convert.c

   1 #include "cache.h"
   2 #include "attr.h"
   3 #include "run-command.h"
   4 #include "quote.h"
   5 #include "sigchain.h"
   6
   7 /*
   8  * convert.c - convert a file when checking it out and checking it in.
   9  *
  10  * This should use the pathname to decide on whether it wants to do some
  11  * more interesting conversions (automatic gzip/unzip, general format
  12  * conversions etc etc), but by default it just does automatic CRLF<->LF
  13  * translation when the "text" attribute or "auto_crlf" option is set.
  14  */
  15
  16 enum crlf_action {
  17         CRLF_GUESS = -1,
  18         CRLF_BINARY = 0,
  19         CRLF_TEXT,
  20         CRLF_INPUT,
  21         CRLF_CRLF,
  22         CRLF_AUTO
  23 };
  24
  25 struct text_stat {
  26         /* NUL, CR, LF and CRLF counts */
  27         unsigned nul, cr, lf, crlf;
  28
  29         /* These are just approximations! */
  30         unsigned printable, nonprintable;
  31 };
  32
  33 static void gather_stats(const char *buf, unsigned long size, struct text_stat *stats)
  34 {
  35         unsigned long i;
  36
  37         memset(stats, 0, sizeof(*stats));
  38
  39         for (i = 0; i < size; i++) {
  40                 unsigned char c = buf[i];
  41                 if (c == '\r') {
  42                         stats->cr++;
  43                         if (i+1 < size && buf[i+1] == '\n')
  44                                 stats->crlf++;
  45                         continue;
  46                 }
  47                 if (c == '\n') {
  48                         stats->lf++;
  49                         continue;
  50                 }
  51                 if (c == 127)
  52                         /* DEL */
  53                         stats->nonprintable++;
  54                 else if (c < 32) {
  55                         switch (c) {
  56                                 /* BS, HT, ESC and FF */
  57                         case '\b': case '\t': case '\033': case '\014':
  58                                 stats->printable++;
  59                                 break;
  60                         case 0:
  61                                 stats->nul++;
  62                                 /* fall through */
  63                         default:
  64                                 stats->nonprintable++;
  65                         }
  66                 }
  67                 else
  68                         stats->printable++;
  69         }
  70
  71         /* If file ends with EOF then don't count this EOF as non-printable. */
  72         if (size >= 1 && buf[size-1] == '\032')
  73                 stats->nonprintable--;
  74 }
  75
  76 /*
  77  * The same heuristics as diff.c::mmfile_is_binary()
  78  */
  79 static int is_binary(unsigned long size, struct text_stat *stats)
  80 {
  81
  82         if (stats->nul)
  83                 return 1;
  84         if ((stats->printable >> 7) < stats->nonprintable)
  85                 return 1;
  86         /*
  87          * Other heuristics? Average line length might be relevant,
  88          * as might LF vs CR vs CRLF counts..
  89          *
  90          * NOTE! It might be normal to have a low ratio of CRLF to LF
  91          * (somebody starts with a LF-only file and edits it with an editor
  92          * that adds CRLF only to lines that are added..). But do  we
  93          * want to support CR-only? Probably not.
  94          */
  95         return 0;
  96 }
  97
  98 static enum eol output_eol(enum crlf_action crlf_action)
  99 {
 100         switch (crlf_action) {
 101         case CRLF_BINARY:
 102                 return EOL_UNSET;
 103         case CRLF_CRLF:
 104                 return EOL_CRLF;
 105         case CRLF_INPUT:
 106                 return EOL_LF;
 107         case CRLF_GUESS:
 108                 if (!auto_crlf)
 109                         return EOL_UNSET;
 110                 /* fall through */
 111         case CRLF_TEXT:
 112         case CRLF_AUTO:
 113                 if (auto_crlf == AUTO_CRLF_TRUE)
 114                         return EOL_CRLF;
 115                 else if (auto_crlf == AUTO_CRLF_INPUT)
 116                         return EOL_LF;
 117                 else if (core_eol == EOL_UNSET)
 118                         return EOL_NATIVE;
 119         }
 120         return core_eol;
 121 }
 122
 123 static void check_safe_crlf(const char *path, enum crlf_action crlf_action,
 124                             struct text_stat *stats, enum safe_crlf checksafe)
 125 {
 126         if (!checksafe)
 127                 return;
 128
 129         if (output_eol(crlf_action) == EOL_LF) {
 130                 /*
 131                  * CRLFs would not be restored by checkout:
 132                  * check if we'd remove CRLFs
 133                  */
 134                 if (stats->crlf) {
 135                         if (checksafe == SAFE_CRLF_WARN)
 136                                 warning("CRLF will be replaced by LF in %s.\nThe file will have its original line endings in your working directory.", path);
 137                         else /* i.e. SAFE_CRLF_FAIL */
 138                                 die("CRLF would be replaced by LF in %s.", path);
 139                 }
 140         } else if (output_eol(crlf_action) == EOL_CRLF) {
 141                 /*
 142                  * CRLFs would be added by checkout:
 143                  * check if we have "naked" LFs
 144                  */
 145                 if (stats->lf != stats->crlf) {
 146                         if (checksafe == SAFE_CRLF_WARN)
 147                                 warning("LF will be replaced by CRLF in %s.\nThe file will have its original line endings in your working directory.", path);
 148                         else /* i.e. SAFE_CRLF_FAIL */
 149                                 die("LF would be replaced by CRLF in %s", path);
 150                 }
 151         }
 152 }
 153
 154 static int has_cr_in_index(const char *path)
 155 {
 156         int pos, len;
 157         unsigned long sz;
 158         enum object_type type;
 159         void *data;
 160         int has_cr;
 161         struct index_state *istate = &the_index;
 162
 163         len = strlen(path);
 164         pos = index_name_pos(istate, path, len);
 165         if (pos < 0) {
 166                 /*
 167                  * We might be in the middle of a merge, in which
 168                  * case we would read stage #2 (ours).
