.
759 builder.AddString(decimal_rep);
760 builder.AddPadding('0', decimal_point - length);
761
762 } else if (0 < decimal_point && decimal_point <= 21) {
763 // ECMA-262 section 9.8.1 step 7.
764 builder.AddSubstring(decimal_rep, decimal_point);
765 builder.AddCharacter('.');
766 builder.AddString(decimal_rep + decimal_point);
767
768 } else if (decimal_point <= 0 && decimal_point > -6) {
769 // ECMA-262 section 9.8.1 step 8.
770 builder.AddString("0.");
771 builder.AddPadding('0', -decimal_point);
772 builder.AddString(decimal_rep);
773
774 } else {
775 // ECMA-262 section 9.8.1 step 9 and 10 combined.
776 builder.AddCharacter(decimal_rep[0]);
777 if (length != 1) {
778 builder.AddCharacter('.');
779 builder.AddString(decimal_rep + 1);
780 }
781 builder.AddCharacter('e');
782 builder.AddCharacter((decimal_point >= 0) ? '+' : '-');
783 int exponent = decimal_point - 1;
784 if (exponent < 0) exponent = -exponent;
785 builder.AddFormatted("%d", exponent);
786 }
787 return builder.Finalize();
788 }
789 }
790 }
791
792
IntToCString(int n,Vector<char> buffer)793 const char* IntToCString(int n, Vector<char> buffer) {
794 bool negative = false;
795 if (n < 0) {
796 // We must not negate the most negative int.
797 if (n == kMinInt) return DoubleToCString(n, buffer);
798 negative = true;
799 n = -n;
800 }
801 // Build the string backwards from the least significant digit.
802 int i = buffer.length();
803 buffer[--i] = '\0';
804 do {
805 buffer[--i] = '0' + (n % 10);
806 n /= 10;
807 } while (n);
808 if (negative) buffer[--i] = '-';
809 return buffer.start() + i;
810 }
811
812
DoubleToFixedCString(double value,int f)813 char* DoubleToFixedCString(double value, int f) {
814 const int kMaxDigitsBeforePoint = 21;
815 const double kFirstNonFixed = 1e21;
816 const int kMaxDigitsAfterPoint = 20;
817 ASSERT(f >= 0);
818 ASSERT(f <= kMaxDigitsAfterPoint);
819
820 bool negative = false;
821 double abs_value = value;
822 if (value < 0) {
823 abs_value = -value;
824 negative = true;
825 }
826
827 // If abs_value has more than kMaxDigitsBeforePoint digits before the point
828 // use the non-fixed conversion routine.
829 if (abs_value >= kFirstNonFixed) {
830 char arr[100];
831 Vector<char> buffer(arr, ARRAY_SIZE(arr));
832 return StrDup(DoubleToCString(value, buffer));
833 }
834
835 // Find a sufficiently precise decimal representation of n.
836 int decimal_point;
837 int sign;
838 // Add space for the '\0' byte.
839 const int kDecimalRepCapacity =
840 kMaxDigitsBeforePoint + kMaxDigitsAfterPoint + 1;
841 char decimal_rep[kDecimalRepCapacity];
842 int decimal_rep_length;
843 DoubleToAscii(value, DTOA_FIXED, f,
844 Vector<char>(decimal_rep, kDecimalRepCapacity),
845 &sign, &decimal_rep_length, &decimal_point);
846
847 // Create a representation that is padded with zeros if needed.
848 int zero_prefix_length = 0;
849 int zero_postfix_length = 0;
850
851 if (decimal_point <= 0) {
852 zero_prefix_length = -decimal_point + 1;
853 decimal_point = 1;
854 }
855
856 if (zero_prefix_length + decimal_rep_length < decimal_point + f) {
857 zero_postfix_length = decimal_point + f - decimal_rep_length -
858 zero_prefix_length;
859 }
860
861 unsigned rep_length =
862 zero_prefix_length + decimal_rep_length + zero_postfix_length;
863 StringBuilder rep_builder(rep_length + 1);
864 rep_builder.AddPadding('0', zero_prefix_length);
865 rep_builder.AddString(decimal_rep);
866 rep_builder.AddPadding('0', zero_postfix_length);
867 char* rep = rep_builder.Finalize();
868
869 // Create the result string by appending a minus and putting in a
870 // decimal point if needed.
871 unsigned result_size = decimal_point + f + 2;
872 StringBuilder builder(result_size + 1);
873 if (negative) builder.AddCharacter('-');
874 builder.AddSubstring(rep, decimal_point);
875 if (f > 0) {
876 builder.AddCharacter('.');
877 builder.AddSubstring(rep + decimal_point, f);
878 }
879 DeleteArray(rep);
880 return builder.Finalize();
881 }
882
883
CreateExponentialRepresentation(char * decimal_rep,int exponent,bool negative,int significant_digits)884 static char* CreateExponentialRepresentation(char* decimal_rep,
885 int exponent,
886 bool negative,
887 int significant_digits) {
888 bool negative_exponent = false;
889 if (exponent < 0) {
890 negative_exponent = true;
891 exponent = -exponent;
892 }
893
894 // Leave room in the result for appending a minus, for a period, the
895 // letter 'e', a minus or a plus depending on the exponent, and a
896 // three digit exponent.
