string.c++ 16.1 KB
Newer Older
Kenton Varda's avatar
Kenton Varda committed
1 2
// Copyright (c) 2013-2014 Sandstorm Development Group, Inc. and contributors
// Licensed under the MIT License:
3
//
Kenton Varda's avatar
Kenton Varda committed
4 5 6 7 8 9
// Permission is hereby granted, free of charge, to any person obtaining a copy
// of this software and associated documentation files (the "Software"), to deal
// in the Software without restriction, including without limitation the rights
// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
// copies of the Software, and to permit persons to whom the Software is
// furnished to do so, subject to the following conditions:
10
//
Kenton Varda's avatar
Kenton Varda committed
11 12
// The above copyright notice and this permission notice shall be included in
// all copies or substantial portions of the Software.
13
//
Kenton Varda's avatar
Kenton Varda committed
14 15 16 17 18 19 20
// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
// THE SOFTWARE.
21 22

#include "string.h"
Kenton Varda's avatar
Kenton Varda committed
23
#include "debug.h"
Kenton Varda's avatar
Kenton Varda committed
24 25 26 27
#include <stdio.h>
#include <float.h>
#include <errno.h>
#include <stdlib.h>
28
#include <stdint.h>
29 30 31

namespace kj {

32 33 34 35 36
#if _MSC_VER
#pragma warning(disable: 4996)
// Warns that sprintf() is buffer-overrunny. We know that, it's cool.
#endif

37
namespace {
38 39 40 41
bool isHex(const char *s) {
  if (*s == '-') s++;
  return s[0] == '0' && (s[1] == 'x' || s[1] == 'X');
}
42 43

long long parseSigned(const StringPtr& s, long long min, long long max) {
44
  KJ_REQUIRE(s != nullptr, "String does not contain valid number", s) { return 0; }
45 46
  char *endPtr;
  errno = 0;
47
  auto value = strtoll(s.begin(), &endPtr, isHex(s.cStr()) ? 16 : 10);
48 49 50
  KJ_REQUIRE(endPtr == s.end(), "String does not contain valid number", s) { return 0; }
  KJ_REQUIRE(errno != ERANGE, "Value out-of-range", s) { return 0; }
  KJ_REQUIRE(value >= min && value <= max, "Value out-of-range", value, min, max) { return 0; }
51 52 53 54
  return value;
}

unsigned long long parseUnsigned(const StringPtr& s, unsigned long long max) {
55
  KJ_REQUIRE(s != nullptr, "String does not contain valid number", s) { return 0; }
56 57
  char *endPtr;
  errno = 0;
58
  auto value = strtoull(s.begin(), &endPtr, isHex(s.cStr()) ? 16 : 10);
59 60 61 62 63
  KJ_REQUIRE(endPtr == s.end(), "String does not contain valid number", s) { return 0; }
  KJ_REQUIRE(errno != ERANGE, "Value out-of-range", s) { return 0; }
  KJ_REQUIRE(value <= max, "Value out-of-range", value, max) { return 0; }
  //strtoull("-1") does not fail with ERANGE
  KJ_REQUIRE(s[0] != '-', "Value out-of-range", s) { return 0; }
64 65 66 67 68
  return value;
}

template <typename T>
T parseInteger(const StringPtr& s) {
69 70 71
  if (static_cast<T>(minValue) < 0) {
    long long min = static_cast<T>(minValue);
    long long max = static_cast<T>(maxValue);
72 73
    return static_cast<T>(parseSigned(s, min, max));
  } else {
74
    unsigned long long max = static_cast<T>(maxValue);
75 76 77 78 79
    return static_cast<T>(parseUnsigned(s, max));
  }
}

double parseDouble(const StringPtr& s) {
80
  KJ_REQUIRE(s != nullptr, "String does not contain valid number", s) { return 0; }
81 82 83
  char *endPtr;
  errno = 0;
  auto value = strtod(s.begin(), &endPtr);
84
  KJ_REQUIRE(endPtr == s.end(), "String does not contain valid floating number", s) { return 0; }
85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106
  return value;
}

