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 } *user_convert, **user_convert_tail;
 433
 434 static int read_convert_config(const char *var, const char *value, void *cb)
 435 {
 436         const char *ep, *name;
 437         int namelen;
 438         struct convert_driver *drv;
 439
 440         /*
 441          * External conversion drivers are configured using
 442          * "filter.<name>.variable".
 443          */
 444         if (prefixcmp(var, "filter.") || (ep = strrchr(var, '.')) == var + 6)
 445                 return 0;
 446         name = var + 7;
 447         namelen = ep - name;
 448         for (drv = user_convert; drv; drv = drv->next)
 449                 if (!strncmp(drv->name, name, namelen) && !drv->name[namelen])
 450                         break;
 451         if (!drv) {
 452                 drv = xcalloc(1, sizeof(struct convert_driver));
 453                 drv->name = xmemdupz(name, namelen);
 454                 *user_convert_tail = drv;
 455                 user_convert_tail = &(drv->next);
 456         }
 457
 458         ep++;
 459
 460         /*
 461          * filter.<name>.smudge and filter.<name>.clean specifies
 462          * the command line:
 463          *
 464          *      command-line
 465          *
 466          * The command-line will not be interpolated in any way.
 467          */
 468
 469         if (!strcmp("smudge", ep))
 470                 return git_config_string(&drv->smudge, var, value);
 471
 472         if (!strcmp("clean", ep))
 473                 return git_config_string(&drv->clean, var, value);
 474
 475         return 0;
 476 }
 477
 478 static int count_ident(const char *cp, unsigned long size)
 479 {
 480         /*
 481          * "$Id: 0000000000000000000000000000000000000000 $" <=> "$Id$"
 482          */
 483         int cnt = 0;
 484         char ch;
 485
 486         while (size) {
 487                 ch = *cp++;
 488                 size--;
 489                 if (ch != '$')
 490                         continue;
 491                 if (size < 3)
 492                         break;
 493                 if (memcmp("Id", cp, 2))
 494                         continue;
 495                 ch = cp[2];
 496                 cp += 3;
 497                 size -= 3;
 498                 if (ch == '$')
 499                         cnt++; /* $Id$ */
 500                 if (ch != ':')
 501                         continue;
 502
 503                 /*
 504                  * "$Id: ... "; scan up to the closing dollar sign and discard.
 505                  */
 506                 while (size) {
 507                         ch = *cp++;
 508                         size--;
 509                         if (ch == '$') {
 510                                 cnt++;
 511                                 break;
 512                         }
 513                         if (ch == '\n')
 514                                 break;
 515                 }
 516         }
 517         return cnt;
 518 }
 519
 520 static int ident_to_git(const char *path, const char *src, size_t len,
 521                         struct strbuf *buf, int ident)
 522 {
 523         char *dst, *dollar;
 524
 525         if (!ident || !count_ident(src, len))
 526                 return 0;
 527
 528         /* only grow if not in place */
 529         if (strbuf_avail(buf) + buf->len < len)
 530                 strbuf_grow(buf, len - buf->len);
 531         dst = buf->buf;
 532         for (;;) {
 533                 dollar = memchr(src, '$', len);
 534                 if (!dollar)
 535                         break;
 536                 memmove(dst, src, dollar + 1 - src);
 537                 dst += dollar + 1 - src;
 538                 len -= dollar + 1 - src;
 539                 src  = dollar + 1;
 540
 541                 if (len > 3 && !memcmp(src, "Id:", 3)) {
 542                         dollar = memchr(src + 3, '$', len - 3);
 543                         if (!dollar)
 544                                 break;
 545                         if (memchr(src + 3, '\n', dollar - src - 3)) {
 546                                 /* Line break before the next dollar. */
 547                                 continue;
 548                         }
 549
 550                         memcpy(dst, "Id$", 3);
 551                         dst += 3;
 552                         len -= dollar + 1 - src;
 553                         src  = dollar + 1;
 554                 }
 555         }
 556         memmove(dst, src, len);
 557         strbuf_setlen(buf, dst + len - buf->buf);
 558         return 1;
 559 }
 560
 561 static int ident_to_worktree(const char *path, const char *src, size_t len,
 562                              struct strbuf *buf, int ident)
 563 {
 564         unsigned char sha1[20];
 565         char *to_free = NULL, *dollar, *spc;
 566         int cnt;
 567
 568         if (!ident)
 569                 return 0;
 570
 571         cnt = count_ident(src, len);
 572         if (!cnt)
 573                 return 0;
 574
 575         /* are we "faking" in place editing ? */
 576         if (src == buf->buf)
 577                 to_free = strbuf_detach(buf, NULL);
 578         hash_sha1_file(src, len, "blob", sha1);
 579
 580         strbuf_grow(buf, len + cnt * 43);
 581         for (;;) {
 582                 /* step 1: run to the next '$' */
 583                 dollar = memchr(src, '$', len);
 584                 if (!dollar)
 585                         break;
 586                 strbuf_add(buf, src, dollar + 1 - src);
 587                 len -= dollar + 1 - src;
 588                 src  = dollar + 1;
 589
 590                 /* step 2: does it looks like a bit like Id:xxx$ or Id$ ? */
 591                 if (len < 3 || memcmp("Id", src, 2))
 592                         continue;
 593
 594                 /* step 3: skip over Id$ or Id:xxxxx$ */
 595                 if (src[2] == '$') {
 596                         src += 3;
 597                         len -= 3;
 598                 } else if (src[2] == ':') {
 599                         /*
 600                          * It's possible that an expanded Id has crept its way into the
 601                          * repository, we cope with that by stripping the expansion out.
