| 1 | // Copyright (C) 2021 The Qt Company Ltd. |
| 2 | // Copyright (C) 2016 Intel Corporation. |
| 3 | // SPDX-License-Identifier: LicenseRef-Qt-Commercial OR LGPL-3.0-only OR GPL-2.0-only OR GPL-3.0-only |
| 4 | |
| 5 | #include "qlocale_tools_p.h" |
| 6 | #include "qdoublescanprint_p.h" |
| 7 | #include "qlocale_p.h" |
| 8 | #include "qstring.h" |
| 9 | |
| 10 | #include <private/qtools_p.h> |
| 11 | #include <private/qnumeric_p.h> |
| 12 | |
| 13 | #include <cstdio> |
| 14 | |
| 15 | #include <ctype.h> |
| 16 | #include <errno.h> |
| 17 | #include <float.h> |
| 18 | #include <limits.h> |
| 19 | #include <math.h> |
| 20 | #include <stdlib.h> |
| 21 | #include <time.h> |
| 22 | |
| 23 | #include <limits> |
| 24 | #include <charconv> |
| 25 | |
| 26 | #if defined(Q_OS_LINUX) && !defined(__UCLIBC__) |
| 27 | # include <fenv.h> |
| 28 | #endif |
| 29 | |
| 30 | // Sizes as defined by the ISO C99 standard - fallback |
| 31 | #ifndef LLONG_MAX |
| 32 | # define LLONG_MAX Q_INT64_C(0x7fffffffffffffff) |
| 33 | #endif |
| 34 | #ifndef LLONG_MIN |
| 35 | # define LLONG_MIN (-LLONG_MAX - Q_INT64_C(1)) |
| 36 | #endif |
| 37 | #ifndef ULLONG_MAX |
| 38 | # define ULLONG_MAX Q_UINT64_C(0xffffffffffffffff) |
| 39 | #endif |
| 40 | |
| 41 | QT_BEGIN_NAMESPACE |
| 42 | |
| 43 | using namespace QtMiscUtils; |
| 44 | |
| 45 | QT_CLOCALE_HOLDER |
| 46 | |
| 47 | void qt_doubleToAscii(double d, QLocaleData::DoubleForm form, int precision, |
| 48 | char *buf, qsizetype bufSize, |
| 49 | bool &sign, int &length, int &decpt) |
| 50 | { |
| 51 | if (bufSize == 0) { |
| 52 | decpt = 0; |
| 53 | sign = d < 0; |
| 54 | length = 0; |
| 55 | return; |
| 56 | } |
| 57 | |
| 58 | // Detect special numbers (nan, +/-inf) |
| 59 | // We cannot use the high-level API of libdouble-conversion as we need to |
| 60 | // apply locale-specific formatting, such as decimal points, grouping |
| 61 | // separators, etc. Because of this, we have to check for infinity and NaN |
| 62 | // before calling DoubleToAscii. |
| 63 | if (qt_is_inf(d)) { |
| 64 | sign = d < 0; |
| 65 | if (bufSize >= 3) { |
| 66 | buf[0] = 'i'; |
| 67 | buf[1] = 'n'; |
| 68 | buf[2] = 'f'; |
| 69 | length = 3; |
| 70 | } else { |
| 71 | length = 0; |
| 72 | } |
| 73 | return; |
| 74 | } else if (qt_is_nan(d)) { |
| 75 | if (bufSize >= 3) { |
| 76 | buf[0] = 'n'; |
| 77 | buf[1] = 'a'; |
| 78 | buf[2] = 'n'; |
| 79 | length = 3; |
| 80 | } else { |
| 81 | length = 0; |
| 82 | } |
| 83 | return; |
| 84 | } |
| 85 | |
| 86 | if (form == QLocaleData::DFSignificantDigits && precision == 0) |
| 87 | precision = 1; // 0 significant digits is silently converted to 1 |
| 88 | |
| 89 | #if !defined(QT_NO_DOUBLECONVERSION) && !defined(QT_BOOTSTRAPPED) |
| 90 | // one digit before the decimal dot, counts as significant digit for DoubleToStringConverter |
| 91 | if (form == QLocaleData::DFExponent && precision >= 0) |
| 92 | ++precision; |
| 93 | |
| 94 | double_conversion::DoubleToStringConverter::DtoaMode mode; |
| 95 | if (precision == QLocale::FloatingPointShortest) { |
| 96 | mode = double_conversion::DoubleToStringConverter::SHORTEST; |
| 97 | } else if (form == QLocaleData::DFSignificantDigits || form == QLocaleData::DFExponent) { |
| 98 | mode = double_conversion::DoubleToStringConverter::PRECISION; |
| 99 | } else { |
| 100 | mode = double_conversion::DoubleToStringConverter::FIXED; |
| 101 | } |
| 102 | // libDoubleConversion is limited to 32-bit lengths. It's ok to cap the buffer size, |
| 103 | // though, because the library will never write 2GiB of chars as output |
| 104 | // (the length out-parameter is just an int, too). |
| 105 | const auto boundedBufferSize = static_cast<int>((std::min)(a: bufSize, b: qsizetype(INT_MAX))); |
| 106 | double_conversion::DoubleToStringConverter::DoubleToAscii(v: d, mode, requested_digits: precision, buffer: buf, |
| 107 | buffer_length: boundedBufferSize, |
| 108 | sign: &sign, length: &length, point: &decpt); |
| 109 | #else // QT_NO_DOUBLECONVERSION || QT_BOOTSTRAPPED |
| 110 | |
| 111 | // Cut the precision at 999, to fit it into the format string. We can't get more than 17 |
| 112 | // significant digits, so anything after that is mostly noise. You do get closer to the "middle" |
| 113 | // of the range covered by the given double with more digits, so to a degree it does make sense |
| 114 | // to honor higher precisions. We define that at more than 999 digits that is not the case. |
| 115 | if (precision > 999) |
| 116 | precision = 999; |
| 117 | else if (precision == QLocale::FloatingPointShortest) |
