convert.c

   1 #include "cache.h"
   2 #include "attr.h"
   3 #include "run-command.h"
   4 #include "quote.h"
   5 #include "sigchain.h"
   6
   7 /*
   8  * convert.c - convert a file when checking it out and checking it in.
   9  *
  10  * This should use the pathname to decide on whether it wants to do some
  11  * more interesting conversions (automatic gzip/unzip, general format
  12  * conversions etc etc), but by default it just does automatic CRLF<->LF
  13  * translation when the "text" attribute or "auto_crlf" option is set.
  14  */
  15
  16 enum crlf_action {
  17         CRLF_GUESS = -1,
  18         CRLF_BINARY = 0,
  19         CRLF_TEXT,
  20         CRLF_INPUT,
  21         CRLF_CRLF,
  22         CRLF_AUTO
  23 };
  24
  25 struct text_stat {
  26         /* NUL, CR, LF and CRLF counts */
  27         unsigned nul, cr, lf, crlf;
  28
  29         /* These are just approximations! */
  30         unsigned printable, nonprintable;
  31 };
  32
  33 static void gather_stats(const char *buf, unsigned long size, struct text_stat *stats)
  34 {
  35         unsigned long i;
  36
  37         memset(stats, 0, sizeof(*stats));
  38
  39         for (i = 0; i < size; i++) {
  40                 unsigned char c = buf[i];
  41                 if (c == '\r') {
  42                         stats->cr++;
  43                         if (i+1 < size && buf[i+1] == '\n')
  44                                 stats->crlf++;
  45                         continue;
  46                 }
  47                 if (c == '\n') {
  48                         stats->lf++;
  49                         continue;
  50                 }
  51                 if (c == 127)
  52                         /* DEL */
  53                         stats->nonprintable++;
  54                 else if (c < 32) {
  55                         switch (c) {
  56                                 /* BS, HT, ESC and FF */
  57                         case '\b': case '\t': case '\033': case '\014':
  58                                 stats->printable++;
  59                                 break;
  60                         case 0:
  61                                 stats->nul++;
  62                                 /* fall through */
  63                         default:
  64                                 stats->nonprintable++;
  65                         }
  66                 }
  67                 else
  68                         stats->printable++;
  69         }
  70
  71         /* If file ends with EOF then don't count this EOF as non-printable. */
  72         if (size >= 1 && buf[size-1] == '\032')
  73                 stats->nonprintable--;
  74 }
  75
  76 /*
  77  * The same heuristics as diff.c::mmfile_is_binary()
  78  */
  79 static int is_binary(unsigned long size, struct text_stat *stats)
  80 {
  81
  82         if (stats->nul)
  83                 return 1;
  84         if ((stats->printable >> 7) < stats->nonprintable)
  85                 return 1;
  86         /*
  87          * Other heuristics? Average line length might be relevant,
  88          * as might LF vs CR vs CRLF counts..
  89          *
  90          * NOTE! It might be normal to have a low ratio of CRLF to LF
  91          * (somebody starts with a LF-only file and edits it with an editor
  92          * that adds CRLF only to lines that are added..). But do  we
  93          * want to support CR-only? Probably not.
  94          */
  95         return 0;
  96 }
  97
  98 static enum eol output_eol(enum crlf_action crlf_action)
  99 {
 100         switch (crlf_action) {
 101         case CRLF_BINARY:
 102                 return EOL_UNSET;
 103         case CRLF_CRLF:
 104                 return EOL_CRLF;
 105         case CRLF_INPUT:
 106                 return EOL_LF;
 107         case CRLF_GUESS:
 108                 if (!auto_crlf)
 109                         return EOL_UNSET;
 110                 /* fall through */
 111         case CRLF_TEXT:
 112         case CRLF_AUTO:
 113                 if (auto_crlf == AUTO_CRLF_TRUE)
 114                         return EOL_CRLF;
 115                 else if (auto_crlf == AUTO_CRLF_INPUT)
 116                         return EOL_LF;
 117                 else if (core_eol == EOL_UNSET)
 118                         return EOL_NATIVE;
 119         }
 120         return core_eol;
 121 }
 122
 123 static void check_safe_crlf(const char *path, enum crlf_action crlf_action,
 124                             struct text_stat *stats, enum safe_crlf checksafe)
 125 {
 126         if (!checksafe)
 127                 return;
 128
 129         if (output_eol(crlf_action) == EOL_LF) {
 130                 /*
 131                  * CRLFs would not be restored by checkout:
 132                  * check if we'd remove CRLFs
 133                  */
 134                 if (stats->crlf) {
 135                         if (checksafe == SAFE_CRLF_WARN)
 136                                 warning("CRLF will be replaced by LF in %s.\nThe file will have its original line endings in your working directory.", path);
 137                         else /* i.e. SAFE_CRLF_FAIL */
 138                                 die("CRLF would be replaced by LF in %s.", path);
 139                 }
 140         } else if (output_eol(crlf_action) == EOL_CRLF) {
 141                 /*
 142                  * CRLFs would be added by checkout:
 143                  * check if we have "naked" LFs
 144                  */
 145                 if (stats->lf != stats->crlf) {
 146                         if (checksafe == SAFE_CRLF_WARN)
 147                                 warning("LF will be replaced by CRLF in %s.\nThe file will have its original line endings in your working directory.", path);
 148                         else /* i.e. SAFE_CRLF_FAIL */
 149                                 die("LF would be replaced by CRLF in %s", path);
 150                 }
 151         }
 152 }
 153
 154 static int has_cr_in_index(const char *path)
 155 {
 156         unsigned long sz;
 157         void *data;
 158         int has_cr;
 159
 160         data = read_blob_data_from_cache(path, &sz);
 161         if (!data)
 162                 return 0;
 163         has_cr = memchr(data, '\r', sz) != NULL;
 164         free(data);
 165         return has_cr;
 166 }
 167
 168 static int crlf_to_git(const char *path, const char *src, size_t len,
 169                        struct strbuf *buf,
 170                        enum crlf_action crlf_action, enum safe_crlf checksafe)
 171 {
 172         struct text_stat stats;
 173         char *dst;
 174
 175         if (crlf_action == CRLF_BINARY ||
 176             (crlf_action == CRLF_GUESS && auto_crlf == AUTO_CRLF_FALSE) ||
 177             (src && !len))
 178                 return 0;
 179
 180         /*
 181          * If we are doing a dry-run and have no source buffer, there is
 182          * nothing to analyze; we must assume we would convert.
