1 //===-- runtime/edit-input.cpp --------------------------------------------===// 2 // 3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. 4 // See https://llvm.org/LICENSE.txt for license information. 5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception 6 // 7 //===----------------------------------------------------------------------===// 8 9 #include "edit-input.h" 10 #include "namelist.h" 11 #include "utf.h" 12 #include "flang/Common/real.h" 13 #include "flang/Common/uint128.h" 14 #include <algorithm> 15 #include <cfenv> 16 17 namespace Fortran::runtime::io { 18 19 template <int LOG2_BASE> 20 static bool EditBOZInput( 21 IoStatementState &io, const DataEdit &edit, void *n, std::size_t bytes) { 22 std::optional<int> remaining; 23 std::optional<char32_t> next{io.PrepareInput(edit, remaining)}; 24 if (*next == '0') { 25 do { 26 next = io.NextInField(remaining, edit); 27 } while (next && *next == '0'); 28 } 29 // Count significant digits after any leading white space & zeroes 30 int digits{0}; 31 for (; next; next = io.NextInField(remaining, edit)) { 32 char32_t ch{*next}; 33 if (ch == ' ' || ch == '\t') { 34 continue; 35 } 36 if (ch >= '0' && ch <= '1') { 37 } else if (LOG2_BASE >= 3 && ch >= '2' && ch <= '7') { 38 } else if (LOG2_BASE >= 4 && ch >= '8' && ch <= '9') { 39 } else if (LOG2_BASE >= 4 && ch >= 'A' && ch <= 'F') { 40 } else if (LOG2_BASE >= 4 && ch >= 'a' && ch <= 'f') { 41 } else { 42 io.GetIoErrorHandler().SignalError( 43 "Bad character '%lc' in B/O/Z input field", ch); 44 return false; 45 } 46 ++digits; 47 } 48 auto significantBytes{static_cast<std::size_t>(digits * LOG2_BASE + 7) / 8}; 49 if (significantBytes > bytes) { 50 io.GetIoErrorHandler().SignalError( 51 "B/O/Z input of %d digits overflows %zd-byte variable", digits, bytes); 52 return false; 53 } 54 // Reset to start of significant digits 55 io.HandleRelativePosition(-digits); 56 remaining.reset(); 57 // Make a second pass now that the digit count is known 58 std::memset(n, 0, bytes); 59 int increment{isHostLittleEndian ? -1 : 1}; 60 auto *data{reinterpret_cast<unsigned char *>(n) + 61 (isHostLittleEndian ? significantBytes - 1 : 0)}; 62 int shift{((digits - 1) * LOG2_BASE) & 7}; 63 if (shift + LOG2_BASE > 8) { 64 shift -= 8; // misaligned octal 65 } 66 while (digits > 0) { 67 char32_t ch{*io.NextInField(remaining, edit)}; 68 int digit{0}; 69 if (ch >= '0' && ch <= '9') { 70 digit = ch - '0'; 71 } else if (ch >= 'A' && ch <= 'F') { 72 digit = ch + 10 - 'A'; 73 } else if (ch >= 'a' && ch <= 'f') { 74 digit = ch + 10 - 'a'; 75 } else { 76 continue; 77 } 78 --digits; 79 if (shift < 0) { 80 shift += 8; 81 if (shift + LOG2_BASE > 8) { // misaligned octal 82 *data |= digit >> (8 - shift); 83 } 84 data += increment; 85 } 86 *data |= digit << shift; 87 shift -= LOG2_BASE; 88 } 89 return true; 90 } 91 92 static inline char32_t GetDecimalPoint(const DataEdit &edit) { 93 return edit.modes.editingFlags & decimalComma ? char32_t{','} : char32_t{'.'}; 94 } 95 96 // Prepares input from a field, and consumes the sign, if any. 97 // Returns true if there's a '-' sign. 98 static bool ScanNumericPrefix(IoStatementState &io, const DataEdit &edit, 99 std::optional<char32_t> &next, std::optional<int> &remaining) { 100 next = io.PrepareInput(edit, remaining); 101 bool negative{false}; 102 if (next) { 103 negative = *next == '-'; 104 if (negative || *next == '+') { 105 io.SkipSpaces(remaining); 106 next = io.NextInField(remaining, edit); 107 } 108 } 109 return negative; 110 } 111 112 bool EditIntegerInput( 113 IoStatementState &io, const DataEdit &edit, void *n, int kind) { 114 RUNTIME_CHECK(io.GetIoErrorHandler(), kind >= 1 && !