git.oblomov.eu Git - git/blob - convert.c

   1 #include "cache.h"
   2 #include "attr.h"
   3 #include "run-command.h"
   4 #include "quote.h"
   5 #include "sigchain.h"
   6 #include "pkt-line.h"
   7 #include "sub-process.h"
   8
   9 /*
  10  * convert.c - convert a file when checking it out and checking it in.
  11  *
  12  * This should use the pathname to decide on whether it wants to do some
  13  * more interesting conversions (automatic gzip/unzip, general format
  14  * conversions etc etc), but by default it just does automatic CRLF<->LF
  15  * translation when the "text" attribute or "auto_crlf" option is set.
  16  */
  17
  18 /* Stat bits: When BIN is set, the txt bits are unset */
  19 #define CONVERT_STAT_BITS_TXT_LF    0x1
  20 #define CONVERT_STAT_BITS_TXT_CRLF  0x2
  21 #define CONVERT_STAT_BITS_BIN       0x4
  22
  23 enum crlf_action {
  24         CRLF_UNDEFINED,
  25         CRLF_BINARY,
  26         CRLF_TEXT,
  27         CRLF_TEXT_INPUT,
  28         CRLF_TEXT_CRLF,
  29         CRLF_AUTO,
  30         CRLF_AUTO_INPUT,
  31         CRLF_AUTO_CRLF
  32 };
  33
  34 struct text_stat {
  35         /* NUL, CR, LF and CRLF counts */
  36         unsigned nul, lonecr, lonelf, crlf;
  37
  38         /* These are just approximations! */
  39         unsigned printable, nonprintable;
  40 };
  41
  42 static void gather_stats(const char *buf, unsigned long size, struct text_stat *stats)
  43 {
  44         unsigned long i;
  45
  46         memset(stats, 0, sizeof(*stats));
  47
  48         for (i = 0; i < size; i++) {
  49                 unsigned char c = buf[i];
  50                 if (c == '\r') {
  51                         if (i+1 < size && buf[i+1] == '\n') {
  52                                 stats->crlf++;
  53                                 i++;
  54                         } else
  55                                 stats->lonecr++;
  56                         continue;
  57                 }
  58                 if (c == '\n') {
  59                         stats->lonelf++;
  60                         continue;
  61                 }
  62                 if (c == 127)
  63                         /* DEL */
  64                         stats->nonprintable++;
  65                 else if (c < 32) {
  66                         switch (c) {
  67                                 /* BS, HT, ESC and FF */
  68                         case '\b': case '\t': case '\033': case '\014':
  69                                 stats->printable++;
  70                                 break;
  71                         case 0:
  72                                 stats->nul++;
  73                                 /* fall through */
  74                         default:
  75                                 stats->nonprintable++;
  76                         }
  77                 }
  78                 else
  79                         stats->printable++;
  80         }
  81
  82         /* If file ends with EOF then don't count this EOF as non-printable. */
  83         if (size >= 1 && buf[size-1] == '\032')
  84                 stats->nonprintable--;
  85 }
  86
  87 /*
  88  * The same heuristics as diff.c::mmfile_is_binary()
  89  * We treat files with bare CR as binary
  90  */
  91 static int convert_is_binary(unsigned long size, const struct text_stat *stats)
  92 {
  93         if (stats->lonecr)
  94                 return 1;
  95         if (stats->nul)
  96                 return 1;
  97         if ((stats->printable >> 7) < stats->nonprintable)
  98                 return 1;
  99         return 0;
 100 }
 101
 102 static unsigned int gather_convert_stats(const char *data, unsigned long size)
 103 {
 104         struct text_stat stats;
 105         int ret = 0;
 106         if (!data || !size)
 107                 return 0;
 108         gather_stats(data, size, &stats);
 109         if (convert_is_binary(size, &stats))
 110                 ret |= CONVERT_STAT_BITS_BIN;
 111         if (stats.crlf)
 112                 ret |= CONVERT_STAT_BITS_TXT_CRLF;
 113         if (stats.lonelf)
 114                 ret |=  CONVERT_STAT_BITS_TXT_LF;
 115
 116         return ret;
 117 }
 118
 119 static const char *gather_convert_stats_ascii(const char *data, unsigned long size)
 120 {
 121         unsigned int convert_stats = gather_convert_stats(data, size);
 122
 123         if (convert_stats & CONVERT_STAT_BITS_BIN)
 124                 return "-text";
 125         switch (convert_stats) {
 126         case CONVERT_STAT_BITS_TXT_LF:
 127                 return "lf";
 128         case CONVERT_STAT_BITS_TXT_CRLF:
 129                 return "crlf";
 130         case CONVERT_STAT_BITS_TXT_LF | CONVERT_STAT_BITS_TXT_CRLF:
 131                 return "mixed";
 132         default:
 133                 return "none";
 134         }
 135 }
 136
 137 const char *get_cached_convert_stats_ascii(const char *path)
 138 {
 139         const char *ret;
 140         unsigned long sz;
 141         void *data = read_blob_data_from_cache(path, &sz);
 142         ret = gather_convert_stats_ascii(data, sz);
 143         free(data);
 144         return ret;
 145 }
 146
 147 const char *get_wt_convert_stats_ascii(const char *path)
 148 {
 149         const char *ret = "";
 150         struct strbuf sb = STRBUF_INIT;
 151         if (strbuf_read_file(&sb, path, 0) >= 0)
 152                 ret = gather_convert_stats_ascii(sb.buf, sb.len);
 153         strbuf_release(&sb);
 154         return ret;
 155 }
 156
 157 static int text_eol_is_crlf(void)
 158 {
 159         if (auto_crlf == AUTO_CRLF_TRUE)
 160                 return 1;
 161         else if (auto_crlf == AUTO_CRLF_INPUT)
 162                 return 0;
 163         if (core_eol == EOL_CRLF)
 164                 return 1;
 165         if (core_eol == EOL_UNSET && EOL_NATIVE == EOL_CRLF)
 166                 return 1;
 167         return 0;
 168 }
 169
 170 static enum eol output_eol(enum crlf_action crlf_action)
 171 {
 172         switch (crlf_action) {
 173         case CRLF_BINARY:
 174                 return EOL_UNSET;
 175         case CRLF_TEXT_CRLF:
 176                 return EOL_CRLF;
 177         case CRLF_TEXT_INPUT:
 178                 return EOL_LF;
 179         case CRLF_UNDEFINED:
 180         case CRLF_AUTO_CRLF:
 181                 return EOL_CRLF;
 182         case CRLF_AUTO_INPUT:
 183                 return EOL_LF;
 184         case CRLF_TEXT:
 185         case CRLF_AUTO:
 186                 /* fall through */
 187                 return text_eol_is_crlf() ? EOL_CRLF : EOL_LF;
 188         }
 189         warning("Illegal crlf_action %d\n", (int)crlf_action);
 190         return core_eol;
 191 }
 192
 193 static void check_safe_crlf(const char *path, enum crlf_action crlf_action,
 194                             struct text_stat *old_stats, struct text_stat *new_stats,
 195                             enum safe_crlf checksafe)
 196 {
 197         if (old_stats->crlf && !new_stats->crlf ) {
 198                 /*
 199                  * CRLFs would not be restored by checkout
 200                  */
 201                 if (checksafe == SAFE_CRLF_WARN)
 202                         warning(_("CRLF will be replaced by LF in %s.\n"
 203                                   "The file will have its original line"
 204                                   " endings in your working directory."), path);
 205                 else /* i.e. SAFE_CRLF_FAIL */
