1 /* File object implementation */ 2 3 #define PY_SSIZE_T_CLEAN 4 #include "Python.h" 5 #include "structmember.h" 6 7 #ifdef HAVE_SYS_TYPES_H 8 #include <sys/types.h> 9 #endif /* HAVE_SYS_TYPES_H */ 10 11 #ifdef MS_WINDOWS 12 #define fileno _fileno 13 /* can simulate truncate with Win32 API functions; see file_truncate */ 14 #define HAVE_FTRUNCATE 15 #define WIN32_LEAN_AND_MEAN 16 #include <windows.h> 17 #endif 18 19 #if defined(PYOS_OS2) && defined(PYCC_GCC) 20 #include <io.h> 21 #endif 22 23 #define BUF(v) PyString_AS_STRING((PyStringObject *)v) 24 25 #ifdef HAVE_ERRNO_H 26 #include <errno.h> 27 #endif 28 29 #ifdef HAVE_GETC_UNLOCKED 30 #define GETC(f) getc_unlocked(f) 31 #define FLOCKFILE(f) flockfile(f) 32 #define FUNLOCKFILE(f) funlockfile(f) 33 #else 34 #define GETC(f) getc(f) 35 #define FLOCKFILE(f) 36 #define FUNLOCKFILE(f) 37 #endif 38 39 /* Bits in f_newlinetypes */ 40 #define NEWLINE_UNKNOWN 0 /* No newline seen, yet */ 41 #define NEWLINE_CR 1 /* \r newline seen */ 42 #define NEWLINE_LF 2 /* \n newline seen */ 43 #define NEWLINE_CRLF 4 /* \r\n newline seen */ 44 45 /* 46 * These macros release the GIL while preventing the f_close() function being 47 * called in the interval between them. For that purpose, a running total of 48 * the number of currently running unlocked code sections is kept in 49 * the unlocked_count field of the PyFileObject. The close() method raises 50 * an IOError if that field is non-zero. See issue #815646, #595601. 51 */ 52 53 #define FILE_BEGIN_ALLOW_THREADS(fobj) \ 54 { \ 55 fobj->unlocked_count++; \ 56 Py_BEGIN_ALLOW_THREADS 57 58 #define FILE_END_ALLOW_THREADS(fobj) \ 59 Py_END_ALLOW_THREADS \ 60 fobj->unlocked_count--; \ 61 assert(fobj->unlocked_count >= 0); \ 62 } 63 64 #define FILE_ABORT_ALLOW_THREADS(fobj) \ 65 Py_BLOCK_THREADS \ 66 fobj->unlocked_count--; \ 67 assert(fobj->unlocked_count >= 0); 68 69 #ifdef __cplusplus 70 extern "C" { 71 #endif 72 73 FILE * 74 PyFile_AsFile(PyObject *f) 75 { 76 if (f == NULL || !PyFile_Check(f)) 77 return NULL; 78 else 79 return ((PyFileObject *)f)->f_fp; 80 } 81 82 void PyFile_IncUseCount(PyFileObject *fobj) 83 { 84 fobj->unlocked_count++; 85 } 86 87 void PyFile_DecUseCount(PyFileObject *fobj) 88 { 89 fobj->unlocked_count--; 90 assert(fobj->unlocked_count >= 0); 91 } 92 93 PyObject * 94 PyFile_Name(PyObject *f) 95 { 96 if (f == NULL || !PyFile_Check(f)) 97 return NULL; 98 else 99 return ((PyFileObject *)f)->f_name; 100 } 101 102 /* This is a safe wrapper around PyObject_Print to print to the FILE 103 of a PyFileObject. PyObject_Print releases the GIL but knows nothing 104 about PyFileObject. */ 105 static int 106 file_PyObject_Print(PyObject *op, PyFileObject *f, int flags) 107 { 108 int result; 109 PyFile_IncUseCount(f); 110 result = PyObject_Print(op, f->f_fp, flags); 111 PyFile_DecUseCount(f); 112 return result; 113 } 114 115 /* On Unix, fopen will succeed for directories. 116 In Python, there should be no file objects referring to 117 directories, so we need a check. */ 118 119 static PyFileObject* 120 dircheck(PyFileObject* f) 121 { 122 #if defined(HAVE_FSTAT) && defined(S_IFDIR) && defined(EISDIR) 123 struct stat buf; 124 if (f->f_fp == NULL) 125 return f; 126 if (fstat(fileno(f->f_fp), &buf) == 0 && 127 S_ISDIR(buf.st_mode)) { 128 char *msg = strerror(EISDIR); 129 PyObject *exc = PyObject_CallFunction(PyExc_IOError, "(isO)", 130 EISDIR, msg, f->f_name); 131 PyErr_SetObject(PyExc_IOError, exc); 132 Py_XDECREF(exc); 133 return NULL; 134 } 135 #endif 136 return f; 137 } 138 139 140 static PyObject * 141 fill_file_fields(PyFileObject *f, FILE *fp, PyObject *name, char *mode, 142 int (*close)(FILE *)) 143 { 144 assert(name != NULL); 145 assert(f != NULL); 146 assert(PyFile_Check(f)); 147 assert(f->f_fp == NULL); 148 149 Py_DECREF(f->f_name); 150 Py_DECREF(f->f_mode); 151 Py_DECREF(f->f_encoding); 152 Py_DECREF(f->f_errors); 153 154 Py_INCREF(name); 155 f->f_name = name; 156 157 f->f_mode = PyString_FromString(mode); 158 159 f->f_close = close; 160 f->f_softspace = 0; 161 f->f_binary = strchr(mode,'b') != NULL; 162 f->f_buf = NULL; 163 f->f_univ_newline = (strchr(mode, 'U') != NULL); 164 f->f_newlinetypes = NEWLINE_UNKNOWN; 165 f->f_skipnextlf = 0; 166 Py_INCREF(Py_None); 167 f->f_encoding = Py_None; 168 Py_INCREF(Py_None); 169 f->f_errors = Py_None; 170 f->readable = f->writable = 0; 171 if (strchr(mode, 'r') != NULL || f->f_univ_newline) 172 f->readable = 1; 173 if (strchr(mode, 'w') != NULL || strchr(mode, 'a') != NULL) 174 f->writable = 1; 175 if (strchr(mode, '+') != NULL) 176 f->readable = f->writable = 1; 177 178 if (f->f_mode == NULL) 179 return NULL; 180 f->f_fp = fp; 181 f = dircheck(f); 182 return (PyObject *) f; 183 } 184 185 #if defined _MSC_VER && _MSC_VER >= 1400 && defined(__STDC_SECURE_LIB__) 186 #define Py_VERIFY_WINNT 187 /* The CRT on windows compiled with Visual Studio 2005 and higher may 188 * assert if given invalid mode strings. This is all fine and well 189 * in static languages like C where the mode string is typcially hard 190 * coded. But in Python, were we pass in the mode string from the user, 191 * we need to verify it first manually 192 */ 193 static int _PyVerify_Mode_WINNT(const char *mode) 194 { 195 /* See if mode string is valid on Windows to avoid hard assertions */ 196 /* remove leading spacese */ 197 int singles = 0; 198 int pairs = 0; 199 int encoding = 0; 200 const char *s, *c; 201 202 while(*mode == ' ') /* strip initial spaces */ 203 ++mode; 204 if (!strchr("rwa", *mode)) /* must start with one of these */ 205 return 0; 206 while (*++mode) { 207 if (*mode == ' ' || *mode == 'N') /* ignore spaces and N */ 208 continue; 209 s = "+TD"; /* each of this can appear only once */ 210 c = strchr(s, *mode); 211 if (c) { 212 ptrdiff_t idx = s-c; 213 if (singles & (1<<idx)) 214 return 0; 215 singles |= (1<<idx); 216 continue; 217 } 218 s = "btcnSR"; /* only one of each letter in the pairs allowed */ 219 c = strchr(s, *mode); 220 if (c) { 221 ptrdiff_t idx = (s-c)/2; 222 if (pairs & (1<<idx)) 223 return 0; 224 pairs |= (1<<idx); 225 continue; 226 } 227 if (*mode == ',') { 228 encoding = 1; 229 break; 230 } 231 return 0; /* found an invalid char */ 232 } 233 234 if (encoding) { 235 char *e[] = {"UTF-8", "UTF-16LE", "UNICODE"}; 236 while (*mode == ' ') 237 ++mode; 238 /* find 'ccs =' */ 239 if (strncmp(mode, "ccs", 3)) 240 return 0; 241 mode += 3; 242 while (*mode == ' ') 243 ++mode; 244 if (*mode != '=') 245 return 0; 246 while (*mode == ' ') 247 ++mode; 248 for(encoding = 0; encoding<_countof(e); ++encoding) { 249 size_t l = strlen(e[encoding]); 250 if (!strncmp(mode, e[encoding], l)) { 251 mode += l; /* found a valid encoding */ 252 break; 253 } 254 } 255 if (encoding == _countof(e)) 256 return 0; 257 } 258 /* skip trailing spaces */ 259 while (*mode == ' ') 260 ++mode; 261 262 return *mode == '\0'; /* must be at the end of the string */ 263 } 264 #endif 265 266 /* check for known incorrect mode strings - problem is, platforms are 267 free to accept any mode characters they like and are supposed to 268 ignore stuff they don't understand... write or append mode with 269 universal newline support is expressly forbidden by PEP 278. 270 Additionally, remove the 'U' from the mode string as platforms 271 won't know what it is. Non-zero return signals an exception */ 272 int 273 _PyFile_SanitizeMode(char *mode) 274 { 275 char *upos; 276 size_t len = strlen(mode); 277 278 if (!len) { 279 PyErr_SetString(PyExc_ValueError, "empty mode string"); 280 return -1; 281 } 282 283 upos = strchr(mode, 'U'); 284 if (upos) { 285 memmove(upos, upos+1, len-(upos-mode)); /* incl null char */ 286 287 if (mode[0] == 'w' || mode[0] == 'a') { 288 PyErr_Format(PyExc_ValueError, "universal newline " 289 "mode can only be used with modes " 290 "starting with 'r'"); 291 return -1; 292 } 293 294 if (mode[0] != 'r') { 295 memmove(mode+1, mode, strlen(mode)+1); 296 mode[0] = 'r'; 297 } 298 299 if (!strchr(mode, 'b')) { 300 memmove(mode+2, mode+1, strlen(mode)); 301 mode[1] = 'b'; 302 } 303 } else if (mode[0] != 'r' && mode[0] != 'w' && mode[0] != 'a') { 304 PyErr_Format(PyExc_ValueError, "mode string must begin with " 305 "one of 'r', 'w', 'a' or 'U', not '%.200s'", mode); 306 return -1; 307 } 308 #ifdef Py_VERIFY_WINNT 309 /* additional checks on NT with visual studio 2005 and higher */ 310 if (!_PyVerify_Mode_WINNT(mode)) { 311 PyErr_Format(PyExc_ValueError, "Invalid mode ('%.50s')", mode); 312 return -1; 313 } 314 #endif 315 return 0; 316 } 317 318 static PyObject * 319 open_the_file(PyFileObject *f, char *name, char *mode) 320 { 321 char *newmode; 322 assert(f != NULL); 323 assert(PyFile_Check(f)); 324 #ifdef MS_WINDOWS 325 /* windows ignores the passed name in order to support Unicode */ 326 assert(f->f_name != NULL); 327 #else 328 assert(name != NULL); 329 #endif 330 assert(mode != NULL); 331 assert(f->f_fp == NULL); 332 333 /* probably need to replace 'U' by 'rb' */ 334 newmode = PyMem_MALLOC(strlen(mode) + 3); 335 if (!newmode) { 336 PyErr_NoMemory(); 337 return NULL; 338 } 339 strcpy(newmode, mode); 340 341 if (_PyFile_SanitizeMode(newmode)) { 342 f = NULL; 343 goto cleanup; 344 } 345 346 /* rexec.py can't stop a user from getting the file() constructor -- 347 all they have to do is get *any* file object f, and then do 348 type(f). Here we prevent them from doing damage with it. */ 349 if (PyEval_GetRestricted()) { 350 PyErr_SetString(PyExc_IOError, 351 "file() constructor not accessible in restricted mode"); 352 f = NULL; 353 goto cleanup; 354 } 355 errno = 0; 356 357 #ifdef MS_WINDOWS 358 if (PyUnicode_Check(f->f_name)) { 359 PyObject *wmode; 360 wmode = PyUnicode_DecodeASCII(newmode, strlen(newmode), NULL); 361 if (f->f_name && wmode) { 362 FILE_BEGIN_ALLOW_THREADS(f) 363 /* PyUnicode_AS_UNICODE OK without thread 364 lock as it is a simple dereference. */ 365 f->f_fp = _wfopen(PyUnicode_AS_UNICODE(f->f_name), 366 PyUnicode_AS_UNICODE(wmode)); 367 FILE_END_ALLOW_THREADS(f) 368 } 369 Py_XDECREF(wmode); 370 } 371 #endif 372 if (NULL == f->f_fp && NULL != name) { 373 FILE_BEGIN_ALLOW_THREADS(f) 374 f->f_fp = fopen(name, newmode); 375 FILE_END_ALLOW_THREADS(f) 376 } 377 378 if (f->f_fp == NULL) { 379 #if defined _MSC_VER && (_MSC_VER < 1400 || !defined(__STDC_SECURE_LIB__)) 380 /* MSVC 6 (Microsoft) leaves errno at 0 for bad mode strings, 381 * across all Windows flavors. When it sets EINVAL varies 382 * across Windows flavors, the exact conditions aren't 383 * documented, and the answer lies in the OS's implementation 384 * of Win32's CreateFile function (whose source is secret). 