 183          */
 184         if (!buf && !src)
 185                 return 1;
 186
 187         gather_stats(src, len, &stats);
 188
 189         if (crlf_action == CRLF_AUTO || crlf_action == CRLF_GUESS) {
 190                 /*
 191                  * We're currently not going to even try to convert stuff
 192                  * that has bare CR characters. Does anybody do that crazy
 193                  * stuff?
 194                  */
 195                 if (stats.cr != stats.crlf)
 196                         return 0;
 197
 198                 /*
 199                  * And add some heuristics for binary vs text, of course...
 200                  */
 201                 if (is_binary(len, &stats))
 202                         return 0;
 203
 204                 if (crlf_action == CRLF_GUESS) {
 205                         /*
 206                          * If the file in the index has any CR in it, do not convert.
 207                          * This is the new safer autocrlf handling.
 208                          */
 209                         if (has_cr_in_index(path))
 210                                 return 0;
 211                 }
 212         }
 213
 214         check_safe_crlf(path, crlf_action, &stats, checksafe);
 215
 216         /* Optimization: No CR? Nothing to convert, regardless. */
 217         if (!stats.cr)
 218                 return 0;
 219
 220         /*
 221          * At this point all of our source analysis is done, and we are sure we
 222          * would convert. If we are in dry-run mode, we can give an answer.
 223          */
 224         if (!buf)
 225                 return 1;
 226
 227         /* only grow if not in place */
 228         if (strbuf_avail(buf) + buf->len < len)
 229                 strbuf_grow(buf, len - buf->len);
 230         dst = buf->buf;
 231         if (crlf_action == CRLF_AUTO || crlf_action == CRLF_GUESS) {
 232                 /*
 233                  * If we guessed, we already know we rejected a file with
 234                  * lone CR, and we can strip a CR without looking at what
 235                  * follow it.
 236                  */
 237                 do {
 238                         unsigned char c = *src++;
 239                         if (c != '\r')
 240                                 *dst++ = c;
 241                 } while (--len);
 242         } else {
 243                 do {
 244                         unsigned char c = *src++;
 245                         if (! (c == '\r' && (1 < len && *src == '\n')))
 246                                 *dst++ = c;
 247                 } while (--len);
 248         }
 249         strbuf_setlen(buf, dst - buf->buf);
 250         return 1;
 251 }
 252
 253 static int crlf_to_worktree(const char *path, const char *src, size_t len,
 254                             struct strbuf *buf, enum crlf_action crlf_action)
 255 {
 256         char *to_free = NULL;
 257         struct text_stat stats;
 258
 259         if (!len || output_eol(crlf_action) != EOL_CRLF)
 260                 return 0;
 261
 262         gather_stats(src, len, &stats);
 263
 264         /* No LF? Nothing to convert, regardless. */
 265         if (!stats.lf)
 266                 return 0;
 267
 268         /* Was it already in CRLF format? */
 269         if (stats.lf == stats.crlf)
 270                 return 0;
 271
 272         if (crlf_action == CRLF_AUTO || crlf_action == CRLF_GUESS) {
 273                 if (crlf_action == CRLF_GUESS) {
 274                         /* If we have any CR or CRLF line endings, we do not touch it */
 275                         /* This is the new safer autocrlf-handling */
 276                         if (stats.cr > 0 || stats.crlf > 0)
 277                                 return 0;
 278                 }
 279
 280                 /* If we have any bare CR characters, we're not going to touch it */
 281                 if (stats.cr != stats.crlf)
 282                         return 0;
 283
 284                 if (is_binary(len, &stats))
 285                         return 0;
 286         }
 287
 288         /* are we "faking" in place editing ? */
 289         if (src == buf->buf)
 290                 to_free = strbuf_detach(buf, NULL);
 291
 292         strbuf_grow(buf, len + stats.lf - stats.crlf);
 293         for (;;) {
 294                 const char *nl = memchr(src, '\n', len);
 295                 if (!nl)
 296                         break;
 297                 if (nl > src && nl[-1] == '\r') {
 298                         strbuf_add(buf, src, nl + 1 - src);
 299                 } else {
 300                         strbuf_add(buf, src, nl - src);
 301                         strbuf_addstr(buf, "\r\n");
 302                 }
 303                 len -= nl + 1 - src;
 304                 src  = nl + 1;
 305         }
 306         strbuf_add(buf, src, len);
 307
 308         free(to_free);
 309         return 1;
 310 }
 311
 312 struct filter_params {
 313         const char *src;
 314         unsigned long size;
 315         int fd;
 316         const char *cmd;
 317         const char *path;
 318 };
 319
 320 static int filter_buffer_or_fd(int in, int out, void *data)
 321 {
 322         /*
 323          * Spawn cmd and feed the buffer contents through its stdin.
