convert.c

   1 #include "cache.h"
   2 #include "attr.h"
   3 #include "run-command.h"
   4 #include "quote.h"
   5
   6 /*
   7  * convert.c - convert a file when checking it out and checking it in.
   8  *
   9  * This should use the pathname to decide on whether it wants to do some
  10  * more interesting conversions (automatic gzip/unzip, general format
  11  * conversions etc etc), but by default it just does automatic CRLF<->LF
  12  * translation when the "text" attribute or "auto_crlf" option is set.
  13  */
  14
  15 enum crlf_action {
  16         CRLF_GUESS = -1,
  17         CRLF_BINARY = 0,
  18         CRLF_TEXT,
  19         CRLF_INPUT,
  20         CRLF_CRLF,
  21         CRLF_AUTO
  22 };
  23
  24 struct text_stat {
  25         /* NUL, CR, LF and CRLF counts */
  26         unsigned nul, cr, lf, crlf;
  27
  28         /* These are just approximations! */
  29         unsigned printable, nonprintable;
  30 };
  31
  32 static void gather_stats(const char *buf, unsigned long size, struct text_stat *stats)
  33 {
  34         unsigned long i;
  35
  36         memset(stats, 0, sizeof(*stats));
  37
  38         for (i = 0; i < size; i++) {
  39                 unsigned char c = buf[i];
  40                 if (c == '\r') {
  41                         stats->cr++;
  42                         if (i+1 < size && buf[i+1] == '\n')
  43                                 stats->crlf++;
  44                         continue;
  45                 }
  46                 if (c == '\n') {
  47                         stats->lf++;
  48                         continue;
  49                 }
  50                 if (c == 127)
  51                         /* DEL */
  52                         stats->nonprintable++;
  53                 else if (c < 32) {
  54                         switch (c) {
  55                                 /* BS, HT, ESC and FF */
  56                         case '\b': case '\t': case '\033': case '\014':
  57                                 stats->printable++;
  58                                 break;
  59                         case 0:
  60                                 stats->nul++;
  61                                 /* fall through */
  62                         default:
  63                                 stats->nonprintable++;
  64                         }
  65                 }
  66                 else
  67                         stats->printable++;
  68         }
  69
  70         /* If file ends with EOF then don't count this EOF as non-printable. */
  71         if (size >= 1 && buf[size-1] == '\032')
  72                 stats->nonprintable--;
  73 }
  74
  75 /*
  76  * The same heuristics as diff.c::mmfile_is_binary()
  77  */
  78 static int is_binary(unsigned long size, struct text_stat *stats)
  79 {
  80
  81         if (stats->nul)
  82                 return 1;
  83         if ((stats->printable >> 7) < stats->nonprintable)
  84                 return 1;
  85         /*
  86          * Other heuristics? Average line length might be relevant,
  87          * as might LF vs CR vs CRLF counts..
  88          *
  89          * NOTE! It might be normal to have a low ratio of CRLF to LF
  90          * (somebody starts with a LF-only file and edits it with an editor
  91          * that adds CRLF only to lines that are added..). But do  we
  92          * want to support CR-only? Probably not.
  93          */
  94         return 0;
  95 }
  96
  97 static enum eol output_eol(enum crlf_action crlf_action)
  98 {
  99         switch (crlf_action) {
 100         case CRLF_BINARY:
 101                 return EOL_UNSET;
 102         case CRLF_CRLF:
 103                 return EOL_CRLF;
 104         case CRLF_INPUT:
 105                 return EOL_LF;
 106         case CRLF_GUESS:
 107                 if (!auto_crlf)
 108                         return EOL_UNSET;
 109                 /* fall through */
 110         case CRLF_TEXT:
 111         case CRLF_AUTO:
 112                 if (auto_crlf == AUTO_CRLF_TRUE)
 113                         return EOL_CRLF;
 114                 else if (auto_crlf == AUTO_CRLF_INPUT)
 115                         return EOL_LF;
 116                 else if (core_eol == EOL_UNSET)
 117                         return EOL_NATIVE;
 118         }
 119         return core_eol;
 120 }
 121
 122 static void check_safe_crlf(const char *path, enum crlf_action crlf_action,
 123                             struct text_stat *stats, enum safe_crlf checksafe)
 124 {
 125         if (!checksafe)
 126                 return;
 127
 128         if (output_eol(crlf_action) == EOL_LF) {
 129                 /*
 130                  * CRLFs would not be restored by checkout:
 131                  * check if we'd remove CRLFs
 132                  */
 133                 if (stats->crlf) {
 134                         if (checksafe == SAFE_CRLF_WARN)
 135                                 warning("CRLF will be replaced by LF in %s.\nThe file will have its original line endings in your working directory.", path);
 136                         else /* i.e. SAFE_CRLF_FAIL */
 137                                 die("CRLF would be replaced by LF in %s.", path);
 138                 }
 139         } else if (output_eol(crlf_action) == EOL_CRLF) {
 140                 /*
 141                  * CRLFs would be added by checkout:
 142                  * check if we have "naked" LFs
 143                  */
 144                 if (stats->lf != stats->crlf) {
 145                         if (checksafe == SAFE_CRLF_WARN)
 146                                 warning("LF will be replaced by CRLF in %s.\nThe file will have its original line endings in your working directory.", path);
 147                         else /* i.e. SAFE_CRLF_FAIL */
 148                                 die("LF would be replaced by CRLF in %s", path);
 149                 }
 150         }
 151 }
 152
 153 static int has_cr_in_index(const char *path)
 154 {
 155         int pos, len;
 156         unsigned long sz;
 157         enum object_type type;
 158         void *data;
 159         int has_cr;
 160         struct index_state *istate = &the_index;
 161
 162         len = strlen(path);
 163         pos = index_name_pos(istate, path, len);
 164         if (pos < 0) {
 165                 /*
 166                  * We might be in the middle of a merge, in which
 167                  * case we would read stage #2 (ours).