385 * Seems the best we can do is map EINVAL to ENOENT. 386 * Starting with Visual Studio .NET 2005, EINVAL is correctly 387 * set by our CRT error handler (set in exceptions.c.) 388 */ 389 if (errno == 0) /* bad mode string */ 390 errno = EINVAL; 391 else if (errno == EINVAL) /* unknown, but not a mode string */ 392 errno = ENOENT; 393 #endif 394 /* EINVAL is returned when an invalid filename or 395 * an invalid mode is supplied. */ 396 if (errno == EINVAL) { 397 PyObject *v; 398 char message[100]; 399 PyOS_snprintf(message, 100, 400 "invalid mode ('%.50s') or filename", mode); 401 v = Py_BuildValue("(isO)", errno, message, f->f_name); 402 if (v != NULL) { 403 PyErr_SetObject(PyExc_IOError, v); 404 Py_DECREF(v); 405 } 406 } 407 else 408 PyErr_SetFromErrnoWithFilenameObject(PyExc_IOError, f->f_name); 409 f = NULL; 410 } 411 if (f != NULL) 412 f = dircheck(f); 413 414 cleanup: 415 PyMem_FREE(newmode); 416 417 return (PyObject *)f; 418 } 419 420 static PyObject * 421 close_the_file(PyFileObject *f) 422 { 423 int sts = 0; 424 int (*local_close)(FILE *); 425 FILE *local_fp = f->f_fp; 426 char *local_setbuf = f->f_setbuf; 427 if (local_fp != NULL) { 428 local_close = f->f_close; 429 if (local_close != NULL && f->unlocked_count > 0) { 430 if (f->ob_refcnt > 0) { 431 PyErr_SetString(PyExc_IOError, 432 "close() called during concurrent " 433 "operation on the same file object."); 434 } else { 435 /* This should not happen unless someone is 436 * carelessly playing with the PyFileObject 437 * struct fields and/or its associated FILE 438 * pointer. */ 439 PyErr_SetString(PyExc_SystemError, 440 "PyFileObject locking error in " 441 "destructor (refcnt <= 0 at close)."); 442 } 443 return NULL; 444 } 445 /* NULL out the FILE pointer before releasing the GIL, because 446 * it will not be valid anymore after the close() function is 447 * called. */ 448 f->f_fp = NULL; 449 if (local_close != NULL) { 450 /* Issue #9295: must temporarily reset f_setbuf so that another 451 thread doesn't free it when running file_close() concurrently. 452 Otherwise this close() will crash when flushing the buffer. */ 453 f->f_setbuf = NULL; 454 Py_BEGIN_ALLOW_THREADS 455 errno = 0; 456 sts = (*local_close)(local_fp); 457 Py_END_ALLOW_THREADS 458 f->f_setbuf = local_setbuf; 459 if (sts == EOF) 460 return PyErr_SetFromErrno(PyExc_IOError); 461 if (sts != 0) 462 return PyInt_FromLong((long)sts); 463 } 464 } 465 Py_RETURN_NONE; 466 } 467 468 PyObject * 469 PyFile_FromFile(FILE *fp, char *name, char *mode, int (*close)(FILE *)) 470 { 471 PyFileObject *f; 472 PyObject *o_name; 473 474 f = (PyFileObject *)PyFile_Type.tp_new(&PyFile_Type, NULL, NULL); 475 if (f == NULL) 476 return NULL; 477 o_name = PyString_FromString(name); 478 if (o_name == NULL) { 479 if (close != NULL && fp != NULL) 480 close(fp); 481 Py_DECREF(f); 482 return NULL; 483 } 484 if (fill_file_fields(f, fp, o_name, mode, close) == NULL) { 485 Py_DECREF(f); 486 Py_DECREF(o_name); 487 return NULL; 488 } 489 Py_DECREF(o_name); 490 return (PyObject *)f; 491 } 492 493 PyObject * 494 PyFile_FromString(char *name, char *mode) 495 { 496 extern int fclose(FILE *); 497 PyFileObject *f; 498 499 f = (PyFileObject *)PyFile_FromFile((FILE *)NULL, name, mode, fclose); 500 if (f != NULL) { 501 if (open_the_file(f, name, mode) == NULL) { 502 Py_DECREF(f); 503 f = NULL; 504 } 505 } 506 return (PyObject *)f; 507 } 508 509 void 510 PyFile_SetBufSize(PyObject *f, int bufsize) 511 { 512 PyFileObject *file = (PyFileObject *)f; 513 if (bufsize >= 0) { 514 int type; 515 switch (bufsize) { 516 case 0: 517 type = _IONBF; 518 break; 519 #ifdef HAVE_SETVBUF 520 case 1: 521 type = _IOLBF; 522 bufsize = BUFSIZ; 523 break; 524 #endif 525 default: 526 type = _IOFBF; 527 #ifndef HAVE_SETVBUF 528 bufsize = BUFSIZ; 529 #endif 530 break; 531 } 532 fflush(file->f_fp); 533 if (type == _IONBF) { 534 PyMem_Free(file->f_setbuf); 535 file->f_setbuf = NULL; 536 } else { 537 file->f_setbuf = (char *)PyMem_Realloc(file->f_setbuf, 538 bufsize); 539 } 540 #ifdef HAVE_SETVBUF 541 setvbuf(file->f_fp, file->f_setbuf, type, bufsize); 542 #else /* !HAVE_SETVBUF */ 543 setbuf(file->f_fp, file->f_setbuf); 544 #endif /* !HAVE_SETVBUF */ 545 } 546 } 547 548 /* Set the encoding used to output Unicode strings. 549 Return 1 on success, 0 on failure. */ 550 551 int 552 PyFile_SetEncoding(PyObject *f, const char *enc) 553 { 554 return PyFile_SetEncodingAndErrors(f, enc, NULL); 555 } 556 557 int 558 PyFile_SetEncodingAndErrors(PyObject *f, const char *enc, char* errors) 559 { 560 PyFileObject *file = (PyFileObject*)f; 561 PyObject *str, *oerrors; 562 563 assert(PyFile_Check(f)); 564 str = PyString_FromString(enc); 565 if (!str) 566 return 0; 567 if (errors) { 568 oerrors = PyString_FromString(errors); 569 if (!oerrors) { 570 Py_DECREF(str); 571 return 0; 572 } 573 } else { 574 oerrors = Py_None; 575 Py_INCREF(Py_None); 576 } 577 Py_DECREF(file->f_encoding); 578 file->f_encoding = str; 579 Py_DECREF(file->f_errors); 580 file->f_errors = oerrors; 581 return 1; 582 } 583 584 static PyObject * 585 err_closed(void) 586 { 587 PyErr_SetString(PyExc_ValueError, "I/O operation on closed file"); 588 return NULL; 589 } 590 591 static PyObject * 592 err_mode(char *action) 593 { 594 PyErr_Format(PyExc_IOError, "File not open for %s", action); 595 return NULL; 596 } 597 598 /* Refuse regular file I/O if there's data in the iteration-buffer. 599 * Mixing them would cause data to arrive out of order, as the read* 600 * methods don't use the iteration buffer. */ 601 static PyObject * 602 err_iterbuffered(void) 603 { 604 PyErr_SetString(PyExc_ValueError, 605 "Mixing iteration and read methods would lose data"); 606 return NULL; 607 } 608 609 static void drop_readahead(PyFileObject *); 610 611 /* Methods */ 612 613 static void 614 file_dealloc(PyFileObject *f) 615 { 616 PyObject *ret; 617 if (f->weakreflist != NULL) 618 PyObject_ClearWeakRefs((PyObject *) f); 619 ret = close_the_file(f); 620 if (!ret) { 621 PySys_WriteStderr("close failed in file object destructor:\n"); 622 PyErr_Print(); 623 } 624 else { 625 Py_DECREF(ret); 626 } 627 PyMem_Free(f->f_setbuf); 628 Py_XDECREF(f->f_name); 629 Py_XDECREF(f->f_mode); 630 Py_XDECREF(f->f_encoding); 631 Py_XDECREF(f->f_errors); 632 drop_readahead(f); 633 Py_TYPE(f)->tp_free((PyObject *)f); 634 } 635 636 static PyObject * 637 file_repr(PyFileObject *f) 638 { 639 PyObject *ret = NULL; 640 PyObject *name = NULL; 641 if (PyUnicode_Check(f->f_name)) { 642 #ifdef Py_USING_UNICODE 643 const char *name_str; 644 name = PyUnicode_AsUnicodeEscapeString(f->f_name); 645 name_str = name ? PyString_AsString(name) : "?"; 646 ret = PyString_FromFormat("<%s file u'%s', mode '%s' at %p>", 647 f->f_fp == NULL ? "closed" : "open", 648 name_str, 649 PyString_AsString(f->f_mode), 650 f); 651 Py_XDECREF(name); 652 return ret; 653 #endif 654 } else { 655 name = PyObject_Repr(f->f_name); 656 if (name == NULL) 657 return NULL; 658 ret = PyString_FromFormat("<%s file %s, mode '%s' at %p>", 659 f->f_fp == NULL ? "closed" : "open", 660 PyString_AsString(name), 661 PyString_AsString(f->f_mode), 662 f); 663 Py_XDECREF(name); 664 return ret; 665 } 666 } 667 668 static PyObject * 669 file_close(PyFileObject *f) 670 { 671 PyObject *sts = close_the_file(f); 672 if (sts) { 673 PyMem_Free(f->f_setbuf); 674 f->f_setbuf = NULL; 675 } 676 return sts; 677 } 678 679 680 /* Our very own off_t-like type, 64-bit if possible */ 681 #if !defined(HAVE_LARGEFILE_SUPPORT) 682 typedef off_t Py_off_t; 683 #elif SIZEOF_OFF_T >= 8 684 typedef off_t Py_off_t; 685 #elif SIZEOF_FPOS_T >= 8 686 typedef fpos_t Py_off_t; 687 #else 688 #error "Large file support, but neither off_t nor fpos_t is large enough." 689 #endif 690 691 692 /* a portable fseek() function 693 return 0 on success, non-zero on failure (with errno set) */ 694 static int 695 _portable_fseek(FILE *fp, Py_off_t offset, int whence) 696 { 697 #if !defined(HAVE_LARGEFILE_SUPPORT) 698 return fseek(fp, offset, whence); 699 #elif defined(HAVE_FSEEKO) && SIZEOF_OFF_T >= 8 700 return fseeko(fp, offset, whence); 701 #elif defined(HAVE_FSEEK64) 702 return fseek64(fp, offset, whence); 703 #elif defined(__BEOS__) 704 return _fseek(fp, offset, whence); 705 #elif SIZEOF_FPOS_T >= 8 706 /* lacking a 64-bit capable fseek(), use a 64-bit capable fsetpos() 707 and fgetpos() to implement fseek()*/ 708 fpos_t pos; 709 switch (whence) { 710 case SEEK_END: 711 #ifdef MS_WINDOWS 712 fflush(fp); 713 if (_lseeki64(fileno(fp), 0, 2) == -1) 714 return -1; 715 #else 716 if (fseek(fp, 0, SEEK_END) != 0) 717 return -1; 718 #endif 719 /* fall through */ 720 case SEEK_CUR: 721 if (fgetpos(fp, &pos) != 0) 722 return -1; 723 offset += pos; 724 break; 725 /* case SEEK_SET: break; */ 726 } 727 return fsetpos(fp, &offset); 728 #else 729 #error "Large file support, but no way to fseek." 730 #endif 731 } 732 733 734 /* a portable ftell() function 735 Return -1 on failure with errno set appropriately, current file 736 position on success */ 737 static Py_off_t 738 _portable_ftell(FILE* fp) 739 { 740 #if !defined(HAVE_LARGEFILE_SUPPORT) 741 return ftell(fp); 742 #elif defined(HAVE_FTELLO) && SIZEOF_OFF_T >= 8 743 return ftello(fp); 744 #elif defined(HAVE_FTELL64) 745 return ftell64(fp); 746 #elif SIZEOF_FPOS_T >= 8 747 fpos_t pos; 748 if (fgetpos(fp, &pos) != 0) 749 return -1; 750 return pos; 751 #else 752 #error "Large file support, but no way to ftell." 753 #endif 754 } 755 756 757 static PyObject * 758 file_seek(PyFileObject *f, PyObject *args) 759 { 760 int whence; 761 int ret; 762 Py_off_t offset; 763 PyObject *offobj, *off_index; 764 765 if (f->f_fp == NULL) 766 return err_closed(); 767 drop_readahead(f); 768 whence = 0; 769 if (!PyArg_ParseTuple(args, "O|i:seek", &offobj, &whence)) 770 return NULL; 771 off_index = PyNumber_Index(offobj); 772 if (!off_index) { 773 if (!PyFloat_Check(offobj)) 774 return NULL; 775 /* Deprecated in 2.6 */ 776 PyErr_Clear(); 777 if (PyErr_WarnEx(PyExc_DeprecationWarning, 778 "integer argument expected, got float", 779 1) < 0) 780 return NULL; 781 off_index = offobj; 782 Py_INCREF(offobj); 783 } 784 #if !defined(HAVE_LARGEFILE_SUPPORT) 785 offset = PyInt_AsLong(off_index); 786 #else 787 offset = PyLong_Check(off_index) ? 