897 unsigned result_size = significant_digits + 7;
898 StringBuilder builder(result_size + 1);
899
900 if (negative) builder.AddCharacter('-');
901 builder.AddCharacter(decimal_rep[0]);
902 if (significant_digits != 1) {
903 builder.AddCharacter('.');
904 builder.AddString(decimal_rep + 1);
905 int rep_length = StrLength(decimal_rep);
906 builder.AddPadding('0', significant_digits - rep_length);
907 }
908
909 builder.AddCharacter('e');
910 builder.AddCharacter(negative_exponent ? '-' : '+');
911 builder.AddFormatted("%d", exponent);
912 return builder.Finalize();
913 }
914
915
916
DoubleToExponentialCString(double value,int f)917 char* DoubleToExponentialCString(double value, int f) {
918 const int kMaxDigitsAfterPoint = 20;
919 // f might be -1 to signal that f was undefined in JavaScript.
920 ASSERT(f >= -1 && f <= kMaxDigitsAfterPoint);
921
922 bool negative = false;
923 if (value < 0) {
924 value = -value;
925 negative = true;
926 }
927
928 // Find a sufficiently precise decimal representation of n.
929 int decimal_point;
930 int sign;
931 // f corresponds to the digits after the point. There is always one digit
932 // before the point. The number of requested_digits equals hence f + 1.
933 // And we have to add one character for the null-terminator.
934 const int kV8DtoaBufferCapacity = kMaxDigitsAfterPoint + 1 + 1;
935 // Make sure that the buffer is big enough, even if we fall back to the
936 // shortest representation (which happens when f equals -1).
937 ASSERT(kBase10MaximalLength <= kMaxDigitsAfterPoint + 1);
938 char decimal_rep[kV8DtoaBufferCapacity];
939 int decimal_rep_length;
940
941 if (f == -1) {
942 DoubleToAscii(value, DTOA_SHORTEST, 0,
943 Vector<char>(decimal_rep, kV8DtoaBufferCapacity),
944 &sign, &decimal_rep_length, &decimal_point);
945 f = decimal_rep_length - 1;
946 } else {
947 DoubleToAscii(value, DTOA_PRECISION, f + 1,
948 Vector<char>(decimal_rep, kV8DtoaBufferCapacity),
949 &sign, &decimal_rep_length, &decimal_point);
950 }
951 ASSERT(decimal_rep_length > 0);
952 ASSERT(decimal_rep_length <= f + 1);
953
954 int exponent = decimal_point - 1;
955 char* result =
956 CreateExponentialRepresentation(decimal_rep, exponent, negative, f+1);
957
958 return result;
959 }
960
961
DoubleToPrecisionCString(double value,int p)962 char* DoubleToPrecisionCString(double value, int p) {
963 const int kMinimalDigits = 1;
964 const int kMaximalDigits = 21;
965 ASSERT(p >= kMinimalDigits && p <= kMaximalDigits);
966 USE(kMinimalDigits);
967
968 bool negative = false;
969 if (value < 0) {
970 value = -value;
971 negative = true;
972 }
973
974 // Find a sufficiently precise decimal representation of n.
975 int decimal_point;
976 int sign;
977 // Add one for the terminating null character.
978 const int kV8DtoaBufferCapacity = kMaximalDigits + 1;
979 char decimal_rep[kV8DtoaBufferCapacity];
980 int decimal_rep_length;
981
982 DoubleToAscii(value, DTOA_PRECISION, p,
983 Vector<char>(decimal_rep, kV8DtoaBufferCapacity),
984 &sign, &decimal_rep_length, &decimal_point);
985 ASSERT(decimal_rep_length <= p);
986
987 int exponent = decimal_point - 1;
988
989 char* result = NULL;
990
991 if (exponent < -6 || exponent >= p) {
992 result =
993 CreateExponentialRepresentation(decimal_rep, exponent, negative, p);
994 } else {
995 // Use fixed notation.
996 //
997 // Leave room in the result for appending a minus, a period and in
998 // the case where decimal_point is not positive for a zero in
999 // front of the period.