 168                  */
 169                 int i;
 170                 for (i = -pos - 1;
 171                      (pos < 0 && i < istate->cache_nr &&
 172                       !strcmp(istate->cache[i]->name, path));
 173                      i++)
 174                         if (ce_stage(istate->cache[i]) == 2)
 175                                 pos = i;
 176         }
 177         if (pos < 0)
 178                 return 0;
 179         data = read_sha1_file(istate->cache[pos]->sha1, &type, &sz);
 180         if (!data || type != OBJ_BLOB) {
 181                 free(data);
 182                 return 0;
 183         }
 184
 185         has_cr = memchr(data, '\r', sz) != NULL;
 186         free(data);
 187         return has_cr;
 188 }
 189
 190 static int crlf_to_git(const char *path, const char *src, size_t len,
 191                        struct strbuf *buf,
 192                        enum crlf_action crlf_action, enum safe_crlf checksafe)
 193 {
 194         struct text_stat stats;
 195         char *dst;
 196
 197         if (crlf_action == CRLF_BINARY ||
 198             (crlf_action == CRLF_GUESS && auto_crlf == AUTO_CRLF_FALSE) || !len)
 199                 return 0;
 200
 201         gather_stats(src, len, &stats);
 202
 203         if (crlf_action == CRLF_AUTO || crlf_action == CRLF_GUESS) {
 204                 /*
 205                  * We're currently not going to even try to convert stuff
 206                  * that has bare CR characters. Does anybody do that crazy
 207                  * stuff?
 208                  */
 209                 if (stats.cr != stats.crlf)
 210                         return 0;
 211
 212                 /*
 213                  * And add some heuristics for binary vs text, of course...
 214                  */
 215                 if (is_binary(len, &stats))
 216                         return 0;
 217
 218                 if (crlf_action == CRLF_GUESS) {
 219                         /*
 220                          * If the file in the index has any CR in it, do not convert.
 221                          * This is the new safer autocrlf handling.
 222                          */
 223                         if (has_cr_in_index(path))
 224                                 return 0;
 225                 }
 226         }
 227
 228         check_safe_crlf(path, crlf_action, &stats, checksafe);
 229
 230         /* Optimization: No CR? Nothing to convert, regardless. */
 231         if (!stats.cr)
 232                 return 0;
 233
 234         /*
 235          * At this point all of our source analysis is done, and we are sure we
 236          * would convert. If we are in dry-run mode, we can give an answer.
 237          */
 238         if (!buf)
 239                 return 1;
 240
 241         /* only grow if not in place */
 242         if (strbuf_avail(buf) + buf->len < len)
 243                 strbuf_grow(buf, len - buf->len);
 244         dst = buf->buf;
 245         if (crlf_action == CRLF_AUTO || crlf_action == CRLF_GUESS) {
 246                 /*
 247                  * If we guessed, we already know we rejected a file with
 248                  * lone CR, and we can strip a CR without looking at what
 249                  * follow it.
 250                  */
 251                 do {
 252                         unsigned char c = *src++;
 253                         if (c != '\r')
 254                                 *dst++ = c;
 255                 } while (--len);
 256         } else {
 257                 do {
 258                         unsigned char c = *src++;
 259                         if (! (c == '\r' && (1 < len && *src == '\n')))
 260                                 *dst++ = c;
 261                 } while (--len);
 262         }
 263         strbuf_setlen(buf, dst - buf->buf);
 264         return 1;
 265 }
 266
 267 static int crlf_to_worktree(const char *path, const char *src, size_t len,
 268                             struct strbuf *buf, enum crlf_action crlf_action)
 269 {
 270         char *to_free = NULL;
 271         struct text_stat stats;
 272
 273         if (!len || output_eol(crlf_action) != EOL_CRLF)
 274                 return 0;
 275
 276         gather_stats(src, len, &stats);
 277
 278         /* No LF? Nothing to convert, regardless. */
 279         if (!stats.lf)
 280                 return 0;
 281
 282         /* Was it already in CRLF format? */
 283         if (stats.lf == stats.crlf)
 284                 return 0;
 285
 286         if (crlf_action == CRLF_AUTO || crlf_action == CRLF_GUESS) {
 287                 if (crlf_action == CRLF_GUESS) {
 288                         /* If we have any CR or CRLF line endings, we do not touch it */
 289                         /* This is the new safer autocrlf-handling */
 290                         if (stats.cr > 0 || stats.crlf > 0)
 291                                 return 0;
 292                 }
 293
 294                 /* If we have any bare CR characters, we're not going to touch it */
 295                 if (stats.cr != stats.crlf)
 296                         return 0;
 297
 298                 if (is_binary(len, &stats))
 299                         return 0;
 300         }
 301
 302         /* are we "faking" in place editing ? */
 303         if (src == buf->buf)
 304                 to_free = strbuf_detach(buf, NULL);
 305
 306         strbuf_grow(buf, len + stats.lf - stats.crlf);
 307         for (;;) {
 308                 const char *nl = memchr(src, '\n', len);
 309                 if (!nl)
 310                         break;
 311                 if (nl > src && nl[-1] == '\r') {
 312                         strbuf_add(buf, src, nl + 1 - src);
 313                 } else {
 314                         strbuf_add(buf, src, nl - src);
 315                         strbuf_addstr(buf, "\r\n");
 316                 }
 317                 len -= nl + 1 - src;
 318                 src  = nl + 1;
 319         }
 320         strbuf_add(buf, src, len);
 321
 322         free(to_free);
 323         return 1;
 324 }
 325
 326 struct filter_params {
 327         const char *src;
 328         unsigned long size;
 329         const char *cmd;
 330         const char *path;
 331 };
 332
 333 static int filter_buffer(int in, int out, void *data)
 334 {
 335         /*
 336          * Spawn cmd and feed the buffer contents through its stdin.
