git.oblomov.eu Git - git/blob - 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) ||
 200             (src && !len))
 201                 return 0;
 202
 203         /*
 204          * If we are doing a dry-run and have no source buffer, there is
 205          * nothing to analyze; we must assume we would convert.
 206          */
 207         if (!buf && !src)
 208                 return 1;
 209
 210         gather_stats(src, len, &stats);
 211
 212         if (crlf_action == CRLF_AUTO || crlf_action == CRLF_GUESS) {
 213                 /*
 214                  * We're currently not going to even try to convert stuff
 215                  * that has bare CR characters. Does anybody do that crazy
 216                  * stuff?
 217                  */
 218                 if (stats.cr != stats.crlf)
 219                         return 0;
 220
 221                 /*
 222                  * And add some heuristics for binary vs text, of course...
 223                  */
 224                 if (is_binary(len, &stats))
 225                         return 0;
 226
 227                 if (crlf_action == CRLF_GUESS) {
 228                         /*
 229                          * If the file in the index has any CR in it, do not convert.
 230                          * This is the new safer autocrlf handling.
 231                          */
 232                         if (has_cr_in_index(path))
 233                                 return 0;
 234                 }
 235         }
 236
 237         check_safe_crlf(path, crlf_action, &stats, checksafe);
 238
 239         /* Optimization: No CR? Nothing to convert, regardless. */
 240         if (!stats.cr)
 241                 return 0;
 242
 243         /*
 244          * At this point all of our source analysis is done, and we are sure we
 245          * would convert. If we are in dry-run mode, we can give an answer.
 246          */
 247         if (!buf)
 248                 return 1;
 249
 250         /* only grow if not in place */
 251         if (strbuf_avail(buf) + buf->len < len)
 252                 strbuf_grow(buf, len - buf->len);
 253         dst = buf->buf;
 254         if (crlf_action == CRLF_AUTO || crlf_action == CRLF_GUESS) {
 255                 /*
 256                  * If we guessed, we already know we rejected a file with
 257                  * lone CR, and we can strip a CR without looking at what
 258                  * follow it.
 259                  */
 260                 do {
 261                         unsigned char c = *src++;
 262                         if (c != '\r')
 263                                 *dst++ = c;
 264                 } while (--len);
 265         } else {
 266                 do {
 267                         unsigned char c = *src++;
 268                         if (! (c == '\r' && (1 < len && *src == '\n')))
 269                                 *dst++ = c;
 270                 } while (--len);
 271         }
 272         strbuf_setlen(buf, dst - buf->buf);
 273         return 1;
 274 }
 275
 276 static int crlf_to_worktree(const char *path, const char *src, size_t len,
 277                             struct strbuf *buf, enum crlf_action crlf_action)
 278 {
 279         char *to_free = NULL;
 280         struct text_stat stats;
 281
 282         if (!len || output_eol(crlf_action) != EOL_CRLF)
 283                 return 0;
 284
 285         gather_stats(src, len, &stats);
 286
 287         /* No LF? Nothing to convert, regardless. */
 288         if (!stats.lf)
 289                 return 0;
 290
 291         /* Was it already in CRLF format? */
 292         if (stats.lf == stats.crlf)
 293                 return 0;
 294
 295         if (crlf_action == CRLF_AUTO || crlf_action == CRLF_GUESS) {
 296                 if (crlf_action == CRLF_GUESS) {
 297                         /* If we have any CR or CRLF line endings, we do not touch it */
 298                         /* This is the new safer autocrlf-handling */
 299                         if (stats.cr > 0 || stats.crlf > 0)
 300                                 return 0;
 301                 }
 302
 303                 /* If we have any bare CR characters, we're not going to touch it */
 304                 if (stats.cr != stats.crlf)
 305                         return 0;
 306
 307                 if (is_binary(len, &stats))
 308                         return 0;
 309         }
 310
 311         /* are we "faking" in place editing ? */
 312         if (src == buf->buf)
 313                 to_free = strbuf_detach(buf, NULL);
 314
 315         strbuf_grow(buf, len + stats.lf - stats.crlf);
 316         for (;;) {
 317                 const char *nl = memchr(src, '\n', len);
 318                 if (!nl)
 319                         break;
 320                 if (nl > src && nl[-1] == '\r') {
 321                         strbuf_add(buf, src, nl + 1 - src);
 322                 } else {
 323                         strbuf_add(buf, src, nl - src);
 324                         strbuf_addstr(buf, "\r\n");
 325                 }
 326                 len -= nl + 1 - src;
 327                 src  = nl + 1;
 328         }
 329         strbuf_add(buf, src, len);
 330
 331         free(to_free);
 332         return 1;
 333 }
 334
 335 struct filter_params {
 336         const char *src;
 337         unsigned long size;
 338         const char *cmd;
 339         const char *path;
 340 };
 341
 342 static int filter_buffer(int in, int out, void *data)
 343 {
 344         /*
 345          * Spawn cmd and feed the buffer contents through its stdin.
