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 #define CAP_DELAY    (1u<<2)
 505
 506 struct cmd2process {
 507         struct subprocess_entry subprocess; /* must be the first member! */
 508         unsigned int supported_capabilities;
 509 };
 510
 511 static int subprocess_map_initialized;
 512 static struct hashmap subprocess_map;
 513
 514 static int start_multi_file_filter_fn(struct subprocess_entry *subprocess)
 515 {
 516         static int versions[] = {2, 0};
 517         static struct subprocess_capability capabilities[] = {
 518                 { "clean",  CAP_CLEAN  },
 519                 { "smudge", CAP_SMUDGE },
 520                 { "delay",  CAP_DELAY  },
 521                 { NULL, 0 }
 522         };
 523         struct cmd2process *entry = (struct cmd2process *)subprocess;
 524         return subprocess_handshake(subprocess, "git-filter", versions, NULL,
 525                                     capabilities,
 526                                     &entry->supported_capabilities);
 527 }
 528
 529 static void handle_filter_error(const struct strbuf *filter_status,
 530                                 struct cmd2process *entry,
 531                                 const unsigned int wanted_capability) {
 532         if (!strcmp(filter_status->buf, "error"))
 533                 ; /* The filter signaled a problem with the file. */
 534         else if (!strcmp(filter_status->buf, "abort") && wanted_capability) {
 535                 /*
 536                  * The filter signaled a permanent problem. Don't try to filter
 537                  * files with the same command for the lifetime of the current
 538                  * Git process.
 539                  */
 540                  entry->supported_capabilities &= ~wanted_capability;
 541         } else {
 542                 /*
 543                  * Something went wrong with the protocol filter.
 544                  * Force shutdown and restart if another blob requires filtering.
 545                  */
 546                 error("external filter '%s' failed", entry->subprocess.cmd);
 547                 subprocess_stop(&subprocess_map, &entry->subprocess);
 548                 free(entry);
 549         }
 550 }
 551
 552 static int apply_multi_file_filter(const char *path, const char *src, size_t len,
 553                                    int fd, struct strbuf *dst, const char *cmd,
 554                                    const unsigned int wanted_capability,
 555                                    struct delayed_checkout *dco)
 556 {
 557         int err;
 558         int can_delay = 0;
 559         struct cmd2process *entry;
 560         struct child_process *process;
 561         struct strbuf nbuf = STRBUF_INIT;
 562         struct strbuf filter_status = STRBUF_INIT;
 563         const char *filter_type;
 564
 565         if (!subprocess_map_initialized) {
 566                 subprocess_map_initialized = 1;
 567                 hashmap_init(&subprocess_map, cmd2process_cmp, NULL, 0);
 568                 entry = NULL;
 569         } else {
 570                 entry = (struct cmd2process *)subprocess_find_entry(&subprocess_map, cmd);
 571         }
 572
 573         fflush(NULL);
 574
 575         if (!entry) {
 576                 entry = xmalloc(sizeof(*entry));
 577                 entry->supported_capabilities = 0;
 578
 579                 if (subprocess_start(&subprocess_map, &entry->subprocess, cmd, start_multi_file_filter_fn)) {
 580                         free(entry);
 581                         return 0;
 582                 }
 583         }
 584         process = &entry->subprocess.process;
 585
 586         if (!(entry->supported_capabilities & wanted_capability))
 587                 return 0;
 588
 589         if (wanted_capability & CAP_CLEAN)
 590                 filter_type = "clean";
 591         else if (wanted_capability & CAP_SMUDGE)
 592                 filter_type = "smudge";
 593         else
 594                 die("unexpected filter type");
 595
 596         sigchain_push(SIGPIPE, SIG_IGN);
 597
 598         assert(strlen(filter_type) < LARGE_PACKET_DATA_MAX - strlen("command=\n"));
 599         err = packet_write_fmt_gently(process->in, "command=%s\n", filter_type);
 600         if (err)
 601                 goto done;
 602
 603         err = strlen(path) > LARGE_PACKET_DATA_MAX - strlen("pathname=\n");
 604         if (err) {
 605                 error("path name too long for external filter");
 606                 goto done;
 607         }
 608
 609         err = packet_write_fmt_gently(process->in, "pathname=%s\n", path);
 610         if (err)
 611                 goto done;
 612
 613         if ((entry->supported_capabilities & CAP_DELAY) &&
 614             dco && dco->state == CE_CAN_DELAY) {
 615                 can_delay = 1;
 616                 err = packet_write_fmt_gently(process->in, "can-delay=1\n");
 617                 if (err)
 618                         goto done;
 619         }
 620