| 118 | precision = std::numeric_limits<double>::max_digits10; // snprintf lacks "shortest" mode |
| 119 | |
| 120 | if (isZero(d)) { |
| 121 | // Negative zero is expected as simple "0", not "-0". We cannot do d < 0, though. |
| 122 | sign = false; |
| 123 | buf[0] = '0'; |
| 124 | length = 1; |
| 125 | decpt = 1; |
| 126 | return; |
| 127 | } else if (d < 0) { |
| 128 | sign = true; |
| 129 | d = -d; |
| 130 | } else { |
| 131 | sign = false; |
| 132 | } |
| 133 | |
| 134 | const int formatLength = 7; // '%', '.', 3 digits precision, 'f', '\0' |
| 135 | char format[formatLength]; |
| 136 | format[formatLength - 1] = '\0'; |
| 137 | format[0] = '%'; |
| 138 | format[1] = '.'; |
| 139 | format[2] = char((precision / 100) % 10) + '0'; |
| 140 | format[3] = char((precision / 10) % 10) + '0'; |
| 141 | format[4] = char(precision % 10) + '0'; |
| 142 | int extraChars; |
| 143 | switch (form) { |
| 144 | case QLocaleData::DFDecimal: |
| 145 | format[formatLength - 2] = 'f'; |
| 146 | // <anything> '.' <precision> '\0' |
| 147 | extraChars = wholePartSpace(d) + 2; |
| 148 | break; |
| 149 | case QLocaleData::DFExponent: |
| 150 | format[formatLength - 2] = 'e'; |
| 151 | // '.', 'e', '-', <exponent> '\0' |
| 152 | extraChars = 7; |
| 153 | break; |
| 154 | case QLocaleData::DFSignificantDigits: |
| 155 | format[formatLength - 2] = 'g'; |
| 156 | |
| 157 | // either the same as in the 'e' case, or '.' and '\0' |
| 158 | // precision covers part before '.' |
| 159 | extraChars = 7; |
| 160 | break; |
| 161 | default: |
| 162 | Q_UNREACHABLE(); |
| 163 | } |
| 164 | |
| 165 | QVarLengthArray<char> target(precision + extraChars); |
| 166 | |
| 167 | length = qDoubleSnprintf(target.data(), target.size(), QT_CLOCALE, format, d); |
| 168 | int firstSignificant = 0; |
| 169 | int decptInTarget = length; |
| 170 | |
| 171 | // Find the first significant digit (not 0), and note any '.' we encounter. |
| 172 | // There is no '-' at the front of target because we made sure d > 0 above. |
| 173 | while (firstSignificant < length) { |
| 174 | if (target[firstSignificant] == '.') |
| 175 | decptInTarget = firstSignificant; |
| 176 | else if (target[firstSignificant] != '0') |
| 177 | break; |
| 178 | ++firstSignificant; |
| 179 | } |
| 180 | |
| 181 | // If no '.' found so far, search the rest of the target buffer for it. |
| 182 | if (decptInTarget == length) |
| 183 | decptInTarget = std::find(target.data() + firstSignificant, target.data() + length, '.') - |
| 184 | target.data(); |
| 185 | |
| 186 | int eSign = length; |
| 187 | if (form != QLocaleData::DFDecimal) { |
| 188 | // In 'e' or 'g' form, look for the 'e'. |
| 189 | eSign = std::find(target.data() + firstSignificant, target.data() + length, 'e') - |
| 190 | target.data(); |
| 191 | |
| 192 | if (eSign < length) { |
| 193 | // If 'e' is found, the final decimal point is determined by the number after 'e'. |
| 194 | // Mind that the final decimal point, decpt, is the offset of the decimal point from the |
| 195 | // start of the resulting string in buf. It may be negative or larger than bufSize, in |
| 196 | // which case the missing digits are zeroes. In the 'e' case decptInTarget is always 1, |
| 197 | // as variants of snprintf always generate numbers with one digit before the '.' then. |
| 198 | // This is why the final decimal point is offset by 1, relative to the number after 'e'. |
| 199 | auto r = qstrntoll(target.data() + eSign + 1, length - eSign - 1, 10); |
| 200 | decpt = r.result + 1; |
| 201 | Q_ASSERT(r.ok()); |
| 202 | Q_ASSERT(r.used + eSign + 1 <= length); |
| 203 | } else { |
| 204 | // No 'e' found, so it's the 'f' form. Variants of snprintf generate numbers with |
| 205 | // potentially multiple digits before the '.', but without decimal exponent then. So we |
| 206 | // get the final decimal point from the position of the '.'. The '.' itself takes up one |
| 207 | // character. We adjust by 1 below if that gets in the way. |
| 208 | decpt = decptInTarget - firstSignificant; |
| 209 | } |
| 210 | } else { |
| 211 | // In 'f' form, there can not be an 'e', so it's enough to look for the '.' |
| 212 | // (and possibly adjust by 1 below) |
| 213 | decpt = decptInTarget - firstSignificant; |
| 214 | } |
| 215 | |
| 216 | // Move the actual digits from the snprintf target to the actual buffer. |
| 217 | if (decptInTarget > firstSignificant) { |
| 218 | // First move the digits before the '.', if any |
| 219 | int lengthBeforeDecpt = decptInTarget - firstSignificant; |