788 PyLong_AsLongLong(off_index) : PyInt_AsLong(off_index); 789 #endif 790 Py_DECREF(off_index); 791 if (PyErr_Occurred()) 792 return NULL; 793 794 FILE_BEGIN_ALLOW_THREADS(f) 795 errno = 0; 796 ret = _portable_fseek(f->f_fp, offset, whence); 797 FILE_END_ALLOW_THREADS(f) 798 799 if (ret != 0) { 800 PyErr_SetFromErrno(PyExc_IOError); 801 clearerr(f->f_fp); 802 return NULL; 803 } 804 f->f_skipnextlf = 0; 805 Py_INCREF(Py_None); 806 return Py_None; 807 } 808 809 810 #ifdef HAVE_FTRUNCATE 811 static PyObject * 812 file_truncate(PyFileObject *f, PyObject *args) 813 { 814 Py_off_t newsize; 815 PyObject *newsizeobj = NULL; 816 Py_off_t initialpos; 817 int ret; 818 819 if (f->f_fp == NULL) 820 return err_closed(); 821 if (!f->writable) 822 return err_mode("writing"); 823 if (!PyArg_UnpackTuple(args, "truncate", 0, 1, &newsizeobj)) 824 return NULL; 825 826 /* Get current file position. If the file happens to be open for 827 * update and the last operation was an input operation, C doesn't 828 * define what the later fflush() will do, but we promise truncate() 829 * won't change the current position (and fflush() *does* change it 830 * then at least on Windows). The easiest thing is to capture 831 * current pos now and seek back to it at the end. 832 */ 833 FILE_BEGIN_ALLOW_THREADS(f) 834 errno = 0; 835 initialpos = _portable_ftell(f->f_fp); 836 FILE_END_ALLOW_THREADS(f) 837 if (initialpos == -1) 838 goto onioerror; 839 840 /* Set newsize to current postion if newsizeobj NULL, else to the 841 * specified value. 842 */ 843 if (newsizeobj != NULL) { 844 #if !defined(HAVE_LARGEFILE_SUPPORT) 845 newsize = PyInt_AsLong(newsizeobj); 846 #else 847 newsize = PyLong_Check(newsizeobj) ? 848 PyLong_AsLongLong(newsizeobj) : 849 PyInt_AsLong(newsizeobj); 850 #endif 851 if (PyErr_Occurred()) 852 return NULL; 853 } 854 else /* default to current position */ 855 newsize = initialpos; 856 857 /* Flush the stream. We're mixing stream-level I/O with lower-level 858 * I/O, and a flush may be necessary to synch both platform views 859 * of the current file state. 860 */ 861 FILE_BEGIN_ALLOW_THREADS(f) 862 errno = 0; 863 ret = fflush(f->f_fp); 864 FILE_END_ALLOW_THREADS(f) 865 if (ret != 0) 866 goto onioerror; 867 868 #ifdef MS_WINDOWS 869 /* MS _chsize doesn't work if newsize doesn't fit in 32 bits, 870 so don't even try using it. */ 871 { 872 HANDLE hFile; 873 874 /* Have to move current pos to desired endpoint on Windows. */ 875 FILE_BEGIN_ALLOW_THREADS(f) 876 errno = 0; 877 ret = _portable_fseek(f->f_fp, newsize, SEEK_SET) != 0; 878 FILE_END_ALLOW_THREADS(f) 879 if (ret) 880 goto onioerror; 881 882 /* Truncate. Note that this may grow the file! */ 883 FILE_BEGIN_ALLOW_THREADS(f) 884 errno = 0; 885 hFile = (HANDLE)_get_osfhandle(fileno(f->f_fp)); 886 ret = hFile == (HANDLE)-1; 887 if (ret == 0) { 888 ret = SetEndOfFile(hFile) == 0; 889 if (ret) 890 errno = EACCES; 891 } 892 FILE_END_ALLOW_THREADS(f) 893 if (ret) 894 goto onioerror; 895 } 896 #else 897 FILE_BEGIN_ALLOW_THREADS(f) 898 errno = 0; 899 ret = ftruncate(fileno(f->f_fp), newsize); 900 FILE_END_ALLOW_THREADS(f) 901 if (ret != 0) 902 goto onioerror; 903 #endif /* !MS_WINDOWS */ 904 905 /* Restore original file position. */ 906 FILE_BEGIN_ALLOW_THREADS(f) 907 errno = 0; 908 ret = _portable_fseek(f->f_fp, initialpos, SEEK_SET) != 0; 909 FILE_END_ALLOW_THREADS(f) 910 if (ret) 911 goto onioerror; 912 913 Py_INCREF(Py_None); 914 return Py_None; 915 916 onioerror: 917 PyErr_SetFromErrno(PyExc_IOError); 918 clearerr(f->f_fp); 919 return NULL; 920 } 921 #endif /* HAVE_FTRUNCATE */ 922 923 static PyObject * 924 file_tell(PyFileObject *f) 925 { 926 Py_off_t pos; 927 928 if (f->f_fp == NULL) 929 return err_closed(); 930 FILE_BEGIN_ALLOW_THREADS(f) 931 errno = 0; 932 pos = _portable_ftell(f->f_fp); 933 FILE_END_ALLOW_THREADS(f) 934 935 if (pos == -1) { 936 PyErr_SetFromErrno(PyExc_IOError); 937 clearerr(f->f_fp); 938 return NULL; 939 } 940 if (f->f_skipnextlf) { 941 int c; 942 c = GETC(f->f_fp); 943 if (c == '\n') { 944 f->f_newlinetypes |= NEWLINE_CRLF; 945 pos++; 946 f->f_skipnextlf = 0; 947 } else if (c != EOF) ungetc(c, f->f_fp); 948 } 949 #if !defined(HAVE_LARGEFILE_SUPPORT) 950 return PyInt_FromLong(pos); 951 #else 952 return PyLong_FromLongLong(pos); 953 #endif 954 } 955 956 static PyObject * 957 file_fileno(PyFileObject *f) 958 { 959 if (f->f_fp == NULL) 960 return err_closed(); 961 return PyInt_FromLong((long) fileno(f->f_fp)); 962 } 963 964 static PyObject * 965 file_flush(PyFileObject *f) 966 { 967 int res; 968 969 if (f->f_fp == NULL) 970 return err_closed(); 971 FILE_BEGIN_ALLOW_THREADS(f) 972 errno = 0; 973 res = fflush(f->f_fp); 974 FILE_END_ALLOW_THREADS(f) 975 if (res != 0) { 976 PyErr_SetFromErrno(PyExc_IOError); 977 clearerr(f->f_fp); 978 return NULL; 979 } 980 Py_INCREF(Py_None); 981 return Py_None; 982 } 983 984 static PyObject * 985 file_isatty(PyFileObject *f) 986 { 987 long res; 988 if (f->f_fp == NULL) 989 return err_closed(); 990 FILE_BEGIN_ALLOW_THREADS(f) 991 res = isatty((int)fileno(f->f_fp)); 992 FILE_END_ALLOW_THREADS(f) 993 return PyBool_FromLong(res); 994 } 995 996 997 #if BUFSIZ < 8192 998 #define SMALLCHUNK 8192 999 #else 1000 #define SMALLCHUNK BUFSIZ 1001 #endif 1002 1003 static size_t 1004 new_buffersize(PyFileObject *f, size_t currentsize) 1005 { 1006 #ifdef HAVE_FSTAT 1007 off_t pos, end; 1008 struct stat st; 1009 if (fstat(fileno(f->f_fp), &st) == 0) { 1010 end = st.st_size; 1011 /* The following is not a bug: we really need to call lseek() 1012 *and* ftell(). The reason is that some stdio libraries 1013 mistakenly flush their buffer when ftell() is called and 1014 the lseek() call it makes fails, thereby throwing away 1015 data that cannot be recovered in any way. To avoid this, 1016 we first test lseek(), and only call ftell() if lseek() 1017 works. We can't use the lseek() value either, because we 1018 need to take the amount of buffered data into account. 1019 (Yet another reason why stdio stinks. :-) */ 1020 pos = lseek(fileno(f->f_fp), 0L, SEEK_CUR); 1021 if (pos >= 0) { 1022 pos = ftell(f->f_fp); 1023 } 1024 if (pos < 0) 1025 clearerr(f->f_fp); 1026 if (end > pos && pos >= 0) 1027 return currentsize + end - pos + 1; 1028 /* Add 1 so if the file were to grow we'd notice. */ 1029 } 1030 #endif 1031 /* Expand the buffer by an amount proportional to the current size, 1032 giving us amortized linear-time behavior. Use a less-than-double 1033 growth factor to avoid excessive allocation. */ 1034 return currentsize + (currentsize >> 3) + 6; 1035 } 1036 1037 #if defined(EWOULDBLOCK) && defined(EAGAIN) && EWOULDBLOCK != EAGAIN 1038 #define BLOCKED_ERRNO(x) ((x) == EWOULDBLOCK || (x) == EAGAIN) 1039 #else 1040 #ifdef EWOULDBLOCK 1041 #define BLOCKED_ERRNO(x) ((x) == EWOULDBLOCK) 1042 #else 1043 #ifdef EAGAIN 1044 #define BLOCKED_ERRNO(x) ((x) == EAGAIN) 1045 #else 1046 #define BLOCKED_ERRNO(x) 0 1047 #endif 1048 #endif 1049 #endif 1050 1051 static PyObject * 1052 file_read(PyFileObject *f, PyObject *args) 1053 { 1054 long bytesrequested = -1; 1055 size_t bytesread, buffersize, chunksize; 1056 PyObject *v; 1057 1058 if (f->f_fp == NULL) 1059 return err_closed(); 1060 if (!f->readable) 1061 return err_mode("reading"); 1062 /* refuse to mix with f.next() */ 1063 if (f->f_buf != NULL && 1064 (f->f_bufend - f->f_bufptr) > 0 && 1065 f->f_buf[0] != '\0') 1066 return err_iterbuffered(); 1067 if (!PyArg_ParseTuple(args, "|l:read", &bytesrequested)) 1068 return NULL; 1069 if (bytesrequested < 0) 1070 buffersize = new_buffersize(f, (size_t)0); 1071 else 1072 buffersize = bytesrequested; 1073 if (buffersize > PY_SSIZE_T_MAX) { 1074 PyErr_SetString(PyExc_OverflowError, 1075 "requested number of bytes is more than a Python string can hold"); 1076 return NULL; 1077 } 1078 v = PyString_FromStringAndSize((char *)NULL, buffersize); 1079 if (v == NULL) 1080 return NULL; 1081 bytesread = 0; 1082 for (;;) { 1083 int interrupted; 1084 FILE_BEGIN_ALLOW_THREADS(f) 1085 errno = 0; 1086 chunksize = Py_UniversalNewlineFread(BUF(v) + bytesread, 1087 buffersize - bytesread, f->f_fp, (PyObject *)f); 1088 interrupted = ferror(f->f_fp) && errno == EINTR; 1089 FILE_END_ALLOW_THREADS(f) 1090 if (interrupted) { 1091 clearerr(f->f_fp); 1092 if (PyErr_CheckSignals()) { 1093 Py_DECREF(v); 1094 return NULL; 1095 } 1096 } 1097 if (chunksize == 0) { 1098 if (interrupted) 1099 continue; 1100 if (!ferror(f->f_fp)) 1101 break; 1102 clearerr(f->f_fp); 1103 /* When in non-blocking mode, data shouldn't 1104 * be discarded if a blocking signal was 1105 * received. That will also happen if 1106 * chunksize != 0, but bytesread < buffersize. */ 1107 if (bytesread > 0 && BLOCKED_ERRNO(errno)) 1108 break; 1109 PyErr_SetFromErrno(PyExc_IOError); 1110 Py_DECREF(v); 1111 return NULL; 1112 } 1113 bytesread += chunksize; 1114 if (bytesread < buffersize && !interrupted) { 1115 clearerr(f->f_fp); 1116 break; 1117 } 1118 if (bytesrequested < 0) { 1119 buffersize = new_buffersize(f, buffersize); 1120 if (_PyString_Resize(&v, buffersize) < 0) 1121 return NULL; 1122 } else { 1123 /* Got what was requested. */ 1124 break; 1125 } 1126 } 1127 if (bytesread != buffersize && _PyString_Resize(&v, bytesread)) 1128 return NULL; 1129 return v; 1130 } 1131 1132 static PyObject * 1133 file_readinto(PyFileObject *f, PyObject *args) 1134 { 1135 char *ptr; 1136 Py_ssize_t ntodo; 1137 Py_ssize_t ndone, nnow; 1138 Py_buffer pbuf; 1139 1140 if (f->f_fp == NULL) 1141 return err_closed(); 1142 if (!f->readable) 1143 return err_mode("reading"); 1144 /* refuse to mix with f.next() */ 1145 if (f->f_buf != NULL && 1146 (f->f_bufend - f->f_bufptr) > 0 && 1147 f->f_buf[0] != '\0') 1148 return err_iterbuffered(); 1149 if (!PyArg_ParseTuple(args, "w*", &pbuf)) 1150 return NULL; 1151 ptr = pbuf.buf; 1152 ntodo = pbuf.len; 1153 ndone = 0; 1154 while (ntodo > 0) { 1155 int interrupted; 1156 FILE_BEGIN_ALLOW_THREADS(f) 1157 errno = 0; 1158 nnow = Py_UniversalNewlineFread(ptr+ndone, ntodo, f->f_fp, 1159 (PyObject *)f); 1160 interrupted = ferror(f->f_fp) && errno == EINTR; 1161 FILE_END_ALLOW_THREADS(f) 1162 if (interrupted) { 1163 clearerr(f->f_fp); 1164 if (PyErr_CheckSignals()) { 1165 PyBuffer_Release(&pbuf); 1166 return NULL; 1167 } 1168 } 1169 if (nnow == 0) { 1170 if (interrupted) 1171 continue; 1172 if (!ferror(f->f_fp)) 1173 break; 1174 PyErr_SetFromErrno(PyExc_IOError); 1175 clearerr(f->f_fp); 1176 PyBuffer_Release(&pbuf); 1177 return NULL; 1178 } 1179 ndone += nnow; 1180 ntodo -= nnow; 1181 } 1182 PyBuffer_Release(&pbuf); 1183 return PyInt_FromSsize_t(ndone); 1184 } 1185 1186 /************************************************************************** 1187 Routine to get next line using platform fgets(). 