convert.c

   1 #define NO_THE_INDEX_COMPATIBILITY_MACROS
   2 #include "cache.h"
   3 #include "config.h"
   4 #include "attr.h"
   5 #include "run-command.h"
   6 #include "quote.h"
   7 #include "sigchain.h"
   8 #include "pkt-line.h"
   9 #include "sub-process.h"
  10
  11 /*
  12  * convert.c - convert a file when checking it out and checking it in.
  13  *
  14  * This should use the pathname to decide on whether it wants to do some
  15  * more interesting conversions (automatic gzip/unzip, general format
  16  * conversions etc etc), but by default it just does automatic CRLF<->LF
  17  * translation when the "text" attribute or "auto_crlf" option is set.
  18  */
  19
  20 /* Stat bits: When BIN is set, the txt bits are unset */
  21 #define CONVERT_STAT_BITS_TXT_LF    0x1
  22 #define CONVERT_STAT_BITS_TXT_CRLF  0x2
  23 #define CONVERT_STAT_BITS_BIN       0x4
  24
  25 enum crlf_action {
  26         CRLF_UNDEFINED,
  27         CRLF_BINARY,
  28         CRLF_TEXT,
  29         CRLF_TEXT_INPUT,
  30         CRLF_TEXT_CRLF,
  31         CRLF_AUTO,
  32         CRLF_AUTO_INPUT,
  33         CRLF_AUTO_CRLF
  34 };
  35
  36 struct text_stat {
  37         /* NUL, CR, LF and CRLF counts */
  38         unsigned nul, lonecr, lonelf, crlf;
  39
  40         /* These are just approximations! */
  41         unsigned printable, nonprintable;
  42 };
  43
  44 static void gather_stats(const char *buf, unsigned long size, struct text_stat *stats)
  45 {
  46         unsigned long i;
  47
  48         memset(stats, 0, sizeof(*stats));
  49
  50         for (i = 0; i < size; i++) {
  51                 unsigned char c = buf[i];
  52                 if (c == '\r') {
  53                         if (i+1 < size && buf[i+1] == '\n') {
  54                                 stats->crlf++;
  55                                 i++;
  56                         } else
  57                                 stats->lonecr++;
  58                         continue;
  59                 }
  60                 if (c == '\n') {
  61                         stats->lonelf++;
  62                         continue;
  63                 }
  64                 if (c == 127)
  65                         /* DEL */
  66                         stats->nonprintable++;
  67                 else if (c < 32) {
  68                         switch (c) {
  69                                 /* BS, HT, ESC and FF */
  70                         case '\b': case '\t': case '\033': case '\014':
  71                                 stats->printable++;
  72                                 break;
  73                         case 0:
  74                                 stats->nul++;
  75                                 /* fall through */
  76                         default:
  77                                 stats->nonprintable++;
  78                         }
  79                 }
  80                 else
  81                         stats->printable++;
  82         }
  83
  84         /* If file ends with EOF then don't count this EOF as non-printable. */
  85         if (size >= 1 && buf[size-1] == '\032')
  86                 stats->nonprintable--;
  87 }
  88
  89 /*
  90  * The same heuristics as diff.c::mmfile_is_binary()
  91  * We treat files with bare CR as binary
  92  */
  93 static int convert_is_binary(unsigned long size, const struct text_stat *stats)
  94 {
  95         if (stats->lonecr)
  96                 return 1;
  97         if (stats->nul)
  98                 return 1;
  99         if ((stats->printable >> 7) < stats->nonprintable)
 100                 return 1;
 101         return 0;
 102 }
 103
 104 static unsigned int gather_convert_stats(const char *data, unsigned long size)
 105 {
 106         struct text_stat stats;
 107         int ret = 0;
 108         if (!data || !size)
 109                 return 0;
 110         gather_stats(data, size, &stats);
 111         if (convert_is_binary(size, &stats))
 112                 ret |= CONVERT_STAT_BITS_BIN;
 113         if (stats.crlf)
 114                 ret |= CONVERT_STAT_BITS_TXT_CRLF;
 115         if (stats.lonelf)
 116                 ret |=  CONVERT_STAT_BITS_TXT_LF;
 117
 118         return ret;
 119 }
 120
 121 static const char *gather_convert_stats_ascii(const char *data, unsigned long size)
 122 {
 123         unsigned int convert_stats = gather_convert_stats(data, size);
 124
 125         if (convert_stats & CONVERT_STAT_BITS_BIN)
 126                 return "-text";
 127         switch (convert_stats) {
 128         case CONVERT_STAT_BITS_TXT_LF:
 129                 return "lf";
 130         case CONVERT_STAT_BITS_TXT_CRLF:
 131                 return "crlf";
 132         case CONVERT_STAT_BITS_TXT_LF | CONVERT_STAT_BITS_TXT_CRLF:
 133                 return "mixed";
 134         default:
 135                 return "none";
 136         }
 137 }
 138
 139 const char *get_cached_convert_stats_ascii(const struct index_state *istate,
 140                                            const char *path)
 141 {
 142         const char *ret;
 143         unsigned long sz;
 144         void *data = read_blob_data_from_index(istate, path, &sz);
 145         ret = gather_convert_stats_ascii(data, sz);
 146         free(data);
 147         return ret;
 148 }
 149
 150 const char *get_wt_convert_stats_ascii(const char *path)
 151 {
 152         const char *ret = "";
 153         struct strbuf sb = STRBUF_INIT;
 154         if (strbuf_read_file(&sb, path, 0) >= 0)
 155                 ret = gather_convert_stats_ascii(sb.buf, sb.len);
 156         strbuf_release(&sb);
 157         return ret;
 158 }
 159
 160 static int text_eol_is_crlf(void)
 161 {
 162         if (auto_crlf == AUTO_CRLF_TRUE)
 163                 return 1;
 164         else if (auto_crlf == AUTO_CRLF_INPUT)
 165                 return 0;
 166         if (core_eol == EOL_CRLF)
 167                 return 1;
 168         if (core_eol == EOL_UNSET && EOL_NATIVE == EOL_CRLF)
 169                 return 1;
 170         return 0;
 171 }
 172
 173 static enum eol output_eol(enum crlf_action crlf_action)
 174 {
 175         switch (crlf_action) {
 176         case CRLF_BINARY:
 177                 return EOL_UNSET;
 178         case CRLF_TEXT_CRLF:
 179                 return EOL_CRLF;
 180         case CRLF_TEXT_INPUT:
 181                 return EOL_LF;
 182         case CRLF_UNDEFINED:
 183         case CRLF_AUTO_CRLF:
 184                 return EOL_CRLF;
 185         case CRLF_AUTO_INPUT:
 186                 return EOL_LF;
 187         case CRLF_TEXT:
 188         case CRLF_AUTO:
 189                 /* fall through */
 190                 return text_eol_is_crlf() ? EOL_CRLF : EOL_LF;
 191         }
 192         warning("Illegal crlf_action %d\n", (int)crlf_action);
 193         return core_eol;
 194 }
 195
 196 static void check_safe_crlf(const char *path, enum crlf_action crlf_action,
 197                             struct text_stat *old_stats, struct text_stat *new_stats,
 198                             enum safe_crlf checksafe)
 199 {
 200         if (old_stats->crlf && !new_stats->crlf ) {
 201                 /*
 202                  * CRLFs would not be restored by checkout
 203                  */
 204                 if (checksafe == SAFE_CRLF_WARN)
 205                         warning(_("CRLF will be replaced by LF in %s.\n"
 206                                   "The file will have its original line"
 207                                   " endings in your working directory."), path);
 208                 else /* i.e. SAFE_CRLF_FAIL */
