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

   1 #define NO_THE_INDEX_COMPATIBILITY_MACROS
   2 #include "cache.h"
   3 #include "attr.h"
   4 #include "run-command.h"
   5 #include "quote.h"
   6 #include "sigchain.h"
   7 #include "pkt-line.h"
   8 #include "sub-process.h"
   9
  10 /*
  11  * convert.c - convert a file when checking it out and checking it in.
  12  *
  13  * This should use the pathname to decide on whether it wants to do some
  14  * more interesting conversions (automatic gzip/unzip, general format
  15  * conversions etc etc), but by default it just does automatic CRLF<->LF
  16  * translation when the "text" attribute or "auto_crlf" option is set.
  17  */
  18
  19 /* Stat bits: When BIN is set, the txt bits are unset */
  20 #define CONVERT_STAT_BITS_TXT_LF    0x1
  21 #define CONVERT_STAT_BITS_TXT_CRLF  0x2
  22 #define CONVERT_STAT_BITS_BIN       0x4
  23
  24 enum crlf_action {
  25         CRLF_UNDEFINED,
  26         CRLF_BINARY,
  27         CRLF_TEXT,
  28         CRLF_TEXT_INPUT,
  29         CRLF_TEXT_CRLF,
  30         CRLF_AUTO,
  31         CRLF_AUTO_INPUT,
  32         CRLF_AUTO_CRLF
  33 };
  34
  35 struct text_stat {
  36         /* NUL, CR, LF and CRLF counts */
  37         unsigned nul, lonecr, lonelf, crlf;
  38
  39         /* These are just approximations! */
  40         unsigned printable, nonprintable;
  41 };
  42
  43 static void gather_stats(const char *buf, unsigned long size, struct text_stat *stats)
  44 {
  45         unsigned long i;
  46
  47         memset(stats, 0, sizeof(*stats));
  48
  49         for (i = 0; i < size; i++) {
  50                 unsigned char c = buf[i];
  51                 if (c == '\r') {
  52                         if (i+1 < size && buf[i+1] == '\n') {
  53                                 stats->crlf++;
  54                                 i++;
  55                         } else
  56                                 stats->lonecr++;
  57                         continue;
  58                 }
  59                 if (c == '\n') {
  60                         stats->lonelf++;
  61                         continue;
  62                 }
  63                 if (c == 127)
  64                         /* DEL */
  65                         stats->nonprintable++;
  66                 else if (c < 32) {
  67                         switch (c) {
  68                                 /* BS, HT, ESC and FF */
  69                         case '\b': case '\t': case '\033': case '\014':
  70                                 stats->printable++;
  71                                 break;
  72                         case 0:
  73                                 stats->nul++;
  74                                 /* fall through */
  75                         default:
  76                                 stats->nonprintable++;
  77                         }
  78                 }
  79                 else
  80                         stats->printable++;
  81         }
  82
  83         /* If file ends with EOF then don't count this EOF as non-printable. */
  84         if (size >= 1 && buf[size-1] == '\032')
  85                 stats->nonprintable--;
  86 }
  87
  88 /*
  89  * The same heuristics as diff.c::mmfile_is_binary()
  90  * We treat files with bare CR as binary
  91  */
  92 static int convert_is_binary(unsigned long size, const struct text_stat *stats)
  93 {
  94         if (stats->lonecr)
  95                 return 1;
  96         if (stats->nul)
  97                 return 1;
  98         if ((stats->printable >> 7) < stats->nonprintable)
  99                 return 1;
 100         return 0;
 101 }
 102
 103 static unsigned int gather_convert_stats(const char *data, unsigned long size)
 104 {
 105         struct text_stat stats;
 106         int ret = 0;
 107         if (!data || !size)
 108                 return 0;
 109         gather_stats(data, size, &stats);
 110         if (convert_is_binary(size, &stats))
 111                 ret |= CONVERT_STAT_BITS_BIN;
 112         if (stats.crlf)
 113                 ret |= CONVERT_STAT_BITS_TXT_CRLF;
 114         if (stats.lonelf)
 115                 ret |=  CONVERT_STAT_BITS_TXT_LF;
 116
 117         return ret;
 118 }
 119
 120 static const char *gather_convert_stats_ascii(const char *data, unsigned long size)
 121 {
 122         unsigned int convert_stats = gather_convert_stats(data, size);
 123
 124         if (convert_stats & CONVERT_STAT_BITS_BIN)
 125                 return "-text";
 126         switch (convert_stats) {
 127         case CONVERT_STAT_BITS_TXT_LF:
 128                 return "lf";
 129         case CONVERT_STAT_BITS_TXT_CRLF:
 130                 return "crlf";
 131         case CONVERT_STAT_BITS_TXT_LF | CONVERT_STAT_BITS_TXT_CRLF:
 132                 return "mixed";
 133         default:
 134                 return "none";
 135         }
 136 }
 137
 138 const char *get_cached_convert_stats_ascii(const struct index_state *istate,
 139                                            const char *path)
 140 {
 141         const char *ret;
 142         unsigned long sz;
 143         void *data = read_blob_data_from_index(istate, path, &sz);
 144         ret = gather_convert_stats_ascii(data, sz);
 145         free(data);
 146         return ret;
 147 }
 148
 149 const char *get_wt_convert_stats_ascii(const char *path)
 150 {
 151         const char *ret = "";
 152         struct strbuf sb = STRBUF_INIT;
 153         if (strbuf_read_file(&sb, path, 0) >= 0)
 154                 ret = gather_convert_stats_ascii(sb.buf, sb.len);
 155         strbuf_release(&sb);
 156         return ret;
 157 }
 158
 159 static int text_eol_is_crlf(void)
 160 {
 161         if (auto_crlf == AUTO_CRLF_TRUE)
 162                 return 1;
 163         else if (auto_crlf == AUTO_CRLF_INPUT)
 164                 return 0;
 165         if (core_eol == EOL_CRLF)
 166                 return 1;
 167         if (core_eol == EOL_UNSET && EOL_NATIVE == EOL_CRLF)
 168                 return 1;
 169         return 0;
 170 }
 171
 172 static enum eol output_eol(enum crlf_action crlf_action)
 173 {
 174         switch (crlf_action) {
 175         case CRLF_BINARY:
 176                 return EOL_UNSET;
 177         case CRLF_TEXT_CRLF:
 178                 return EOL_CRLF;
 179         case CRLF_TEXT_INPUT:
 180                 return EOL_LF;
 181         case CRLF_UNDEFINED:
 182         case CRLF_AUTO_CRLF:
 183                 return EOL_CRLF;
 184         case CRLF_AUTO_INPUT:
 185                 return EOL_LF;
 186         case CRLF_TEXT:
 187         case CRLF_AUTO:
 188                 /* fall through */
 189                 return text_eol_is_crlf() ? EOL_CRLF : EOL_LF;
 190         }
 191         warning("Illegal crlf_action %d\n", (int)crlf_action);
 192         return core_eol;
 193 }
 194
 195 static void check_safe_crlf(const char *path, enum crlf_action crlf_action,
 196                             struct text_stat *old_stats, struct text_stat *new_stats,
 197                             enum safe_crlf checksafe)
 198 {
 199         if (old_stats->crlf && !new_stats->crlf ) {
 200                 /*
 201                  * CRLFs would not be restored by checkout
 202                  */
 203                 if (checksafe == SAFE_CRLF_WARN)
 204                         warning(_("CRLF will be replaced by LF in %s.\n"
 205                                   "The file will have its original line"
 206                                   " endings in your working directory."), path);
 207                 else /* i.e. SAFE_CRLF_FAIL */