| 220 | memcpy(buf, target.data() + firstSignificant, qMin(lengthBeforeDecpt, bufSize)); |
| 221 | if (eSign > decptInTarget && lengthBeforeDecpt < bufSize) { |
| 222 | // Then move any remaining digits, until 'e' |
| 223 | memcpy(buf + lengthBeforeDecpt, target.data() + decptInTarget + 1, |
| 224 | qMin(eSign - decptInTarget - 1, bufSize - lengthBeforeDecpt)); |
| 225 | // The final length of the output is the distance between the first significant digit |
| 226 | // and 'e' minus 1, for the '.', except if the buffer is smaller. |
| 227 | length = qMin(eSign - firstSignificant - 1, bufSize); |
| 228 | } else { |
| 229 | // 'e' was before the decpt or things didn't fit. Don't subtract the '.' from the length. |
| 230 | length = qMin(eSign - firstSignificant, bufSize); |
| 231 | } |
| 232 | } else { |
| 233 | if (eSign > firstSignificant) { |
| 234 | // If there are any significant digits at all, they are all after the '.' now. |
| 235 | // Just copy them straight away. |
| 236 | memcpy(buf, target.data() + firstSignificant, qMin(eSign - firstSignificant, bufSize)); |
| 237 | |
| 238 | // The decimal point was before the first significant digit, so we were one off above. |
| 239 | // Consider 0.1 - buf will be just '1', and decpt should be 0. But |
| 240 | // "decptInTarget - firstSignificant" will yield -1. |
| 241 | ++decpt; |
| 242 | length = qMin(eSign - firstSignificant, bufSize); |
| 243 | } else { |
| 244 | // No significant digits means the number is just 0. |
| 245 | buf[0] = '0'; |
| 246 | length = 1; |
| 247 | decpt = 1; |
| 248 | } |
| 249 | } |
| 250 | #endif // QT_NO_DOUBLECONVERSION || QT_BOOTSTRAPPED |
| 251 | while (length > 1 && buf[length - 1] == '0') // drop trailing zeroes |
| 252 | --length; |
| 253 | } |
| 254 | |
| 255 | QSimpleParsedNumber<double> qt_asciiToDouble(const char *num, qsizetype numLen, |
| 256 | StrayCharacterMode strayCharMode) |
| 257 | { |
| 258 | if (numLen <= 0) |
| 259 | return {}; |
| 260 | |
| 261 | // We have to catch NaN before because we need NaN as marker for "garbage" in the |
| 262 | // libdouble-conversion case and, in contrast to libdouble-conversion or sscanf, we don't allow |
| 263 | // "-nan" or "+nan" |
| 264 | if (char c = *num; numLen >= 3 |
| 265 | && (c == '-' || c == '+' || c == 'I' || c == 'i' || c == 'N' || c == 'n')) { |
| 266 | bool negative = (c == '-'); |
| 267 | bool hasSign = negative || (c == '+'); |
| 268 | qptrdiff offset = 0; |
| 269 | if (hasSign) { |
| 270 | offset = 1; |
| 271 | c = num[offset]; |
| 272 | } |
| 273 | |
| 274 | if (c > '9') { |
| 275 | auto lowered = [](char c) { |
| 276 | // this will mangle non-letters, but none can become a letter |
| 277 | return c | 0x20; |
| 278 | }; |
| 279 | |
| 280 | // Found a non-digit, so this MUST be either "inf", "+inf", "-inf" |
| 281 | // or "nan". Anything else is an invalid parse and we don't need to |
| 282 | // feed it to the converter below. |
| 283 | if (numLen != offset + 3) |
| 284 | return {}; |
| 285 | |
| 286 | c = lowered(c); |
| 287 | char c2 = lowered(num[offset + 1]); |
| 288 | char c3 = lowered(num[offset + 2]); |
| 289 | if (c == 'i' && c2 == 'n' && c3 == 'f') |
| 290 | return { .result: negative ? -qt_inf() : qt_inf(), .used: offset + 3 }; |
| 291 | else if (c == 'n' && c2 == 'a' && c3 == 'n' && !hasSign) |
| 292 | return { .result: qt_qnan(), .used: 3 }; |
| 293 | return {}; |
| 294 | } |
| 295 | } |
| 296 | |
| 297 | double d = 0.0; |
| 298 | int processed; |
| 299 | #if !defined(QT_NO_DOUBLECONVERSION) && !defined(QT_BOOTSTRAPPED) |
| 300 | int conv_flags = double_conversion::StringToDoubleConverter::NO_FLAGS; |
| 301 | if (strayCharMode == TrailingJunkAllowed) { |
| 302 | conv_flags = double_conversion::StringToDoubleConverter::ALLOW_TRAILING_JUNK; |
| 303 | } else if (strayCharMode == WhitespacesAllowed) { |
| 304 | conv_flags = double_conversion::StringToDoubleConverter::ALLOW_LEADING_SPACES |
| 305 | | double_conversion::StringToDoubleConverter::ALLOW_TRAILING_SPACES; |
| 306 | } |
| 307 | double_conversion::StringToDoubleConverter conv(conv_flags, 0.0, qt_qnan(), nullptr, nullptr); |
| 308 | if (int(numLen) != numLen) { |
| 309 | // a number over 2 GB in length is silly, just assume it isn't valid |
| 310 | return {}; |
| 311 | } else { |
| 312 | d = conv.StringToDouble(buffer: num, length: int(numLen), processed_characters_count: &processed); |
| 313 | } |
| 314 | |
| 315 | if (!qt_is_finite(d)) { |
| 316 | if (qt_is_nan(d)) { |
| 317 | // Garbage found. We don't accept it and return 0. |
| 318 | return {}; |