 209                         die(_("CRLF would be replaced by LF in %s."), path);
 210         } else if (old_stats->lonelf && !new_stats->lonelf ) {
 211                 /*
 212                  * CRLFs would be added by checkout
 213                  */
 214                 if (checksafe == SAFE_CRLF_WARN)
 215                         warning(_("LF will be replaced by CRLF in %s.\n"
 216                                   "The file will have its original line"
 217                                   " endings in your working directory."), path);
 218                 else /* i.e. SAFE_CRLF_FAIL */
 219                         die(_("LF would be replaced by CRLF in %s"), path);
 220         }
 221 }
 222
 223 static int has_cr_in_index(const struct index_state *istate, const char *path)
 224 {
 225         unsigned long sz;
 226         void *data;
 227         int has_cr;
 228
 229         data = read_blob_data_from_index(istate, path, &sz);
 230         if (!data)
 231                 return 0;
 232         has_cr = memchr(data, '\r', sz) != NULL;
 233         free(data);
 234         return has_cr;
 235 }
 236
 237 static int will_convert_lf_to_crlf(size_t len, struct text_stat *stats,
 238                                    enum crlf_action crlf_action)
 239 {
 240         if (output_eol(crlf_action) != EOL_CRLF)
 241                 return 0;
 242         /* No "naked" LF? Nothing to convert, regardless. */
 243         if (!stats->lonelf)
 244                 return 0;
 245
 246         if (crlf_action == CRLF_AUTO || crlf_action == CRLF_AUTO_INPUT || crlf_action == CRLF_AUTO_CRLF) {
 247                 /* If we have any CR or CRLF line endings, we do not touch it */
 248                 /* This is the new safer autocrlf-handling */
 249                 if (stats->lonecr || stats->crlf)
 250                         return 0;
 251
 252                 if (convert_is_binary(len, stats))
 253                         return 0;
 254         }
 255         return 1;
 256
 257 }
 258
 259 static int crlf_to_git(const struct index_state *istate,
 260                        const char *path, const char *src, size_t len,
 261                        struct strbuf *buf,
 262                        enum crlf_action crlf_action, enum safe_crlf checksafe)
 263 {
 264         struct text_stat stats;
 265         char *dst;
 266         int convert_crlf_into_lf;
 267
 268         if (crlf_action == CRLF_BINARY ||
 269             (src && !len))
 270                 return 0;
 271
 272         /*
 273          * If we are doing a dry-run and have no source buffer, there is
 274          * nothing to analyze; we must assume we would convert.
 275          */
 276         if (!buf && !src)
 277                 return 1;
 278
 279         gather_stats(src, len, &stats);
 280         /* Optimization: No CRLF? Nothing to convert, regardless. */
 281         convert_crlf_into_lf = !!stats.crlf;
 282
 283         if (crlf_action == CRLF_AUTO || crlf_action == CRLF_AUTO_INPUT || crlf_action == CRLF_AUTO_CRLF) {
 284                 if (convert_is_binary(len, &stats))
 285                         return 0;
 286                 /*
 287                  * If the file in the index has any CR in it, do not
 288                  * convert.  This is the new safer autocrlf handling,
 289                  * unless we want to renormalize in a merge or
 290                  * cherry-pick.
 291                  */
 292                 if ((checksafe != SAFE_CRLF_RENORMALIZE) &&
 293                     has_cr_in_index(istate, path))
 294                         convert_crlf_into_lf = 0;
 295         }
 296         if ((checksafe == SAFE_CRLF_WARN ||
 297             (checksafe == SAFE_CRLF_FAIL)) && len) {
 298                 struct text_stat new_stats;
 299                 memcpy(&new_stats, &stats, sizeof(new_stats));
 300                 /* simulate "git add" */
 301                 if (convert_crlf_into_lf) {
 302                         new_stats.lonelf += new_stats.crlf;
 303                         new_stats.crlf = 0;
 304                 }
 305                 /* simulate "git checkout" */
 306                 if (will_convert_lf_to_crlf(len, &new_stats, crlf_action)) {
 307                         new_stats.crlf += new_stats.lonelf;
 308                         new_stats.lonelf = 0;
 309                 }
 310                 check_safe_crlf(path, crlf_action, &stats, &new_stats, checksafe);
 311         }
 312         if (!convert_crlf_into_lf)
 313                 return 0;
 314
 315         /*
 316          * At this point all of our source analysis is done, and we are sure we
 317          * would convert. If we are in dry-run mode, we can give an answer.
 318          */
 319         if (!buf)
 320                 return 1;
 321
 322         /* only grow if not in place */
 323         if (strbuf_avail(buf) + buf->len < len)
 324                 strbuf_grow(buf, len - buf->len);
 325         dst = buf->buf;
 326         if (crlf_action == CRLF_AUTO || crlf_action == CRLF_AUTO_INPUT || crlf_action == CRLF_AUTO_CRLF) {
 327                 /*
 328                  * If we guessed, we already know we rejected a file with
 329                  * lone CR, and we can strip a CR without looking at what
 330                  * follow it.
 331                  */
 332                 do {
 333                         unsigned char c = *src++;
 334                         if (c != '\r')
 335                                 *dst++ = c;
 336                 } while (--len);
 337         } else {
 338                 do {
 339                         unsigned char c = *src++;
 340                         if (! (c == '\r' && (1 < len && *src == '\n')))
 341                                 *dst++ = c;
 342                 } while (--len);
 343         }
 344         strbuf_setlen(buf, dst - buf->buf);
 345         return 1;
 346 }
 347
 348 static int crlf_to_worktree(const char *path, const char *src, size_t len,
 349                             struct strbuf *buf, enum crlf_action crlf_action)
 350 {
 351         char *to_free = NULL;
 352         struct text_stat stats;
 353
 354         if (!len || output_eol(crlf_action) != EOL_CRLF)
 355                 return 0;
 356
 357         gather_stats(src, len, &stats);
 358         if (!will_convert_lf_to_crlf(len, &stats, crlf_action))
 359                 return 0;
 360
 361         /* are we "faking" in place editing ? */
 362         if (src == buf->buf)
 363                 to_free = strbuf_detach(buf, NULL);
 364
 365         strbuf_grow(buf, len + stats.lonelf);
 366         for (;;) {
 367                 const char *nl = memchr(src, '\n', len);
 368                 if (!nl)
 369                         break;
 370                 if (nl > src && nl[-1] == '\r') {
 371                         strbuf_add(buf, src, nl + 1 - src);
 372                 } else {
 373                         strbuf_add(buf, src, nl - src);
 374                         strbuf_addstr(buf, "\r\n");
 375                 }
 376                 len -= nl + 1 - src;
 377                 src  = nl + 1;
 378         }
 379         strbuf_add(buf, src, len);
 380
 381         free(to_free);
 382         return 1;
 383 }
 384
 385 struct filter_params {
 386         const char *src;
 387         unsigned long size;
 388         int fd;
 389         const char *cmd;
 390         const char *path;
 391 };
 392
 393 static int filter_buffer_or_fd(int in, int out, void *data)
 394 {
 395         /*
 396          * Spawn cmd and feed the buffer contents through its stdin.
