Coverage Report

Created: 2023-09-25 06:27

/src/abseil-cpp/absl/strings/numbers.cc
Line
Count
Source (jump to first uncovered line)
1
// Copyright 2017 The Abseil Authors.
2
//
3
// Licensed under the Apache License, Version 2.0 (the "License");
4
// you may not use this file except in compliance with the License.
5
// You may obtain a copy of the License at
6
//
7
//      https://www.apache.org/licenses/LICENSE-2.0
8
//
9
// Unless required by applicable law or agreed to in writing, software
10
// distributed under the License is distributed on an "AS IS" BASIS,
11
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12
// See the License for the specific language governing permissions and
13
// limitations under the License.
14
15
// This file contains string processing functions related to
16
// numeric values.
17
18
#include "absl/strings/numbers.h"
19
20
#include <algorithm>
21
#include <cassert>
22
#include <cfloat>  // for DBL_DIG and FLT_DIG
23
#include <cmath>   // for HUGE_VAL
24
#include <cstdint>
25
#include <cstdio>
26
#include <cstdlib>
27
#include <cstring>
28
#include <iterator>
29
#include <limits>
30
#include <system_error>  // NOLINT(build/c++11)
31
#include <utility>
32
33
#include "absl/base/attributes.h"
34
#include "absl/base/config.h"
35
#include "absl/base/internal/endian.h"
36
#include "absl/base/internal/raw_logging.h"
37
#include "absl/base/optimization.h"
38
#include "absl/numeric/bits.h"
39
#include "absl/numeric/int128.h"
40
#include "absl/strings/ascii.h"
41
#include "absl/strings/charconv.h"
42
#include "absl/strings/match.h"
43
#include "absl/strings/string_view.h"
44
45
namespace absl {
46
ABSL_NAMESPACE_BEGIN
47
48
0
bool SimpleAtof(absl::string_view str, float* out) {
49
0
  *out = 0.0;
50
0
  str = StripAsciiWhitespace(str);
51
  // std::from_chars doesn't accept an initial +, but SimpleAtof does, so if one
52
  // is present, skip it, while avoiding accepting "+-0" as valid.
53
0
  if (!str.empty() && str[0] == '+') {
54
0
    str.remove_prefix(1);
55
0
    if (!str.empty() && str[0] == '-') {
56
0
      return false;
57
0
    }
58
0
  }
59
0
  auto result = absl::from_chars(str.data(), str.data() + str.size(), *out);
60
0
  if (result.ec == std::errc::invalid_argument) {
61
0
    return false;
62
0
  }
63
0
  if (result.ptr != str.data() + str.size()) {
64
    // not all non-whitespace characters consumed
65
0
    return false;
66
0
  }
67
  // from_chars() with DR 3081's current wording will return max() on
68
  // overflow.  SimpleAtof returns infinity instead.
69
0
  if (result.ec == std::errc::result_out_of_range) {
70
0
    if (*out > 1.0) {
71
0
      *out = std::numeric_limits<float>::infinity();
72
0
    } else if (*out < -1.0) {
73
0
      *out = -std::numeric_limits<float>::infinity();
74
0
    }
75
0
  }
76
0
  return true;
77
0
}
78
79
11.9M
bool SimpleAtod(absl::string_view str, double* out) {
80
11.9M
  *out = 0.0;
81
11.9M
  str = StripAsciiWhitespace(str);
82
  // std::from_chars doesn't accept an initial +, but SimpleAtod does, so if one
83
  // is present, skip it, while avoiding accepting "+-0" as valid.
84
11.9M
  if (!str.empty() && str[0] == '+') {
85
824
    str.remove_prefix(1);
86
824
    if (!str.empty() && str[0] == '-') {
87
1
      return false;
88
1
    }
89
824
  }
90
11.9M
  auto result = absl::from_chars(str.data(), str.data() + str.size(), *out);
91
11.9M
  if (result.ec == std::errc::invalid_argument) {
92
2.29k
    return false;
93
2.29k
  }
94
11.9M
  if (result.ptr != str.data() + str.size()) {
95
    // not all non-whitespace characters consumed
96
534
    return false;
97
534
  }
98
  // from_chars() with DR 3081's current wording will return max() on
99
  // overflow.  SimpleAtod returns infinity instead.
100
11.9M
  if (result.ec == std::errc::result_out_of_range) {
101
5.33k
    if (*out > 1.0) {
102
3.64k
      *out = std::numeric_limits<double>::infinity();
103
3.64k
    } else if (*out < -1.0) {
104
275
      *out = -std::numeric_limits<double>::infinity();
105
275
    }
106
5.33k
  }
107
11.9M
  return true;
108
11.9M
}
109
110
0
bool SimpleAtob(absl::string_view str, bool* out) {
111
0
  ABSL_RAW_CHECK(out != nullptr, "Output pointer must not be nullptr.");
112
0
  if (EqualsIgnoreCase(str, "true") || EqualsIgnoreCase(str, "t") ||
113
0
      EqualsIgnoreCase(str, "yes") || EqualsIgnoreCase(str, "y") ||
114
0
      EqualsIgnoreCase(str, "1")) {
115
0
    *out = true;
116
0
    return true;
117
0
  }
118
0
  if (EqualsIgnoreCase(str, "false") || EqualsIgnoreCase(str, "f") ||
119
0
      EqualsIgnoreCase(str, "no") || EqualsIgnoreCase(str, "n") ||
120
0
      EqualsIgnoreCase(str, "0")) {
121
0
    *out = false;
122
0
    return true;
123
0
  }
124
0
  return false;
125
0
}
126
127
// ----------------------------------------------------------------------
128
// FastIntToBuffer() overloads
129
//
130
// Like the Fast*ToBuffer() functions above, these are intended for speed.
131
// Unlike the Fast*ToBuffer() functions, however, these functions write
132
// their output to the beginning of the buffer.  The caller is responsible
133
// for ensuring that the buffer has enough space to hold the output.
134
//
135
// Returns a pointer to the end of the string (i.e. the null character
136
// terminating the string).
137
// ----------------------------------------------------------------------
138
139
namespace {
140
141
// Various routines to encode integers to strings.
142
143
// We split data encodings into a group of 2 digits, 4 digits, 8 digits as
144
// it's easier to combine powers of two into scalar arithmetic.
145
146
// Previous implementation used a lookup table of 200 bytes for every 2 bytes
147
// and it was memory bound, any L1 cache miss would result in a much slower
148
// result. When benchmarking with a cache eviction rate of several percent,
149
// this implementation proved to be better.
150
151
// These constants represent '00', '0000' and '00000000' as ascii strings in
152
// integers. We can add these numbers if we encode to bytes from 0 to 9. as
153
// 'i' = '0' + i for 0 <= i <= 9.
154
constexpr uint32_t kTwoZeroBytes = 0x0101 * '0';
155
constexpr uint64_t kFourZeroBytes = 0x01010101 * '0';
156
constexpr uint64_t kEightZeroBytes = 0x0101010101010101ull * '0';
157
158
// * 103 / 1024 is a division by 10 for values from 0 to 99. It's also a
159
// division of a structure [k takes 2 bytes][m takes 2 bytes], then * 103 / 1024
160
// will be [k / 10][m / 10]. It allows parallel division.
161
constexpr uint64_t kDivisionBy10Mul = 103u;
162
constexpr uint64_t kDivisionBy10Div = 1 << 10;
163
164
// * 10486 / 1048576 is a division by 100 for values from 0 to 9999.
165
constexpr uint64_t kDivisionBy100Mul = 10486u;
166
constexpr uint64_t kDivisionBy100Div = 1 << 20;
167
168
// Encode functions write the ASCII output of input `n` to `out_str`.