| 319 | } else { |
| 320 | // Overflow. That's not OK, but we still return infinity. |
| 321 | return { .result: d, .used: -processed }; |
| 322 | } |
| 323 | } |
| 324 | #else |
| 325 | // ::digits10 is 19, but ::max() is 18'446'744'073'709'551'615ULL - go, figure... |
| 326 | constexpr auto maxDigitsForULongLong = 1 + std::numeric_limits<unsigned long long>::digits10; |
| 327 | // need to ensure that we don't read more than numLen of input: |
| 328 | char fmt[1 + maxDigitsForULongLong + 4 + 1]; |
| 329 | std::snprintf(fmt, sizeof fmt, "%s%llu%s" , |
| 330 | "%" , static_cast<unsigned long long>(numLen), "lf%n" ); |
| 331 | |
| 332 | if (qDoubleSscanf(num, QT_CLOCALE, fmt, &d, &processed) < 1) |
| 333 | processed = 0; |
| 334 | |
| 335 | if ((strayCharMode == TrailingJunkProhibited && processed != numLen) || qt_is_nan(d)) { |
| 336 | // Implementation defined nan symbol or garbage found. We don't accept it. |
| 337 | return {}; |
| 338 | } |
| 339 | |
| 340 | if (!qt_is_finite(d)) { |
| 341 | // Overflow. Check for implementation-defined infinity symbols and reject them. |
| 342 | // We assume that any infinity symbol has to contain a character that cannot be part of a |
| 343 | // "normal" number (that is 0-9, ., -, +, e). |
| 344 | for (int i = 0; i < processed; ++i) { |
| 345 | char c = num[i]; |
| 346 | if ((c < '0' || c > '9') && c != '.' && c != '-' && c != '+' && c != 'e' && c != 'E') { |
| 347 | // Garbage found |
| 348 | return {}; |
| 349 | } |
| 350 | } |
| 351 | return { d, -processed }; |
| 352 | } |
| 353 | #endif // !defined(QT_NO_DOUBLECONVERSION) && !defined(QT_BOOTSTRAPPED) |
| 354 | |
| 355 | // Otherwise we would have gotten NaN or sorted it out above. |
| 356 | Q_ASSERT(strayCharMode == TrailingJunkAllowed || processed == numLen); |
| 357 | |
| 358 | // Check if underflow has occurred. |
| 359 | if (isZero(d)) { |
| 360 | for (int i = 0; i < processed; ++i) { |
| 361 | if (num[i] >= '1' && num[i] <= '9') { |
| 362 | // if a digit before any 'e' is not 0, then a non-zero number was intended. |
| 363 | return {.result: d, .used: -processed}; |
| 364 | } else if (num[i] == 'e' || num[i] == 'E') { |
| 365 | break; |
| 366 | } |
| 367 | } |
| 368 | } |
| 369 | return { .result: d, .used: processed }; |
| 370 | } |
| 371 | |
| 372 | /* Detect base if 0 and, if base is hex or bin, skip over 0x/0b prefixes */ |
| 373 | static auto scanPrefix(const char *p, const char *stop, int base) |
| 374 | { |
| 375 | struct R |
| 376 | { |
| 377 | const char *next; |
| 378 | int base; |
| 379 | }; |
| 380 | if (p < stop && isAsciiDigit(c: *p)) { |
| 381 | if (*p == '0') { |
| 382 | const char *x_or_b = p + 1; |
| 383 | if (x_or_b < stop) { |
| 384 | switch (*x_or_b) { |
| 385 | case 'b': |
| 386 | case 'B': |
| 387 | if (base == 0) |
| 388 | base = 2; |
| 389 | if (base == 2) |
| 390 | p += 2; |
| 391 | return R{.next: p, .base: base}; |
| 392 | case 'x': |
| 393 | case 'X': |
| 394 | if (base == 0) |
| 395 | base = 16; |
| 396 | if (base == 16) |
| 397 | p += 2; |
| 398 | return R{.next: p, .base: base}; |
| 399 | } |
| 400 | } |
| 401 | if (base == 0) |
| 402 | base = 8; |
| 403 | } else if (base == 0) { |
| 404 | base = 10; |
| 405 | } |
| 406 | Q_ASSERT(base); |
| 407 | } |
| 408 | return R{.next: p, .base: base}; |
| 409 | } |
| 410 | |
| 411 | static bool isDigitForBase(char d, int base) |
| 412 | { |
| 413 | if (d < '0') |
| 414 | return false; |
| 415 | if (d - '0' < qMin(a: base, b: 10)) |
| 416 | return true; |
| 417 | if (base > 10) { |
| 418 | d |= 0x20; // tolower |
| 419 | return d >= 'a' && d < 'a' + base - 10; |
| 420 | } |
| 421 | return false; |
| 422 | } |
| 423 | |
| 424 | QSimpleParsedNumber<qulonglong> qstrntoull(const char *begin, qsizetype size, int base) |
| 425 | { |
| 426 | const char *p = begin, *const stop = begin + size; |
| 427 | while (p < stop && ascii_isspace(c: *p)) |
| 428 | ++p; |
| 429 | unsigned long long result = 0; |
| 430 | if (p >= stop || *p == '-') |
| 431 | return { }; |
| 432 | const auto prefix = scanPrefix(p: *p == '+' ? p + 1 : p, stop, base); |
| 433 | if (!prefix.base || prefix.next >= stop) |
| 434 | return { }; |
| 435 | |
| 436 | const auto res = std::from_chars(first: prefix.next, last: stop, value&: result, base: prefix.base); |
| 437 | if (res.ec != std::errc{}) |
| 438 | return { }; |
| 439 | return { .result: result, .used: res.ptr == prefix.next ? 0 : res.ptr - begin }; |
| 440 | } |
| 441 | |
| 442 | QSimpleParsedNumber<qlonglong> qstrntoll(const char *begin, qsizetype size, int base) |
| 443 | { |