 206                         die(_("CRLF would be replaced by LF in %s."), path);
 207         } else if (old_stats->lonelf && !new_stats->lonelf ) {
 208                 /*
 209                  * CRLFs would be added by checkout
 210                  */
 211                 if (checksafe == SAFE_CRLF_WARN)
 212                         warning(_("LF will be replaced by CRLF in %s.\n"
 213                                   "The file will have its original line"
 214                                   " endings in your working directory."), path);
 215                 else /* i.e. SAFE_CRLF_FAIL */
 216                         die(_("LF would be replaced by CRLF in %s"), path);
 217         }
 218 }
 219
 220 static int has_cr_in_index(const char *path)
 221 {
 222         unsigned long sz;
 223         void *data;
 224         int has_cr;
 225
 226         data = read_blob_data_from_cache(path, &sz);
 227         if (!data)
 228                 return 0;
 229         has_cr = memchr(data, '\r', sz) != NULL;
 230         free(data);
 231         return has_cr;
 232 }
 233
 234 static int will_convert_lf_to_crlf(size_t len, struct text_stat *stats,
 235                                    enum crlf_action crlf_action)
 236 {
 237         if (output_eol(crlf_action) != EOL_CRLF)
 238                 return 0;
 239         /* No "naked" LF? Nothing to convert, regardless. */
 240         if (!stats->lonelf)
 241                 return 0;
 242
 243         if (crlf_action == CRLF_AUTO || crlf_action == CRLF_AUTO_INPUT || crlf_action == CRLF_AUTO_CRLF) {
 244                 /* If we have any CR or CRLF line endings, we do not touch it */
 245                 /* This is the new safer autocrlf-handling */
 246                 if (stats->lonecr || stats->crlf)
 247                         return 0;
 248
 249                 if (convert_is_binary(len, stats))
 250                         return 0;
 251         }
 252         return 1;
 253
 254 }
 255
 256 static int crlf_to_git(const char *path, const char *src, size_t len,
 257                        struct strbuf *buf,
 258                        enum crlf_action crlf_action, enum safe_crlf checksafe)
 259 {
 260         struct text_stat stats;
 261         char *dst;
 262         int convert_crlf_into_lf;
 263
 264         if (crlf_action == CRLF_BINARY ||
 265             (src && !len))
 266                 return 0;
 267
 268         /*
 269          * If we are doing a dry-run and have no source buffer, there is
 270          * nothing to analyze; we must assume we would convert.
 271          */
 272         if (!buf && !src)
 273                 return 1;
 274
 275         gather_stats(src, len, &stats);
 276         /* Optimization: No CRLF? Nothing to convert, regardless. */
 277         convert_crlf_into_lf = !!stats.crlf;
 278
 279         if (crlf_action == CRLF_AUTO || crlf_action == CRLF_AUTO_INPUT || crlf_action == CRLF_AUTO_CRLF) {
 280                 if (convert_is_binary(len, &stats))
 281                         return 0;
 282                 /*
 283                  * If the file in the index has any CR in it, do not
 284                  * convert.  This is the new safer autocrlf handling,
 285                  * unless we want to renormalize in a merge or
 286                  * cherry-pick.
 287                  */
 288                 if ((checksafe != SAFE_CRLF_RENORMALIZE) && has_cr_in_index(path))
 289                         convert_crlf_into_lf = 0;
 290         }
 291         if ((checksafe == SAFE_CRLF_WARN ||
 292             (checksafe == SAFE_CRLF_FAIL)) && len) {
 293                 struct text_stat new_stats;
 294                 memcpy(&new_stats, &stats, sizeof(new_stats));
 295                 /* simulate "git add" */
 296                 if (convert_crlf_into_lf) {
 297                         new_stats.lonelf += new_stats.crlf;
 298                         new_stats.crlf = 0;
 299                 }
 300                 /* simulate "git checkout" */
 301                 if (will_convert_lf_to_crlf(len, &new_stats, crlf_action)) {
 302                         new_stats.crlf += new_stats.lonelf;
 303                         new_stats.lonelf = 0;
 304                 }
 305                 check_safe_crlf(path, crlf_action, &stats, &new_stats, checksafe);
 306         }
 307         if (!convert_crlf_into_lf)
 308                 return 0;
 309
 310         /*
 311          * At this point all of our source analysis is done, and we are sure we
 312          * would convert. If we are in dry-run mode, we can give an answer.
 313          */
 314         if (!buf)
 315                 return 1;
 316
 317         /* only grow if not in place */
 318         if (strbuf_avail(buf) + buf->len < len)
 319                 strbuf_grow(buf, len - buf->len);
 320         dst = buf->buf;
 321         if (crlf_action == CRLF_AUTO || crlf_action == CRLF_AUTO_INPUT || crlf_action == CRLF_AUTO_CRLF) {
 322                 /*
 323                  * If we guessed, we already know we rejected a file with
 324                  * lone CR, and we can strip a CR without looking at what
 325                  * follow it.
 326                  */
 327                 do {
 328                         unsigned char c = *src++;
 329                         if (c != '\r')
 330                                 *dst++ = c;
 331                 } while (--len);
 332         } else {
 333                 do {
 334                         unsigned char c = *src++;
 335                         if (! (c == '\r' && (1 < len && *src == '\n')))
 336                                 *dst++ = c;
 337                 } while (--len);
 338         }
 339         strbuf_setlen(buf, dst - buf->buf);
 340         return 1;
 341 }
 342
 343 static int crlf_to_worktree(const char *path, const char *src, size_t len,
 344                             struct strbuf *buf, enum crlf_action crlf_action)
 345 {
 346         char *to_free = NULL;
 347         struct text_stat stats;
 348
 349         if (!len || output_eol(crlf_action) != EOL_CRLF)
 350                 return 0;
 351
 352         gather_stats(src, len, &stats);
 353         if (!will_convert_lf_to_crlf(len, &stats, crlf_action))
 354                 return 0;
 355
 356         /* are we "faking" in place editing ? */
 357         if (src == buf->buf)
 358                 to_free = strbuf_detach(buf, NULL);
 359
 360         strbuf_grow(buf, len + stats.lonelf);
 361         for (;;) {
 362                 const char *nl = memchr(src, '\n', len);
 363                 if (!nl)
 364                         break;
 365                 if (nl > src && nl[-1] == '\r') {
 366                         strbuf_add(buf, src, nl + 1 - src);
 367                 } else {
 368                         strbuf_add(buf, src, nl - src);
 369                         strbuf_addstr(buf, "\r\n");
 370                 }
 371                 len -= nl + 1 - src;
 372                 src  = nl + 1;
 373         }
 374         strbuf_add(buf, src, len);
 375
 376         free(to_free);
 377         return 1;
 378 }
 379
 380 struct filter_params {
 381         const char *src;
 382         unsigned long size;
 383         int fd;
 384         const char *cmd;
 385         const char *path;
 386 };
 387
 388 static int filter_buffer_or_fd(int in, int out, void *data)
 389 {
 390         /*
 391          * Spawn cmd and feed the buffer contents through its stdin.