 621         err = packet_flush_gently(process->in);
 622         if (err)
 623                 goto done;
 624
 625         if (fd >= 0)
 626                 err = write_packetized_from_fd(fd, process->in);
 627         else
 628                 err = write_packetized_from_buf(src, len, process->in);
 629         if (err)
 630                 goto done;
 631
 632         err = subprocess_read_status(process->out, &filter_status);
 633         if (err)
 634                 goto done;
 635
 636         if (can_delay && !strcmp(filter_status.buf, "delayed")) {
 637                 string_list_insert(&dco->filters, cmd);
 638                 string_list_insert(&dco->paths, path);
 639         } else {
 640                 /* The filter got the blob and wants to send us a response. */
 641                 err = strcmp(filter_status.buf, "success");
 642                 if (err)
 643                         goto done;
 644
 645                 err = read_packetized_to_strbuf(process->out, &nbuf) < 0;
 646                 if (err)
 647                         goto done;
 648
 649                 err = subprocess_read_status(process->out, &filter_status);
 650                 if (err)
 651                         goto done;
 652
 653                 err = strcmp(filter_status.buf, "success");
 654         }
 655
 656 done:
 657         sigchain_pop(SIGPIPE);
 658
 659         if (err)
 660                 handle_filter_error(&filter_status, entry, wanted_capability);
 661         else
 662                 strbuf_swap(dst, &nbuf);
 663         strbuf_release(&nbuf);
 664         return !err;
 665 }
 666
 667
 668 int async_query_available_blobs(const char *cmd, struct string_list *available_paths)
 669 {
 670         int err;
 671         char *line;
 672         struct cmd2process *entry;
 673         struct child_process *process;
 674         struct strbuf filter_status = STRBUF_INIT;
 675
 676         assert(subprocess_map_initialized);
 677         entry = (struct cmd2process *)subprocess_find_entry(&subprocess_map, cmd);
 678         if (!entry) {
 679                 error("external filter '%s' is not available anymore although "
 680                       "not all paths have been filtered", cmd);
 681                 return 0;
 682         }
 683         process = &entry->subprocess.process;
 684         sigchain_push(SIGPIPE, SIG_IGN);
 685
 686         err = packet_write_fmt_gently(
 687                 process->in, "command=list_available_blobs\n");
 688         if (err)
 689                 goto done;
 690
 691         err = packet_flush_gently(process->in);
 692         if (err)
 693                 goto done;
 694
 695         while ((line = packet_read_line(process->out, NULL))) {
 696                 const char *path;
 697                 if (skip_prefix(line, "pathname=", &path))
 698                         string_list_insert(available_paths, xstrdup(path));
 699                 else
 700                         ; /* ignore unknown keys */
 701         }
 702
 703         err = subprocess_read_status(process->out, &filter_status);
 704         if (err)
 705                 goto done;
 706
 707         err = strcmp(filter_status.buf, "success");
 708
 709 done:
 710         sigchain_pop(SIGPIPE);
 711
 712         if (err)
 713                 handle_filter_error(&filter_status, entry, 0);
 714         return !err;
 715 }
 716
 717 static struct convert_driver {
 718         const char *name;
 719         struct convert_driver *next;
 720         const char *smudge;
 721         const char *clean;
 722         const char *process;
 723         int required;
 724 } *user_convert, **user_convert_tail;
 725
 726 static int apply_filter(const char *path, const char *src, size_t len,
 727                         int fd, struct strbuf *dst, struct convert_driver *drv,
 728                         const unsigned int wanted_capability,
 729                         struct delayed_checkout *dco)
 730 {
 731         const char *cmd = NULL;
 732
 733         if (!drv)
 734                 return 0;
 735
 736         if (!dst)
 737                 return 1;
 738
 739         if ((wanted_capability & CAP_CLEAN) && !drv->process && drv->clean)
 740                 cmd = drv->clean;
 741         else if ((wanted_capability & CAP_SMUDGE) && !drv->process && drv->smudge)
 742                 cmd = drv->smudge;
 743
 744         if (cmd && *cmd)
 745                 return apply_single_file_filter(path, src, len, fd, dst, cmd);
 746         else if (drv->process && *drv->process)
 747                 return apply_multi_file_filter(path, src, len, fd, dst,
 748                         drv->process, wanted_capability, dco);
 749
 750         return 0;
 751 }
 752
 753 static int read_convert_config(const char *var, const char *value, void *cb)
 754 {
 755         const char *key, *name;
 756         int namelen;
 757         struct convert_driver *drv;
 758
 759         /*
 760          * External conversion drivers are configured using
 761          * "filter.<name>.variable".