 208                         die(_("CRLF would be replaced by LF in %s."), path);
 209         } else if (old_stats->lonelf && !new_stats->lonelf ) {
 210                 /*
 211                  * CRLFs would be added by checkout
 212                  */
 213                 if (checksafe == SAFE_CRLF_WARN)
 214                         warning(_("LF will be replaced by CRLF in %s.\n"
 215                                   "The file will have its original line"
 216                                   " endings in your working directory."), path);
 217                 else /* i.e. SAFE_CRLF_FAIL */
 218                         die(_("LF would be replaced by CRLF in %s"), path);
 219         }
 220 }
 221
 222 static int has_cr_in_index(const struct index_state *istate, const char *path)
 223 {
 224         unsigned long sz;
 225         void *data;
 226         int has_cr;
 227
 228         data = read_blob_data_from_index(istate, path, &sz);
 229         if (!data)
 230                 return 0;
 231         has_cr = memchr(data, '\r', sz) != NULL;
 232         free(data);
 233         return has_cr;
 234 }
 235
 236 static int will_convert_lf_to_crlf(size_t len, struct text_stat *stats,
 237                                    enum crlf_action crlf_action)
 238 {
 239         if (output_eol(crlf_action) != EOL_CRLF)
 240                 return 0;
 241         /* No "naked" LF? Nothing to convert, regardless. */
 242         if (!stats->lonelf)
 243                 return 0;
 244
 245         if (crlf_action == CRLF_AUTO || crlf_action == CRLF_AUTO_INPUT || crlf_action == CRLF_AUTO_CRLF) {
 246                 /* If we have any CR or CRLF line endings, we do not touch it */
 247                 /* This is the new safer autocrlf-handling */
 248                 if (stats->lonecr || stats->crlf)
 249                         return 0;
 250
 251                 if (convert_is_binary(len, stats))
 252                         return 0;
 253         }
 254         return 1;
 255
 256 }
 257
 258 static int crlf_to_git(const struct index_state *istate,
 259                        const char *path, const char *src, size_t len,
 260                        struct strbuf *buf,
 261                        enum crlf_action crlf_action, enum safe_crlf checksafe)
 262 {
 263         struct text_stat stats;
 264         char *dst;
 265         int convert_crlf_into_lf;
 266
 267         if (crlf_action == CRLF_BINARY ||
 268             (src && !len))
 269                 return 0;
 270
 271         /*
 272          * If we are doing a dry-run and have no source buffer, there is
 273          * nothing to analyze; we must assume we would convert.
 274          */
 275         if (!buf && !src)
 276                 return 1;
 277
 278         gather_stats(src, len, &stats);
 279         /* Optimization: No CRLF? Nothing to convert, regardless. */
 280         convert_crlf_into_lf = !!stats.crlf;
 281
 282         if (crlf_action == CRLF_AUTO || crlf_action == CRLF_AUTO_INPUT || crlf_action == CRLF_AUTO_CRLF) {
 283                 if (convert_is_binary(len, &stats))
 284                         return 0;
 285                 /*
 286                  * If the file in the index has any CR in it, do not
 287                  * convert.  This is the new safer autocrlf handling,
 288                  * unless we want to renormalize in a merge or
 289                  * cherry-pick.
 290                  */
 291                 if ((checksafe != SAFE_CRLF_RENORMALIZE) &&
 292                     has_cr_in_index(istate, path))
 293                         convert_crlf_into_lf = 0;
 294         }
 295         if ((checksafe == SAFE_CRLF_WARN ||
 296             (checksafe == SAFE_CRLF_FAIL)) && len) {
 297                 struct text_stat new_stats;
 298                 memcpy(&new_stats, &stats, sizeof(new_stats));
 299                 /* simulate "git add" */
 300                 if (convert_crlf_into_lf) {
 301                         new_stats.lonelf += new_stats.crlf;
 302                         new_stats.crlf = 0;
 303                 }
 304                 /* simulate "git checkout" */
 305                 if (will_convert_lf_to_crlf(len, &new_stats, crlf_action)) {
 306                         new_stats.crlf += new_stats.lonelf;
 307                         new_stats.lonelf = 0;
 308                 }
 309                 check_safe_crlf(path, crlf_action, &stats, &new_stats, checksafe);
 310         }
 311         if (!convert_crlf_into_lf)
 312                 return 0;
 313
 314         /*
 315          * At this point all of our source analysis is done, and we are sure we
 316          * would convert. If we are in dry-run mode, we can give an answer.
 317          */
 318         if (!buf)
 319                 return 1;
 320
 321         /* only grow if not in place */
 322         if (strbuf_avail(buf) + buf->len < len)
 323                 strbuf_grow(buf, len - buf->len);
 324         dst = buf->buf;
 325         if (crlf_action == CRLF_AUTO || crlf_action == CRLF_AUTO_INPUT || crlf_action == CRLF_AUTO_CRLF) {
 326                 /*
 327                  * If we guessed, we already know we rejected a file with
 328                  * lone CR, and we can strip a CR without looking at what
 329                  * follow it.
 330                  */
 331                 do {
 332                         unsigned char c = *src++;
 333                         if (c != '\r')
 334                                 *dst++ = c;
 335                 } while (--len);
 336         } else {
 337                 do {
 338                         unsigned char c = *src++;
 339                         if (! (c == '\r' && (1 < len && *src == '\n')))
 340                                 *dst++ = c;
 341                 } while (--len);
 342         }
 343         strbuf_setlen(buf, dst - buf->buf);
 344         return 1;
 345 }
 346
 347 static int crlf_to_worktree(const char *path, const char *src, size_t len,
 348                             struct strbuf *buf, enum crlf_action crlf_action)
 349 {
 350         char *to_free = NULL;
 351         struct text_stat stats;
 352
 353         if (!len || output_eol(crlf_action) != EOL_CRLF)
 354                 return 0;
 355
 356         gather_stats(src, len, &stats);
 357         if (!will_convert_lf_to_crlf(len, &stats, crlf_action))
 358                 return 0;
 359
 360         /* are we "faking" in place editing ? */
 361         if (src == buf->buf)
 362                 to_free = strbuf_detach(buf, NULL);
 363
 364         strbuf_grow(buf, len + stats.lonelf);
 365         for (;;) {
 366                 const char *nl = memchr(src, '\n', len);
 367                 if (!nl)
 368                         break;
 369                 if (nl > src && nl[-1] == '\r') {
 370                         strbuf_add(buf, src, nl + 1 - src);
 371                 } else {
 372                         strbuf_add(buf, src, nl - src);
 373                         strbuf_addstr(buf, "\r\n");
 374                 }
 375                 len -= nl + 1 - src;
 376                 src  = nl + 1;
 377         }
 378         strbuf_add(buf, src, len);
 379
 380         free(to_free);
 381         return 1;
 382 }
 383
 384 struct filter_params {
 385         const char *src;
 386         unsigned long size;
 387         int fd;
 388         const char *cmd;
 389         const char *path;
 390 };
 391
 392 static int filter_buffer_or_fd(int in, int out, void *data)
 393 {
 394         /*
 395          * Spawn cmd and feed the buffer contents through its stdin.