 392          */
 393         struct child_process child_process = CHILD_PROCESS_INIT;
 394         struct filter_params *params = (struct filter_params *)data;
 395         int write_err, status;
 396         const char *argv[] = { NULL, NULL };
 397
 398         /* apply % substitution to cmd */
 399         struct strbuf cmd = STRBUF_INIT;
 400         struct strbuf path = STRBUF_INIT;
 401         struct strbuf_expand_dict_entry dict[] = {
 402                 { "f", NULL, },
 403                 { NULL, NULL, },
 404         };
 405
 406         /* quote the path to preserve spaces, etc. */
 407         sq_quote_buf(&path, params->path);
 408         dict[0].value = path.buf;
 409
 410         /* expand all %f with the quoted path */
 411         strbuf_expand(&cmd, params->cmd, strbuf_expand_dict_cb, &dict);
 412         strbuf_release(&path);
 413
 414         argv[0] = cmd.buf;
 415
 416         child_process.argv = argv;
 417         child_process.use_shell = 1;
 418         child_process.in = -1;
 419         child_process.out = out;
 420
 421         if (start_command(&child_process))
 422                 return error("cannot fork to run external filter '%s'", params->cmd);
 423
 424         sigchain_push(SIGPIPE, SIG_IGN);
 425
 426         if (params->src) {
 427                 write_err = (write_in_full(child_process.in,
 428                                            params->src, params->size) < 0);
 429                 if (errno == EPIPE)
 430                         write_err = 0;
 431         } else {
 432                 write_err = copy_fd(params->fd, child_process.in);
 433                 if (write_err == COPY_WRITE_ERROR && errno == EPIPE)
 434                         write_err = 0;
 435         }
 436
 437         if (close(child_process.in))
 438                 write_err = 1;
 439         if (write_err)
 440                 error("cannot feed the input to external filter '%s'", params->cmd);
 441
 442         sigchain_pop(SIGPIPE);
 443
 444         status = finish_command(&child_process);
 445         if (status)
 446                 error("external filter '%s' failed %d", params->cmd, status);
 447
 448         strbuf_release(&cmd);
 449         return (write_err || status);
 450 }
 451
 452 static int apply_single_file_filter(const char *path, const char *src, size_t len, int fd,
 453                         struct strbuf *dst, const char *cmd)
 454 {
 455         /*
 456          * Create a pipeline to have the command filter the buffer's
 457          * contents.
 458          *
 459          * (child --> cmd) --> us
 460          */
 461         int err = 0;
 462         struct strbuf nbuf = STRBUF_INIT;
 463         struct async async;
 464         struct filter_params params;
 465
 466         memset(&async, 0, sizeof(async));
 467         async.proc = filter_buffer_or_fd;
 468         async.data = &params;
 469         async.out = -1;
 470         params.src = src;
 471         params.size = len;
 472         params.fd = fd;
 473         params.cmd = cmd;
 474         params.path = path;
 475
 476         fflush(NULL);
 477         if (start_async(&async))
 478                 return 0;       /* error was already reported */
 479
 480         if (strbuf_read(&nbuf, async.out, len) < 0) {
 481                 err = error("read from external filter '%s' failed", cmd);
 482         }
 483         if (close(async.out)) {
 484                 err = error("read from external filter '%s' failed", cmd);
 485         }
 486         if (finish_async(&async)) {
 487                 err = error("external filter '%s' failed", cmd);
 488         }
 489
 490         if (!err) {
 491                 strbuf_swap(dst, &nbuf);
 492         }
 493         strbuf_release(&nbuf);
 494         return !err;
 495 }
 496
 497 #define CAP_CLEAN    (1u<<0)
 498 #define CAP_SMUDGE   (1u<<1)
 499
 500 struct cmd2process {
 501         struct subprocess_entry subprocess; /* must be the first member! */
 502         unsigned int supported_capabilities;
 503 };
 504
 505 static int subprocess_map_initialized;
 506 static struct hashmap subprocess_map;
 507
 508 static int start_multi_file_filter_fn(struct subprocess_entry *subprocess)
 509 {
 510         int err;
 511         struct cmd2process *entry = (struct cmd2process *)subprocess;
 512         struct string_list cap_list = STRING_LIST_INIT_NODUP;
 513         char *cap_buf;
 514         const char *cap_name;
 515         struct child_process *process = &subprocess->process;
 516         const char *cmd = subprocess->cmd;
 517
 518         sigchain_push(SIGPIPE, SIG_IGN);
 519
 520         err = packet_writel(process->in, "git-filter-client", "version=2", NULL);
 521         if (err)
 522                 goto done;
 523
 524         err = strcmp(packet_read_line(process->out, NULL), "git-filter-server");
 525         if (err) {
 526                 error("external filter '%s' does not support filter protocol version 2", cmd);
 527                 goto done;
 528         }
 529         err = strcmp(packet_read_line(process->out, NULL), "version=2");
 530         if (err)
 531                 goto done;
 532         err = packet_read_line(process->out, NULL) != NULL;
 533         if (err)
 534                 goto done;
 535
 536         err = packet_writel(process->in, "capability=clean", "capability=smudge", NULL);
 537
 538         for (;;) {
 539                 cap_buf = packet_read_line(process->out, NULL);
 540                 if (!cap_buf)
 541                         break;
 542                 string_list_split_in_place(&cap_list, cap_buf, '=', 1);
 543
 544                 if (cap_list.nr != 2 || strcmp(cap_list.items[0].string, "capability"))
 545                         continue;
 546
 547                 cap_name = cap_list.items[1].string;
 548                 if (!strcmp(cap_name, "clean")) {
 549                         entry->supported_capabilities |= CAP_CLEAN;
 550                 } else if (!strcmp(cap_name, "smudge")) {
 551                         entry->supported_capabilities |= CAP_SMUDGE;
 552                 } else {
 553                         warning(
 554                                 "external filter '%s' requested unsupported filter capability '%s'",
 555                                 cmd, cap_name
 556                         );
 557                 }
 558
 559                 string_list_clear(&cap_list, 0);
 560         }
 561
 562 done:
 563         sigchain_pop(SIGPIPE);
 564
 565         return err;
 566 }
 567
 568 static int apply_multi_file_filter(const char *path, const char *src, size_t len,
 569                                    int fd, struct strbuf *dst, const char *cmd,
 570                                    const unsigned int wanted_capability)
 571 {
 572         int err;
 573         struct cmd2process *entry;
 574         struct child_process *process;
 575         struct strbuf nbuf = STRBUF_INIT;
 576         struct strbuf filter_status = STRBUF_INIT;
 577         const char *filter_type;
 578
 579         if (!subprocess_map_initialized) {
 580                 subprocess_map_initialized = 1;
 581                 hashmap_init(&subprocess_map, (hashmap_cmp_fn) cmd2process_cmp, 0);
 582                 entry = NULL;
 583         } else {
 584                 entry = (struct cmd2process *)subprocess_find_entry(&subprocess_map, cmd);
 585         }
 586
 587         fflush(NULL);
 588
 589         if (!entry) {