1188 1189 Under MSVC 6: 1190 1191 + MS threadsafe getc is very slow (multiple layers of function calls before+ 1192 after each character, to lock+unlock the stream). 1193 + The stream-locking functions are MS-internal -- can't access them from user 1194 code. 1195 + There's nothing Tim could find in the MS C or platform SDK libraries that 1196 can worm around this. 1197 + MS fgets locks/unlocks only once per line; it's the only hook we have. 1198 1199 So we use fgets for speed(!), despite that it's painful. 1200 1201 MS realloc is also slow. 1202 1203 Reports from other platforms on this method vs getc_unlocked (which MS doesn't 1204 have): 1205 Linux a wash 1206 Solaris a wash 1207 Tru64 Unix getline_via_fgets significantly faster 1208 1209 CAUTION: The C std isn't clear about this: in those cases where fgets 1210 writes something into the buffer, can it write into any position beyond the 1211 required trailing null byte? MSVC 6 fgets does not, and no platform is (yet) 1212 known on which it does; and it would be a strange way to code fgets. Still, 1213 getline_via_fgets may not work correctly if it does. The std test 1214 test_bufio.py should fail if platform fgets() routinely writes beyond the 1215 trailing null byte. #define DONT_USE_FGETS_IN_GETLINE to disable this code. 1216 **************************************************************************/ 1217 1218 /* Use this routine if told to, or by default on non-get_unlocked() 1219 * platforms unless told not to. Yikes! Let's spell that out: 1220 * On a platform with getc_unlocked(): 1221 * By default, use getc_unlocked(). 1222 * If you want to use fgets() instead, #define USE_FGETS_IN_GETLINE. 1223 * On a platform without getc_unlocked(): 1224 * By default, use fgets(). 1225 * If you don't want to use fgets(), #define DONT_USE_FGETS_IN_GETLINE. 1226 */ 1227 #if !defined(USE_FGETS_IN_GETLINE) && !defined(HAVE_GETC_UNLOCKED) 1228 #define USE_FGETS_IN_GETLINE 1229 #endif 1230 1231 #if defined(DONT_USE_FGETS_IN_GETLINE) && defined(USE_FGETS_IN_GETLINE) 1232 #undef USE_FGETS_IN_GETLINE 1233 #endif 1234 1235 #ifdef USE_FGETS_IN_GETLINE 1236 static PyObject* 1237 getline_via_fgets(PyFileObject *f, FILE *fp) 1238 { 1239 /* INITBUFSIZE is the maximum line length that lets us get away with the fast 1240 * no-realloc, one-fgets()-call path. Boosting it isn't free, because we have 1241 * to fill this much of the buffer with a known value in order to figure out 1242 * how much of the buffer fgets() overwrites. So if INITBUFSIZE is larger 1243 * than "most" lines, we waste time filling unused buffer slots. 100 is 1244 * surely adequate for most peoples' email archives, chewing over source code, 1245 * etc -- "regular old text files". 1246 * MAXBUFSIZE is the maximum line length that lets us get away with the less 1247 * fast (but still zippy) no-realloc, two-fgets()-call path. See above for 1248 * cautions about boosting that. 300 was chosen because the worst real-life 1249 * text-crunching job reported on Python-Dev was a mail-log crawler where over 1250 * half the lines were 254 chars. 1251 */ 1252 #define INITBUFSIZE 100 1253 #define MAXBUFSIZE 300 1254 char* p; /* temp */ 1255 char buf[MAXBUFSIZE]; 1256 PyObject* v; /* the string object result */ 1257 char* pvfree; /* address of next free slot */ 1258 char* pvend; /* address one beyond last free slot */ 1259 size_t nfree; /* # of free buffer slots; pvend-pvfree */ 1260 size_t total_v_size; /* total # of slots in buffer */ 1261 size_t increment; /* amount to increment the buffer */ 1262 size_t prev_v_size; 1263 1264 /* Optimize for normal case: avoid _PyString_Resize if at all 1265 * possible via first reading into stack buffer "buf". 1266 */ 1267 total_v_size = INITBUFSIZE; /* start small and pray */ 1268 pvfree = buf; 1269 for (;;) { 1270 FILE_BEGIN_ALLOW_THREADS(f) 1271 pvend = buf + total_v_size; 1272 nfree = pvend - pvfree; 1273 memset(pvfree, '\n', nfree); 1274 assert(nfree < INT_MAX); /* Should be atmost MAXBUFSIZE */ 1275 p = fgets(pvfree, (int)nfree, fp); 1276 FILE_END_ALLOW_THREADS(f) 1277 1278 if (p == NULL) { 1279 clearerr(fp); 1280 if (PyErr_CheckSignals()) 1281 return NULL; 1282 v = PyString_FromStringAndSize(buf, pvfree - buf); 1283 return v; 1284 } 1285 /* fgets read *something* */ 1286 p = memchr(pvfree, '\n', nfree); 1287 if (p != NULL) { 1288 /* Did the \n come from fgets or from us? 1289 * Since fgets stops at the first \n, and then writes 1290 * \0, if it's from fgets a \0 must be next. But if 1291 * that's so, it could not have come from us, since 1292 * the \n's we filled the buffer with have only more 1293 * \n's to the right. 1294 */ 1295 if (p+1 < pvend && *(p+1) == '\0') { 1296 /* It's from fgets: we win! In particular, 1297 * we haven't done any mallocs yet, and can 1298 * build the final result on the first try. 1299 */ 1300 ++p; /* include \n from fgets */ 1301 } 1302 else { 1303 /* Must be from us: fgets didn't fill the 1304 * buffer and didn't find a newline, so it 1305 * must be the last and newline-free line of 1306 * the file. 1307 */ 1308 assert(p > pvfree && *(p-1) == '\0'); 1309 --p; /* don't include \0 from fgets */ 1310 } 1311 v = PyString_FromStringAndSize(buf, p - buf); 1312 return v; 1313 } 1314 /* yuck: fgets overwrote all the newlines, i.e. the entire 1315 * buffer. So this line isn't over yet, or maybe it is but 1316 * we're exactly at EOF. If we haven't already, try using the 1317 * rest of the stack buffer. 1318 */ 1319 assert(*(pvend-1) == '\0'); 1320 if (pvfree == buf) { 1321 pvfree = pvend - 1; /* overwrite trailing null */ 1322 total_v_size = MAXBUFSIZE; 1323 } 1324 else 1325 break; 1326 } 1327 1328 /* The stack buffer isn't big enough; malloc a string object and read 1329 * into its buffer. 1330 */ 1331 total_v_size = MAXBUFSIZE << 1; 1332 v = PyString_FromStringAndSize((char*)NULL, (int)total_v_size); 1333 if (v == NULL) 1334 return v; 1335 /* copy over everything except the last null byte */ 1336 memcpy(BUF(v), buf, MAXBUFSIZE-1); 1337 pvfree = BUF(v) + MAXBUFSIZE - 1; 1338 1339 /* Keep reading stuff into v; if it ever ends successfully, break 1340 * after setting p one beyond the end of the line. The code here is 1341 * very much like the code above, except reads into v's buffer; see 1342 * the code above for detailed comments about the logic. 1343 */ 1344 for (;;) { 1345 FILE_BEGIN_ALLOW_THREADS(f) 1346 pvend = BUF(v) + total_v_size; 1347 nfree = pvend - pvfree; 1348 memset(pvfree, '\n', nfree); 1349 assert(nfree < INT_MAX); 1350 p = fgets(pvfree, (int)nfree, fp); 1351 FILE_END_ALLOW_THREADS(f) 1352 1353 if (p == NULL) { 1354 clearerr(fp); 1355 if (PyErr_CheckSignals()) { 1356 Py_DECREF(v); 1357 return NULL; 1358 } 1359 p = pvfree; 1360 break; 1361 } 1362 p = memchr(pvfree, '\n', nfree); 1363 if (p != NULL) { 1364 if (p+1 < pvend && *(p+1) == '\0') { 1365 /* \n came from fgets */ 1366 ++p; 1367 break; 1368 } 1369 /* \n came from us; last line of file, no newline */ 1370 assert(p > pvfree && *(p-1) == '\0'); 1371 --p; 1372 break; 1373 } 1374 /* expand buffer and try again */ 1375 assert(*(pvend-1) == '\0'); 1376 increment = total_v_size >> 2; /* mild exponential growth */ 1377 prev_v_size = total_v_size; 1378 total_v_size += increment; 1379 /* check for overflow */ 1380 if (total_v_size <= prev_v_size || 1381 total_v_size > PY_SSIZE_T_MAX) { 1382 PyErr_SetString(PyExc_OverflowError, 1383 "line is longer than a Python string can hold"); 1384 Py_DECREF(v); 1385 return NULL; 1386 } 1387 if (_PyString_Resize(&v, (int)total_v_size) < 0) 1388 return NULL; 1389 /* overwrite the trailing null byte */ 1390 pvfree = BUF(v) + (prev_v_size - 1); 1391 } 1392 if (BUF(v) + total_v_size != p && _PyString_Resize(&v, p - BUF(v))) 1393 return NULL; 1394 return v; 1395 #undef INITBUFSIZE 1396 #undef MAXBUFSIZE 1397 } 1398 #endif /* ifdef USE_FGETS_IN_GETLINE */ 1399 1400 /* Internal routine to get a line. 1401 Size argument interpretation: 1402 > 0: max length; 1403 <= 0: read arbitrary line 1404 */ 1405 1406 static PyObject * 1407 get_line(PyFileObject *f, int n) 1408 { 1409 FILE *fp = f->f_fp; 1410 int c; 1411 char *buf, *end; 1412 size_t total_v_size; /* total # of slots in buffer */ 1413 size_t used_v_size; /* # used slots in buffer */ 1414 size_t increment; /* amount to increment the buffer */ 1415 PyObject *v; 1416 int newlinetypes = f->f_newlinetypes; 1417 int skipnextlf = f->f_skipnextlf; 1418 int univ_newline = f->f_univ_newline; 1419 1420 #if defined(USE_FGETS_IN_GETLINE) 1421 if (n <= 0 && !univ_newline ) 1422 return getline_via_fgets(f, fp); 1423 #endif 1424 total_v_size = n > 0 ? n : 100; 1425 v = PyString_FromStringAndSize((char *)NULL, total_v_size); 1426 if (v == NULL) 1427 return NULL; 1428 buf = BUF(v); 1429 end = buf + total_v_size; 1430 1431 for (;;) { 1432 FILE_BEGIN_ALLOW_THREADS(f) 1433 FLOCKFILE(fp); 1434 if (univ_newline) { 1435 c = 'x'; /* Shut up gcc warning */ 1436 while ( buf != end && (c = GETC(fp)) != EOF ) { 1437 if (skipnextlf ) { 1438 skipnextlf = 0; 1439 if (c == '\n') { 1440 /* Seeing a \n here with 1441 * skipnextlf true means we 1442 * saw a \r before. 1443 */ 1444 newlinetypes |= NEWLINE_CRLF; 1445 c = GETC(fp); 1446 if (c == EOF) break; 1447 } else { 1448 newlinetypes |= NEWLINE_CR; 1449 } 1450 } 1451 if (c == '\r') { 1452 skipnextlf = 1; 1453 c = '\n'; 1454 } else if ( c == '\n') 1455 newlinetypes |= NEWLINE_LF; 1456 *buf++ = c; 1457 if (c == '\n') break; 1458 } 1459 if (c == EOF) { 1460 if (ferror(fp) && errno == EINTR) { 1461 FUNLOCKFILE(fp); 1462 FILE_ABORT_ALLOW_THREADS(f) 1463 f->f_newlinetypes = newlinetypes; 1464 f->f_skipnextlf = skipnextlf; 1465 1466 if (PyErr_CheckSignals()) { 1467 Py_DECREF(v); 1468 return NULL; 1469 } 1470 /* We executed Python signal handlers and got no exception. 