169
3.24M
inline char* EncodeHundred(uint32_t n, char* out_str) {
170
3.24M
  int num_digits = static_cast<int>(n - 10) >> 8;
171
3.24M
  uint32_t div10 = (n * kDivisionBy10Mul) / kDivisionBy10Div;
172
3.24M
  uint32_t mod10 = n - 10u * div10;
173
3.24M
  uint32_t base = kTwoZeroBytes + div10 + (mod10 << 8);
174
3.24M
  base >>= num_digits & 8;
175
3.24M
  little_endian::Store16(out_str, static_cast<uint16_t>(base));
176
3.24M
  return out_str + 2 + num_digits;
177
3.24M
}
178
179
0
inline char* EncodeTenThousand(uint32_t n, char* out_str) {
180
  // We split lower 2 digits and upper 2 digits of n into 2 byte consecutive
181
  // blocks. 123 ->  [\0\1][\0\23]. We divide by 10 both blocks
182
  // (it's 1 division + zeroing upper bits), and compute modulo 10 as well "in
183
  // parallel". Then we combine both results to have both ASCII digits,
184
  // strip trailing zeros, add ASCII '0000' and return.
185
0
  uint32_t div100 = (n * kDivisionBy100Mul) / kDivisionBy100Div;
186
0
  uint32_t mod100 = n - 100ull * div100;
187
0
  uint32_t hundreds = (mod100 << 16) + div100;
188
0
  uint32_t tens = (hundreds * kDivisionBy10Mul) / kDivisionBy10Div;
189
0
  tens &= (0xFull << 16) | 0xFull;
190
0
  tens += (hundreds - 10ull * tens) << 8;
191
0
  ABSL_ASSUME(tens != 0);
192
  // The result can contain trailing zero bits, we need to strip them to a first
193
  // significant byte in a final representation. For example, for n = 123, we
194
  // have tens to have representation \0\1\2\3. We do `& -8` to round
195
  // to a multiple to 8 to strip zero bytes, not all zero bits.
196
  // countr_zero to help.
197
  // 0 minus 8 to make MSVC happy.
198
0
  uint32_t zeroes = static_cast<uint32_t>(absl::countr_zero(tens)) & (0 - 8u);
199
0
  tens += kFourZeroBytes;
200
0
  tens >>= zeroes;
201
0
  little_endian::Store32(out_str, tens);
202
0
  return out_str + sizeof(tens) - zeroes / 8;
203
0
}
204
205
// Helper function to produce an ASCII representation of `i`.
206
//
207
// Function returns an 8-byte integer which when summed with `kEightZeroBytes`,
208
// can be treated as a printable buffer with ascii representation of `i`,
209
// possibly with leading zeros.
210
//
211
// Example:
212
//
213
//  uint64_t buffer = PrepareEightDigits(102030) + kEightZeroBytes;
214
//  char* ascii = reinterpret_cast<char*>(&buffer);
215
//  // Note two leading zeros:
216
//  EXPECT_EQ(absl::string_view(ascii, 8), "00102030");
217
//
218
// Pre-condition: `i` must be less than 100000000.
219
12.9M
inline uint64_t PrepareEightDigits(uint32_t i) {
220
12.9M
  ABSL_ASSUME(i < 10000'0000);
221
  // Prepare 2 blocks of 4 digits "in parallel".
222
0
  uint32_t hi = i / 10000;
223
12.9M
  uint32_t lo = i % 10000;
224
12.9M
  uint64_t merged = hi | (uint64_t{lo} << 32);
225
12.9M
  uint64_t div100 = ((merged * kDivisionBy100Mul) / kDivisionBy100Div) &
226
12.9M
                    ((0x7Full << 32) | 0x7Full);
227
12.9M
  uint64_t mod100 = merged - 100ull * div100;
228
12.9M
  uint64_t hundreds = (mod100 << 16) + div100;
229
12.9M
  uint64_t tens = (hundreds * kDivisionBy10Mul) / kDivisionBy10Div;
230
12.9M
  tens &= (0xFull << 48) | (0xFull << 32) | (0xFull << 16) | 0xFull;
231
12.9M
  tens += (hundreds - 10ull * tens) << 8;
232
12.9M
  return tens;
233
12.9M
}
234
235
inline ABSL_ATTRIBUTE_ALWAYS_INLINE char* EncodeFullU32(uint32_t n,
236
12.9M
                                                        char* out_str) {
237
12.9M
  if (n < 10) {
238
37
    *out_str = static_cast<char>('0' + n);
239
37
    return out_str + 1;
240
37
  }
241
12.9M
  if (n < 100'000'000) {
242
9.72M
    uint64_t bottom = PrepareEightDigits(n);
243
9.72M
    ABSL_ASSUME(bottom != 0);
244
    // 0 minus 8 to make MSVC happy.
245
0
    uint32_t zeroes =
246
9.72M
        static_cast<uint32_t>(absl::countr_zero(bottom)) & (0 - 8u);
247
9.72M
    little_endian::Store64(out_str, (bottom + kEightZeroBytes) >> zeroes);
248
9.72M
    return out_str + sizeof(bottom) - zeroes / 8;
249
9.72M
  }
250
3.24M
  uint32_t div08 = n / 100'000'000;
251
3.24M
  uint32_t mod08 = n % 100'000'000;
252
3.24M
  uint64_t bottom = PrepareEightDigits(mod08) + kEightZeroBytes;
253
3.24M
  out_str = EncodeHundred(div08, out_str);
254
3.24M
  little_endian::Store64(out_str, bottom);
255
3.24M
  return out_str + sizeof(bottom);
256
12.9M
}
257
258
inline ABSL_ATTRIBUTE_ALWAYS_INLINE char* EncodeFullU64(uint64_t i,
259
0
                                                        char* buffer) {
260
0
  if (i <= std::numeric_limits<uint32_t>::max()) {
261
0
    return EncodeFullU32(static_cast<uint32_t>(i), buffer);
262
0
  }
263
0
  uint32_t mod08;
264
0
  if (i < 1'0000'0000'0000'0000ull) {
265
0
    uint32_t div08 = static_cast<uint32_t>(i / 100'000'000ull);
266
0
    mod08 =  static_cast<uint32_t>(i % 100'000'000ull);
267
0
    buffer = EncodeFullU32(div08, buffer);
268
0
  } else {
269
0
    uint64_t div08 = i / 100'000'000ull;
270
0
    mod08 =  static_cast<uint32_t>(i % 100'000'000ull);
271
0
    uint32_t div016 = static_cast<uint32_t>(div08 / 100'000'000ull);
272
0
    uint32_t div08mod08 = static_cast<uint32_t>(div08 % 100'000'000ull);
273
0
    uint64_t mid_result = PrepareEightDigits(div08mod08) + kEightZeroBytes;
274
0
    buffer = EncodeTenThousand(div016, buffer);
275
0
    little_endian::Store64(buffer, mid_result);
276
0
    buffer += sizeof(mid_result);
277
0
  }
278
0
  uint64_t mod_result = PrepareEightDigits(mod08) + kEightZeroBytes;
279
0
  little_endian::Store64(buffer, mod_result);
280
0
  return buffer + sizeof(mod_result);
281
0
}
282
283
}  // namespace
284
285
0
void numbers_internal::PutTwoDigits(uint32_t i, char* buf) {
286
0
  assert(i < 100);
287
0
  uint32_t base = kTwoZeroBytes;
288
0
  uint32_t div10 = (i * kDivisionBy10Mul) / kDivisionBy10Div;
289
0
  uint32_t mod10 = i - 10u * div10;
290
0
  base += div10 + (mod10 << 8);
291
0
  little_endian::Store16(buf, static_cast<uint16_t>(base));
292
0
}
293
294
0
char* numbers_internal::FastIntToBuffer(uint32_t n, char* out_str) {
295
0
  out_str = EncodeFullU32(n, out_str);
296
0
  *out_str = '\0';
297
0
  return out_str;
298
0
}
299
300
12.9M
char* numbers_internal::FastIntToBuffer(int32_t i, char* buffer) {
301
12.9M
  uint32_t u = static_cast<uint32_t>(i);
302
12.9M
  if (i < 0) {
303
0
    *buffer++ = '-';
304
    // We need to do the negation in modular (i.e., "unsigned")
305
    // arithmetic; MSVC++ apparently warns for plain "-u", so
306
    // we write the equivalent expression "0 - u" instead.