 590                 entry = xmalloc(sizeof(*entry));
 591                 entry->supported_capabilities = 0;
 592
 593                 if (subprocess_start(&subprocess_map, &entry->subprocess, cmd, start_multi_file_filter_fn)) {
 594                         free(entry);
 595                         return 0;
 596                 }
 597         }
 598         process = &entry->subprocess.process;
 599
 600         if (!(wanted_capability & entry->supported_capabilities))
 601                 return 0;
 602
 603         if (CAP_CLEAN & wanted_capability)
 604                 filter_type = "clean";
 605         else if (CAP_SMUDGE & wanted_capability)
 606                 filter_type = "smudge";
 607         else
 608                 die("unexpected filter type");
 609
 610         sigchain_push(SIGPIPE, SIG_IGN);
 611
 612         assert(strlen(filter_type) < LARGE_PACKET_DATA_MAX - strlen("command=\n"));
 613         err = packet_write_fmt_gently(process->in, "command=%s\n", filter_type);
 614         if (err)
 615                 goto done;
 616
 617         err = strlen(path) > LARGE_PACKET_DATA_MAX - strlen("pathname=\n");
 618         if (err) {
 619                 error("path name too long for external filter");
 620                 goto done;
 621         }
 622
 623         err = packet_write_fmt_gently(process->in, "pathname=%s\n", path);
 624         if (err)
 625                 goto done;
 626
 627         err = packet_flush_gently(process->in);
 628         if (err)
 629                 goto done;
 630
 631         if (fd >= 0)
 632                 err = write_packetized_from_fd(fd, process->in);
 633         else
 634                 err = write_packetized_from_buf(src, len, process->in);
 635         if (err)
 636                 goto done;
 637
 638         subprocess_read_status(process->out, &filter_status);
 639         err = strcmp(filter_status.buf, "success");
 640         if (err)
 641                 goto done;
 642
 643         err = read_packetized_to_strbuf(process->out, &nbuf) < 0;
 644         if (err)
 645                 goto done;
 646
 647         subprocess_read_status(process->out, &filter_status);
 648         err = strcmp(filter_status.buf, "success");
 649
 650 done:
 651         sigchain_pop(SIGPIPE);
 652
 653         if (err) {
 654                 if (!strcmp(filter_status.buf, "error")) {
 655                         /* The filter signaled a problem with the file. */
 656                 } else if (!strcmp(filter_status.buf, "abort")) {
 657                         /*
 658                          * The filter signaled a permanent problem. Don't try to filter
 659                          * files with the same command for the lifetime of the current
 660                          * Git process.
 661                          */
 662                          entry->supported_capabilities &= ~wanted_capability;
 663                 } else {
 664                         /*
 665                          * Something went wrong with the protocol filter.
 666                          * Force shutdown and restart if another blob requires filtering.
 667                          */
 668                         error("external filter '%s' failed", cmd);
 669                         subprocess_stop(&subprocess_map, &entry->subprocess);
 670                         free(entry);
 671                 }
 672         } else {
 673                 strbuf_swap(dst, &nbuf);
 674         }
 675         strbuf_release(&nbuf);
 676         return !err;
 677 }
 678
 679 static struct convert_driver {
 680         const char *name;
 681         struct convert_driver *next;
 682         const char *smudge;
 683         const char *clean;
 684         const char *process;
 685         int required;
 686 } *user_convert, **user_convert_tail;
 687
 688 static int apply_filter(const char *path, const char *src, size_t len,
 689                         int fd, struct strbuf *dst, struct convert_driver *drv,
 690                         const unsigned int wanted_capability)
 691 {
 692         const char *cmd = NULL;
 693
 694         if (!drv)
 695                 return 0;
 696
 697         if (!dst)
 698                 return 1;
 699
 700         if ((CAP_CLEAN & wanted_capability) && !drv->process && drv->clean)
 701                 cmd = drv->clean;
 702         else if ((CAP_SMUDGE & wanted_capability) && !drv->process && drv->smudge)
 703                 cmd = drv->smudge;
 704
 705         if (cmd && *cmd)
 706                 return apply_single_file_filter(path, src, len, fd, dst, cmd);
 707         else if (drv->process && *drv->process)
 708                 return apply_multi_file_filter(path, src, len, fd, dst, drv->process, wanted_capability);
 709
 710         return 0;
 711 }
 712
 713 static int read_convert_config(const char *var, const char *value, void *cb)
 714 {
 715         const char *key, *name;
 716         int namelen;
 717         struct convert_driver *drv;
 718
 719         /*
 720          * External conversion drivers are configured using
 721          * "filter.<name>.variable".
 722          */
 723         if (parse_config_key(var, "filter", &name, &namelen, &key) < 0 || !name)
 724                 return 0;
 725         for (drv = user_convert; drv; drv = drv->next)
 726                 if (!strncmp(drv->name, name, namelen) && !drv->name[namelen])
 727                         break;
 728         if (!drv) {
 729                 drv = xcalloc(1, sizeof(struct convert_driver));
 730                 drv->name = xmemdupz(name, namelen);
 731                 *user_convert_tail = drv;
 732                 user_convert_tail = &(drv->next);
 733         }
 734
 735         /*
 736          * filter.<name>.smudge and filter.<name>.clean specifies
 737          * the command line:
 738          *
 739          *      command-line
 740          *
 741          * The command-line will not be interpolated in any way.
 742          */
 743
 744         if (!strcmp("smudge", key))
 745                 return git_config_string(&drv->smudge, var, value);
 746
 747         if (!strcmp("clean", key))
 748                 return git_config_string(&drv->clean, var, value);
 749
 750         if (!strcmp("process", key))
 751                 return git_config_string(&drv->process, var, value);
 752
 753         if (!strcmp("required", key)) {
 754                 drv->required = git_config_bool(var, value);
 755                 return 0;
 756         }
 757
 758         return 0;
 759 }
 760
 761 static int count_ident(const char *cp, unsigned long size)
 762 {
 763         /*
 764          * "$Id: 0000000000000000000000000000000000000000 $" <=> "$Id$"
 765          */
 766         int cnt = 0;
 767         char ch;
 768
 769         while (size) {
 770                 ch = *cp++;
 771                 size--;
 772                 if (ch != '$')
 773                         continue;
 774                 if (size < 3)
 775                         break;
 776                 if (memcmp("Id", cp, 2))
 777                         continue;
 778                 ch = cp[2];
 779                 cp += 3;
 780                 size -= 3;
 781                 if (ch == '$')
 782                         cnt++; /* $Id$ */
 783                 if (ch != ':')
 784                         continue;
 785
 786                 /*
 787                  * "$Id: ... "; scan up to the closing dollar sign and discard.