 602                          * This is probably not a good idea, since it will cause changes
 603                          * on checkout, which won't go away by stash, but let's keep it
 604                          * for git-style ids.
 605                          */
 606                         dollar = memchr(src + 3, '$', len - 3);
 607                         if (!dollar) {
 608                                 /* incomplete keyword, no more '$', so just quit the loop */
 609                                 break;
 610                         }
 611
 612                         if (memchr(src + 3, '\n', dollar - src - 3)) {
 613                                 /* Line break before the next dollar. */
 614                                 continue;
 615                         }
 616
 617                         spc = memchr(src + 4, ' ', dollar - src - 4);
 618                         if (spc && spc < dollar-1) {
 619                                 /* There are spaces in unexpected places.
 620                                  * This is probably an id from some other
 621                                  * versioning system. Keep it for now.
 622                                  */
 623                                 continue;
 624                         }
 625
 626                         len -= dollar + 1 - src;
 627                         src  = dollar + 1;
 628                 } else {
 629                         /* it wasn't a "Id$" or "Id:xxxx$" */
 630                         continue;
 631                 }
 632
 633                 /* step 4: substitute */
 634                 strbuf_addstr(buf, "Id: ");
 635                 strbuf_add(buf, sha1_to_hex(sha1), 40);
 636                 strbuf_addstr(buf, " $");
 637         }
 638         strbuf_add(buf, src, len);
 639
 640         free(to_free);
 641         return 1;
 642 }
 643
 644 static enum crlf_action git_path_check_crlf(const char *path, struct git_attr_check *check)
 645 {
 646         const char *value = check->value;
 647
 648         if (ATTR_TRUE(value))
 649                 return CRLF_TEXT;
 650         else if (ATTR_FALSE(value))
 651                 return CRLF_BINARY;
 652         else if (ATTR_UNSET(value))
 653                 ;
 654         else if (!strcmp(value, "input"))
 655                 return CRLF_INPUT;
 656         else if (!strcmp(value, "auto"))
 657                 return CRLF_AUTO;
 658         return CRLF_GUESS;
 659 }
 660
 661 static enum eol git_path_check_eol(const char *path, struct git_attr_check *check)
 662 {
 663         const char *value = check->value;
 664
 665         if (ATTR_UNSET(value))
 666                 ;
 667         else if (!strcmp(value, "lf"))
 668                 return EOL_LF;
 669         else if (!strcmp(value, "crlf"))
 670                 return EOL_CRLF;
 671         return EOL_UNSET;
 672 }
 673
 674 static struct convert_driver *git_path_check_convert(const char *path,
 675                                              struct git_attr_check *check)
 676 {
 677         const char *value = check->value;
 678         struct convert_driver *drv;
 679
 680         if (ATTR_TRUE(value) || ATTR_FALSE(value) || ATTR_UNSET(value))
 681                 return NULL;
 682         for (drv = user_convert; drv; drv = drv->next)
 683                 if (!strcmp(value, drv->name))
 684                         return drv;
 685         return NULL;
 686 }
 687
 688 static int git_path_check_ident(const char *path, struct git_attr_check *check)
 689 {