 337          */
 338         struct child_process child_process;
 339         struct filter_params *params = (struct filter_params *)data;
 340         int write_err, status;
 341         const char *argv[] = { NULL, NULL };
 342
 343         /* apply % substitution to cmd */
 344         struct strbuf cmd = STRBUF_INIT;
 345         struct strbuf path = STRBUF_INIT;
 346         struct strbuf_expand_dict_entry dict[] = {
 347                 { "f", NULL, },
 348                 { NULL, NULL, },
 349         };
 350
 351         /* quote the path to preserve spaces, etc. */
 352         sq_quote_buf(&path, params->path);
 353         dict[0].value = path.buf;
 354
 355         /* expand all %f with the quoted path */
 356         strbuf_expand(&cmd, params->cmd, strbuf_expand_dict_cb, &dict);
 357         strbuf_release(&path);
 358
 359         argv[0] = cmd.buf;
 360
 361         memset(&child_process, 0, sizeof(child_process));
 362         child_process.argv = argv;
 363         child_process.use_shell = 1;
 364         child_process.in = -1;
 365         child_process.out = out;
 366
 367         if (start_command(&child_process))
 368                 return error("cannot fork to run external filter %s", params->cmd);
 369
 370         write_err = (write_in_full(child_process.in, params->src, params->size) < 0);
 371         if (close(child_process.in))
 372                 write_err = 1;
 373         if (write_err)
 374                 error("cannot feed the input to external filter %s", params->cmd);
 375
 376         status = finish_command(&child_process);
 377         if (status)
 378                 error("external filter %s failed %d", params->cmd, status);
 379
 380         strbuf_release(&cmd);
 381         return (write_err || status);
 382 }
 383
 384 static int apply_filter(const char *path, const char *src, size_t len,
 385                         struct strbuf *dst, const char *cmd)
 386 {
 387         /*
 388          * Create a pipeline to have the command filter the buffer's
 389          * contents.
 390          *
 391          * (child --> cmd) --> us
 392          */
 393         int ret = 1;
 394         struct strbuf nbuf = STRBUF_INIT;
 395         struct async async;
 396         struct filter_params params;
 397
 398         if (!cmd)
 399                 return 0;
 400
 401         if (!dst)
 402                 return 1;
 403
 404         memset(&async, 0, sizeof(async));
 405         async.proc = filter_buffer;
 406         async.data = &params;
 407         async.out = -1;
 408         params.src = src;
 409         params.size = len;
 410         params.cmd = cmd;
 411         params.path = path;
 412
 413         fflush(NULL);
 414         if (start_async(&async))
 415                 return 0;       /* error was already reported */
 416
 417         if (strbuf_read(&nbuf, async.out, len) < 0) {
 418                 error("read from external filter %s failed", cmd);
 419                 ret = 0;
 420         }
 421         if (close(async.out)) {
 422                 error("read from external filter %s failed", cmd);
 423                 ret = 0;
 424         }
 425         if (finish_async(&async)) {
 426                 error("external filter %s failed", cmd);
 427                 ret = 0;
 428         }
 429
 430         if (ret) {
 431                 strbuf_swap(dst, &nbuf);
 432         }
 433         strbuf_release(&nbuf);
 434         return ret;
 435 }
 436
 437 static struct convert_driver {
 438         const char *name;
 439         struct convert_driver *next;
 440         const char *smudge;
 441         const char *clean;
 442 } *user_convert, **user_convert_tail;
 443
 444 static int read_convert_config(const char *var, const char *value, void *cb)
 445 {
 446         const char *ep, *name;
 447         int namelen;
 448         struct convert_driver *drv;
 449
 450         /*
 451          * External conversion drivers are configured using
 452          * "filter.<name>.variable".
 453          */
 454         if (prefixcmp(var, "filter.") || (ep = strrchr(var, '.')) == var + 6)
 455                 return 0;
 456         name = var + 7;
 457         namelen = ep - name;
 458         for (drv = user_convert; drv; drv = drv->next)
 459                 if (!strncmp(drv->name, name, namelen) && !drv->name[namelen])
 460                         break;
 461         if (!drv) {
 462                 drv = xcalloc(1, sizeof(struct convert_driver));
 463                 drv->name = xmemdupz(name, namelen);
 464                 *user_convert_tail = drv;
 465                 user_convert_tail = &(drv->next);
 466         }
 467
 468         ep++;
 469
 470         /*
 471          * filter.<name>.smudge and filter.<name>.clean specifies
 472          * the command line:
 473          *
 474          *      command-line
 475          *
 476          * The command-line will not be interpolated in any way.