 788                  */
 789                 while (size) {
 790                         ch = *cp++;
 791                         size--;
 792                         if (ch == '$') {
 793                                 cnt++;
 794                                 break;
 795                         }
 796                         if (ch == '\n')
 797                                 break;
 798                 }
 799         }
 800         return cnt;
 801 }
 802
 803 static int ident_to_git(const char *path, const char *src, size_t len,
 804                         struct strbuf *buf, int ident)
 805 {
 806         char *dst, *dollar;
 807
 808         if (!ident || (src && !count_ident(src, len)))
 809                 return 0;
 810
 811         if (!buf)
 812                 return 1;
 813
 814         /* only grow if not in place */
 815         if (strbuf_avail(buf) + buf->len < len)
 816                 strbuf_grow(buf, len - buf->len);
 817         dst = buf->buf;
 818         for (;;) {
 819                 dollar = memchr(src, '$', len);
 820                 if (!dollar)
 821                         break;
 822                 memmove(dst, src, dollar + 1 - src);
 823                 dst += dollar + 1 - src;
 824                 len -= dollar + 1 - src;
 825                 src  = dollar + 1;
 826
 827                 if (len > 3 && !memcmp(src, "Id:", 3)) {
 828                         dollar = memchr(src + 3, '$', len - 3);
 829                         if (!dollar)
 830                                 break;
 831                         if (memchr(src + 3, '\n', dollar - src - 3)) {
 832                                 /* Line break before the next dollar. */
 833                                 continue;
 834                         }
 835
 836                         memcpy(dst, "Id$", 3);
 837                         dst += 3;
 838                         len -= dollar + 1 - src;
 839                         src  = dollar + 1;
 840                 }
 841         }
 842         memmove(dst, src, len);
 843         strbuf_setlen(buf, dst + len - buf->buf);
 844         return 1;
 845 }
 846
 847 static int ident_to_worktree(const char *path, const char *src, size_t len,
 848                              struct strbuf *buf, int ident)
 849 {
 850         unsigned char sha1[20];
 851         char *to_free = NULL, *dollar, *spc;
 852         int cnt;
 853
 854         if (!ident)
 855                 return 0;
 856
 857         cnt = count_ident(src, len);
 858         if (!cnt)
 859                 return 0;
 860
 861         /* are we "faking" in place editing ? */
 862         if (src == buf->buf)
 863                 to_free = strbuf_detach(buf, NULL);
 864         hash_sha1_file(src, len, "blob", sha1);
 865
 866         strbuf_grow(buf, len + cnt * 43);
 867         for (;;) {
 868                 /* step 1: run to the next '$' */
 869                 dollar = memchr(src, '$', len);
 870                 if (!dollar)
 871                         break;
 872                 strbuf_add(buf, src, dollar + 1 - src);
 873                 len -= dollar + 1 - src;
 874                 src  = dollar + 1;
 875
 876                 /* step 2: does it looks like a bit like Id:xxx$ or Id$ ? */
 877                 if (len < 3 || memcmp("Id", src, 2))
 878                         continue;
 879
 880                 /* step 3: skip over Id$ or Id:xxxxx$ */
 881                 if (src[2] == '$') {
 882                         src += 3;
 883                         len -= 3;
 884                 } else if (src[2] == ':') {
 885                         /*
 886                          * It's possible that an expanded Id has crept its way into the
 887                          * repository, we cope with that by stripping the expansion out.
 888                          * This is probably not a good idea, since it will cause changes
 889                          * on checkout, which won't go away by stash, but let's keep it
 890                          * for git-style ids.
 891                          */
 892                         dollar = memchr(src + 3, '$', len - 3);
 893                         if (!dollar) {
 894                                 /* incomplete keyword, no more '$', so just quit the loop */
 895                                 break;
 896                         }
 897
 898                         if (memchr(src + 3, '\n', dollar - src - 3)) {
 899                                 /* Line break before the next dollar. */
 900                                 continue;
 901                         }
 902
 903                         spc = memchr(src + 4, ' ', dollar - src - 4);
 904                         if (spc && spc < dollar-1) {
 905                                 /* There are spaces in unexpected places.
 906                                  * This is probably an id from some other
 907                                  * versioning system. Keep it for now.
 908                                  */
 909                                 continue;
 910                         }
 911
 912                         len -= dollar + 1 - src;
 913                         src  = dollar + 1;
 914                 } else {
 915                         /* it wasn't a "Id$" or "Id:xxxx$" */
 916                         continue;
 917                 }
 918
 919                 /* step 4: substitute */
 920                 strbuf_addstr(buf, "Id: ");
 921                 strbuf_add(buf, sha1_to_hex(sha1), 40);
 922                 strbuf_addstr(buf, " $");
 923         }
 924         strbuf_add(buf, src, len);
 925
 926         free(to_free);
 927         return 1;
 928 }
 929
 930 static enum crlf_action git_path_check_crlf(struct attr_check_item *check)
 931 {
 932         const char *value = check->value;
 933
 934         if (ATTR_TRUE(value))
 935                 return CRLF_TEXT;
 936         else if (ATTR_FALSE(value))
 937                 return CRLF_BINARY;
 938         else if (ATTR_UNSET(value))
 939                 ;
 940         else if (!strcmp(value, "input"))
 941                 return CRLF_TEXT_INPUT;
 942         else if (!strcmp(value, "auto"))
 943                 return CRLF_AUTO;
 944         return CRLF_UNDEFINED;
 945 }
 946
 947 static enum eol git_path_check_eol(struct attr_check_item *check)
 948 {
 949         const char *value = check->value;
 950
 951         if (ATTR_UNSET(value))
 952                 ;
 953         else if (!strcmp(value, "lf"))
 954                 return EOL_LF;
 955         else if (!strcmp(value, "crlf"))
 956                 return EOL_CRLF;
 957         return EOL_UNSET;
 958 }
 959
 960 static struct convert_driver *git_path_check_convert(struct attr_check_item *check)
 961 {
 962         const char *value = check->value;
 963         struct convert_driver *drv;
 964
 965         if (ATTR_TRUE(value) || ATTR_FALSE(value) || ATTR_UNSET(value))
 966                 return NULL;
 967         for (drv = user_convert; drv; drv = drv->next)
 968                 if (!strcmp(value, drv->name))
 969                         return drv;
 970         return NULL;
 971 }
 972
 973 static int git_path_check_ident(struct attr_check_item *check)
 974 {
 975         const char *value = check->value;
 976
 977         return !!ATTR_TRUE(value);
 978 }
 979
 980 struct conv_attrs {
 981         struct convert_driver *drv;
 982         enum crlf_action attr_action; /* What attr says */
 983         enum crlf_action crlf_action; /* When no attr is set, use core.autocrlf */
 984         int ident;
 985 };
 986
 987 static void convert_attrs(struct conv_attrs *ca, const char *path)
 988 {
 989         static struct attr_check *check;
 990
 991         if (!check) {
 992                 check = attr_check_initl("crlf", "ident", "filter",
 993                                          "eol", "text", NULL);
 994                 user_convert_tail = &user_convert;
 995                 git_config(read_convert_config, NULL);
 996         }
 997
 998         if (!git_check_attr(path, check)) {
 999                 struct attr_check_item *ccheck = check->items;
1000                 ca->crlf_action = git_path_check_crlf(ccheck + 4);
1001                 if (ca->crlf_action == CRLF_UNDEFINED)
1002                         ca->crlf_action = git_path_check_crlf(ccheck + 0);
1003                 ca->attr_action = ca->crlf_action;
1004                 ca->ident = git_path_check_ident(ccheck + 1);
1005                 ca->drv = git_path_check_convert(ccheck + 2);
1006                 if (ca->crlf_action != CRLF_BINARY) {
1007                         enum eol eol_attr = git_path_check_eol(ccheck + 3);
1008                         if (ca->crlf_action == CRLF_AUTO && eol_attr == EOL_LF)
1009                                 ca->crlf_action = CRLF_AUTO_INPUT;
1010                         else if (ca->crlf_action == CRLF_AUTO && eol_attr == EOL_CRLF)
1011                                 ca->crlf_action = CRLF_AUTO_CRLF;
1012                         else if (eol_attr == EOL_LF)
1013                                 ca->crlf_action = CRLF_TEXT_INPUT;
1014                         else if (eol_attr == EOL_CRLF)
1015                                 ca->crlf_action = CRLF_TEXT_CRLF;
1016                 }
1017                 ca->attr_action = ca->crlf_action;
1018         } else {
1019                 ca->drv = NULL;
1020                 ca->crlf_action = CRLF_UNDEFINED;
1021                 ca->ident = 0;
1022         }
1023         if (ca->crlf_action == CRLF_TEXT)
1024                 ca->crlf_action = text_eol_is_crlf() ? CRLF_TEXT_CRLF : CRLF_TEXT_INPUT;
1025         if (ca->crlf_action == CRLF_UNDEFINED && auto_crlf == AUTO_CRLF_FALSE)
1026                 ca->crlf_action = CRLF_BINARY;
1027         if (ca->crlf_action == CRLF_UNDEFINED && auto_crlf == AUTO_CRLF_TRUE)
1028                 ca->crlf_action = CRLF_AUTO_CRLF;
1029         if (ca->crlf_action == CRLF_UNDEFINED && auto_crlf == AUTO_CRLF_INPUT)
1030                 ca->crlf_action = CRLF_AUTO_INPUT;
1031 }
1032
1033 int would_convert_to_git_filter_fd(const char *path)
1034 {
1035         struct conv_attrs ca;
1036
1037         convert_attrs(&ca, path);
1038         if (!ca.drv)
1039                 return 0;
1040
1041         /*
1042          * Apply a filter to an fd only if the filter is required to succeed.