 762          */
 763         if (parse_config_key(var, "filter", &name, &namelen, &key) < 0 || !name)
 764                 return 0;
 765         for (drv = user_convert; drv; drv = drv->next)
 766                 if (!strncmp(drv->name, name, namelen) && !drv->name[namelen])
 767                         break;
 768         if (!drv) {
 769                 drv = xcalloc(1, sizeof(struct convert_driver));
 770                 drv->name = xmemdupz(name, namelen);
 771                 *user_convert_tail = drv;
 772                 user_convert_tail = &(drv->next);
 773         }
 774
 775         /*
 776          * filter.<name>.smudge and filter.<name>.clean specifies
 777          * the command line:
 778          *
 779          *      command-line
 780          *
 781          * The command-line will not be interpolated in any way.
 782          */
 783
 784         if (!strcmp("smudge", key))
 785                 return git_config_string(&drv->smudge, var, value);
 786
 787         if (!strcmp("clean", key))
 788                 return git_config_string(&drv->clean, var, value);
 789
 790         if (!strcmp("process", key))
 791                 return git_config_string(&drv->process, var, value);
 792
 793         if (!strcmp("required", key)) {
 794                 drv->required = git_config_bool(var, value);
 795                 return 0;
 796         }
 797
 798         return 0;
 799 }
 800
 801 static int count_ident(const char *cp, unsigned long size)
 802 {
 803         /*
 804          * "$Id: 0000000000000000000000000000000000000000 $" <=> "$Id$"
 805          */
 806         int cnt = 0;
 807         char ch;
 808
 809         while (size) {
 810                 ch = *cp++;
 811                 size--;
 812                 if (ch != '$')
 813                         continue;
 814                 if (size < 3)
 815                         break;
 816                 if (memcmp("Id", cp, 2))
 817                         continue;
 818                 ch = cp[2];
 819                 cp += 3;
 820                 size -= 3;
 821                 if (ch == '$')
 822                         cnt++; /* $Id$ */
 823                 if (ch != ':')
 824                         continue;
 825
 826                 /*
 827                  * "$Id: ... "; scan up to the closing dollar sign and discard.