(kind & (kind - 1))); 115 switch (edit.descriptor) { 116 case DataEdit::ListDirected: 117 if (IsNamelistName(io)) { 118 return false; 119 } 120 break; 121 case 'G': 122 case 'I': 123 break; 124 case 'B': 125 return EditBOZInput<1>(io, edit, n, kind); 126 case 'O': 127 return EditBOZInput<3>(io, edit, n, kind); 128 case 'Z': 129 return EditBOZInput<4>(io, edit, n, kind); 130 case 'A': // legacy extension 131 return EditCharacterInput(io, edit, reinterpret_cast<char *>(n), kind); 132 default: 133 io.GetIoErrorHandler().SignalError(IostatErrorInFormat, 134 "Data edit descriptor '%c' may not be used with an INTEGER data item", 135 edit.descriptor); 136 return false; 137 } 138 std::optional<int> remaining; 139 std::optional<char32_t> next; 140 bool negate{ScanNumericPrefix(io, edit, next, remaining)}; 141 common::UnsignedInt128 value{0}; 142 bool any{negate}; 143 for (; next; next = io.NextInField(remaining, edit)) { 144 char32_t ch{*next}; 145 if (ch == ' ' || ch == '\t') { 146 if (edit.modes.editingFlags & blankZero) { 147 ch = '0'; // BZ mode - treat blank as if it were zero 148 } else { 149 continue; 150 } 151 } 152 int digit{0}; 153 if (ch >= '0' && ch <= '9') { 154 digit = ch - '0'; 155 } else { 156 io.GetIoErrorHandler().SignalError( 157 "Bad character '%lc' in INTEGER input field", ch); 158 return false; 159 } 160 value *= 10; 161 value += digit; 162 any = true; 163 } 164 if (negate) { 165 value = -value; 166 } 167 if (any || !io.GetConnectionState().IsAtEOF()) { 168 std::memcpy(n, &value, kind); // a blank field means zero 169 } 170 return any; 171 } 172 173 // Parses a REAL input number from the input source as a normalized 174 // fraction into a supplied buffer -- there's an optional '-', a 175 // decimal point, and at least one digit. The adjusted exponent value 176 // is returned in a reference argument. The returned value is the number 177 // of characters that (should) have been written to the buffer -- this can 178 // be larger than the buffer size and can indicate overflow. Replaces 179 // blanks with zeroes if appropriate. 180 static int ScanRealInput(char *buffer, int bufferSize, IoStatementState &io, 181 const DataEdit &edit, int &exponent) { 182 std::optional<int> remaining; 183 std::optional<char32_t> next; 184 int got{0}; 185 std::optional<int> decimalPoint; 186 auto Put{[&](char ch) -> void { 187 if (got < bufferSize) { 188 buffer[got] = ch; 189 } 190 ++got; 191 }}; 192 if (ScanNumericPrefix(io, edit, next, remaining)) { 193 Put('-'); 194 } 195 bool bzMode{(edit.modes.editingFlags & blankZero) != 0}; 196 if (!next || (!bzMode && *next == ' ')) { // empty/blank field means zero 197 remaining.reset(); 198 if (!io.GetConnectionState().IsAtEOF()) { 199 Put('0'); 200 } 201 return got; 202 } 203 char32_t decimal{GetDecimalPoint(edit)}; 204 char32_t first{*next >= 'a' && *next <= 'z' ? *next + 'A' - 'a' : *next}; 205 if (first == 'N' || first == 'I') { 206 // NaN or infinity - convert to upper case 207 // Subtle: a blank field of digits could be followed by 'E' or 'D', 208 for (; next && 209 ((*next >= 'a' && *next <= 'z') || (*next >= 'A' && *next <= 'Z')); 210 next = io.NextInField(remaining, edit)) { 211 if (*next >= 'a' && *next <= 'z') { 212 Put(*next - 'a' + 'A'); 213 } else { 214 Put(*next); 215 } 216 } 217 if (next && *next == '(') { // NaN(...) 