1043          * We must die if the filter fails, because the original data before
1044          * filtering is not available.
1045          */
1046         if (!ca.drv->required)
1047                 return 0;
1048
1049         return apply_filter(path, NULL, 0, -1, NULL, ca.drv, CAP_CLEAN);
1050 }
1051
1052 const char *get_convert_attr_ascii(const char *path)
1053 {
1054         struct conv_attrs ca;
1055
1056         convert_attrs(&ca, path);
1057         switch (ca.attr_action) {
1058         case CRLF_UNDEFINED:
1059                 return "";
1060         case CRLF_BINARY:
1061                 return "-text";
1062         case CRLF_TEXT:
1063                 return "text";
1064         case CRLF_TEXT_INPUT:
1065                 return "text eol=lf";
1066         case CRLF_TEXT_CRLF:
1067                 return "text eol=crlf";
1068         case CRLF_AUTO:
1069                 return "text=auto";
1070         case CRLF_AUTO_CRLF:
1071                 return "text=auto eol=crlf";
1072         case CRLF_AUTO_INPUT:
1073                 return "text=auto eol=lf";
1074         }
1075         return "";
1076 }
1077
1078 int convert_to_git(const char *path, const char *src, size_t len,
1079                    struct strbuf *dst, enum safe_crlf checksafe)
1080 {
1081         int ret = 0;
1082         struct conv_attrs ca;
1083
1084         convert_attrs(&ca, path);
1085
1086         ret |= apply_filter(path, src, len, -1, dst, ca.drv, CAP_CLEAN);
1087         if (!ret && ca.drv && ca.drv->required)
1088                 die("%s: clean filter '%s' failed", path, ca.drv->name);
1089
1090         if (ret && dst) {
1091                 src = dst->buf;
1092                 len = dst->len;
1093         }
1094         ret |= crlf_to_git(path, src, len, dst, ca.crlf_action, checksafe);
1095         if (ret && dst) {
1096                 src = dst->buf;
1097                 len = dst->len;
1098         }
1099         return ret | ident_to_git(path, src, len, dst, ca.ident);
1100 }
1101
1102 void convert_to_git_filter_fd(const char *path, int fd, struct strbuf *dst,
1103                               enum safe_crlf checksafe)
1104 {
1105         struct conv_attrs ca;
1106         convert_attrs(&ca, path);
1107
1108         assert(ca.drv);
1109         assert(ca.drv->clean || ca.drv->process);
1110
1111         if (!apply_filter(path, NULL, 0, fd, dst, ca.drv, CAP_CLEAN))
1112                 die("%s: clean filter '%s' failed", path, ca.drv->name);
1113
1114         crlf_to_git(path, dst->buf, dst->len, dst, ca.crlf_action, checksafe);
1115         ident_to_git(path, dst->buf, dst->len, dst, ca.ident);
1116 }
1117
1118 static int convert_to_working_tree_internal(const char *path, const char *src,
1119                                             size_t len, struct strbuf *dst,
1120                                             int normalizing)
1121 {
1122         int ret = 0, ret_filter = 0;
1123         struct conv_attrs ca;
1124
1125         convert_attrs(&ca, path);
1126
1127         ret |= ident_to_worktree(path, src, len, dst, ca.ident);
1128         if (ret) {
1129                 src = dst->buf;
1130                 len = dst->len;
1131         }
1132         /*
1133          * CRLF conversion can be skipped if normalizing, unless there
1134          * is a smudge or process filter (even if the process filter doesn't
1135          * support smudge).  The filters might expect CRLFs.
1136          */
1137         if ((ca.drv && (ca.drv->smudge || ca.drv->process)) || !normalizing) {
1138                 ret |= crlf_to_worktree(path, src, len, dst, ca.crlf_action);
1139                 if (ret) {
1140                         src = dst->buf;
1141                         len = dst->len;
1142                 }
1143         }
1144
1145         ret_filter = apply_filter(path, src, len, -1, dst, ca.drv, CAP_SMUDGE);
1146         if (!ret_filter && ca.drv && ca.drv->required)
1147                 die("%s: smudge filter %s failed", path, ca.drv->name);
1148
1149         return ret | ret_filter;
1150 }
1151
1152 int convert_to_working_tree(const char *path, const char *src, size_t len, struct strbuf *dst)
1153 {
1154         return convert_to_working_tree_internal(path, src, len, dst, 0);
1155 }
1156
1157 int renormalize_buffer(const char *path, const char *src, size_t len, struct strbuf *dst)
1158 {
1159         int ret = convert_to_working_tree_internal(path, src, len, dst, 1);
1160         if (ret) {
1161                 src = dst->buf;
1162                 len = dst->len;
1163         }
1164         return ret | convert_to_git(path, src, len, dst, SAFE_CRLF_RENORMALIZE);
1165 }
1166
1167 /*****************************************************************
1168  *
1169  * Streaming conversion support
1170  *
1171  *****************************************************************/
1172
1173 typedef int (*filter_fn)(struct stream_filter *,
1174                          const char *input, size_t *isize_p,
1175                          char *output, size_t *osize_p);
1176 typedef void (*free_fn)(struct stream_filter *);
1177
1178 struct stream_filter_vtbl {
1179         filter_fn filter;
1180         free_fn free;
1181 };
1182
1183 struct stream_filter {
1184         struct stream_filter_vtbl *vtbl;
1185 };
1186
1187 static int null_filter_fn(struct stream_filter *filter,
1188                           const char *input, size_t *isize_p,
1189                           char *output, size_t *osize_p)
1190 {
1191         size_t count;
1192
1193         if (!input)
1194                 return 0; /* we do not keep any states */
1195         count = *isize_p;
1196         if (*osize_p < count)
1197                 count = *osize_p;
1198         if (count) {
1199                 memmove(output, input, count);
1200                 *isize_p -= count;
1201                 *osize_p -= count;
1202         }
1203         return 0;
1204 }
1205
1206 static void null_free_fn(struct stream_filter *filter)
1207 {
1208         ; /* nothing -- null instances are shared */
1209 }
1210
1211 static struct stream_filter_vtbl null_vtbl = {
1212         null_filter_fn,
1213         null_free_fn,
1214 };
1215
1216 static struct stream_filter null_filter_singleton = {
1217         &null_vtbl,
1218 };
1219
1220 int is_null_stream_filter(struct stream_filter *filter)
1221 {
1222         return filter == &null_filter_singleton;
1223 }
1224
1225
1226 /*
1227  * LF-to-CRLF filter
1228  */
1229
1230 struct lf_to_crlf_filter {
1231         struct stream_filter filter;
1232         unsigned has_held:1;
1233         char held;
1234 };
1235
1236 static int lf_to_crlf_filter_fn(struct stream_filter *filter,
1237                                 const char *input, size_t *isize_p,
1238                                 char *output, size_t *osize_p)
1239 {
1240         size_t count, o = 0;
1241         struct lf_to_crlf_filter *lf_to_crlf = (struct lf_to_crlf_filter *)filter;
1242
1243         /*
1244          * We may be holding onto the CR to see if it is followed by a
1245          * LF, in which case we would need to go to the main loop.