 396          */
 397         struct child_process child_process = CHILD_PROCESS_INIT;
 398         struct filter_params *params = (struct filter_params *)data;
 399         int write_err, status;
 400         const char *argv[] = { NULL, NULL };
 401
 402         /* apply % substitution to cmd */
 403         struct strbuf cmd = STRBUF_INIT;
 404         struct strbuf path = STRBUF_INIT;
 405         struct strbuf_expand_dict_entry dict[] = {
 406                 { "f", NULL, },
 407                 { NULL, NULL, },
 408         };
 409
 410         /* quote the path to preserve spaces, etc. */
 411         sq_quote_buf(&path, params->path);
 412         dict[0].value = path.buf;
 413
 414         /* expand all %f with the quoted path */
 415         strbuf_expand(&cmd, params->cmd, strbuf_expand_dict_cb, &dict);
 416         strbuf_release(&path);
 417
 418         argv[0] = cmd.buf;
 419
 420         child_process.argv = argv;
 421         child_process.use_shell = 1;
 422         child_process.in = -1;
 423         child_process.out = out;
 424
 425         if (start_command(&child_process))
 426                 return error("cannot fork to run external filter '%s'", params->cmd);
 427
 428         sigchain_push(SIGPIPE, SIG_IGN);
 429
 430         if (params->src) {
 431                 write_err = (write_in_full(child_process.in,
 432                                            params->src, params->size) < 0);
 433                 if (errno == EPIPE)
 434                         write_err = 0;
 435         } else {
 436                 write_err = copy_fd(params->fd, child_process.in);
 437                 if (write_err == COPY_WRITE_ERROR && errno == EPIPE)
 438                         write_err = 0;
 439         }
 440
 441         if (close(child_process.in))
 442                 write_err = 1;
 443         if (write_err)
 444                 error("cannot feed the input to external filter '%s'", params->cmd);
 445
 446         sigchain_pop(SIGPIPE);
 447
 448         status = finish_command(&child_process);
 449         if (status)
 450                 error("external filter '%s' failed %d", params->cmd, status);
 451
 452         strbuf_release(&cmd);
 453         return (write_err || status);
 454 }
 455
 456 static int apply_single_file_filter(const char *path, const char *src, size_t len, int fd,
 457                         struct strbuf *dst, const char *cmd)
 458 {
 459         /*
 460          * Create a pipeline to have the command filter the buffer's
 461          * contents.
 462          *
 463          * (child --> cmd) --> us
 464          */
 465         int err = 0;
 466         struct strbuf nbuf = STRBUF_INIT;
 467         struct async async;
 468         struct filter_params params;
 469
 470         memset(&async, 0, sizeof(async));
 471         async.proc = filter_buffer_or_fd;
 472         async.data = &params;
 473         async.out = -1;
 474         params.src = src;
 475         params.size = len;
 476         params.fd = fd;
 477         params.cmd = cmd;
 478         params.path = path;
 479
 480         fflush(NULL);
 481         if (start_async(&async))
 482                 return 0;       /* error was already reported */
 483
 484         if (strbuf_read(&nbuf, async.out, len) < 0) {
 485                 err = error("read from external filter '%s' failed", cmd);
 486         }
 487         if (close(async.out)) {
 488                 err = error("read from external filter '%s' failed", cmd);
 489         }
 490         if (finish_async(&async)) {
 491                 err = error("external filter '%s' failed", cmd);
 492         }
 493
 494         if (!err) {
 495                 strbuf_swap(dst, &nbuf);
 496         }
 497         strbuf_release(&nbuf);
 498         return !err;
 499 }
 500
 501 #define CAP_CLEAN    (1u<<0)
 502 #define CAP_SMUDGE   (1u<<1)
 503
 504 struct cmd2process {
 505         struct subprocess_entry subprocess; /* must be the first member! */
 506         unsigned int supported_capabilities;
 507 };
 508
 509 static int subprocess_map_initialized;
 510 static struct hashmap subprocess_map;
 511
 512 static int start_multi_file_filter_fn(struct subprocess_entry *subprocess)
 513 {
 514         int err;
 515         struct cmd2process *entry = (struct cmd2process *)subprocess;
 516         struct string_list cap_list = STRING_LIST_INIT_NODUP;
 517         char *cap_buf;
 518         const char *cap_name;
 519         struct child_process *process = &subprocess->process;
 520         const char *cmd = subprocess->cmd;
 521
 522         sigchain_push(SIGPIPE, SIG_IGN);
 523
 524         err = packet_writel(process->in, "git-filter-client", "version=2", NULL);
 525         if (err)
 526                 goto done;
 527
 528         err = strcmp(packet_read_line(process->out, NULL), "git-filter-server");
 529         if (err) {
 530                 error("external filter '%s' does not support filter protocol version 2", cmd);
 531                 goto done;
 532         }
 533         err = strcmp(packet_read_line(process->out, NULL), "version=2");
 534         if (err)
 535                 goto done;
 536         err = packet_read_line(process->out, NULL) != NULL;
 537         if (err)
 538                 goto done;
 539
 540         err = packet_writel(process->in, "capability=clean", "capability=smudge", NULL);
 541
 542         for (;;) {
 543                 cap_buf = packet_read_line(process->out, NULL);
 544                 if (!cap_buf)
 545                         break;
 546                 string_list_split_in_place(&cap_list, cap_buf, '=', 1);
 547
 548                 if (cap_list.nr != 2 || strcmp(cap_list.items[0].string, "capability"))
 549                         continue;
 550
 551                 cap_name = cap_list.items[1].string;
 552                 if (!strcmp(cap_name, "clean")) {
 553                         entry->supported_capabilities |= CAP_CLEAN;
 554                 } else if (!strcmp(cap_name, "smudge")) {
 555                         entry->supported_capabilities |= CAP_SMUDGE;
 556                 } else {
 557                         warning(
 558                                 "external filter '%s' requested unsupported filter capability '%s'",
 559                                 cmd, cap_name
 560                         );
 561                 }
 562
 563                 string_list_clear(&cap_list, 0);
 564         }
 565
 566 done:
 567         sigchain_pop(SIGPIPE);
 568
 569         return err;
 570 }
 571
 572 static int apply_multi_file_filter(const char *path, const char *src, size_t len,
 573                                    int fd, struct strbuf *dst, const char *cmd,
 574                                    const unsigned int wanted_capability)
 575 {
 576         int err;
 577         struct cmd2process *entry;
 578         struct child_process *process;
 579         struct strbuf nbuf = STRBUF_INIT;
 580         struct strbuf filter_status = STRBUF_INIT;
 581         const char *filter_type;
 582
 583         if (!subprocess_map_initialized) {
 584                 subprocess_map_initialized = 1;
 585                 hashmap_init(&subprocess_map, (hashmap_cmp_fn) cmd2process_cmp, 0);
 586                 entry = NULL;
 587         } else {
 588                 entry = (struct cmd2process *)subprocess_find_entry(&subprocess_map, cmd);
 589         }
 590
 591         fflush(NULL);
 592
 593         if (!entry) {
 594                 entry = xmalloc(sizeof(*entry));