 397          */
 398         struct child_process child_process = CHILD_PROCESS_INIT;
 399         struct filter_params *params = (struct filter_params *)data;
 400         int write_err, status;
 401         const char *argv[] = { NULL, NULL };
 402
 403         /* apply % substitution to cmd */
 404         struct strbuf cmd = STRBUF_INIT;
 405         struct strbuf path = STRBUF_INIT;
 406         struct strbuf_expand_dict_entry dict[] = {
 407                 { "f", NULL, },
 408                 { NULL, NULL, },
 409         };
 410
 411         /* quote the path to preserve spaces, etc. */
 412         sq_quote_buf(&path, params->path);
 413         dict[0].value = path.buf;
 414
 415         /* expand all %f with the quoted path */
 416         strbuf_expand(&cmd, params->cmd, strbuf_expand_dict_cb, &dict);
 417         strbuf_release(&path);
 418
 419         argv[0] = cmd.buf;
 420
 421         child_process.argv = argv;
 422         child_process.use_shell = 1;
 423         child_process.in = -1;
 424         child_process.out = out;
 425
 426         if (start_command(&child_process))
 427                 return error("cannot fork to run external filter '%s'", params->cmd);
 428
 429         sigchain_push(SIGPIPE, SIG_IGN);
 430
 431         if (params->src) {
 432                 write_err = (write_in_full(child_process.in,
 433                                            params->src, params->size) < 0);
 434                 if (errno == EPIPE)
 435                         write_err = 0;
 436         } else {
 437                 write_err = copy_fd(params->fd, child_process.in);
 438                 if (write_err == COPY_WRITE_ERROR && errno == EPIPE)
 439                         write_err = 0;
 440         }
 441
 442         if (close(child_process.in))
 443                 write_err = 1;
 444         if (write_err)
 445                 error("cannot feed the input to external filter '%s'", params->cmd);
 446
 447         sigchain_pop(SIGPIPE);
 448
 449         status = finish_command(&child_process);
 450         if (status)
 451                 error("external filter '%s' failed %d", params->cmd, status);
 452
 453         strbuf_release(&cmd);
 454         return (write_err || status);
 455 }
 456
 457 static int apply_single_file_filter(const char *path, const char *src, size_t len, int fd,
 458                         struct strbuf *dst, const char *cmd)
 459 {
 460         /*
 461          * Create a pipeline to have the command filter the buffer's
 462          * contents.
 463          *
 464          * (child --> cmd) --> us
 465          */
 466         int err = 0;
 467         struct strbuf nbuf = STRBUF_INIT;
 468         struct async async;
 469         struct filter_params params;
 470
 471         memset(&async, 0, sizeof(async));
 472         async.proc = filter_buffer_or_fd;
 473         async.data = &params;
 474         async.out = -1;
 475         params.src = src;
 476         params.size = len;
 477         params.fd = fd;
 478         params.cmd = cmd;
 479         params.path = path;
 480
 481         fflush(NULL);
 482         if (start_async(&async))
 483                 return 0;       /* error was already reported */
 484
 485         if (strbuf_read(&nbuf, async.out, len) < 0) {
 486                 err = error("read from external filter '%s' failed", cmd);
 487         }
 488         if (close(async.out)) {
 489                 err = error("read from external filter '%s' failed", cmd);
 490         }
 491         if (finish_async(&async)) {
 492                 err = error("external filter '%s' failed", cmd);
 493         }
 494
 495         if (!err) {
 496                 strbuf_swap(dst, &nbuf);
 497         }
 498         strbuf_release(&nbuf);
 499         return !err;
 500 }
 501
 502 #define CAP_CLEAN    (1u<<0)
 503 #define CAP_SMUDGE   (1u<<1)
 504
 505 struct cmd2process {
 506         struct subprocess_entry subprocess; /* must be the first member! */
 507         unsigned int supported_capabilities;
 508 };
 509
 510 static int subprocess_map_initialized;
 511 static struct hashmap subprocess_map;
 512
 513 static int start_multi_file_filter_fn(struct subprocess_entry *subprocess)
 514 {
 515         int err;
 516         struct cmd2process *entry = (struct cmd2process *)subprocess;
 517         struct string_list cap_list = STRING_LIST_INIT_NODUP;
 518         char *cap_buf;
 519         const char *cap_name;
 520         struct child_process *process = &subprocess->process;
 521         const char *cmd = subprocess->cmd;
 522
 523         sigchain_push(SIGPIPE, SIG_IGN);
 524
 525         err = packet_writel(process->in, "git-filter-client", "version=2", NULL);
 526         if (err)
 527                 goto done;
 528
 529         err = strcmp(packet_read_line(process->out, NULL), "git-filter-server");
 530         if (err) {
 531                 error("external filter '%s' does not support filter protocol version 2", cmd);
 532                 goto done;
 533         }
 534         err = strcmp(packet_read_line(process->out, NULL), "version=2");
 535         if (err)
 536                 goto done;
 537         err = packet_read_line(process->out, NULL) != NULL;
 538         if (err)
 539                 goto done;
 540
 541         err = packet_writel(process->in, "capability=clean", "capability=smudge", NULL);
 542
 543         for (;;) {
 544                 cap_buf = packet_read_line(process->out, NULL);
 545                 if (!cap_buf)
 546                         break;
 547                 string_list_split_in_place(&cap_list, cap_buf, '=', 1);
 548
 549                 if (cap_list.nr != 2 || strcmp(cap_list.items[0].string, "capability"))
 550                         continue;
 551
 552                 cap_name = cap_list.items[1].string;
 553                 if (!strcmp(cap_name, "clean")) {
 554                         entry->supported_capabilities |= CAP_CLEAN;
 555                 } else if (!strcmp(cap_name, "smudge")) {
 556                         entry->supported_capabilities |= CAP_SMUDGE;
 557                 } else {
 558                         warning(
 559                                 "external filter '%s' requested unsupported filter capability '%s'",
 560                                 cmd, cap_name
 561                         );
 562                 }
 563
 564                 string_list_clear(&cap_list, 0);
 565         }
 566
 567 done:
 568         sigchain_pop(SIGPIPE);
 569
 570         return err;
 571 }
 572
 573 static int apply_multi_file_filter(const char *path, const char *src, size_t len,
 574                                    int fd, struct strbuf *dst, const char *cmd,
 575                                    const unsigned int wanted_capability)
 576 {
 577         int err;
 578         struct cmd2process *entry;
 579         struct child_process *process;
 580         struct strbuf nbuf = STRBUF_INIT;
 581         struct strbuf filter_status = STRBUF_INIT;
 582         const char *filter_type;
 583
 584         if (!subprocess_map_initialized) {
 585                 subprocess_map_initialized = 1;
 586                 hashmap_init(&subprocess_map, (hashmap_cmp_fn) cmd2process_cmp,
 587                              NULL, 0);
 588                 entry = NULL;
 589         } else {
 590                 entry = (struct cmd2process *)subprocess_find_entry(&subprocess_map, cmd);
 591         }
 592
 593         fflush(NULL);
 594
 595         if (!entry) {
 596                 entry = xmalloc(sizeof(*entry));