218 Put('('); 219 int depth{1}; 220 do { 221 next = io.NextInField(remaining, edit); 222 if (!next) { 223 break; 224 } else if (*next == '(') { 225 ++depth; 226 } else if (*next == ')') { 227 --depth; 228 } 229 Put(*next); 230 } while (depth > 0); 231 } 232 exponent = 0; 233 } else if (first == decimal || (first >= '0' && first <= '9') || 234 (bzMode && (first == ' ' || first == '\t')) || first == 'E' || 235 first == 'D' || first == 'Q') { 236 Put('.'); // input field is normalized to a fraction 237 auto start{got}; 238 for (; next; next = io.NextInField(remaining, edit)) { 239 char32_t ch{*next}; 240 if (ch == ' ' || ch == '\t') { 241 if (bzMode) { 242 ch = '0'; // BZ mode - treat blank as if it were zero 243 } else { 244 continue; 245 } 246 } 247 if (ch == '0' && got == start && !decimalPoint) { 248 // omit leading zeroes before the decimal 249 } else if (ch >= '0' && ch <= '9') { 250 Put(ch); 251 } else if (ch == decimal && !decimalPoint) { 252 // the decimal point is *not* copied to the buffer 253 decimalPoint = got - start; // # of digits before the decimal point 254 } else { 255 break; 256 } 257 } 258 if (got == start) { 259 // Nothing but zeroes and maybe a decimal point. F'2018 requires 260 // at least one digit, but F'77 did not, and a bare "." shows up in 261 // the FCVS suite. 262 Put('0'); // emit at least one digit 263 } 264 if (next && 265 (*next == 'e' || *next == 'E' || *next == 'd' || *next == 'D' || 266 *next == 'q' || *next == 'Q')) { 267 // Optional exponent letter. Blanks are allowed between the 268 // optional exponent letter and the exponent value. 269 io.SkipSpaces(remaining); 270 next = io.NextInField(remaining, edit); 271 } 272 // The default exponent is -kP, but the scale factor doesn't affect 273 // an explicit exponent. 274 exponent = -edit.modes.scale; 275 if (next && 276 (*next == '-' || *next == '+' || (*next >= '0' && *next <= '9') || 277 *next == ' ' || *next == '\t')) { 278 bool negExpo{*next == '-'}; 279 if (negExpo || *next == '+') { 280 next = io.NextInField(remaining, edit); 281 } 282 for (exponent = 0; next; next = io.NextInField(remaining, edit)) { 283 if (*next >= '0' && *next <= '9') { 284 exponent = 10 * exponent + *next - '0'; 285 } else if (*next == ' ' || *next == '\t') { 286 if (bzMode) { 287 exponent = 10 * exponent; 288 } 289 } else { 290 break; 291 } 292 } 293 if (negExpo) { 294 exponent = -exponent; 295 } 296 } 297 if (decimalPoint) { 298 exponent += *decimalPoint; 299 } else { 300 // When no decimal point (or comma) appears in the value, the 'd' 301 // part of the edit descriptor must be interpreted as the number of 302 // digits in the value to be interpreted as being to the *right* of 303 // the assumed decimal point (13.7.2.3.2) 304 exponent += got - start - edit.digits.value_or(0); 305 } 306 } else { 307 // TODO: hex FP input 308 exponent = 0; 309 return 0; 310 } 311 // Consume the trailing ')' of a list-directed or NAMELIST complex 312 // input value. 