 595                 entry->supported_capabilities = 0;
 596
 597                 if (subprocess_start(&subprocess_map, &entry->subprocess, cmd, start_multi_file_filter_fn)) {
 598                         free(entry);
 599                         return 0;
 600                 }
 601         }
 602         process = &entry->subprocess.process;
 603
 604         if (!(wanted_capability & entry->supported_capabilities))
 605                 return 0;
 606
 607         if (CAP_CLEAN & wanted_capability)
 608                 filter_type = "clean";
 609         else if (CAP_SMUDGE & wanted_capability)
 610                 filter_type = "smudge";
 611         else
 612                 die("unexpected filter type");
 613
 614         sigchain_push(SIGPIPE, SIG_IGN);
 615
 616         assert(strlen(filter_type) < LARGE_PACKET_DATA_MAX - strlen("command=\n"));
 617         err = packet_write_fmt_gently(process->in, "command=%s\n", filter_type);
 618         if (err)
 619                 goto done;
 620
 621         err = strlen(path) > LARGE_PACKET_DATA_MAX - strlen("pathname=\n");
 622         if (err) {
 623                 error("path name too long for external filter");
 624                 goto done;
 625         }
 626
 627         err = packet_write_fmt_gently(process->in, "pathname=%s\n", path);
 628         if (err)
 629                 goto done;
 630
 631         err = packet_flush_gently(process->in);
 632         if (err)
 633                 goto done;
 634
 635         if (fd >= 0)
 636                 err = write_packetized_from_fd(fd, process->in);
 637         else
 638                 err = write_packetized_from_buf(src, len, process->in);
 639         if (err)
 640                 goto done;
 641
 642         err = subprocess_read_status(process->out, &filter_status);
 643         if (err)
 644                 goto done;
 645
 646         err = strcmp(filter_status.buf, "success");
 647         if (err)
 648                 goto done;
 649
 650         err = read_packetized_to_strbuf(process->out, &nbuf) < 0;
 651         if (err)
 652                 goto done;
 653
 654         err = subprocess_read_status(process->out, &filter_status);
 655         if (err)
 656                 goto done;
 657
 658         err = strcmp(filter_status.buf, "success");
 659
 660 done:
 661         sigchain_pop(SIGPIPE);
 662
 663         if (err) {
 664                 if (!strcmp(filter_status.buf, "error")) {
 665                         /* The filter signaled a problem with the file. */
 666                 } else if (!strcmp(filter_status.buf, "abort")) {
 667                         /*
 668                          * The filter signaled a permanent problem. Don't try to filter
 669                          * files with the same command for the lifetime of the current
 670                          * Git process.
 671                          */
 672                          entry->supported_capabilities &= ~wanted_capability;
 673                 } else {
 674                         /*
 675                          * Something went wrong with the protocol filter.
 676                          * Force shutdown and restart if another blob requires filtering.
 677                          */
 678                         error("external filter '%s' failed", cmd);
 679                         subprocess_stop(&subprocess_map, &entry->subprocess);
 680                         free(entry);
 681                 }
 682         } else {
 683                 strbuf_swap(dst, &nbuf);
 684         }
 685         strbuf_release(&nbuf);
 686         return !err;
 687 }
 688
 689 static struct convert_driver {
 690         const char *name;
 691         struct convert_driver *next;
 692         const char *smudge;
 693         const char *clean;
 694         const char *process;
 695         int required;
 696 } *user_convert, **user_convert_tail;
 697
 698 static int apply_filter(const char *path, const char *src, size_t len,
 699                         int fd, struct strbuf *dst, struct convert_driver *drv,
 700                         const unsigned int wanted_capability)
 701 {
 702         const char *cmd = NULL;
 703
 704         if (!drv)
 705                 return 0;
 706
 707         if (!dst)
 708                 return 1;
 709
 710         if ((CAP_CLEAN & wanted_capability) && !drv->process && drv->clean)
 711                 cmd = drv->clean;
 712         else if ((CAP_SMUDGE & wanted_capability) && !drv->process && drv->smudge)
 713                 cmd = drv->smudge;
 714
 715         if (cmd && *cmd)
 716                 return apply_single_file_filter(path, src, len, fd, dst, cmd);
 717         else if (drv->process && *drv->process)
 718                 return apply_multi_file_filter(path, src, len, fd, dst, drv->process, wanted_capability);
 719
 720         return 0;
 721 }
 722
 723 static int read_convert_config(const char *var, const char *value, void *cb)
 724 {
 725         const char *key, *name;
 726         int namelen;
 727         struct convert_driver *drv;
 728
 729         /*
 730          * External conversion drivers are configured using
 731          * "filter.<name>.variable".
 732          */
 733         if (parse_config_key(var, "filter", &name, &namelen, &key) < 0 || !name)
 734                 return 0;
 735         for (drv = user_convert; drv; drv = drv->next)
 736                 if (!strncmp(drv->name, name, namelen) && !drv->name[namelen])
 737                         break;
 738         if (!drv) {
 739                 drv = xcalloc(1, sizeof(struct convert_driver));
 740                 drv->name = xmemdupz(name, namelen);
 741                 *user_convert_tail = drv;
 742                 user_convert_tail = &(drv->next);
 743         }
 744
 745         /*
 746          * filter.<name>.smudge and filter.<name>.clean specifies
 747          * the command line:
 748          *
 749          *      command-line
 750          *
 751          * The command-line will not be interpolated in any way.
 752          */
 753
 754         if (!strcmp("smudge", key))
 755                 return git_config_string(&drv->smudge, var, value);
 756
 757         if (!strcmp("clean", key))
 758                 return git_config_string(&drv->clean, var, value);
 759
 760         if (!strcmp("process", key))
 761                 return git_config_string(&drv->process, var, value);
 762
 763         if (!strcmp("required", key)) {
 764                 drv->required = git_config_bool(var, value);
 765                 return 0;
 766         }
 767
 768         return 0;
 769 }
 770
 771 static int count_ident(const char *cp, unsigned long size)
 772 {
 773         /*
 774          * "$Id: 0000000000000000000000000000000000000000 $" <=> "$Id$"
 775          */
 776         int cnt = 0;
 777         char ch;
 778
 779         while (size) {
 780                 ch = *cp++;
 781                 size--;
 782                 if (ch != '$')
 783                         continue;
 784                 if (size < 3)
 785                         break;
 786                 if (memcmp("Id", cp, 2))
 787                         continue;
 788                 ch = cp[2];
 789                 cp += 3;
 790                 size -= 3;
 791                 if (ch == '$')
 792                         cnt++; /* $Id$ */
 793                 if (ch != ':')
 794                         continue;
 795
 796                 /*
 797                  * "$Id: ... "; scan up to the closing dollar sign and discard.
