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 } *user_convert, **user_convert_tail;
 438
 439 static int read_convert_config(const char *var, const char *value, void *cb)
 440 {
 441         const char *ep, *name;
 442         int namelen;
 443         struct convert_driver *drv;
 444
 445         /*
 446          * External conversion drivers are configured using
 447          * "filter.<name>.variable".
 448          */
 449         if (prefixcmp(var, "filter.") || (ep = strrchr(var, '.')) == var + 6)
 450                 return 0;
 451         name = var + 7;
 452         namelen = ep - name;
 453         for (drv = user_convert; drv; drv = drv->next)
 454                 if (!strncmp(drv->name, name, namelen) && !drv->name[namelen])
 455                         break;
 456         if (!drv) {
 457                 drv = xcalloc(1, sizeof(struct convert_driver));
 458                 drv->name = xmemdupz(name, namelen);
 459                 *user_convert_tail = drv;
 460                 user_convert_tail = &(drv->next);
 461         }
 462
 463         ep++;
 464
 465         /*
 466          * filter.<name>.smudge and filter.<name>.clean specifies
 467          * the command line:
 468          *
 469          *      command-line
 470          *
 471          * The command-line will not be interpolated in any way.
 472          */
 473
 474         if (!strcmp("smudge", ep))
 475                 return git_config_string(&drv->smudge, var, value);
 476
 477         if (!strcmp("clean", ep))
 478                 return git_config_string(&drv->clean, var, value);
 479
 480         return 0;
 481 }
 482
 483 static int count_ident(const char *cp, unsigned long size)
 484 {
 485         /*
 486          * "$Id: 0000000000000000000000000000000000000000 $" <=> "$Id$"
 487          */
 488         int cnt = 0;
 489         char ch;
 490
 491         while (size) {
 492                 ch = *cp++;
 493                 size--;
 494                 if (ch != '$')
 495                         continue;
 496                 if (size < 3)
 497                         break;
 498                 if (memcmp("Id", cp, 2))
 499                         continue;
 500                 ch = cp[2];
 501                 cp += 3;
 502                 size -= 3;
 503                 if (ch == '$')
 504                         cnt++; /* $Id$ */
 505                 if (ch != ':')
 506                         continue;
 507
 508                 /*
 509                  * "$Id: ... "; scan up to the closing dollar sign and discard.
 510                  */
 511                 while (size) {
 512                         ch = *cp++;
 513                         size--;
 514                         if (ch == '$') {
 515                                 cnt++;
 516                                 break;
 517                         }
 518                         if (ch == '\n')
 519                                 break;
 520                 }
 521         }
 522         return cnt;
 523 }
 524
 525 static int ident_to_git(const char *path, const char *src, size_t len,
 526                         struct strbuf *buf, int ident)
 527 {
 528         char *dst, *dollar;
 529
 530         if (!ident || !count_ident(src, len))
 531                 return 0;
 532
 533         /* only grow if not in place */
 534         if (strbuf_avail(buf) + buf->len < len)
 535                 strbuf_grow(buf, len - buf->len);
 536         dst = buf->buf;
 537         for (;;) {
 538                 dollar = memchr(src, '$', len);
 539                 if (!dollar)
 540                         break;
 541                 memmove(dst, src, dollar + 1 - src);
 542                 dst += dollar + 1 - src;
 543                 len -= dollar + 1 - src;
 544                 src  = dollar + 1;
 545
 546                 if (len > 3 && !memcmp(src, "Id:", 3)) {
 547                         dollar = memchr(src + 3, '$', len - 3);
 548                         if (!dollar)
 549                                 break;
 550                         if (memchr(src + 3, '\n', dollar - src - 3)) {
 551                                 /* Line break before the next dollar. */
 552                                 continue;
 553                         }
 554
 555                         memcpy(dst, "Id$", 3);
 556                         dst += 3;
 557                         len -= dollar + 1 - src;
 558                         src  = dollar + 1;
 559                 }
 560         }
 561         memmove(dst, src, len);