307
0
    u = 0 - u;
308
0
  }
309
12.9M
  buffer = EncodeFullU32(u, buffer);
310
12.9M
  *buffer = '\0';
311
12.9M
  return buffer;
312
12.9M
}
313
314
0
char* numbers_internal::FastIntToBuffer(uint64_t i, char* buffer) {
315
0
  buffer = EncodeFullU64(i, buffer);
316
0
  *buffer = '\0';
317
0
  return buffer;
318
0
}
319
320
0
char* numbers_internal::FastIntToBuffer(int64_t i, char* buffer) {
321
0
  uint64_t u = static_cast<uint64_t>(i);
322
0
  if (i < 0) {
323
0
    *buffer++ = '-';
324
    // We need to do the negation in modular (i.e., "unsigned")
325
    // arithmetic; MSVC++ apparently warns for plain "-u", so
326
    // we write the equivalent expression "0 - u" instead.
327
0
    u = 0 - u;
328
0
  }
329
0
  buffer = EncodeFullU64(u, buffer);
330
0
  *buffer = '\0';
331
0
  return buffer;
332
0
}
333
334
// Given a 128-bit number expressed as a pair of uint64_t, high half first,
335
// return that number multiplied by the given 32-bit value.  If the result is
336
// too large to fit in a 128-bit number, divide it by 2 until it fits.
337
static std::pair<uint64_t, uint64_t> Mul32(std::pair<uint64_t, uint64_t> num,
338
0
                                           uint32_t mul) {
339
0
  uint64_t bits0_31 = num.second & 0xFFFFFFFF;
340
0
  uint64_t bits32_63 = num.second >> 32;
341
0
  uint64_t bits64_95 = num.first & 0xFFFFFFFF;
342
0
  uint64_t bits96_127 = num.first >> 32;
343
344
  // The picture so far: each of these 64-bit values has only the lower 32 bits
345
  // filled in.
346
  // bits96_127:          [ 00000000 xxxxxxxx ]
347
  // bits64_95:                    [ 00000000 xxxxxxxx ]
348
  // bits32_63:                             [ 00000000 xxxxxxxx ]
349
  // bits0_31:                                       [ 00000000 xxxxxxxx ]
350
351
0
  bits0_31 *= mul;
352
0
  bits32_63 *= mul;
353
0
  bits64_95 *= mul;
354
0
  bits96_127 *= mul;
355
356
  // Now the top halves may also have value, though all 64 of their bits will
357
  // never be set at the same time, since they are a result of a 32x32 bit
358
  // multiply.  This makes the carry calculation slightly easier.
359
  // bits96_127:          [ mmmmmmmm | mmmmmmmm ]
360
  // bits64_95:                    [ | mmmmmmmm mmmmmmmm | ]
361
  // bits32_63:                      |        [ mmmmmmmm | mmmmmmmm ]
362
  // bits0_31:                       |                 [ | mmmmmmmm mmmmmmmm ]
363
  // eventually:        [ bits128_up | ...bits64_127.... | ..bits0_63... ]
364
365
0
  uint64_t bits0_63 = bits0_31 + (bits32_63 << 32);
366
0
  uint64_t bits64_127 = bits64_95 + (bits96_127 << 32) + (bits32_63 >> 32) +
367
0
                        (bits0_63 < bits0_31);
368
0
  uint64_t bits128_up = (bits96_127 >> 32) + (bits64_127 < bits64_95);
369
0
  if (bits128_up == 0) return {bits64_127, bits0_63};
370
371
0
  auto shift = static_cast<unsigned>(bit_width(bits128_up));
372
0
  uint64_t lo = (bits0_63 >> shift) + (bits64_127 << (64 - shift));
373
0
  uint64_t hi = (bits64_127 >> shift) + (bits128_up << (64 - shift));
374
0
  return {hi, lo};
375
0
}
376
377
// Compute num * 5 ^ expfive, and return the first 128 bits of the result,
378
// where the first bit is always a one.  So PowFive(1, 0) starts 0b100000,
379
// PowFive(1, 1) starts 0b101000, PowFive(1, 2) starts 0b110010, etc.
380
0
static std::pair<uint64_t, uint64_t> PowFive(uint64_t num, int expfive) {
381
0
  std::pair<uint64_t, uint64_t> result = {num, 0};
382
0
  while (expfive >= 13) {
383
    // 5^13 is the highest power of five that will fit in a 32-bit integer.
384
0
    result = Mul32(result, 5 * 5 * 5 * 5 * 5 * 5 * 5 * 5 * 5 * 5 * 5 * 5 * 5);
385
0
    expfive -= 13;
386
0
  }
387
0
  constexpr uint32_t powers_of_five[13] = {
388
0
      1,
389
0
      5,
390
0
      5 * 5,
391
0
      5 * 5 * 5,
392
0
      5 * 5 * 5 * 5,
393
0
      5 * 5 * 5 * 5 * 5,
394
0
      5 * 5 * 5 * 5 * 5 * 5,
395
0
      5 * 5 * 5 * 5 * 5 * 5 * 5,
396
0
      5 * 5 * 5 * 5 * 5 * 5 * 5 * 5,
397
0
      5 * 5 * 5 * 5 * 5 * 5 * 5 * 5 * 5,
398
0
      5 * 5 * 5 * 5 * 5 * 5 * 5 * 5 * 5 * 5,
399
0
      5 * 5 * 5 * 5 * 5 * 5 * 5 * 5 * 5 * 5 * 5,
400
0
      5 * 5 * 5 * 5 * 5 * 5 * 5 * 5 * 5 * 5 * 5 * 5};
401
0
  result = Mul32(result, powers_of_five[expfive & 15]);
402
0
  int shift = countl_zero(result.first);
403
0
  if (shift != 0) {
404
0
    result.first = (result.first << shift) + (result.second >> (64 - shift));
405
0
    result.second = (result.second << shift);
406
0
  }
407
0
  return result;
408
0
}
409
410
struct ExpDigits {
411
  int32_t exponent;
412
  char digits[6];
413
};
414
415
// SplitToSix converts value, a positive double-precision floating-point number,
416
// into a base-10 exponent and 6 ASCII digits, where the first digit is never
417
// zero.  For example, SplitToSix(1) returns an exponent of zero and a digits
418
// array of {'1', '0', '0', '0', '0', '0'}.  If value is exactly halfway between
419
// two possible representations, e.g. value = 100000.5, then "round to even" is
420
// performed.
421
0
static ExpDigits SplitToSix(const double value) {
422
0
  ExpDigits exp_dig;
423
0
  int exp = 5;
424
0
  double d = value;
425
  // First step: calculate a close approximation of the output, where the
426
  // value d will be between 100,000 and 999,999, representing the digits
427
  // in the output ASCII array, and exp is the base-10 exponent.  It would be
428
  // faster to use a table here, and to look up the base-2 exponent of value,
429
  // however value is an IEEE-754 64-bit number, so the table would have 2,000
430
  // entries, which is not cache-friendly.
