git.oblomov.eu Git - git/blob - 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, 0);
 587                 entry = NULL;
 588         } else {
 589                 entry = (struct cmd2process *)subprocess_find_entry(&subprocess_map, cmd);
 590         }
 591
 592         fflush(NULL);
 593
 594         if (!entry) {
 595                 entry = xmalloc(sizeof(*entry));
 596                 entry->supported_capabilities = 0;
 597
 598                 if (subprocess_start(&subprocess_map, &entry->subprocess, cmd, start_multi_file_filter_fn)) {
 599                         free(entry);
 600                         return 0;
 601                 }
 602         }
 603         process = &entry->subprocess.process;
 604
 605         if (!(wanted_capability & entry->supported_capabilities))
 606                 return 0;
 607
 608         if (CAP_CLEAN & wanted_capability)
 609                 filter_type = "clean";
 610         else if (CAP_SMUDGE & wanted_capability)
 611                 filter_type = "smudge";
 612         else
 613                 die("unexpected filter type");
 614
 615         sigchain_push(SIGPIPE, SIG_IGN);
 616
 617         assert(strlen(filter_type) < LARGE_PACKET_DATA_MAX - strlen("command=\n"));
 618         err = packet_write_fmt_gently(process->in, "command=%s\n", filter_type);
 619         if (err)
 620                 goto done;
 621
 622         err = strlen(path) > LARGE_PACKET_DATA_MAX - strlen("pathname=\n");
 623         if (err) {
 624                 error("path name too long for external filter");
 625                 goto done;
 626         }
 627
 628         err = packet_write_fmt_gently(process->in, "pathname=%s\n", path);
 629         if (err)
 630                 goto done;
 631
 632         err = packet_flush_gently(process->in);
 633         if (err)
 634                 goto done;
 635
 636         if (fd >= 0)
 637                 err = write_packetized_from_fd(fd, process->in);
 638         else
 639                 err = write_packetized_from_buf(src, len, process->in);
 640         if (err)
 641                 goto done;
 642
 643         err = subprocess_read_status(process->out, &filter_status);
 644         if (err)
 645                 goto done;
 646
 647         err = strcmp(filter_status.buf, "success");
 648         if (err)
 649                 goto done;
 650
 651         err = read_packetized_to_strbuf(process->out, &nbuf) < 0;
 652         if (err)
 653                 goto done;
 654
 655         err = subprocess_read_status(process->out, &filter_status);
 656         if (err)
 657                 goto done;
 658
 659         err = strcmp(filter_status.buf, "success");
 660
 661 done:
 662         sigchain_pop(SIGPIPE);
 663
 664         if (err) {
 665                 if (!strcmp(filter_status.buf, "error")) {
 666                         /* The filter signaled a problem with the file. */
 667                 } else if (!strcmp(filter_status.buf, "abort")) {
 668                         /*
 669                          * The filter signaled a permanent problem. Don't try to filter
 670                          * files with the same command for the lifetime of the current
 671                          * Git process.
 672                          */
 673                          entry->supported_capabilities &= ~wanted_capability;
 674                 } else {
 675                         /*
 676                          * Something went wrong with the protocol filter.
 677                          * Force shutdown and restart if another blob requires filtering.
 678                          */
 679                         error("external filter '%s' failed", cmd);
 680                         subprocess_stop(&subprocess_map, &entry->subprocess);
 681                         free(entry);
 682                 }
 683         } else {
 684                 strbuf_swap(dst, &nbuf);
 685         }
 686         strbuf_release(&nbuf);
 687         return !err;
 688 }
 689
 690 static struct convert_driver {
 691         const char *name;
 692         struct convert_driver *next;
 693         const char *smudge;
 694         const char *clean;
 695         const char *process;
 696         int required;
 697 } *user_convert, **user_convert_tail;
 698
 699 static int apply_filter(const char *path, const char *src, size_t len,
 700                         int fd, struct strbuf *dst, struct convert_driver *drv,
 701                         const unsigned int wanted_capability)
 702 {
 703         const char *cmd = NULL;
 704
 705         if (!drv)
 706                 return 0;
 707
 708         if (!dst)
 709                 return 1;
 710
 711         if ((CAP_CLEAN & wanted_capability) && !drv->process && drv->clean)
 712                 cmd = drv->clean;
 713         else if ((CAP_SMUDGE & wanted_capability) && !drv->process && drv->smudge)
 714                 cmd = drv->smudge;
 715
 716         if (cmd && *cmd)
 717                 return apply_single_file_filter(path, src, len, fd, dst, cmd);
 718         else if (drv->process && *drv->process)
 719                 return apply_multi_file_filter(path, src, len, fd, dst, drv->process, wanted_capability);
 720
 721         return 0;
 722 }
 723
 724 static int read_convert_config(const char *var, const char *value, void *cb)
 725 {
 726         const char *key, *name;
 727         int namelen;
 728         struct convert_driver *drv;
 729
 730         /*
 731          * External conversion drivers are configured using
 732          * "filter.<name>.variable".