 324          */
 325         struct child_process child_process = CHILD_PROCESS_INIT;
 326         struct filter_params *params = (struct filter_params *)data;
 327         int write_err, status;
 328         const char *argv[] = { NULL, NULL };
 329
 330         /* apply % substitution to cmd */
 331         struct strbuf cmd = STRBUF_INIT;
 332         struct strbuf path = STRBUF_INIT;
 333         struct strbuf_expand_dict_entry dict[] = {
 334                 { "f", NULL, },
 335                 { NULL, NULL, },
 336         };
 337
 338         /* quote the path to preserve spaces, etc. */
 339         sq_quote_buf(&path, params->path);
 340         dict[0].value = path.buf;
 341
 342         /* expand all %f with the quoted path */
 343         strbuf_expand(&cmd, params->cmd, strbuf_expand_dict_cb, &dict);
 344         strbuf_release(&path);
 345
 346         argv[0] = cmd.buf;
 347
 348         child_process.argv = argv;
 349         child_process.use_shell = 1;
 350         child_process.in = -1;
 351         child_process.out = out;
 352
 353         if (start_command(&child_process))
 354                 return error("cannot fork to run external filter %s", params->cmd);
 355
 356         sigchain_push(SIGPIPE, SIG_IGN);
 357
 358         if (params->src) {
 359                 write_err = (write_in_full(child_process.in,
 360                                            params->src, params->size) < 0);
 361                 if (errno == EPIPE)
 362                         write_err = 0;
 363         } else {
 364                 write_err = copy_fd(params->fd, child_process.in);
 365                 if (write_err == COPY_WRITE_ERROR && errno == EPIPE)
 366                         write_err = 0;
 367         }
 368
 369         if (close(child_process.in))
 370                 write_err = 1;
 371         if (write_err)
 372                 error("cannot feed the input to external filter %s", params->cmd);
 373
 374         sigchain_pop(SIGPIPE);
 375
 376         status = finish_command(&child_process);
 377         if (status)
 378                 error("external filter %s failed %d", params->cmd, status);
 379
 380         strbuf_release(&cmd);
 381         return (write_err || status);
 382 }
 383
 384 static int apply_filter(const char *path, const char *src, size_t len, int fd,
 385                         struct strbuf *dst, const char *cmd)
 386 {
 387         /*
 388          * Create a pipeline to have the command filter the buffer's
 389          * contents.
 390          *
 391          * (child --> cmd) --> us
 392          */
 393         int ret = 1;
 394         struct strbuf nbuf = STRBUF_INIT;
 395         struct async async;
 396         struct filter_params params;
 397
 398         if (!cmd)
 399                 return 0;
 400
 401         if (!dst)
 402                 return 1;
 403
 404         memset(&async, 0, sizeof(async));
 405         async.proc = filter_buffer_or_fd;
 406         async.data = &params;
 407         async.out = -1;
 408         params.src = src;
 409         params.size = len;
 410         params.fd = fd;
 411         params.cmd = cmd;
 412         params.path = path;
 413
 414         fflush(NULL);
 415         if (start_async(&async))
 416                 return 0;       /* error was already reported */
 417
 418         if (strbuf_read(&nbuf, async.out, len) < 0) {
 419                 error("read from external filter %s failed", cmd);
 420                 ret = 0;
 421         }
 422         if (close(async.out)) {
 423                 error("read from external filter %s failed", cmd);
 424                 ret = 0;
 425         }
 426         if (finish_async(&async)) {
 427                 error("external filter %s failed", cmd);
 428                 ret = 0;
 429         }
 430
 431         if (ret) {
 432                 strbuf_swap(dst, &nbuf);
 433         }
 434         strbuf_release(&nbuf);
 435         return ret;
 436 }
 437
 438 static struct convert_driver {
 439         const char *name;
 440         struct convert_driver *next;
 441         const char *smudge;
 442         const char *clean;
 443         int required;
 444 } *user_convert, **user_convert_tail;
 445
 446 static int read_convert_config(const char *var, const char *value, void *cb)
 447 {
 448         const char *key, *name;
 449         int namelen;
 450         struct convert_driver *drv;
 451
 452         /*
 453          * External conversion drivers are configured using
 454          * "filter.<name>.variable".
 455          */
 456         if (parse_config_key(var, "filter", &name, &namelen, &key) < 0 || !name)
 457                 return 0;
 458         for (drv = user_convert; drv; drv = drv->next)
 459                 if (!strncmp(drv->name, name, namelen) && !drv->name[namelen])
 460                         break;
 461         if (!drv) {
 462                 drv = xcalloc(1, sizeof(struct convert_driver));
 463                 drv->name = xmemdupz(name, namelen);
 464                 *user_convert_tail = drv;
 465                 user_convert_tail = &(drv->next);
 466         }
 467
 468         /*
 469          * filter.<name>.smudge and filter.<name>.clean specifies
 470          * the command line:
 471          *
 472          *      command-line
 473          *
 474          * The command-line will not be interpolated in any way.
 475          */
 476
 477         if (!strcmp("smudge", key))
 478                 return git_config_string(&drv->smudge, var, value);
 479
 480         if (!strcmp("clean", key))
 481                 return git_config_string(&drv->clean, var, value);
 482
 483         if (!strcmp("required", key)) {
 484                 drv->required = git_config_bool(var, value);
 485                 return 0;
 486         }
 487
 488         return 0;
 489 }
 490
 491 static int count_ident(const char *cp, unsigned long size)
 492 {
 493         /*
 494          * "$Id: 0000000000000000000000000000000000000000 $" <=> "$Id$"
 495          */
 496         int cnt = 0;
 497         char ch;
 498
 499         while (size) {
 500                 ch = *cp++;
 501                 size--;
 502                 if (ch != '$')
 503                         continue;
 504                 if (size < 3)
 505                         break;
 506                 if (memcmp("Id", cp, 2))
 507                         continue;
 508                 ch = cp[2];
 509                 cp += 3;
 510                 size -= 3;
 511                 if (ch == '$')
 512                         cnt++; /* $Id$ */
 513                 if (ch != ':')
 514                         continue;
 515
 516                 /*
 517                  * "$Id: ... "; scan up to the closing dollar sign and discard.