431
0
  if (d >= 999999.5) {
432
0
    if (d >= 1e+261) exp += 256, d *= 1e-256;
433
0
    if (d >= 1e+133) exp += 128, d *= 1e-128;
434
0
    if (d >= 1e+69) exp += 64, d *= 1e-64;
435
0
    if (d >= 1e+37) exp += 32, d *= 1e-32;
436
0
    if (d >= 1e+21) exp += 16, d *= 1e-16;
437
0
    if (d >= 1e+13) exp += 8, d *= 1e-8;
438
0
    if (d >= 1e+9) exp += 4, d *= 1e-4;
439
0
    if (d >= 1e+7) exp += 2, d *= 1e-2;
440
0
    if (d >= 1e+6) exp += 1, d *= 1e-1;
441
0
  } else {
442
0
    if (d < 1e-250) exp -= 256, d *= 1e256;
443
0
    if (d < 1e-122) exp -= 128, d *= 1e128;
444
0
    if (d < 1e-58) exp -= 64, d *= 1e64;
445
0
    if (d < 1e-26) exp -= 32, d *= 1e32;
446
0
    if (d < 1e-10) exp -= 16, d *= 1e16;
447
0
    if (d < 1e-2) exp -= 8, d *= 1e8;
448
0
    if (d < 1e+2) exp -= 4, d *= 1e4;
449
0
    if (d < 1e+4) exp -= 2, d *= 1e2;
450
0
    if (d < 1e+5) exp -= 1, d *= 1e1;
451
0
  }
452
  // At this point, d is in the range [99999.5..999999.5) and exp is in the
453
  // range [-324..308]. Since we need to round d up, we want to add a half
454
  // and truncate.
455
  // However, the technique above may have lost some precision, due to its
456
  // repeated multiplication by constants that each may be off by half a bit
457
  // of precision.  This only matters if we're close to the edge though.
458
  // Since we'd like to know if the fractional part of d is close to a half,
459
  // we multiply it by 65536 and see if the fractional part is close to 32768.
460
  // (The number doesn't have to be a power of two,but powers of two are faster)
461
0
  uint64_t d64k = d * 65536;
462
0
  uint32_t dddddd;  // A 6-digit decimal integer.
463
0
  if ((d64k % 65536) == 32767 || (d64k % 65536) == 32768) {
464
    // OK, it's fairly likely that precision was lost above, which is
465
    // not a surprise given only 52 mantissa bits are available.  Therefore
466
    // redo the calculation using 128-bit numbers.  (64 bits are not enough).
467
468
    // Start out with digits rounded down; maybe add one below.
469
0
    dddddd = static_cast<uint32_t>(d64k / 65536);
470
471
    // mantissa is a 64-bit integer representing M.mmm... * 2^63.  The actual
472
    // value we're representing, of course, is M.mmm... * 2^exp2.
473
0
    int exp2;
474
0
    double m = std::frexp(value, &exp2);
475
0
    uint64_t mantissa = m * (32768.0 * 65536.0 * 65536.0 * 65536.0);
476
    // std::frexp returns an m value in the range [0.5, 1.0), however we
477
    // can't multiply it by 2^64 and convert to an integer because some FPUs
478
    // throw an exception when converting an number higher than 2^63 into an
479
    // integer - even an unsigned 64-bit integer!  Fortunately it doesn't matter
480
    // since m only has 52 significant bits anyway.
481
0
    mantissa <<= 1;
482
0
    exp2 -= 64;  // not needed, but nice for debugging
483
484
    // OK, we are here to compare:
485
    //     (dddddd + 0.5) * 10^(exp-5)  vs.  mantissa * 2^exp2
486
    // so we can round up dddddd if appropriate.  Those values span the full
487
    // range of 600 orders of magnitude of IEE 64-bit floating-point.
488
    // Fortunately, we already know they are very close, so we don't need to
489
    // track the base-2 exponent of both sides.  This greatly simplifies the
490
    // the math since the 2^exp2 calculation is unnecessary and the power-of-10
491
    // calculation can become a power-of-5 instead.
492
493
0
    std::pair<uint64_t, uint64_t> edge, val;
494
0
    if (exp >= 6) {
495
      // Compare (dddddd + 0.5) * 5 ^ (exp - 5) to mantissa
496
      // Since we're tossing powers of two, 2 * dddddd + 1 is the
497
      // same as dddddd + 0.5
498
0
      edge = PowFive(2 * dddddd + 1, exp - 5);
499
500
0
      val.first = mantissa;
501
0
      val.second = 0;
502
0
    } else {
503
      // We can't compare (dddddd + 0.5) * 5 ^ (exp - 5) to mantissa as we did
504
      // above because (exp - 5) is negative.  So we compare (dddddd + 0.5) to
505
      // mantissa * 5 ^ (5 - exp)
506
0
      edge = PowFive(2 * dddddd + 1, 0);
507
508
0
      val = PowFive(mantissa, 5 - exp);
509
0
    }
510
    // printf("exp=%d %016lx %016lx vs %016lx %016lx\n", exp, val.first,
511
    //        val.second, edge.first, edge.second);
512
0
    if (val > edge) {
513
0
      dddddd++;
514
0
    } else if (val == edge) {
515
0
      dddddd += (dddddd & 1);
516
0
    }
517
0
  } else {
518
    // Here, we are not close to the edge.
519
0
    dddddd = static_cast<uint32_t>((d64k + 32768) / 65536);
520
0
  }
521
0
  if (dddddd == 1000000) {
522
0
    dddddd = 100000;
523
0
    exp += 1;
524
0
  }
525
0
  exp_dig.exponent = exp;
526
527
0
  uint32_t two_digits = dddddd / 10000;
528
0
  dddddd -= two_digits * 10000;
529
0
  numbers_internal::PutTwoDigits(two_digits, &exp_dig.digits[0]);
530
531
0
  two_digits = dddddd / 100;
532
0
  dddddd -= two_digits * 100;
533
0
  numbers_internal::PutTwoDigits(two_digits, &exp_dig.digits[2]);
534
535
0
  numbers_internal::PutTwoDigits(dddddd, &exp_dig.digits[4]);
536
0
  return exp_dig;
537
0
}
538
539
// Helper function for fast formatting of floating-point.
540
// The result is the same as "%g", a.k.a. "%.6g".
541
0
size_t numbers_internal::SixDigitsToBuffer(double d, char* const buffer) {
542
0
  static_assert(std::numeric_limits<float>::is_iec559,
543
0
                "IEEE-754/IEC-559 support only");
544
545
0
  char* out = buffer;  // we write data to out, incrementing as we go, but
546
                       // FloatToBuffer always returns the address of the buffer
547
                       // passed in.