 733          */
 734         if (parse_config_key(var, "filter", &name, &namelen, &key) < 0 || !name)
 735                 return 0;
 736         for (drv = user_convert; drv; drv = drv->next)
 737                 if (!strncmp(drv->name, name, namelen) && !drv->name[namelen])
 738                         break;
 739         if (!drv) {
 740                 drv = xcalloc(1, sizeof(struct convert_driver));
 741                 drv->name = xmemdupz(name, namelen);
 742                 *user_convert_tail = drv;
 743                 user_convert_tail = &(drv->next);
 744         }
 745
 746         /*
 747          * filter.<name>.smudge and filter.<name>.clean specifies
 748          * the command line:
 749          *
 750          *      command-line
 751          *
 752          * The command-line will not be interpolated in any way.
 753          */
 754
 755         if (!strcmp("smudge", key))
 756                 return git_config_string(&drv->smudge, var, value);
 757
 758         if (!strcmp("clean", key))
 759                 return git_config_string(&drv->clean, var, value);
 760
 761         if (!strcmp("process", key))
 762                 return git_config_string(&drv->process, var, value);
 763
 764         if (!strcmp("required", key)) {
 765                 drv->required = git_config_bool(var, value);
 766                 return 0;
 767         }
 768
 769         return 0;
 770 }
 771
 772 static int count_ident(const char *cp, unsigned long size)
 773 {
 774         /*
 775          * "$Id: 0000000000000000000000000000000000000000 $" <=> "$Id$"
 776          */
 777         int cnt = 0;
 778         char ch;
 779
 780         while (size) {
 781                 ch = *cp++;
 782                 size--;
 783                 if (ch != '$')
 784                         continue;
 785                 if (size < 3)
 786                         break;
 787                 if (memcmp("Id", cp, 2))
 788                         continue;
 789                 ch = cp[2];
 790                 cp += 3;
 791                 size -= 3;
 792                 if (ch == '$')
 793                         cnt++; /* $Id$ */
 794                 if (ch != ':')
 795                         continue;
 796
 797                 /*
 798                  * "$Id: ... "; scan up to the closing dollar sign and discard.
 799                  */
 800                 while (size) {
 801                         ch = *cp++;
 802                         size--;
 803                         if (ch == '$') {
 804                                 cnt++;
 805                                 break;
 806                         }
 807                         if (ch == '\n')
 808                                 break;
 809                 }
 810         }
 811         return cnt;
 812 }
 813
 814 static int ident_to_git(const char *path, const char *src, size_t len,
 815                         struct strbuf *buf, int ident)
 816 {
 817         char *dst, *dollar;
 818
 819         if (!ident || (src && !count_ident(src, len)))
 820                 return 0;
 821
 822         if (!buf)
 823                 return 1;
 824
 825         /* only grow if not in place */
 826         if (strbuf_avail(buf) + buf->len < len)
 827                 strbuf_grow(buf, len - buf->len);
 828         dst = buf->buf;
 829         for (;;) {
 830                 dollar = memchr(src, '$', len);
 831                 if (!dollar)
 832                         break;
 833                 memmove(dst, src, dollar + 1 - src);
 834                 dst += dollar + 1 - src;
 835                 len -= dollar + 1 - src;
 836                 src  = dollar + 1;
 837
 838                 if (len > 3 && !memcmp(src, "Id:", 3)) {
 839                         dollar = memchr(src + 3, '$', len - 3);
 840                         if (!dollar)
 841                                 break;
 842                         if (memchr(src + 3, '\n', dollar - src - 3)) {
 843                                 /* Line break before the next dollar. */
 844                                 continue;
 845                         }
 846
 847                         memcpy(dst, "Id$", 3);
 848                         dst += 3;
 849                         len -= dollar + 1 - src;
 850                         src  = dollar + 1;
 851                 }
 852         }
 853         memmove(dst, src, len);
 854         strbuf_setlen(buf, dst + len - buf->buf);
 855         return 1;
 856 }
 857
 858 static int ident_to_worktree(const char *path, const char *src, size_t len,
 859                              struct strbuf *buf, int ident)
 860 {
 861         unsigned char sha1[20];
 862         char *to_free = NULL, *dollar, *spc;
 863         int cnt;
 864
 865         if (!ident)
 866                 return 0;
 867
 868         cnt = count_ident(src, len);
 869         if (!cnt)
 870                 return 0;
 871
 872         /* are we "faking" in place editing ? */
 873         if (src == buf->buf)
 874                 to_free = strbuf_detach(buf, NULL);
 875         hash_sha1_file(src, len, "blob", sha1);
 876
 877         strbuf_grow(buf, len + cnt * 43);
 878         for (;;) {
 879                 /* step 1: run to the next '$' */
 880                 dollar = memchr(src, '$', len);
 881                 if (!dollar)
 882                         break;
 883                 strbuf_add(buf, src, dollar + 1 - src);
 884                 len -= dollar + 1 - src;
 885                 src  = dollar + 1;
 886
 887                 /* step 2: does it looks like a bit like Id:xxx$ or Id$ ? */
 888                 if (len < 3 || memcmp("Id", src, 2))
 889                         continue;
 890
 891                 /* step 3: skip over Id$ or Id:xxxxx$ */
 892                 if (src[2] == '$') {
 893                         src += 3;
 894                         len -= 3;
 895                 } else if (src[2] == ':') {
 896                         /*
 897                          * It's possible that an expanded Id has crept its way into the
 898                          * repository, we cope with that by stripping the expansion out.