 690         const char *value = check->value;
 691
 692         return !!ATTR_TRUE(value);
 693 }
 694
 695 static enum crlf_action input_crlf_action(enum crlf_action text_attr, enum eol eol_attr)
 696 {
 697         if (text_attr == CRLF_BINARY)
 698                 return CRLF_BINARY;
 699         if (eol_attr == EOL_LF)
 700                 return CRLF_INPUT;
 701         if (eol_attr == EOL_CRLF)
 702                 return CRLF_CRLF;
 703         return text_attr;
 704 }
 705
 706 struct conv_attrs {
 707         struct convert_driver *drv;
 708         enum crlf_action crlf_action;
 709         enum eol eol_attr;
 710         int ident;
 711 };
 712
 713 static const char *conv_attr_name[] = {
 714         "crlf", "ident", "filter", "eol", "text",
 715 };
 716 #define NUM_CONV_ATTRS ARRAY_SIZE(conv_attr_name)
 717
 718 static void convert_attrs(struct conv_attrs *ca, const char *path)
 719 {
 720         int i;
 721         static struct git_attr_check ccheck[NUM_CONV_ATTRS];
 722
 723         if (!ccheck[0].attr) {
 724                 for (i = 0; i < NUM_CONV_ATTRS; i++)
 725                         ccheck[i].attr = git_attr(conv_attr_name[i]);
 726                 user_convert_tail = &user_convert;
 727                 git_config(read_convert_config, NULL);
 728         }
 729
 730         if (!git_check_attr(path, NUM_CONV_ATTRS, ccheck)) {
 731                 ca->crlf_action = git_path_check_crlf(path, ccheck + 4);
 732                 if (ca->crlf_action == CRLF_GUESS)
 733                         ca->crlf_action = git_path_check_crlf(path, ccheck + 0);
 734                 ca->ident = git_path_check_ident(path, ccheck + 1);
 735                 ca->drv = git_path_check_convert(path, ccheck + 2);
 736                 ca->eol_attr = git_path_check_eol(path, ccheck + 3);
 737         } else {
 738                 ca->drv = NULL;
 739                 ca->crlf_action = CRLF_GUESS;
 740                 ca->eol_attr = EOL_UNSET;
 741                 ca->ident = 0;
 742         }
 743 }
 744
 745 int convert_to_git(const char *path, const char *src, size_t len,
 746                    struct strbuf *dst, enum safe_crlf checksafe)
 747 {
 748         int ret = 0;
 749         const char *filter = NULL;
 750         struct conv_attrs ca;
 751
 752         convert_attrs(&ca, path);
 753         if (ca.drv)
 754                 filter = ca.drv->clean;
 755
 756         ret |= apply_filter(path, src, len, dst, filter);
 757         if (ret) {
 758                 src = dst->buf;
 759                 len = dst->len;
 760         }
 761         ca.crlf_action = input_crlf_action(ca.crlf_action, ca.eol_attr);
 762         ret |= crlf_to_git(path, src, len, dst, ca.crlf_action, checksafe);
 763         if (ret) {
 764                 src = dst->buf;
 765                 len = dst->len;
 766         }
 767         return ret | ident_to_git(path, src, len, dst, ca.ident);
 768 }
 769
 770 static int convert_to_working_tree_internal(const char *path, const char *src,
 771                                             size_t len, struct strbuf *dst,
 772                                             int normalizing)
 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->smudge;
 781
 782         ret |= ident_to_worktree(path, src, len, dst, ca.ident);
 783         if (ret) {
 784                 src = dst->buf;
 785                 len = dst->len;
 786         }
 787         /*
 788          * CRLF conversion can be skipped if normalizing, unless there
 789          * is a smudge filter.  The filter might expect CRLFs.