1000 unsigned result_size = (decimal_point <= 0)
1001 ? -decimal_point + p + 3
1002 : p + 2;
1003 StringBuilder builder(result_size + 1);
1004 if (negative) builder.AddCharacter('-');
1005 if (decimal_point <= 0) {
1006 builder.AddString("0.");
1007 builder.AddPadding('0', -decimal_point);
1008 builder.AddString(decimal_rep);
1009 builder.AddPadding('0', p - decimal_rep_length);
1010 } else {
1011 const int m = Min(decimal_rep_length, decimal_point);
1012 builder.AddSubstring(decimal_rep, m);
1013 builder.AddPadding('0', decimal_point - decimal_rep_length);
1014 if (decimal_point < p) {
1015 builder.AddCharacter('.');
1016 const int extra = negative ? 2 : 1;
1017 if (decimal_rep_length > decimal_point) {
1018 const int len = StrLength(decimal_rep + decimal_point);
1019 const int n = Min(len, p - (builder.position() - extra));
1020 builder.AddSubstring(decimal_rep + decimal_point, n);
1021 }
1022 builder.AddPadding('0', extra + (p - builder.position()));
1023 }
1024 }
1025 result = builder.Finalize();
1026 }
1027
1028 return result;
1029 }
1030
1031
DoubleToRadixCString(double value,int radix)1032 char* DoubleToRadixCString(double value, int radix) {
1033 ASSERT(radix >= 2 && radix <= 36);
1034
1035 // Character array used for conversion.
1036 static const char chars[] = "0123456789abcdefghijklmnopqrstuvwxyz";
1037
1038 // Buffer for the integer part of the result. 1024 chars is enough
1039 // for max integer value in radix 2. We need room for a sign too.
1040 static const int kBufferSize = 1100;
1041 char integer_buffer[kBufferSize];
1042 integer_buffer[kBufferSize - 1] = '\0';
1043
1044 // Buffer for the decimal part of the result. We only generate up
1045 // to kBufferSize - 1 chars for the decimal part.
1046 char decimal_buffer[kBufferSize];
1047 decimal_buffer[kBufferSize - 1] = '\0';
1048
1049 // Make sure the value is positive.
1050 bool is_negative = value < 0.0;
1051 if (is_negative) value = -value;
1052
1053 // Get the integer part and the decimal part.
1054 double integer_part = floor(value);
1055 double decimal_part = value - integer_part;
1056
1057 // Convert the integer part starting from the back. Always generate
1058 // at least one digit.
1059 int integer_pos = kBufferSize - 2;
1060 do {
1061 integer_buffer[integer_pos--] =
1062 chars[static_cast<int>(modulo(integer_part, radix))];
1063 integer_part /= radix;
1064 } while (integer_part >= 1.0);
1065 // Sanity check.
1066 ASSERT(integer_pos > 0);
1067 // Add sign if needed.
1068 if (is_negative) integer_buffer[integer_pos--] = '-';
1069
1070 // Convert the decimal part. Repeatedly multiply by the radix to
1071 // generate the next char. Never generate more than kBufferSize - 1
1072 // chars.
1073 //
1074 // TODO(1093998): We will often generate a full decimal_buffer of
1075 // chars because hitting zero will often not happen. The right
1076 // solution would be to continue until the string representation can
1077 // be read back and yield the original value. To implement this
1078 // efficiently, we probably have to modify dtoa.
1079 int decimal_pos = 0;
1080 while ((decimal_part > 0.0) && (decimal_pos < kBufferSize - 1)) {
1081 decimal_part *= radix;
1082 decimal_buffer[decimal_pos++] =
1083 chars[static_cast<int>(floor(decimal_part))];
1084 decimal_part -= floor(decimal_part);
1085 }
1086 decimal_buffer[decimal_pos] = '\0';
1087
1088 // Compute the result size.
1089 int integer_part_size = kBufferSize - 2 - integer_pos;
1090 // Make room for zero termination.
1091 unsigned result_size = integer_part_size + decimal_pos;
1092 // If the number has a decimal part, leave room for the period.
1093 if (decimal_pos > 0) result_size++;
1094 // Allocate result and fill in the parts.
1095 StringBuilder builder(result_size + 1);
1096 builder.AddSubstring(integer_buffer + integer_pos + 1, integer_part_size);
1097 if (decimal_pos > 0) builder.AddCharacter('.');
1098 builder.AddSubstring(decimal_buffer, decimal_pos);
1099 return builder.Finalize();
1100 }
1101
1102
1103 static Mutex* dtoa_lock_one = OS::CreateMutex();
1104 static Mutex* dtoa_lock_zero = OS::CreateMutex();
1105
1106
1107 } } // namespace v8::internal
1108
1109
1110 extern "C" {
ACQUIRE_DTOA_LOCK(int n)1111 void ACQUIRE_DTOA_LOCK(int n) {
1112 ASSERT(n == 0 || n == 1);
1113 (n == 0 ? v8::internal::dtoa_lock_zero : v8::internal::dtoa_lock_one)->Lock();
1114 }
1115
1116
FREE_DTOA_LOCK(int n)1117 void FREE_DTOA_LOCK(int n) {
1118 ASSERT(n == 0 || n == 1);
1119 (n == 0 ? v8::internal::dtoa_lock_zero : v8::internal::dtoa_lock_one)->
1120 Unlock();
1121 }
1122 }
1123