 562         strbuf_setlen(buf, dst + len - buf->buf);
 563         return 1;
 564 }
 565
 566 static int ident_to_worktree(const char *path, const char *src, size_t len,
 567                              struct strbuf *buf, int ident)
 568 {
 569         unsigned char sha1[20];
 570         char *to_free = NULL, *dollar, *spc;
 571         int cnt;
 572
 573         if (!ident)
 574                 return 0;
 575
 576         cnt = count_ident(src, len);
 577         if (!cnt)
 578                 return 0;
 579
 580         /* are we "faking" in place editing ? */
 581         if (src == buf->buf)
 582                 to_free = strbuf_detach(buf, NULL);
 583         hash_sha1_file(src, len, "blob", sha1);
 584
 585         strbuf_grow(buf, len + cnt * 43);
 586         for (;;) {
 587                 /* step 1: run to the next '$' */
 588                 dollar = memchr(src, '$', len);
 589                 if (!dollar)
 590                         break;
 591                 strbuf_add(buf, src, dollar + 1 - src);
 592                 len -= dollar + 1 - src;
 593                 src  = dollar + 1;
 594
 595                 /* step 2: does it looks like a bit like Id:xxx$ or Id$ ? */
 596                 if (len < 3 || memcmp("Id", src, 2))
 597                         continue;
 598
 599                 /* step 3: skip over Id$ or Id:xxxxx$ */
 600                 if (src[2] == '$') {
 601                         src += 3;
 602                         len -= 3;
 603                 } else if (src[2] == ':') {
 604                         /*
 605                          * It's possible that an expanded Id has crept its way into the
 606                          * repository, we cope with that by stripping the expansion out.
 607                          * This is probably not a good idea, since it will cause changes
 608                          * on checkout, which won't go away by stash, but let's keep it
 609                          * for git-style ids.
 610                          */
 611                         dollar = memchr(src + 3, '$', len - 3);
 612                         if (!dollar) {
 613                                 /* incomplete keyword, no more '$', so just quit the loop */
 614                                 break;
 615                         }
 616
 617                         if (memchr(src + 3, '\n', dollar - src - 3)) {
 618                                 /* Line break before the next dollar. */
 619                                 continue;
 620                         }
 621
 622                         spc = memchr(src + 4, ' ', dollar - src - 4);
 623                         if (spc && spc < dollar-1) {
 624                                 /* There are spaces in unexpected places.
 625                                  * This is probably an id from some other
 626                                  * versioning system. Keep it for now.
 627                                  */
 628                                 continue;
 629                         }
 630
 631                         len -= dollar + 1 - src;
 632                         src  = dollar + 1;
 633                 } else {
 634                         /* it wasn't a "Id$" or "Id:xxxx$" */
 635                         continue;
 636                 }
 637
 638                 /* step 4: substitute */
 639                 strbuf_addstr(buf, "Id: ");
 640                 strbuf_add(buf, sha1_to_hex(sha1), 40);
 641                 strbuf_addstr(buf, " $");
 642         }
 643         strbuf_add(buf, src, len);
 644
 645         free(to_free);
 646         return 1;
 647 }
 648
 649 static enum crlf_action git_path_check_crlf(const char *path, struct git_attr_check *check)
 650 {
 651         const char *value = check->value;
 652
 653         if (ATTR_TRUE(value))
 654                 return CRLF_TEXT;
 655         else if (ATTR_FALSE(value))
 656                 return CRLF_BINARY;
 657         else if (ATTR_UNSET(value))
 658                 ;
 659         else if (!strcmp(value, "input"))
 660                 return CRLF_INPUT;
 661         else if (!strcmp(value, "auto"))
 662                 return CRLF_AUTO;
 663         return CRLF_GUESS;
 664 }
 665
 666 static enum eol git_path_check_eol(const char *path, struct git_attr_check *check)
 667 {
 668         const char *value = check->value;
 669
 670         if (ATTR_UNSET(value))
 671                 ;
 672         else if (!strcmp(value, "lf"))
 673                 return EOL_LF;
