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