 597                 entry->supported_capabilities = 0;
 598
 599                 if (subprocess_start(&subprocess_map, &entry->subprocess, cmd, start_multi_file_filter_fn)) {
 600                         free(entry);
 601                         return 0;
 602                 }
 603         }
 604         process = &entry->subprocess.process;
 605
 606         if (!(wanted_capability & entry->supported_capabilities))
 607                 return 0;
 608
 609         if (CAP_CLEAN & wanted_capability)
 610                 filter_type = "clean";
 611         else if (CAP_SMUDGE & wanted_capability)
 612                 filter_type = "smudge";
 613         else
 614                 die("unexpected filter type");
 615
 616         sigchain_push(SIGPIPE, SIG_IGN);
 617
 618         assert(strlen(filter_type) < LARGE_PACKET_DATA_MAX - strlen("command=\n"));
 619         err = packet_write_fmt_gently(process->in, "command=%s\n", filter_type);
 620         if (err)
 621                 goto done;
 622
 623         err = strlen(path) > LARGE_PACKET_DATA_MAX - strlen("pathname=\n");
 624         if (err) {
 625                 error("path name too long for external filter");
 626                 goto done;
 627         }
 628
 629         err = packet_write_fmt_gently(process->in, "pathname=%s\n", path);
 630         if (err)
 631                 goto done;
 632
 633         err = packet_flush_gently(process->in);
 634         if (err)
 635                 goto done;
 636
 637         if (fd >= 0)
 638                 err = write_packetized_from_fd(fd, process->in);
 639         else
 640                 err = write_packetized_from_buf(src, len, process->in);
 641         if (err)
 642                 goto done;
 643
 644         err = subprocess_read_status(process->out, &filter_status);
 645         if (err)
 646                 goto done;
 647
 648         err = strcmp(filter_status.buf, "success");
 649         if (err)
 650                 goto done;
 651
 652         err = read_packetized_to_strbuf(process->out, &nbuf) < 0;
 653         if (err)
 654                 goto done;
 655
 656         err = subprocess_read_status(process->out, &filter_status);
 657         if (err)
 658                 goto done;
 659
 660         err = strcmp(filter_status.buf, "success");
 661
 662 done:
 663         sigchain_pop(SIGPIPE);
 664
 665         if (err) {
 666                 if (!strcmp(filter_status.buf, "error")) {
 667                         /* The filter signaled a problem with the file. */
 668                 } else if (!strcmp(filter_status.buf, "abort")) {
 669                         /*
 670                          * The filter signaled a permanent problem. Don't try to filter
 671                          * files with the same command for the lifetime of the current
 672                          * Git process.
 673                          */
 674                          entry->supported_capabilities &= ~wanted_capability;
 675                 } else {
 676                         /*
 677                          * Something went wrong with the protocol filter.
 678                          * Force shutdown and restart if another blob requires filtering.
 679                          */
 680                         error("external filter '%s' failed", cmd);
 681                         subprocess_stop(&subprocess_map, &entry->subprocess);
 682                         free(entry);
 683                 }
 684         } else {
 685                 strbuf_swap(dst, &nbuf);
 686         }
 687         strbuf_release(&nbuf);
 688         return !err;
 689 }
 690
 691 static struct convert_driver {
 692         const char *name;
 693         struct convert_driver *next;
 694         const char *smudge;
 695         const char *clean;
 696         const char *process;
 697         int required;
 698 } *user_convert, **user_convert_tail;
 699
 700 static int apply_filter(const char *path, const char *src, size_t len,
 701                         int fd, struct strbuf *dst, struct convert_driver *drv,
 702                         const unsigned int wanted_capability)
 703 {
 704         const char *cmd = NULL;
 705
 706         if (!drv)
 707                 return 0;
 708
 709         if (!dst)
 710                 return 1;
 711
 712         if ((CAP_CLEAN & wanted_capability) && !drv->process && drv->clean)
 713                 cmd = drv->clean;
 714         else if ((CAP_SMUDGE & wanted_capability) && !drv->process && drv->smudge)
 715                 cmd = drv->smudge;
 716
 717         if (cmd && *cmd)
 718                 return apply_single_file_filter(path, src, len, fd, dst, cmd);
 719         else if (drv->process && *drv->process)
 720                 return apply_multi_file_filter(path, src, len, fd, dst, drv->process, wanted_capability);
 721
 722         return 0;
 723 }
 724
 725 static int read_convert_config(const char *var, const char *value, void *cb)
 726 {
 727         const char *key, *name;
 728         int namelen;
 729         struct convert_driver *drv;
 730
 731         /*
 732          * External conversion drivers are configured using
 733          * "filter.<name>.variable".
 734          */
 735         if (parse_config_key(var, "filter", &name, &namelen, &key) < 0 || !name)
 736                 return 0;
 737         for (drv = user_convert; drv; drv = drv->next)
 738                 if (!strncmp(drv->name, name, namelen) && !drv->name[namelen])
 739                         break;
 740         if (!drv) {
 741                 drv = xcalloc(1, sizeof(struct convert_driver));
 742                 drv->name = xmemdupz(name, namelen);
 743                 *user_convert_tail = drv;
 744                 user_convert_tail = &(drv->next);
 745         }
 746
 747         /*
 748          * filter.<name>.smudge and filter.<name>.clean specifies
 749          * the command line:
 750          *
 751          *      command-line
 752          *
 753          * The command-line will not be interpolated in any way.
 754          */
 755
 756         if (!strcmp("smudge", key))
 757                 return git_config_string(&drv->smudge, var, value);
 758
 759         if (!strcmp("clean", key))
 760                 return git_config_string(&drv->clean, var, value);
 761
 762         if (!strcmp("process", key))
 763                 return git_config_string(&drv->process, var, value);
 764
 765         if (!strcmp("required", key)) {
 766                 drv->required = git_config_bool(var, value);
 767                 return 0;
 768         }
 769
 770         return 0;
 771 }
 772
 773 static int count_ident(const char *cp, unsigned long size)
 774 {
 775         /*
 776          * "$Id: 0000000000000000000000000000000000000000 $" <=> "$Id$"
 777          */
 778         int cnt = 0;
 779         char ch;
 780
 781         while (size) {
 782                 ch = *cp++;
 783                 size--;
 784                 if (ch != '$')
 785                         continue;
 786                 if (size < 3)
 787                         break;
 788                 if (memcmp("Id", cp, 2))
 789                         continue;
 790                 ch = cp[2];
 791                 cp += 3;
 792                 size -= 3;
 793                 if (ch == '$')
 794                         cnt++; /* $Id$ */
 795                 if (ch != ':')
 796                         continue;
 797
 798                 /*
 799                  * "$Id: ... "; scan up to the closing dollar sign and discard.