} // namespace

#define PARSE_AS_INTEGER(T) \
    template <> T StringPtr::parseAs<T>() const { return parseInteger<T>(*this); }
PARSE_AS_INTEGER(char);
PARSE_AS_INTEGER(signed char);
PARSE_AS_INTEGER(unsigned char);
PARSE_AS_INTEGER(short);
PARSE_AS_INTEGER(unsigned short);
PARSE_AS_INTEGER(int);
PARSE_AS_INTEGER(unsigned int);
PARSE_AS_INTEGER(long);
PARSE_AS_INTEGER(unsigned long);
PARSE_AS_INTEGER(long long);
PARSE_AS_INTEGER(unsigned long long);
#undef PARSE_AS_INTEGER
template <> double StringPtr::parseAs<double>() const { return parseDouble(*this); }
template <> float StringPtr::parseAs<float>() const { return parseDouble(*this); }

Kenton Varda's avatar
Kenton Varda committed
107
String heapString(size_t size) {
108
  char* buffer = _::HeapArrayDisposer::allocate<char>(size + 1);
Kenton Varda's avatar
Kenton Varda committed
109
  buffer[size] = '\0';
110
  return String(buffer, size, _::HeapArrayDisposer::instance);
111 112
}

Kenton Varda's avatar
Kenton Varda committed
113
String heapString(const char* value, size_t size) {
114
  char* buffer = _::HeapArrayDisposer::allocate<char>(size + 1);
115 116 117
  if (size != 0u) {
    memcpy(buffer, value, size);
  }
118
  buffer[size] = '\0';
119
  return String(buffer, size, _::HeapArrayDisposer::instance);
120 121
}

122 123 124 125
template <typename T>
static CappedArray<char, sizeof(T) * 2 + 1> hexImpl(T i) {
  // We don't use sprintf() because it's not async-signal-safe (for strPreallocated()).
  CappedArray<char, sizeof(T) * 2 + 1> result;
126 127
  uint8_t reverse[sizeof(T) * 2];
  uint8_t* p = reverse;
128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145
  if (i == 0) {
    *p++ = 0;
  } else {
    while (i > 0) {
      *p++ = i % 16;
      i /= 16;
    }
  }

  char* p2 = result.begin();
  while (p > reverse) {
    *p2++ = "0123456789abcdef"[*--p];
  }
  result.setSize(p2 - result.begin());
  return result;
}

#define HEXIFY_INT(type) \
146
CappedArray<char, sizeof(type) * 2 + 1> hex(type i) { \
147
  return hexImpl<type>(i); \
Kenton Varda's avatar
Kenton Varda committed
148 149
}

150 151 152 153 154
HEXIFY_INT(unsigned char);
HEXIFY_INT(unsigned short);
HEXIFY_INT(unsigned int);
HEXIFY_INT(unsigned long);
HEXIFY_INT(unsigned long long);
Kenton Varda's avatar
Kenton Varda committed
155 156 157

#undef HEXIFY_INT

158
namespace _ {  // private
Kenton Varda's avatar
Kenton Varda committed
159

160 161 162 163
StringPtr Stringifier::operator*(decltype(nullptr)) const {
  return "nullptr";
}

Kenton Varda's avatar
Kenton Varda committed
164 165 166 167
StringPtr Stringifier::operator*(bool b) const {
  return b ? StringPtr("true") : StringPtr("false");
}

168 169 170 171 172 173
template <typename T, typename Unsigned>
static CappedArray<char, sizeof(T) * 3 + 2> stringifyImpl(T i) {
  // We don't use sprintf() because it's not async-signal-safe (for strPreallocated()).
  CappedArray<char, sizeof(T) * 3 + 2> result;
  bool negative = i < 0;
  Unsigned u = negative ? -i : i;
174 175
  uint8_t reverse[sizeof(T) * 3 + 1];
  uint8_t* p = reverse;
176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194
  if (u == 0) {
    *p++ = 0;
  } else {
    while (u > 0) {
      *p++ = u % 10;
      u /= 10;
    }
  }

  char* p2 = result.begin();
  if (negative) *p2++ = '-';
  while (p > reverse) {
    *p2++ = '0' + *--p;
  }
  result.setSize(p2 - result.begin());
  return result;
}