 169                  */
 170                 int i;
 171                 for (i = -pos - 1;
 172                      (pos < 0 && i < istate->cache_nr &&
 173                       !strcmp(istate->cache[i]->name, path));
 174                      i++)
 175                         if (ce_stage(istate->cache[i]) == 2)
 176                                 pos = i;
 177         }
 178         if (pos < 0)
 179                 return 0;
 180         data = read_sha1_file(istate->cache[pos]->sha1, &type, &sz);
 181         if (!data || type != OBJ_BLOB) {
 182                 free(data);
 183                 return 0;
 184         }
 185
 186         has_cr = memchr(data, '\r', sz) != NULL;
 187         free(data);
 188         return has_cr;
 189 }
 190
 191 static int crlf_to_git(const char *path, const char *src, size_t len,
 192                        struct strbuf *buf,
 193                        enum crlf_action crlf_action, enum safe_crlf checksafe)
 194 {
 195         struct text_stat stats;
 196         char *dst;
 197
 198         if (crlf_action == CRLF_BINARY ||
 199             (crlf_action == CRLF_GUESS && auto_crlf == AUTO_CRLF_FALSE) || !len)
 200                 return 0;
 201
 202         gather_stats(src, len, &stats);
 203
 204         if (crlf_action == CRLF_AUTO || crlf_action == CRLF_GUESS) {
 205                 /*
 206                  * We're currently not going to even try to convert stuff
 207                  * that has bare CR characters. Does anybody do that crazy
 208                  * stuff?
 209                  */
 210                 if (stats.cr != stats.crlf)
 211                         return 0;
 212
 213                 /*
 214                  * And add some heuristics for binary vs text, of course...
 215                  */
 216                 if (is_binary(len, &stats))
 217                         return 0;
 218
 219                 if (crlf_action == CRLF_GUESS) {
 220                         /*
 221                          * If the file in the index has any CR in it, do not convert.
 222                          * This is the new safer autocrlf handling.
 223                          */
 224                         if (has_cr_in_index(path))
 225                                 return 0;
 226                 }
 227         }
 228
 229         check_safe_crlf(path, crlf_action, &stats, checksafe);
 230
 231         /* Optimization: No CR? Nothing to convert, regardless. */
 232         if (!stats.cr)
 233                 return 0;
 234
 235         /* only grow if not in place */
 236         if (strbuf_avail(buf) + buf->len < len)
 237                 strbuf_grow(buf, len - buf->len);
 238         dst = buf->buf;
 239         if (crlf_action == CRLF_AUTO || crlf_action == CRLF_GUESS) {
 240                 /*
 241                  * If we guessed, we already know we rejected a file with
 242                  * lone CR, and we can strip a CR without looking at what
 243                  * follow it.
 244                  */
 245                 do {
 246                         unsigned char c = *src++;
 247                         if (c != '\r')
 248                                 *dst++ = c;
 249                 } while (--len);
 250         } else {
 251                 do {
 252                         unsigned char c = *src++;
 253                         if (! (c == '\r' && (1 < len && *src == '\n')))
 254                                 *dst++ = c;
 255                 } while (--len);
 256         }
 257         strbuf_setlen(buf, dst - buf->buf);
 258         return 1;
 259 }
 260
 261 static int crlf_to_worktree(const char *path, const char *src, size_t len,
 262                             struct strbuf *buf, enum crlf_action crlf_action)
 263 {
 264         char *to_free = NULL;
 265         struct text_stat stats;
 266
 267         if (!len || output_eol(crlf_action) != EOL_CRLF)
 268                 return 0;
 269
 270         gather_stats(src, len, &stats);
 271
 272         /* No LF? Nothing to convert, regardless. */
 273         if (!stats.lf)
 274                 return 0;
 275
 276         /* Was it already in CRLF format? */
 277         if (stats.lf == stats.crlf)
 278                 return 0;
 279
 280         if (crlf_action == CRLF_AUTO || crlf_action == CRLF_GUESS) {
 281                 if (crlf_action == CRLF_GUESS) {
 282                         /* If we have any CR or CRLF line endings, we do not touch it */
 283                         /* This is the new safer autocrlf-handling */
 284                         if (stats.cr > 0 || stats.crlf > 0)
 285                                 return 0;
 286                 }
 287
 288                 /* If we have any bare CR characters, we're not going to touch it */
 289                 if (stats.cr != stats.crlf)
 290                         return 0;
 291
 292                 if (is_binary(len, &stats))
 293                         return 0;
 294         }
 295
 296         /* are we "faking" in place editing ? */
 297         if (src == buf->buf)
 298                 to_free = strbuf_detach(buf, NULL);
 299
 300         strbuf_grow(buf, len + stats.lf - stats.crlf);
 301         for (;;) {
 302                 const char *nl = memchr(src, '\n', len);
 303                 if (!nl)
 304                         break;
 305                 if (nl > src && nl[-1] == '\r') {
 306                         strbuf_add(buf, src, nl + 1 - src);
 307                 } else {
 308                         strbuf_add(buf, src, nl - src);
 309                         strbuf_addstr(buf, "\r\n");
 310                 }
 311                 len -= nl + 1 - src;
 312                 src  = nl + 1;
 313         }
 314         strbuf_add(buf, src, len);
 315
 316         free(to_free);
 317         return 1;
 318 }
 319
 320 struct filter_params {
 321         const char *src;
 322         unsigned long size;
 323         const char *cmd;
 324         const char *path;
 325 };
 326
 327 static int filter_buffer(int in, int out, void *data)
 328 {
 329         /*
 330          * Spawn cmd and feed the buffer contents through its stdin.