313 if (edit.descriptor == DataEdit::ListDirectedImaginaryPart) { 314 if (next && (*next == ' ' || *next == '\t')) { 315 next = io.NextInField(remaining, edit); 316 } 317 if (!next) { // NextInField fails on separators like ')' 318 std::size_t byteCount{0}; 319 next = io.GetCurrentChar(byteCount); 320 if (next && *next == ')') { 321 io.HandleRelativePosition(byteCount); 322 } 323 } 324 } else if (remaining) { 325 while (next && (*next == ' ' || *next == '\t')) { 326 next = io.NextInField(remaining, edit); 327 } 328 if (next) { 329 return 0; // error: unused nonblank character in fixed-width field 330 } 331 } 332 return got; 333 } 334 335 static void RaiseFPExceptions(decimal::ConversionResultFlags flags) { 336 #undef RAISE 337 #ifdef feraisexcept // a macro in some environments; omit std:: 338 #define RAISE feraiseexcept 339 #else 340 #define RAISE std::feraiseexcept 341 #endif 342 if (flags & decimal::ConversionResultFlags::Overflow) { 343 RAISE(FE_OVERFLOW); 344 } 345 if (flags & decimal::ConversionResultFlags::Inexact) { 346 RAISE(FE_INEXACT); 347 } 348 if (flags & decimal::ConversionResultFlags::Invalid) { 349 RAISE(FE_INVALID); 350 } 351 #undef RAISE 352 } 353 354 // If no special modes are in effect and the form of the input value 355 // that's present in the input stream is acceptable to the decimal->binary 356 // converter without modification, this fast path for real input 357 // saves time by avoiding memory copies and reformatting of the exponent. 358 template <int PRECISION> 359 static bool TryFastPathRealInput( 360 IoStatementState &io, const DataEdit &edit, void *n) { 361 if (edit.modes.editingFlags & (blankZero | decimalComma)) { 362 return false; 363 } 364 if (edit.modes.scale != 0) { 365 return false; 366 } 367 const char *str{nullptr}; 368 std::size_t got{io.GetNextInputBytes(str)}; 369 if (got == 0 || str == nullptr || 370 !io.GetConnectionState().recordLength.has_value()) { 371 return false; // could not access reliably-terminated input stream 372 } 373 const char *p{str}; 374 std::int64_t maxConsume{ 375 std::min<std::int64_t>(got, edit.width.value_or(got))}; 376 const char *limit{str + maxConsume}; 377 decimal::ConversionToBinaryResult<PRECISION> converted{ 378 decimal::ConvertToBinary<PRECISION>(p, edit.modes.round, limit)}; 379 if (converted.flags & decimal::Invalid) { 380 return false; 381 } 382 if (edit.digits.value_or(0) != 0) { 383 // Edit descriptor is Fw.d (or other) with d != 0, which 384 // implies scaling 385 const char *q{str}; 386 for (; q < limit; ++q) { 387 if (*q == '.' || *q == 'n' || *q == 'N') { 388 break; 389 } 390 } 391 if (q == limit) { 392 // No explicit decimal point, and not NaN/Inf. 393 return false; 394 } 395 } 396 for (; p < limit && (*p == ' ' || *p == '\t'); ++p) { 397 } 398 if (edit.descriptor == DataEdit::ListDirectedImaginaryPart) { 399 // Need to consume a trailing ')' and any white space after 400 if (p >= limit || *p != ')') { 401 return false; 402 } 403 for (++p; p < limit && (*p == ' ' || *p == '\t'); ++p) { 404 } 405 } 406 if (edit.width && p < str + *edit.width) { 407 return false; // unconverted characters remain in fixed width field 408 } 409 // Success on the fast path! 410 *reinterpret_cast<decimal::BinaryFloatingPointNumber<PRECISION> *>(n) = 411 converted.binary; 412 io.HandleRelativePosition(p - str); 413 // Set FP exception flags 414 if (converted.flags != decimal::ConversionResultFlags::Exact) { 415 RaiseFPExceptions(converted.flags); 416 } 417 return true; 418 } 419 420 template <int KIND> 421 bool EditCommonRealInput(IoStatementState &io, const DataEdit &edit, void *n) { 422 constexpr int binaryPrecision{common::PrecisionOfRealKind(KIND)}; 423 if (TryFastPathRealInput<binaryPrecision>(io, edit, n)) { 