 674         else if (!strcmp(value, "crlf"))
 675                 return EOL_CRLF;
 676         return EOL_UNSET;
 677 }
 678
 679 static struct convert_driver *git_path_check_convert(const char *path,
 680                                              struct git_attr_check *check)
 681 {
 682         const char *value = check->value;
 683         struct convert_driver *drv;
 684
 685         if (ATTR_TRUE(value) || ATTR_FALSE(value) || ATTR_UNSET(value))
 686                 return NULL;
 687         for (drv = user_convert; drv; drv = drv->next)
 688                 if (!strcmp(value, drv->name))
 689                         return drv;
 690         return NULL;
 691 }
 692
 693 static int git_path_check_ident(const char *path, struct git_attr_check *check)
 694 {
 695         const char *value = check->value;
 696
 697         return !!ATTR_TRUE(value);
 698 }
 699
 700 static enum crlf_action input_crlf_action(enum crlf_action text_attr, enum eol eol_attr)
 701 {
 702         if (text_attr == CRLF_BINARY)
 703                 return CRLF_BINARY;
 704         if (eol_attr == EOL_LF)
 705                 return CRLF_INPUT;
 706         if (eol_attr == EOL_CRLF)
 707                 return CRLF_CRLF;
 708         return text_attr;
 709 }
 710
 711 struct conv_attrs {
 712         struct convert_driver *drv;
 713         enum crlf_action crlf_action;
 714         enum eol eol_attr;
 715         int ident;
 716 };
 717
 718 static const char *conv_attr_name[] = {
 719         "crlf", "ident", "filter", "eol", "text",
 720 };
 721 #define NUM_CONV_ATTRS ARRAY_SIZE(conv_attr_name)
 722
 723 static void convert_attrs(struct conv_attrs *ca, const char *path)
 724 {
 725         int i;
 726         static struct git_attr_check ccheck[NUM_CONV_ATTRS];
 727
 728         if (!ccheck[0].attr) {
 729                 for (i = 0; i < NUM_CONV_ATTRS; i++)
 730                         ccheck[i].attr = git_attr(conv_attr_name[i]);
 731                 user_convert_tail = &user_convert;
 732                 git_config(read_convert_config, NULL);
 733         }
 734
 735         if (!git_check_attr(path, NUM_CONV_ATTRS, ccheck)) {
 736                 ca->crlf_action = git_path_check_crlf(path, ccheck + 4);
 737                 if (ca->crlf_action == CRLF_GUESS)
 738                         ca->crlf_action = git_path_check_crlf(path, ccheck + 0);
 739                 ca->ident = git_path_check_ident(path, ccheck + 1);
 740                 ca->drv = git_path_check_convert(path, ccheck + 2);
 741                 ca->eol_attr = git_path_check_eol(path, ccheck + 3);
 742         } else {
 743                 ca->drv = NULL;
 744                 ca->crlf_action = CRLF_GUESS;
 745                 ca->eol_attr = EOL_UNSET;
 746                 ca->ident = 0;
 747         }
 748 }
 749
 750 int convert_to_git(const char *path, const char *src, size_t len,
 751                    struct strbuf *dst, enum safe_crlf checksafe)
 752 {
 753         int ret = 0;
 754         const char *filter = NULL;
 755         struct conv_attrs ca;
 756
 757         convert_attrs(&ca, path);
 758         if (ca.drv)
 759                 filter = ca.drv->clean;
 760
 761         ret |= apply_filter(path, src, len, dst, filter);
 762         if (ret) {
 763                 src = dst->buf;
 764                 len = dst->len;
 765         }
 766         ca.crlf_action = input_crlf_action(ca.crlf_action, ca.eol_attr);
 767         ret |= crlf_to_git(path, src, len, dst, ca.crlf_action, checksafe);
 768         if (ret) {
 769                 src = dst->buf;
 770                 len = dst->len;
 771         }
 772         return ret | ident_to_git(path, src, len, dst, ca.ident);
 773 }
 774
 775 static int convert_to_working_tree_internal(const char *path, const char *src,
 776                                             size_t len, struct strbuf *dst,
 777                                             int normalizing)
 778 {
 779         int ret = 0;
 780         const char *filter = NULL;
 781         struct conv_attrs ca;
 782
 783         convert_attrs(&ca, path);
 784         if (ca.drv)
 785                 filter = ca.drv->smudge;
 786
 787         ret |= ident_to_worktree(path, src, len, dst, ca.ident);
 788         if (ret) {
 789                 src = dst->buf;
 790                 len = dst->len;
 791         }
 792         /*
 793          * CRLF conversion can be skipped if normalizing, unless there
 794          * is a smudge filter.  The filter might expect CRLFs.