 331          */
 332         struct child_process child_process;
 333         struct filter_params *params = (struct filter_params *)data;
 334         int write_err, status;
 335         const char *argv[] = { NULL, NULL };
 336
 337         /* apply % substitution to cmd */
 338         struct strbuf cmd = STRBUF_INIT;
 339         struct strbuf path = STRBUF_INIT;
 340         struct strbuf_expand_dict_entry dict[] = {
 341                 { "f", NULL, },
 342                 { NULL, NULL, },
 343         };
 344
 345         /* quote the path to preserve spaces, etc. */
 346         sq_quote_buf(&path, params->path);
 347         dict[0].value = path.buf;
 348
 349         /* expand all %f with the quoted path */
 350         strbuf_expand(&cmd, params->cmd, strbuf_expand_dict_cb, &dict);
 351         strbuf_release(&path);
 352
 353         argv[0] = cmd.buf;
 354
 355         memset(&child_process, 0, sizeof(child_process));
 356         child_process.argv = argv;
 357         child_process.use_shell = 1;
 358         child_process.in = -1;
 359         child_process.out = out;
 360
 361         if (start_command(&child_process))
 362                 return error("cannot fork to run external filter %s", params->cmd);
 363
 364         sigchain_push(SIGPIPE, SIG_IGN);
 365
 366         write_err = (write_in_full(child_process.in, params->src, params->size) < 0);
 367         if (close(child_process.in))
 368                 write_err = 1;
 369         if (write_err)
 370                 error("cannot feed the input to external filter %s", params->cmd);
 371
 372         sigchain_pop(SIGPIPE);
 373
 374         status = finish_command(&child_process);
 375         if (status)
 376                 error("external filter %s failed %d", params->cmd, status);
 377
 378         strbuf_release(&cmd);
 379         return (write_err || status);
 380 }
 381
 382 static int apply_filter(const char *path, const char *src, size_t len,
 383                         struct strbuf *dst, const char *cmd)
 384 {
 385         /*
 386          * Create a pipeline to have the command filter the buffer's
 387          * contents.
 388          *
 389          * (child --> cmd) --> us
 390          */
 391         int ret = 1;
 392         struct strbuf nbuf = STRBUF_INIT;
 393         struct async async;
 394         struct filter_params params;
 395
 396         if (!cmd)
 397                 return 0;
 398
 399         memset(&async, 0, sizeof(async));
 400         async.proc = filter_buffer;
 401         async.data = &params;
 402         async.out = -1;
 403         params.src = src;
 404         params.size = len;
 405         params.cmd = cmd;
 406         params.path = path;
 407
 408         fflush(NULL);
 409         if (start_async(&async))
 410                 return 0;       /* error was already reported */
 411
 412         if (strbuf_read(&nbuf, async.out, len) < 0) {
 413                 error("read from external filter %s failed", cmd);
 414                 ret = 0;
 415         }
 416         if (close(async.out)) {
 417                 error("read from external filter %s failed", cmd);
 418                 ret = 0;
 419         }
 420         if (finish_async(&async)) {
 421                 error("external filter %s failed", cmd);
 422                 ret = 0;
 423         }
 424
 425         if (ret) {
 426                 strbuf_swap(dst, &nbuf);
 427         }
 428         strbuf_release(&nbuf);
 429         return ret;
 430 }
 431
 432 static struct convert_driver {
 433         const char *name;
 434         struct convert_driver *next;
 435         const char *smudge;
 436         const char *clean;
 437         int required;
 438 } *user_convert, **user_convert_tail;
 439
 440 static int read_convert_config(const char *var, const char *value, void *cb)
 441 {
 442         const char *ep, *name;
 443         int namelen;
 444         struct convert_driver *drv;
 445
 446         /*
 447          * External conversion drivers are configured using
 448          * "filter.<name>.variable".
 449          */
 450         if (prefixcmp(var, "filter.") || (ep = strrchr(var, '.')) == var + 6)
 451                 return 0;
 452         name = var + 7;
 453         namelen = ep - name;
 454         for (drv = user_convert; drv; drv = drv->next)
 455                 if (!strncmp(drv->name, name, namelen) && !drv->name[namelen])
 456                         break;
 457         if (!drv) {
 458                 drv = xcalloc(1, sizeof(struct convert_driver));
 459                 drv->name = xmemdupz(name, namelen);
 460                 *user_convert_tail = drv;
 461                 user_convert_tail = &(drv->next);
 462         }
 463
 464         ep++;
 465
 466         /*
 467          * filter.<name>.smudge and filter.<name>.clean specifies
 468          * the command line:
 469          *
 470          *      command-line
 471          *
 472          * The command-line will not be interpolated in any way.