548
549
0
  if (std::isnan(d)) {
550
0
    strcpy(out, "nan");  // NOLINT(runtime/printf)
551
0
    return 3;
552
0
  }
553
0
  if (d == 0) {  // +0 and -0 are handled here
554
0
    if (std::signbit(d)) *out++ = '-';
555
0
    *out++ = '0';
556
0
    *out = 0;
557
0
    return static_cast<size_t>(out - buffer);
558
0
  }
559
0
  if (d < 0) {
560
0
    *out++ = '-';
561
0
    d = -d;
562
0
  }
563
0
  if (d > std::numeric_limits<double>::max()) {
564
0
    strcpy(out, "inf");  // NOLINT(runtime/printf)
565
0
    return static_cast<size_t>(out + 3 - buffer);
566
0
  }
567
568
0
  auto exp_dig = SplitToSix(d);
569
0
  int exp = exp_dig.exponent;
570
0
  const char* digits = exp_dig.digits;
571
0
  out[0] = '0';
572
0
  out[1] = '.';
573
0
  switch (exp) {
574
0
    case 5:
575
0
      memcpy(out, &digits[0], 6), out += 6;
576
0
      *out = 0;
577
0
      return static_cast<size_t>(out - buffer);
578
0
    case 4:
579
0
      memcpy(out, &digits[0], 5), out += 5;
580
0
      if (digits[5] != '0') {
581
0
        *out++ = '.';
582
0
        *out++ = digits[5];
583
0
      }
584
0
      *out = 0;
585
0
      return static_cast<size_t>(out - buffer);
586
0
    case 3:
587
0
      memcpy(out, &digits[0], 4), out += 4;
588
0
      if ((digits[5] | digits[4]) != '0') {
589
0
        *out++ = '.';
590
0
        *out++ = digits[4];
591
0
        if (digits[5] != '0') *out++ = digits[5];
592
0
      }
593
0
      *out = 0;
594
0
      return static_cast<size_t>(out - buffer);
595
0
    case 2:
596
0
      memcpy(out, &digits[0], 3), out += 3;
597
0
      *out++ = '.';
598
0
      memcpy(out, &digits[3], 3);
599
0
      out += 3;
600
0
      while (out[-1] == '0') --out;
601
0
      if (out[-1] == '.') --out;
602
0
      *out = 0;
603
0
      return static_cast<size_t>(out - buffer);
604
0
    case 1:
605
0
      memcpy(out, &digits[0], 2), out += 2;
606
0
      *out++ = '.';
607
0
      memcpy(out, &digits[2], 4);
608
0
      out += 4;
609
0
      while (out[-1] == '0') --out;
610
0
      if (out[-1] == '.') --out;
611
0
      *out = 0;
612
0
      return static_cast<size_t>(out - buffer);
613
0
    case 0:
614
0
      memcpy(out, &digits[0], 1), out += 1;
615
0
      *out++ = '.';
616
0
      memcpy(out, &digits[1], 5);
617
0
      out += 5;
618
0
      while (out[-1] == '0') --out;
619
0
      if (out[-1] == '.') --out;
620
0
      *out = 0;
621
0
      return static_cast<size_t>(out - buffer);
622
0
    case -4:
623
0
      out[2] = '0';
624
0
      ++out;
625
0
      ABSL_FALLTHROUGH_INTENDED;
626
0
    case -3:
627
0
      out[2] = '0';
628
0
      ++out;
629
0
      ABSL_FALLTHROUGH_INTENDED;
630
0
    case -2:
631
0
      out[2] = '0';
632
0
      ++out;
633
0
      ABSL_FALLTHROUGH_INTENDED;
634
0
    case -1:
635
0
      out += 2;
636
0
      memcpy(out, &digits[0], 6);
637
0
      out += 6;
638
0
      while (out[-1] == '0') --out;
639
0
      *out = 0;
640
0
      return static_cast<size_t>(out - buffer);
641
0
  }
642
0
  assert(exp < -4 || exp >= 6);
643
0
  out[0] = digits[0];
644
0
  assert(out[1] == '.');
645
0
  out += 2;
646
0
  memcpy(out, &digits[1], 5), out += 5;
647
0
  while (out[-1] == '0') --out;
648
0
  if (out[-1] == '.') --out;
649
0
  *out++ = 'e';
650
0
  if (exp > 0) {
651
0
    *out++ = '+';
652
0
  } else {
653
0
    *out++ = '-';
654
0
    exp = -exp;
655
0
  }
656
0
  if (exp > 99) {
657
0
    int dig1 = exp / 100;
658
0
    exp -= dig1 * 100;
659
0
    *out++ = '0' + static_cast<char>(dig1);
660
0
  }
661
0
  PutTwoDigits(static_cast<uint32_t>(exp), out);
662
0
  out += 2;
663
0
  *out = 0;
664
0
  return static_cast<size_t>(out - buffer);
665
0
}
666
667
namespace {
668
// Represents integer values of digits.
669
// Uses 36 to indicate an invalid character since we support
670
// bases up to 36.
671
static const int8_t kAsciiToInt[256] = {
672
    36, 36, 36, 36, 36, 36, 36, 36, 36, 36, 36, 36, 36, 36, 36, 36,  // 16 36s.
673
    36, 36, 36, 36, 36, 36, 36, 36, 36, 36, 36, 36, 36, 36, 36, 36, 36, 36, 36,
674
    36, 36, 36, 36, 36, 36, 36, 36, 36, 36, 36, 36, 36, 0,  1,  2,  3,  4,  5,
675
    6,  7,  8,  9,  36, 36, 36, 36, 36, 36, 36, 10, 11, 12, 13, 14, 15, 16, 17,
676
    18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36,
677
    36, 36, 36, 36, 36, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23,
678
    24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 36, 36, 36, 36, 36, 36,
679
    36, 36, 36, 36, 36, 36, 36, 36, 36, 36, 36, 36, 36, 36, 36, 36, 36, 36, 36,
680
    36, 36, 36, 36, 36, 36, 36, 36, 36, 36, 36, 36, 36, 36, 36, 36, 36, 36, 36,
681
    36, 36, 36, 36, 36, 36, 36, 36, 36, 36, 36, 36, 36, 36, 36, 36, 36, 36, 36,
682
    36, 36, 36, 36, 36, 36, 36, 36, 36, 36, 36, 36, 36, 36, 36, 36, 36, 36, 36,
683
    36, 36, 36, 36, 36, 36, 36, 36, 36, 36, 36, 36, 36, 36, 36, 36, 36, 36, 36,
684
    36, 36, 36, 36, 36, 36, 36, 36, 36, 36, 36, 36, 36, 36, 36, 36, 36, 36, 36,
685
    36, 36, 36, 36, 36, 36, 36, 36, 36, 36, 36, 36};
686
687
// Parse the sign and optional hex or oct prefix in text.
688
inline bool safe_parse_sign_and_base(absl::string_view* text /*inout*/,
689
                                     int* base_ptr /*inout*/,
690
0
                                     bool* negative_ptr /*output*/) {
691
0
  if (text->data() == nullptr) {
692
0
    return false;
693
0
  }
694
695
0
  const char* start = text->data();
696
0
  const char* end = start + text->size();
697
0
  int base = *base_ptr;
698
699
  // Consume whitespace.
700
0
  while (start < end &&
701
0
         absl::ascii_isspace(static_cast<unsigned char>(start[0]))) {
702
0
    ++start;
703
0
  }
704
0
  while (start < end &&
705
0
         absl::ascii_isspace(static_cast<unsigned char>(end[-1]))) {
706
0
    --end;
707
0
  }
708
0
  if (start >= end) {
709
0
    return false;
710
0
  }
711
712
  // Consume sign.
713
0
  *negative_ptr = (start[0] == '-');
714
0
  if (*negative_ptr || start[0] == '+') {
715
0
    ++start;
716
0
    if (start >= end) {
717
0
      return false;
718
0
    }
719
0
  }
720
721
  // Consume base-dependent prefix.
722
  //  base 0: "0x" -> base 16, "0" -> base 8, default -> base 10
723
  //  base 16: "0x" -> base 16
724
  // Also validate the base.
725
0
  if (base == 0) {
726
0
    if (end - start >= 2 && start[0] == '0' &&
727
0
        (start[1] == 'x' || start[1] == 'X')) {
728
0
      base = 16;
729
0
      start += 2;
730
0
      if (start >= end) {
731
        // "0x" with no digits after is invalid.