 518                  */
 519                 while (size) {
 520                         ch = *cp++;
 521                         size--;
 522                         if (ch == '$') {
 523                                 cnt++;
 524                                 break;
 525                         }
 526                         if (ch == '\n')
 527                                 break;
 528                 }
 529         }
 530         return cnt;
 531 }
 532
 533 static int ident_to_git(const char *path, const char *src, size_t len,
 534                         struct strbuf *buf, int ident)
 535 {
 536         char *dst, *dollar;
 537
 538         if (!ident || (src && !count_ident(src, len)))
 539                 return 0;
 540
 541         if (!buf)
 542                 return 1;
 543
 544         /* only grow if not in place */
 545         if (strbuf_avail(buf) + buf->len < len)
 546                 strbuf_grow(buf, len - buf->len);
 547         dst = buf->buf;
 548         for (;;) {
 549                 dollar = memchr(src, '$', len);
 550                 if (!dollar)
 551                         break;
 552                 memmove(dst, src, dollar + 1 - src);
 553                 dst += dollar + 1 - src;
 554                 len -= dollar + 1 - src;
 555                 src  = dollar + 1;
 556
 557                 if (len > 3 && !memcmp(src, "Id:", 3)) {
 558                         dollar = memchr(src + 3, '$', len - 3);
 559                         if (!dollar)
 560                                 break;
 561                         if (memchr(src + 3, '\n', dollar - src - 3)) {
 562                                 /* Line break before the next dollar. */
 563                                 continue;
 564                         }
 565
 566                         memcpy(dst, "Id$", 3);
 567                         dst += 3;
 568                         len -= dollar + 1 - src;
 569                         src  = dollar + 1;
 570                 }
 571         }
 572         memmove(dst, src, len);
 573         strbuf_setlen(buf, dst + len - buf->buf);
 574         return 1;
 575 }
 576
 577 static int ident_to_worktree(const char *path, const char *src, size_t len,
 578                              struct strbuf *buf, int ident)
 579 {
 580         unsigned char sha1[20];
 581         char *to_free = NULL, *dollar, *spc;
 582         int cnt;
 583
 584         if (!ident)
 585                 return 0;
 586
 587         cnt = count_ident(src, len);
 588         if (!cnt)
 589                 return 0;
 590
 591         /* are we "faking" in place editing ? */
 592         if (src == buf->buf)
 593                 to_free = strbuf_detach(buf, NULL);
 594         hash_sha1_file(src, len, "blob", sha1);
 595
 596         strbuf_grow(buf, len + cnt * 43);
 597         for (;;) {
 598                 /* step 1: run to the next '$' */
 599                 dollar = memchr(src, '$', len);
 600                 if (!dollar)
 601                         break;
 602                 strbuf_add(buf, src, dollar + 1 - src);
 603                 len -= dollar + 1 - src;
 604                 src  = dollar + 1;
 605
 606                 /* step 2: does it looks like a bit like Id:xxx$ or Id$ ? */
 607                 if (len < 3 || memcmp("Id", src, 2))
 608                         continue;
 609
 610                 /* step 3: skip over Id$ or Id:xxxxx$ */
 611                 if (src[2] == '$') {
 612                         src += 3;
 613                         len -= 3;
 614                 } else if (src[2] == ':') {
 615                         /*
 616                          * It's possible that an expanded Id has crept its way into the
 617                          * repository, we cope with that by stripping the expansion out.
 618                          * This is probably not a good idea, since it will cause changes
 619                          * on checkout, which won't go away by stash, but let's keep it
 620                          * for git-style ids.
 621                          */
 622                         dollar = memchr(src + 3, '$', len - 3);
 623                         if (!dollar) {
 624                                 /* incomplete keyword, no more '$', so just quit the loop */
 625                                 break;
 626                         }
 627
 628                         if (memchr(src + 3, '\n', dollar - src - 3)) {
 629                                 /* Line break before the next dollar. */
 630                                 continue;
 631                         }
 632
 633                         spc = memchr(src + 4, ' ', dollar - src - 4);
 634                         if (spc && spc < dollar-1) {
 635                                 /* There are spaces in unexpected places.
 636                                  * This is probably an id from some other
 637                                  * versioning system. Keep it for now.