| 444 | const char *p = begin, *const stop = begin + size; |
| 445 | while (p < stop && ascii_isspace(c: *p)) |
| 446 | ++p; |
| 447 | // Frustratingly, std::from_chars() doesn't cope with a 0x prefix that might |
| 448 | // be between the sign and digits, so we have to handle that for it, which |
| 449 | // means we can't use its ability to read LLONG_MIN directly; see below. |
| 450 | const bool negate = p < stop && *p == '-'; |
| 451 | if (negate || (p < stop && *p == '+')) |
| 452 | ++p; |
| 453 | |
| 454 | const auto prefix = scanPrefix(p, stop, base); |
| 455 | // Must check for digit, as from_chars() will accept a sign, which would be |
| 456 | // a second sign, that we should reject. |
| 457 | if (!prefix.base || prefix.next >= stop || !isDigitForBase(d: *prefix.next, base: prefix.base)) |
| 458 | return { }; |
| 459 | |
| 460 | long long result = 0; |
| 461 | auto res = std::from_chars(first: prefix.next, last: stop, value&: result, base: prefix.base); |
| 462 | if (negate && res.ec == std::errc::result_out_of_range) { |
| 463 | // Maybe LLONG_MIN: |
| 464 | unsigned long long check = 0; |
| 465 | res = std::from_chars(first: prefix.next, last: stop, value&: check, base: prefix.base); |
| 466 | if (res.ec == std::errc{} && check + std::numeric_limits<long long>::min() == 0) |
| 467 | return { .result: std::numeric_limits<long long>::min(), .used: res.ptr - begin }; |
| 468 | return { }; |
| 469 | } |
| 470 | if (res.ec != std::errc{}) |
| 471 | return { }; |
| 472 | return { .result: negate ? -result : result, .used: res.ptr - begin }; |
| 473 | } |
| 474 | |
| 475 | template <typename Char> |
| 476 | static Q_ALWAYS_INLINE void qulltoString_helper(qulonglong number, int base, Char *&p) |
| 477 | { |
| 478 | // Performance-optimized code. Compiler can generate faster code when base is known. |
| 479 | switch (base) { |
| 480 | #define BIG_BASE_LOOP(b) \ |
| 481 | do { \ |
| 482 | const int r = number % b; \ |
| 483 | *--p = Char((r < 10 ? '0' : 'a' - 10) + r); \ |
| 484 | number /= b; \ |
| 485 | } while (number) |
| 486 | #ifndef __OPTIMIZE_SIZE__ |
| 487 | # define SMALL_BASE_LOOP(b) \ |
| 488 | do { \ |
| 489 | *--p = Char('0' + number % b); \ |
| 490 | number /= b; \ |
| 491 | } while (number) |
| 492 | |
| 493 | case 2: SMALL_BASE_LOOP(2); break; |
| 494 | case 8: SMALL_BASE_LOOP(8); break; |
| 495 | case 10: SMALL_BASE_LOOP(10); break; |
| 496 | case 16: BIG_BASE_LOOP(16); break; |
| 497 | #undef SMALL_BASE_LOOP |
| 498 | #endif |
| 499 | default: BIG_BASE_LOOP(base); break; |
| 500 | #undef BIG_BASE_LOOP |
| 501 | } |
| 502 | } |
| 503 | |
| 504 | // This is technically "qulonglong to ascii", but that name's taken |
| 505 | QString qulltoBasicLatin(qulonglong number, int base, bool negative) |
| 506 | { |
| 507 | if (number == 0) |
| 508 | return QStringLiteral("0" ); |
| 509 | // Length of MIN_LLONG with the sign in front is 65; we never need surrogate pairs. |
| 510 | // We do not need a terminator. |
| 511 | const unsigned maxlen = 65; |
| 512 | static_assert(CHAR_BIT * sizeof(number) + 1 <= maxlen); |
| 513 | Q_DECL_UNINITIALIZED char16_t buff[maxlen]; |
| 514 | char16_t *const end = buff + maxlen, *p = end; |
| 515 | |
| 516 | qulltoString_helper<char16_t>(number, base, p); |
| 517 | if (negative) |
| 518 | *--p = u'-'; |
| 519 | |
| 520 | return QString(reinterpret_cast<QChar *>(p), end - p); |
| 521 | } |
| 522 | |
| 523 | QString qulltoa(qulonglong number, int base, const QStringView zero) |
| 524 | { |
| 525 | // Length of MAX_ULLONG in base 2 is 64; and we may need a surrogate pair |
| 526 | // per digit. We do not need a terminator. |
| 527 | const unsigned maxlen = 128; |
| 528 | static_assert(CHAR_BIT * sizeof(number) <= maxlen); |
| 529 | Q_DECL_UNINITIALIZED char16_t buff[maxlen]; |
| 530 | char16_t *const end = buff + maxlen, *p = end; |
| 531 | |
| 532 | if (base != 10 || zero == u"0" ) { |
| 533 | qulltoString_helper<char16_t>(number, base, p); |
| 534 | } else if (zero.size() && !zero.at(n: 0).isSurrogate()) { |
| 535 | const char16_t zeroUcs2 = zero.at(n: 0).unicode(); |
| 536 | while (number != 0) { |
| 537 | *(--p) = unicodeForDigit(digit: number % base, zero: zeroUcs2); |
| 538 | |
| 539 | number /= base; |
| 540 | } |
| 541 | } else if (zero.size() == 2 && zero.at(n: 0).isHighSurrogate()) { |
| 542 | const char32_t zeroUcs4 = QChar::surrogateToUcs4(high: zero.at(n: 0), low: zero.at(n: 1)); |
| 543 | while (number != 0) { |
| 544 | const char32_t digit = unicodeForDigit(digit: number % base, zero: zeroUcs4); |