 899                          * This is probably not a good idea, since it will cause changes
 900                          * on checkout, which won't go away by stash, but let's keep it
 901                          * for git-style ids.
 902                          */
 903                         dollar = memchr(src + 3, '$', len - 3);
 904                         if (!dollar) {
 905                                 /* incomplete keyword, no more '$', so just quit the loop */
 906                                 break;
 907                         }
 908
 909                         if (memchr(src + 3, '\n', dollar - src - 3)) {
 910                                 /* Line break before the next dollar. */
 911                                 continue;
 912                         }
 913
 914                         spc = memchr(src + 4, ' ', dollar - src - 4);
 915                         if (spc && spc < dollar-1) {
 916                                 /* There are spaces in unexpected places.
 917                                  * This is probably an id from some other
 918                                  * versioning system. Keep it for now.
 919                                  */
 920                                 continue;
 921                         }
 922
 923                         len -= dollar + 1 - src;
 924                         src  = dollar + 1;
 925                 } else {
 926                         /* it wasn't a "Id$" or "Id:xxxx$" */
 927                         continue;
 928                 }
 929
 930                 /* step 4: substitute */
 931                 strbuf_addstr(buf, "Id: ");
 932                 strbuf_add(buf, sha1_to_hex(sha1), 40);
 933                 strbuf_addstr(buf, " $");
 934         }
 935         strbuf_add(buf, src, len);
 936
 937         free(to_free);
 938         return 1;
 939 }
 940
 941 static enum crlf_action git_path_check_crlf(struct attr_check_item *check)
 942 {
 943         const char *value = check->value;
 944
 945         if (ATTR_TRUE(value))
 946                 return CRLF_TEXT;
 947         else if (ATTR_FALSE(value))
 948                 return CRLF_BINARY;
 949         else if (ATTR_UNSET(value))
 950                 ;
 951         else if (!strcmp(value, "input"))
 952                 return CRLF_TEXT_INPUT;
 953         else if (!strcmp(value, "auto"))
 954                 return CRLF_AUTO;
 955         return CRLF_UNDEFINED;
 956 }
 957
 958 static enum eol git_path_check_eol(struct attr_check_item *check)
 959 {
 960         const char *value = check->value;
 961
 962         if (ATTR_UNSET(value))
 963                 ;
 964         else if (!strcmp(value, "lf"))
 965                 return EOL_LF;
 966         else if (!strcmp(value, "crlf"))
 967                 return EOL_CRLF;
 968         return EOL_UNSET;
 969 }
 970
 971 static struct convert_driver *git_path_check_convert(struct attr_check_item *check)
 972 {
 973         const char *value = check->value;
 974         struct convert_driver *drv;
 975
 976         if (ATTR_TRUE(value) || ATTR_FALSE(value) || ATTR_UNSET(value))
 977                 return NULL;
 978         for (drv = user_convert; drv; drv = drv->next)
 979                 if (!strcmp(value, drv->name))
 980                         return drv;
 981         return NULL;
 982 }
 983
 984 static int git_path_check_ident(struct attr_check_item *check)
 985 {
 986         const char *value = check->value;
 987
 988         return !!ATTR_TRUE(value);
 989 }
 990
 991 struct conv_attrs {
 992         struct convert_driver *drv;
 993         enum crlf_action attr_action; /* What attr says */
 994         enum crlf_action crlf_action; /* When no attr is set, use core.autocrlf */
 995         int ident;
 996 };
 997
 998 static void convert_attrs(struct conv_attrs *ca, const char *path)
 999 {
1000         static struct attr_check *check;
1001
1002         if (!check) {
1003                 check = attr_check_initl("crlf", "ident", "filter",
1004                                          "eol", "text", NULL);
1005                 user_convert_tail = &user_convert;
1006                 git_config(read_convert_config, NULL);
1007         }
1008
1009         if (!git_check_attr(path, check)) {
1010                 struct attr_check_item *ccheck = check->items;
1011                 ca->crlf_action = git_path_check_crlf(ccheck + 4);
1012                 if (ca->crlf_action == CRLF_UNDEFINED)
1013                         ca->crlf_action = git_path_check_crlf(ccheck + 0);
1014                 ca->attr_action = ca->crlf_action;