1246          * Otherwise, just emit it to the output stream.
1247          */
1248         if (lf_to_crlf->has_held && (lf_to_crlf->held != '\r' || !input)) {
1249                 output[o++] = lf_to_crlf->held;
1250                 lf_to_crlf->has_held = 0;
1251         }
1252
1253         /* We are told to drain */
1254         if (!input) {
1255                 *osize_p -= o;
1256                 return 0;
1257         }
1258
1259         count = *isize_p;
1260         if (count || lf_to_crlf->has_held) {
1261                 size_t i;
1262                 int was_cr = 0;
1263
1264                 if (lf_to_crlf->has_held) {
1265                         was_cr = 1;
1266                         lf_to_crlf->has_held = 0;
1267                 }
1268
1269                 for (i = 0; o < *osize_p && i < count; i++) {
1270                         char ch = input[i];
1271
1272                         if (ch == '\n') {
1273                                 output[o++] = '\r';
1274                         } else if (was_cr) {
1275                                 /*
1276                                  * Previous round saw CR and it is not followed
1277                                  * by a LF; emit the CR before processing the
1278                                  * current character.
1279                                  */
1280                                 output[o++] = '\r';
1281                         }
1282
1283                         /*
1284                          * We may have consumed the last output slot,
1285                          * in which case we need to break out of this
1286                          * loop; hold the current character before
1287                          * returning.
1288                          */
1289                         if (*osize_p <= o) {
1290                                 lf_to_crlf->has_held = 1;
1291                                 lf_to_crlf->held = ch;
1292                                 continue; /* break but increment i */
1293                         }
1294
1295                         if (ch == '\r') {
1296                                 was_cr = 1;
1297                                 continue;
1298                         }
1299
1300                         was_cr = 0;
1301                         output[o++] = ch;
1302                 }
1303
1304                 *osize_p -= o;
1305                 *isize_p -= i;
1306
1307                 if (!lf_to_crlf->has_held && was_cr) {
1308                         lf_to_crlf->has_held = 1;
1309                         lf_to_crlf->held = '\r';
1310                 }
1311         }
1312         return 0;
1313 }
1314
1315 static void lf_to_crlf_free_fn(struct stream_filter *filter)
1316 {
1317         free(filter);
1318 }
1319
1320 static struct stream_filter_vtbl lf_to_crlf_vtbl = {
1321         lf_to_crlf_filter_fn,
1322         lf_to_crlf_free_fn,
1323 };
1324
1325 static struct stream_filter *lf_to_crlf_filter(void)
1326 {
1327         struct lf_to_crlf_filter *lf_to_crlf = xcalloc(1, sizeof(*lf_to_crlf));
1328
1329         lf_to_crlf->filter.vtbl = &lf_to_crlf_vtbl;
1330         return (struct stream_filter *)lf_to_crlf;
1331 }
1332
1333 /*
1334  * Cascade filter
1335  */
1336 #define FILTER_BUFFER 1024
1337 struct cascade_filter {
1338         struct stream_filter filter;
1339         struct stream_filter *one;
1340         struct stream_filter *two;
1341         char buf[FILTER_BUFFER];
1342         int end, ptr;
1343 };
1344
1345 static int cascade_filter_fn(struct stream_filter *filter,
1346                              const char *input, size_t *isize_p,
1347                              char *output, size_t *osize_p)
1348 {
1349         struct cascade_filter *cas = (struct cascade_filter *) filter;
1350         size_t filled = 0;
1351         size_t sz = *osize_p;
1352         size_t to_feed, remaining;
1353
1354         /*
1355          * input -- (one) --> buf -- (two) --> output
1356          */
1357         while (filled < sz) {
1358                 remaining = sz - filled;
1359
1360                 /* do we already have something to feed two with? */
1361                 if (cas->ptr < cas->end) {
1362                         to_feed = cas->end - cas->ptr;
1363                         if (stream_filter(cas->two,
1364                                           cas->buf + cas->ptr, &to_feed,
1365                                           output + filled, &remaining))
1366                                 return -1;
1367                         cas->ptr += (cas->end - cas->ptr) - to_feed;
1368                         filled = sz - remaining;
1369                         continue;
1370                 }
1371
1372                 /* feed one from upstream and have it emit into our buffer */
1373                 to_feed = input ? *isize_p : 0;
1374                 if (input && !to_feed)
1375                         break;
1376                 remaining = sizeof(cas->buf);
1377                 if (stream_filter(cas->one,
1378                                   input, &to_feed,
1379                                   cas->buf, &remaining))
1380                         return -1;
1381                 cas->end = sizeof(cas->buf) - remaining;
1382                 cas->ptr = 0;
1383                 if (input) {
1384                         size_t fed = *isize_p - to_feed;
1385                         *isize_p -= fed;
1386                         input += fed;
1387                 }
1388
1389                 /* do we know that we drained one completely? */
1390                 if (input || cas->end)
1391                         continue;
1392
1393                 /* tell two to drain; we have nothing more to give it */
1394                 to_feed = 0;
1395                 remaining = sz - filled;
1396                 if (stream_filter(cas->two,
1397                                   NULL, &to_feed,
1398                                   output + filled, &remaining))
1399                         return -1;
1400                 if (remaining == (sz - filled))
1401                         break; /* completely drained two */
1402                 filled = sz - remaining;
1403         }
1404         *osize_p -= filled;
1405         return 0;
1406 }
1407
1408 static void cascade_free_fn(struct stream_filter *filter)
1409 {
1410         struct cascade_filter *cas = (struct cascade_filter *)filter;
1411         free_stream_filter(cas->one);
1412         free_stream_filter(cas->two);
1413         free(filter);
1414 }
1415
1416 static struct stream_filter_vtbl cascade_vtbl = {
1417         cascade_filter_fn,
1418         cascade_free_fn,
1419 };
1420
1421 static struct stream_filter *cascade_filter(struct stream_filter *one,
1422                                             struct stream_filter *two)
1423 {
1424         struct cascade_filter *cascade;
1425
1426         if (!one || is_null_stream_filter(one))
1427                 return two;
1428         if (!two || is_null_stream_filter(two))
1429                 return one;
1430
1431         cascade = xmalloc(sizeof(*cascade));
1432         cascade->one = one;