 346          */
 347         struct child_process child_process;
 348         struct filter_params *params = (struct filter_params *)data;
 349         int write_err, status;
 350         const char *argv[] = { NULL, NULL };
 351
 352         /* apply % substitution to cmd */
 353         struct strbuf cmd = STRBUF_INIT;
 354         struct strbuf path = STRBUF_INIT;
 355         struct strbuf_expand_dict_entry dict[] = {
 356                 { "f", NULL, },
 357                 { NULL, NULL, },
 358         };
 359
 360         /* quote the path to preserve spaces, etc. */
 361         sq_quote_buf(&path, params->path);
 362         dict[0].value = path.buf;
 363
 364         /* expand all %f with the quoted path */
 365         strbuf_expand(&cmd, params->cmd, strbuf_expand_dict_cb, &dict);
 366         strbuf_release(&path);
 367
 368         argv[0] = cmd.buf;
 369
 370         memset(&child_process, 0, sizeof(child_process));
 371         child_process.argv = argv;
 372         child_process.use_shell = 1;
 373         child_process.in = -1;
 374         child_process.out = out;
 375
 376         if (start_command(&child_process))
 377                 return error("cannot fork to run external filter %s", params->cmd);
 378
 379         sigchain_push(SIGPIPE, SIG_IGN);
 380
 381         write_err = (write_in_full(child_process.in, params->src, params->size) < 0);
 382         if (close(child_process.in))
 383                 write_err = 1;
 384         if (write_err)
 385                 error("cannot feed the input to external filter %s", params->cmd);
 386
 387         sigchain_pop(SIGPIPE);
 388
 389         status = finish_command(&child_process);
 390         if (status)
 391                 error("external filter %s failed %d", params->cmd, status);
 392
 393         strbuf_release(&cmd);
 394         return (write_err || status);
 395 }
 396
 397 static int apply_filter(const char *path, const char *src, size_t len,
 398                         struct strbuf *dst, const char *cmd)
 399 {
 400         /*
 401          * Create a pipeline to have the command filter the buffer's
 402          * contents.
 403          *
 404          * (child --> cmd) --> us
 405          */
 406         int ret = 1;
 407         struct strbuf nbuf = STRBUF_INIT;
 408         struct async async;
 409         struct filter_params params;
 410
 411         if (!cmd)
 412                 return 0;
 413
 414         if (!dst)
 415                 return 1;
 416
 417         memset(&async, 0, sizeof(async));
 418         async.proc = filter_buffer;
 419         async.data = &params;
 420         async.out = -1;
 421         params.src = src;
 422         params.size = len;
 423         params.cmd = cmd;
 424         params.path = path;
 425
 426         fflush(NULL);
 427         if (start_async(&async))
 428                 return 0;       /* error was already reported */
 429
 430         if (strbuf_read(&nbuf, async.out, len) < 0) {
 431                 error("read from external filter %s failed", cmd);
 432                 ret = 0;
 433         }
 434         if (close(async.out)) {
 435                 error("read from external filter %s failed", cmd);
 436                 ret = 0;
 437         }
 438         if (finish_async(&async)) {
 439                 error("external filter %s failed", cmd);
 440                 ret = 0;
 441         }
 442
 443         if (ret) {
 444                 strbuf_swap(dst, &nbuf);
 445         }
 446         strbuf_release(&nbuf);
 447         return ret;
 448 }
 449
 450 static struct convert_driver {
 451         const char *name;
 452         struct convert_driver *next;
 453         const char *smudge;
 454         const char *clean;
 455 } *user_convert, **user_convert_tail;
 456
 457 static int read_convert_config(const char *var, const char *value, void *cb)
 458 {
 459         const char *ep, *name;
 460         int namelen;
 461         struct convert_driver *drv;
 462
 463         /*
 464          * External conversion drivers are configured using
 465          * "filter.<name>.variable".
 466          */
 467         if (prefixcmp(var, "filter.") || (ep = strrchr(var, '.')) == var + 6)
 468                 return 0;
 469         name = var + 7;
 470         namelen = ep - name;
 471         for (drv = user_convert; drv; drv = drv->next)
 472                 if (!strncmp(drv->name, name, namelen) && !drv->name[namelen])
 473                         break;
 474         if (!drv) {
 475                 drv = xcalloc(1, sizeof(struct convert_driver));
 476                 drv->name = xmemdupz(name, namelen);
 477                 *user_convert_tail = drv;
 478                 user_convert_tail = &(drv->next);
 479         }
 480
 481         ep++;
 482
 483         /*
 484          * filter.<name>.smudge and filter.<name>.clean specifies
 485          * the command line:
 486          *
 487          *      command-line
 488          *
 489          * The command-line will not be interpolated in any way.