1015                 ca->ident = git_path_check_ident(ccheck + 1);
1016                 ca->drv = git_path_check_convert(ccheck + 2);
1017                 if (ca->crlf_action != CRLF_BINARY) {
1018                         enum eol eol_attr = git_path_check_eol(ccheck + 3);
1019                         if (ca->crlf_action == CRLF_AUTO && eol_attr == EOL_LF)
1020                                 ca->crlf_action = CRLF_AUTO_INPUT;
1021                         else if (ca->crlf_action == CRLF_AUTO && eol_attr == EOL_CRLF)
1022                                 ca->crlf_action = CRLF_AUTO_CRLF;
1023                         else if (eol_attr == EOL_LF)
1024                                 ca->crlf_action = CRLF_TEXT_INPUT;
1025                         else if (eol_attr == EOL_CRLF)
1026                                 ca->crlf_action = CRLF_TEXT_CRLF;
1027                 }
1028                 ca->attr_action = ca->crlf_action;
1029         } else {
1030                 ca->drv = NULL;
1031                 ca->crlf_action = CRLF_UNDEFINED;
1032                 ca->ident = 0;
1033         }
1034         if (ca->crlf_action == CRLF_TEXT)
1035                 ca->crlf_action = text_eol_is_crlf() ? CRLF_TEXT_CRLF : CRLF_TEXT_INPUT;
1036         if (ca->crlf_action == CRLF_UNDEFINED && auto_crlf == AUTO_CRLF_FALSE)
1037                 ca->crlf_action = CRLF_BINARY;
1038         if (ca->crlf_action == CRLF_UNDEFINED && auto_crlf == AUTO_CRLF_TRUE)
1039                 ca->crlf_action = CRLF_AUTO_CRLF;
1040         if (ca->crlf_action == CRLF_UNDEFINED && auto_crlf == AUTO_CRLF_INPUT)
1041                 ca->crlf_action = CRLF_AUTO_INPUT;
1042 }
1043
1044 int would_convert_to_git_filter_fd(const char *path)
1045 {
1046         struct conv_attrs ca;
1047
1048         convert_attrs(&ca, path);
1049         if (!ca.drv)
1050                 return 0;
1051
1052         /*
1053          * Apply a filter to an fd only if the filter is required to succeed.
1054          * We must die if the filter fails, because the original data before
1055          * filtering is not available.
1056          */
1057         if (!ca.drv->required)
1058                 return 0;
1059
1060         return apply_filter(path, NULL, 0, -1, NULL, ca.drv, CAP_CLEAN);
1061 }
1062
1063 const char *get_convert_attr_ascii(const char *path)
1064 {
1065         struct conv_attrs ca;
1066
1067         convert_attrs(&ca, path);
1068         switch (ca.attr_action) {
1069         case CRLF_UNDEFINED:
1070                 return "";
1071         case CRLF_BINARY:
1072                 return "-text";
1073         case CRLF_TEXT:
1074                 return "text";
1075         case CRLF_TEXT_INPUT:
1076                 return "text eol=lf";
1077         case CRLF_TEXT_CRLF:
1078                 return "text eol=crlf";
1079         case CRLF_AUTO:
1080                 return "text=auto";
1081         case CRLF_AUTO_CRLF:
1082                 return "text=auto eol=crlf";
1083         case CRLF_AUTO_INPUT:
1084                 return "text=auto eol=lf";
1085         }
1086         return "";
1087 }
1088
1089 int convert_to_git(const struct index_state *istate,
1090                    const char *path, const char *src, size_t len,
1091                    struct strbuf *dst, enum safe_crlf checksafe)
1092 {
1093         int ret = 0;
1094         struct conv_attrs ca;
1095
1096         convert_attrs(&ca, path);
1097
1098         ret |= apply_filter(path, src, len, -1, dst, ca.drv, CAP_CLEAN);
1099         if (!ret && ca.drv && ca.drv->required)
1100                 die("%s: clean filter '%s' failed", path, ca.drv->name);
1101
1102         if (ret && dst) {
1103                 src = dst->buf;
1104                 len = dst->len;
1105         }
1106         ret |= crlf_to_git(istate, path, src, len, dst, ca.crlf_action, checksafe);
1107         if (ret && dst) {
1108                 src = dst->buf;
1109                 len = dst->len;
1110         }
1111         return ret | ident_to_git(path, src, len, dst, ca.ident);
1112 }
1113
1114 void convert_to_git_filter_fd(const struct index_state *istate,
1115                               const char *path, int fd, struct strbuf *dst,
1116                               enum safe_crlf checksafe)
1117 {
1118         struct conv_attrs ca;
1119         convert_attrs(&ca, path);
1120
1121         assert(ca.drv);
1122         assert(ca.drv->clean || ca.drv->process);
1123
1124         if (!apply_filter(path, NULL, 0, fd, dst, ca.drv, CAP_CLEAN))
1125                 die("%s: clean filter '%s' failed", path, ca.drv->name);
1126
1127         crlf_to_git(istate, path, dst->buf, dst->len, dst, ca.crlf_action, checksafe);