1471 * Now back to reading the line where we left off. */ 1472 clearerr(fp); 1473 continue; 1474 } 1475 if (skipnextlf) 1476 newlinetypes |= NEWLINE_CR; 1477 } 1478 } else /* If not universal newlines use the normal loop */ 1479 while ((c = GETC(fp)) != EOF && 1480 (*buf++ = c) != '\n' && 1481 buf != end) 1482 ; 1483 FUNLOCKFILE(fp); 1484 FILE_END_ALLOW_THREADS(f) 1485 f->f_newlinetypes = newlinetypes; 1486 f->f_skipnextlf = skipnextlf; 1487 if (c == '\n') 1488 break; 1489 if (c == EOF) { 1490 if (ferror(fp)) { 1491 if (errno == EINTR) { 1492 if (PyErr_CheckSignals()) { 1493 Py_DECREF(v); 1494 return NULL; 1495 } 1496 /* We executed Python signal handlers and got no exception. 1497 * Now back to reading the line where we left off. */ 1498 clearerr(fp); 1499 continue; 1500 } 1501 PyErr_SetFromErrno(PyExc_IOError); 1502 clearerr(fp); 1503 Py_DECREF(v); 1504 return NULL; 1505 } 1506 clearerr(fp); 1507 if (PyErr_CheckSignals()) { 1508 Py_DECREF(v); 1509 return NULL; 1510 } 1511 break; 1512 } 1513 /* Must be because buf == end */ 1514 if (n > 0) 1515 break; 1516 used_v_size = total_v_size; 1517 increment = total_v_size >> 2; /* mild exponential growth */ 1518 total_v_size += increment; 1519 if (total_v_size > PY_SSIZE_T_MAX) { 1520 PyErr_SetString(PyExc_OverflowError, 1521 "line is longer than a Python string can hold"); 1522 Py_DECREF(v); 1523 return NULL; 1524 } 1525 if (_PyString_Resize(&v, total_v_size) < 0) 1526 return NULL; 1527 buf = BUF(v) + used_v_size; 1528 end = BUF(v) + total_v_size; 1529 } 1530 1531 used_v_size = buf - BUF(v); 1532 if (used_v_size != total_v_size && _PyString_Resize(&v, used_v_size)) 1533 return NULL; 1534 return v; 1535 } 1536 1537 /* External C interface */ 1538 1539 PyObject * 1540 PyFile_GetLine(PyObject *f, int n) 1541 { 1542 PyObject *result; 1543 1544 if (f == NULL) { 1545 PyErr_BadInternalCall(); 1546 return NULL; 1547 } 1548 1549 if (PyFile_Check(f)) { 1550 PyFileObject *fo = (PyFileObject *)f; 1551 if (fo->f_fp == NULL) 1552 return err_closed(); 1553 if (!fo->readable) 1554 return err_mode("reading"); 1555 /* refuse to mix with f.next() */ 1556 if (fo->f_buf != NULL && 1557 (fo->f_bufend - fo->f_bufptr) > 0 && 1558 fo->f_buf[0] != '\0') 1559 return err_iterbuffered(); 1560 result = get_line(fo, n); 1561 } 1562 else { 1563 PyObject *reader; 1564 PyObject *args; 1565 1566 reader = PyObject_GetAttrString(f, "readline"); 1567 if (reader == NULL) 1568 return NULL; 1569 if (n <= 0) 1570 args = PyTuple_New(0); 1571 else 1572 args = Py_BuildValue("(i)", n); 1573 if (args == NULL) { 1574 Py_DECREF(reader); 1575 return NULL; 1576 } 1577 result = PyEval_CallObject(reader, args); 1578 Py_DECREF(reader); 1579 Py_DECREF(args); 1580 if (result != NULL && !PyString_Check(result) && 1581 !PyUnicode_Check(result)) { 1582 Py_DECREF(result); 1583 result = NULL; 1584 PyErr_SetString(PyExc_TypeError, 1585 "object.readline() returned non-string"); 1586 } 1587 } 1588 1589 if (n < 0 && result != NULL && PyString_Check(result)) { 1590 char *s = PyString_AS_STRING(result); 1591 Py_ssize_t len = PyString_GET_SIZE(result); 1592 if (len == 0) { 1593 Py_DECREF(result); 1594 result = NULL; 1595 PyErr_SetString(PyExc_EOFError, 1596 "EOF when reading a line"); 1597 } 1598 else if (s[len-1] == '\n') { 1599 if (result->ob_refcnt == 1) { 1600 if (_PyString_Resize(&result, len-1)) 1601 return NULL; 1602 } 1603 else { 1604 PyObject *v; 1605 v = PyString_FromStringAndSize(s, len-1); 1606 Py_DECREF(result); 1607 result = v; 1608 } 1609 } 1610 } 1611 #ifdef Py_USING_UNICODE 1612 if (n < 0 && result != NULL && PyUnicode_Check(result)) { 1613 Py_UNICODE *s = PyUnicode_AS_UNICODE(result); 1614 Py_ssize_t len = PyUnicode_GET_SIZE(result); 1615 if (len == 0) { 1616 Py_DECREF(result); 1617 result = NULL; 1618 PyErr_SetString(PyExc_EOFError, 1619 "EOF when reading a line"); 1620 } 1621 else if (s[len-1] == '\n') { 1622 if (result->ob_refcnt == 1) 1623 PyUnicode_Resize(&result, len-1); 1624 else { 1625 PyObject *v; 1626 v = PyUnicode_FromUnicode(s, len-1); 1627 Py_DECREF(result); 1628 result = v; 1629 } 1630 } 1631 } 1632 #endif 1633 return result; 1634 } 1635 1636 /* Python method */ 1637 1638 static PyObject * 1639 file_readline(PyFileObject *f, PyObject *args) 1640 { 1641 int n = -1; 1642 1643 if (f->f_fp == NULL) 1644 return err_closed(); 1645 if (!f->readable) 1646 return err_mode("reading"); 1647 /* refuse to mix with f.next() */ 1648 if (f->f_buf != NULL && 1649 (f->f_bufend - f->f_bufptr) > 0 && 1650 f->f_buf[0] != '\0') 1651 return err_iterbuffered(); 1652 if (!PyArg_ParseTuple(args, "|i:readline", &n)) 1653 return NULL; 1654 if (n == 0) 1655 return PyString_FromString(""); 1656 if (n < 0) 1657 n = 0; 1658 return get_line(f, n); 1659 } 1660 1661 static PyObject * 1662 file_readlines(PyFileObject *f, PyObject *args) 1663 { 1664 long sizehint = 0; 1665 PyObject *list = NULL; 1666 PyObject *line; 1667 char small_buffer[SMALLCHUNK]; 1668 char *buffer = small_buffer; 1669 size_t buffersize = SMALLCHUNK; 1670 PyObject *big_buffer = NULL; 1671 size_t nfilled = 0; 1672 size_t nread; 1673 size_t totalread = 0; 1674 char *p, *q, *end; 1675 int err; 1676 int shortread = 0; /* bool, did the previous read come up short? */ 1677 1678 if (f->f_fp == NULL) 1679 return err_closed(); 1680 if (!f->readable) 1681 return err_mode("reading"); 1682 /* refuse to mix with f.next() */ 1683 if (f->f_buf != NULL && 1684 (f->f_bufend - f->f_bufptr) > 0 && 1685 f->f_buf[0] != '\0') 1686 return err_iterbuffered(); 1687 if (!PyArg_ParseTuple(args, "|l:readlines", &sizehint)) 1688 return NULL; 1689 if ((list = PyList_New(0)) == NULL) 1690 return NULL; 1691 for (;;) { 1692 if (shortread) 1693 nread = 0; 1694 else { 1695 FILE_BEGIN_ALLOW_THREADS(f) 1696 errno = 0; 1697 nread = Py_UniversalNewlineFread(buffer+nfilled, 1698 buffersize-nfilled, f->f_fp, (PyObject *)f); 1699 FILE_END_ALLOW_THREADS(f) 1700 shortread = (nread < buffersize-nfilled); 1701 } 1702 if (nread == 0) { 1703 sizehint = 0; 1704 if (!ferror(f->f_fp)) 1705 break; 1706 if (errno == EINTR) { 1707 if (PyErr_CheckSignals()) { 1708 goto error; 1709 } 1710 clearerr(f->f_fp); 1711 shortread = 0; 1712 continue; 1713 } 1714 PyErr_SetFromErrno(PyExc_IOError); 1715 clearerr(f->f_fp); 1716 goto error; 1717 } 1718 totalread += nread; 1719 p = (char *)memchr(buffer+nfilled, '\n', nread); 1720 if (p == NULL) { 1721 /* Need a larger buffer to fit this line */ 1722 nfilled += nread; 1723 buffersize *= 2; 1724 if (buffersize > PY_SSIZE_T_MAX) { 1725 PyErr_SetString(PyExc_OverflowError, 1726 "line is longer than a Python string can hold"); 1727 goto error; 1728 } 1729 if (big_buffer == NULL) { 1730 /* Create the big buffer */ 1731 big_buffer = PyString_FromStringAndSize( 1732 NULL, buffersize); 1733 if (big_buffer == NULL) 1734 goto error; 1735 buffer = PyString_AS_STRING(big_buffer); 1736 memcpy(buffer, small_buffer, nfilled); 1737 } 1738 else { 1739 /* Grow the big buffer */ 1740 if ( _PyString_Resize(&big_buffer, buffersize) < 0 ) 1741 goto error; 1742 buffer = PyString_AS_STRING(big_buffer); 1743 } 1744 continue; 1745 } 1746 end = buffer+nfilled+nread; 1747 q = buffer; 1748 do { 1749 /* Process complete lines */ 1750 p++; 1751 line = PyString_FromStringAndSize(q, p-q); 1752 if (line == NULL) 1753 goto error; 1754 err = PyList_Append(list, line); 1755 Py_DECREF(line); 1756 if (err != 0) 1757 goto error; 1758 q = p; 1759 p = (char *)memchr(q, '\n', end-q); 1760 } while (p != NULL); 1761 /* Move the remaining incomplete line to the start */ 1762 nfilled = end-q; 1763 memmove(buffer, q, nfilled); 1764 if (sizehint > 0) 1765 if (totalread >= (size_t)sizehint) 1766 break; 1767 } 1768 if (nfilled != 0) { 1769 /* Partial last line */ 1770 line = PyString_FromStringAndSize(buffer, nfilled); 1771 if (line == NULL) 1772 goto error; 1773 if (sizehint > 0) { 1774 /* Need to complete the last line */ 1775 PyObject *rest = get_line(f, 0); 1776 if (rest == NULL) { 1777 Py_DECREF(line); 1778 goto error; 1779 } 1780 PyString_Concat(&line, rest); 1781 Py_DECREF(rest); 1782 if (line == NULL) 1783 goto error; 1784 } 1785 err = PyList_Append(list, line); 1786 Py_DECREF(line); 1787 if (err != 0) 1788 goto error; 1789 } 1790 1791 cleanup: 1792 Py_XDECREF(big_buffer); 1793 return list; 1794 1795 error: 1796 Py_CLEAR(list); 1797 goto cleanup; 1798 } 1799 1800 static PyObject * 1801 file_write(PyFileObject *f, PyObject *args) 1802 { 1803 Py_buffer pbuf; 1804 const char *s; 1805 Py_ssize_t n, n2; 1806 PyObject *encoded = NULL; 1807 int err_flag = 0, err; 1808 1809 if (f->f_fp == NULL) 1810 return err_closed(); 1811 if (!f->writable) 1812 return err_mode("writing"); 1813 if (f->f_binary) { 1814 if (!PyArg_ParseTuple(args, "s*", &pbuf)) 1815 return NULL; 1816 s = pbuf.buf; 1817 n = pbuf.len; 1818 } 1819 else { 1820 PyObject *text; 1821 if (!PyArg_ParseTuple(args, "O", &text)) 1822 return NULL; 1823 1824 if (PyString_Check(text)) { 1825 s = PyString_AS_STRING(text); 1826 n = PyString_GET_SIZE(text); 1827 #ifdef Py_USING_UNICODE 1828 } else if (PyUnicode_Check(text)) { 1829 const char *encoding, *errors; 1830 if (f->f_encoding != Py_None) 1831 encoding = PyString_AS_STRING(f->f_encoding); 1832 else 1833 encoding = PyUnicode_GetDefaultEncoding(); 1834 if (f->f_errors != Py_None) 1835 errors = PyString_AS_STRING(f->f_errors); 1836 else 1837 errors = "strict"; 1838 encoded = PyUnicode_AsEncodedString(text, encoding, errors); 1839 if (encoded == NULL) 1840 return NULL; 1841 s = PyString_AS_STRING(encoded); 1842 n = PyString_GET_SIZE(encoded); 1843 #endif 1844 } else { 1845 if (PyObject_AsCharBuffer(text, &s, &n)) 1846 return NULL; 1847 } 1848 } 1849 f->f_softspace = 0; 1850 FILE_BEGIN_ALLOW_THREADS(f) 1851 errno = 0; 1852 n2 = fwrite(s, 1, n, f->f_fp); 1853 if (n2 != n || ferror(f->f_fp)) { 1854 err_flag = 1; 1855 err = errno; 1856 } 1857 FILE_END_ALLOW_THREADS(f) 1858 Py_XDECREF(encoded); 1859 if (f->f_binary) 1860 PyBuffer_Release(&pbuf); 1861 if (err_flag) { 1862 errno = err; 1863 PyErr_SetFromErrno(PyExc_IOError); 1864 clearerr(f->f_fp); 1865 return NULL; 1866 } 1867 Py_INCREF(Py_None); 1868 return