 798                  */
 799                 while (size) {
 800                         ch = *cp++;
 801                         size--;
 802                         if (ch == '$') {
 803                                 cnt++;
 804                                 break;
 805                         }
 806                         if (ch == '\n')
 807                                 break;
 808                 }
 809         }
 810         return cnt;
 811 }
 812
 813 static int ident_to_git(const char *path, const char *src, size_t len,
 814                         struct strbuf *buf, int ident)
 815 {
 816         char *dst, *dollar;
 817
 818         if (!ident || (src && !count_ident(src, len)))
 819                 return 0;
 820
 821         if (!buf)
 822                 return 1;
 823
 824         /* only grow if not in place */
 825         if (strbuf_avail(buf) + buf->len < len)
 826                 strbuf_grow(buf, len - buf->len);
 827         dst = buf->buf;
 828         for (;;) {
 829                 dollar = memchr(src, '$', len);
 830                 if (!dollar)
 831                         break;
 832                 memmove(dst, src, dollar + 1 - src);
 833                 dst += dollar + 1 - src;
 834                 len -= dollar + 1 - src;
 835                 src  = dollar + 1;
 836
 837                 if (len > 3 && !memcmp(src, "Id:", 3)) {
 838                         dollar = memchr(src + 3, '$', len - 3);
 839                         if (!dollar)
 840                                 break;
 841                         if (memchr(src + 3, '\n', dollar - src - 3)) {
 842                                 /* Line break before the next dollar. */
 843                                 continue;
 844                         }
 845
 846                         memcpy(dst, "Id$", 3);
 847                         dst += 3;
 848                         len -= dollar + 1 - src;
 849                         src  = dollar + 1;
 850                 }
 851         }
 852         memmove(dst, src, len);
 853         strbuf_setlen(buf, dst + len - buf->buf);
 854         return 1;
 855 }
 856
 857 static int ident_to_worktree(const char *path, const char *src, size_t len,
 858                              struct strbuf *buf, int ident)
 859 {
 860         unsigned char sha1[20];
 861         char *to_free = NULL, *dollar, *spc;
 862         int cnt;
 863
 864         if (!ident)
 865                 return 0;
 866
 867         cnt = count_ident(src, len);
 868         if (!cnt)
 869                 return 0;
 870
 871         /* are we "faking" in place editing ? */
 872         if (src == buf->buf)
 873                 to_free = strbuf_detach(buf, NULL);
 874         hash_sha1_file(src, len, "blob", sha1);
 875
 876         strbuf_grow(buf, len + cnt * 43);
 877         for (;;) {
 878                 /* step 1: run to the next '$' */
 879                 dollar = memchr(src, '$', len);
 880                 if (!dollar)
 881                         break;
 882                 strbuf_add(buf, src, dollar + 1 - src);
 883                 len -= dollar + 1 - src;
 884                 src  = dollar + 1;
 885
 886                 /* step 2: does it looks like a bit like Id:xxx$ or Id$ ? */
 887                 if (len < 3 || memcmp("Id", src, 2))
 888                         continue;
 889
 890                 /* step 3: skip over Id$ or Id:xxxxx$ */
 891                 if (src[2] == '$') {
 892                         src += 3;
 893                         len -= 3;
 894                 } else if (src[2] == ':') {
 895                         /*
 896                          * It's possible that an expanded Id has crept its way into the
 897                          * repository, we cope with that by stripping the expansion out.
 898                          * This is probably not a good idea, since it will cause changes
 899                          * on checkout, which won't go away by stash, but let's keep it
 900                          * for git-style ids.
 901                          */
 902                         dollar = memchr(src + 3, '$', len - 3);
 903                         if (!dollar) {
 904                                 /* incomplete keyword, no more '$', so just quit the loop */
 905                                 break;
 906                         }
 907
 908                         if (memchr(src + 3, '\n', dollar - src - 3)) {
 909                                 /* Line break before the next dollar. */
 910                                 continue;
 911                         }
 912
 913                         spc = memchr(src + 4, ' ', dollar - src - 4);
 914                         if (spc && spc < dollar-1) {
 915                                 /* There are spaces in unexpected places.
 916                                  * This is probably an id from some other
 917                                  * versioning system. Keep it for now.
 918                                  */
 919                                 continue;
 920                         }
 921
 922                         len -= dollar + 1 - src;
 923                         src  = dollar + 1;
 924                 } else {
 925                         /* it wasn't a "Id$" or "Id:xxxx$" */
 926                         continue;
 927                 }
 928
 929                 /* step 4: substitute */
 930                 strbuf_addstr(buf, "Id: ");
 931                 strbuf_add(buf, sha1_to_hex(sha1), 40);
 932                 strbuf_addstr(buf, " $");
 933         }
 934         strbuf_add(buf, src, len);
 935
 936         free(to_free);
 937         return 1;
 938 }
 939
 940 static enum crlf_action git_path_check_crlf(struct attr_check_item *check)
 941 {
 942         const char *value = check->value;
 943
 944         if (ATTR_TRUE(value))
 945                 return CRLF_TEXT;
 946         else if (ATTR_FALSE(value))
 947                 return CRLF_BINARY;
 948         else if (ATTR_UNSET(value))
 949                 ;
 950         else if (!strcmp(value, "input"))
 951                 return CRLF_TEXT_INPUT;
 952         else if (!strcmp(value, "auto"))
 953                 return CRLF_AUTO;
 954         return CRLF_UNDEFINED;
 955 }
 956
 957 static enum eol git_path_check_eol(struct attr_check_item *check)
 958 {
 959         const char *value = check->value;
 960
 961         if (ATTR_UNSET(value))
 962                 ;
 963         else if (!strcmp(value, "lf"))
 964                 return EOL_LF;
 965         else if (!strcmp(value, "crlf"))
 966                 return EOL_CRLF;
 967         return EOL_UNSET;
 968 }
 969
 970 static struct convert_driver *git_path_check_convert(struct attr_check_item *check)
 971 {
 972         const char *value = check->value;
 973         struct convert_driver *drv;
 974
 975         if (ATTR_TRUE(value) || ATTR_FALSE(value) || ATTR_UNSET(value))
 976                 return NULL;
 977         for (drv = user_convert; drv; drv = drv->next)
 978                 if (!strcmp(value, drv->name))
 979                         return drv;
 980         return NULL;
 981 }
 982
 983 static int git_path_check_ident(struct attr_check_item *check)
 984 {
 985         const char *value = check->value;
 986
 987         return !!ATTR_TRUE(value);
 988 }
 989
 990 struct conv_attrs {
 991         struct convert_driver *drv;
 992         enum crlf_action attr_action; /* What attr says */
 993         enum crlf_action crlf_action; /* When no attr is set, use core.autocrlf */
 994         int ident;
 995 };
 996
 997 static void convert_attrs(struct conv_attrs *ca, const char *path)
 998 {
 999         static struct attr_check *check;
1000
1001         if (!check) {
1002                 check = attr_check_initl("crlf", "ident", "filter",
1003                                          "eol", "text", NULL);
1004                 user_convert_tail = &user_convert;
1005                 git_config(read_convert_config, NULL);
1006         }
1007
1008         if (!git_check_attr(path, check)) {
1009                 struct attr_check_item *ccheck = check->items;
1010                 ca->crlf_action = git_path_check_crlf(ccheck + 4);
1011                 if (ca->crlf_action == CRLF_UNDEFINED)
1012                         ca->crlf_action = git_path_check_crlf(ccheck + 0);
1013                 ca->attr_action = ca->crlf_action;
1014                 ca->ident = git_path_check_ident(ccheck + 1);
1015                 ca->drv = git_path_check_convert(ccheck + 2);
1016                 if (ca->crlf_action != CRLF_BINARY) {
1017                         enum eol eol_attr = git_path_check_eol(ccheck + 3);
1018                         if (ca->crlf_action == CRLF_AUTO && eol_attr == EOL_LF)
1019                                 ca->crlf_action = CRLF_AUTO_INPUT;
1020                         else if (ca->crlf_action == CRLF_AUTO && eol_attr == EOL_CRLF)
1021                                 ca->crlf_action = CRLF_AUTO_CRLF;
1022                         else if (eol_attr == EOL_LF)
1023                                 ca->crlf_action = CRLF_TEXT_INPUT;
1024                         else if (eol_attr == EOL_CRLF)
1025                                 ca->crlf_action = CRLF_TEXT_CRLF;
1026                 }
1027                 ca->attr_action = ca->crlf_action;
1028         } else {
1029                 ca->drv = NULL;
1030                 ca->crlf_action = CRLF_UNDEFINED;
1031                 ca->ident = 0;
1032         }
1033         if (ca->crlf_action == CRLF_TEXT)
1034                 ca->crlf_action = text_eol_is_crlf() ? CRLF_TEXT_CRLF : CRLF_TEXT_INPUT;
1035         if (ca->crlf_action == CRLF_UNDEFINED && auto_crlf == AUTO_CRLF_FALSE)
1036                 ca->crlf_action = CRLF_BINARY;
1037         if (ca->crlf_action == CRLF_UNDEFINED && auto_crlf == AUTO_CRLF_TRUE)
1038                 ca->crlf_action = CRLF_AUTO_CRLF;
1039         if (ca->crlf_action == CRLF_UNDEFINED && auto_crlf == AUTO_CRLF_INPUT)
1040                 ca->crlf_action = CRLF_AUTO_INPUT;
1041 }
1042
1043 int would_convert_to_git_filter_fd(const char *path)
1044 {
1045         struct conv_attrs ca;
1046
1047         convert_attrs(&ca, path);
1048         if (!ca.drv)
1049                 return 0;
1050
1051         /*
1052          * Apply a filter to an fd only if the filter is required to succeed.