 800                  */
 801                 while (size) {
 802                         ch = *cp++;
 803                         size--;
 804                         if (ch == '$') {
 805                                 cnt++;
 806                                 break;
 807                         }
 808                         if (ch == '\n')
 809                                 break;
 810                 }
 811         }
 812         return cnt;
 813 }
 814
 815 static int ident_to_git(const char *path, const char *src, size_t len,
 816                         struct strbuf *buf, int ident)
 817 {
 818         char *dst, *dollar;
 819
 820         if (!ident || (src && !count_ident(src, len)))
 821                 return 0;
 822
 823         if (!buf)
 824                 return 1;
 825
 826         /* only grow if not in place */
 827         if (strbuf_avail(buf) + buf->len < len)
 828                 strbuf_grow(buf, len - buf->len);
 829         dst = buf->buf;
 830         for (;;) {
 831                 dollar = memchr(src, '$', len);
 832                 if (!dollar)
 833                         break;
 834                 memmove(dst, src, dollar + 1 - src);
 835                 dst += dollar + 1 - src;
 836                 len -= dollar + 1 - src;
 837                 src  = dollar + 1;
 838
 839                 if (len > 3 && !memcmp(src, "Id:", 3)) {
 840                         dollar = memchr(src + 3, '$', len - 3);
 841                         if (!dollar)
 842                                 break;
 843                         if (memchr(src + 3, '\n', dollar - src - 3)) {
 844                                 /* Line break before the next dollar. */
 845                                 continue;
 846                         }
 847
 848                         memcpy(dst, "Id$", 3);
 849                         dst += 3;
 850                         len -= dollar + 1 - src;
 851                         src  = dollar + 1;
 852                 }
 853         }
 854         memmove(dst, src, len);
 855         strbuf_setlen(buf, dst + len - buf->buf);
 856         return 1;
 857 }
 858
 859 static int ident_to_worktree(const char *path, const char *src, size_t len,
 860                              struct strbuf *buf, int ident)
 861 {
 862         unsigned char sha1[20];
 863         char *to_free = NULL, *dollar, *spc;
 864         int cnt;
 865
 866         if (!ident)
 867                 return 0;
 868
 869         cnt = count_ident(src, len);
 870         if (!cnt)
 871                 return 0;
 872
 873         /* are we "faking" in place editing ? */
 874         if (src == buf->buf)
 875                 to_free = strbuf_detach(buf, NULL);
 876         hash_sha1_file(src, len, "blob", sha1);
 877
 878         strbuf_grow(buf, len + cnt * 43);
 879         for (;;) {
 880                 /* step 1: run to the next '$' */
 881                 dollar = memchr(src, '$', len);
 882                 if (!dollar)
 883                         break;
 884                 strbuf_add(buf, src, dollar + 1 - src);
 885                 len -= dollar + 1 - src;
 886                 src  = dollar + 1;
 887
 888                 /* step 2: does it looks like a bit like Id:xxx$ or Id$ ? */
 889                 if (len < 3 || memcmp("Id", src, 2))
 890                         continue;
 891
 892                 /* step 3: skip over Id$ or Id:xxxxx$ */
 893                 if (src[2] == '$') {
 894                         src += 3;
 895                         len -= 3;
 896                 } else if (src[2] == ':') {
 897                         /*
 898                          * It's possible that an expanded Id has crept its way into the
 899                          * repository, we cope with that by stripping the expansion out.
 900                          * This is probably not a good idea, since it will cause changes
 901                          * on checkout, which won't go away by stash, but let's keep it
 902                          * for git-style ids.
 903                          */
 904                         dollar = memchr(src + 3, '$', len - 3);
 905                         if (!dollar) {
 906                                 /* incomplete keyword, no more '$', so just quit the loop */
 907                                 break;
 908                         }
 909
 910                         if (memchr(src + 3, '\n', dollar - src - 3)) {
 911                                 /* Line break before the next dollar. */
 912                                 continue;
 913                         }
 914
 915                         spc = memchr(src + 4, ' ', dollar - src - 4);
 916                         if (spc && spc < dollar-1) {
 917                                 /* There are spaces in unexpected places.
 918                                  * This is probably an id from some other
 919                                  * versioning system. Keep it for now.
 920                                  */
 921                                 continue;
 922                         }
 923
 924                         len -= dollar + 1 - src;
 925                         src  = dollar + 1;
 926                 } else {
 927                         /* it wasn't a "Id$" or "Id:xxxx$" */
 928                         continue;
 929                 }
 930
 931                 /* step 4: substitute */
 932                 strbuf_addstr(buf, "Id: ");
 933                 strbuf_add(buf, sha1_to_hex(sha1), 40);
 934                 strbuf_addstr(buf, " $");
 935         }
 936         strbuf_add(buf, src, len);
 937
 938         free(to_free);
 939         return 1;
 940 }
 941
 942 static enum crlf_action git_path_check_crlf(struct attr_check_item *check)
 943 {
 944         const char *value = check->value;
 945
 946         if (ATTR_TRUE(value))
 947                 return CRLF_TEXT;
 948         else if (ATTR_FALSE(value))
 949                 return CRLF_BINARY;
 950         else if (ATTR_UNSET(value))
 951                 ;
 952         else if (!strcmp(value, "input"))
 953                 return CRLF_TEXT_INPUT;
 954         else if (!strcmp(value, "auto"))
 955                 return CRLF_AUTO;
 956         return CRLF_UNDEFINED;
 957 }
 958
 959 static enum eol git_path_check_eol(struct attr_check_item *check)
 960 {
 961         const char *value = check->value;
 962
 963         if (ATTR_UNSET(value))
 964                 ;
 965         else if (!strcmp(value, "lf"))
 966                 return EOL_LF;
 967         else if (!strcmp(value, "crlf"))
 968                 return EOL_CRLF;
 969         return EOL_UNSET;
 970 }
 971
 972 static struct convert_driver *git_path_check_convert(struct attr_check_item *check)
 973 {
 974         const char *value = check->value;
 975         struct convert_driver *drv;
 976
 977         if (ATTR_TRUE(value) || ATTR_FALSE(value) || ATTR_UNSET(value))
 978                 return NULL;
 979         for (drv = user_convert; drv; drv = drv->next)
 980                 if (!strcmp(value, drv->name))
 981                         return drv;
 982         return NULL;
 983 }
 984
 985 static int git_path_check_ident(struct attr_check_item *check)
 986 {
 987         const char *value = check->value;
 988
 989         return !!ATTR_TRUE(value);
 990 }
 991
 992 struct conv_attrs {
 993         struct convert_driver *drv;
 994         enum crlf_action attr_action; /* What attr says */
 995         enum crlf_action crlf_action; /* When no attr is set, use core.autocrlf */
 996         int ident;
 997 };
 998
 999 static void convert_attrs(struct conv_attrs *ca, const char *path)
1000 {
1001         static struct attr_check *check;
1002
1003         if (!check) {
1004                 check = attr_check_initl("crlf", "ident", "filter",
1005                                          "eol", "text", NULL);
1006                 user_convert_tail = &user_convert;
1007                 git_config(read_convert_config, NULL);
1008         }
1009
1010         if (!git_check_attr(path, check)) {
1011                 struct attr_check_item *ccheck = check->items;
1012                 ca->crlf_action = git_path_check_crlf(ccheck + 4);
1013                 if (ca->crlf_action == CRLF_UNDEFINED)
1014                         ca->crlf_action = git_path_check_crlf(ccheck + 0);
1015                 ca->attr_action = ca->crlf_action;
1016                 ca->ident = git_path_check_ident(ccheck + 1);
1017                 ca->drv = git_path_check_convert(ccheck + 2);
1018                 if (ca->crlf_action != CRLF_BINARY) {
1019                         enum eol eol_attr = git_path_check_eol(ccheck + 3);
1020                         if (ca->crlf_action == CRLF_AUTO && eol_attr == EOL_LF)
1021                                 ca->crlf_action = CRLF_AUTO_INPUT;
1022                         else if (ca->crlf_action == CRLF_AUTO && eol_attr == EOL_CRLF)
1023                                 ca->crlf_action = CRLF_AUTO_CRLF;
1024                         else if (eol_attr == EOL_LF)
1025                                 ca->crlf_action = CRLF_TEXT_INPUT;
1026                         else if (eol_attr == EOL_CRLF)
1027                                 ca->crlf_action = CRLF_TEXT_CRLF;
1028                 }
1029                 ca->attr_action = ca->crlf_action;
1030         } else {
1031                 ca->drv = NULL;
1032                 ca->crlf_action = CRLF_UNDEFINED;
1033                 ca->ident = 0;
1034         }
1035         if (ca->crlf_action == CRLF_TEXT)
1036                 ca->crlf_action = text_eol_is_crlf() ? CRLF_TEXT_CRLF : CRLF_TEXT_INPUT;
1037         if (ca->crlf_action == CRLF_UNDEFINED && auto_crlf == AUTO_CRLF_FALSE)
1038                 ca->crlf_action = CRLF_BINARY;
1039         if (ca->crlf_action == CRLF_UNDEFINED && auto_crlf == AUTO_CRLF_TRUE)
1040                 ca->crlf_action = CRLF_AUTO_CRLF;
1041         if (ca->crlf_action == CRLF_UNDEFINED && auto_crlf == AUTO_CRLF_INPUT)
1042                 ca->crlf_action = CRLF_AUTO_INPUT;
1043 }
1044
1045 int would_convert_to_git_filter_fd(const char *path)
1046 {
1047         struct conv_attrs ca;
1048
1049         convert_attrs(&ca, path);
1050         if (!ca.drv)
1051                 return 0;
1052
1053         /*
1054          * Apply a filter to an fd only if the filter is required to succeed.