 795          */
 796         if (filter || !normalizing) {
 797                 ca.crlf_action = input_crlf_action(ca.crlf_action, ca.eol_attr);
 798                 ret |= crlf_to_worktree(path, src, len, dst, ca.crlf_action);
 799                 if (ret) {
 800                         src = dst->buf;
 801                         len = dst->len;
 802                 }
 803         }
 804         return ret | apply_filter(path, src, len, dst, filter);
 805 }
 806
 807 int convert_to_working_tree(const char *path, const char *src, size_t len, struct strbuf *dst)
 808 {
 809         return convert_to_working_tree_internal(path, src, len, dst, 0);
 810 }
 811
 812 int renormalize_buffer(const char *path, const char *src, size_t len, struct strbuf *dst)
 813 {
 814         int ret = convert_to_working_tree_internal(path, src, len, dst, 1);
 815         if (ret) {
 816                 src = dst->buf;
 817                 len = dst->len;
 818         }
 819         return ret | convert_to_git(path, src, len, dst, SAFE_CRLF_FALSE);
 820 }
 821
 822 /*****************************************************************
 823  *
 824  * Streaming converison support
 825  *
 826  *****************************************************************/
 827
 828 typedef int (*filter_fn)(struct stream_filter *,
 829                          const char *input, size_t *isize_p,
 830                          char *output, size_t *osize_p);
 831 typedef void (*free_fn)(struct stream_filter *);
 832
 833 struct stream_filter_vtbl {
 834         filter_fn filter;
 835         free_fn free;
 836 };
 837
 838 struct stream_filter {
 839         struct stream_filter_vtbl *vtbl;
 840 };
 841
 842 static int null_filter_fn(struct stream_filter *filter,
 843                           const char *input, size_t *isize_p,
 844                           char *output, size_t *osize_p)
 845 {
 846         size_t count;
 847
 848         if (!input)
 849                 return 0; /* we do not keep any states */
 850         count = *isize_p;
 851         if (*osize_p < count)
 852                 count = *osize_p;
 853         if (count) {
 854                 memmove(output, input, count);
 855                 *isize_p -= count;
 856                 *osize_p -= count;
 857         }
 858         return 0;
 859 }
 860
 861 static void null_free_fn(struct stream_filter *filter)
 862 {
 863         ; /* nothing -- null instances are shared */
 864 }
 865
 866 static struct stream_filter_vtbl null_vtbl = {
 867         null_filter_fn,
 868         null_free_fn,
 869 };
 870
 871 static struct stream_filter null_filter_singleton = {
 872         &null_vtbl,
 873 };
 874
 875 int is_null_stream_filter(struct stream_filter *filter)
 876 {
 877         return filter == &null_filter_singleton;
 878 }
 879
 880
 881 /*
 882  * LF-to-CRLF filter
 883  */
 884
 885 struct lf_to_crlf_filter {
 886         struct stream_filter filter;
 887         unsigned has_held:1;
 888         char held;
 889 };
 890
 891 static int lf_to_crlf_filter_fn(struct stream_filter *filter,
 892                                 const char *input, size_t *isize_p,
 893                                 char *output, size_t *osize_p)
 894 {
 895         size_t count, o = 0;
 896         struct lf_to_crlf_filter *lf_to_crlf = (struct lf_to_crlf_filter *)filter;
 897
 898         /*
 899          * We may be holding onto the CR to see if it is followed by a
 900          * LF, in which case we would need to go to the main loop.
 901          * Otherwise, just emit it to the output stream.
 902          */
 903         if (lf_to_crlf->has_held && (lf_to_crlf->held != '\r' || !input)) {
 904                 output[o++] = lf_to_crlf->held;
 905                 lf_to_crlf->has_held = 0;
 906         }
 907
 908         /* We are told to drain */
 909         if (!input) {
 910                 *osize_p -= o;
 911                 return 0;
 912         }
 913
 914         count = *isize_p;
 915         if (count || lf_to_crlf->has_held) {
 916                 size_t i;
 917                 int was_cr = 0;
 918
 919                 if (lf_to_crlf->has_held) {
 920                         was_cr = 1;
 921                         lf_to_crlf->has_held = 0;
 922                 }
 923
 924                 for (i = 0; o < *osize_p && i < count; i++) {
 925                         char ch = input[i];
 926
 927                         if (ch == '\n') {
 928                                 output[o++] = '\r';
 929                         } else if (was_cr) {
 930                                 /*
 931                                  * Previous round saw CR and it is not followed
 932                                  * by a LF; emit the CR before processing the
 933                                  * current character.