 477          */
 478
 479         if (!strcmp("smudge", ep))
 480                 return git_config_string(&drv->smudge, var, value);
 481
 482         if (!strcmp("clean", ep))
 483                 return git_config_string(&drv->clean, var, value);
 484
 485         return 0;
 486 }
 487
 488 static int count_ident(const char *cp, unsigned long size)
 489 {
 490         /*
 491          * "$Id: 0000000000000000000000000000000000000000 $" <=> "$Id$"
 492          */
 493         int cnt = 0;
 494         char ch;
 495
 496         while (size) {
 497                 ch = *cp++;
 498                 size--;
 499                 if (ch != '$')
 500                         continue;
 501                 if (size < 3)
 502                         break;
 503                 if (memcmp("Id", cp, 2))
 504                         continue;
 505                 ch = cp[2];
 506                 cp += 3;
 507                 size -= 3;
 508                 if (ch == '$')
 509                         cnt++; /* $Id$ */
 510                 if (ch != ':')
 511                         continue;
 512
 513                 /*
 514                  * "$Id: ... "; scan up to the closing dollar sign and discard.
 515                  */
 516                 while (size) {
 517                         ch = *cp++;
 518                         size--;
 519                         if (ch == '$') {
 520                                 cnt++;
 521                                 break;
 522                         }
 523                         if (ch == '\n')
 524                                 break;
 525                 }
 526         }
 527         return cnt;
 528 }
 529
 530 static int ident_to_git(const char *path, const char *src, size_t len,
 531                         struct strbuf *buf, int ident)
 532 {
 533         char *dst, *dollar;
 534
 535         if (!ident || !count_ident(src, len))
 536                 return 0;
 537
 538         if (!buf)
 539                 return 1;
 540
 541         /* only grow if not in place */
 542         if (strbuf_avail(buf) + buf->len < len)
 543                 strbuf_grow(buf, len - buf->len);
 544         dst = buf->buf;
 545         for (;;) {
 546                 dollar = memchr(src, '$', len);
 547                 if (!dollar)
 548                         break;
 549                 memmove(dst, src, dollar + 1 - src);
 550                 dst += dollar + 1 - src;
 551                 len -= dollar + 1 - src;
 552                 src  = dollar + 1;
 553
 554                 if (len > 3 && !memcmp(src, "Id:", 3)) {
 555                         dollar = memchr(src + 3, '$', len - 3);
 556                         if (!dollar)
 557                                 break;
 558                         if (memchr(src + 3, '\n', dollar - src - 3)) {
 559                                 /* Line break before the next dollar. */
 560                                 continue;
 561                         }
 562
 563                         memcpy(dst, "Id$", 3);
 564                         dst += 3;
 565                         len -= dollar + 1 - src;
 566                         src  = dollar + 1;
 567                 }
 568         }
 569         memmove(dst, src, len);
 570         strbuf_setlen(buf, dst + len - buf->buf);
 571         return 1;
 572 }
 573
 574 static int ident_to_worktree(const char *path, const char *src, size_t len,
 575                              struct strbuf *buf, int ident)
 576 {
 577         unsigned char sha1[20];
 578         char *to_free = NULL, *dollar, *spc;
 579         int cnt;
 580
 581         if (!ident)
 582                 return 0;
 583
 584         cnt = count_ident(src, len);
 585         if (!cnt)
 586                 return 0;
 587
 588         /* are we "faking" in place editing ? */
 589         if (src == buf->buf)
 590                 to_free = strbuf_detach(buf, NULL);
 591         hash_sha1_file(src, len, "blob", sha1);
 592
 593         strbuf_grow(buf, len + cnt * 43);
 594         for (;;) {
 595                 /* step 1: run to the next '$' */
 596                 dollar = memchr(src, '$', len);
 597                 if (!dollar)
 598                         break;
 599                 strbuf_add(buf, src, dollar + 1 - src);
 600                 len -= dollar + 1 - src;
 601                 src  = dollar + 1;
 602
 603                 /* step 2: does it looks like a bit like Id:xxx$ or Id$ ? */
 604                 if (len < 3 || memcmp("Id", src, 2))
 605                         continue;
 606
 607                 /* step 3: skip over Id$ or Id:xxxxx$ */
 608                 if (src[2] == '$') {
 609                         src += 3;
 610                         len -= 3;
 611                 } else if (src[2] == ':') {
 612                         /*
 613                          * It's possible that an expanded Id has crept its way into the
 614                          * repository, we cope with that by stripping the expansion out.
 615                          * This is probably not a good idea, since it will cause changes
 616                          * on checkout, which won't go away by stash, but let's keep it
 617                          * for git-style ids.
 618                          */
 619                         dollar = memchr(src + 3, '$', len - 3);
 620                         if (!dollar) {
 621                                 /* incomplete keyword, no more '$', so just quit the loop */
 622                                 break;
 623                         }
 624
 625                         if (memchr(src + 3, '\n', dollar - src - 3)) {
 626                                 /* Line break before the next dollar. */
 627                                 continue;
 628                         }
 629
 630                         spc = memchr(src + 4, ' ', dollar - src - 4);
 631                         if (spc && spc < dollar-1) {
 632                                 /* There are spaces in unexpected places.
 633                                  * This is probably an id from some other
 634                                  * versioning system. Keep it for now.