 490          */
 491
 492         if (!strcmp("smudge", ep))
 493                 return git_config_string(&drv->smudge, var, value);
 494
 495         if (!strcmp("clean", ep))
 496                 return git_config_string(&drv->clean, var, value);
 497
 498         return 0;
 499 }
 500
 501 static int count_ident(const char *cp, unsigned long size)
 502 {
 503         /*
 504          * "$Id: 0000000000000000000000000000000000000000 $" <=> "$Id$"
 505          */
 506         int cnt = 0;
 507         char ch;
 508
 509         while (size) {
 510                 ch = *cp++;
 511                 size--;
 512                 if (ch != '$')
 513                         continue;
 514                 if (size < 3)
 515                         break;
 516                 if (memcmp("Id", cp, 2))
 517                         continue;
 518                 ch = cp[2];
 519                 cp += 3;
 520                 size -= 3;
 521                 if (ch == '$')
 522                         cnt++; /* $Id$ */
 523                 if (ch != ':')
 524                         continue;
 525
 526                 /*
 527                  * "$Id: ... "; scan up to the closing dollar sign and discard.
 528                  */
 529                 while (size) {
 530                         ch = *cp++;
 531                         size--;
 532                         if (ch == '$') {
 533                                 cnt++;
 534                                 break;
 535                         }
 536                         if (ch == '\n')
 537                                 break;
 538                 }
 539         }
 540         return cnt;
 541 }
 542
 543 static int ident_to_git(const char *path, const char *src, size_t len,
 544                         struct strbuf *buf, int ident)
 545 {
 546         char *dst, *dollar;
 547
 548         if (!ident || (src && !count_ident(src, len)))
 549                 return 0;
 550
 551         if (!buf)
 552                 return 1;
 553
 554         /* only grow if not in place */
 555         if (strbuf_avail(buf) + buf->len < len)
 556                 strbuf_grow(buf, len - buf->len);
 557         dst = buf->buf;
 558         for (;;) {
 559                 dollar = memchr(src, '$', len);
 560                 if (!dollar)
 561                         break;
 562                 memmove(dst, src, dollar + 1 - src);
 563                 dst += dollar + 1 - src;
 564                 len -= dollar + 1 - src;
 565                 src  = dollar + 1;
 566
 567                 if (len > 3 && !memcmp(src, "Id:", 3)) {
 568                         dollar = memchr(src + 3, '$', len - 3);
 569                         if (!dollar)
 570                                 break;
 571                         if (memchr(src + 3, '\n', dollar - src - 3)) {
 572                                 /* Line break before the next dollar. */
 573                                 continue;
 574                         }
 575
 576                         memcpy(dst, "Id$", 3);
 577                         dst += 3;
 578                         len -= dollar + 1 - src;
 579                         src  = dollar + 1;
 580                 }
 581         }
 582         memmove(dst, src, len);
 583         strbuf_setlen(buf, dst + len - buf->buf);
 584         return 1;
 585 }
 586
 587 static int ident_to_worktree(const char *path, const char *src, size_t len,
 588                              struct strbuf *buf, int ident)
 589 {
 590         unsigned char sha1[20];
 591         char *to_free = NULL, *dollar, *spc;
 592         int cnt;
 593
 594         if (!ident)
 595                 return 0;
 596
 597         cnt = count_ident(src, len);
 598         if (!cnt)
 599                 return 0;
 600
 601         /* are we "faking" in place editing ? */
 602         if (src == buf->buf)
 603                 to_free = strbuf_detach(buf, NULL);
 604         hash_sha1_file(src, len, "blob", sha1);
 605
 606         strbuf_grow(buf, len + cnt * 43);
 607         for (;;) {
 608                 /* step 1: run to the next '$' */
 609                 dollar = memchr(src, '$', len);
 610                 if (!dollar)
 611                         break;
 612                 strbuf_add(buf, src, dollar + 1 - src);
 613                 len -= dollar + 1 - src;
 614                 src  = dollar + 1;
 615
 616                 /* step 2: does it looks like a bit like Id:xxx$ or Id$ ? */
 617                 if (len < 3 || memcmp("Id", src, 2))
 618                         continue;
 619
 620                 /* step 3: skip over Id$ or Id:xxxxx$ */
 621                 if (src[2] == '$') {
 622                         src += 3;
 623                         len -= 3;
 624                 } else if (src[2] == ':') {
 625                         /*
 626                          * It's possible that an expanded Id has crept its way into the
 627                          * repository, we cope with that by stripping the expansion out.