 638                                  */
 639                                 continue;
 640                         }
 641
 642                         len -= dollar + 1 - src;
 643                         src  = dollar + 1;
 644                 } else {
 645                         /* it wasn't a "Id$" or "Id:xxxx$" */
 646                         continue;
 647                 }
 648
 649                 /* step 4: substitute */
 650                 strbuf_addstr(buf, "Id: ");
 651                 strbuf_add(buf, sha1_to_hex(sha1), 40);
 652                 strbuf_addstr(buf, " $");
 653         }
 654         strbuf_add(buf, src, len);
 655
 656         free(to_free);
 657         return 1;
 658 }
 659
 660 static enum crlf_action git_path_check_crlf(const char *path, struct git_attr_check *check)
 661 {
 662         const char *value = check->value;
 663
 664         if (ATTR_TRUE(value))
 665                 return CRLF_TEXT;
 666         else if (ATTR_FALSE(value))
 667                 return CRLF_BINARY;
 668         else if (ATTR_UNSET(value))
 669                 ;
 670         else if (!strcmp(value, "input"))
 671                 return CRLF_INPUT;
 672         else if (!strcmp(value, "auto"))
 673                 return CRLF_AUTO;
 674         return CRLF_GUESS;
 675 }
 676
 677 static enum eol git_path_check_eol(const char *path, struct git_attr_check *check)
 678 {
 679         const char *value = check->value;
 680
 681         if (ATTR_UNSET(value))
 682                 ;
 683         else if (!strcmp(value, "lf"))
 684                 return EOL_LF;
 685         else if (!strcmp(value, "crlf"))
 686                 return EOL_CRLF;
 687         return EOL_UNSET;
 688 }
 689
 690 static struct convert_driver *git_path_check_convert(const char *path,
 691                                              struct git_attr_check *check)
 692 {
 693         const char *value = check->value;
 694         struct convert_driver *drv;
 695
 696         if (ATTR_TRUE(value) || ATTR_FALSE(value) || ATTR_UNSET(value))
 697                 return NULL;
 698         for (drv = user_convert; drv; drv = drv->next)
 699                 if (!strcmp(value, drv->name))
 700                         return drv;
 701         return NULL;
 702 }
 703
 704 static int git_path_check_ident(const char *path, struct git_attr_check *check)
 705 {
 706         const char *value = check->value;
 707
 708         return !!ATTR_TRUE(value);
 709 }
 710
 711 static enum crlf_action input_crlf_action(enum crlf_action text_attr, enum eol eol_attr)
 712 {
 713         if (text_attr == CRLF_BINARY)
 714                 return CRLF_BINARY;
 715         if (eol_attr == EOL_LF)
 716                 return CRLF_INPUT;
 717         if (eol_attr == EOL_CRLF)
 718                 return CRLF_CRLF;
 719         return text_attr;
 720 }
 721
 722 struct conv_attrs {
 723         struct convert_driver *drv;
 724         enum crlf_action crlf_action;
 725         enum eol eol_attr;
 726         int ident;
 727 };
 728
 729 static const char *conv_attr_name[] = {
 730         "crlf", "ident", "filter", "eol", "text",
 731 };
 732 #define NUM_CONV_ATTRS ARRAY_SIZE(conv_attr_name)
 733
 734 static void convert_attrs(struct conv_attrs *ca, const char *path)
 735 {
 736         int i;
 737         static struct git_attr_check ccheck[NUM_CONV_ATTRS];
 738
 739         if (!ccheck[0].attr) {
 740                 for (i = 0; i < NUM_CONV_ATTRS; i++)
 741                         ccheck[i].attr = git_attr(conv_attr_name[i]);
 742                 user_convert_tail = &user_convert;
 743                 git_config(read_convert_config, NULL);
 744         }
 745
 746         if (!git_check_attr(path, NUM_CONV_ATTRS, ccheck)) {
 747                 ca->crlf_action = git_path_check_crlf(path, ccheck + 4);
 748                 if (ca->crlf_action == CRLF_GUESS)
 749                         ca->crlf_action = git_path_check_crlf(path, ccheck + 0);
 750                 ca->ident = git_path_check_ident(path, ccheck + 1);
 751                 ca->drv = git_path_check_convert(path, ccheck + 2);
 752                 ca->eol_attr = git_path_check_eol(path, ccheck + 3);
 753         } else {
 754                 ca->drv = NULL;
 755                 ca->crlf_action = CRLF_GUESS;
 756                 ca->eol_attr = EOL_UNSET;
 757                 ca->ident = 0;
 758         }
 759 }
 760
 761 int would_convert_to_git_filter_fd(const char *path)
 762 {
 763         struct conv_attrs ca;
 764
 765         convert_attrs(&ca, path);
 766         if (!ca.drv)
 767                 return 0;
 768
 769         /*
 770          * Apply a filter to an fd only if the filter is required to succeed.
 771          * We must die if the filter fails, because the original data before
 772          * filtering is not available.
 773          */
 774         if (!ca.drv->required)
 775                 return 0;
 776
 777         return apply_filter(path, NULL, 0, -1, NULL, ca.drv->clean);
 778 }
 779
 780 int convert_to_git(const char *path, const char *src, size_t len,
 781                    struct strbuf *dst, enum safe_crlf checksafe)
 782 {
 783         int ret = 0;
 784         const char *filter = NULL;
 785         int required = 0;
 786         struct conv_attrs ca;
 787
 788         convert_attrs(&ca, path);
 789         if (ca.drv) {
 790                 filter = ca.drv->clean;
 791                 required = ca.drv->required;
 792         }
 793
 794         ret |= apply_filter(path, src, len, -1, dst, filter);
 795         if (!ret && required)
 796                 die("%s: clean filter '%s' failed", path, ca.drv->name);
 797
 798         if (ret && dst) {
 799                 src = dst->buf;
 800                 len = dst->len;
 801         }
 802         ca.crlf_action = input_crlf_action(ca.crlf_action, ca.eol_attr);
 803         ret |= crlf_to_git(path, src, len, dst, ca.crlf_action, checksafe);
 804         if (ret && dst) {
 805                 src = dst->buf;
 806                 len = dst->len;
 807         }
 808         return ret | ident_to_git(path, src, len, dst, ca.ident);
 809 }
 810
 811 void convert_to_git_filter_fd(const char *path, int fd, struct strbuf *dst,
 812                               enum safe_crlf checksafe)
 813 {
 814         struct conv_attrs ca;
 815         convert_attrs(&ca, path);
 816
 817         assert(ca.drv);
 818         assert(ca.drv->clean);
 819
 820         if (!apply_filter(path, NULL, 0, fd, dst, ca.drv->clean))
 821                 die("%s: clean filter '%s' failed", path, ca.drv->name);
 822
 823         ca.crlf_action = input_crlf_action(ca.crlf_action, ca.eol_attr);
 824         crlf_to_git(path, dst->buf, dst->len, dst, ca.crlf_action, checksafe);
 825         ident_to_git(path, dst->buf, dst->len, dst, ca.ident);
 826 }
 827
 828 static int convert_to_working_tree_internal(const char *path, const char *src,
 829                                             size_t len, struct strbuf *dst,
 830                                             int normalizing)
 831 {
 832         int ret = 0, ret_filter = 0;
 833         const char *filter = NULL;
 834         int required = 0;
 835         struct conv_attrs ca;
 836
 837         convert_attrs(&ca, path);
 838         if (ca.drv) {
 839                 filter = ca.drv->smudge;
 840                 required = ca.drv->required;
 841         }
 842
 843         ret |= ident_to_worktree(path, src, len, dst, ca.ident);
 844         if (ret) {
 845                 src = dst->buf;
 846                 len = dst->len;
 847         }
 848         /*
 849          * CRLF conversion can be skipped if normalizing, unless there
 850          * is a smudge filter.  The filter might expect CRLFs.