424 return true; 425 } 426 // Fast path wasn't available or didn't work; go the more general route 427 static constexpr int maxDigits{ 428 common::MaxDecimalConversionDigits(binaryPrecision)}; 429 static constexpr int bufferSize{maxDigits + 18}; 430 char buffer[bufferSize]; 431 int exponent{0}; 432 int got{ScanRealInput(buffer, maxDigits + 2, io, edit, exponent)}; 433 if (got >= maxDigits + 2) { 434 io.GetIoErrorHandler().Crash("EditCommonRealInput: buffer was too small"); 435 return false; 436 } 437 if (got == 0) { 438 io.GetIoErrorHandler().SignalError(IostatBadRealInput); 439 return false; 440 } 441 bool hadExtra{got > maxDigits}; 442 if (exponent != 0) { 443 buffer[got++] = 'e'; 444 if (exponent < 0) { 445 buffer[got++] = '-'; 446 exponent = -exponent; 447 } 448 if (exponent > 9999) { 449 exponent = 9999; // will convert to +/-Inf 450 } 451 if (exponent > 999) { 452 int dig{exponent / 1000}; 453 buffer[got++] = '0' + dig; 454 int rest{exponent - 1000 * dig}; 455 dig = rest / 100; 456 buffer[got++] = '0' + dig; 457 rest -= 100 * dig; 458 dig = rest / 10; 459 buffer[got++] = '0' + dig; 460 buffer[got++] = '0' + (rest - 10 * dig); 461 } else if (exponent > 99) { 462 int dig{exponent / 100}; 463 buffer[got++] = '0' + dig; 464 int rest{exponent - 100 * dig}; 465 dig = rest / 10; 466 buffer[got++] = '0' + dig; 467 buffer[got++] = '0' + (rest - 10 * dig); 468 } else if (exponent > 9) { 469 int dig{exponent / 10}; 470 buffer[got++] = '0' + dig; 471 buffer[got++] = '0' + (exponent - 10 * dig); 472 } else { 473 buffer[got++] = '0' + exponent; 474 } 475 } 476 buffer[got] = '\0'; 477 const char *p{buffer}; 478 decimal::ConversionToBinaryResult<binaryPrecision> converted{ 479 decimal::ConvertToBinary<binaryPrecision>(p, edit.modes.round)}; 480 if (hadExtra) { 481 converted.flags = static_cast<enum decimal::ConversionResultFlags>( 482 converted.flags | decimal::Inexact); 483 } 484 if (*p) { // unprocessed junk after value 485 io.GetIoErrorHandler().SignalError(IostatBadRealInput); 486 return false; 487 } 488 *reinterpret_cast<decimal::BinaryFloatingPointNumber<binaryPrecision> *>(n) = 489 converted.binary; 490 // Set FP exception flags 491 if (converted.flags != decimal::ConversionResultFlags::Exact) { 492 RaiseFPExceptions(converted.flags); 493 } 494 return true; 495 } 496 497 template <int KIND> 498 bool EditRealInput(IoStatementState &io, const DataEdit &edit, void *n) { 499 switch (edit.descriptor) { 500 case DataEdit::ListDirected: 501 if (IsNamelistName(io)) { 502 return false; 503 } 504 return EditCommonRealInput<KIND>(io, edit, n); 505 case DataEdit::ListDirectedRealPart: 506 case DataEdit::ListDirectedImaginaryPart: 507 case 'F': 508 case 'E': // incl. EN, ES, & EX 509 case 'D': 510 case 'G': 511 return EditCommonRealInput<KIND>(io, edit, n); 512 case 'B': 513 return EditBOZInput<1>(io, edit, n, 514 common::BitsForBinaryPrecision(common::PrecisionOfRealKind(KIND)) >> 3); 515 case 'O': 516 return EditBOZInput<3>(io, edit, n, 517 common::BitsForBinaryPrecision(common::PrecisionOfRealKind(KIND)) >> 3); 518 case 'Z': 519 return EditBOZInput<4>(io, edit, n, 520 common::BitsForBinaryPrecision(common::PrecisionOfRealKind(KIND)) >> 3); 521 case 'A': // legacy extension 522 return EditCharacterInput(io, edit, reinterpret_cast<char *>(n), KIND); 523 default: 524 io.GetIoErrorHandler().SignalError(IostatErrorInFormat, 525 "Data edit descriptor '%c' may not be used for REAL input", 526 edit.descriptor); 527 return false; 528 } 529 } 530 531 // 13.7.3 in Fortran 2018 532 bool EditLogicalInput(IoStatementState &io, const DataEdit &edit, bool &x) { 533 switch (edit.descriptor) { 534 case DataEdit::ListDirected: 535 if (IsNamelistName(io)) { 536 return false; 537 } 538 break; 539 case 'L': 540 case 'G': 541 break; 542 default: 543 io.GetIoErrorHandler().SignalError(IostatErrorInFormat, 544 "Data edit descriptor '%c' may not be used for LOGICAL input", 545 edit.descriptor); 546 return false; 547 } 548 std::optional<int> remaining; 549 std::optional<char32_t> next{io.PrepareInput(edit, remaining)}; 550 if (next && *next == '.') { // skip optional period 551 next = io.NextInField(remaining, edit); 552 } 553 if (!next) { 554 io.GetIoErrorHandler().SignalError("Empty LOGICAL input field"); 555 return false; 556 } 557 switch (*next) { 558 case 'T': 559 case 't': 560 x = true; 561 break; 562 case 'F': 563 case 'f': 564 x = false; 565 break; 566 default: 567 io.GetIoErrorHandler().SignalError( 568 "Bad character '%lc' in LOGICAL input field", *next); 569 return false; 570 } 571 if (remaining) { // ignore the rest of the field 572 io.HandleRelativePosition(*remaining); 573 } else if (edit.descriptor == DataEdit::ListDirected) { 574 while (io.NextInField(remaining, edit)) { // discard rest of field 575 } 576 } 577 return true; 578 } 579 580 // See 13.10.3.1 paragraphs 7-9 in Fortran 2018 581 template <typename CHAR> 582 static bool EditDelimitedCharacterInput( 583 IoStatementState &io, CHAR *x, std::size_t length, char32_t delimiter) { 584 bool result{true}; 585 while (true) { 586 std::size_t byteCount{0}; 587 auto ch{io.GetCurrentChar(byteCount)}; 588 if (!ch) { 589 if (io.AdvanceRecord()) { 590 continue; 591 } else { 592 result = false; // EOF in character value 593 break; 594 } 595 } 596 io.HandleRelativePosition(byteCount); 597 if (*ch == delimiter) { 598 auto next{io.GetCurrentChar(byteCount)}; 599 if (next && *next == delimiter) { 600 // Repeated delimiter: use as character value 601 io.HandleRelativePosition(byteCount); 602 } else { 603 break; // closing delimiter 604 } 605 } 606 if (length > 0) { 607 *x++ = *ch; 608 --length; 609 } 610 } 611 std::fill_n(x, length, ' '); 612 return result; 613 } 614 615 template <typename CHAR> 616 static bool EditListDirectedCharacterInput( 617 IoStatementState &io, CHAR *x, std::size_t length, const DataEdit &edit) { 618 std::size_t byteCount{0}; 619 auto ch{io.GetCurrentChar(byteCount)}; 620 if (ch && (*ch == '\'' || *ch == '"')) { 621 io.HandleRelativePosition(byteCount); 622 return EditDelimitedCharacterInput(io, x, length, *ch); 623 } 624 if (IsNamelistName(io) || io.GetConnectionState().IsAtEOF()) { 625 return false; 626 } 627 // Undelimited list-directed character input: stop at a value separator 628 // or the end of the current record. Subtlety: the "remaining" count 629 // here is a dummy that's used to avoid the interpretation of separators 630 // in NextInField. 