 628                          * This is probably not a good idea, since it will cause changes
 629                          * on checkout, which won't go away by stash, but let's keep it
 630                          * for git-style ids.
 631                          */
 632                         dollar = memchr(src + 3, '$', len - 3);
 633                         if (!dollar) {
 634                                 /* incomplete keyword, no more '$', so just quit the loop */
 635                                 break;
 636                         }
 637
 638                         if (memchr(src + 3, '\n', dollar - src - 3)) {
 639                                 /* Line break before the next dollar. */
 640                                 continue;
 641                         }
 642
 643                         spc = memchr(src + 4, ' ', dollar - src - 4);
 644                         if (spc && spc < dollar-1) {
 645                                 /* There are spaces in unexpected places.
 646                                  * This is probably an id from some other
 647                                  * versioning system. Keep it for now.
 648                                  */
 649                                 continue;
 650                         }
 651
 652                         len -= dollar + 1 - src;
 653                         src  = dollar + 1;
 654                 } else {
 655                         /* it wasn't a "Id$" or "Id:xxxx$" */
 656                         continue;
 657                 }
 658
 659                 /* step 4: substitute */
 660                 strbuf_addstr(buf, "Id: ");
 661                 strbuf_add(buf, sha1_to_hex(sha1), 40);
 662                 strbuf_addstr(buf, " $");
 663         }
 664         strbuf_add(buf, src, len);
 665
 666         free(to_free);
 667         return 1;
 668 }
 669
 670 static enum crlf_action git_path_check_crlf(const char *path, struct git_attr_check *check)
 671 {
 672         const char *value = check->value;
 673
 674         if (ATTR_TRUE(value))
 675                 return CRLF_TEXT;
 676         else if (ATTR_FALSE(value))
 677                 return CRLF_BINARY;
 678         else if (ATTR_UNSET(value))
 679                 ;
 680         else if (!strcmp(value, "input"))
 681                 return CRLF_INPUT;
 682         else if (!strcmp(value, "auto"))
 683                 return CRLF_AUTO;
 684         return CRLF_GUESS;
 685 }
 686
 687 static enum eol git_path_check_eol(const char *path, struct git_attr_check *check)
 688 {
 689         const char *value = check->value;
 690
 691         if (ATTR_UNSET(value))
 692                 ;
 693         else if (!strcmp(value, "lf"))
 694                 return EOL_LF;
 695         else if (!strcmp(value, "crlf"))
 696                 return EOL_CRLF;
 697         return EOL_UNSET;
 698 }
 699
 700 static struct convert_driver *git_path_check_convert(const char *path,
 701                                              struct git_attr_check *check)
 702 {
 703         const char *value = check->value;
 704         struct convert_driver *drv;
 705
 706         if (ATTR_TRUE(value) || ATTR_FALSE(value) || ATTR_UNSET(value))
 707                 return NULL;
 708         for (drv = user_convert; drv; drv = drv->next)
 709                 if (!strcmp(value, drv->name))
 710                         return drv;
 711         return NULL;
 712 }
 713
 714 static int git_path_check_ident(const char *path, struct git_attr_check *check)
 715 {
 716         const char *value = check->value;
 717
 718         return !!ATTR_TRUE(value);
 719 }
 720
 721 static enum crlf_action input_crlf_action(enum crlf_action text_attr, enum eol eol_attr)
 722 {
 723         if (text_attr == CRLF_BINARY)
 724                 return CRLF_BINARY;
 725         if (eol_attr == EOL_LF)
 726                 return CRLF_INPUT;
 727         if (eol_attr == EOL_CRLF)
 728                 return CRLF_CRLF;
 729         return text_attr;
 730 }
 731
 732 struct conv_attrs {
 733         struct convert_driver *drv;
 734         enum crlf_action crlf_action;
 735         enum eol eol_attr;
 736         int ident;
 737 };
 738
 739 static const char *conv_attr_name[] = {
 740         "crlf", "ident", "filter", "eol", "text",
 741 };
 742 #define NUM_CONV_ATTRS ARRAY_SIZE(conv_attr_name)
 743
 744 static void convert_attrs(struct conv_attrs *ca, const char *path)
 745 {
 746         int i;
 747         static struct git_attr_check ccheck[NUM_CONV_ATTRS];
 748
 749         if (!ccheck[0].attr) {
 750                 for (i = 0; i < NUM_CONV_ATTRS; i++)
 751                         ccheck[i].attr = git_attr(conv_attr_name[i]);
 752                 user_convert_tail = &user_convert;
 753                 git_config(read_convert_config, NULL);
 754         }
 755
 756         if (!git_check_attr(path, NUM_CONV_ATTRS, ccheck)) {
 757                 ca->crlf_action = git_path_check_crlf(path, ccheck + 4);
 758                 if (ca->crlf_action == CRLF_GUESS)
 759                         ca->crlf_action = git_path_check_crlf(path, ccheck + 0);
 760                 ca->ident = git_path_check_ident(path, ccheck + 1);
 761                 ca->drv = git_path_check_convert(path, ccheck + 2);
 762                 ca->eol_attr = git_path_check_eol(path, ccheck + 3);
 763         } else {
 764                 ca->drv = NULL;