 635                                  */
 636                                 continue;
 637                         }
 638
 639                         len -= dollar + 1 - src;
 640                         src  = dollar + 1;
 641                 } else {
 642                         /* it wasn't a "Id$" or "Id:xxxx$" */
 643                         continue;
 644                 }
 645
 646                 /* step 4: substitute */
 647                 strbuf_addstr(buf, "Id: ");
 648                 strbuf_add(buf, sha1_to_hex(sha1), 40);
 649                 strbuf_addstr(buf, " $");
 650         }
 651         strbuf_add(buf, src, len);
 652
 653         free(to_free);
 654         return 1;
 655 }
 656
 657 static enum crlf_action git_path_check_crlf(const char *path, struct git_attr_check *check)
 658 {
 659         const char *value = check->value;
 660
 661         if (ATTR_TRUE(value))
 662                 return CRLF_TEXT;
 663         else if (ATTR_FALSE(value))
 664                 return CRLF_BINARY;
 665         else if (ATTR_UNSET(value))
 666                 ;
 667         else if (!strcmp(value, "input"))
 668                 return CRLF_INPUT;
 669         else if (!strcmp(value, "auto"))
 670                 return CRLF_AUTO;
 671         return CRLF_GUESS;
 672 }
 673
 674 static enum eol git_path_check_eol(const char *path, struct git_attr_check *check)
 675 {
 676         const char *value = check->value;
 677
 678         if (ATTR_UNSET(value))
 679                 ;
 680         else if (!strcmp(value, "lf"))
 681                 return EOL_LF;
 682         else if (!strcmp(value, "crlf"))
 683                 return EOL_CRLF;
 684         return EOL_UNSET;
 685 }
 686
 687 static struct convert_driver *git_path_check_convert(const char *path,
 688                                              struct git_attr_check *check)
 689 {
 690         const char *value = check->value;
 691         struct convert_driver *drv;
 692
 693         if (ATTR_TRUE(value) || ATTR_FALSE(value) || ATTR_UNSET(value))
 694                 return NULL;
 695         for (drv = user_convert; drv; drv = drv->next)
 696                 if (!strcmp(value, drv->name))
 697                         return drv;
 698         return NULL;
 699 }
 700
 701 static int git_path_check_ident(const char *path, struct git_attr_check *check)
 702 {
 703         const char *value = check->value;
 704
 705         return !!ATTR_TRUE(value);
 706 }
 707
 708 static enum crlf_action input_crlf_action(enum crlf_action text_attr, enum eol eol_attr)
 709 {
 710         if (text_attr == CRLF_BINARY)
 711                 return CRLF_BINARY;
 712         if (eol_attr == EOL_LF)
 713                 return CRLF_INPUT;
 714         if (eol_attr == EOL_CRLF)
 715                 return CRLF_CRLF;
 716         return text_attr;
 717 }
 718
 719 struct conv_attrs {
 720         struct convert_driver *drv;
 721         enum crlf_action crlf_action;
 722         enum eol eol_attr;
 723         int ident;
 724 };
 725
 726 static const char *conv_attr_name[] = {
 727         "crlf", "ident", "filter", "eol", "text",
 728 };
 729 #define NUM_CONV_ATTRS ARRAY_SIZE(conv_attr_name)
 730
 731 static void convert_attrs(struct conv_attrs *ca, const char *path)
 732 {
 733         int i;
 734         static struct git_attr_check ccheck[NUM_CONV_ATTRS];
 735
 736         if (!ccheck[0].attr) {
 737                 for (i = 0; i < NUM_CONV_ATTRS; i++)
 738                         ccheck[i].attr = git_attr(conv_attr_name[i]);
 739                 user_convert_tail = &user_convert;
 740                 git_config(read_convert_config, NULL);
 741         }
 742
 743         if (!git_check_attr(path, NUM_CONV_ATTRS, ccheck)) {
 744                 ca->crlf_action = git_path_check_crlf(path, ccheck + 4);
 745                 if (ca->crlf_action == CRLF_GUESS)
 746                         ca->crlf_action = git_path_check_crlf(path, ccheck + 0);
 747                 ca->ident = git_path_check_ident(path, ccheck + 1);
 748                 ca->drv = git_path_check_convert(path, ccheck + 2);
 749                 ca->eol_attr = git_path_check_eol(path, ccheck + 3);
 750         } else {
 751                 ca->drv = NULL;
 752                 ca->crlf_action = CRLF_GUESS;
 753                 ca->eol_attr = EOL_UNSET;
 754                 ca->ident = 0;
 755         }
 756 }
 757
 758 int convert_to_git(const char *path, const char *src, size_t len,
 759                    struct strbuf *dst, enum safe_crlf checksafe)
 760 {
 761         int ret = 0;
 762         const char *filter = NULL;
 763         struct conv_attrs ca;
 764
 765         convert_attrs(&ca, path);
 766         if (ca.drv)
 767                 filter = ca.drv->clean;
 768
 769         ret |= apply_filter(path, src, len, dst, filter);
 770         if (ret && dst) {
 771                 src = dst->buf;
 772                 len = dst->len;
 773         }
 774         ca.crlf_action = input_crlf_action(ca.crlf_action, ca.eol_attr);
 775         ret |= crlf_to_git(path, src, len, dst, ca.crlf_action, checksafe);
 776         if (ret && dst) {
 777                 src = dst->buf;
 778                 len = dst->len;
 779         }
 780         return ret | ident_to_git(path, src, len, dst, ca.ident);
 781 }
 782
 783 static int convert_to_working_tree_internal(const char *path, const char *src,
 784                                             size_t len, struct strbuf *dst,
 785                                             int normalizing)
 786 {
 787         int ret = 0;
 788         const char *filter = NULL;
 789         struct conv_attrs ca;
 790
 791         convert_attrs(&ca, path);
 792         if (ca.drv)
 793                 filter = ca.drv->smudge;
 794
 795         ret |= ident_to_worktree(path, src, len, dst, ca.ident);
 796         if (ret) {
 797                 src = dst->buf;
 798                 len = dst->len;
 799         }
 800         /*
 801          * CRLF conversion can be skipped if normalizing, unless there
 802          * is a smudge filter.  The filter might expect CRLFs.