| 545 | |
| 546 | *(--p) = QChar::lowSurrogate(ucs4: digit); |
| 547 | *(--p) = QChar::highSurrogate(ucs4: digit); |
| 548 | |
| 549 | number /= base; |
| 550 | } |
| 551 | } else { // zero should always be either a non-surrogate or a surrogate pair: |
| 552 | Q_UNREACHABLE_RETURN(QString()); |
| 553 | } |
| 554 | |
| 555 | return QString(reinterpret_cast<QChar *>(p), end - p); |
| 556 | } |
| 557 | |
| 558 | char *qulltoa2(char *p, qulonglong n, int base) |
| 559 | { |
| 560 | #if defined(QT_CHECK_RANGE) |
| 561 | if (base < 2 || base > 36) { |
| 562 | qWarning("QByteArray::setNum: Invalid base %d" , base); |
| 563 | base = 10; |
| 564 | } |
| 565 | #endif |
| 566 | qulltoString_helper(number: n, base, p); |
| 567 | return p; |
| 568 | } |
| 569 | |
| 570 | /*! |
| 571 | \internal |
| 572 | |
| 573 | Converts the initial portion of the string pointed to by \a s00 to a double, |
| 574 | using the 'C' locale. The function sets the pointer pointed to by \a se to |
| 575 | point to the character past the last character converted. |
| 576 | */ |
| 577 | double qstrntod(const char *s00, qsizetype len, const char **se, bool *ok) |
| 578 | { |
| 579 | auto r = qt_asciiToDouble(num: s00, numLen: len, strayCharMode: TrailingJunkAllowed); |
| 580 | if (se) |
| 581 | *se = s00 + (r.used < 0 ? -r.used : r.used); |
| 582 | if (ok) |
| 583 | *ok = r.ok(); |
| 584 | return r.result; |
| 585 | } |
| 586 | |
| 587 | QString qdtoa(qreal d, int *decpt, int *sign) |
| 588 | { |
| 589 | bool nonNullSign = false; |
| 590 | int nonNullDecpt = 0; |
| 591 | int length = 0; |
| 592 | |
| 593 | // Some versions of libdouble-conversion like an extra digit, probably for '\0' |
| 594 | constexpr qsizetype digits = std::numeric_limits<double>::max_digits10 + 1; |
| 595 | char result[digits]; |
| 596 | qt_doubleToAscii(d, form: QLocaleData::DFSignificantDigits, precision: QLocale::FloatingPointShortest, |
| 597 | buf: result, bufSize: digits, sign&: nonNullSign, length, decpt&: nonNullDecpt); |
| 598 | |
| 599 | if (sign) |
| 600 | *sign = nonNullSign ? 1 : 0; |
| 601 | if (decpt) |
| 602 | *decpt = nonNullDecpt; |
| 603 | |
| 604 | return QLatin1StringView(result, length); |
| 605 | } |
| 606 | |
| 607 | static QLocaleData::DoubleForm resolveFormat(int precision, int decpt, qsizetype length) |
| 608 | { |
| 609 | bool useDecimal; |
| 610 | if (precision == QLocale::FloatingPointShortest) { |
| 611 | // Find out which representation is shorter. |
| 612 | // Set bias to everything added to exponent form but not |
| 613 | // decimal, minus the converse. |
| 614 | |
| 615 | // Exponent adds separator, sign and two exponents: |
| 616 | int bias = 2 + 2; |
| 617 | if (length <= decpt && length > 1) |
| 618 | ++bias; |
| 619 | else if (length == 1 && decpt <= 0) |
| 620 | --bias; |
| 621 | |
| 622 | // When 0 < decpt <= length, the forms have equal digit |
| 623 | // counts, plus things bias has taken into account; |
| 624 | // otherwise decimal form's digit count is right-padded with |
| 625 | // zeros to decpt, when decpt is positive, otherwise it's |
| 626 | // left-padded with 1 - decpt zeros. |
| 627 | if (decpt <= 0) |
| 628 | useDecimal = 1 - decpt <= bias; |
| 629 | else if (decpt <= length) |
| 630 | useDecimal = true; |
| 631 | else |
| 632 | useDecimal = decpt <= length + bias; |
| 633 | } else { |
| 634 | // X == decpt - 1, POSIX's P; -4 <= X < P iff -4 < decpt <= P |
| 635 | Q_ASSERT(precision >= 0); |
| 636 | useDecimal = decpt > -4 && decpt <= (precision ? precision : 1); |
| 637 | } |
| 638 | return useDecimal ? QLocaleData::DFDecimal : QLocaleData::DFExponent; |
| 639 | } |
| 640 | |
| 641 | static constexpr int digits(int number) |
| 642 | { |
| 643 | Q_ASSERT(number >= 0); |
| 644 | if (Q_LIKELY(number < 1000)) |
| 645 | return number < 10 ? 1 : number < 100 ? 2 : 3; |
| 646 | int i = 3; |
| 647 | for (number /= 1000; number; number /= 10) |
| 648 | ++i; |
| 649 | return i; |
| 650 | } |
| 651 | |
| 652 | // Used generically for both QString and QByteArray |
| 653 | template <typename T> |
| 654 | static T dtoString(double d, QLocaleData::DoubleForm form, int precision, bool uppercase) |
| 655 | { |
| 656 | // Undocumented: aside from F.P.Shortest, precision < 0 is treated as |
| 657 | // default, 6 - same as printf(). |
| 658 | if (precision != QLocale::FloatingPointShortest && precision < 0) |
| 659 | precision = 6; |
| 660 | |
| 661 | using D = std::numeric_limits<double>; |
| 662 | // 1 is for the null-terminator |
| 663 | constexpr int MaxDigits = 1 + qMax(a: D::max_exponent10, b: D::digits10 - D::min_exponent10); |