 790          */
 791         if (filter || !normalizing) {
 792                 ca.crlf_action = input_crlf_action(ca.crlf_action, ca.eol_attr);
 793                 ret |= crlf_to_worktree(path, src, len, dst, ca.crlf_action);
 794                 if (ret) {
 795                         src = dst->buf;
 796                         len = dst->len;
 797                 }
 798         }
 799         return ret | apply_filter(path, src, len, dst, filter);
 800 }
 801
 802 int convert_to_working_tree(const char *path, const char *src, size_t len, struct strbuf *dst)
 803 {
 804         return convert_to_working_tree_internal(path, src, len, dst, 0);
 805 }
 806
 807 int renormalize_buffer(const char *path, const char *src, size_t len, struct strbuf *dst)
 808 {
 809         int ret = convert_to_working_tree_internal(path, src, len, dst, 1);
 810         if (ret) {
 811                 src = dst->buf;
 812                 len = dst->len;
 813         }
 814         return ret | convert_to_git(path, src, len, dst, SAFE_CRLF_FALSE);
 815 }
 816
 817 /*****************************************************************
 818  *
 819  * Streaming converison support
 820  *
 821  *****************************************************************/
 822
 823 typedef int (*filter_fn)(struct stream_filter *,
 824                          const char *input, size_t *isize_p,
 825                          char *output, size_t *osize_p);
 826 typedef void (*free_fn)(struct stream_filter *);
 827
 828 struct stream_filter_vtbl {
 829         filter_fn filter;
 830         free_fn free;
 831 };
 832
 833 struct stream_filter {
 834         struct stream_filter_vtbl *vtbl;
 835 };
 836
 837 static int null_filter_fn(struct stream_filter *filter,
 838                           const char *input, size_t *isize_p,
 839                           char *output, size_t *osize_p)
 840 {
 841         size_t count;
 842
 843         if (!input)
 844                 return 0; /* we do not keep any states */
 845         count = *isize_p;
 846         if (*osize_p < count)
 847                 count = *osize_p;
 848         if (count) {
 849                 memmove(output, input, count);
 850                 *isize_p -= count;
 851                 *osize_p -= count;
 852         }
 853         return 0;
 854 }
 855
 856 static void null_free_fn(struct stream_filter *filter)
 857 {
 858         ; /* nothing -- null instances are shared */
 859 }
 860
 861 static struct stream_filter_vtbl null_vtbl = {
 862         null_filter_fn,
 863         null_free_fn,
 864 };
 865
 866 static struct stream_filter null_filter_singleton = {
 867         &null_vtbl,
 868 };
 869
 870 int is_null_stream_filter(struct stream_filter *filter)
 871 {
 872         return filter == &null_filter_singleton;
 873 }
 874
 875
 876 /*
 877  * LF-to-CRLF filter
 878  */
 879 static int lf_to_crlf_filter_fn(struct stream_filter *filter,
 880                                 const char *input, size_t *isize_p,
 881                                 char *output, size_t *osize_p)
 882 {
 883         size_t count;
 884
 885         if (!input)
 886                 return 0; /* we do not keep any states */
 887         count = *isize_p;
 888         if (count) {
 889                 size_t i, o;
 890                 for (i = o = 0; o < *osize_p && i < count; i++) {
 891                         char ch = input[i];
 892                         if (ch == '\n') {
 893                                 if (o + 1 < *osize_p)
 894                                         output[o++] = '\r';
 895                                 else
 896                                         break;
 897                         }
 898                         output[o++] = ch;
 899                 }
 900
 901                 *osize_p -= o;
 902                 *isize_p -= i;
 903         }
 904         return 0;
 905 }
 906
 907 static struct stream_filter_vtbl lf_to_crlf_vtbl = {
 908         lf_to_crlf_filter_fn,
 909         null_free_fn,
 910 };
 911
 912 static struct stream_filter lf_to_crlf_filter_singleton = {
 913         &lf_to_crlf_vtbl,
 914 };
 915
 916
 917 /*
 918  * Cascade filter
 919  */
 920 #define FILTER_BUFFER 1024
 921 struct cascade_filter {
 922         struct stream_filter filter;
 923         struct stream_filter *one;
 924         struct stream_filter *two;
 925         char buf[FILTER_BUFFER];
 926         int end, ptr;
 927 };
 928
 929 static int cascade_filter_fn(struct stream_filter *filter,
 930                              const char *input, size_t *isize_p,
 931                              char *output, size_t *osize_p)
 932 {
 933         struct cascade_filter *cas = (struct cascade_filter *) filter;
 934         size_t filled = 0;
 935         size_t sz = *osize_p;
 936         size_t to_feed, remaining;
 937
 938         /*
 939          * input -- (one) --> buf -- (two) --> output
 940          */
 941         while (filled < sz) {
 942                 remaining = sz - filled;
 943
 944                 /* do we already have something to feed two with? */
 945                 if (cas->ptr < cas->end) {
 946                         to_feed = cas->end - cas->ptr;
 947                         if (stream_filter(cas->two,
 948                                           cas->buf + cas->ptr, &to_feed,
 949                                           output + filled, &remaining))
 950                                 return -1;
 951                         cas->ptr += (cas->end - cas->ptr) - to_feed;