 765                 ca->crlf_action = CRLF_GUESS;
 766                 ca->eol_attr = EOL_UNSET;
 767                 ca->ident = 0;
 768         }
 769 }
 770
 771 int convert_to_git(const char *path, const char *src, size_t len,
 772                    struct strbuf *dst, enum safe_crlf checksafe)
 773 {
 774         int ret = 0;
 775         const char *filter = NULL;
 776         struct conv_attrs ca;
 777
 778         convert_attrs(&ca, path);
 779         if (ca.drv)
 780                 filter = ca.drv->clean;
 781
 782         ret |= apply_filter(path, src, len, dst, filter);
 783         if (ret && dst) {
 784                 src = dst->buf;
 785                 len = dst->len;
 786         }
 787         ca.crlf_action = input_crlf_action(ca.crlf_action, ca.eol_attr);
 788         ret |= crlf_to_git(path, src, len, dst, ca.crlf_action, checksafe);
 789         if (ret && dst) {
 790                 src = dst->buf;
 791                 len = dst->len;
 792         }
 793         return ret | ident_to_git(path, src, len, dst, ca.ident);
 794 }
 795
 796 static int convert_to_working_tree_internal(const char *path, const char *src,
 797                                             size_t len, struct strbuf *dst,
 798                                             int normalizing)
 799 {
 800         int ret = 0;
 801         const char *filter = NULL;
 802         struct conv_attrs ca;
 803
 804         convert_attrs(&ca, path);
 805         if (ca.drv)
 806                 filter = ca.drv->smudge;
 807
 808         ret |= ident_to_worktree(path, src, len, dst, ca.ident);
 809         if (ret) {
 810                 src = dst->buf;
 811                 len = dst->len;
 812         }
 813         /*
 814          * CRLF conversion can be skipped if normalizing, unless there
 815          * is a smudge filter.  The filter might expect CRLFs.
 816          */
 817         if (filter || !normalizing) {
 818                 ca.crlf_action = input_crlf_action(ca.crlf_action, ca.eol_attr);
 819                 ret |= crlf_to_worktree(path, src, len, dst, ca.crlf_action);
 820                 if (ret) {
 821                         src = dst->buf;
 822                         len = dst->len;
 823                 }
 824         }
 825         return ret | apply_filter(path, src, len, dst, filter);
 826 }
 827
 828 int convert_to_working_tree(const char *path, const char *src, size_t len, struct strbuf *dst)
 829 {
 830         return convert_to_working_tree_internal(path, src, len, dst, 0);
 831 }
 832
 833 int renormalize_buffer(const char *path, const char *src, size_t len, struct strbuf *dst)
 834 {
 835         int ret = convert_to_working_tree_internal(path, src, len, dst, 1);
 836         if (ret) {
 837                 src = dst->buf;
 838                 len = dst->len;
 839         }
 840         return ret | convert_to_git(path, src, len, dst, SAFE_CRLF_FALSE);
 841 }
 842
 843 /*****************************************************************
 844  *
 845  * Streaming converison support
 846  *
 847  *****************************************************************/
 848
 849 typedef int (*filter_fn)(struct stream_filter *,
 850                          const char *input, size_t *isize_p,
 851                          char *output, size_t *osize_p);
 852 typedef void (*free_fn)(struct stream_filter *);
 853
 854 struct stream_filter_vtbl {
 855         filter_fn filter;
 856         free_fn free;
 857 };
 858
 859 struct stream_filter {
 860         struct stream_filter_vtbl *vtbl;
 861 };
 862
 863 static int null_filter_fn(struct stream_filter *filter,
 864                           const char *input, size_t *isize_p,
 865                           char *output, size_t *osize_p)
 866 {
 867         size_t count;
 868
 869         if (!input)
 870                 return 0; /* we do not keep any states */
 871         count = *isize_p;
 872         if (*osize_p < count)
 873                 count = *osize_p;
 874         if (count) {
 875                 memmove(output, input, count);
 876                 *isize_p -= count;
 877                 *osize_p -= count;
 878         }
 879         return 0;
 880 }
 881
 882 static void null_free_fn(struct stream_filter *filter)
 883 {
 884         ; /* nothing -- null instances are shared */
 885 }
 886
 887 static struct stream_filter_vtbl null_vtbl = {
 888         null_filter_fn,
 889         null_free_fn,
 890 };
 891
 892 static struct stream_filter null_filter_singleton = {
 893         &null_vtbl,
 894 };
 895
 896 int is_null_stream_filter(struct stream_filter *filter)
 897 {
 898         return filter == &null_filter_singleton;
 899 }
 900
 901
 902 /*
 903  * LF-to-CRLF filter
 904  */
 905
 906 struct lf_to_crlf_filter {
 907         struct stream_filter filter;
 908         unsigned has_held:1;
 909         char held;
 910 };
 911
 912 static int lf_to_crlf_filter_fn(struct stream_filter *filter,
 913                                 const char *input, size_t *isize_p,
 914                                 char *output, size_t *osize_p)
 915 {
 916         size_t count, o = 0;
 917         struct lf_to_crlf_filter *lf_to_crlf = (struct lf_to_crlf_filter *)filter;
 918
 919         /*
 920          * We may be holding onto the CR to see if it is followed by a
 921          * LF, in which case we would need to go to the main loop.
