convert.c

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