1128         ident_to_git(path, dst->buf, dst->len, dst, ca.ident);
1129 }
1130
1131 static int convert_to_working_tree_internal(const char *path, const char *src,
1132                                             size_t len, struct strbuf *dst,
1133                                             int normalizing)
1134 {
1135         int ret = 0, ret_filter = 0;
1136         struct conv_attrs ca;
1137
1138         convert_attrs(&ca, path);
1139
1140         ret |= ident_to_worktree(path, src, len, dst, ca.ident);
1141         if (ret) {
1142                 src = dst->buf;
1143                 len = dst->len;
1144         }
1145         /*
1146          * CRLF conversion can be skipped if normalizing, unless there
1147          * is a smudge or process filter (even if the process filter doesn't
1148          * support smudge).  The filters might expect CRLFs.
1149          */
1150         if ((ca.drv && (ca.drv->smudge || ca.drv->process)) || !normalizing) {
1151                 ret |= crlf_to_worktree(path, src, len, dst, ca.crlf_action);
1152                 if (ret) {
1153                         src = dst->buf;
1154                         len = dst->len;
1155                 }
1156         }
1157
1158         ret_filter = apply_filter(path, src, len, -1, dst, ca.drv, CAP_SMUDGE);
1159         if (!ret_filter && ca.drv && ca.drv->required)
1160                 die("%s: smudge filter %s failed", path, ca.drv->name);
1161
1162         return ret | ret_filter;
1163 }
1164
1165 int convert_to_working_tree(const char *path, const char *src, size_t len, struct strbuf *dst)
1166 {
1167         return convert_to_working_tree_internal(path, src, len, dst, 0);
1168 }
1169
1170 int renormalize_buffer(const struct index_state *istate, const char *path,
1171                        const char *src, size_t len, struct strbuf *dst)
1172 {
1173         int ret = convert_to_working_tree_internal(path, src, len, dst, 1);
1174         if (ret) {
1175                 src = dst->buf;
1176                 len = dst->len;
1177         }
1178         return ret | convert_to_git(istate, path, src, len, dst, SAFE_CRLF_RENORMALIZE);
1179 }
1180
1181 /*****************************************************************
1182  *
1183  * Streaming conversion support
1184  *
1185  *****************************************************************/
1186
1187 typedef int (*filter_fn)(struct stream_filter *,
1188                          const char *input, size_t *isize_p,
1189                          char *output, size_t *osize_p);
1190 typedef void (*free_fn)(struct stream_filter *);
1191
1192 struct stream_filter_vtbl {
1193         filter_fn filter;
1194         free_fn free;
1195 };
1196
1197 struct stream_filter {
1198         struct stream_filter_vtbl *vtbl;
1199 };
1200
1201 static int null_filter_fn(struct stream_filter *filter,
1202                           const char *input, size_t *isize_p,
1203                           char *output, size_t *osize_p)
1204 {
1205         size_t count;
1206
1207         if (!input)
1208                 return 0; /* we do not keep any states */
1209         count = *isize_p;
1210         if (*osize_p < count)
1211                 count = *osize_p;
1212         if (count) {
1213                 memmove(output, input, count);
1214                 *isize_p -= count;
1215                 *osize_p -= count;
1216         }
1217         return 0;
1218 }
1219
1220 static void null_free_fn(struct stream_filter *filter)
1221 {
1222         ; /* nothing -- null instances are shared */
1223 }
1224
1225 static struct stream_filter_vtbl null_vtbl = {
1226         null_filter_fn,
1227         null_free_fn,
1228 };
1229
1230 static struct stream_filter null_filter_singleton = {
1231         &null_vtbl,
1232 };
1233
1234 int is_null_stream_filter(struct stream_filter *filter)
1235 {
1236         return filter == &null_filter_singleton;
1237 }
1238
1239
1240 /*
1241  * LF-to-CRLF filter
1242  */
1243
1244 struct lf_to_crlf_filter {
1245         struct stream_filter filter;
1246         unsigned has_held:1;
1247         char held;
1248 };
1249
1250 static int lf_to_crlf_filter_fn(struct stream_filter *filter,
1251                                 const char *input, size_t *isize_p,
1252                                 char *output, size_t *osize_p)
1253 {
1254         size_t count, o = 0;
1255         struct lf_to_crlf_filter *lf_to_crlf = (struct lf_to_crlf_filter *)filter;
1256
1257         /*
1258          * We may be holding onto the CR to see if it is followed by a
1259          * LF, in which case we would need to go to the main loop.