 922          * Otherwise, just emit it to the output stream.
 923          */
 924         if (lf_to_crlf->has_held && (lf_to_crlf->held != '\r' || !input)) {
 925                 output[o++] = lf_to_crlf->held;
 926                 lf_to_crlf->has_held = 0;
 927         }
 928
 929         /* We are told to drain */
 930         if (!input) {
 931                 *osize_p -= o;
 932                 return 0;
 933         }
 934
 935         count = *isize_p;
 936         if (count || lf_to_crlf->has_held) {
 937                 size_t i;
 938                 int was_cr = 0;
 939
 940                 if (lf_to_crlf->has_held) {
 941                         was_cr = 1;
 942                         lf_to_crlf->has_held = 0;
 943                 }
 944
 945                 for (i = 0; o < *osize_p && i < count; i++) {
 946                         char ch = input[i];
 947
 948                         if (ch == '\n') {
 949                                 output[o++] = '\r';
 950                         } else if (was_cr) {
 951                                 /*
 952                                  * Previous round saw CR and it is not followed
 953                                  * by a LF; emit the CR before processing the
 954                                  * current character.
 955                                  */
 956                                 output[o++] = '\r';
 957                         }
 958
 959                         /*
 960                          * We may have consumed the last output slot,
 961                          * in which case we need to break out of this
 962                          * loop; hold the current character before
 963                          * returning.
 964                          */
 965                         if (*osize_p <= o) {
 966                                 lf_to_crlf->has_held = 1;
 967                                 lf_to_crlf->held = ch;
 968                                 continue; /* break but increment i */
 969                         }
 970
 971                         if (ch == '\r') {
 972                                 was_cr = 1;
 973                                 continue;
 974                         }
 975
 976                         was_cr = 0;
 977                         output[o++] = ch;
 978                 }
 979
 980                 *osize_p -= o;
 981                 *isize_p -= i;
 982
 983                 if (!lf_to_crlf->has_held && was_cr) {
 984                         lf_to_crlf->has_held = 1;
 985                         lf_to_crlf->held = '\r';
 986                 }
 987         }
 988         return 0;
 989 }
 990
 991 static void lf_to_crlf_free_fn(struct stream_filter *filter)
 992 {
 993         free(filter);
 994 }
 995
 996 static struct stream_filter_vtbl lf_to_crlf_vtbl = {
 997         lf_to_crlf_filter_fn,
 998         lf_to_crlf_free_fn,
 999 };
1000
1001 static struct stream_filter *lf_to_crlf_filter(void)
1002 {
1003         struct lf_to_crlf_filter *lf_to_crlf = xcalloc(1, sizeof(*lf_to_crlf));
1004
1005         lf_to_crlf->filter.vtbl = &lf_to_crlf_vtbl;
1006         return (struct stream_filter *)lf_to_crlf;
1007 }
1008
1009 /*
1010  * Cascade filter
1011  */
1012 #define FILTER_BUFFER 1024
1013 struct cascade_filter {
1014         struct stream_filter filter;
1015         struct stream_filter *one;
1016         struct stream_filter *two;
1017         char buf[FILTER_BUFFER];
1018         int end, ptr;
1019 };
1020
1021 static int cascade_filter_fn(struct stream_filter *filter,
1022                              const char *input, size_t *isize_p,
1023                              char *output, size_t *osize_p)
1024 {
1025         struct cascade_filter *cas = (struct cascade_filter *) filter;
1026         size_t filled = 0;
1027         size_t sz = *osize_p;
1028         size_t to_feed, remaining;
1029
1030         /*
1031          * input -- (one) --> buf -- (two) --> output
1032          */
1033         while (filled < sz) {
1034                 remaining = sz - filled;
1035
1036                 /* do we already have something to feed two with? */
1037                 if (cas->ptr < cas->end) {
1038                         to_feed = cas->end - cas->ptr;
1039                         if (stream_filter(cas->two,
1040                                           cas->buf + cas->ptr, &to_feed,
1041                                           output + filled, &remaining))
1042                                 return -1;
1043                         cas->ptr += (cas->end - cas->ptr) - to_feed;
1044                         filled = sz - remaining;
1045                         continue;
1046                 }
1047
1048                 /* feed one from upstream and have it emit into our buffer */
1049                 to_feed = input ? *isize_p : 0;
1050                 if (input && !to_feed)
1051                         break;
1052                 remaining = sizeof(cas->buf);
1053                 if (stream_filter(cas->one,