1433         cascade->two = two;
1434         cascade->end = cascade->ptr = 0;
1435         cascade->filter.vtbl = &cascade_vtbl;
1436         return (struct stream_filter *)cascade;
1437 }
1438
1439 /*
1440  * ident filter
1441  */
1442 #define IDENT_DRAINING (-1)
1443 #define IDENT_SKIPPING (-2)
1444 struct ident_filter {
1445         struct stream_filter filter;
1446         struct strbuf left;
1447         int state;
1448         char ident[45]; /* ": x40 $" */
1449 };
1450
1451 static int is_foreign_ident(const char *str)
1452 {
1453         int i;
1454
1455         if (!skip_prefix(str, "$Id: ", &str))
1456                 return 0;
1457         for (i = 0; str[i]; i++) {
1458                 if (isspace(str[i]) && str[i+1] != '$')
1459                         return 1;
1460         }
1461         return 0;
1462 }
1463
1464 static void ident_drain(struct ident_filter *ident, char **output_p, size_t *osize_p)
1465 {
1466         size_t to_drain = ident->left.len;
1467
1468         if (*osize_p < to_drain)
1469                 to_drain = *osize_p;
1470         if (to_drain) {
1471                 memcpy(*output_p, ident->left.buf, to_drain);
1472                 strbuf_remove(&ident->left, 0, to_drain);
1473                 *output_p += to_drain;
1474                 *osize_p -= to_drain;
1475         }
1476         if (!ident->left.len)
1477                 ident->state = 0;
1478 }
1479
1480 static int ident_filter_fn(struct stream_filter *filter,
1481                            const char *input, size_t *isize_p,
1482                            char *output, size_t *osize_p)
1483 {
1484         struct ident_filter *ident = (struct ident_filter *)filter;
1485         static const char head[] = "$Id";
1486
1487         if (!input) {
1488                 /* drain upon eof */
1489                 switch (ident->state) {
1490                 default:
1491                         strbuf_add(&ident->left, head, ident->state);
1492                 case IDENT_SKIPPING:
1493                         /* fallthru */
1494                 case IDENT_DRAINING:
1495                         ident_drain(ident, &output, osize_p);
1496                 }
1497                 return 0;
1498         }
1499
1500         while (*isize_p || (ident->state == IDENT_DRAINING)) {
1501                 int ch;
1502
1503                 if (ident->state == IDENT_DRAINING) {
1504                         ident_drain(ident, &output, osize_p);
1505                         if (!*osize_p)
1506                                 break;
1507                         continue;
1508                 }
1509
1510                 ch = *(input++);
1511                 (*isize_p)--;
1512
1513                 if (ident->state == IDENT_SKIPPING) {
1514                         /*
1515                          * Skipping until '$' or LF, but keeping them
1516                          * in case it is a foreign ident.
1517                          */
1518                         strbuf_addch(&ident->left, ch);
1519                         if (ch != '\n' && ch != '$')
1520                                 continue;
1521                         if (ch == '$' && !is_foreign_ident(ident->left.buf)) {
1522                                 strbuf_setlen(&ident->left, sizeof(head) - 1);
1523                                 strbuf_addstr(&ident->left, ident->ident);
1524                         }
1525                         ident->state = IDENT_DRAINING;
1526                         continue;
1527                 }
1528
1529                 if (ident->state < sizeof(head) &&
1530                     head[ident->state] == ch) {
1531                         ident->state++;
1532                         continue;
1533                 }
1534
1535                 if (ident->state)
1536                         strbuf_add(&ident->left, head, ident->state);
1537                 if (ident->state == sizeof(head) - 1) {
1538                         if (ch != ':' && ch != '$') {
1539                                 strbuf_addch(&ident->left, ch);
1540                                 ident->state = 0;
1541                                 continue;
1542                         }
1543
1544                         if (ch == ':') {
1545                                 strbuf_addch(&ident->left, ch);
1546                                 ident->state = IDENT_SKIPPING;
1547                         } else {
1548                                 strbuf_addstr(&ident->left, ident->ident);
1549                                 ident->state = IDENT_DRAINING;
1550                         }
1551                         continue;
1552                 }
1553
1554                 strbuf_addch(&ident->left, ch);
1555                 ident->state = IDENT_DRAINING;
1556         }
1557         return 0;
1558 }
1559
1560 static void ident_free_fn(struct stream_filter *filter)
1561 {
1562         struct ident_filter *ident = (struct ident_filter *)filter;
1563         strbuf_release(&ident->left);
1564         free(filter);
1565 }
1566
1567 static struct stream_filter_vtbl ident_vtbl = {
1568         ident_filter_fn,
1569         ident_free_fn,
1570 };
1571
1572 static struct stream_filter *ident_filter(const unsigned char *sha1)
1573 {
1574         struct ident_filter *ident = xmalloc(sizeof(*ident));
1575
1576         xsnprintf(ident->ident, sizeof(ident->ident),
1577                   ": %s $", sha1_to_hex(sha1));
1578         strbuf_init(&ident->left, 0);
1579         ident->filter.vtbl = &ident_vtbl;
1580         ident->state = 0;
1581         return (struct stream_filter *)ident;
1582 }
1583
1584 /*
1585  * Return an appropriately constructed filter for the path, or NULL if
1586  * the contents cannot be filtered without reading the whole thing
1587  * in-core.
1588  *
1589  * Note that you would be crazy to set CRLF, smuge/clean or ident to a
1590  * large binary blob you would want us not to slurp into the memory!
1591  */
1592 struct stream_filter *get_stream_filter(const char *path, const unsigned char *sha1)
1593 {
1594         struct conv_attrs ca;
1595         struct stream_filter *filter = NULL;
1596
1597         convert_attrs(&ca, path);
1598         if (ca.drv && (ca.drv->process || ca.drv->smudge || ca.drv->clean))
1599                 return NULL;
1600
1601         if (ca.crlf_action == CRLF_AUTO || ca.crlf_action == CRLF_AUTO_CRLF)
1602                 return NULL;
1603
1604         if (ca.ident)
1605                 filter = ident_filter(sha1);
1606
1607         if (output_eol(ca.crlf_action) == EOL_CRLF)
1608                 filter = cascade_filter(filter, lf_to_crlf_filter());
1609         else
1610                 filter = cascade_filter(filter, &null_filter_singleton);
1611
1612         return filter;
1613 }
1614
1615 void free_stream_filter(struct stream_filter *filter)
1616 {
1617         filter->vtbl->free(filter);
1618 }
1619
1620 int stream_filter(struct stream_filter *filter,
1621                   const char *input, size_t *isize_p,
1622                   char *output, size_t *osize_p)
1623 {
1624         return filter->vtbl->filter(filter, input, isize_p, output, osize_p);
1625 }