 803          */
 804         if (filter || !normalizing) {
 805                 ca.crlf_action = input_crlf_action(ca.crlf_action, ca.eol_attr);
 806                 ret |= crlf_to_worktree(path, src, len, dst, ca.crlf_action);
 807                 if (ret) {
 808                         src = dst->buf;
 809                         len = dst->len;
 810                 }
 811         }
 812         return ret | apply_filter(path, src, len, dst, filter);
 813 }
 814
 815 int convert_to_working_tree(const char *path, const char *src, size_t len, struct strbuf *dst)
 816 {
 817         return convert_to_working_tree_internal(path, src, len, dst, 0);
 818 }
 819
 820 int renormalize_buffer(const char *path, const char *src, size_t len, struct strbuf *dst)
 821 {
 822         int ret = convert_to_working_tree_internal(path, src, len, dst, 1);
 823         if (ret) {
 824                 src = dst->buf;
 825                 len = dst->len;
 826         }
 827         return ret | convert_to_git(path, src, len, dst, SAFE_CRLF_FALSE);
 828 }
 829
 830 /*****************************************************************
 831  *
 832  * Streaming converison support
 833  *
 834  *****************************************************************/
 835
 836 typedef int (*filter_fn)(struct stream_filter *,
 837                          const char *input, size_t *isize_p,
 838                          char *output, size_t *osize_p);
 839 typedef void (*free_fn)(struct stream_filter *);
 840
 841 struct stream_filter_vtbl {
 842         filter_fn filter;
 843         free_fn free;
 844 };
 845
 846 struct stream_filter {
 847         struct stream_filter_vtbl *vtbl;
 848 };
 849
 850 static int null_filter_fn(struct stream_filter *filter,
 851                           const char *input, size_t *isize_p,
 852                           char *output, size_t *osize_p)
 853 {
 854         size_t count;
 855
 856         if (!input)
 857                 return 0; /* we do not keep any states */
 858         count = *isize_p;
 859         if (*osize_p < count)
 860                 count = *osize_p;
 861         if (count) {
 862                 memmove(output, input, count);
 863                 *isize_p -= count;
 864                 *osize_p -= count;
 865         }
 866         return 0;
 867 }
 868
 869 static void null_free_fn(struct stream_filter *filter)
 870 {
 871         ; /* nothing -- null instances are shared */
 872 }
 873
 874 static struct stream_filter_vtbl null_vtbl = {
 875         null_filter_fn,
 876         null_free_fn,
 877 };
 878
 879 static struct stream_filter null_filter_singleton = {
 880         &null_vtbl,
 881 };
 882
 883 int is_null_stream_filter(struct stream_filter *filter)
 884 {
 885         return filter == &null_filter_singleton;
 886 }
 887
 888
 889 /*
 890  * LF-to-CRLF filter
 891  */
 892
 893 struct lf_to_crlf_filter {
 894         struct stream_filter filter;
 895         unsigned has_held:1;
 896         char held;
 897 };
 898
 899 static int lf_to_crlf_filter_fn(struct stream_filter *filter,
 900                                 const char *input, size_t *isize_p,
 901                                 char *output, size_t *osize_p)
 902 {
 903         size_t count, o = 0;
 904         struct lf_to_crlf_filter *lf_to_crlf = (struct lf_to_crlf_filter *)filter;
 905
 906         /*
 907          * We may be holding onto the CR to see if it is followed by a
 908          * LF, in which case we would need to go to the main loop.
 909          * Otherwise, just emit it to the output stream.
 910          */
 911         if (lf_to_crlf->has_held && (lf_to_crlf->held != '\r' || !input)) {
 912                 output[o++] = lf_to_crlf->held;
 913                 lf_to_crlf->has_held = 0;
 914         }
 915
 916         /* We are told to drain */
 917         if (!input) {
 918                 *osize_p -= o;
 919                 return 0;
 920         }
 921
 922         count = *isize_p;
 923         if (count || lf_to_crlf->has_held) {
 924                 size_t i;
 925                 int was_cr = 0;
 926
 927                 if (lf_to_crlf->has_held) {
 928                         was_cr = 1;
 929                         lf_to_crlf->has_held = 0;
 930                 }
 931
 932                 for (i = 0; o < *osize_p && i < count; i++) {
 933                         char ch = input[i];
 934
 935                         if (ch == '\n') {
 936                                 output[o++] = '\r';
 937                         } else if (was_cr) {
 938                                 /*
 939                                  * Previous round saw CR and it is not followed
 940                                  * by a LF; emit the CR before processing the
 941                                  * current character.
 942                                  */
 943                                 output[o++] = '\r';
 944                         }
 945
 946                         /*
 947                          * We may have consumed the last output slot,
 948                          * in which case we need to break out of this
 949                          * loop; hold the current character before
 950                          * returning.