 473          */
 474
 475         if (!strcmp("smudge", ep))
 476                 return git_config_string(&drv->smudge, var, value);
 477
 478         if (!strcmp("clean", ep))
 479                 return git_config_string(&drv->clean, var, value);
 480
 481         if (!strcmp("required", ep)) {
 482                 drv->required = git_config_bool(var, value);
 483                 return 0;
 484         }
 485
 486         return 0;
 487 }
 488
 489 static int count_ident(const char *cp, unsigned long size)
 490 {
 491         /*
 492          * "$Id: 0000000000000000000000000000000000000000 $" <=> "$Id$"
 493          */
 494         int cnt = 0;
 495         char ch;
 496
 497         while (size) {
 498                 ch = *cp++;
 499                 size--;
 500                 if (ch != '$')
 501                         continue;
 502                 if (size < 3)
 503                         break;
 504                 if (memcmp("Id", cp, 2))
 505                         continue;
 506                 ch = cp[2];
 507                 cp += 3;
 508                 size -= 3;
 509                 if (ch == '$')
 510                         cnt++; /* $Id$ */
 511                 if (ch != ':')
 512                         continue;
 513
 514                 /*
 515                  * "$Id: ... "; scan up to the closing dollar sign and discard.
 516                  */
 517                 while (size) {
 518                         ch = *cp++;
 519                         size--;
 520                         if (ch == '$') {
 521                                 cnt++;
 522                                 break;
 523                         }
 524                         if (ch == '\n')
 525                                 break;
 526                 }
 527         }
 528         return cnt;
 529 }
 530
 531 static int ident_to_git(const char *path, const char *src, size_t len,
 532                         struct strbuf *buf, int ident)
 533 {
 534         char *dst, *dollar;
 535
 536         if (!ident || !count_ident(src, len))
 537                 return 0;
 538
 539         /* only grow if not in place */
 540         if (strbuf_avail(buf) + buf->len < len)
 541                 strbuf_grow(buf, len - buf->len);
 542         dst = buf->buf;
 543         for (;;) {
 544                 dollar = memchr(src, '$', len);
 545                 if (!dollar)
 546                         break;
 547                 memmove(dst, src, dollar + 1 - src);
 548                 dst += dollar + 1 - src;
 549                 len -= dollar + 1 - src;
 550                 src  = dollar + 1;
 551
 552                 if (len > 3 && !memcmp(src, "Id:", 3)) {
 553                         dollar = memchr(src + 3, '$', len - 3);
 554                         if (!dollar)
 555                                 break;
 556                         if (memchr(src + 3, '\n', dollar - src - 3)) {
 557                                 /* Line break before the next dollar. */
 558                                 continue;
 559                         }
 560
 561                         memcpy(dst, "Id$", 3);
 562                         dst += 3;
 563                         len -= dollar + 1 - src;
 564                         src  = dollar + 1;
 565                 }
 566         }
 567         memmove(dst, src, len);
 568         strbuf_setlen(buf, dst + len - buf->buf);
 569         return 1;
 570 }
 571
 572 static int ident_to_worktree(const char *path, const char *src, size_t len,
 573                              struct strbuf *buf, int ident)
 574 {
 575         unsigned char sha1[20];
 576         char *to_free = NULL, *dollar, *spc;
 577         int cnt;
 578
 579         if (!ident)
 580                 return 0;
 581
 582         cnt = count_ident(src, len);
 583         if (!cnt)
 584                 return 0;
 585
 586         /* are we "faking" in place editing ? */
 587         if (src == buf->buf)
 588                 to_free = strbuf_detach(buf, NULL);
 589         hash_sha1_file(src, len, "blob", sha1);
 590
 591         strbuf_grow(buf, len + cnt * 43);
 592         for (;;) {
 593                 /* step 1: run to the next '$' */
 594                 dollar = memchr(src, '$', len);
 595                 if (!dollar)
 596                         break;
 597                 strbuf_add(buf, src, dollar + 1 - src);
 598                 len -= dollar + 1 - src;
 599                 src  = dollar + 1;
 600
 601                 /* step 2: does it looks like a bit like Id:xxx$ or Id$ ? */
 602                 if (len < 3 || memcmp("Id", src, 2))
 603                         continue;
 604
 605                 /* step 3: skip over Id$ or Id:xxxxx$ */
 606                 if (src[2] == '$') {
 607                         src += 3;
 608                         len -= 3;
 609                 } else if (src[2] == ':') {
 610                         /*
 611                          * It's possible that an expanded Id has crept its way into the
 612                          * repository, we cope with that by stripping the expansion out.
 613                          * This is probably not a good idea, since it will cause changes
 614                          * on checkout, which won't go away by stash, but let's keep it
 615                          * for git-style ids.
 616                          */
 617                         dollar = memchr(src + 3, '$', len - 3);
 618                         if (!dollar) {
 619                                 /* incomplete keyword, no more '$', so just quit the loop */
 620                                 break;
 621                         }
 622
 623                         if (memchr(src + 3, '\n', dollar - src - 3)) {
 624                                 /* Line break before the next dollar. */
 625                                 continue;
 626                         }
 627
 628                         spc = memchr(src + 4, ' ', dollar - src - 4);
 629                         if (spc && spc < dollar-1) {
 630                                 /* There are spaces in unexpected places.
 631                                  * This is probably an id from some other
 632                                  * versioning system. Keep it for now.