1053          * We must die if the filter fails, because the original data before
1054          * filtering is not available.
1055          */
1056         if (!ca.drv->required)
1057                 return 0;
1058
1059         return apply_filter(path, NULL, 0, -1, NULL, ca.drv, CAP_CLEAN);
1060 }
1061
1062 const char *get_convert_attr_ascii(const char *path)
1063 {
1064         struct conv_attrs ca;
1065
1066         convert_attrs(&ca, path);
1067         switch (ca.attr_action) {
1068         case CRLF_UNDEFINED:
1069                 return "";
1070         case CRLF_BINARY:
1071                 return "-text";
1072         case CRLF_TEXT:
1073                 return "text";
1074         case CRLF_TEXT_INPUT:
1075                 return "text eol=lf";
1076         case CRLF_TEXT_CRLF:
1077                 return "text eol=crlf";
1078         case CRLF_AUTO:
1079                 return "text=auto";
1080         case CRLF_AUTO_CRLF:
1081                 return "text=auto eol=crlf";
1082         case CRLF_AUTO_INPUT:
1083                 return "text=auto eol=lf";
1084         }
1085         return "";
1086 }
1087
1088 int convert_to_git(const struct index_state *istate,
1089                    const char *path, const char *src, size_t len,
1090                    struct strbuf *dst, enum safe_crlf checksafe)
1091 {
1092         int ret = 0;
1093         struct conv_attrs ca;
1094
1095         convert_attrs(&ca, path);
1096
1097         ret |= apply_filter(path, src, len, -1, dst, ca.drv, CAP_CLEAN);
1098         if (!ret && ca.drv && ca.drv->required)
1099                 die("%s: clean filter '%s' failed", path, ca.drv->name);
1100
1101         if (ret && dst) {
1102                 src = dst->buf;
1103                 len = dst->len;
1104         }
1105         ret |= crlf_to_git(istate, path, src, len, dst, ca.crlf_action, checksafe);
1106         if (ret && dst) {
1107                 src = dst->buf;
1108                 len = dst->len;
1109         }
1110         return ret | ident_to_git(path, src, len, dst, ca.ident);
1111 }
1112
1113 void convert_to_git_filter_fd(const struct index_state *istate,
1114                               const char *path, int fd, struct strbuf *dst,
1115                               enum safe_crlf checksafe)
1116 {
1117         struct conv_attrs ca;
1118         convert_attrs(&ca, path);
1119
1120         assert(ca.drv);
1121         assert(ca.drv->clean || ca.drv->process);
1122
1123         if (!apply_filter(path, NULL, 0, fd, dst, ca.drv, CAP_CLEAN))
1124                 die("%s: clean filter '%s' failed", path, ca.drv->name);
1125
1126         crlf_to_git(istate, path, dst->buf, dst->len, dst, ca.crlf_action, checksafe);
1127         ident_to_git(path, dst->buf, dst->len, dst, ca.ident);
1128 }
1129
1130 static int convert_to_working_tree_internal(const char *path, const char *src,
1131                                             size_t len, struct strbuf *dst,
1132                                             int normalizing)
1133 {
1134         int ret = 0, ret_filter = 0;
1135         struct conv_attrs ca;
1136
1137         convert_attrs(&ca, path);
1138
1139         ret |= ident_to_worktree(path, src, len, dst, ca.ident);
1140         if (ret) {
1141                 src = dst->buf;
1142                 len = dst->len;
1143         }
1144         /*
1145          * CRLF conversion can be skipped if normalizing, unless there
1146          * is a smudge or process filter (even if the process filter doesn't
1147          * support smudge).  The filters might expect CRLFs.
1148          */
1149         if ((ca.drv && (ca.drv->smudge || ca.drv->process)) || !normalizing) {
1150                 ret |= crlf_to_worktree(path, src, len, dst, ca.crlf_action);
1151                 if (ret) {
1152                         src = dst->buf;
1153                         len = dst->len;
1154                 }
1155         }
1156
1157         ret_filter = apply_filter(path, src, len, -1, dst, ca.drv, CAP_SMUDGE);
1158         if (!ret_filter && ca.drv && ca.drv->required)
1159                 die("%s: smudge filter %s failed", path, ca.drv->name);
1160
1161         return ret | ret_filter;
1162 }
1163
1164 int convert_to_working_tree(const char *path, const char *src, size_t len, struct strbuf *dst)
1165 {
1166         return convert_to_working_tree_internal(path, src, len, dst, 0);
1167 }
1168
1169 int renormalize_buffer(const struct index_state *istate, const char *path,
1170                        const char *src, size_t len, struct strbuf *dst)
1171 {
1172         int ret = convert_to_working_tree_internal(path, src, len, dst, 1);
1173         if (ret) {
1174                 src = dst->buf;
1175                 len = dst->len;
1176         }
1177         return ret | convert_to_git(istate, path, src, len, dst, SAFE_CRLF_RENORMALIZE);
1178 }
1179
1180 /*****************************************************************
1181  *
1182  * Streaming conversion support
1183  *
1184  *****************************************************************/
1185
1186 typedef int (*filter_fn)(struct stream_filter *,
1187                          const char *input, size_t *isize_p,
1188                          char *output, size_t *osize_p);
1189 typedef void (*free_fn)(struct stream_filter *);
1190
1191 struct stream_filter_vtbl {
1192         filter_fn filter;
1193         free_fn free;
1194 };
1195
1196 struct stream_filter {
1197         struct stream_filter_vtbl *vtbl;
1198 };
1199
1200 static int null_filter_fn(struct stream_filter *filter,
1201                           const char *input, size_t *isize_p,
1202                           char *output, size_t *osize_p)
1203 {
1204         size_t count;
1205
1206         if (!input)
1207                 return 0; /* we do not keep any states */
1208         count = *isize_p;
1209         if (*osize_p < count)
1210                 count = *osize_p;
1211         if (count) {
1212                 memmove(output, input, count);
1213                 *isize_p -= count;
1214                 *osize_p -= count;
1215         }
1216         return 0;
1217 }
1218
1219 static void null_free_fn(struct stream_filter *filter)
1220 {
1221         ; /* nothing -- null instances are shared */
1222 }
1223
1224 static struct stream_filter_vtbl null_vtbl = {
1225         null_filter_fn,
1226         null_free_fn,
1227 };
1228
1229 static struct stream_filter null_filter_singleton = {
1230         &null_vtbl,
1231 };
1232
1233 int is_null_stream_filter(struct stream_filter *filter)
1234 {
1235         return filter == &null_filter_singleton;
1236 }
1237
1238
1239 /*
1240  * LF-to-CRLF filter
1241  */
1242
1243 struct lf_to_crlf_filter {
1244         struct stream_filter filter;
1245         unsigned has_held:1;
1246         char held;
1247 };
1248
1249 static int lf_to_crlf_filter_fn(struct stream_filter *filter,
1250                                 const char *input, size_t *isize_p,
1251                                 char *output, size_t *osize_p)
1252 {
1253         size_t count, o = 0;
1254         struct lf_to_crlf_filter *lf_to_crlf = (struct lf_to_crlf_filter *)filter;
1255
1256         /*
1257          * We may be holding onto the CR to see if it is followed by a
1258          * LF, in which case we would need to go to the main loop.