1260          * Otherwise, just emit it to the output stream.
1261          */
1262         if (lf_to_crlf->has_held && (lf_to_crlf->held != '\r' || !input)) {
1263                 output[o++] = lf_to_crlf->held;
1264                 lf_to_crlf->has_held = 0;
1265         }
1266
1267         /* We are told to drain */
1268         if (!input) {
1269                 *osize_p -= o;
1270                 return 0;
1271         }
1272
1273         count = *isize_p;
1274         if (count || lf_to_crlf->has_held) {
1275                 size_t i;
1276                 int was_cr = 0;
1277
1278                 if (lf_to_crlf->has_held) {
1279                         was_cr = 1;
1280                         lf_to_crlf->has_held = 0;
1281                 }
1282
1283                 for (i = 0; o < *osize_p && i < count; i++) {
1284                         char ch = input[i];
1285
1286                         if (ch == '\n') {
1287                                 output[o++] = '\r';
1288                         } else if (was_cr) {
1289                                 /*
1290                                  * Previous round saw CR and it is not followed
1291                                  * by a LF; emit the CR before processing the
1292                                  * current character.
1293                                  */
1294                                 output[o++] = '\r';
1295                         }
1296
1297                         /*
1298                          * We may have consumed the last output slot,
1299                          * in which case we need to break out of this
1300                          * loop; hold the current character before
1301                          * returning.
1302                          */
1303                         if (*osize_p <= o) {
1304                                 lf_to_crlf->has_held = 1;
1305                                 lf_to_crlf->held = ch;
1306                                 continue; /* break but increment i */
1307                         }
1308
1309                         if (ch == '\r') {
1310                                 was_cr = 1;
1311                                 continue;
1312                         }
1313
1314                         was_cr = 0;
1315                         output[o++] = ch;
1316                 }
1317
1318                 *osize_p -= o;
1319                 *isize_p -= i;
1320
1321                 if (!lf_to_crlf->has_held && was_cr) {
1322                         lf_to_crlf->has_held = 1;
1323                         lf_to_crlf->held = '\r';
1324                 }
1325         }
1326         return 0;
1327 }
1328
1329 static void lf_to_crlf_free_fn(struct stream_filter *filter)
1330 {
1331         free(filter);
1332 }
1333
1334 static struct stream_filter_vtbl lf_to_crlf_vtbl = {
1335         lf_to_crlf_filter_fn,
1336         lf_to_crlf_free_fn,
1337 };
1338
1339 static struct stream_filter *lf_to_crlf_filter(void)
1340 {
1341         struct lf_to_crlf_filter *lf_to_crlf = xcalloc(1, sizeof(*lf_to_crlf));
1342
1343         lf_to_crlf->filter.vtbl = &lf_to_crlf_vtbl;
1344         return (struct stream_filter *)lf_to_crlf;
1345 }
1346
1347 /*
1348  * Cascade filter
1349  */
1350 #define FILTER_BUFFER 1024
1351 struct cascade_filter {
1352         struct stream_filter filter;
1353         struct stream_filter *one;
1354         struct stream_filter *two;
1355         char buf[FILTER_BUFFER];
1356         int end, ptr;
1357 };
1358
1359 static int cascade_filter_fn(struct stream_filter *filter,
1360                              const char *input, size_t *isize_p,
1361                              char *output, size_t *osize_p)
1362 {
1363         struct cascade_filter *cas = (struct cascade_filter *) filter;
1364         size_t filled = 0;
1365         size_t sz = *osize_p;
1366         size_t to_feed, remaining;
1367
1368         /*
1369          * input -- (one) --> buf -- (two) --> output
1370          */
1371         while (filled < sz) {
1372                 remaining = sz - filled;
1373
1374                 /* do we already have something to feed two with? */
1375                 if (cas->ptr < cas->end) {
1376                         to_feed = cas->end - cas->ptr;
1377                         if (stream_filter(cas->two,
1378                                           cas->buf + cas->ptr, &to_feed,
1379                                           output + filled, &remaining))
1380                                 return -1;
1381                         cas->ptr += (cas->end - cas->ptr) - to_feed;
1382                         filled = sz - remaining;
1383                         continue;
1384                 }
1385
1386                 /* feed one from upstream and have it emit into our buffer */
1387                 to_feed = input ? *isize_p : 0;
1388                 if (input && !to_feed)
1389                         break;
1390                 remaining = sizeof(cas->buf);
1391                 if (stream_filter(cas->one,