 934                                  */
 935                                 output[o++] = '\r';
 936                         }
 937
 938                         /*
 939                          * We may have consumed the last output slot,
 940                          * in which case we need to break out of this
 941                          * loop; hold the current character before
 942                          * returning.
 943                          */
 944                         if (*osize_p <= o) {
 945                                 lf_to_crlf->has_held = 1;
 946                                 lf_to_crlf->held = ch;
 947                                 continue; /* break but increment i */
 948                         }
 949
 950                         if (ch == '\r') {
 951                                 was_cr = 1;
 952                                 continue;
 953                         }
 954
 955                         was_cr = 0;
 956                         output[o++] = ch;
 957                 }
 958
 959                 *osize_p -= o;
 960                 *isize_p -= i;
 961
 962                 if (!lf_to_crlf->has_held && was_cr) {
 963                         lf_to_crlf->has_held = 1;
 964                         lf_to_crlf->held = '\r';
 965                 }
 966         }
 967         return 0;
 968 }
 969
 970 static void lf_to_crlf_free_fn(struct stream_filter *filter)
 971 {
 972         free(filter);
 973 }
 974
 975 static struct stream_filter_vtbl lf_to_crlf_vtbl = {
 976         lf_to_crlf_filter_fn,
 977         lf_to_crlf_free_fn,
 978 };
 979
 980 static struct stream_filter *lf_to_crlf_filter(void)
 981 {
 982         struct lf_to_crlf_filter *lf_to_crlf = xcalloc(1, sizeof(*lf_to_crlf));
 983
 984         lf_to_crlf->filter.vtbl = &lf_to_crlf_vtbl;
 985         return (struct stream_filter *)lf_to_crlf;
 986 }
 987
 988 /*
 989  * Cascade filter
 990  */
 991 #define FILTER_BUFFER 1024
 992 struct cascade_filter {
 993         struct stream_filter filter;
 994         struct stream_filter *one;
 995         struct stream_filter *two;
 996         char buf[FILTER_BUFFER];
 997         int end, ptr;
 998 };
 999
1000 static int cascade_filter_fn(struct stream_filter *filter,
1001                              const char *input, size_t *isize_p,
1002                              char *output, size_t *osize_p)
1003 {
1004         struct cascade_filter *cas = (struct cascade_filter *) filter;
1005         size_t filled = 0;
1006         size_t sz = *osize_p;
1007         size_t to_feed, remaining;
1008
1009         /*
1010          * input -- (one) --> buf -- (two) --> output
1011          */
1012         while (filled < sz) {
1013                 remaining = sz - filled;
1014
1015                 /* do we already have something to feed two with? */
1016                 if (cas->ptr < cas->end) {
1017                         to_feed = cas->end - cas->ptr;
1018                         if (stream_filter(cas->two,
1019                                           cas->buf + cas->ptr, &to_feed,
1020                                           output + filled, &remaining))
1021                                 return -1;
1022                         cas->ptr += (cas->end - cas->ptr) - to_feed;
1023                         filled = sz - remaining;
1024                         continue;
1025                 }
1026
1027                 /* feed one from upstream and have it emit into our buffer */
1028                 to_feed = input ? *isize_p : 0;
1029                 if (input && !to_feed)
1030                         break;
1031                 remaining = sizeof(cas->buf);
1032                 if (stream_filter(cas->one,
1033                                   input, &to_feed,
1034                                   cas->buf, &remaining))
1035                         return -1;
1036                 cas->end = sizeof(cas->buf) - remaining;
1037                 cas->ptr = 0;
1038                 if (input) {
1039                         size_t fed = *isize_p - to_feed;
1040                         *isize_p -= fed;
1041                         input += fed;
1042                 }
1043
1044                 /* do we know that we drained one completely? */
1045                 if (input || cas->end)
1046                         continue;
1047