732
0
        return false;
733
0
      }
734
0
    } else if (end - start >= 1 && start[0] == '0') {
735
0
      base = 8;
736
0
      start += 1;
737
0
    } else {
738
0
      base = 10;
739
0
    }
740
0
  } else if (base == 16) {
741
0
    if (end - start >= 2 && start[0] == '0' &&
742
0
        (start[1] == 'x' || start[1] == 'X')) {
743
0
      start += 2;
744
0
      if (start >= end) {
745
        // "0x" with no digits after is invalid.
746
0
        return false;
747
0
      }
748
0
    }
749
0
  } else if (base >= 2 && base <= 36) {
750
    // okay
751
0
  } else {
752
0
    return false;
753
0
  }
754
0
  *text = absl::string_view(start, static_cast<size_t>(end - start));
755
0
  *base_ptr = base;
756
0
  return true;
757
0
}
758
759
// Consume digits.
760
//
761
// The classic loop:
762
//
763
//   for each digit
764
//     value = value * base + digit
765
//   value *= sign
766
//
767
// The classic loop needs overflow checking.  It also fails on the most
768
// negative integer, -2147483648 in 32-bit two's complement representation.
769
//
770
// My improved loop:
771
//
772
//  if (!negative)
773
//    for each digit
774
//      value = value * base
775
//      value = value + digit
776
//  else
777
//    for each digit
778
//      value = value * base
779
//      value = value - digit
780
//
781
// Overflow checking becomes simple.
782
783
// Lookup tables per IntType:
784
// vmax/base and vmin/base are precomputed because division costs at least 8ns.
785
// TODO(junyer): Doing this per base instead (i.e. an array of structs, not a
786
// struct of arrays) would probably be better in terms of d-cache for the most
787
// commonly used bases.
788
template <typename IntType>
789
struct LookupTables {
790
  ABSL_CONST_INIT static const IntType kVmaxOverBase[];
791
  ABSL_CONST_INIT static const IntType kVminOverBase[];
792
};
793
794
// An array initializer macro for X/base where base in [0, 36].
795
// However, note that lookups for base in [0, 1] should never happen because
796
// base has been validated to be in [2, 36] by safe_parse_sign_and_base().
797
#define X_OVER_BASE_INITIALIZER(X)                                        \
798
  {                                                                       \
799
    0, 0, X / 2, X / 3, X / 4, X / 5, X / 6, X / 7, X / 8, X / 9, X / 10, \
800
        X / 11, X / 12, X / 13, X / 14, X / 15, X / 16, X / 17, X / 18,   \
801
        X / 19, X / 20, X / 21, X / 22, X / 23, X / 24, X / 25, X / 26,   \
802
        X / 27, X / 28, X / 29, X / 30, X / 31, X / 32, X / 33, X / 34,   \
803
        X / 35, X / 36,                                                   \
804
  }
805
806
// This kVmaxOverBase is generated with
807
//  for (int base = 2; base < 37; ++base) {
808
//    absl::uint128 max = std::numeric_limits<absl::uint128>::max();
809
//    auto result = max / base;
810
//    std::cout << "    MakeUint128(" << absl::Uint128High64(result) << "u, "
811
//              << absl::Uint128Low64(result) << "u),\n";
812
//  }
813
// See https://godbolt.org/z/aneYsb
814
//
815
// uint128& operator/=(uint128) is not constexpr, so hardcode the resulting
816
// array to avoid a static initializer.
817
template <>
818
ABSL_CONST_INIT const uint128 LookupTables<uint128>::kVmaxOverBase[] = {
819
    0,
820
    0,
821
    MakeUint128(9223372036854775807u, 18446744073709551615u),
822
    MakeUint128(6148914691236517205u, 6148914691236517205u),
823
    MakeUint128(4611686018427387903u, 18446744073709551615u),
824
    MakeUint128(3689348814741910323u, 3689348814741910323u),
825
    MakeUint128(3074457345618258602u, 12297829382473034410u),
826
    MakeUint128(2635249153387078802u, 5270498306774157604u),
827
    MakeUint128(2305843009213693951u, 18446744073709551615u),
828
    MakeUint128(2049638230412172401u, 14347467612885206812u),
829
    MakeUint128(1844674407370955161u, 11068046444225730969u),
830
    MakeUint128(1676976733973595601u, 8384883669867978007u),
831
    MakeUint128(1537228672809129301u, 6148914691236517205u),
832
    MakeUint128(1418980313362273201u, 4256940940086819603u),
833
    MakeUint128(1317624576693539401u, 2635249153387078802u),
834
    MakeUint128(1229782938247303441u, 1229782938247303441u),
835
    MakeUint128(1152921504606846975u, 18446744073709551615u),
836
    MakeUint128(1085102592571150095u, 1085102592571150095u),
837
    MakeUint128(1024819115206086200u, 16397105843297379214u),
838
    MakeUint128(970881267037344821u, 16504981539634861972u),
839
    MakeUint128(922337203685477580u, 14757395258967641292u),
840
    MakeUint128(878416384462359600u, 14054662151397753612u),
841
    MakeUint128(838488366986797800u, 13415813871788764811u),
842
    MakeUint128(802032351030850070u, 4812194106185100421u),
843
    MakeUint128(768614336404564650u, 12297829382473034410u),
844
    MakeUint128(737869762948382064u, 11805916207174113034u),
845
    MakeUint128(709490156681136600u, 11351842506898185609u),
846
    MakeUint128(683212743470724133u, 17080318586768103348u),
847
    MakeUint128(658812288346769700u, 10540996613548315209u),
848
    MakeUint128(636094623231363848u, 15266270957552732371u),
849
    MakeUint128(614891469123651720u, 9838263505978427528u),
850
    MakeUint128(595056260442243600u, 9520900167075897608u),
851
    MakeUint128(576460752303423487u, 18446744073709551615u),
852
    MakeUint128(558992244657865200u, 8943875914525843207u),
853
    MakeUint128(542551296285575047u, 9765923333140350855u),
854
    MakeUint128(527049830677415760u, 8432797290838652167u),
855
    MakeUint128(512409557603043100u, 8198552921648689607u),
856
};
857
858
// This kVmaxOverBase generated with
859
//   for (int base = 2; base < 37; ++base) {
860
//    absl::int128 max = std::numeric_limits<absl::int128>::max();
861
//    auto result = max / base;
862
//    std::cout << "\tMakeInt128(" << absl::Int128High64(result) << ", "
863
//              << absl::Int128Low64(result) << "u),\n";
864
//  }
865
// See https://godbolt.org/z/7djYWz
866
//
867
// int128& operator/=(int128) is not constexpr, so hardcode the resulting array
868
// to avoid a static initializer.