| 664 | |
| 665 | // "maxDigits" above is a reasonable estimate, though we may need more due to extra precision |
| 666 | int bufSize = 1; |
| 667 | if (precision == QLocale::FloatingPointShortest) |
| 668 | bufSize += D::max_digits10; |
| 669 | else if (form == QLocaleData::DFDecimal && qt_is_finite(d)) |
| 670 | bufSize += wholePartSpace(d: qAbs(t: d)) + precision; |
| 671 | else // Add extra digit due to different interpretations of precision. |
| 672 | bufSize += qMax(a: 2, b: precision) + 1; // Must also be big enough for "nan" or "inf" |
| 673 | |
| 674 | // Reserve `MaxDigits` on the stack, which is a reasonable estimate; |
| 675 | // but we may need more due to extra precision, which we cannot know at compile-time. |
| 676 | QVarLengthArray<char, MaxDigits> buffer(bufSize); |
| 677 | bool negative = false; |
| 678 | int length = 0; |
| 679 | int decpt = 0; |
| 680 | qt_doubleToAscii(d, form, precision, buf: buffer.data(), bufSize: buffer.size(), sign&: negative, length, decpt); |
| 681 | QLatin1StringView view(buffer.data(), length); |
| 682 | const bool succinct = form == QLocaleData::DFSignificantDigits; |
| 683 | qsizetype total = (negative ? 1 : 0) + length; |
| 684 | if (qt_is_finite(d)) { |
| 685 | if (succinct) |
| 686 | form = resolveFormat(precision, decpt, length: view.size()); |
| 687 | |
| 688 | switch (form) { |
| 689 | case QLocaleData::DFExponent: |
| 690 | total += 3; // (.e+) The '.' may not be needed, but we would only overestimate by 1 char |
| 691 | // Exponents: we guarantee at least 2 |
| 692 | total += std::max(a: 2, b: digits(number: std::abs(x: decpt - 1))); |
| 693 | // "length - 1" because one of the digits will always be before the decimal point |
| 694 | if (int = precision - (length - 1); extraPrecision > 0 && !succinct) |
| 695 | total += extraPrecision; // some requested zero-padding |
| 696 | break; |
| 697 | case QLocaleData::DFDecimal: |
| 698 | if (decpt <= 0) // leading "0." and zeros |
| 699 | total += 2 - decpt; |
| 700 | else if (decpt < length) // just the dot |
| 701 | total += 1; |
| 702 | else // trailing zeros (and no dot, unless we require extra precision): |
| 703 | total += decpt - length; |
| 704 | |
| 705 | if (precision > 0 && !succinct) { |
| 706 | // May need trailing zeros to satisfy precision: |
| 707 | if (decpt < length) |
| 708 | total += std::max(a: 0, b: precision - length + decpt); |
| 709 | else // and a dot to separate them: |
| 710 | total += 1 + precision; |
| 711 | } |
| 712 | break; |
| 713 | case QLocaleData::DFSignificantDigits: |
| 714 | Q_UNREACHABLE(); // Handled earlier |
| 715 | } |
| 716 | } |
| 717 | |
| 718 | constexpr bool IsQString = std::is_same_v<T, QString>; |
| 719 | using Char = std::conditional_t<IsQString, char16_t, char>; |
| 720 | |
| 721 | T result; |
| 722 | result.reserve(total); |
| 723 | |
| 724 | if (negative && !isZero(d)) // We don't return "-0" |
| 725 | result.append(Char('-')); |
| 726 | if (!qt_is_finite(d)) { |
| 727 | result.append(view); |
| 728 | if (uppercase) |
| 729 | result = std::move(result).toUpper(); |
| 730 | } else { |
| 731 | switch (form) { |
| 732 | case QLocaleData::DFExponent: { |
| 733 | result.append(view.first(n: 1)); |
| 734 | view = view.sliced(pos: 1); |
| 735 | if (!view.isEmpty() || (!succinct && precision > 0)) { |
| 736 | result.append(Char('.')); |
| 737 | result.append(view); |
| 738 | if (qsizetype pad = precision - view.size(); !succinct && pad > 0) { |
| 739 | for (int i = 0; i < pad; ++i) |
| 740 | result.append(Char('0')); |
| 741 | } |
| 742 | } |
| 743 | int exponent = decpt - 1; |
| 744 | result.append(Char(uppercase ? 'E' : 'e')); |
| 745 | result.append(Char(exponent < 0 ? '-' : '+')); |
| 746 | exponent = std::abs(x: exponent); |
| 747 | Q_ASSERT(exponent <= D::max_exponent10 + D::max_digits10); |
| 748 | int exponentDigits = digits(number: exponent); |
| 749 | // C's printf guarantees a two-digit exponent, and so do we: |
| 750 | if (exponentDigits == 1) |
| 751 | result.append(Char('0')); |
| 752 | result.resize(result.size() + exponentDigits); |
| 753 | auto location = reinterpret_cast<Char *>(result.end()); |
| 754 | qulltoString_helper<Char>(exponent, 10, location); |
| 755 | break; |
| 756 | } |
| 757 | case QLocaleData::DFDecimal: |
| 758 | if (decpt < 0) { |
| 759 | if constexpr (IsQString) |
| 760 | result.append(u"0.0" ); |
| 761 | else |
| 762 | result.append("0.0" ); |
| 763 | while (++decpt < 0) |
| 764 | result.append(Char('0')); |