1259          * Otherwise, just emit it to the output stream.
1260          */
1261         if (lf_to_crlf->has_held && (lf_to_crlf->held != '\r' || !input)) {
1262                 output[o++] = lf_to_crlf->held;
1263                 lf_to_crlf->has_held = 0;
1264         }
1265
1266         /* We are told to drain */
1267         if (!input) {
1268                 *osize_p -= o;
1269                 return 0;
1270         }
1271
1272         count = *isize_p;
1273         if (count || lf_to_crlf->has_held) {
1274                 size_t i;
1275                 int was_cr = 0;
1276
1277                 if (lf_to_crlf->has_held) {
1278                         was_cr = 1;
1279                         lf_to_crlf->has_held = 0;
1280                 }
1281
1282                 for (i = 0; o < *osize_p && i < count; i++) {
1283                         char ch = input[i];
1284
1285                         if (ch == '\n') {
1286                                 output[o++] = '\r';
1287                         } else if (was_cr) {
1288                                 /*
1289                                  * Previous round saw CR and it is not followed
1290                                  * by a LF; emit the CR before processing the
1291                                  * current character.
1292                                  */
1293                                 output[o++] = '\r';
1294                         }
1295
1296                         /*
1297                          * We may have consumed the last output slot,
1298                          * in which case we need to break out of this
1299                          * loop; hold the current character before
1300                          * returning.
1301                          */
1302                         if (*osize_p <= o) {
1303                                 lf_to_crlf->has_held = 1;
1304                                 lf_to_crlf->held = ch;
1305                                 continue; /* break but increment i */
1306                         }
1307
1308                         if (ch == '\r') {
1309                                 was_cr = 1;
1310                                 continue;
1311                         }
1312
1313                         was_cr = 0;
1314                         output[o++] = ch;
1315                 }
1316
1317                 *osize_p -= o;
1318                 *isize_p -= i;
1319
1320                 if (!lf_to_crlf->has_held && was_cr) {
1321                         lf_to_crlf->has_held = 1;
1322                         lf_to_crlf->held = '\r';
1323                 }
1324         }
1325         return 0;
1326 }
1327
1328 static void lf_to_crlf_free_fn(struct stream_filter *filter)
1329 {
1330         free(filter);
1331 }
1332
1333 static struct stream_filter_vtbl lf_to_crlf_vtbl = {
1334         lf_to_crlf_filter_fn,
1335         lf_to_crlf_free_fn,
1336 };
1337
1338 static struct stream_filter *lf_to_crlf_filter(void)
1339 {
1340         struct lf_to_crlf_filter *lf_to_crlf = xcalloc(1, sizeof(*lf_to_crlf));
1341
1342         lf_to_crlf->filter.vtbl = &lf_to_crlf_vtbl;
1343         return (struct stream_filter *)lf_to_crlf;
1344 }
1345
1346 /*
1347  * Cascade filter
1348  */
1349 #define FILTER_BUFFER 1024
1350 struct cascade_filter {
1351         struct stream_filter filter;
1352         struct stream_filter *one;
1353         struct stream_filter *two;
1354         char buf[FILTER_BUFFER];
1355         int end, ptr;
1356 };
1357
1358 static int cascade_filter_fn(struct stream_filter *filter,
1359                              const char *input, size_t *isize_p,
1360                              char *output, size_t *osize_p)
1361 {
1362         struct cascade_filter *cas = (struct cascade_filter *) filter;
1363         size_t filled = 0;
1364         size_t sz = *osize_p;
1365         size_t to_feed, remaining;
1366
1367         /*
1368          * input -- (one) --> buf -- (two) --> output
1369          */
1370         while (filled < sz) {
1371                 remaining = sz - filled;
1372
1373                 /* do we already have something to feed two with? */
1374                 if (cas->ptr < cas->end) {
1375                         to_feed = cas->end - cas->ptr;
1376                         if (stream_filter(cas->two,
1377                                           cas->buf + cas->ptr, &to_feed,
1378                                           output + filled, &remaining))
1379                                 return -1;
1380                         cas->ptr += (cas->end - cas->ptr) - to_feed;
1381                         filled = sz - remaining;
1382                         continue;
1383                 }
1384
1385                 /* feed one from upstream and have it emit into our buffer */
1386                 to_feed = input ? *isize_p : 0;
1387                 if (input && !to_feed)
1388                         break;
1389                 remaining = sizeof(cas->buf);
1390                 if (stream_filter(cas->one,
1391                                   input, &to_feed,
1392                                   cas->buf, &remaining))
1393                         return -1;
1394                 cas->end = sizeof(cas->buf) - remaining;
1395                 cas->ptr = 0;
1396                 if (input) {
1397                         size_t fed = *isize_p - to_feed;
1398                         *isize_p -= fed;
1399                         input += fed;
1400                 }
1401
1402                 /* do we know that we drained one completely? */
1403                 if (input || cas->end)
1404                         continue;
1405
1406                 /* tell two to drain; we have nothing more to give it */
1407                 to_feed = 0;
1408                 remaining = sz - filled;
1409                 if (stream_filter(cas->two,
1410                                   NULL, &to_feed,
1411                                   output + filled, &remaining))
1412                         return -1;
1413                 if (remaining == (sz - filled))
1414                         break; /* completely drained two */
1415                 filled = sz - remaining;
1416         }
1417         *osize_p -= filled;
1418         return 0;
1419 }
1420
1421 static void cascade_free_fn(struct stream_filter *filter)
1422 {
1423         struct cascade_filter *cas = (struct cascade_filter *)filter;
1424         free_stream_filter(cas->one);
1425         free_stream_filter(cas->two);
1426         free(filter);
1427 }
1428
1429 static struct stream_filter_vtbl cascade_vtbl = {
1430         cascade_filter_fn,
1431         cascade_free_fn,
1432 };
1433
1434 static struct stream_filter *cascade_filter(struct stream_filter *one,
1435                                             struct stream_filter *two)
1436 {
1437         struct cascade_filter *cascade;
1438
1439         if (!one || is_null_stream_filter(one))
1440                 return two;
1441         if (!two || is_null_stream_filter(two))
1442                 return one;
1443
1444         cascade = xmalloc(sizeof(*cascade));
1445         cascade->one = one;