869
template <>
870
ABSL_CONST_INIT const int128 LookupTables<int128>::kVmaxOverBase[] = {
871
    0,
872
    0,
873
    MakeInt128(4611686018427387903, 18446744073709551615u),
874
    MakeInt128(3074457345618258602, 12297829382473034410u),
875
    MakeInt128(2305843009213693951, 18446744073709551615u),
876
    MakeInt128(1844674407370955161, 11068046444225730969u),
877
    MakeInt128(1537228672809129301, 6148914691236517205u),
878
    MakeInt128(1317624576693539401, 2635249153387078802u),
879
    MakeInt128(1152921504606846975, 18446744073709551615u),
880
    MakeInt128(1024819115206086200, 16397105843297379214u),
881
    MakeInt128(922337203685477580, 14757395258967641292u),
882
    MakeInt128(838488366986797800, 13415813871788764811u),
883
    MakeInt128(768614336404564650, 12297829382473034410u),
884
    MakeInt128(709490156681136600, 11351842506898185609u),
885
    MakeInt128(658812288346769700, 10540996613548315209u),
886
    MakeInt128(614891469123651720, 9838263505978427528u),
887
    MakeInt128(576460752303423487, 18446744073709551615u),
888
    MakeInt128(542551296285575047, 9765923333140350855u),
889
    MakeInt128(512409557603043100, 8198552921648689607u),
890
    MakeInt128(485440633518672410, 17475862806672206794u),
891
    MakeInt128(461168601842738790, 7378697629483820646u),
892
    MakeInt128(439208192231179800, 7027331075698876806u),
893
    MakeInt128(419244183493398900, 6707906935894382405u),
894
    MakeInt128(401016175515425035, 2406097053092550210u),
895
    MakeInt128(384307168202282325, 6148914691236517205u),
896
    MakeInt128(368934881474191032, 5902958103587056517u),
897
    MakeInt128(354745078340568300, 5675921253449092804u),
898
    MakeInt128(341606371735362066, 17763531330238827482u),
899
    MakeInt128(329406144173384850, 5270498306774157604u),
900
    MakeInt128(318047311615681924, 7633135478776366185u),
901
    MakeInt128(307445734561825860, 4919131752989213764u),
902
    MakeInt128(297528130221121800, 4760450083537948804u),
903
    MakeInt128(288230376151711743, 18446744073709551615u),
904
    MakeInt128(279496122328932600, 4471937957262921603u),
905
    MakeInt128(271275648142787523, 14106333703424951235u),
906
    MakeInt128(263524915338707880, 4216398645419326083u),
907
    MakeInt128(256204778801521550, 4099276460824344803u),
908
};
909
910
// This kVminOverBase generated with
911
//  for (int base = 2; base < 37; ++base) {
912
//    absl::int128 min = std::numeric_limits<absl::int128>::min();
913
//    auto result = min / base;
914
//    std::cout << "\tMakeInt128(" << absl::Int128High64(result) << ", "
915
//              << absl::Int128Low64(result) << "u),\n";
916
//  }
917
//
918
// See https://godbolt.org/z/7djYWz
919
//
920
// int128& operator/=(int128) is not constexpr, so hardcode the resulting array
921
// to avoid a static initializer.
922
template <>
923
ABSL_CONST_INIT const int128 LookupTables<int128>::kVminOverBase[] = {
924
    0,
925
    0,
926
    MakeInt128(-4611686018427387904, 0u),
927
    MakeInt128(-3074457345618258603, 6148914691236517206u),
928
    MakeInt128(-2305843009213693952, 0u),
929
    MakeInt128(-1844674407370955162, 7378697629483820647u),
930
    MakeInt128(-1537228672809129302, 12297829382473034411u),
931
    MakeInt128(-1317624576693539402, 15811494920322472814u),
932
    MakeInt128(-1152921504606846976, 0u),
933
    MakeInt128(-1024819115206086201, 2049638230412172402u),
934
    MakeInt128(-922337203685477581, 3689348814741910324u),
935
    MakeInt128(-838488366986797801, 5030930201920786805u),
936
    MakeInt128(-768614336404564651, 6148914691236517206u),
937
    MakeInt128(-709490156681136601, 7094901566811366007u),
938
    MakeInt128(-658812288346769701, 7905747460161236407u),
939
    MakeInt128(-614891469123651721, 8608480567731124088u),
940
    MakeInt128(-576460752303423488, 0u),
941
    MakeInt128(-542551296285575048, 8680820740569200761u),
942
    MakeInt128(-512409557603043101, 10248191152060862009u),
943
    MakeInt128(-485440633518672411, 970881267037344822u),
944
    MakeInt128(-461168601842738791, 11068046444225730970u),
945
    MakeInt128(-439208192231179801, 11419412998010674810u),
946
    MakeInt128(-419244183493398901, 11738837137815169211u),
947
    MakeInt128(-401016175515425036, 16040647020617001406u),
948
    MakeInt128(-384307168202282326, 12297829382473034411u),
949
    MakeInt128(-368934881474191033, 12543785970122495099u),
950
    MakeInt128(-354745078340568301, 12770822820260458812u),
951
    MakeInt128(-341606371735362067, 683212743470724134u),
952
    MakeInt128(-329406144173384851, 13176245766935394012u),
953
    MakeInt128(-318047311615681925, 10813608594933185431u),
954
    MakeInt128(-307445734561825861, 13527612320720337852u),
955
    MakeInt128(-297528130221121801, 13686293990171602812u),
956
    MakeInt128(-288230376151711744, 0u),
957
    MakeInt128(-279496122328932601, 13974806116446630013u),
958
    MakeInt128(-271275648142787524, 4340410370284600381u),
959
    MakeInt128(-263524915338707881, 14230345428290225533u),
960
    MakeInt128(-256204778801521551, 14347467612885206813u),
961
};
962
963
template <typename IntType>
964
ABSL_CONST_INIT const IntType LookupTables<IntType>::kVmaxOverBase[] =
965
    X_OVER_BASE_INITIALIZER(std::numeric_limits<IntType>::max());
966
967
template <typename IntType>
968
ABSL_CONST_INIT const IntType LookupTables<IntType>::kVminOverBase[] =
969
    X_OVER_BASE_INITIALIZER(std::numeric_limits<IntType>::min());
970
971
#undef X_OVER_BASE_INITIALIZER
972
973
template <typename IntType>
974
inline bool safe_parse_positive_int(absl::string_view text, int base,
975
0
                                    IntType* value_p) {
976
0
  IntType value = 0;
977
0
  const IntType vmax = std::numeric_limits<IntType>::max();
978
0
  assert(vmax > 0);
979
0
  assert(base >= 0);
980
0
  const IntType base_inttype = static_cast<IntType>(base);
981
0
  assert(vmax >= base_inttype);
982
0
  const IntType vmax_over_base = LookupTables<IntType>::kVmaxOverBase[base];
983
0
  assert(base < 2 ||
984
0
         std::numeric_limits<IntType>::max() / base_inttype == vmax_over_base);
985
0
  const char* start = text.data();
986
0
  const char* end = start + text.size();
987
  // loop over digits
988
0
  for (; start < end; ++start) {
989
0
    unsigned char c = static_cast<unsigned char>(start[0]);
990
0
    IntType digit = static_cast<IntType>(kAsciiToInt[c]);
991
0
    if (digit >= base_inttype) {
992
0
      *value_p = value;
993
0
      return false;
994
0
    }
995
0
    if (value > vmax_over_base) {
996
0
      *value_p = vmax;
997
0
      return false;
998
0
    }
999
0
    value *= base_inttype;
1000
0
    if (value > vmax - digit) {
1001
0
      *value_p = vmax;
1002
0
      return false;
1003
0
    }
1004
0
    value += digit;
1005
0
  }
1006
0
  *value_p = value;
1007
0
  return true;
1008
0
}
Unexecuted instantiation: numbers.cc:bool absl::(anonymous namespace)::safe_parse_positive_int<int>(absl::string_view, int, int*)
Unexecuted instantiation: numbers.cc:bool absl::(anonymous namespace)::safe_parse_positive_int<long>(absl::string_view, int, long*)
Unexecuted instantiation: numbers.cc:bool absl::(anonymous namespace)::safe_parse_positive_int<absl::int128>(absl::string_view, int, absl::int128*)
Unexecuted instantiation: numbers.cc:bool absl::(anonymous namespace)::safe_parse_positive_int<unsigned int>(absl::string_view, int, unsigned int*)
Unexecuted instantiation: numbers.cc:bool absl::(anonymous namespace)::safe_parse_positive_int<unsigned long>(absl::string_view, int, unsigned long*)
Unexecuted instantiation: numbers.cc:bool absl::(anonymous namespace)::safe_parse_positive_int<absl::uint128>(absl::string_view, int, absl::uint128*)
1009
1010
template <typename IntType>
1011
inline bool safe_parse_negative_int(absl::string_view text, int base,
1012
0
                                    IntType* value_p) {
1013
0
  IntType value = 0;
1014
0
  const IntType vmin = std::numeric_limits<IntType>::min();
1015
0
  assert(vmin < 0);
1016
0
  assert(vmin <= 0 - base);
1017
0
  IntType vmin_over_base = LookupTables<IntType>::kVminOverBase[base];
1018
0
  assert(base < 2 ||
1019
0
         std::numeric_limits<IntType>::min() / base == vmin_over_base);
1020
  // 2003 c++ standard [expr.mul]
1021
  // "... the sign of the remainder is implementation-defined."