 951                          */
 952                         if (*osize_p <= o) {
 953                                 lf_to_crlf->has_held = 1;
 954                                 lf_to_crlf->held = ch;
 955                                 continue; /* break but increment i */
 956                         }
 957
 958                         if (ch == '\r') {
 959                                 was_cr = 1;
 960                                 continue;
 961                         }
 962
 963                         was_cr = 0;
 964                         output[o++] = ch;
 965                 }
 966
 967                 *osize_p -= o;
 968                 *isize_p -= i;
 969
 970                 if (!lf_to_crlf->has_held && was_cr) {
 971                         lf_to_crlf->has_held = 1;
 972                         lf_to_crlf->held = '\r';
 973                 }
 974         }
 975         return 0;
 976 }
 977
 978 static void lf_to_crlf_free_fn(struct stream_filter *filter)
 979 {
 980         free(filter);
 981 }
 982
 983 static struct stream_filter_vtbl lf_to_crlf_vtbl = {
 984         lf_to_crlf_filter_fn,
 985         lf_to_crlf_free_fn,
 986 };
 987
 988 static struct stream_filter *lf_to_crlf_filter(void)
 989 {
 990         struct lf_to_crlf_filter *lf_to_crlf = xcalloc(1, sizeof(*lf_to_crlf));
 991
 992         lf_to_crlf->filter.vtbl = &lf_to_crlf_vtbl;
 993         return (struct stream_filter *)lf_to_crlf;
 994 }
 995
 996 /*
 997  * Cascade filter
 998  */
 999 #define FILTER_BUFFER 1024
1000 struct cascade_filter {
1001         struct stream_filter filter;
1002         struct stream_filter *one;
1003         struct stream_filter *two;
1004         char buf[FILTER_BUFFER];
1005         int end, ptr;
1006 };
1007
1008 static int cascade_filter_fn(struct stream_filter *filter,
1009                              const char *input, size_t *isize_p,
1010                              char *output, size_t *osize_p)
1011 {
1012         struct cascade_filter *cas = (struct cascade_filter *) filter;
1013         size_t filled = 0;
1014         size_t sz = *osize_p;
1015         size_t to_feed, remaining;
1016
1017         /*
1018          * input -- (one) --> buf -- (two) --> output
1019          */
1020         while (filled < sz) {
1021                 remaining = sz - filled;
1022
1023                 /* do we already have something to feed two with? */
1024                 if (cas->ptr < cas->end) {
1025                         to_feed = cas->end - cas->ptr;
1026                         if (stream_filter(cas->two,
1027                                           cas->buf + cas->ptr, &to_feed,
1028                                           output + filled, &remaining))
1029                                 return -1;
1030                         cas->ptr += (cas->end - cas->ptr) - to_feed;
1031                         filled = sz - remaining;
1032                         continue;
1033                 }
1034
1035                 /* feed one from upstream and have it emit into our buffer */
1036                 to_feed = input ? *isize_p : 0;
1037                 if (input && !to_feed)
1038                         break;
1039                 remaining = sizeof(cas->buf);
1040                 if (stream_filter(cas->one,
1041                                   input, &to_feed,
1042                                   cas->buf, &remaining))
1043                         return -1;
1044                 cas->end = sizeof(cas->buf) - remaining;
1045                 cas->ptr = 0;
1046                 if (input) {
1047                         size_t fed = *isize_p - to_feed;
1048                         *isize_p -= fed;
1049                         input += fed;
1050                 }
1051
1052                 /* do we know that we drained one completely? */
1053                 if (input || cas->end)
1054                         continue;
1055
1056                 /* tell two to drain; we have nothing more to give it */
1057                 to_feed = 0;
1058                 remaining = sz - filled;
1059                 if (stream_filter(cas->two,
1060                                   NULL, &to_feed,
1061                                   output + filled, &remaining))
1062                         return -1;
1063                 if (remaining == (sz - filled))
1064                         break; /* completely drained two */
1065                 filled = sz - remaining;
1066         }
1067         *osize_p -= filled;
1068         return 0;
1069 }
1070
1071 static void cascade_free_fn(struct stream_filter *filter)
1072 {
1073         struct cascade_filter *cas = (struct cascade_filter *)filter;
1074         free_stream_filter(cas->one);
1075         free_stream_filter(cas->two);
1076         free(filter);
1077 }
1078
1079 static struct stream_filter_vtbl cascade_vtbl = {
1080         cascade_filter_fn,
1081         cascade_free_fn,
1082 };
1083
1084 static struct stream_filter *cascade_filter(struct stream_filter *one,
1085                                             struct stream_filter *two)
1086 {
1087         struct cascade_filter *cascade;
1088
1089         if (!one || is_null_stream_filter(one))
1090                 return two;
1091         if (!two || is_null_stream_filter(two))
1092                 return one;
1093
1094         cascade = xmalloc(sizeof(*cascade));
1095         cascade->one = one;
1096         cascade->two = two;
1097         cascade->end = cascade->ptr = 0;
1098         cascade->filter.vtbl = &cascade_vtbl;
1099         return (struct stream_filter *)cascade;
1100 }
1101
1102 /*
1103  * ident filter
1104  */
1105 #define IDENT_DRAINING (-1)
1106 #define IDENT_SKIPPING (-2)
1107 struct ident_filter {
1108         struct stream_filter filter;
1109         struct strbuf left;
1110         int state;
1111         char ident[45]; /* ": x40 $" */
1112 };
1113
1114 static int is_foreign_ident(const char *str)
1115 {
1116         int i;
1117
1118         if (prefixcmp(str, "$Id: "))
1119                 return 0;
1120         for (i = 5; str[i]; i++) {
1121                 if (isspace(str[i]) && str[i+1] != '$')
1122                         return 1;
1123         }
1124         return 0;
1125 }
1126
1127 static void ident_drain(struct ident_filter *ident, char **output_p, size_t *osize_p)
1128 {
1129         size_t to_drain = ident->left.len;
1130
1131         if (*osize_p < to_drain)
1132                 to_drain = *osize_p;
1133         if (to_drain) {
1134                 memcpy(*output_p, ident->left.buf, to_drain);
1135                 strbuf_remove(&ident->left, 0, to_drain);
1136                 *output_p += to_drain;
1137                 *osize_p -= to_drain;
1138         }
1139         if (!ident->left.len)
1140                 ident->state = 0;
1141 }
1142
1143 static int ident_filter_fn(struct stream_filter *filter,
1144                            const char *input, size_t *isize_p,
1145                            char *output, size_t *osize_p)
1146 {
1147         struct ident_filter *ident = (struct ident_filter *)filter;
1148         static const char head[] = "$Id";
1149
1150         if (!input) {
1151                 /* drain upon eof */
1152                 switch (ident->state) {
1153                 default:
1154                         strbuf_add(&ident->left, head, ident->state);
1155                 case IDENT_SKIPPING:
1156                         /* fallthru */
1157                 case IDENT_DRAINING:
1158                         ident_drain(ident, &output, osize_p);
1159                 }
1160                 return 0;
1161         }
1162
1163         while (*isize_p || (ident->state == IDENT_DRAINING)) {
1164                 int ch;
1165
1166                 if (ident->state == IDENT_DRAINING) {
1167                         ident_drain(ident, &output, osize_p);
1168                         if (!*osize_p)
1169                                 break;
1170                         continue;
1171                 }
1172
1173                 ch = *(input++);
1174                 (*isize_p)--;
1175
1176                 if (ident->state == IDENT_SKIPPING) {
1177                         /*
1178                          * Skipping until '$' or LF, but keeping them
1179                          * in case it is a foreign ident.