#define STRINGIFY_INT(type, unsigned) \
195
CappedArray<char, sizeof(type) * 3 + 2> Stringifier::operator*(type i) const { \
196
  return stringifyImpl<type, unsigned>(i); \
Kenton Varda's avatar
Kenton Varda committed
197 198
}

199 200 201 202 203 204 205 206 207 208
STRINGIFY_INT(signed char, uint);
STRINGIFY_INT(unsigned char, uint);
STRINGIFY_INT(short, uint);
STRINGIFY_INT(unsigned short, uint);
STRINGIFY_INT(int, uint);
STRINGIFY_INT(unsigned int, uint);
STRINGIFY_INT(long, unsigned long);
STRINGIFY_INT(unsigned long, unsigned long);
STRINGIFY_INT(long long, unsigned long long);
STRINGIFY_INT(unsigned long long, unsigned long long);
Kenton Varda's avatar
Kenton Varda committed
209 210 211

#undef STRINGIFY_INT

212 213 214 215
CappedArray<char, sizeof(const void*) * 2 + 1> Stringifier::operator*(const void* i) const { \
  return hexImpl<uintptr_t>(reinterpret_cast<uintptr_t>(i));
}

Kenton Varda's avatar
Kenton Varda committed
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
namespace {

// ----------------------------------------------------------------------
// DoubleToBuffer()
// FloatToBuffer()
//    Copied from Protocol Buffers, (C) Google, BSD license.
//    Kenton wrote this code originally.  The following commentary is
//    from the original.
//
//    Description: converts a double or float to a string which, if
//    passed to NoLocaleStrtod(), will produce the exact same original double
//    (except in case of NaN; all NaNs are considered the same value).
//    We try to keep the string short but it's not guaranteed to be as
//    short as possible.
//
//    DoubleToBuffer() and FloatToBuffer() write the text to the given
//    buffer and return it.  The buffer must be at least
//    kDoubleToBufferSize bytes for doubles and kFloatToBufferSize
//    bytes for floats.  kFastToBufferSize is also guaranteed to be large
//    enough to hold either.
//
//    We want to print the value without losing precision, but we also do
//    not want to print more digits than necessary.  This turns out to be
//    trickier than it sounds.  Numbers like 0.2 cannot be represented
//    exactly in binary.  If we print 0.2 with a very large precision,
//    e.g. "%.50g", we get "0.2000000000000000111022302462515654042363167".
//    On the other hand, if we set the precision too low, we lose
//    significant digits when printing numbers that actually need them.
//    It turns out there is no precision value that does the right thing
//    for all numbers.
//
//    Our strategy is to first try printing with a precision that is never
//    over-precise, then parse the result with strtod() to see if it
//    matches.  If not, we print again with a precision that will always
//    give a precise result, but may use more digits than necessary.
//
//    An arguably better strategy would be to use the algorithm described
//    in "How to Print Floating-Point Numbers Accurately" by Steele &
//    White, e.g. as implemented by David M. Gay's dtoa().  It turns out,
//    however, that the following implementation is about as fast as
//    DMG's code.  Furthermore, DMG's code locks mutexes, which means it
//    will not scale well on multi-core machines.  DMG's code is slightly
//    more accurate (in that it will never use more digits than
//    necessary), but this is probably irrelevant for most users.
//
//    Rob Pike and Ken Thompson also have an implementation of dtoa() in
//    third_party/fmt/fltfmt.cc.  Their implementation is similar to this
//    one in that it makes guesses and then uses strtod() to check them.
//    Their implementation is faster because they use their own code to
//    generate the digits in the first place rather than use snprintf(),
//    thus avoiding format string parsing overhead.  However, this makes
//    it considerably more complicated than the following implementation,
//    and it is embedded in a larger library.  If speed turns out to be
//    an issue, we could re-implement this in terms of their
//    implementation.
// ----------------------------------------------------------------------

#ifdef _WIN32
// MSVC has only _snprintf, not snprintf.
//
// MinGW has both snprintf and _snprintf, but they appear to be different
// functions.  The former is buggy.  When invoked like so:
//   char buffer[32];
//   snprintf(buffer, 32, "%.*g\n", FLT_DIG, 1.23e10f);
// it prints "1.23000e+10".  This is plainly wrong:  %g should never print
// trailing zeros after the decimal point.  For some reason this bug only
// occurs with some input values, not all.  In any case, _snprintf does the
// right thing, so we use it.
#define snprintf _snprintf
#endif

inline bool IsNaN(double value) {
  // NaN is never equal to anything, even itself.
  return value != value;
}