 952                         filled = sz - remaining;
 953                         continue;
 954                 }
 955
 956                 /* feed one from upstream and have it emit into our buffer */
 957                 to_feed = input ? *isize_p : 0;
 958                 if (input && !to_feed)
 959                         break;
 960                 remaining = sizeof(cas->buf);
 961                 if (stream_filter(cas->one,
 962                                   input, &to_feed,
 963                                   cas->buf, &remaining))
 964                         return -1;
 965                 cas->end = sizeof(cas->buf) - remaining;
 966                 cas->ptr = 0;
 967                 if (input) {
 968                         size_t fed = *isize_p - to_feed;
 969                         *isize_p -= fed;
 970                         input += fed;
 971                 }
 972
 973                 /* do we know that we drained one completely? */
 974                 if (input || cas->end)
 975                         continue;
 976
 977                 /* tell two to drain; we have nothing more to give it */
 978                 to_feed = 0;
 979                 remaining = sz - filled;
 980                 if (stream_filter(cas->two,
 981                                   NULL, &to_feed,
 982                                   output + filled, &remaining))
 983                         return -1;
 984                 if (remaining == (sz - filled))
 985                         break; /* completely drained two */
 986                 filled = sz - remaining;
 987         }
 988         *osize_p -= filled;
 989         return 0;
 990 }
 991
 992 static void cascade_free_fn(struct stream_filter *filter)
 993 {
 994         struct cascade_filter *cas = (struct cascade_filter *)filter;
 995         free_stream_filter(cas->one);
 996         free_stream_filter(cas->two);
 997         free(filter);
 998 }
 999
1000 static struct stream_filter_vtbl cascade_vtbl = {
1001         cascade_filter_fn,
1002         cascade_free_fn,
1003 };
1004
1005 static struct stream_filter *cascade_filter(struct stream_filter *one,
1006                                             struct stream_filter *two)
1007 {
1008         struct cascade_filter *cascade;
1009
1010         if (!one || is_null_stream_filter(one))
1011                 return two;
1012         if (!two || is_null_stream_filter(two))
1013                 return one;
1014
1015         cascade = xmalloc(sizeof(*cascade));
1016         cascade->one = one;
1017         cascade->two = two;
1018         cascade->end = cascade->ptr = 0;
1019         cascade->filter.vtbl = &cascade_vtbl;
1020         return (struct stream_filter *)cascade;
1021 }
1022
1023 /*
1024  * ident filter
1025  */
1026 #define IDENT_DRAINING (-1)
1027 #define IDENT_SKIPPING (-2)
1028 struct ident_filter {
1029         struct stream_filter filter;
1030         struct strbuf left;
1031         int state;
1032         char ident[45]; /* ": x40 $" */
1033 };
1034
1035 static int is_foreign_ident(const char *str)
1036 {
1037         int i;
1038
1039         if (prefixcmp(str, "$Id: "))
1040                 return 0;
1041         for (i = 5; str[i]; i++) {
1042                 if (isspace(str[i]) && str[i+1] != '$')
1043                         return 1;
1044         }
1045         return 0;
1046 }
1047
1048 static void ident_drain(struct ident_filter *ident, char **output_p, size_t *osize_p)
1049 {
1050         size_t to_drain = ident->left.len;
1051
1052         if (*osize_p < to_drain)
1053                 to_drain = *osize_p;
1054         if (to_drain) {
1055                 memcpy(*output_p, ident->left.buf, to_drain);
1056                 strbuf_remove(&ident->left, 0, to_drain);
1057                 *output_p += to_drain;
1058                 *osize_p -= to_drain;
1059         }
1060         if (!ident->left.len)
1061                 ident->state = 0;
1062 }
1063
1064 static int ident_filter_fn(struct stream_filter *filter,
1065                            const char *input, size_t *isize_p,
1066                            char *output, size_t *osize_p)
1067 {
1068         struct ident_filter *ident = (struct ident_filter *)filter;
1069         static const char head[] = "$Id";
1070
1071         if (!input) {
1072                 /* drain upon eof */
1073                 switch (ident->state) {
1074                 default:
1075                         strbuf_add(&ident->left, head, ident->state);
1076                 case IDENT_SKIPPING:
1077                         /* fallthru */
1078                 case IDENT_DRAINING:
1079                         ident_drain(ident, &output, osize_p);
1080                 }
1081                 return 0;
1082         }
1083
1084         while (*isize_p || (ident->state == IDENT_DRAINING)) {
1085                 int ch;
1086
1087                 if (ident->state == IDENT_DRAINING) {
1088                         ident_drain(ident, &output, osize_p);
1089                         if (!*osize_p)
1090                                 break;
1091                         continue;
1092                 }
1093
1094                 ch = *(input++);
1095                 (*isize_p)--;
1096
1097                 if (ident->state == IDENT_SKIPPING) {
1098                         /*
1099                          * Skipping until '$' or LF, but keeping them
1100                          * in case it is a foreign ident.