1392                                   input, &to_feed,
1393                                   cas->buf, &remaining))
1394                         return -1;
1395                 cas->end = sizeof(cas->buf) - remaining;
1396                 cas->ptr = 0;
1397                 if (input) {
1398                         size_t fed = *isize_p - to_feed;
1399                         *isize_p -= fed;
1400                         input += fed;
1401                 }
1402
1403                 /* do we know that we drained one completely? */
1404                 if (input || cas->end)
1405                         continue;
1406
1407                 /* tell two to drain; we have nothing more to give it */
1408                 to_feed = 0;
1409                 remaining = sz - filled;
1410                 if (stream_filter(cas->two,
1411                                   NULL, &to_feed,
1412                                   output + filled, &remaining))
1413                         return -1;
1414                 if (remaining == (sz - filled))
1415                         break; /* completely drained two */
1416                 filled = sz - remaining;
1417         }
1418         *osize_p -= filled;
1419         return 0;
1420 }
1421
1422 static void cascade_free_fn(struct stream_filter *filter)
1423 {
1424         struct cascade_filter *cas = (struct cascade_filter *)filter;
1425         free_stream_filter(cas->one);
1426         free_stream_filter(cas->two);
1427         free(filter);
1428 }
1429
1430 static struct stream_filter_vtbl cascade_vtbl = {
1431         cascade_filter_fn,
1432         cascade_free_fn,
1433 };
1434
1435 static struct stream_filter *cascade_filter(struct stream_filter *one,
1436                                             struct stream_filter *two)
1437 {
1438         struct cascade_filter *cascade;
1439
1440         if (!one || is_null_stream_filter(one))
1441                 return two;
1442         if (!two || is_null_stream_filter(two))
1443                 return one;
1444
1445         cascade = xmalloc(sizeof(*cascade));
1446         cascade->one = one;
1447         cascade->two = two;
1448         cascade->end = cascade->ptr = 0;
1449         cascade->filter.vtbl = &cascade_vtbl;
1450         return (struct stream_filter *)cascade;
1451 }
1452
1453 /*
1454  * ident filter
1455  */
1456 #define IDENT_DRAINING (-1)
1457 #define IDENT_SKIPPING (-2)
1458 struct ident_filter {
1459         struct stream_filter filter;
1460         struct strbuf left;
1461         int state;
1462         char ident[45]; /* ": x40 $" */
1463 };
1464
1465 static int is_foreign_ident(const char *str)
1466 {
1467         int i;
1468
1469         if (!skip_prefix(str, "$Id: ", &str))
1470                 return 0;
1471         for (i = 0; str[i]; i++) {
1472                 if (isspace(str[i]) && str[i+1] != '$')
1473                         return 1;
1474         }
1475         return 0;
1476 }
1477
1478 static void ident_drain(struct ident_filter *ident, char **output_p, size_t *osize_p)
1479 {
1480         size_t to_drain = ident->left.len;
1481
1482         if (*osize_p < to_drain)
1483                 to_drain = *osize_p;
1484         if (to_drain) {
1485                 memcpy(*output_p, ident->left.buf, to_drain);
1486                 strbuf_remove(&ident->left, 0, to_drain);
1487                 *output_p += to_drain;
1488                 *osize_p -= to_drain;
1489         }
1490         if (!ident->left.len)
1491                 ident->state = 0;
1492 }
1493
1494 static int ident_filter_fn(struct stream_filter *filter,
1495                            const char *input, size_t *isize_p,
1496                            char *output, size_t *osize_p)
1497 {
1498         struct ident_filter *ident = (struct ident_filter *)filter;
1499         static const char head[] = "$Id";
1500
1501         if (!input) {
1502                 /* drain upon eof */
1503                 switch (ident->state) {
1504                 default:
1505                         strbuf_add(&ident->left, head, ident->state);
1506                 case IDENT_SKIPPING:
1507                         /* fallthru */
1508                 case IDENT_DRAINING:
1509                         ident_drain(ident, &output, osize_p);
1510                 }
1511                 return 0;
1512         }
1513
1514         while (*isize_p || (ident->state == IDENT_DRAINING)) {
1515                 int ch;
1516
1517                 if (ident->state == IDENT_DRAINING) {
1518                         ident_drain(ident, &output, osize_p);
1519                         if (!*osize_p)
1520                                 break;
1521                         continue;
1522                 }
1523
1524                 ch = *(input++);
1525                 (*isize_p)--;
1526
1527                 if (ident->state == IDENT_SKIPPING) {
1528                         /*
1529                          * Skipping until '$' or LF, but keeping them
1530                          * in case it is a foreign ident.