| 765 | result.append(view); |
| 766 | if (!succinct) { |
| 767 | auto numDecimals = result.size() - 2 - (negative ? 1 : 0); |
| 768 | for (qsizetype i = numDecimals; i < precision; ++i) |
| 769 | result.append(Char('0')); |
| 770 | } |
| 771 | } else { |
| 772 | if (decpt > view.size()) { |
| 773 | result.append(view); |
| 774 | const int sign = negative ? 1 : 0; |
| 775 | while (result.size() - sign < decpt) |
| 776 | result.append(Char('0')); |
| 777 | view = {}; |
| 778 | } else if (decpt) { |
| 779 | result.append(view.first(n: decpt)); |
| 780 | view = view.sliced(pos: decpt); |
| 781 | } else { |
| 782 | result.append(Char('0')); |
| 783 | } |
| 784 | if (!view.isEmpty() || (!succinct && view.size() < precision)) { |
| 785 | result.append(Char('.')); |
| 786 | result.append(view); |
| 787 | if (!succinct) { |
| 788 | for (qsizetype i = view.size(); i < precision; ++i) |
| 789 | result.append(Char('0')); |
| 790 | } |
| 791 | } |
| 792 | } |
| 793 | break; |
| 794 | case QLocaleData::DFSignificantDigits: |
| 795 | Q_UNREACHABLE(); // taken care of earlier |
| 796 | break; |
| 797 | } |
| 798 | } |
| 799 | Q_ASSERT(total >= result.size()); // No reallocations are needed |
| 800 | return result; |
| 801 | } |
| 802 | |
| 803 | QString qdtoBasicLatin(double d, QLocaleData::DoubleForm form, int precision, bool uppercase) |
| 804 | { |
| 805 | return dtoString<QString>(d, form, precision, uppercase); |
| 806 | } |
| 807 | |
| 808 | QByteArray qdtoAscii(double d, QLocaleData::DoubleForm form, int precision, bool uppercase) |
| 809 | { |
| 810 | return dtoString<QByteArray>(d, form, precision, uppercase); |
| 811 | } |
| 812 | |
| 813 | #if defined(QT_SUPPORTS_INT128) || defined(QT_USE_MSVC_INT128) |
| 814 | static inline quint64 toUInt64(qinternaluint128 v) |
| 815 | { |
| 816 | #if defined(QT_USE_MSVC_INT128) |
| 817 | return quint64(v._Word[0]); |
| 818 | #else |
| 819 | return quint64(v); |
| 820 | #endif |
| 821 | } |
| 822 | |
| 823 | QString quint128toBasicLatin(qinternaluint128 number, int base) |
| 824 | { |
| 825 | // We divide our 128-bit number into parts that we can do text |
| 826 | // concatenation with. This list is the maximum power of the |
| 827 | // base that is less than 2^64. |
| 828 | static constexpr auto dividers = []() constexpr { |
| 829 | std::array<quint64, 35> bases {}; |
| 830 | for (int base = 2; base <= 36; ++base) { |
| 831 | quint64 v = base; |
| 832 | while (v * base > v) |
| 833 | v *= base; |
| 834 | bases[base - 2] = v; |
| 835 | } |
| 836 | return bases; |
| 837 | }(); |
| 838 | static constexpr auto digitCounts = []() constexpr { |
| 839 | std::array<quint8, 35> digits{}; |
| 840 | for (int base = 2; base <= 36; ++base) { |
| 841 | quint64 v = base; |
| 842 | int i = 0; |
| 843 | for (i = 0; v * base > v; ++i) |
| 844 | v *= base; |
| 845 | digits[base - 2] = i; |
| 846 | } |
| 847 | return digits; |
| 848 | }(); |
| 849 | |
| 850 | QString result; |
| 851 | |
| 852 | constexpr unsigned flags = QLocaleData::NoFlags; |
| 853 | const QLocaleData *dd = QLocaleData::c(); |
| 854 | |
| 855 | // special base cases: |
| 856 | constexpr int Width = -1; |
| 857 | if (base == 2 || base == 4 || base == 16) { |
| 858 | // 2^64 is a power of 2, 4 and 16 |
| 859 | result = dd->unsLongLongToString(l: quint64(number), precision: 64, base, width: Width, flags); |
| 860 | result.prepend(s: dd->unsLongLongToString(l: quint64(number >> 64), precision: -1, base, width: Width, flags)); |
| 861 | } else { |
| 862 | int digitCount = digitCounts[base - 2]; |
| 863 | quint64 divider = dividers[base - 2]; |
| 864 | quint64 lower = toUInt64(v: number % divider); |
| 865 | number /= divider; |
| 866 | while (number) { |
| 867 | result.prepend(s: dd->unsLongLongToString(l: lower, precision: digitCount, base, width: Width, flags)); |
| 868 | lower = toUInt64(v: number % divider); |
| 869 | number /= divider; |
| 870 | } |
| 871 | result.prepend(s: dd->unsLongLongToString(l: lower, precision: -1, base, width: Width, flags)); |
| 872 | } |
| 873 | return result; |
| 874 | } |
| 875 | |
| 876 | QString qint128toBasicLatin(qinternalint128 number, int base) |
| 877 | { |
| 878 | const bool negative = number < 0; |
| 879 | if (negative) |
| 880 | number *= -1; |
| 881 | QString result = quint128toBasicLatin(number: qinternaluint128(number), base); |
| 882 | if (negative) |
| 883 | result.prepend(c: u'-'); |
| 884 | return result; |
| 885 | } |
| 886 | #endif // defined(QT_SUPPORTS_INT128) || defined(QT_USE_MSVC_INT128) |
| 887 | |
| 888 | QT_END_NAMESPACE |
| 889 | |