Py_None; 1869 } 1870 1871 static PyObject * 1872 file_writelines(PyFileObject *f, PyObject *seq) 1873 { 1874 #define CHUNKSIZE 1000 1875 PyObject *list, *line; 1876 PyObject *it; /* iter(seq) */ 1877 PyObject *result; 1878 int index, islist; 1879 Py_ssize_t i, j, nwritten, len; 1880 1881 assert(seq != NULL); 1882 if (f->f_fp == NULL) 1883 return err_closed(); 1884 if (!f->writable) 1885 return err_mode("writing"); 1886 1887 result = NULL; 1888 list = NULL; 1889 islist = PyList_Check(seq); 1890 if (islist) 1891 it = NULL; 1892 else { 1893 it = PyObject_GetIter(seq); 1894 if (it == NULL) { 1895 PyErr_SetString(PyExc_TypeError, 1896 "writelines() requires an iterable argument"); 1897 return NULL; 1898 } 1899 /* From here on, fail by going to error, to reclaim "it". */ 1900 list = PyList_New(CHUNKSIZE); 1901 if (list == NULL) 1902 goto error; 1903 } 1904 1905 /* Strategy: slurp CHUNKSIZE lines into a private list, 1906 checking that they are all strings, then write that list 1907 without holding the interpreter lock, then come back for more. */ 1908 for (index = 0; ; index += CHUNKSIZE) { 1909 if (islist) { 1910 Py_XDECREF(list); 1911 list = PyList_GetSlice(seq, index, index+CHUNKSIZE); 1912 if (list == NULL) 1913 goto error; 1914 j = PyList_GET_SIZE(list); 1915 } 1916 else { 1917 for (j = 0; j < CHUNKSIZE; j++) { 1918 line = PyIter_Next(it); 1919 if (line == NULL) { 1920 if (PyErr_Occurred()) 1921 goto error; 1922 break; 1923 } 1924 PyList_SetItem(list, j, line); 1925 } 1926 /* The iterator might have closed the file on us. */ 1927 if (f->f_fp == NULL) { 1928 err_closed(); 1929 goto error; 1930 } 1931 } 1932 if (j == 0) 1933 break; 1934 1935 /* Check that all entries are indeed strings. If not, 1936 apply the same rules as for file.write() and 1937 convert the results to strings. This is slow, but 1938 seems to be the only way since all conversion APIs 1939 could potentially execute Python code. */ 1940 for (i = 0; i < j; i++) { 1941 PyObject *v = PyList_GET_ITEM(list, i); 1942 if (!PyString_Check(v)) { 1943 const char *buffer; 1944 int res; 1945 if (f->f_binary) { 1946 res = PyObject_AsReadBuffer(v, (const void**)&buffer, &len); 1947 } else { 1948 res = PyObject_AsCharBuffer(v, &buffer, &len); 1949 } 1950 if (res) { 1951 PyErr_SetString(PyExc_TypeError, 1952 "writelines() argument must be a sequence of strings"); 1953 goto error; 1954 } 1955 line = PyString_FromStringAndSize(buffer, 1956 len); 1957 if (line == NULL) 1958 goto error; 1959 Py_DECREF(v); 1960 PyList_SET_ITEM(list, i, line); 1961 } 1962 } 1963 1964 /* Since we are releasing the global lock, the 1965 following code may *not* execute Python code. */ 1966 f->f_softspace = 0; 1967 FILE_BEGIN_ALLOW_THREADS(f) 1968 errno = 0; 1969 for (i = 0; i < j; i++) { 1970 line = PyList_GET_ITEM(list, i); 1971 len = PyString_GET_SIZE(line); 1972 nwritten = fwrite(PyString_AS_STRING(line), 1973 1, len, f->f_fp); 1974 if (nwritten != len) { 1975 FILE_ABORT_ALLOW_THREADS(f) 1976 PyErr_SetFromErrno(PyExc_IOError); 1977 clearerr(f->f_fp); 1978 goto error; 1979 } 1980 } 1981 FILE_END_ALLOW_THREADS(f) 1982 1983 if (j < CHUNKSIZE) 1984 break; 1985 } 1986 1987 Py_INCREF(Py_None); 1988 result = Py_None; 1989 error: 1990 Py_XDECREF(list); 1991 Py_XDECREF(it); 1992 return result; 1993 #undef CHUNKSIZE 1994 } 1995 1996 static PyObject * 1997 file_self(PyFileObject *f) 1998 { 1999 if (f->f_fp == NULL) 2000 return err_closed(); 2001 Py_INCREF(f); 2002 return (PyObject *)f; 2003 } 2004 2005 static PyObject * 2006 file_xreadlines(PyFileObject *f) 2007 { 2008 if (PyErr_WarnPy3k("f.xreadlines() not supported in 3.x, " 2009 "try 'for line in f' instead", 1) < 0) 2010 return NULL; 2011 return file_self(f); 2012 } 2013 2014 static PyObject * 2015 file_exit(PyObject *f, PyObject *args) 2016 { 2017 PyObject *ret = PyObject_CallMethod(f, "close", NULL); 2018 if (!ret) 2019 /* If error occurred, pass through */ 2020 return NULL; 2021 Py_DECREF(ret); 2022 /* We cannot return the result of close since a true 2023 * value will be interpreted as "yes, swallow the 2024 * exception if one was raised inside the with block". */ 2025 Py_RETURN_NONE; 2026 } 2027 2028 PyDoc_STRVAR(readline_doc, 2029 "readline([size]) -> next line from the file, as a string.\n" 2030 "\n" 2031 "Retain newline. A non-negative size argument limits the maximum\n" 2032 "number of bytes to return (an incomplete line may be returned then).\n" 2033 "Return an empty string at EOF."); 2034 2035 PyDoc_STRVAR(read_doc, 2036 "read([size]) -> read at most size bytes, returned as a string.\n" 2037 "\n" 2038 "If the size argument is negative or omitted, read until EOF is reached.\n" 2039 "Notice that when in non-blocking mode, less data than what was requested\n" 2040 "may be returned, even if no size parameter was given."); 2041 2042 PyDoc_STRVAR(write_doc, 2043 "write(str) -> None. Write string str to file.\n" 2044 "\n" 2045 "Note that due to buffering, flush() or close() may be needed before\n" 2046 "the file on disk reflects the data written."); 2047 2048 PyDoc_STRVAR(fileno_doc, 2049 "fileno() -> integer \"file descriptor\".\n" 2050 "\n" 2051 "This is needed for lower-level file interfaces, such os.read()."); 2052 2053 PyDoc_STRVAR(seek_doc, 2054 "seek(offset[, whence]) -> None. Move to new file position.\n" 2055 "\n" 2056 "Argument offset is a byte count. Optional argument whence defaults to\n" 2057 "0 (offset from start of file, offset should be >= 0); other values are 1\n" 2058 "(move relative to current position, positive or negative), and 2 (move\n" 2059 "relative to end of file, usually negative, although many platforms allow\n" 2060 "seeking beyond the end of a file). If the file is opened in text mode,\n" 2061 "only offsets returned by tell() are legal. Use of other offsets causes\n" 2062 "undefined behavior." 2063 "\n" 2064 "Note that not all file objects are seekable."); 2065 2066 #ifdef HAVE_FTRUNCATE 2067 PyDoc_STRVAR(truncate_doc, 2068 "truncate([size]) -> None. Truncate the file to at most size bytes.\n" 2069 "\n" 2070 "Size defaults to the current file position, as returned by tell()."); 2071 #endif 2072 2073 PyDoc_STRVAR(tell_doc, 2074 "tell() -> current file position, an integer (may be a long integer)."); 2075 2076 PyDoc_STRVAR(readinto_doc, 2077 "readinto() -> Undocumented. Don't use this; it may go away."); 2078 2079 PyDoc_STRVAR(readlines_doc, 2080 "readlines([size]) -> list of strings, each a line from the file.\n" 2081 "\n" 2082 "Call readline() repeatedly and return a list of the lines so read.\n" 2083 "The optional size argument, if given, is an approximate bound on the\n" 2084 "total number of bytes in the lines returned."); 2085 2086 PyDoc_STRVAR(xreadlines_doc, 2087 "xreadlines() -> returns self.\n" 2088 "\n" 2089 "For backward compatibility. File objects now include the performance\n" 2090 "optimizations previously implemented in the xreadlines module."); 2091 2092 PyDoc_STRVAR(writelines_doc, 2093 "writelines(sequence_of_strings) -> None. Write the strings to the file.\n" 2094 "\n" 2095 "Note that newlines are not added. The sequence can be any iterable object\n" 2096 "producing strings. This is equivalent to calling write() for each string."); 2097 2098 PyDoc_STRVAR(flush_doc, 2099 "flush() -> None. Flush the internal I/O buffer."); 2100 2101 PyDoc_STRVAR(close_doc, 2102 "close() -> None or (perhaps) an integer. Close the file.\n" 2103 "\n" 2104 "Sets data attribute .closed to True. A closed file cannot be used for\n" 2105 "further I/O operations. close() may be called more than once without\n" 2106 "error. Some kinds of file objects (for example, opened by popen())\n" 2107 "may return an exit status upon closing."); 2108 2109 PyDoc_STRVAR(isatty_doc, 2110 "isatty() -> true or false. True if the file is connected to a tty device."); 2111 2112 PyDoc_STRVAR(enter_doc, 2113 "__enter__() -> self."); 2114 2115 PyDoc_STRVAR(exit_doc, 2116 "__exit__(*excinfo) -> None. Closes the file."); 2117 2118 static PyMethodDef file_methods[] = { 2119 {"readline", (PyCFunction)file_readline, METH_VARARGS, readline_doc}, 2120 {"read", (PyCFunction)file_read, METH_VARARGS, read_doc}, 2121 {"write", (PyCFunction)file_write, METH_VARARGS, write_doc}, 2122 {"fileno", (PyCFunction)file_fileno, METH_NOARGS, fileno_doc}, 2123 {"seek", (PyCFunction)file_seek, METH_VARARGS, seek_doc}, 2124 #ifdef HAVE_FTRUNCATE 2125 {"truncate", (PyCFunction)file_truncate, METH_VARARGS, truncate_doc}, 2126 #endif 2127 {"tell", (PyCFunction)file_tell, METH_NOARGS, tell_doc}, 2128 {"readinto", (PyCFunction)file_readinto, METH_VARARGS, readinto_doc}, 2129 {"readlines", (PyCFunction)file_readlines, METH_VARARGS, readlines_doc}, 2130 {"xreadlines",(PyCFunction)file_xreadlines, METH_NOARGS, xreadlines_doc}, 2131 {"writelines",(PyCFunction)file_writelines, METH_O, writelines_doc}, 2132 {"flush", (PyCFunction)file_flush, METH_NOARGS, flush_doc}, 2133 {"close", (PyCFunction)file_close, METH_NOARGS, close_doc}, 2134 {"isatty", (PyCFunction)file_isatty, METH_NOARGS, isatty_doc}, 2135 {"__enter__", (PyCFunction)file_self, METH_NOARGS, enter_doc}, 2136 {"__exit__", (PyCFunction)file_exit, METH_VARARGS, exit_doc}, 2137 {NULL, NULL} /* sentinel */ 2138 }; 2139 2140 #define OFF(x) offsetof(PyFileObject, x) 2141 2142 static PyMemberDef file_memberlist[] = { 2143 {"mode", T_OBJECT, OFF(f_mode), RO, 2144 "file mode ('r', 'U', 'w', 'a', possibly with 'b' or '+' added)"}, 2145 {"name", T_OBJECT, OFF(f_name), RO, 2146 "file name"}, 2147 {"encoding", T_OBJECT, OFF(f_encoding), RO, 2148 "file encoding"}, 2149 {"errors", T_OBJECT, OFF(f_errors), RO, 2150 "Unicode error handler"}, 2151 /* getattr(f, "closed") is implemented without this table */ 2152 {NULL} /* Sentinel */ 2153 }; 2154 2155 static PyObject * 2156 get_closed(PyFileObject *f, void *closure) 2157 { 2158 return PyBool_FromLong((long)(f->f_fp == 0)); 2159 } 2160 static PyObject * 2161 get_newlines(PyFileObject *f, void *closure) 2162 { 2163 switch (f->f_newlinetypes) { 2164 case NEWLINE_UNKNOWN: 2165 Py_INCREF(Py_None); 2166 return Py_None; 2167 case NEWLINE_CR: 2168 return PyString_FromString("\r"); 2169 case NEWLINE_LF: 2170 return PyString_FromString("\n"); 2171 case NEWLINE_CR|NEWLINE_LF: 2172 return Py_BuildValue("(ss)", "\r", "\n"); 2173 case