1022
  // Although (vmin/base)*base + vmin%base is always vmin.
1023
  // 2011 c++ standard tightens the spec but we cannot rely on it.
1024
  // TODO(junyer): Handle this in the lookup table generation.
1025
0
  if (vmin % base > 0) {
1026
0
    vmin_over_base += 1;
1027
0
  }
1028
0
  const char* start = text.data();
1029
0
  const char* end = start + text.size();
1030
  // loop over digits
1031
0
  for (; start < end; ++start) {
1032
0
    unsigned char c = static_cast<unsigned char>(start[0]);
1033
0
    int digit = kAsciiToInt[c];
1034
0
    if (digit >= base) {
1035
0
      *value_p = value;
1036
0
      return false;
1037
0
    }
1038
0
    if (value < vmin_over_base) {
1039
0
      *value_p = vmin;
1040
0
      return false;
1041
0
    }
1042
0
    value *= base;
1043
0
    if (value < vmin + digit) {
1044
0
      *value_p = vmin;
1045
0
      return false;
1046
0
    }
1047
0
    value -= digit;
1048
0
  }
1049
0
  *value_p = value;
1050
0
  return true;
1051
0
}
Unexecuted instantiation: numbers.cc:bool absl::(anonymous namespace)::safe_parse_negative_int<int>(absl::string_view, int, int*)
Unexecuted instantiation: numbers.cc:bool absl::(anonymous namespace)::safe_parse_negative_int<long>(absl::string_view, int, long*)
Unexecuted instantiation: numbers.cc:bool absl::(anonymous namespace)::safe_parse_negative_int<absl::int128>(absl::string_view, int, absl::int128*)
1052
1053
// Input format based on POSIX.1-2008 strtol
1054
// http://pubs.opengroup.org/onlinepubs/9699919799/functions/strtol.html
1055
template <typename IntType>
1056
inline bool safe_int_internal(absl::string_view text, IntType* value_p,
1057
0
                              int base) {
1058
0
  *value_p = 0;
1059
0
  bool negative;
1060
0
  if (!safe_parse_sign_and_base(&text, &base, &negative)) {
1061
0
    return false;
1062
0
  }
1063
0
  if (!negative) {
1064
0
    return safe_parse_positive_int(text, base, value_p);
1065
0
  } else {
1066
0
    return safe_parse_negative_int(text, base, value_p);
1067
0
  }
1068
0
}
Unexecuted instantiation: numbers.cc:bool absl::(anonymous namespace)::safe_int_internal<int>(absl::string_view, int*, int)
Unexecuted instantiation: numbers.cc:bool absl::(anonymous namespace)::safe_int_internal<long>(absl::string_view, long*, int)
Unexecuted instantiation: numbers.cc:bool absl::(anonymous namespace)::safe_int_internal<absl::int128>(absl::string_view, absl::int128*, int)
1069
1070
template <typename IntType>
1071
inline bool safe_uint_internal(absl::string_view text, IntType* value_p,
1072
0
                               int base) {
1073
0
  *value_p = 0;
1074
0
  bool negative;
1075
0
  if (!safe_parse_sign_and_base(&text, &base, &negative) || negative) {
1076
0
    return false;
1077
0
  }
1078
0
  return safe_parse_positive_int(text, base, value_p);
1079
0
}
Unexecuted instantiation: numbers.cc:bool absl::(anonymous namespace)::safe_uint_internal<unsigned int>(absl::string_view, unsigned int*, int)
Unexecuted instantiation: numbers.cc:bool absl::(anonymous namespace)::safe_uint_internal<unsigned long>(absl::string_view, unsigned long*, int)
Unexecuted instantiation: numbers.cc:bool absl::(anonymous namespace)::safe_uint_internal<absl::uint128>(absl::string_view, absl::uint128*, int)
1080
}  // anonymous namespace
1081
1082
namespace numbers_internal {
1083
1084
// Digit conversion.
1085
ABSL_CONST_INIT ABSL_DLL const char kHexChar[] =
1086
    "0123456789abcdef";
1087
1088
ABSL_CONST_INIT ABSL_DLL const char kHexTable[513] =
1089
    "000102030405060708090a0b0c0d0e0f"
1090
    "101112131415161718191a1b1c1d1e1f"
1091
    "202122232425262728292a2b2c2d2e2f"
1092
    "303132333435363738393a3b3c3d3e3f"
1093
    "404142434445464748494a4b4c4d4e4f"
1094
    "505152535455565758595a5b5c5d5e5f"
1095
    "606162636465666768696a6b6c6d6e6f"
1096
    "707172737475767778797a7b7c7d7e7f"
1097
    "808182838485868788898a8b8c8d8e8f"
1098
    "909192939495969798999a9b9c9d9e9f"
1099
    "a0a1a2a3a4a5a6a7a8a9aaabacadaeaf"
1100
    "b0b1b2b3b4b5b6b7b8b9babbbcbdbebf"
1101
    "c0c1c2c3c4c5c6c7c8c9cacbcccdcecf"
1102
    "d0d1d2d3d4d5d6d7d8d9dadbdcdddedf"
1103
    "e0e1e2e3e4e5e6e7e8e9eaebecedeeef"
1104
    "f0f1f2f3f4f5f6f7f8f9fafbfcfdfeff";
1105
1106
0
bool safe_strto32_base(absl::string_view text, int32_t* value, int base) {
1107
0
  return safe_int_internal<int32_t>(text, value, base);
1108
0
}
1109
1110
0
bool safe_strto64_base(absl::string_view text, int64_t* value, int base) {
1111
0
  return safe_int_internal<int64_t>(text, value, base);
1112
0
}
1113
1114
0
bool safe_strto128_base(absl::string_view text, int128* value, int base) {
1115
0
  return safe_int_internal<absl::int128>(text, value, base);
1116
0
}
1117
1118
0
bool safe_strtou32_base(absl::string_view text, uint32_t* value, int base) {
1119
0
  return safe_uint_internal<uint32_t>(text, value, base);
1120
0
}
1121
1122
0
bool safe_strtou64_base(absl::string_view text, uint64_t* value, int base) {
1123
0
  return safe_uint_internal<uint64_t>(text, value, base);
1124
0
}
1125
1126
0
bool safe_strtou128_base(absl::string_view text, uint128* value, int base) {
1127
0
  return safe_uint_internal<absl::uint128>(text, value, base);
1128
0
}
1129
1130
}  // namespace numbers_internal
1131
ABSL_NAMESPACE_END
1132
}  // namespace absl