// In practice, doubles should never need more than 24 bytes and floats
// should never need more than 14 (including null terminators), but we
// overestimate to be safe.
static const int kDoubleToBufferSize = 32;
static const int kFloatToBufferSize = 24;

static inline bool IsValidFloatChar(char c) {
  return ('0' <= c && c <= '9') ||
         c == 'e' || c == 'E' ||
         c == '+' || c == '-';
}

void DelocalizeRadix(char* buffer) {
  // Fast check:  if the buffer has a normal decimal point, assume no
  // translation is needed.
  if (strchr(buffer, '.') != NULL) return;

  // Find the first unknown character.
  while (IsValidFloatChar(*buffer)) ++buffer;

  if (*buffer == '\0') {
    // No radix character found.
    return;
  }

  // We are now pointing at the locale-specific radix character.  Replace it
  // with '.'.
  *buffer = '.';
  ++buffer;

  if (!IsValidFloatChar(*buffer) && *buffer != '\0') {
    // It appears the radix was a multi-byte character.  We need to remove the
    // extra bytes.
    char* target = buffer;
    do { ++buffer; } while (!IsValidFloatChar(*buffer) && *buffer != '\0');
    memmove(target, buffer, strlen(buffer) + 1);
  }
}

void RemovePlus(char* buffer) {
  // Remove any + characters because they are redundant and ugly.

  for (;;) {
    buffer = strchr(buffer, '+');
    if (buffer == NULL) {
      return;
    }
    memmove(buffer, buffer + 1, strlen(buffer + 1) + 1);
  }
}

343
#if _WIN32
344 345 346 347
void RemoveE0(char* buffer) {
  // Remove redundant leading 0's after an e, e.g. 1e012. Seems to appear on
  // Windows.

348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365
  // Find and skip 'e'.
  char* ptr = strchr(buffer, 'e');
  if (ptr == nullptr) return;
  ++ptr;

  // Skip '-'.
  if (*ptr == '-') ++ptr;

  // Skip '0's.
  char* ptr2 = ptr;
  while (*ptr2 == '0') ++ptr2;

  // If we went past the last digit, back up one.
  if (*ptr2 < '0' || *ptr2 > '9') --ptr2;

  // Move bytes backwards.
  if (ptr2 > ptr) {
    memmove(ptr, ptr2, strlen(ptr2) + 1);
366 367
  }
}
368
#endif
369

Kenton Varda's avatar
Kenton Varda committed
370 371 372 373 374 375 376
char* DoubleToBuffer(double value, char* buffer) {
  // DBL_DIG is 15 for IEEE-754 doubles, which are used on almost all
  // platforms these days.  Just in case some system exists where DBL_DIG
  // is significantly larger -- and risks overflowing our buffer -- we have
  // this assert.
  static_assert(DBL_DIG < 20, "DBL_DIG is too big.");

377
  if (value == inf()) {
Kenton Varda's avatar
Kenton Varda committed
378 379
    strcpy(buffer, "inf");
    return buffer;
380
  } else if (value == -inf()) {
Kenton Varda's avatar
Kenton Varda committed
381 382 383 384 385 386 387
    strcpy(buffer, "-inf");
    return buffer;
  } else if (IsNaN(value)) {
    strcpy(buffer, "nan");
    return buffer;
  }

388
  int snprintf_result KJ_UNUSED =
Kenton Varda's avatar
Kenton Varda committed
389 390 391 392
    snprintf(buffer, kDoubleToBufferSize, "%.*g", DBL_DIG, value);

  // The snprintf should never overflow because the buffer is significantly
  // larger than the precision we asked for.
393
  KJ_DASSERT(snprintf_result > 0 && snprintf_result < kDoubleToBufferSize);
Kenton Varda's avatar
Kenton Varda committed
394 395 396 397 398 399 400 401 402