NEWLINE_CRLF: 2174 return PyString_FromString("\r\n"); 2175 case NEWLINE_CR|NEWLINE_CRLF: 2176 return Py_BuildValue("(ss)", "\r", "\r\n"); 2177 case NEWLINE_LF|NEWLINE_CRLF: 2178 return Py_BuildValue("(ss)", "\n", "\r\n"); 2179 case NEWLINE_CR|NEWLINE_LF|NEWLINE_CRLF: 2180 return Py_BuildValue("(sss)", "\r", "\n", "\r\n"); 2181 default: 2182 PyErr_Format(PyExc_SystemError, 2183 "Unknown newlines value 0x%x\n", 2184 f->f_newlinetypes); 2185 return NULL; 2186 } 2187 } 2188 2189 static PyObject * 2190 get_softspace(PyFileObject *f, void *closure) 2191 { 2192 if (PyErr_WarnPy3k("file.softspace not supported in 3.x", 1) < 0) 2193 return NULL; 2194 return PyInt_FromLong(f->f_softspace); 2195 } 2196 2197 static int 2198 set_softspace(PyFileObject *f, PyObject *value) 2199 { 2200 int new; 2201 if (PyErr_WarnPy3k("file.softspace not supported in 3.x", 1) < 0) 2202 return -1; 2203 2204 if (value == NULL) { 2205 PyErr_SetString(PyExc_TypeError, 2206 "can't delete softspace attribute"); 2207 return -1; 2208 } 2209 2210 new = PyInt_AsLong(value); 2211 if (new == -1 && PyErr_Occurred()) 2212 return -1; 2213 f->f_softspace = new; 2214 return 0; 2215 } 2216 2217 static PyGetSetDef file_getsetlist[] = { 2218 {"closed", (getter)get_closed, NULL, "True if the file is closed"}, 2219 {"newlines", (getter)get_newlines, NULL, 2220 "end-of-line convention used in this file"}, 2221 {"softspace", (getter)get_softspace, (setter)set_softspace, 2222 "flag indicating that a space needs to be printed; used by print"}, 2223 {0}, 2224 }; 2225 2226 static void 2227 drop_readahead(PyFileObject *f) 2228 { 2229 if (f->f_buf != NULL) { 2230 PyMem_Free(f->f_buf); 2231 f->f_buf = NULL; 2232 } 2233 } 2234 2235 /* Make sure that file has a readahead buffer with at least one byte 2236 (unless at EOF) and no more than bufsize. Returns negative value on 2237 error, will set MemoryError if bufsize bytes cannot be allocated. */ 2238 static int 2239 readahead(PyFileObject *f, Py_ssize_t bufsize) 2240 { 2241 Py_ssize_t chunksize; 2242 2243 if (f->f_buf != NULL) { 2244 if( (f->f_bufend - f->f_bufptr) >= 1) 2245 return 0; 2246 else 2247 drop_readahead(f); 2248 } 2249 if ((f->f_buf = (char *)PyMem_Malloc(bufsize)) == NULL) { 2250 PyErr_NoMemory(); 2251 return -1; 2252 } 2253 FILE_BEGIN_ALLOW_THREADS(f) 2254 errno = 0; 2255 chunksize = Py_UniversalNewlineFread( 2256 f->f_buf, bufsize, f->f_fp, (PyObject *)f); 2257 FILE_END_ALLOW_THREADS(f) 2258 if (chunksize == 0) { 2259 if (ferror(f->f_fp)) { 2260 PyErr_SetFromErrno(PyExc_IOError); 2261 clearerr(f->f_fp); 2262 drop_readahead(f); 2263 return -1; 2264 } 2265 } 2266 f->f_bufptr = f->f_buf; 2267 f->f_bufend = f->f_buf + chunksize; 2268 return 0; 2269 } 2270 2271 /* Used by file_iternext. The returned string will start with 'skip' 2272 uninitialized bytes followed by the remainder of the line. Don't be 2273 horrified by the recursive call: maximum recursion depth is limited by 2274 logarithmic buffer growth to about 50 even when reading a 1gb line. */ 2275 2276 static PyStringObject * 2277 readahead_get_line_skip(PyFileObject *f, Py_ssize_t skip, Py_ssize_t bufsize) 2278 { 2279 PyStringObject* s; 2280 char *bufptr; 2281 char *buf; 2282 Py_ssize_t len; 2283 2284 if (f->f_buf == NULL) 2285 if (readahead(f, bufsize) < 0) 2286 return NULL; 2287 2288 len = f->f_bufend - f->f_bufptr; 2289 if (len == 0) 2290 return (PyStringObject *) 2291 PyString_FromStringAndSize(NULL, skip); 2292 bufptr = (char *)memchr(f->f_bufptr, '\n', len); 2293 if (bufptr != NULL) { 2294 bufptr++; /* Count the '\n' */ 2295 len = bufptr - f->f_bufptr; 2296 s = (PyStringObject *) 2297 PyString_FromStringAndSize(NULL, skip + len); 2298 if (s == NULL) 2299 return NULL; 2300 memcpy(PyString_AS_STRING(s) + skip, f->f_bufptr, len); 2301 f->f_bufptr = bufptr; 2302 if (bufptr == f->f_bufend) 2303 drop_readahead(f); 2304 } else { 2305 bufptr = f->f_bufptr; 2306 buf = f->f_buf; 2307 f->f_buf = NULL; /* Force new readahead buffer */ 2308 assert(len <= PY_SSIZE_T_MAX - skip); 2309 s = readahead_get_line_skip(f, skip + len, bufsize + (bufsize>>2)); 2310 if (s == NULL) { 2311 PyMem_Free(buf); 2312 return NULL; 2313 } 2314 memcpy(PyString_AS_STRING(s) + skip, bufptr, len); 2315 PyMem_Free(buf); 2316 } 2317 return s; 2318 } 2319 2320 /* A larger buffer size may actually decrease performance. */ 2321 #define READAHEAD_BUFSIZE 8192 2322 2323 static PyObject * 2324 file_iternext(PyFileObject *f) 2325 { 2326 PyStringObject* l; 2327 2328 if (f->f_fp == NULL) 2329 return err_closed(); 2330 if (!f->readable) 2331 return err_mode("reading"); 2332 2333 l = readahead_get_line_skip(f, 0, READAHEAD_BUFSIZE); 2334 if (l == NULL || PyString_GET_SIZE(l) == 0) { 2335 Py_XDECREF(l); 2336 return NULL; 2337 } 2338 return (PyObject *)l; 2339 } 2340 2341 2342 static PyObject * 2343 file_new(PyTypeObject *type, PyObject *args, PyObject *kwds) 2344 { 2345 PyObject *self; 2346 static PyObject *not_yet_string; 2347 2348 assert(type != NULL && type->tp_alloc != NULL); 2349 2350 if (not_yet_string == NULL) { 2351 not_yet_string = PyString_InternFromString("<uninitialized file>"); 2352 if (not_yet_string == NULL) 2353 return NULL; 2354 } 2355 2356 self = type->tp_alloc(type, 0); 2357 if (self != NULL) { 2358 /* Always fill in the name and mode, so that nobody else 2359 needs to special-case NULLs there. */ 2360 Py_INCREF(not_yet_string); 2361 ((PyFileObject *)self)->f_name = not_yet_string; 2362 Py_INCREF(not_yet_string); 2363 ((PyFileObject *)self)->f_mode = not_yet_string; 2364 Py_INCREF(Py_None); 2365 ((PyFileObject *)self)->f_encoding = Py_None; 2366 Py_INCREF(Py_None); 2367 ((PyFileObject *)self)->f_errors = Py_None; 2368 ((PyFileObject *)self)->weakreflist = NULL; 2369 ((PyFileObject *)self)->unlocked_count = 0; 2370 } 2371 return self; 2372 } 2373 2374 static int 2375 file_init(PyObject *self, PyObject *args, PyObject *kwds) 2376 { 2377 PyFileObject *foself = (PyFileObject *)self; 2378 int ret = 0; 2379 static char *kwlist[] = {"name", "mode", "buffering", 0}; 2380 char *name = NULL; 2381 char *mode = "r"; 2382 int bufsize = -1; 2383 int wideargument = 0; 2384 #ifdef MS_WINDOWS 2385 PyObject *po; 2386 #endif 2387 2388 assert(PyFile_Check(self)); 2389 if (foself->f_fp != NULL) { 2390 /* Have to close the existing file first. */ 2391 PyObject *closeresult = file_close(foself); 2392 if (closeresult == NULL) 2393 return -1; 2394 Py_DECREF(closeresult); 2395 } 2396 2397 #ifdef MS_WINDOWS 2398 if (PyArg_ParseTupleAndKeywords(args, kwds, "U|si:file", 2399 kwlist, &po, &mode, &bufsize)) { 2400 wideargument = 1; 2401 if (fill_file_fields(foself, NULL, po, mode, 2402 fclose) == NULL) 2403 goto Error; 2404 } else { 2405 /* Drop the argument parsing error as narrow 2406 strings are also valid. */ 2407 PyErr_Clear(); 2408 } 2409 #endif 2410 2411 if (!wideargument) { 2412 PyObject *o_name; 2413 2414 if (!PyArg_ParseTupleAndKeywords(args, kwds, "et|si:file", kwlist, 2415 Py_FileSystemDefaultEncoding, 2416 &name, 2417 &mode, &bufsize)) 2418 return -1; 2419 2420 /* We parse again to get the name as a PyObject */ 2421 if (!PyArg_ParseTupleAndKeywords(args, kwds, "O|si:file", 2422 kwlist, &o_name, &mode, 2423 &bufsize)) 2424 goto Error; 2425 2426 if (fill_file_fields(foself, NULL, o_name, mode, 2427 fclose) == NULL) 2428 goto Error; 2429 } 2430 if (open_the_file(foself, name, mode) == NULL) 2431 goto Error; 2432 foself->f_setbuf = NULL; 2433 PyFile_SetBufSize(self, bufsize); 2434 goto Done; 2435 2436 Error: 2437 ret = -1; 2438 /* fall through */ 2439 Done: 2440 PyMem_Free(name); /* free the encoded string */ 2441 return ret; 2442 } 2443 2444 PyDoc_VAR(file_doc) = 2445 PyDoc_STR( 2446 "file(name[, mode[, buffering]]) -> file object\n" 2447 "\n" 2448 "Open a file. The mode can be 'r', 'w' or 'a' for reading (default),\n" 2449 "writing or appending. The file will be created if it doesn't exist\n" 2450 "when opened for writing or appending; it will be truncated when\n" 2451 "opened for writing. Add a 'b' to the mode for binary files.\n" 2452 "Add a '+' to the mode to allow simultaneous reading and writing.\n" 2453 "If the buffering argument is given, 0 means unbuffered, 1 means line\n" 2454 "buffered, and larger numbers specify the buffer size. The preferred way\n" 2455 "to open a file is with the builtin open() function.\n" 2456 ) 2457 PyDoc_STR( 2458 "Add a 'U' to mode to open the file for input with universal newline\n" 2459 "support. Any line ending in the input file will be seen as a '\\n'\n" 2460 "in Python. Also, a file so opened gains the attribute 'newlines';\n" 2461 "the value for this attribute is one of None (no newline read yet),\n" 2462 "'\\r', '\\n', '\\r\\n' or a tuple containing all the newline types seen.\n" 2463 "\n" 2464 "'U' cannot be combined with 'w' or '+' mode.\n" 2465 ); 2466 2467 PyTypeObject PyFile_Type = { 2468 PyVarObject_HEAD_INIT(&PyType_Type, 0) 2469 "file", 2470 sizeof(PyFileObject), 2471 0, 2472 (destructor)file_dealloc, /* tp_dealloc */ 2473 0, /* tp_print */ 2474 0, /* tp_getattr */ 2475 0, /* tp_setattr */ 2476 0, /* tp_compare */ 2477 (reprfunc)file_repr, /* tp_repr */ 2478 0, /* tp_as_number */ 2479 0, /* tp_as_sequence */ 2480 0, /* tp_as_mapping */ 2481 0, /* tp_hash */ 2482 0, /* tp_call */ 2483 0, /* tp_str */ 2484 PyObject_GenericGetAttr, /* tp_getattro */ 2485 /* softspace is writable: we must supply tp_setattro */ 2486 PyObject_GenericSetAttr, /* tp_setattro */ 2487 0, /* tp_as_buffer */ 2488 Py_TPFLAGS_DEFAULT | Py_TPFLAGS_BASETYPE | Py_TPFLAGS_HAVE_WEAKREFS, /* tp_flags */ 2489 file_doc, /* tp_doc */ 2490 0, /* tp_traverse */ 2491 0, /* tp_clear */ 2492 0, /* tp_richcompare */ 2493 offsetof(PyFileObject, weakreflist), /* tp_weaklistoffset */ 2494 (getiterfunc)file_self, /* tp_iter */ 2495 (iternextfunc)file_iternext, /* tp_iternext */ 2496 file_methods, /* tp_methods */ 2497 file_memberlist, /* tp_members */ 2498 file_getsetlist, /* tp_getset */ 2499 0, /* tp_base */ 2500 0, /* tp_dict */ 2501 0, /* tp_descr_get */ 2502 0, /* tp_descr_set */ 2503 0, /* tp_dictoffset */ 2504 file_init, /* tp_init */ 2505 PyType_GenericAlloc, /* tp_alloc */ 2506 file_new, /* tp_new */ 2507 PyObject_Del, /* tp_free */ 2508 }; 2509 2510 /* Interface for the 'soft space' between print items. */ 2511 2512 int 2513 PyFile_SoftSpace(PyObject *f, int newflag) 2514 { 2515 long oldflag = 0; 2516 if (f == NULL) { 2517 /* Do nothing */ 2518 } 2519 else if (PyFile_Check(f)) { 2520 oldflag = ((PyFileObject *)f)->f_softspace; 2521 ((PyFileObject *)f)->f_softspace = newflag; 2522 } 2523 else { 2524 PyObject *v; 2525 v = PyObject_GetAttrString(f, "softspace"); 2526 if (v == NULL) 2527 PyErr_Clear(); 2528 else { 2529 if (PyInt_Check(v)) 2530 oldflag = PyInt_AsLong(v); 2531 assert(oldflag < INT_MAX); 2532 Py_DECREF(v); 2533 } 2534 v = PyInt_FromLong((long)newflag); 2535 if (v == NULL) 2536 PyErr_Clear(); 2537 else { 2538 if (PyObject_SetAttrString(f, "softspace", v) != 0) 2539 PyErr_Clear(); 2540 Py_DECREF(v); 2541 } 2542 } 2543 return (int)oldflag; 2544 } 2545 2546 /* Interfaces to write objects/strings to file-like objects */ 2547 2548 int 2549 PyFile_WriteObject(PyObject *v, PyObject *f, int flags) 2550 { 2551 PyObject *writer, *value, *args, *result; 2552 if (f == NULL) { 2553 PyErr_SetString(PyExc_TypeError, "writeobject with NULL file"); 2554 return -1; 2555 } 2556 else if (PyFile_Check(f)) { 2557 PyFileObject *fobj = (PyFileObject *) f; 2558 #ifdef Py_USING_UNICODE 2559 PyObject *enc = fobj->f_encoding; 2560 int result; 2561 #endif 2562 if (fobj->f_fp == NULL) { 2563 err_closed(); 2564 return -1; 2565 } 2566 #ifdef Py_USING_UNICODE 2567 if ((flags & Py_PRINT_RAW) && 2568 PyUnicode_Check(v) && enc != Py_None) { 2569 char *cenc = PyString_AS_STRING(enc); 2570 char *errors = fobj->f_errors == Py_None ? 