1531                          */
1532                         strbuf_addch(&ident->left, ch);
1533                         if (ch != '\n' && ch != '$')
1534                                 continue;
1535                         if (ch == '$' && !is_foreign_ident(ident->left.buf)) {
1536                                 strbuf_setlen(&ident->left, sizeof(head) - 1);
1537                                 strbuf_addstr(&ident->left, ident->ident);
1538                         }
1539                         ident->state = IDENT_DRAINING;
1540                         continue;
1541                 }
1542
1543                 if (ident->state < sizeof(head) &&
1544                     head[ident->state] == ch) {
1545                         ident->state++;
1546                         continue;
1547                 }
1548
1549                 if (ident->state)
1550                         strbuf_add(&ident->left, head, ident->state);
1551                 if (ident->state == sizeof(head) - 1) {
1552                         if (ch != ':' && ch != '$') {
1553                                 strbuf_addch(&ident->left, ch);
1554                                 ident->state = 0;
1555                                 continue;
1556                         }
1557
1558                         if (ch == ':') {
1559                                 strbuf_addch(&ident->left, ch);
1560                                 ident->state = IDENT_SKIPPING;
1561                         } else {
1562                                 strbuf_addstr(&ident->left, ident->ident);
1563                                 ident->state = IDENT_DRAINING;
1564                         }
1565                         continue;
1566                 }
1567
1568                 strbuf_addch(&ident->left, ch);
1569                 ident->state = IDENT_DRAINING;
1570         }
1571         return 0;
1572 }
1573
1574 static void ident_free_fn(struct stream_filter *filter)
1575 {
1576         struct ident_filter *ident = (struct ident_filter *)filter;
1577         strbuf_release(&ident->left);
1578         free(filter);
1579 }
1580
1581 static struct stream_filter_vtbl ident_vtbl = {
1582         ident_filter_fn,
1583         ident_free_fn,
1584 };
1585
1586 static struct stream_filter *ident_filter(const unsigned char *sha1)
1587 {
1588         struct ident_filter *ident = xmalloc(sizeof(*ident));
1589
1590         xsnprintf(ident->ident, sizeof(ident->ident),
1591                   ": %s $", sha1_to_hex(sha1));
1592         strbuf_init(&ident->left, 0);
1593         ident->filter.vtbl = &ident_vtbl;
1594         ident->state = 0;
1595         return (struct stream_filter *)ident;
1596 }
1597
1598 /*
1599  * Return an appropriately constructed filter for the path, or NULL if
1600  * the contents cannot be filtered without reading the whole thing
1601  * in-core.
1602  *
1603  * Note that you would be crazy to set CRLF, smuge/clean or ident to a
1604  * large binary blob you would want us not to slurp into the memory!
1605  */
1606 struct stream_filter *get_stream_filter(const char *path, const unsigned char *sha1)
1607 {
1608         struct conv_attrs ca;
1609         struct stream_filter *filter = NULL;
1610
1611         convert_attrs(&ca, path);
1612         if (ca.drv && (ca.drv->process || ca.drv->smudge || ca.drv->clean))
1613                 return NULL;
1614
1615         if (ca.crlf_action == CRLF_AUTO || ca.crlf_action == CRLF_AUTO_CRLF)
1616                 return NULL;
1617
1618         if (ca.ident)
1619                 filter = ident_filter(sha1);
1620
1621         if (output_eol(ca.crlf_action) == EOL_CRLF)
1622                 filter = cascade_filter(filter, lf_to_crlf_filter());
1623         else
1624                 filter = cascade_filter(filter, &null_filter_singleton);
1625
1626         return filter;
1627 }
1628
1629 void free_stream_filter(struct stream_filter *filter)
1630 {
1631         filter->vtbl->free(filter);
1632 }
1633
1634 int stream_filter(struct stream_filter *filter,
1635                   const char *input, size_t *isize_p,
1636                   char *output, size_t *osize_p)
1637 {
1638         return filter->vtbl->filter(filter, input, isize_p, output, osize_p);
1639 }