 828                  */
 829                 while (size) {
 830                         ch = *cp++;
 831                         size--;
 832                         if (ch == '$') {
 833                                 cnt++;
 834                                 break;
 835                         }
 836                         if (ch == '\n')
 837                                 break;
 838                 }
 839         }
 840         return cnt;
 841 }
 842
 843 static int ident_to_git(const char *path, const char *src, size_t len,
 844                         struct strbuf *buf, int ident)
 845 {
 846         char *dst, *dollar;
 847
 848         if (!ident || (src && !count_ident(src, len)))
 849                 return 0;
 850
 851         if (!buf)
 852                 return 1;
 853
 854         /* only grow if not in place */
 855         if (strbuf_avail(buf) + buf->len < len)
 856                 strbuf_grow(buf, len - buf->len);
 857         dst = buf->buf;
 858         for (;;) {
 859                 dollar = memchr(src, '$', len);
 860                 if (!dollar)
 861                         break;
 862                 memmove(dst, src, dollar + 1 - src);
 863                 dst += dollar + 1 - src;
 864                 len -= dollar + 1 - src;
 865                 src  = dollar + 1;
 866
 867                 if (len > 3 && !memcmp(src, "Id:", 3)) {
 868                         dollar = memchr(src + 3, '$', len - 3);
 869                         if (!dollar)
 870                                 break;
 871                         if (memchr(src + 3, '\n', dollar - src - 3)) {
 872                                 /* Line break before the next dollar. */
 873                                 continue;
 874                         }
 875
 876                         memcpy(dst, "Id$", 3);
 877                         dst += 3;
 878                         len -= dollar + 1 - src;
 879                         src  = dollar + 1;
 880                 }
 881         }
 882         memmove(dst, src, len);
 883         strbuf_setlen(buf, dst + len - buf->buf);
 884         return 1;
 885 }
 886
 887 static int ident_to_worktree(const char *path, const char *src, size_t len,
 888                              struct strbuf *buf, int ident)
 889 {
 890         unsigned char sha1[20];
 891         char *to_free = NULL, *dollar, *spc;
 892         int cnt;
 893
 894         if (!ident)
 895                 return 0;
 896
 897         cnt = count_ident(src, len);
 898         if (!cnt)
 899                 return 0;
 900
 901         /* are we "faking" in place editing ? */
 902         if (src == buf->buf)
 903                 to_free = strbuf_detach(buf, NULL);
 904         hash_sha1_file(src, len, "blob", sha1);
 905
 906         strbuf_grow(buf, len + cnt * 43);
 907         for (;;) {
 908                 /* step 1: run to the next '$' */
 909                 dollar = memchr(src, '$', len);
 910                 if (!dollar)
 911                         break;
 912                 strbuf_add(buf, src, dollar + 1 - src);
 913                 len -= dollar + 1 - src;
 914                 src  = dollar + 1;
 915
 916                 /* step 2: does it looks like a bit like Id:xxx$ or Id$ ? */
 917                 if (len < 3 || memcmp("Id", src, 2))
 918                         continue;
 919
 920                 /* step 3: skip over Id$ or Id:xxxxx$ */
 921                 if (src[2] == '$') {
 922                         src += 3;
 923                         len -= 3;
 924                 } else if (src[2] == ':') {
 925                         /*
 926                          * It's possible that an expanded Id has crept its way into the
 927                          * repository, we cope with that by stripping the expansion out.
 928                          * This is probably not a good idea, since it will cause changes
 929                          * on checkout, which won't go away by stash, but let's keep it
 930                          * for git-style ids.
 931                          */
 932                         dollar = memchr(src + 3, '$', len - 3);
 933                         if (!dollar) {
 934                                 /* incomplete keyword, no more '$', so just quit the loop */
 935                                 break;
 936                         }
 937
 938                         if (memchr(src + 3, '\n', dollar - src - 3)) {
 939                                 /* Line break before the next dollar. */
 940                                 continue;
 941                         }
 942
 943                         spc = memchr(src + 4, ' ', dollar - src - 4);
 944                         if (spc && spc < dollar-1) {
 945                                 /* There are spaces in unexpected places.
 946                                  * This is probably an id from some other
 947                                  * versioning system. Keep it for now.