 851          */
 852         if (filter || !normalizing) {
 853                 ca.crlf_action = input_crlf_action(ca.crlf_action, ca.eol_attr);
 854                 ret |= crlf_to_worktree(path, src, len, dst, ca.crlf_action);
 855                 if (ret) {
 856                         src = dst->buf;
 857                         len = dst->len;
 858                 }
 859         }
 860
 861         ret_filter = apply_filter(path, src, len, -1, dst, filter);
 862         if (!ret_filter && required)
 863                 die("%s: smudge filter %s failed", path, ca.drv->name);
 864
 865         return ret | ret_filter;
 866 }
 867
 868 int convert_to_working_tree(const char *path, const char *src, size_t len, struct strbuf *dst)
 869 {
 870         return convert_to_working_tree_internal(path, src, len, dst, 0);
 871 }
 872
 873 int renormalize_buffer(const char *path, const char *src, size_t len, struct strbuf *dst)
 874 {
 875         int ret = convert_to_working_tree_internal(path, src, len, dst, 1);
 876         if (ret) {
 877                 src = dst->buf;
 878                 len = dst->len;
 879         }
 880         return ret | convert_to_git(path, src, len, dst, SAFE_CRLF_FALSE);
 881 }
 882
 883 /*****************************************************************
 884  *
 885  * Streaming conversion support
 886  *
 887  *****************************************************************/
 888
 889 typedef int (*filter_fn)(struct stream_filter *,
 890                          const char *input, size_t *isize_p,
 891                          char *output, size_t *osize_p);
 892 typedef void (*free_fn)(struct stream_filter *);
 893
 894 struct stream_filter_vtbl {
 895         filter_fn filter;
 896         free_fn free;
 897 };
 898
 899 struct stream_filter {
 900         struct stream_filter_vtbl *vtbl;
 901 };
 902
 903 static int null_filter_fn(struct stream_filter *filter,
 904                           const char *input, size_t *isize_p,
 905                           char *output, size_t *osize_p)
 906 {
 907         size_t count;
 908
 909         if (!input)
 910                 return 0; /* we do not keep any states */
 911         count = *isize_p;
 912         if (*osize_p < count)
 913                 count = *osize_p;
 914         if (count) {
 915                 memmove(output, input, count);
 916                 *isize_p -= count;
 917                 *osize_p -= count;
 918         }
 919         return 0;
 920 }
 921
 922 static void null_free_fn(struct stream_filter *filter)
 923 {
 924         ; /* nothing -- null instances are shared */
 925 }
 926
 927 static struct stream_filter_vtbl null_vtbl = {
 928         null_filter_fn,
 929         null_free_fn,
 930 };
 931
 932 static struct stream_filter null_filter_singleton = {
 933         &null_vtbl,
 934 };
 935
 936 int is_null_stream_filter(struct stream_filter *filter)
 937 {
 938         return filter == &null_filter_singleton;
 939 }
 940
 941
 942 /*
 943  * LF-to-CRLF filter
 944  */
 945
 946 struct lf_to_crlf_filter {
 947         struct stream_filter filter;
 948         unsigned has_held:1;
 949         char held;
 950 };
 951
 952 static int lf_to_crlf_filter_fn(struct stream_filter *filter,
 953                                 const char *input, size_t *isize_p,
 954                                 char *output, size_t *osize_p)
 955 {
 956         size_t count, o = 0;
 957         struct lf_to_crlf_filter *lf_to_crlf = (struct lf_to_crlf_filter *)filter;
 958
 959         /*
 960          * We may be holding onto the CR to see if it is followed by a
 961          * LF, in which case we would need to go to the main loop.
 962          * Otherwise, just emit it to the output stream.
 963          */
 964         if (lf_to_crlf->has_held && (lf_to_crlf->held != '\r' || !input)) {
 965                 output[o++] = lf_to_crlf->held;
 966                 lf_to_crlf->has_held = 0;
 967         }
 968
 969         /* We are told to drain */
 970         if (!input) {
 971                 *osize_p -= o;
 972                 return 0;
 973         }
 974
 975         count = *isize_p;
 976         if (count || lf_to_crlf->has_held) {
 977                 size_t i;
 978                 int was_cr = 0;
 979
 980                 if (lf_to_crlf->has_held) {
 981                         was_cr = 1;
 982                         lf_to_crlf->has_held = 0;
 983                 }
 984
 985                 for (i = 0; o < *osize_p && i < count; i++) {
 986                         char ch = input[i];
 987
 988                         if (ch == '\n') {
 989                                 output[o++] = '\r';
 990                         } else if (was_cr) {
 991                                 /*
 992                                  * Previous round saw CR and it is not followed
 993                                  * by a LF; emit the CR before processing the
 994                                  * current character.
 995                                  */
 996                                 output[o++] = '\r';
 997                         }
 998
 999                         /*
1000                          * We may have consumed the last output slot,
1001                          * in which case we need to break out of this
1002                          * loop; hold the current character before
1003                          * returning.