1180                          */
1181                         strbuf_addch(&ident->left, ch);
1182                         if (ch != '\n' && ch != '$')
1183                                 continue;
1184                         if (ch == '$' && !is_foreign_ident(ident->left.buf)) {
1185                                 strbuf_setlen(&ident->left, sizeof(head) - 1);
1186                                 strbuf_addstr(&ident->left, ident->ident);
1187                         }
1188                         ident->state = IDENT_DRAINING;
1189                         continue;
1190                 }
1191
1192                 if (ident->state < sizeof(head) &&
1193                     head[ident->state] == ch) {
1194                         ident->state++;
1195                         continue;
1196                 }
1197
1198                 if (ident->state)
1199                         strbuf_add(&ident->left, head, ident->state);
1200                 if (ident->state == sizeof(head) - 1) {
1201                         if (ch != ':' && ch != '$') {
1202                                 strbuf_addch(&ident->left, ch);
1203                                 ident->state = 0;
1204                                 continue;
1205                         }
1206
1207                         if (ch == ':') {
1208                                 strbuf_addch(&ident->left, ch);
1209                                 ident->state = IDENT_SKIPPING;
1210                         } else {
1211                                 strbuf_addstr(&ident->left, ident->ident);
1212                                 ident->state = IDENT_DRAINING;
1213                         }
1214                         continue;
1215                 }
1216
1217                 strbuf_addch(&ident->left, ch);
1218                 ident->state = IDENT_DRAINING;
1219         }
1220         return 0;
1221 }
1222
1223 static void ident_free_fn(struct stream_filter *filter)
1224 {
1225         struct ident_filter *ident = (struct ident_filter *)filter;
1226         strbuf_release(&ident->left);
1227         free(filter);
1228 }
1229
1230 static struct stream_filter_vtbl ident_vtbl = {
1231         ident_filter_fn,
1232         ident_free_fn,
1233 };
1234
1235 static struct stream_filter *ident_filter(const unsigned char *sha1)
1236 {
1237         struct ident_filter *ident = xmalloc(sizeof(*ident));
1238
1239         sprintf(ident->ident, ": %s $", sha1_to_hex(sha1));
1240         strbuf_init(&ident->left, 0);
1241         ident->filter.vtbl = &ident_vtbl;
1242         ident->state = 0;
1243         return (struct stream_filter *)ident;
1244 }
1245
1246 /*
1247  * Return an appropriately constructed filter for the path, or NULL if
1248  * the contents cannot be filtered without reading the whole thing
1249  * in-core.
1250  *
1251  * Note that you would be crazy to set CRLF, smuge/clean or ident to a
1252  * large binary blob you would want us not to slurp into the memory!
1253  */
1254 struct stream_filter *get_stream_filter(const char *path, const unsigned char *sha1)
1255 {
1256         struct conv_attrs ca;
1257         enum crlf_action crlf_action;
1258         struct stream_filter *filter = NULL;
1259
1260         convert_attrs(&ca, path);
1261
1262         if (ca.drv && (ca.drv->smudge || ca.drv->clean))
1263                 return filter;
1264
1265         if (ca.ident)
1266                 filter = ident_filter(sha1);
1267
1268         crlf_action = input_crlf_action(ca.crlf_action, ca.eol_attr);
1269
1270         if ((crlf_action == CRLF_BINARY) || (crlf_action == CRLF_INPUT) ||
1271             (crlf_action == CRLF_GUESS && auto_crlf == AUTO_CRLF_FALSE))
1272                 filter = cascade_filter(filter, &null_filter_singleton);
1273
1274         else if (output_eol(crlf_action) == EOL_CRLF &&
1275                  !(crlf_action == CRLF_AUTO || crlf_action == CRLF_GUESS))
1276                 filter = cascade_filter(filter, lf_to_crlf_filter());
1277
1278         return filter;
1279 }
1280
1281 void free_stream_filter(struct stream_filter *filter)
1282 {
1283         filter->vtbl->free(filter);
1284 }
1285
1286 int stream_filter(struct stream_filter *filter,
1287                   const char *input, size_t *isize_p,
1288                   char *output, size_t *osize_p)
1289 {
1290         return filter->vtbl->filter(filter, input, isize_p, output, osize_p);
1291 }