 633                                  */
 634                                 continue;
 635                         }
 636
 637                         len -= dollar + 1 - src;
 638                         src  = dollar + 1;
 639                 } else {
 640                         /* it wasn't a "Id$" or "Id:xxxx$" */
 641                         continue;
 642                 }
 643
 644                 /* step 4: substitute */
 645                 strbuf_addstr(buf, "Id: ");
 646                 strbuf_add(buf, sha1_to_hex(sha1), 40);
 647                 strbuf_addstr(buf, " $");
 648         }
 649         strbuf_add(buf, src, len);
 650
 651         free(to_free);
 652         return 1;
 653 }
 654
 655 static enum crlf_action git_path_check_crlf(const char *path, struct git_attr_check *check)
 656 {
 657         const char *value = check->value;
 658
 659         if (ATTR_TRUE(value))
 660                 return CRLF_TEXT;
 661         else if (ATTR_FALSE(value))
 662                 return CRLF_BINARY;
 663         else if (ATTR_UNSET(value))
 664                 ;
 665         else if (!strcmp(value, "input"))
 666                 return CRLF_INPUT;
 667         else if (!strcmp(value, "auto"))
 668                 return CRLF_AUTO;
 669         return CRLF_GUESS;
 670 }
 671
 672 static enum eol git_path_check_eol(const char *path, struct git_attr_check *check)
 673 {
 674         const char *value = check->value;
 675
 676         if (ATTR_UNSET(value))
 677                 ;
 678         else if (!strcmp(value, "lf"))
 679                 return EOL_LF;
 680         else if (!strcmp(value, "crlf"))
 681                 return EOL_CRLF;
 682         return EOL_UNSET;
 683 }
 684
 685 static struct convert_driver *git_path_check_convert(const char *path,
 686                                              struct git_attr_check *check)
 687 {
 688         const char *value = check->value;
 689         struct convert_driver *drv;
 690
 691         if (ATTR_TRUE(value) || ATTR_FALSE(value) || ATTR_UNSET(value))
 692                 return NULL;
 693         for (drv = user_convert; drv; drv = drv->next)
 694                 if (!strcmp(value, drv->name))
 695                         return drv;
 696         return NULL;
 697 }
 698
 699 static int git_path_check_ident(const char *path, struct git_attr_check *check)
 700 {
 701         const char *value = check->value;
 702
 703         return !!ATTR_TRUE(value);
 704 }
 705
 706 static enum crlf_action input_crlf_action(enum crlf_action text_attr, enum eol eol_attr)
 707 {
 708         if (text_attr == CRLF_BINARY)
 709                 return CRLF_BINARY;
 710         if (eol_attr == EOL_LF)
 711                 return CRLF_INPUT;
 712         if (eol_attr == EOL_CRLF)
 713                 return CRLF_CRLF;
 714         return text_attr;
 715 }
 716
 717 struct conv_attrs {
 718         struct convert_driver *drv;
 719         enum crlf_action crlf_action;
 720         enum eol eol_attr;
 721         int ident;
 722 };
 723
 724 static const char *conv_attr_name[] = {
 725         "crlf", "ident", "filter", "eol", "text",
 726 };
 727 #define NUM_CONV_ATTRS ARRAY_SIZE(conv_attr_name)
 728
 729 static void convert_attrs(struct conv_attrs *ca, const char *path)
 730 {
 731         int i;
 732         static struct git_attr_check ccheck[NUM_CONV_ATTRS];
 733
 734         if (!ccheck[0].attr) {
 735                 for (i = 0; i < NUM_CONV_ATTRS; i++)
 736                         ccheck[i].attr = git_attr(conv_attr_name[i]);
 737                 user_convert_tail = &user_convert;
 738                 git_config(read_convert_config, NULL);
 739         }
 740
 741         if (!git_check_attr(path, NUM_CONV_ATTRS, ccheck)) {
 742                 ca->crlf_action = git_path_check_crlf(path, ccheck + 4);
 743                 if (ca->crlf_action == CRLF_GUESS)
 744                         ca->crlf_action = git_path_check_crlf(path, ccheck + 0);
 745                 ca->ident = git_path_check_ident(path, ccheck + 1);
 746                 ca->drv = git_path_check_convert(path, ccheck + 2);
 747                 ca->eol_attr = git_path_check_eol(path, ccheck + 3);
 748         } else {
 749                 ca->drv = NULL;
 750                 ca->crlf_action = CRLF_GUESS;
 751                 ca->eol_attr = EOL_UNSET;
 752                 ca->ident = 0;
 753         }
 754 }
 755
 756 int convert_to_git(const char *path, const char *src, size_t len,
 757                    struct strbuf *dst, enum safe_crlf checksafe)
 758 {
 759         int ret = 0;
 760         const char *filter = NULL;
 761         int required = 0;
 762         struct conv_attrs ca;
 763
 764         convert_attrs(&ca, path);
 765         if (ca.drv) {
 766                 filter = ca.drv->clean;
 767                 required = ca.drv->required;
 768         }
 769
 770         ret |= apply_filter(path, src, len, dst, filter);
 771         if (!ret && required)
 772                 die("%s: clean filter '%s' failed", path, ca.drv->name);
 773
 774         if (ret) {
 775                 src = dst->buf;
 776                 len = dst->len;
 777         }
 778         ca.crlf_action = input_crlf_action(ca.crlf_action, ca.eol_attr);
 779         ret |= crlf_to_git(path, src, len, dst, ca.crlf_action, checksafe);
 780         if (ret) {
 781                 src = dst->buf;
 782                 len = dst->len;
 783         }
 784         return ret | ident_to_git(path, src, len, dst, ca.ident);
 785 }
 786
 787 static int convert_to_working_tree_internal(const char *path, const char *src,
 788                                             size_t len, struct strbuf *dst,
 789                                             int normalizing)
 790 {
 791         int ret = 0, ret_filter = 0;
 792         const char *filter = NULL;
 793         int required = 0;
 794         struct conv_attrs ca;
 795
 796         convert_attrs(&ca, path);
 797         if (ca.drv) {
 798                 filter = ca.drv->smudge;
 799                 required = ca.drv->required;
 800         }
 801
 802         ret |= ident_to_worktree(path, src, len, dst, ca.ident);
 803         if (ret) {
 804                 src = dst->buf;
 805                 len = dst->len;
 806         }
 807         /*
 808          * CRLF conversion can be skipped if normalizing, unless there
 809          * is a smudge filter.  The filter might expect CRLFs.