 948                                  */
 949                                 continue;
 950                         }
 951
 952                         len -= dollar + 1 - src;
 953                         src  = dollar + 1;
 954                 } else {
 955                         /* it wasn't a "Id$" or "Id:xxxx$" */
 956                         continue;
 957                 }
 958
 959                 /* step 4: substitute */
 960                 strbuf_addstr(buf, "Id: ");
 961                 strbuf_add(buf, sha1_to_hex(sha1), 40);
 962                 strbuf_addstr(buf, " $");
 963         }
 964         strbuf_add(buf, src, len);
 965
 966         free(to_free);
 967         return 1;
 968 }
 969
 970 static enum crlf_action git_path_check_crlf(struct attr_check_item *check)
 971 {
 972         const char *value = check->value;
 973
 974         if (ATTR_TRUE(value))
 975                 return CRLF_TEXT;
 976         else if (ATTR_FALSE(value))
 977                 return CRLF_BINARY;
 978         else if (ATTR_UNSET(value))
 979                 ;
 980         else if (!strcmp(value, "input"))
 981                 return CRLF_TEXT_INPUT;
 982         else if (!strcmp(value, "auto"))
 983                 return CRLF_AUTO;
 984         return CRLF_UNDEFINED;
 985 }
 986
 987 static enum eol git_path_check_eol(struct attr_check_item *check)
 988 {
 989         const char *value = check->value;
 990
 991         if (ATTR_UNSET(value))
 992                 ;
 993         else if (!strcmp(value, "lf"))
 994                 return EOL_LF;
 995         else if (!strcmp(value, "crlf"))
 996                 return EOL_CRLF;
 997         return EOL_UNSET;
 998 }
 999
1000 static struct convert_driver *git_path_check_convert(struct attr_check_item *check)
1001 {
1002         const char *value = check->value;
1003         struct convert_driver *drv;
1004
1005         if (ATTR_TRUE(value) || ATTR_FALSE(value) || ATTR_UNSET(value))
1006                 return NULL;
1007         for (drv = user_convert; drv; drv = drv->next)
1008                 if (!strcmp(value, drv->name))
1009                         return drv;
1010         return NULL;
1011 }
1012
1013 static int git_path_check_ident(struct attr_check_item *check)
1014 {
1015         const char *value = check->value;
1016
1017         return !!ATTR_TRUE(value);
1018 }
1019
1020 struct conv_attrs {
1021         struct convert_driver *drv;
1022         enum crlf_action attr_action; /* What attr says */
1023         enum crlf_action crlf_action; /* When no attr is set, use core.autocrlf */
1024         int ident;
1025 };
1026
1027 static void convert_attrs(struct conv_attrs *ca, const char *path)
1028 {
1029         static struct attr_check *check;
1030
1031         if (!check) {
1032                 check = attr_check_initl("crlf", "ident", "filter",
1033                                          "eol", "text", NULL);
1034                 user_convert_tail = &user_convert;
1035                 git_config(read_convert_config, NULL);
1036         }
1037
1038         if (!git_check_attr(path, check)) {
1039                 struct attr_check_item *ccheck = check->items;
1040                 ca->crlf_action = git_path_check_crlf(ccheck + 4);
1041                 if (ca->crlf_action == CRLF_UNDEFINED)
1042                         ca->crlf_action = git_path_check_crlf(ccheck + 0);
1043                 ca->attr_action = ca->crlf_action;
1044                 ca->ident = git_path_check_ident(ccheck + 1);
1045                 ca->drv = git_path_check_convert(ccheck + 2);
1046                 if (ca->crlf_action != CRLF_BINARY) {
1047                         enum eol eol_attr = git_path_check_eol(ccheck + 3);
1048                         if (ca->crlf_action == CRLF_AUTO && eol_attr == EOL_LF)
1049                                 ca->crlf_action = CRLF_AUTO_INPUT;
1050                         else if (ca->crlf_action == CRLF_AUTO && eol_attr == EOL_CRLF)
1051                                 ca->crlf_action = CRLF_AUTO_CRLF;
1052                         else if (eol_attr == EOL_LF)
1053                                 ca->crlf_action = CRLF_TEXT_INPUT;
1054                         else if (eol_attr == EOL_CRLF)
1055                                 ca->crlf_action = CRLF_TEXT_CRLF;
1056                 }
1057                 ca->attr_action = ca->crlf_action;
1058         } else {
1059                 ca->drv = NULL;
1060                 ca->crlf_action = CRLF_UNDEFINED;
1061                 ca->ident = 0;
1062         }
1063         if (ca->crlf_action == CRLF_TEXT)
1064                 ca->crlf_action = text_eol_is_crlf() ? CRLF_TEXT_CRLF : CRLF_TEXT_INPUT;
1065         if (ca->crlf_action == CRLF_UNDEFINED && auto_crlf == AUTO_CRLF_FALSE)
1066                 ca->crlf_action = CRLF_BINARY;
1067         if (ca->crlf_action == CRLF_UNDEFINED && auto_crlf == AUTO_CRLF_TRUE)
1068                 ca->crlf_action = CRLF_AUTO_CRLF;
1069         if (ca->crlf_action == CRLF_UNDEFINED && auto_crlf == AUTO_CRLF_INPUT)
1070                 ca->crlf_action = CRLF_AUTO_INPUT;
1071 }
1072
1073 int would_convert_to_git_filter_fd(const char *path)
1074 {
1075         struct conv_attrs ca;
1076
1077         convert_attrs(&ca, path);
1078         if (!ca.drv)
1079                 return 0;
1080
1081         /*
1082          * Apply a filter to an fd only if the filter is required to succeed.
