convert.c

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