 810          */
 811         if (filter || !normalizing) {
 812                 ca.crlf_action = input_crlf_action(ca.crlf_action, ca.eol_attr);
 813                 ret |= crlf_to_worktree(path, src, len, dst, ca.crlf_action);
 814                 if (ret) {
 815                         src = dst->buf;
 816                         len = dst->len;
 817                 }
 818         }
 819
 820         ret_filter = apply_filter(path, src, len, dst, filter);
 821         if (!ret_filter && required)
 822                 die("%s: smudge filter %s failed", path, ca.drv->name);
 823
 824         return ret | ret_filter;
 825 }
 826
 827 int convert_to_working_tree(const char *path, const char *src, size_t len, struct strbuf *dst)
 828 {
 829         return convert_to_working_tree_internal(path, src, len, dst, 0);
 830 }
 831
 832 int renormalize_buffer(const char *path, const char *src, size_t len, struct strbuf *dst)
 833 {
 834         int ret = convert_to_working_tree_internal(path, src, len, dst, 1);
 835         if (ret) {
 836                 src = dst->buf;
 837                 len = dst->len;
 838         }
 839         return ret | convert_to_git(path, src, len, dst, SAFE_CRLF_FALSE);
 840 }
 841
 842 /*****************************************************************
 843  *
 844  * Streaming converison support
 845  *
 846  *****************************************************************/
 847
 848 typedef int (*filter_fn)(struct stream_filter *,
 849                          const char *input, size_t *isize_p,
 850                          char *output, size_t *osize_p);
 851 typedef void (*free_fn)(struct stream_filter *);
 852
 853 struct stream_filter_vtbl {
 854         filter_fn filter;
 855         free_fn free;
 856 };
 857
 858 struct stream_filter {
 859         struct stream_filter_vtbl *vtbl;
 860 };
 861
 862 static int null_filter_fn(struct stream_filter *filter,
 863                           const char *input, size_t *isize_p,
 864                           char *output, size_t *osize_p)
 865 {
 866         size_t count;
 867
 868         if (!input)
 869                 return 0; /* we do not keep any states */
 870         count = *isize_p;
 871         if (*osize_p < count)
 872                 count = *osize_p;
 873         if (count) {
 874                 memmove(output, input, count);
 875                 *isize_p -= count;
 876                 *osize_p -= count;
 877         }
 878         return 0;
 879 }
 880
 881 static void null_free_fn(struct stream_filter *filter)
 882 {
 883         ; /* nothing -- null instances are shared */
 884 }
 885
 886 static struct stream_filter_vtbl null_vtbl = {
 887         null_filter_fn,
 888         null_free_fn,
 889 };
 890
 891 static struct stream_filter null_filter_singleton = {
 892         &null_vtbl,
 893 };
 894
 895 int is_null_stream_filter(struct stream_filter *filter)
 896 {
 897         return filter == &null_filter_singleton;
 898 }
 899
 900
 901 /*
 902  * LF-to-CRLF filter
 903  */
 904
 905 struct lf_to_crlf_filter {
 906         struct stream_filter filter;
 907         unsigned has_held:1;
 908         char held;
 909 };
 910
 911 static int lf_to_crlf_filter_fn(struct stream_filter *filter,
 912                                 const char *input, size_t *isize_p,
 913                                 char *output, size_t *osize_p)
 914 {
 915         size_t count, o = 0;
 916         struct lf_to_crlf_filter *lf_to_crlf = (struct lf_to_crlf_filter *)filter;
 917
 918         /*
 919          * We may be holding onto the CR to see if it is followed by a
 920          * LF, in which case we would need to go to the main loop.
 921          * Otherwise, just emit it to the output stream.
 922          */
 923         if (lf_to_crlf->has_held && (lf_to_crlf->held != '\r' || !input)) {
 924                 output[o++] = lf_to_crlf->held;
 925                 lf_to_crlf->has_held = 0;
 926         }
 927
 928         /* We are told to drain */
 929         if (!input) {
 930                 *osize_p -= o;
 931                 return 0;
 932         }
 933
 934         count = *isize_p;
 935         if (count || lf_to_crlf->has_held) {
 936                 size_t i;
 937                 int was_cr = 0;
 938
 939                 if (lf_to_crlf->has_held) {
 940                         was_cr = 1;
 941                         lf_to_crlf->has_held = 0;
 942                 }
 943
 944                 for (i = 0; o < *osize_p && i < count; i++) {
 945                         char ch = input[i];
 946
 947                         if (ch == '\n') {
 948                                 output[o++] = '\r';
 949                         } else if (was_cr) {
 950                                 /*
 951                                  * Previous round saw CR and it is not followed
 952                                  * by a LF; emit the CR before processing the
 953                                  * current character.
 954                                  */
 955                                 output[o++] = '\r';
 956                         }
 957
 958                         /*
 959                          * We may have consumed the last output slot,
 960                          * in which case we need to break out of this
 961                          * loop; hold the current character before
 962                          * returning.