1055          * We must die if the filter fails, because the original data before
1056          * filtering is not available.
1057          */
1058         if (!ca.drv->required)
1059                 return 0;
1060
1061         return apply_filter(path, NULL, 0, -1, NULL, ca.drv, CAP_CLEAN);
1062 }
1063
1064 const char *get_convert_attr_ascii(const char *path)
1065 {
1066         struct conv_attrs ca;
1067
1068         convert_attrs(&ca, path);
1069         switch (ca.attr_action) {
1070         case CRLF_UNDEFINED:
1071                 return "";
1072         case CRLF_BINARY:
1073                 return "-text";
1074         case CRLF_TEXT:
1075                 return "text";
1076         case CRLF_TEXT_INPUT:
1077                 return "text eol=lf";
1078         case CRLF_TEXT_CRLF:
1079                 return "text eol=crlf";
1080         case CRLF_AUTO:
1081                 return "text=auto";
1082         case CRLF_AUTO_CRLF:
1083                 return "text=auto eol=crlf";
1084         case CRLF_AUTO_INPUT:
1085                 return "text=auto eol=lf";
1086         }
1087         return "";
1088 }
1089
1090 int convert_to_git(const struct index_state *istate,
1091                    const char *path, const char *src, size_t len,
1092                    struct strbuf *dst, enum safe_crlf checksafe)
1093 {
1094         int ret = 0;
1095         struct conv_attrs ca;
1096
1097         convert_attrs(&ca, path);
1098
1099         ret |= apply_filter(path, src, len, -1, dst, ca.drv, CAP_CLEAN);
1100         if (!ret && ca.drv && ca.drv->required)
1101                 die("%s: clean filter '%s' failed", path, ca.drv->name);
1102
1103         if (ret && dst) {
1104                 src = dst->buf;
1105                 len = dst->len;
1106         }
1107         ret |= crlf_to_git(istate, path, src, len, dst, ca.crlf_action, checksafe);
1108         if (ret && dst) {
1109                 src = dst->buf;
1110                 len = dst->len;
1111         }
1112         return ret | ident_to_git(path, src, len, dst, ca.ident);
1113 }
1114
1115 void convert_to_git_filter_fd(const struct index_state *istate,
1116                               const char *path, int fd, struct strbuf *dst,
1117                               enum safe_crlf checksafe)
1118 {
1119         struct conv_attrs ca;
1120         convert_attrs(&ca, path);
1121
1122         assert(ca.drv);
1123         assert(ca.drv->clean || ca.drv->process);
1124
1125         if (!apply_filter(path, NULL, 0, fd, dst, ca.drv, CAP_CLEAN))
1126                 die("%s: clean filter '%s' failed", path, ca.drv->name);
1127
1128         crlf_to_git(istate, path, dst->buf, dst->len, dst, ca.crlf_action, checksafe);
1129         ident_to_git(path, dst->buf, dst->len, dst, ca.ident);
1130 }
1131
1132 static int convert_to_working_tree_internal(const char *path, const char *src,
1133                                             size_t len, struct strbuf *dst,
1134                                             int normalizing)
1135 {
1136         int ret = 0, ret_filter = 0;
1137         struct conv_attrs ca;
1138
1139         convert_attrs(&ca, path);
1140
1141         ret |= ident_to_worktree(path, src, len, dst, ca.ident);
1142         if (ret) {
1143                 src = dst->buf;
1144                 len = dst->len;
1145         }
1146         /*
1147          * CRLF conversion can be skipped if normalizing, unless there
1148          * is a smudge or process filter (even if the process filter doesn't
1149          * support smudge).  The filters might expect CRLFs.
1150          */
1151         if ((ca.drv && (ca.drv->smudge || ca.drv->process)) || !normalizing) {
1152                 ret |= crlf_to_worktree(path, src, len, dst, ca.crlf_action);
1153                 if (ret) {
1154                         src = dst->buf;
1155                         len = dst->len;
1156                 }
1157         }
1158
1159         ret_filter = apply_filter(path, src, len, -1, dst, ca.drv, CAP_SMUDGE);
1160         if (!ret_filter && ca.drv && ca.drv->required)
1161                 die("%s: smudge filter %s failed", path, ca.drv->name);
1162
1163         return ret | ret_filter;
1164 }
1165
1166 int convert_to_working_tree(const char *path, const char *src, size_t len, struct strbuf *dst)
1167 {
1168         return convert_to_working_tree_internal(path, src, len, dst, 0);
1169 }
1170
1171 int renormalize_buffer(const struct index_state *istate, const char *path,
1172                        const char *src, size_t len, struct strbuf *dst)
1173 {
1174         int ret = convert_to_working_tree_internal(path, src, len, dst, 1);
1175         if (ret) {
1176                 src = dst->buf;
1177                 len = dst->len;
1178         }
1179         return ret | convert_to_git(istate, path, src, len, dst, SAFE_CRLF_RENORMALIZE);
1180 }
1181
1182 /*****************************************************************
1183  *
1184  * Streaming conversion support
1185  *
1186  *****************************************************************/
1187
1188 typedef int (*filter_fn)(struct stream_filter *,
1189                          const char *input, size_t *isize_p,
1190                          char *output, size_t *osize_p);
1191 typedef void (*free_fn)(struct stream_filter *);
1192
1193 struct stream_filter_vtbl {
1194         filter_fn filter;
1195         free_fn free;
1196 };
1197
1198 struct stream_filter {
1199         struct stream_filter_vtbl *vtbl;
1200 };
1201
1202 static int null_filter_fn(struct stream_filter *filter,
1203                           const char *input, size_t *isize_p,
1204                           char *output, size_t *osize_p)
1205 {
1206         size_t count;
1207
1208         if (!input)
1209                 return 0; /* we do not keep any states */
1210         count = *isize_p;
1211         if (*osize_p < count)
1212                 count = *osize_p;
1213         if (count) {
1214                 memmove(output, input, count);
1215                 *isize_p -= count;
1216                 *osize_p -= count;
1217         }
1218         return 0;
1219 }
1220
1221 static void null_free_fn(struct stream_filter *filter)
1222 {
1223         ; /* nothing -- null instances are shared */
1224 }
1225
1226 static struct stream_filter_vtbl null_vtbl = {
1227         null_filter_fn,
1228         null_free_fn,
1229 };
1230
1231 static struct stream_filter null_filter_singleton = {
1232         &null_vtbl,
1233 };
1234
1235 int is_null_stream_filter(struct stream_filter *filter)
1236 {
1237         return filter == &null_filter_singleton;
1238 }
1239
1240
1241 /*
1242  * LF-to-CRLF filter
1243  */
1244
1245 struct lf_to_crlf_filter {
1246         struct stream_filter filter;
1247         unsigned has_held:1;
1248         char held;
1249 };
1250
1251 static int lf_to_crlf_filter_fn(struct stream_filter *filter,
1252                                 const char *input, size_t *isize_p,
1253                                 char *output, size_t *osize_p)
1254 {
1255         size_t count, o = 0;
1256         struct lf_to_crlf_filter *lf_to_crlf = (struct lf_to_crlf_filter *)filter;
1257
1258         /*
1259          * We may be holding onto the CR to see if it is followed by a
1260          * LF, in which case we would need to go to the main loop.