1004                          */
1005                         if (*osize_p <= o) {
1006                                 lf_to_crlf->has_held = 1;
1007                                 lf_to_crlf->held = ch;
1008                                 continue; /* break but increment i */
1009                         }
1010
1011                         if (ch == '\r') {
1012                                 was_cr = 1;
1013                                 continue;
1014                         }
1015
1016                         was_cr = 0;
1017                         output[o++] = ch;
1018                 }
1019
1020                 *osize_p -= o;
1021                 *isize_p -= i;
1022
1023                 if (!lf_to_crlf->has_held && was_cr) {
1024                         lf_to_crlf->has_held = 1;
1025                         lf_to_crlf->held = '\r';
1026                 }
1027         }
1028         return 0;
1029 }
1030
1031 static void lf_to_crlf_free_fn(struct stream_filter *filter)
1032 {
1033         free(filter);
1034 }
1035
1036 static struct stream_filter_vtbl lf_to_crlf_vtbl = {
1037         lf_to_crlf_filter_fn,
1038         lf_to_crlf_free_fn,
1039 };
1040
1041 static struct stream_filter *lf_to_crlf_filter(void)
1042 {
1043         struct lf_to_crlf_filter *lf_to_crlf = xcalloc(1, sizeof(*lf_to_crlf));
1044
1045         lf_to_crlf->filter.vtbl = &lf_to_crlf_vtbl;
1046         return (struct stream_filter *)lf_to_crlf;
1047 }
1048
1049 /*
1050  * Cascade filter
1051  */
1052 #define FILTER_BUFFER 1024
1053 struct cascade_filter {
1054         struct stream_filter filter;
1055         struct stream_filter *one;
1056         struct stream_filter *two;
1057         char buf[FILTER_BUFFER];
1058         int end, ptr;
1059 };
1060
1061 static int cascade_filter_fn(struct stream_filter *filter,
1062                              const char *input, size_t *isize_p,
1063                              char *output, size_t *osize_p)
1064 {
1065         struct cascade_filter *cas = (struct cascade_filter *) filter;
1066         size_t filled = 0;
1067         size_t sz = *osize_p;
1068         size_t to_feed, remaining;
1069
1070         /*
1071          * input -- (one) --> buf -- (two) --> output
1072          */
1073         while (filled < sz) {
1074                 remaining = sz - filled;
1075
1076                 /* do we already have something to feed two with? */
1077                 if (cas->ptr < cas->end) {
1078                         to_feed = cas->end - cas->ptr;
1079                         if (stream_filter(cas->two,
1080                                           cas->buf + cas->ptr, &to_feed,
1081                                           output + filled, &remaining))
1082                                 return -1;
1083                         cas->ptr += (cas->end - cas->ptr) - to_feed;
1084                         filled = sz - remaining;
1085                         continue;
1086                 }
1087
1088                 /* feed one from upstream and have it emit into our buffer */
1089                 to_feed = input ? *isize_p : 0;
1090                 if (input && !to_feed)
1091                         break;
1092                 remaining = sizeof(cas->buf);
1093                 if (stream_filter(cas->one,
1094                                   input, &to_feed,
1095                                   cas->buf, &remaining))
1096                         return -1;
1097                 cas->end = sizeof(cas->buf) - remaining;
1098                 cas->ptr = 0;
1099                 if (input) {
1100                         size_t fed = *isize_p - to_feed;
1101                         *isize_p -= fed;
1102                         input += fed;
1103                 }
1104
1105                 /* do we know that we drained one completely? */
1106                 if (input || cas->end)
1107                         continue;
1108
1109                 /* tell two to drain; we have nothing more to give it */
1110                 to_feed = 0;
1111                 remaining = sz - filled;
1112                 if (stream_filter(cas->two,
1113                                   NULL, &to_feed,
1114                                   output + filled, &remaining))
1115                         return -1;
1116                 if (remaining == (sz - filled))
1117                         break; /* completely drained two */
1118                 filled = sz - remaining;
1119         }
1120         *osize_p -= filled;
1121         return 0;
1122 }
1123
1124 static void cascade_free_fn(struct stream_filter *filter)
1125 {
1126         struct cascade_filter *cas = (struct cascade_filter *)filter;
1127         free_stream_filter(cas->one);
1128         free_stream_filter(cas->two);
1129         free(filter);
1130 }
1131
1132 static struct stream_filter_vtbl cascade_vtbl = {
1133         cascade_filter_fn,
1134         cascade_free_fn,
1135 };
1136
1137 static struct stream_filter *cascade_filter(struct stream_filter *one,
1138                                             struct stream_filter *two)
1139 {
1140         struct cascade_filter *cascade;
1141
1142         if (!one || is_null_stream_filter(one))
1143                 return two;
1144         if (!two || is_null_stream_filter(two))
1145                 return one;
1146
1147         cascade = xmalloc(sizeof(*cascade));
1148         cascade->one = one;
1149         cascade->two = two;
1150         cascade->end = cascade->ptr = 0;
1151         cascade->filter.vtbl = &cascade_vtbl;
1152         return (struct stream_filter *)cascade;
1153 }
1154
1155 /*
1156  * ident filter
1157  */
1158 #define IDENT_DRAINING (-1)
1159 #define IDENT_SKIPPING (-2)
1160 struct ident_filter {
1161         struct stream_filter filter;
1162         struct strbuf left;
1163         int state;
1164         char ident[45]; /* ": x40 $" */
1165 };
1166
1167 static int is_foreign_ident(const char *str)
1168 {
1169         int i;
1170
1171         if (!skip_prefix(str, "$Id: ", &str))
1172                 return 0;
1173         for (i = 0; str[i]; i++) {
1174                 if (isspace(str[i]) && str[i+1] != '$')
1175                         return 1;
1176         }
1177         return 0;
1178 }
1179
1180 static void ident_drain(struct ident_filter *ident, char **output_p, size_t *osize_p)
1181 {
1182         size_t to_drain = ident->left.len;
1183
1184         if (*osize_p < to_drain)
1185                 to_drain = *osize_p;
1186         if (to_drain) {
1187                 memcpy(*output_p, ident->left.buf, to_drain);
1188                 strbuf_remove(&ident->left, 0, to_drain);
1189                 *output_p += to_drain;
1190                 *osize_p -= to_drain;
1191         }
1192         if (!ident->left.len)
1193                 ident->state = 0;
1194 }
1195
1196 static int ident_filter_fn(struct stream_filter *filter,
1197                            const char *input, size_t *isize_p,
1198                            char *output, size_t *osize_p)
1199 {
1200         struct ident_filter *ident = (struct ident_filter *)filter;
1201         static const char head[] = "$Id";
1202
1203         if (!input) {
1204                 /* drain upon eof */
1205                 switch (ident->state) {
1206                 default:
1207                         strbuf_add(&ident->left, head, ident->state);
1208                 case IDENT_SKIPPING:
1209                         /* fallthru */
1210                 case IDENT_DRAINING:
1211                         ident_drain(ident, &output, osize_p);
1212                 }
1213                 return 0;
1214         }
1215
1216         while (*isize_p || (ident->state == IDENT_DRAINING)) {
1217                 int ch;
1218
1219                 if (ident->state == IDENT_DRAINING) {
1220                         ident_drain(ident, &output, osize_p);
1221                         if (!*osize_p)
1222                                 break;
1223                         continue;
1224                 }
1225
1226                 ch = *(input++);
1227                 (*isize_p)--;
1228
1229                 if (ident->state == IDENT_SKIPPING) {
1230                         /*
1231                          * Skipping until '$' or LF, but keeping them
1232                          * in case it is a foreign ident.