1048                 /* tell two to drain; we have nothing more to give it */
1049                 to_feed = 0;
1050                 remaining = sz - filled;
1051                 if (stream_filter(cas->two,
1052                                   NULL, &to_feed,
1053                                   output + filled, &remaining))
1054                         return -1;
1055                 if (remaining == (sz - filled))
1056                         break; /* completely drained two */
1057                 filled = sz - remaining;
1058         }
1059         *osize_p -= filled;
1060         return 0;
1061 }
1062
1063 static void cascade_free_fn(struct stream_filter *filter)
1064 {
1065         struct cascade_filter *cas = (struct cascade_filter *)filter;
1066         free_stream_filter(cas->one);
1067         free_stream_filter(cas->two);
1068         free(filter);
1069 }
1070
1071 static struct stream_filter_vtbl cascade_vtbl = {
1072         cascade_filter_fn,
1073         cascade_free_fn,
1074 };
1075
1076 static struct stream_filter *cascade_filter(struct stream_filter *one,
1077                                             struct stream_filter *two)
1078 {
1079         struct cascade_filter *cascade;
1080
1081         if (!one || is_null_stream_filter(one))
1082                 return two;
1083         if (!two || is_null_stream_filter(two))
1084                 return one;
1085
1086         cascade = xmalloc(sizeof(*cascade));
1087         cascade->one = one;
1088         cascade->two = two;
1089         cascade->end = cascade->ptr = 0;
1090         cascade->filter.vtbl = &cascade_vtbl;
1091         return (struct stream_filter *)cascade;
1092 }
1093
1094 /*
1095  * ident filter
1096  */
1097 #define IDENT_DRAINING (-1)
1098 #define IDENT_SKIPPING (-2)
1099 struct ident_filter {
1100         struct stream_filter filter;
1101         struct strbuf left;
1102         int state;
1103         char ident[45]; /* ": x40 $" */
1104 };
1105
1106 static int is_foreign_ident(const char *str)
1107 {
1108         int i;
1109
1110         if (prefixcmp(str, "$Id: "))
1111                 return 0;
1112         for (i = 5; str[i]; i++) {
1113                 if (isspace(str[i]) && str[i+1] != '$')
1114                         return 1;
1115         }
1116         return 0;
1117 }
1118
1119 static void ident_drain(struct ident_filter *ident, char **output_p, size_t *osize_p)
1120 {
1121         size_t to_drain = ident->left.len;
1122
1123         if (*osize_p < to_drain)
1124                 to_drain = *osize_p;
1125         if (to_drain) {
1126                 memcpy(*output_p, ident->left.buf, to_drain);
1127                 strbuf_remove(&ident->left, 0, to_drain);
1128                 *output_p += to_drain;
1129                 *osize_p -= to_drain;
1130         }
1131         if (!ident->left.len)
1132                 ident->state = 0;
1133 }
1134
1135 static int ident_filter_fn(struct stream_filter *filter,
1136                            const char *input, size_t *isize_p,
1137                            char *output, size_t *osize_p)
1138 {
1139         struct ident_filter *ident = (struct ident_filter *)filter;
1140         static const char head[] = "$Id";
1141
1142         if (!input) {
1143                 /* drain upon eof */
1144                 switch (ident->state) {
1145                 default:
1146                         strbuf_add(&ident->left, head, ident->state);
1147                 case IDENT_SKIPPING:
1148                         /* fallthru */
1149                 case IDENT_DRAINING:
1150                         ident_drain(ident, &output, osize_p);
1151                 }
1152                 return 0;
1153         }
1154
1155         while (*isize_p || (ident->state == IDENT_DRAINING)) {
1156                 int ch;
1157
1158                 if (ident->state == IDENT_DRAINING) {
1159                         ident_drain(ident, &output, osize_p);
1160                         if (!*osize_p)
1161                                 break;
1162                         continue;
1163                 }
1164
1165                 ch = *(input++);
1166                 (*isize_p)--;
1167
1168                 if (ident->state == IDENT_SKIPPING) {
1169                         /*
1170                          * Skipping until '$' or LF, but keeping them
1171                          * in case it is a foreign ident.