 963                          */
 964                         if (*osize_p <= o) {
 965                                 lf_to_crlf->has_held = 1;
 966                                 lf_to_crlf->held = ch;
 967                                 continue; /* break but increment i */
 968                         }
 969
 970                         if (ch == '\r') {
 971                                 was_cr = 1;
 972                                 continue;
 973                         }
 974
 975                         was_cr = 0;
 976                         output[o++] = ch;
 977                 }
 978
 979                 *osize_p -= o;
 980                 *isize_p -= i;
 981
 982                 if (!lf_to_crlf->has_held && was_cr) {
 983                         lf_to_crlf->has_held = 1;
 984                         lf_to_crlf->held = '\r';
 985                 }
 986         }
 987         return 0;
 988 }
 989
 990 static void lf_to_crlf_free_fn(struct stream_filter *filter)
 991 {
 992         free(filter);
 993 }
 994
 995 static struct stream_filter_vtbl lf_to_crlf_vtbl = {
 996         lf_to_crlf_filter_fn,
 997         lf_to_crlf_free_fn,
 998 };
 999
1000 static struct stream_filter *lf_to_crlf_filter(void)
1001 {
1002         struct lf_to_crlf_filter *lf_to_crlf = xcalloc(1, sizeof(*lf_to_crlf));
1003
1004         lf_to_crlf->filter.vtbl = &lf_to_crlf_vtbl;
1005         return (struct stream_filter *)lf_to_crlf;
1006 }
1007
1008 /*
1009  * Cascade filter
1010  */
1011 #define FILTER_BUFFER 1024
1012 struct cascade_filter {
1013         struct stream_filter filter;
1014         struct stream_filter *one;
1015         struct stream_filter *two;
1016         char buf[FILTER_BUFFER];
1017         int end, ptr;
1018 };
1019
1020 static int cascade_filter_fn(struct stream_filter *filter,
1021                              const char *input, size_t *isize_p,
1022                              char *output, size_t *osize_p)
1023 {
1024         struct cascade_filter *cas = (struct cascade_filter *) filter;
1025         size_t filled = 0;
1026         size_t sz = *osize_p;
1027         size_t to_feed, remaining;
1028
1029         /*
1030          * input -- (one) --> buf -- (two) --> output
1031          */
1032         while (filled < sz) {
1033                 remaining = sz - filled;
1034
1035                 /* do we already have something to feed two with? */
1036                 if (cas->ptr < cas->end) {
1037                         to_feed = cas->end - cas->ptr;
1038                         if (stream_filter(cas->two,
1039                                           cas->buf + cas->ptr, &to_feed,
1040                                           output + filled, &remaining))
1041                                 return -1;
1042                         cas->ptr += (cas->end - cas->ptr) - to_feed;
1043                         filled = sz - remaining;
1044                         continue;
1045                 }
1046
1047                 /* feed one from upstream and have it emit into our buffer */
1048                 to_feed = input ? *isize_p : 0;
1049                 if (input && !to_feed)
1050                         break;
1051                 remaining = sizeof(cas->buf);
1052                 if (stream_filter(cas->one,
1053                                   input, &to_feed,
1054                                   cas->buf, &remaining))
1055                         return -1;
1056                 cas->end = sizeof(cas->buf) - remaining;
1057                 cas->ptr = 0;
1058                 if (input) {
1059                         size_t fed = *isize_p - to_feed;
1060                         *isize_p -= fed;
1061                         input += fed;
1062                 }
1063
1064                 /* do we know that we drained one completely? */
1065                 if (input || cas->end)
1066                         continue;
1067
1068                 /* tell two to drain; we have nothing more to give it */
1069                 to_feed = 0;
1070                 remaining = sz - filled;
1071                 if (stream_filter(cas->two,
1072                                   NULL, &to_feed,
1073                                   output + filled, &remaining))
1074                         return -1;
1075                 if (remaining == (sz - filled))
1076                         break; /* completely drained two */
1077                 filled = sz - remaining;
1078         }
1079         *osize_p -= filled;
1080         return 0;
1081 }
1082
1083 static void cascade_free_fn(struct stream_filter *filter)
1084 {
1085         struct cascade_filter *cas = (struct cascade_filter *)filter;
1086         free_stream_filter(cas->one);
1087         free_stream_filter(cas->two);
1088         free(filter);
1089 }
1090
1091 static struct stream_filter_vtbl cascade_vtbl = {
1092         cascade_filter_fn,
1093         cascade_free_fn,
1094 };
1095
1096 static struct stream_filter *cascade_filter(struct stream_filter *one,
1097                                             struct stream_filter *two)
1098 {
1099         struct cascade_filter *cascade;
1100
1101         if (!one || is_null_stream_filter(one))
1102                 return two;
1103         if (!two || is_null_stream_filter(two))
1104                 return one;
1105
1106         cascade = xmalloc(sizeof(*cascade));
1107         cascade->one = one;
1108         cascade->two = two;
1109         cascade->end = cascade->ptr = 0;
1110         cascade->filter.vtbl = &cascade_vtbl;
1111         return (struct stream_filter *)cascade;
1112 }
1113
1114 /*
1115  * ident filter
1116  */
1117 #define IDENT_DRAINING (-1)
1118 #define IDENT_SKIPPING (-2)
1119 struct ident_filter {
1120         struct stream_filter filter;
1121         struct strbuf left;
1122         int state;
1123         char ident[45]; /* ": x40 $" */
1124 };
1125
1126 static int is_foreign_ident(const char *str)
1127 {
1128         int i;
1129
1130         if (prefixcmp(str, "$Id: "))
1131                 return 0;
1132         for (i = 5; str[i]; i++) {
1133                 if (isspace(str[i]) && str[i+1] != '$')
1134                         return 1;
1135         }
1136         return 0;
1137 }
1138
1139 static void ident_drain(struct ident_filter *ident, char **output_p, size_t *osize_p)
1140 {
1141         size_t to_drain = ident->left.len;
1142
1143         if (*osize_p < to_drain)
1144                 to_drain = *osize_p;
1145         if (to_drain) {
1146                 memcpy(*output_p, ident->left.buf, to_drain);
1147                 strbuf_remove(&ident->left, 0, to_drain);
1148                 *output_p += to_drain;
1149                 *osize_p -= to_drain;
1150         }
1151         if (!ident->left.len)
1152                 ident->state = 0;
1153 }
1154
1155 static int ident_filter_fn(struct stream_filter *filter,
1156                            const char *input, size_t *isize_p,
1157                            char *output, size_t *osize_p)
1158 {
1159         struct ident_filter *ident = (struct ident_filter *)filter;
1160         static const char head[] = "$Id";
1161
1162         if (!input) {
1163                 /* drain upon eof */
1164                 switch (ident->state) {
1165                 default:
1166                         strbuf_add(&ident->left, head, ident->state);
1167                 case IDENT_SKIPPING:
1168                         /* fallthru */
1169                 case IDENT_DRAINING:
1170                         ident_drain(ident, &output, osize_p);
1171                 }
1172                 return 0;
1173         }
1174
1175         while (*isize_p || (ident->state == IDENT_DRAINING)) {
1176                 int ch;
1177
1178                 if (ident->state == IDENT_DRAINING) {
1179                         ident_drain(ident, &output, osize_p);
1180                         if (!*osize_p)
1181                                 break;
1182                         continue;
1183                 }
1184
1185                 ch = *(input++);
1186                 (*isize_p)--;
1187
1188                 if (ident->state == IDENT_SKIPPING) {
1189                         /*
1190                          * Skipping until '$' or LF, but keeping them
1191                          * in case it is a foreign ident.