1054                                   input, &to_feed,
1055                                   cas->buf, &remaining))
1056                         return -1;
1057                 cas->end = sizeof(cas->buf) - remaining;
1058                 cas->ptr = 0;
1059                 if (input) {
1060                         size_t fed = *isize_p - to_feed;
1061                         *isize_p -= fed;
1062                         input += fed;
1063                 }
1064
1065                 /* do we know that we drained one completely? */
1066                 if (input || cas->end)
1067                         continue;
1068
1069                 /* tell two to drain; we have nothing more to give it */
1070                 to_feed = 0;
1071                 remaining = sz - filled;
1072                 if (stream_filter(cas->two,
1073                                   NULL, &to_feed,
1074                                   output + filled, &remaining))
1075                         return -1;
1076                 if (remaining == (sz - filled))
1077                         break; /* completely drained two */
1078                 filled = sz - remaining;
1079         }
1080         *osize_p -= filled;
1081         return 0;
1082 }
1083
1084 static void cascade_free_fn(struct stream_filter *filter)
1085 {
1086         struct cascade_filter *cas = (struct cascade_filter *)filter;
1087         free_stream_filter(cas->one);
1088         free_stream_filter(cas->two);
1089         free(filter);
1090 }
1091
1092 static struct stream_filter_vtbl cascade_vtbl = {
1093         cascade_filter_fn,
1094         cascade_free_fn,
1095 };
1096
1097 static struct stream_filter *cascade_filter(struct stream_filter *one,
1098                                             struct stream_filter *two)
1099 {
1100         struct cascade_filter *cascade;
1101
1102         if (!one || is_null_stream_filter(one))
1103                 return two;
1104         if (!two || is_null_stream_filter(two))
1105                 return one;
1106
1107         cascade = xmalloc(sizeof(*cascade));
1108         cascade->one = one;
1109         cascade->two = two;
1110         cascade->end = cascade->ptr = 0;
1111         cascade->filter.vtbl = &cascade_vtbl;
1112         return (struct stream_filter *)cascade;
1113 }
1114
1115 /*
1116  * ident filter
1117  */
1118 #define IDENT_DRAINING (-1)
1119 #define IDENT_SKIPPING (-2)
1120 struct ident_filter {
1121         struct stream_filter filter;
1122         struct strbuf left;
1123         int state;
1124         char ident[45]; /* ": x40 $" */
1125 };
1126
1127 static int is_foreign_ident(const char *str)
1128 {
1129         int i;
1130
1131         if (prefixcmp(str, "$Id: "))
1132                 return 0;
1133         for (i = 5; str[i]; i++) {
1134                 if (isspace(str[i]) && str[i+1] != '$')
1135                         return 1;
1136         }
1137         return 0;
1138 }
1139
1140 static void ident_drain(struct ident_filter *ident, char **output_p, size_t *osize_p)
1141 {
1142         size_t to_drain = ident->left.len;
1143
1144         if (*osize_p < to_drain)
1145                 to_drain = *osize_p;
1146         if (to_drain) {
1147                 memcpy(*output_p, ident->left.buf, to_drain);
1148                 strbuf_remove(&ident->left, 0, to_drain);
1149                 *output_p += to_drain;
1150                 *osize_p -= to_drain;
1151         }
1152         if (!ident->left.len)
1153                 ident->state = 0;
1154 }
1155
1156 static int ident_filter_fn(struct stream_filter *filter,
1157                            const char *input, size_t *isize_p,
1158                            char *output, size_t *osize_p)
1159 {
1160         struct ident_filter *ident = (struct ident_filter *)filter;
1161         static const char head[] = "$Id";
1162
1163         if (!input) {
1164                 /* drain upon eof */
1165                 switch (ident->state) {
1166                 default:
1167                         strbuf_add(&ident->left, head, ident->state);
1168                 case IDENT_SKIPPING:
1169                         /* fallthru */
1170                 case IDENT_DRAINING:
1171                         ident_drain(ident, &output, osize_p);
1172                 }
1173                 return 0;
1174         }
1175
1176         while (*isize_p || (ident->state == IDENT_DRAINING)) {
1177                 int ch;
1178
1179                 if (ident->state == IDENT_DRAINING) {
1180                         ident_drain(ident, &output, osize_p);
1181                         if (!*osize_p)
1182                                 break;
1183                         continue;
1184                 }
1185
1186                 ch = *(input++);
1187                 (*isize_p)--;
1188
1189                 if (ident->state == IDENT_SKIPPING) {
1190                         /*
1191                          * Skipping until '$' or LF, but keeping them
1192                          * in case it is a foreign ident.