  // We need to make parsed_value volatile in order to force the compiler to
  // write it out to the stack.  Otherwise, it may keep the value in a
  // register, and if it does that, it may keep it as a long double instead
  // of a double.  This long double may have extra bits that make it compare
  // unequal to "value" even though it would be exactly equal if it were
  // truncated to a double.
  volatile double parsed_value = strtod(buffer, NULL);
  if (parsed_value != value) {
403
    int snprintf_result2 KJ_UNUSED =
Kenton Varda's avatar
Kenton Varda committed
404 405 406
      snprintf(buffer, kDoubleToBufferSize, "%.*g", DBL_DIG+2, value);

    // Should never overflow; see above.
407
    KJ_DASSERT(snprintf_result2 > 0 && snprintf_result2 < kDoubleToBufferSize);
Kenton Varda's avatar
Kenton Varda committed
408 409 410 411
  }

  DelocalizeRadix(buffer);
  RemovePlus(buffer);
412 413 414
#if _WIN32
  RemoveE0(buffer);
#endif // _WIN32
Kenton Varda's avatar
Kenton Varda committed
415 416 417 418 419 420 421
  return buffer;
}

bool safe_strtof(const char* str, float* value) {
  char* endptr;
  errno = 0;  // errno only gets set on errors
#if defined(_WIN32) || defined (__hpux)  // has no strtof()
422
  *value = static_cast<float>(strtod(str, &endptr));
Kenton Varda's avatar
Kenton Varda committed
423 424 425 426 427 428 429 430 431 432 433 434 435
#else
  *value = strtof(str, &endptr);
#endif
  return *str != 0 && *endptr == 0 && errno == 0;
}

char* FloatToBuffer(float value, char* buffer) {
  // FLT_DIG is 6 for IEEE-754 floats, which are used on almost all
  // platforms these days.  Just in case some system exists where FLT_DIG
  // is significantly larger -- and risks overflowing our buffer -- we have
  // this assert.
  static_assert(FLT_DIG < 10, "FLT_DIG is too big");

436
  if (value == inf()) {
Kenton Varda's avatar
Kenton Varda committed
437 438
    strcpy(buffer, "inf");
    return buffer;
439
  } else if (value == -inf()) {
Kenton Varda's avatar
Kenton Varda committed
440 441 442 443 444 445 446
    strcpy(buffer, "-inf");
    return buffer;
  } else if (IsNaN(value)) {
    strcpy(buffer, "nan");
    return buffer;
  }

447
  int snprintf_result KJ_UNUSED =
Kenton Varda's avatar
Kenton Varda committed
448 449 450 451
    snprintf(buffer, kFloatToBufferSize, "%.*g", FLT_DIG, value);

  // The snprintf should never overflow because the buffer is significantly
  // larger than the precision we asked for.
452
  KJ_DASSERT(snprintf_result > 0 && snprintf_result < kFloatToBufferSize);
Kenton Varda's avatar
Kenton Varda committed
453 454 455

  float parsed_value;
  if (!safe_strtof(buffer, &parsed_value) || parsed_value != value) {
456
    int snprintf_result2 KJ_UNUSED =
Kenton Varda's avatar
Kenton Varda committed
457 458 459
      snprintf(buffer, kFloatToBufferSize, "%.*g", FLT_DIG+2, value);

    // Should never overflow; see above.
460
    KJ_DASSERT(snprintf_result2 > 0 && snprintf_result2 < kFloatToBufferSize);
Kenton Varda's avatar
Kenton Varda committed
461 462 463 464
  }

  DelocalizeRadix(buffer);
  RemovePlus(buffer);
465 466 467
#if _WIN32
  RemoveE0(buffer);
#endif // _WIN32
Kenton Varda's avatar
Kenton Varda committed
468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484
  return buffer;
}

}  // namespace

CappedArray<char, kFloatToBufferSize> Stringifier::operator*(float f) const {
  CappedArray<char, kFloatToBufferSize> result;
  result.setSize(strlen(FloatToBuffer(f, result.begin())));
  return result;
}

CappedArray<char, kDoubleToBufferSize> Stringifier::operator*(double f) const {
  CappedArray<char, kDoubleToBufferSize> result;
  result.setSize(strlen(DoubleToBuffer(f, result.begin())));
  return result;
}

485
}  // namespace _ (private)
486
}  // namespace kj