2571 "strict" : PyString_AS_STRING(fobj->f_errors); 2572 value = PyUnicode_AsEncodedString(v, cenc, errors); 2573 if (value == NULL) 2574 return -1; 2575 } else { 2576 value = v; 2577 Py_INCREF(value); 2578 } 2579 result = file_PyObject_Print(value, fobj, flags); 2580 Py_DECREF(value); 2581 return result; 2582 #else 2583 return file_PyObject_Print(v, fobj, flags); 2584 #endif 2585 } 2586 writer = PyObject_GetAttrString(f, "write"); 2587 if (writer == NULL) 2588 return -1; 2589 if (flags & Py_PRINT_RAW) { 2590 if (PyUnicode_Check(v)) { 2591 value = v; 2592 Py_INCREF(value); 2593 } else 2594 value = PyObject_Str(v); 2595 } 2596 else 2597 value = PyObject_Repr(v); 2598 if (value == NULL) { 2599 Py_DECREF(writer); 2600 return -1; 2601 } 2602 args = PyTuple_Pack(1, value); 2603 if (args == NULL) { 2604 Py_DECREF(value); 2605 Py_DECREF(writer); 2606 return -1; 2607 } 2608 result = PyEval_CallObject(writer, args); 2609 Py_DECREF(args); 2610 Py_DECREF(value); 2611 Py_DECREF(writer); 2612 if (result == NULL) 2613 return -1; 2614 Py_DECREF(result); 2615 return 0; 2616 } 2617 2618 int 2619 PyFile_WriteString(const char *s, PyObject *f) 2620 { 2621 2622 if (f == NULL) { 2623 /* Should be caused by a pre-existing error */ 2624 if (!PyErr_Occurred()) 2625 PyErr_SetString(PyExc_SystemError, 2626 "null file for PyFile_WriteString"); 2627 return -1; 2628 } 2629 else if (PyFile_Check(f)) { 2630 PyFileObject *fobj = (PyFileObject *) f; 2631 FILE *fp = PyFile_AsFile(f); 2632 if (fp == NULL) { 2633 err_closed(); 2634 return -1; 2635 } 2636 FILE_BEGIN_ALLOW_THREADS(fobj) 2637 fputs(s, fp); 2638 FILE_END_ALLOW_THREADS(fobj) 2639 return 0; 2640 } 2641 else if (!PyErr_Occurred()) { 2642 PyObject *v = PyString_FromString(s); 2643 int err; 2644 if (v == NULL) 2645 return -1; 2646 err = PyFile_WriteObject(v, f, Py_PRINT_RAW); 2647 Py_DECREF(v); 2648 return err; 2649 } 2650 else 2651 return -1; 2652 } 2653 2654 /* Try to get a file-descriptor from a Python object. If the object 2655 is an integer or long integer, its value is returned. If not, the 2656 object's fileno() method is called if it exists; the method must return 2657 an integer or long integer, which is returned as the file descriptor value. 2658 -1 is returned on failure. 2659 */ 2660 2661 int PyObject_AsFileDescriptor(PyObject *o) 2662 { 2663 int fd; 2664 PyObject *meth; 2665 2666 if (PyInt_Check(o)) { 2667 fd = _PyInt_AsInt(o); 2668 } 2669 else if (PyLong_Check(o)) { 2670 fd = _PyLong_AsInt(o); 2671 } 2672 else if ((meth = PyObject_GetAttrString(o, "fileno")) != NULL) 2673 { 2674 PyObject *fno = PyEval_CallObject(meth, NULL); 2675 Py_DECREF(meth); 2676 if (fno == NULL) 2677 return -1; 2678 2679 if (PyInt_Check(fno)) { 2680 fd = _PyInt_AsInt(fno); 2681 Py_DECREF(fno); 2682 } 2683 else if (PyLong_Check(fno)) { 2684 fd = _PyLong_AsInt(fno); 2685 Py_DECREF(fno); 2686 } 2687 else { 2688 PyErr_SetString(PyExc_TypeError, 2689 "fileno() returned a non-integer"); 2690 Py_DECREF(fno); 2691 return -1; 2692 } 2693 } 2694 else { 2695 PyErr_SetString(PyExc_TypeError, 2696 "argument must be an int, or have a fileno() method."); 2697 return -1; 2698 } 2699 2700 if (fd < 0) { 2701 PyErr_Format(PyExc_ValueError, 2702 "file descriptor cannot be a negative integer (%i)", 2703 fd); 2704 return -1; 2705 } 2706 return fd; 2707 } 2708 2709 /* From here on we need access to the real fgets and fread */ 2710 #undef fgets 2711 #undef fread 2712 2713 /* 2714 ** Py_UniversalNewlineFgets is an fgets variation that understands 2715 ** all of \r, \n and \r\n conventions. 2716 ** The stream should be opened in binary mode. 2717 ** If fobj is NULL the routine always does newline conversion, and 2718 ** it may peek one char ahead to gobble the second char in \r\n. 2719 ** If fobj is non-NULL it must be a PyFileObject. In this case there 2720 ** is no readahead but in stead a flag is used to skip a following 2721 ** \n on the next read. Also, if the file is open in binary mode 2722 ** the whole conversion is skipped. Finally, the routine keeps track of 2723 ** the different types of newlines seen. 2724 ** Note that we need no error handling: fgets() treats error and eof 2725 ** identically. 2726 */ 2727 char * 2728 Py_UniversalNewlineFgets(char *buf, int n, FILE *stream, PyObject *fobj) 2729 { 2730 char *p = buf; 2731 int c; 2732 int newlinetypes = 0; 2733 int skipnextlf = 0; 2734 int univ_newline = 1; 2735 2736 if (fobj) { 2737 if (!PyFile_Check(fobj)) { 2738 errno = ENXIO; /* What can you do... */ 2739 return NULL; 2740 } 2741 univ_newline = ((PyFileObject *)fobj)->f_univ_newline; 2742 if ( !univ_newline ) 2743 return fgets(buf, n, stream); 2744 newlinetypes = ((PyFileObject *)fobj)->f_newlinetypes; 2745 skipnextlf = ((PyFileObject *)fobj)->f_skipnextlf; 2746 } 2747 FLOCKFILE(stream); 2748 c = 'x'; /* Shut up gcc warning */ 2749 while (--n > 0 && (c = GETC(stream)) != EOF ) { 2750 if (skipnextlf ) { 2751 skipnextlf = 0; 2752 if (c == '\n') { 2753 /* Seeing a \n here with skipnextlf true 2754 ** means we saw a \r before. 2755 */ 2756 newlinetypes |= NEWLINE_CRLF; 2757 c = GETC(stream); 2758 if (c == EOF) break; 2759 } else { 2760 /* 2761 ** Note that c == EOF also brings us here, 2762 ** so we're okay if the last char in the file 2763 ** is a CR. 2764 */ 2765 newlinetypes |= NEWLINE_CR; 2766 } 2767 } 2768 if (c == '\r') { 2769 /* A \r is translated into a \n, and we skip 2770 ** an adjacent \n, if any. We don't set the 2771 ** newlinetypes flag until we've seen the next char. 2772 */ 2773 skipnextlf = 1; 2774 c = '\n'; 2775 } else if ( c == '\n') { 2776 newlinetypes |= NEWLINE_LF; 2777 } 2778 *p++ = c; 2779 if (c == '\n') break; 2780 } 2781 if ( c == EOF && skipnextlf ) 2782 newlinetypes |= NEWLINE_CR; 2783 FUNLOCKFILE(stream); 2784 *p = '\0'; 2785 if (fobj) { 2786 ((PyFileObject *)fobj)->f_newlinetypes = newlinetypes; 2787 ((PyFileObject *)fobj)->f_skipnextlf = skipnextlf; 2788 } else if ( skipnextlf ) { 2789 /* If we have no file object we cannot save the 2790 ** skipnextlf flag. We have to readahead, which 2791 ** will cause a pause if we're reading from an 2792 ** interactive stream, but that is very unlikely 2793 ** unless we're doing something silly like 2794 ** execfile("/dev/tty"). 2795 */ 2796 c = GETC(stream); 2797 if ( c != '\n' ) 2798 ungetc(c, stream); 2799 } 2800 if (p == buf) 2801 return NULL; 2802 return buf; 2803 } 2804 2805 /* 2806 ** Py_UniversalNewlineFread is an fread variation that understands 2807 ** all of \r, \n and \r\n conventions. 2808 ** The stream should be opened in binary mode. 2809 ** fobj must be a PyFileObject. In this case there 2810 ** is no readahead but in stead a flag is used to skip a following 2811 ** \n on the next read. Also, if the file is open in binary mode 2812 ** the whole conversion is skipped. Finally, the routine keeps track of 2813 ** the different types of newlines seen. 2814 */ 2815 size_t 2816 Py_UniversalNewlineFread(char *buf, size_t n, 2817 FILE *stream, PyObject *fobj) 2818 { 2819 char *dst = buf; 2820 PyFileObject *f = (PyFileObject *)fobj; 2821 int newlinetypes, skipnextlf; 2822 2823 assert(buf != NULL); 2824 assert(stream != NULL); 2825 2826 if (!fobj || !PyFile_Check(fobj)) { 2827 errno = ENXIO; /* What can you do... */ 2828 return 0; 2829 } 2830 if (!f->f_univ_newline) 2831 return fread(buf, 1, n, stream); 2832 newlinetypes = f->f_newlinetypes; 2833 skipnextlf = f->f_skipnextlf; 2834 /* Invariant: n is the number of bytes remaining to be filled 2835 * in the buffer. 2836 */ 2837 while (n) { 2838 size_t nread; 2839 int shortread; 2840 char *src = dst; 2841 2842 nread = fread(dst, 1, n, stream); 2843 assert(nread <= n); 2844 if (nread == 0) 2845 break; 2846 2847 n -= nread; /* assuming 1 byte out for each in; will adjust */ 2848 shortread = n != 0; /* true iff EOF or error */ 2849 while (nread--) { 2850 char c = *src++; 2851 if (c == '\r') { 2852 /* Save as LF and set flag to skip next LF. */ 2853 *dst++ = '\n'; 2854 skipnextlf = 1; 2855 } 2856 else if (skipnextlf && c == '\n') { 2857 /* Skip LF, and remember we saw CR LF. */ 2858 skipnextlf = 0; 2859 newlinetypes |= NEWLINE_CRLF; 2860 ++n; 2861 } 2862 else { 2863 /* Normal char to be stored in buffer. Also 2864 * update the newlinetypes flag if either this 2865 * is an LF or the previous char was a CR. 2866 */ 2867 if (c == '\n') 2868 newlinetypes |= NEWLINE_LF; 2869 else if (skipnextlf) 2870 newlinetypes |= NEWLINE_CR; 2871 *dst++ = c; 2872 skipnextlf = 0; 2873 } 2874 } 2875 if (shortread) { 2876 /* If this is EOF, update type flags. */ 2877 if (skipnextlf && feof(stream)) 2878 newlinetypes |= NEWLINE_CR; 2879 break; 2880 } 2881 } 2882 f->f_newlinetypes = newlinetypes; 2883 f->f_skipnextlf = skipnextlf; 2884 return dst - buf; 2885 } 2886 2887 #ifdef __cplusplus 2888 } 2889 #endif 2890