1233                          */
1234                         strbuf_addch(&ident->left, ch);
1235                         if (ch != '\n' && ch != '$')
1236                                 continue;
1237                         if (ch == '$' && !is_foreign_ident(ident->left.buf)) {
1238                                 strbuf_setlen(&ident->left, sizeof(head) - 1);
1239                                 strbuf_addstr(&ident->left, ident->ident);
1240                         }
1241                         ident->state = IDENT_DRAINING;
1242                         continue;
1243                 }
1244
1245                 if (ident->state < sizeof(head) &&
1246                     head[ident->state] == ch) {
1247                         ident->state++;
1248                         continue;
1249                 }
1250
1251                 if (ident->state)
1252                         strbuf_add(&ident->left, head, ident->state);
1253                 if (ident->state == sizeof(head) - 1) {
1254                         if (ch != ':' && ch != '$') {
1255                                 strbuf_addch(&ident->left, ch);
1256                                 ident->state = 0;
1257                                 continue;
1258                         }
1259
1260                         if (ch == ':') {
1261                                 strbuf_addch(&ident->left, ch);
1262                                 ident->state = IDENT_SKIPPING;
1263                         } else {
1264                                 strbuf_addstr(&ident->left, ident->ident);
1265                                 ident->state = IDENT_DRAINING;
1266                         }
1267                         continue;
1268                 }
1269
1270                 strbuf_addch(&ident->left, ch);
1271                 ident->state = IDENT_DRAINING;
1272         }
1273         return 0;
1274 }
1275
1276 static void ident_free_fn(struct stream_filter *filter)
1277 {
1278         struct ident_filter *ident = (struct ident_filter *)filter;
1279         strbuf_release(&ident->left);
1280         free(filter);
1281 }
1282
1283 static struct stream_filter_vtbl ident_vtbl = {
1284         ident_filter_fn,
1285         ident_free_fn,
1286 };
1287
1288 static struct stream_filter *ident_filter(const unsigned char *sha1)
1289 {
1290         struct ident_filter *ident = xmalloc(sizeof(*ident));
1291
1292         xsnprintf(ident->ident, sizeof(ident->ident),
1293                   ": %s $", sha1_to_hex(sha1));
1294         strbuf_init(&ident->left, 0);
1295         ident->filter.vtbl = &ident_vtbl;
1296         ident->state = 0;
1297         return (struct stream_filter *)ident;
1298 }
1299
1300 /*
1301  * Return an appropriately constructed filter for the path, or NULL if
1302  * the contents cannot be filtered without reading the whole thing
1303  * in-core.
1304  *
1305  * Note that you would be crazy to set CRLF, smuge/clean or ident to a
1306  * large binary blob you would want us not to slurp into the memory!
1307  */
1308 struct stream_filter *get_stream_filter(const char *path, const unsigned char *sha1)
1309 {
1310         struct conv_attrs ca;
1311         enum crlf_action crlf_action;
1312         struct stream_filter *filter = NULL;
1313
1314         convert_attrs(&ca, path);
1315
1316         if (ca.drv && (ca.drv->smudge || ca.drv->clean))
1317                 return filter;
1318
1319         if (ca.ident)
1320                 filter = ident_filter(sha1);
1321
1322         crlf_action = input_crlf_action(ca.crlf_action, ca.eol_attr);
1323
1324         if ((crlf_action == CRLF_BINARY) || (crlf_action == CRLF_INPUT) ||
1325             (crlf_action == CRLF_GUESS && auto_crlf == AUTO_CRLF_FALSE))
1326                 filter = cascade_filter(filter, &null_filter_singleton);
1327
1328         else if (output_eol(crlf_action) == EOL_CRLF &&
1329                  !(crlf_action == CRLF_AUTO || crlf_action == CRLF_GUESS))
1330                 filter = cascade_filter(filter, lf_to_crlf_filter());
1331
1332         return filter;
1333 }
1334
1335 void free_stream_filter(struct stream_filter *filter)
1336 {
1337         filter->vtbl->free(filter);
1338 }
1339
1340 int stream_filter(struct stream_filter *filter,
1341                   const char *input, size_t *isize_p,
1342                   char *output, size_t *osize_p)
1343 {
1344         return filter->vtbl->filter(filter, input, isize_p, output, osize_p);
1345 }