1101                          */
1102                         strbuf_addch(&ident->left, ch);
1103                         if (ch != '\n' && ch != '$')
1104                                 continue;
1105                         if (ch == '$' && !is_foreign_ident(ident->left.buf)) {
1106                                 strbuf_setlen(&ident->left, sizeof(head) - 1);
1107                                 strbuf_addstr(&ident->left, ident->ident);
1108                         }
1109                         ident->state = IDENT_DRAINING;
1110                         continue;
1111                 }
1112
1113                 if (ident->state < sizeof(head) &&
1114                     head[ident->state] == ch) {
1115                         ident->state++;
1116                         continue;
1117                 }
1118
1119                 if (ident->state)
1120                         strbuf_add(&ident->left, head, ident->state);
1121                 if (ident->state == sizeof(head) - 1) {
1122                         if (ch != ':' && ch != '$') {
1123                                 strbuf_addch(&ident->left, ch);
1124                                 ident->state = 0;
1125                                 continue;
1126                         }
1127
1128                         if (ch == ':') {
1129                                 strbuf_addch(&ident->left, ch);
1130                                 ident->state = IDENT_SKIPPING;
1131                         } else {
1132                                 strbuf_addstr(&ident->left, ident->ident);
1133                                 ident->state = IDENT_DRAINING;
1134                         }
1135                         continue;
1136                 }
1137
1138                 strbuf_addch(&ident->left, ch);
1139                 ident->state = IDENT_DRAINING;
1140         }
1141         return 0;
1142 }
1143
1144 static void ident_free_fn(struct stream_filter *filter)
1145 {
1146         struct ident_filter *ident = (struct ident_filter *)filter;
1147         strbuf_release(&ident->left);
1148         free(filter);
1149 }
1150
1151 static struct stream_filter_vtbl ident_vtbl = {
1152         ident_filter_fn,
1153         ident_free_fn,
1154 };
1155
1156 static struct stream_filter *ident_filter(const unsigned char *sha1)
1157 {
1158         struct ident_filter *ident = xmalloc(sizeof(*ident));
1159
1160         sprintf(ident->ident, ": %s $", sha1_to_hex(sha1));
1161         strbuf_init(&ident->left, 0);
1162         ident->filter.vtbl = &ident_vtbl;
1163         ident->state = 0;
1164         return (struct stream_filter *)ident;
1165 }
1166
1167 /*
1168  * Return an appropriately constructed filter for the path, or NULL if
1169  * the contents cannot be filtered without reading the whole thing
1170  * in-core.
1171  *
1172  * Note that you would be crazy to set CRLF, smuge/clean or ident to a
1173  * large binary blob you would want us not to slurp into the memory!
1174  */
1175 struct stream_filter *get_stream_filter(const char *path, const unsigned char *sha1)
1176 {
1177         struct conv_attrs ca;
1178         enum crlf_action crlf_action;
1179         struct stream_filter *filter = NULL;
1180
1181         convert_attrs(&ca, path);
1182
1183         if (ca.drv && (ca.drv->smudge || ca.drv->clean))
1184                 return filter;
1185
1186         if (ca.ident)
1187                 filter = ident_filter(sha1);
1188
1189         crlf_action = input_crlf_action(ca.crlf_action, ca.eol_attr);
1190
1191         if ((crlf_action == CRLF_BINARY) || (crlf_action == CRLF_INPUT) ||
1192             (crlf_action == CRLF_GUESS && auto_crlf == AUTO_CRLF_FALSE))
1193                 filter = cascade_filter(filter, &null_filter_singleton);
1194
1195         else if (output_eol(crlf_action) == EOL_CRLF &&
1196                  !(crlf_action == CRLF_AUTO || crlf_action == CRLF_GUESS))
1197                 filter = cascade_filter(filter, &lf_to_crlf_filter_singleton);
1198
1199         return filter;
1200 }
1201
1202 void free_stream_filter(struct stream_filter *filter)
1203 {
1204         filter->vtbl->free(filter);
1205 }
1206
1207 int stream_filter(struct stream_filter *filter,
1208                   const char *input, size_t *isize_p,
1209                   char *output, size_t *osize_p)
1210 {
1211         return filter->vtbl->filter(filter, input, isize_p, output, osize_p);
1212 }