1172                          */
1173                         strbuf_addch(&ident->left, ch);
1174                         if (ch != '\n' && ch != '$')
1175                                 continue;
1176                         if (ch == '$' && !is_foreign_ident(ident->left.buf)) {
1177                                 strbuf_setlen(&ident->left, sizeof(head) - 1);
1178                                 strbuf_addstr(&ident->left, ident->ident);
1179                         }
1180                         ident->state = IDENT_DRAINING;
1181                         continue;
1182                 }
1183
1184                 if (ident->state < sizeof(head) &&
1185                     head[ident->state] == ch) {
1186                         ident->state++;
1187                         continue;
1188                 }
1189
1190                 if (ident->state)
1191                         strbuf_add(&ident->left, head, ident->state);
1192                 if (ident->state == sizeof(head) - 1) {
1193                         if (ch != ':' && ch != '$') {
1194                                 strbuf_addch(&ident->left, ch);
1195                                 ident->state = 0;
1196                                 continue;
1197                         }
1198
1199                         if (ch == ':') {
1200                                 strbuf_addch(&ident->left, ch);
1201                                 ident->state = IDENT_SKIPPING;
1202                         } else {
1203                                 strbuf_addstr(&ident->left, ident->ident);
1204                                 ident->state = IDENT_DRAINING;
1205                         }
1206                         continue;
1207                 }
1208
1209                 strbuf_addch(&ident->left, ch);
1210                 ident->state = IDENT_DRAINING;
1211         }
1212         return 0;
1213 }
1214
1215 static void ident_free_fn(struct stream_filter *filter)
1216 {
1217         struct ident_filter *ident = (struct ident_filter *)filter;
1218         strbuf_release(&ident->left);
1219         free(filter);
1220 }
1221
1222 static struct stream_filter_vtbl ident_vtbl = {
1223         ident_filter_fn,
1224         ident_free_fn,
1225 };
1226
1227 static struct stream_filter *ident_filter(const unsigned char *sha1)
1228 {
1229         struct ident_filter *ident = xmalloc(sizeof(*ident));
1230
1231         sprintf(ident->ident, ": %s $", sha1_to_hex(sha1));
1232         strbuf_init(&ident->left, 0);
1233         ident->filter.vtbl = &ident_vtbl;
1234         ident->state = 0;
1235         return (struct stream_filter *)ident;
1236 }
1237
1238 /*
1239  * Return an appropriately constructed filter for the path, or NULL if
1240  * the contents cannot be filtered without reading the whole thing
1241  * in-core.
1242  *
1243  * Note that you would be crazy to set CRLF, smuge/clean or ident to a
1244  * large binary blob you would want us not to slurp into the memory!
1245  */
1246 struct stream_filter *get_stream_filter(const char *path, const unsigned char *sha1)
1247 {
1248         struct conv_attrs ca;
1249         enum crlf_action crlf_action;
1250         struct stream_filter *filter = NULL;
1251
1252         convert_attrs(&ca, path);
1253
1254         if (ca.drv && (ca.drv->smudge || ca.drv->clean))
1255                 return filter;
1256
1257         if (ca.ident)
1258                 filter = ident_filter(sha1);
1259
1260         crlf_action = input_crlf_action(ca.crlf_action, ca.eol_attr);
1261
1262         if ((crlf_action == CRLF_BINARY) || (crlf_action == CRLF_INPUT) ||
1263             (crlf_action == CRLF_GUESS && auto_crlf == AUTO_CRLF_FALSE))
1264                 filter = cascade_filter(filter, &null_filter_singleton);
1265
1266         else if (output_eol(crlf_action) == EOL_CRLF &&
1267                  !(crlf_action == CRLF_AUTO || crlf_action == CRLF_GUESS))
1268                 filter = cascade_filter(filter, lf_to_crlf_filter());
1269
1270         return filter;
1271 }
1272
1273 void free_stream_filter(struct stream_filter *filter)
1274 {
1275         filter->vtbl->free(filter);
1276 }
1277
1278 int stream_filter(struct stream_filter *filter,
1279                   const char *input, size_t *isize_p,
1280                   char *output, size_t *osize_p)
1281 {
1282         return filter->vtbl->filter(filter, input, isize_p, output, osize_p);
1283 }