631 std::optional<int> remaining{length > 0 ? maxUTF8Bytes : 0}; 632 while (std::optional<char32_t> next{io.NextInField(remaining, edit)}) { 633 switch (*next) { 634 case ' ': 635 case '\t': 636 case ',': 637 case ';': 638 case '/': 639 remaining = 0; // value separator: stop 640 break; 641 default: 642 *x++ = *next; 643 remaining = --length > 0 ? maxUTF8Bytes : 0; 644 } 645 } 646 std::fill_n(x, length, ' '); 647 return true; 648 } 649 650 template <typename CHAR> 651 bool EditCharacterInput( 652 IoStatementState &io, const DataEdit &edit, CHAR *x, std::size_t length) { 653 switch (edit.descriptor) { 654 case DataEdit::ListDirected: 655 return EditListDirectedCharacterInput(io, x, length, edit); 656 case 'A': 657 case 'G': 658 break; 659 case 'B': 660 return EditBOZInput<1>(io, edit, x, length * sizeof *x); 661 case 'O': 662 return EditBOZInput<3>(io, edit, x, length * sizeof *x); 663 case 'Z': 664 return EditBOZInput<4>(io, edit, x, length * sizeof *x); 665 default: 666 io.GetIoErrorHandler().SignalError(IostatErrorInFormat, 667 "Data edit descriptor '%c' may not be used with a CHARACTER data item", 668 edit.descriptor); 669 return false; 670 } 671 const ConnectionState &connection{io.GetConnectionState()}; 672 if (connection.IsAtEOF()) { 673 return false; 674 } 675 std::size_t remaining{length}; 676 if (edit.width && *edit.width > 0) { 677 remaining = *edit.width; 678 } 679 // When the field is wider than the variable, we drop the leading 680 // characters. When the variable is wider than the field, there's 681 // trailing padding. 682 const char *input{nullptr}; 683 std::size_t ready{0}; 684 bool hitEnd{false}; 685 // Skip leading bytes. 686 // These bytes don't count towards INQUIRE(IOLENGTH=). 687 std::size_t skip{remaining > length ? remaining - length : 0}; 688 // Transfer payload bytes; these do count. 689 while (remaining > 0) { 690 if (ready == 0) { 691 ready = io.GetNextInputBytes(input); 692 if (ready == 0) { 693 hitEnd = true; 694 break; 695 } 696 } 697 std::size_t chunk; 698 bool skipping{skip > 0}; 699 if (connection.isUTF8) { 700 chunk = MeasureUTF8Bytes(*input); 701 if (skipping) { 702 --skip; 703 } else if (auto ucs{DecodeUTF8(input)}) { 704 *x++ = *ucs; 705 --length; 706 } else if (chunk == 0) { 707 // error recovery: skip bad encoding 708 chunk = 1; 709 } 710 --remaining; 711 } else { 712 if (skipping) { 713 chunk = std::min<std::size_t>(skip, ready); 714 skip -= chunk; 715 } else { 716 chunk = std::min<std::size_t>(remaining, ready); 717 std::memcpy(x, input, chunk); 718 x += chunk; 719 length -= chunk; 720 } 721 remaining -= chunk; 722 } 723 input += chunk; 724 if (!skipping) { 725 io.GotChar(chunk); 726 } 727 io.HandleRelativePosition(chunk); 728 ready -= chunk; 729 } 730 // Pad the remainder of the input variable, if any. 731 std::fill_n(x, length, ' '); 732 if (hitEnd) { 733 io.CheckForEndOfRecord(); // signal any needed error 734 } 735 return true; 736 } 737 738 template bool EditRealInput<2>(IoStatementState &, const DataEdit &, void *); 739 template bool EditRealInput<3>(IoStatementState &, const DataEdit &, void *); 740 template bool EditRealInput<4>(IoStatementState &, const DataEdit &, void *); 741 template bool EditRealInput<8>(IoStatementState &, const DataEdit &, void *); 742 template bool EditRealInput<10>(IoStatementState &, const DataEdit &, void *); 743 // TODO: double/double 744 template bool EditRealInput<16>(IoStatementState &, const DataEdit &, void *); 745 746 template bool EditCharacterInput( 747 IoStatementState &, const DataEdit &, char *, std::size_t); 748 template bool EditCharacterInput( 749 IoStatementState &, const DataEdit &, char16_t *, std::size_t); 750 template bool EditCharacterInput( 751 IoStatementState &, const DataEdit &, char32_t *, std::size_t); 752 753 } // namespace Fortran::runtime::io 754