1192                          */
1193                         strbuf_addch(&ident->left, ch);
1194                         if (ch != '\n' && ch != '$')
1195                                 continue;
1196                         if (ch == '$' && !is_foreign_ident(ident->left.buf)) {
1197                                 strbuf_setlen(&ident->left, sizeof(head) - 1);
1198                                 strbuf_addstr(&ident->left, ident->ident);
1199                         }
1200                         ident->state = IDENT_DRAINING;
1201                         continue;
1202                 }
1203
1204                 if (ident->state < sizeof(head) &&
1205                     head[ident->state] == ch) {
1206                         ident->state++;
1207                         continue;
1208                 }
1209
1210                 if (ident->state)
1211                         strbuf_add(&ident->left, head, ident->state);
1212                 if (ident->state == sizeof(head) - 1) {
1213                         if (ch != ':' && ch != '$') {
1214                                 strbuf_addch(&ident->left, ch);
1215                                 ident->state = 0;
1216                                 continue;
1217                         }
1218
1219                         if (ch == ':') {
1220                                 strbuf_addch(&ident->left, ch);
1221                                 ident->state = IDENT_SKIPPING;
1222                         } else {
1223                                 strbuf_addstr(&ident->left, ident->ident);
1224                                 ident->state = IDENT_DRAINING;
1225                         }
1226                         continue;
1227                 }
1228
1229                 strbuf_addch(&ident->left, ch);
1230                 ident->state = IDENT_DRAINING;
1231         }
1232         return 0;
1233 }
1234
1235 static void ident_free_fn(struct stream_filter *filter)
1236 {
1237         struct ident_filter *ident = (struct ident_filter *)filter;
1238         strbuf_release(&ident->left);
1239         free(filter);
1240 }
1241
1242 static struct stream_filter_vtbl ident_vtbl = {
1243         ident_filter_fn,
1244         ident_free_fn,
1245 };
1246
1247 static struct stream_filter *ident_filter(const unsigned char *sha1)
1248 {
1249         struct ident_filter *ident = xmalloc(sizeof(*ident));
1250
1251         sprintf(ident->ident, ": %s $", sha1_to_hex(sha1));
1252         strbuf_init(&ident->left, 0);
1253         ident->filter.vtbl = &ident_vtbl;
1254         ident->state = 0;
1255         return (struct stream_filter *)ident;
1256 }
1257
1258 /*
1259  * Return an appropriately constructed filter for the path, or NULL if
1260  * the contents cannot be filtered without reading the whole thing
1261  * in-core.
1262  *
1263  * Note that you would be crazy to set CRLF, smuge/clean or ident to a
1264  * large binary blob you would want us not to slurp into the memory!
1265  */
1266 struct stream_filter *get_stream_filter(const char *path, const unsigned char *sha1)
1267 {
1268         struct conv_attrs ca;
1269         enum crlf_action crlf_action;
1270         struct stream_filter *filter = NULL;
1271
1272         convert_attrs(&ca, path);
1273
1274         if (ca.drv && (ca.drv->smudge || ca.drv->clean))
1275                 return filter;
1276
1277         if (ca.ident)
1278                 filter = ident_filter(sha1);
1279
1280         crlf_action = input_crlf_action(ca.crlf_action, ca.eol_attr);
1281
1282         if ((crlf_action == CRLF_BINARY) || (crlf_action == CRLF_INPUT) ||
1283             (crlf_action == CRLF_GUESS && auto_crlf == AUTO_CRLF_FALSE))
1284                 filter = cascade_filter(filter, &null_filter_singleton);
1285
1286         else if (output_eol(crlf_action) == EOL_CRLF &&
1287                  !(crlf_action == CRLF_AUTO || crlf_action == CRLF_GUESS))
1288                 filter = cascade_filter(filter, lf_to_crlf_filter());
1289
1290         return filter;
1291 }
1292
1293 void free_stream_filter(struct stream_filter *filter)
1294 {
1295         filter->vtbl->free(filter);
1296 }
1297
1298 int stream_filter(struct stream_filter *filter,
1299                   const char *input, size_t *isize_p,
1300                   char *output, size_t *osize_p)
1301 {
1302         return filter->vtbl->filter(filter, input, isize_p, output, osize_p);
1303 }