1446         cascade->two = two;
1447         cascade->end = cascade->ptr = 0;
1448         cascade->filter.vtbl = &cascade_vtbl;
1449         return (struct stream_filter *)cascade;
1450 }
1451
1452 /*
1453  * ident filter
1454  */
1455 #define IDENT_DRAINING (-1)
1456 #define IDENT_SKIPPING (-2)
1457 struct ident_filter {
1458         struct stream_filter filter;
1459         struct strbuf left;
1460         int state;
1461         char ident[45]; /* ": x40 $" */
1462 };
1463
1464 static int is_foreign_ident(const char *str)
1465 {
1466         int i;
1467
1468         if (!skip_prefix(str, "$Id: ", &str))
1469                 return 0;
1470         for (i = 0; str[i]; i++) {
1471                 if (isspace(str[i]) && str[i+1] != '$')
1472                         return 1;
1473         }
1474         return 0;
1475 }
1476
1477 static void ident_drain(struct ident_filter *ident, char **output_p, size_t *osize_p)
1478 {
1479         size_t to_drain = ident->left.len;
1480
1481         if (*osize_p < to_drain)
1482                 to_drain = *osize_p;
1483         if (to_drain) {
1484                 memcpy(*output_p, ident->left.buf, to_drain);
1485                 strbuf_remove(&ident->left, 0, to_drain);
1486                 *output_p += to_drain;
1487                 *osize_p -= to_drain;
1488         }
1489         if (!ident->left.len)
1490                 ident->state = 0;
1491 }
1492
1493 static int ident_filter_fn(struct stream_filter *filter,
1494                            const char *input, size_t *isize_p,
1495                            char *output, size_t *osize_p)
1496 {
1497         struct ident_filter *ident = (struct ident_filter *)filter;
1498         static const char head[] = "$Id";
1499
1500         if (!input) {
1501                 /* drain upon eof */
1502                 switch (ident->state) {
1503                 default:
1504                         strbuf_add(&ident->left, head, ident->state);
1505                 case IDENT_SKIPPING:
1506                         /* fallthru */
1507                 case IDENT_DRAINING:
1508                         ident_drain(ident, &output, osize_p);
1509                 }
1510                 return 0;
1511         }
1512
1513         while (*isize_p || (ident->state == IDENT_DRAINING)) {
1514                 int ch;
1515
1516                 if (ident->state == IDENT_DRAINING) {
1517                         ident_drain(ident, &output, osize_p);
1518                         if (!*osize_p)
1519                                 break;
1520                         continue;
1521                 }
1522
1523                 ch = *(input++);
1524                 (*isize_p)--;
1525
1526                 if (ident->state == IDENT_SKIPPING) {
1527                         /*
1528                          * Skipping until '$' or LF, but keeping them
1529                          * in case it is a foreign ident.
1530                          */
1531                         strbuf_addch(&ident->left, ch);
1532                         if (ch != '\n' && ch != '$')
1533                                 continue;
1534                         if (ch == '$' && !is_foreign_ident(ident->left.buf)) {
1535                                 strbuf_setlen(&ident->left, sizeof(head) - 1);
1536                                 strbuf_addstr(&ident->left, ident->ident);
1537                         }
1538                         ident->state = IDENT_DRAINING;
1539                         continue;
1540                 }
1541
1542                 if (ident->state < sizeof(head) &&
1543                     head[ident->state] == ch) {
1544                         ident->state++;
1545                         continue;
1546                 }
1547
1548                 if (ident->state)
1549                         strbuf_add(&ident->left, head, ident->state);
1550                 if (ident->state == sizeof(head) - 1) {
1551                         if (ch != ':' && ch != '$') {
1552                                 strbuf_addch(&ident->left, ch);
1553                                 ident->state = 0;
1554                                 continue;
1555                         }
1556
1557                         if (ch == ':') {
1558                                 strbuf_addch(&ident->left, ch);
1559                                 ident->state = IDENT_SKIPPING;
1560                         } else {
1561                                 strbuf_addstr(&ident->left, ident->ident);
1562                                 ident->state = IDENT_DRAINING;
1563                         }
1564                         continue;
1565                 }
1566
1567                 strbuf_addch(&ident->left, ch);
1568                 ident->state = IDENT_DRAINING;
1569         }
1570         return 0;
1571 }
1572
1573 static void ident_free_fn(struct stream_filter *filter)
1574 {
1575         struct ident_filter *ident = (struct ident_filter *)filter;
1576         strbuf_release(&ident->left);
1577         free(filter);
1578 }
1579
1580 static struct stream_filter_vtbl ident_vtbl = {
1581         ident_filter_fn,
1582         ident_free_fn,
1583 };
1584
1585 static struct stream_filter *ident_filter(const unsigned char *sha1)
1586 {
1587         struct ident_filter *ident = xmalloc(sizeof(*ident));
1588
1589         xsnprintf(ident->ident, sizeof(ident->ident),
1590                   ": %s $", sha1_to_hex(sha1));
1591         strbuf_init(&ident->left, 0);
1592         ident->filter.vtbl = &ident_vtbl;
1593         ident->state = 0;
1594         return (struct stream_filter *)ident;
1595 }
1596
1597 /*
1598  * Return an appropriately constructed filter for the path, or NULL if
1599  * the contents cannot be filtered without reading the whole thing
1600  * in-core.
1601  *
1602  * Note that you would be crazy to set CRLF, smuge/clean or ident to a
1603  * large binary blob you would want us not to slurp into the memory!
1604  */
1605 struct stream_filter *get_stream_filter(const char *path, const unsigned char *sha1)
1606 {
1607         struct conv_attrs ca;
1608         struct stream_filter *filter = NULL;
1609
1610         convert_attrs(&ca, path);
1611         if (ca.drv && (ca.drv->process || ca.drv->smudge || ca.drv->clean))
1612                 return NULL;
1613
1614         if (ca.crlf_action == CRLF_AUTO || ca.crlf_action == CRLF_AUTO_CRLF)
1615                 return NULL;
1616
1617         if (ca.ident)
1618                 filter = ident_filter(sha1);
1619
1620         if (output_eol(ca.crlf_action) == EOL_CRLF)
1621                 filter = cascade_filter(filter, lf_to_crlf_filter());
1622         else
1623                 filter = cascade_filter(filter, &null_filter_singleton);
1624
1625         return filter;
1626 }
1627
1628 void free_stream_filter(struct stream_filter *filter)
1629 {
1630         filter->vtbl->free(filter);
1631 }
1632
1633 int stream_filter(struct stream_filter *filter,
1634                   const char *input, size_t *isize_p,
1635                   char *output, size_t *osize_p)
1636 {
1637         return filter->vtbl->filter(filter, input, isize_p, output, osize_p);
1638 }