1193                          */
1194                         strbuf_addch(&ident->left, ch);
1195                         if (ch != '\n' && ch != '$')
1196                                 continue;
1197                         if (ch == '$' && !is_foreign_ident(ident->left.buf)) {
1198                                 strbuf_setlen(&ident->left, sizeof(head) - 1);
1199                                 strbuf_addstr(&ident->left, ident->ident);
1200                         }
1201                         ident->state = IDENT_DRAINING;
1202                         continue;
1203                 }
1204
1205                 if (ident->state < sizeof(head) &&
1206                     head[ident->state] == ch) {
1207                         ident->state++;
1208                         continue;
1209                 }
1210
1211                 if (ident->state)
1212                         strbuf_add(&ident->left, head, ident->state);
1213                 if (ident->state == sizeof(head) - 1) {
1214                         if (ch != ':' && ch != '$') {
1215                                 strbuf_addch(&ident->left, ch);
1216                                 ident->state = 0;
1217                                 continue;
1218                         }
1219
1220                         if (ch == ':') {
1221                                 strbuf_addch(&ident->left, ch);
1222                                 ident->state = IDENT_SKIPPING;
1223                         } else {
1224                                 strbuf_addstr(&ident->left, ident->ident);
1225                                 ident->state = IDENT_DRAINING;
1226                         }
1227                         continue;
1228                 }
1229
1230                 strbuf_addch(&ident->left, ch);
1231                 ident->state = IDENT_DRAINING;
1232         }
1233         return 0;
1234 }
1235
1236 static void ident_free_fn(struct stream_filter *filter)
1237 {
1238         struct ident_filter *ident = (struct ident_filter *)filter;
1239         strbuf_release(&ident->left);
1240         free(filter);
1241 }
1242
1243 static struct stream_filter_vtbl ident_vtbl = {
1244         ident_filter_fn,
1245         ident_free_fn,
1246 };
1247
1248 static struct stream_filter *ident_filter(const unsigned char *sha1)
1249 {
1250         struct ident_filter *ident = xmalloc(sizeof(*ident));
1251
1252         sprintf(ident->ident, ": %s $", sha1_to_hex(sha1));
1253         strbuf_init(&ident->left, 0);
1254         ident->filter.vtbl = &ident_vtbl;
1255         ident->state = 0;
1256         return (struct stream_filter *)ident;
1257 }
1258
1259 /*
1260  * Return an appropriately constructed filter for the path, or NULL if
1261  * the contents cannot be filtered without reading the whole thing
1262  * in-core.
1263  *
1264  * Note that you would be crazy to set CRLF, smuge/clean or ident to a
1265  * large binary blob you would want us not to slurp into the memory!
1266  */
1267 struct stream_filter *get_stream_filter(const char *path, const unsigned char *sha1)
1268 {
1269         struct conv_attrs ca;
1270         enum crlf_action crlf_action;
1271         struct stream_filter *filter = NULL;
1272
1273         convert_attrs(&ca, path);
1274
1275         if (ca.drv && (ca.drv->smudge || ca.drv->clean))
1276                 return filter;
1277
1278         if (ca.ident)
1279                 filter = ident_filter(sha1);
1280
1281         crlf_action = input_crlf_action(ca.crlf_action, ca.eol_attr);
1282
1283         if ((crlf_action == CRLF_BINARY) || (crlf_action == CRLF_INPUT) ||
1284             (crlf_action == CRLF_GUESS && auto_crlf == AUTO_CRLF_FALSE))
1285                 filter = cascade_filter(filter, &null_filter_singleton);
1286
1287         else if (output_eol(crlf_action) == EOL_CRLF &&
1288                  !(crlf_action == CRLF_AUTO || crlf_action == CRLF_GUESS))
1289                 filter = cascade_filter(filter, lf_to_crlf_filter());
1290
1291         return filter;
1292 }
1293
1294 void free_stream_filter(struct stream_filter *filter)
1295 {
1296         filter->vtbl->free(filter);
1297 }
1298
1299 int stream_filter(struct stream_filter *filter,
1300                   const char *input, size_t *isize_p,
1301                   char *output, size_t *osize_p)
1302 {
1303         return filter->vtbl->filter(filter, input, isize_p, output, osize_p);
1304 }