1261          * Otherwise, just emit it to the output stream.
1262          */
1263         if (lf_to_crlf->has_held && (lf_to_crlf->held != '\r' || !input)) {
1264                 output[o++] = lf_to_crlf->held;
1265                 lf_to_crlf->has_held = 0;
1266         }
1267
1268         /* We are told to drain */
1269         if (!input) {
1270                 *osize_p -= o;
1271                 return 0;
1272         }
1273
1274         count = *isize_p;
1275         if (count || lf_to_crlf->has_held) {
1276                 size_t i;
1277                 int was_cr = 0;
1278
1279                 if (lf_to_crlf->has_held) {
1280                         was_cr = 1;
1281                         lf_to_crlf->has_held = 0;
1282                 }
1283
1284                 for (i = 0; o < *osize_p && i < count; i++) {
1285                         char ch = input[i];
1286
1287                         if (ch == '\n') {
1288                                 output[o++] = '\r';
1289                         } else if (was_cr) {
1290                                 /*
1291                                  * Previous round saw CR and it is not followed
1292                                  * by a LF; emit the CR before processing the
1293                                  * current character.
1294                                  */
1295                                 output[o++] = '\r';
1296                         }
1297
1298                         /*
1299                          * We may have consumed the last output slot,
1300                          * in which case we need to break out of this
1301                          * loop; hold the current character before
1302                          * returning.
1303                          */
1304                         if (*osize_p <= o) {
1305                                 lf_to_crlf->has_held = 1;
1306                                 lf_to_crlf->held = ch;
1307                                 continue; /* break but increment i */
1308                         }
1309
1310                         if (ch == '\r') {
1311                                 was_cr = 1;
1312                                 continue;
1313                         }
1314
1315                         was_cr = 0;
1316                         output[o++] = ch;
1317                 }
1318
1319                 *osize_p -= o;
1320                 *isize_p -= i;
1321
1322                 if (!lf_to_crlf->has_held && was_cr) {
1323                         lf_to_crlf->has_held = 1;
1324                         lf_to_crlf->held = '\r';
1325                 }
1326         }
1327         return 0;
1328 }
1329
1330 static void lf_to_crlf_free_fn(struct stream_filter *filter)
1331 {
1332         free(filter);
1333 }
1334
1335 static struct stream_filter_vtbl lf_to_crlf_vtbl = {
1336         lf_to_crlf_filter_fn,
1337         lf_to_crlf_free_fn,
1338 };
1339
1340 static struct stream_filter *lf_to_crlf_filter(void)
1341 {
1342         struct lf_to_crlf_filter *lf_to_crlf = xcalloc(1, sizeof(*lf_to_crlf));
1343
1344         lf_to_crlf->filter.vtbl = &lf_to_crlf_vtbl;
1345         return (struct stream_filter *)lf_to_crlf;
1346 }
1347
1348 /*
1349  * Cascade filter
1350  */
1351 #define FILTER_BUFFER 1024
1352 struct cascade_filter {
1353         struct stream_filter filter;
1354         struct stream_filter *one;
1355         struct stream_filter *two;
1356         char buf[FILTER_BUFFER];
1357         int end, ptr;
1358 };
1359
1360 static int cascade_filter_fn(struct stream_filter *filter,
1361                              const char *input, size_t *isize_p,
1362                              char *output, size_t *osize_p)
1363 {
1364         struct cascade_filter *cas = (struct cascade_filter *) filter;
1365         size_t filled = 0;
1366         size_t sz = *osize_p;
1367         size_t to_feed, remaining;
1368
1369         /*
1370          * input -- (one) --> buf -- (two) --> output
1371          */
1372         while (filled < sz) {
1373                 remaining = sz - filled;
1374
1375                 /* do we already have something to feed two with? */
1376                 if (cas->ptr < cas->end) {
1377                         to_feed = cas->end - cas->ptr;
1378                         if (stream_filter(cas->two,
1379                                           cas->buf + cas->ptr, &to_feed,
1380                                           output + filled, &remaining))
1381                                 return -1;
1382                         cas->ptr += (cas->end - cas->ptr) - to_feed;
1383                         filled = sz - remaining;
1384                         continue;
1385                 }
1386
1387                 /* feed one from upstream and have it emit into our buffer */
1388                 to_feed = input ? *isize_p : 0;
1389                 if (input && !to_feed)
1390                         break;
1391                 remaining = sizeof(cas->buf);
1392                 if (stream_filter(cas->one,
1393                                   input, &to_feed,
1394                                   cas->buf, &remaining))
1395                         return -1;
1396                 cas->end = sizeof(cas->buf) - remaining;
1397                 cas->ptr = 0;
1398                 if (input) {
1399                         size_t fed = *isize_p - to_feed;
1400                         *isize_p -= fed;
1401                         input += fed;
1402                 }
1403
1404                 /* do we know that we drained one completely? */
1405                 if (input || cas->end)
1406                         continue;
1407
1408                 /* tell two to drain; we have nothing more to give it */
1409                 to_feed = 0;
1410                 remaining = sz - filled;
1411                 if (stream_filter(cas->two,
1412                                   NULL, &to_feed,
1413                                   output + filled, &remaining))
1414                         return -1;
1415                 if (remaining == (sz - filled))
1416                         break; /* completely drained two */
1417                 filled = sz - remaining;
1418         }
1419         *osize_p -= filled;
1420         return 0;
1421 }
1422
1423 static void cascade_free_fn(struct stream_filter *filter)
1424 {
1425         struct cascade_filter *cas = (struct cascade_filter *)filter;
1426         free_stream_filter(cas->one);
1427         free_stream_filter(cas->two);
1428         free(filter);
1429 }
1430
1431 static struct stream_filter_vtbl cascade_vtbl = {
1432         cascade_filter_fn,
1433         cascade_free_fn,
1434 };
1435
1436 static struct stream_filter *cascade_filter(struct stream_filter *one,
1437                                             struct stream_filter *two)
1438 {
1439         struct cascade_filter *cascade;
1440
1441         if (!one || is_null_stream_filter(one))
1442                 return two;
1443         if (!two || is_null_stream_filter(two))
1444                 return one;
1445
1446         cascade = xmalloc(sizeof(*cascade));
1447         cascade->one = one;
1448         cascade->two = two;
1449         cascade->end = cascade->ptr = 0;
1450         cascade->filter.vtbl = &cascade_vtbl;
1451         return (struct stream_filter *)cascade;
1452 }
1453
1454 /*
1455  * ident filter
1456  */
1457 #define IDENT_DRAINING (-1)
1458 #define IDENT_SKIPPING (-2)
1459 struct ident_filter {
1460         struct stream_filter filter;
1461         struct strbuf left;
1462         int state;
1463         char ident[45]; /* ": x40 $" */
1464 };
1465
1466 static int is_foreign_ident(const char *str)
1467 {
1468         int i;
1469
1470         if (!skip_prefix(str, "$Id: ", &str))
1471                 return 0;
1472         for (i = 0; str[i]; i++) {
1473                 if (isspace(str[i]) && str[i+1] != '$')
1474                         return 1;
1475         }
1476         return 0;
1477 }
1478
1479 static void ident_drain(struct ident_filter *ident, char **output_p, size_t *osize_p)
1480 {
1481         size_t to_drain = ident->left.len;
1482
1483         if (*osize_p < to_drain)
1484                 to_drain = *osize_p;
1485         if (to_drain) {
1486                 memcpy(*output_p, ident->left.buf, to_drain);
1487                 strbuf_remove(&ident->left, 0, to_drain);
1488                 *output_p += to_drain;
1489                 *osize_p -= to_drain;
1490         }
1491         if (!ident->left.len)
1492                 ident->state = 0;
1493 }
1494
1495 static int ident_filter_fn(struct stream_filter *filter,
1496                            const char *input, size_t *isize_p,
1497                            char *output, size_t *osize_p)
1498 {
1499         struct ident_filter *ident = (struct ident_filter *)filter;
1500         static const char head[] = "$Id";
1501
1502         if (!input) {
1503                 /* drain upon eof */
1504                 switch (ident->state) {
1505                 default:
1506                         strbuf_add(&ident->left, head, ident->state);
1507                 case IDENT_SKIPPING:
1508                         /* fallthru */
1509                 case IDENT_DRAINING:
1510                         ident_drain(ident, &output, osize_p);
1511                 }
1512                 return 0;
1513         }
1514
1515         while (*isize_p || (ident->state == IDENT_DRAINING)) {
1516                 int ch;
1517
1518                 if (ident->state == IDENT_DRAINING) {
1519                         ident_drain(ident, &output, osize_p);
1520                         if (!*osize_p)
1521                                 break;
1522                         continue;
1523                 }
1524
1525                 ch = *(input++);
1526                 (*isize_p)--;
1527
1528                 if (ident->state == IDENT_SKIPPING) {
1529                         /*
1530                          * Skipping until '$' or LF, but keeping them
1531                          * in case it is a foreign ident.
1532                          */
1533                         strbuf_addch(&ident->left, ch);
1534                         if (ch != '\n' && ch != '$')
1535                                 continue;
1536                         if (ch == '$' && !is_foreign_ident(ident->left.buf)) {
1537                                 strbuf_setlen(&ident->left, sizeof(head) - 1);
1538                                 strbuf_addstr(&ident->left, ident->ident);
1539                         }
1540                         ident->state = IDENT_DRAINING;
1541                         continue;
1542                 }
1543
1544                 if (ident->state < sizeof(head) &&
1545                     head[ident->state] == ch) {
1546                         ident->state++;
1547                         continue;
1548                 }
1549
1550                 if (ident->state)
1551                         strbuf_add(&ident->left, head, ident->state);
1552                 if (ident->state == sizeof(head) - 1) {
1553                         if (ch != ':' && ch != '$') {
1554                                 strbuf_addch(&ident->left, ch);
1555                                 ident->state = 0;
1556                                 continue;
1557                         }
1558
1559                         if (ch == ':') {
1560                                 strbuf_addch(&ident->left, ch);
1561                                 ident->state = IDENT_SKIPPING;
1562                         } else {
1563                                 strbuf_addstr(&ident->left, ident->ident);
1564                                 ident->state = IDENT_DRAINING;
1565                         }
1566                         continue;
1567                 }
1568
1569                 strbuf_addch(&ident->left, ch);
1570                 ident->state = IDENT_DRAINING;
1571         }
1572         return 0;
1573 }
1574
1575 static void ident_free_fn(struct stream_filter *filter)
1576 {
1577         struct ident_filter *ident = (struct ident_filter *)filter;
1578         strbuf_release(&ident->left);
1579         free(filter);
1580 }
1581
1582 static struct stream_filter_vtbl ident_vtbl = {
1583         ident_filter_fn,
1584         ident_free_fn,
1585 };
1586
1587 static struct stream_filter *ident_filter(const unsigned char *sha1)
1588 {
1589         struct ident_filter *ident = xmalloc(sizeof(*ident));
1590
1591         xsnprintf(ident->ident, sizeof(ident->ident),
1592                   ": %s $", sha1_to_hex(sha1));
1593         strbuf_init(&ident->left, 0);
1594         ident->filter.vtbl = &ident_vtbl;
1595         ident->state = 0;
1596         return (struct stream_filter *)ident;
1597 }
1598
1599 /*
1600  * Return an appropriately constructed filter for the path, or NULL if
1601  * the contents cannot be filtered without reading the whole thing
1602  * in-core.
1603  *
1604  * Note that you would be crazy to set CRLF, smuge/clean or ident to a
1605  * large binary blob you would want us not to slurp into the memory!
1606  */
1607 struct stream_filter *get_stream_filter(const char *path, const unsigned char *sha1)
1608 {
1609         struct conv_attrs ca;
1610         struct stream_filter *filter = NULL;
1611
1612         convert_attrs(&ca, path);
1613         if (ca.drv && (ca.drv->process || ca.drv->smudge || ca.drv->clean))
1614                 return NULL;
1615
1616         if (ca.crlf_action == CRLF_AUTO || ca.crlf_action == CRLF_AUTO_CRLF)
1617                 return NULL;
1618
1619         if (ca.ident)
1620                 filter = ident_filter(sha1);
1621
1622         if (output_eol(ca.crlf_action) == EOL_CRLF)
1623                 filter = cascade_filter(filter, lf_to_crlf_filter());
1624         else
1625                 filter = cascade_filter(filter, &null_filter_singleton);
1626
1627         return filter;
1628 }
1629
1630 void free_stream_filter(struct stream_filter *filter)
1631 {
1632         filter->vtbl->free(filter);
1633 }
1634
1635 int stream_filter(struct stream_filter *filter,
1636                   const char *input, size_t *isize_p,
1637                   char *output, size_t *osize_p)
1638 {
1639         return filter->vtbl->filter(filter, input, isize_p, output, osize_p);
1640 }