1083          * We must die if the filter fails, because the original data before
1084          * filtering is not available.
1085          */
1086         if (!ca.drv->required)
1087                 return 0;
1088
1089         return apply_filter(path, NULL, 0, -1, NULL, ca.drv, CAP_CLEAN, NULL);
1090 }
1091
1092 const char *get_convert_attr_ascii(const char *path)
1093 {
1094         struct conv_attrs ca;
1095
1096         convert_attrs(&ca, path);
1097         switch (ca.attr_action) {
1098         case CRLF_UNDEFINED:
1099                 return "";
1100         case CRLF_BINARY:
1101                 return "-text";
1102         case CRLF_TEXT:
1103                 return "text";
1104         case CRLF_TEXT_INPUT:
1105                 return "text eol=lf";
1106         case CRLF_TEXT_CRLF:
1107                 return "text eol=crlf";
1108         case CRLF_AUTO:
1109                 return "text=auto";
1110         case CRLF_AUTO_CRLF:
1111                 return "text=auto eol=crlf";
1112         case CRLF_AUTO_INPUT:
1113                 return "text=auto eol=lf";
1114         }
1115         return "";
1116 }
1117
1118 int convert_to_git(const struct index_state *istate,
1119                    const char *path, const char *src, size_t len,
1120                    struct strbuf *dst, enum safe_crlf checksafe)
1121 {
1122         int ret = 0;
1123         struct conv_attrs ca;
1124
1125         convert_attrs(&ca, path);
1126
1127         ret |= apply_filter(path, src, len, -1, dst, ca.drv, CAP_CLEAN, NULL);
1128         if (!ret && ca.drv && ca.drv->required)
1129                 die("%s: clean filter '%s' failed", path, ca.drv->name);
1130
1131         if (ret && dst) {
1132                 src = dst->buf;
1133                 len = dst->len;
1134         }
1135         if (checksafe != SAFE_CRLF_KEEP_CRLF) {
1136                 ret |= crlf_to_git(istate, path, src, len, dst, ca.crlf_action, checksafe);
1137                 if (ret && dst) {
1138                         src = dst->buf;
1139                         len = dst->len;
1140                 }
1141         }
1142         return ret | ident_to_git(path, src, len, dst, ca.ident);
1143 }
1144
1145 void convert_to_git_filter_fd(const struct index_state *istate,
1146                               const char *path, int fd, struct strbuf *dst,
1147                               enum safe_crlf checksafe)
1148 {
1149         struct conv_attrs ca;
1150         convert_attrs(&ca, path);
1151
1152         assert(ca.drv);
1153         assert(ca.drv->clean || ca.drv->process);
1154
1155         if (!apply_filter(path, NULL, 0, fd, dst, ca.drv, CAP_CLEAN, NULL))
1156                 die("%s: clean filter '%s' failed", path, ca.drv->name);
1157
1158         crlf_to_git(istate, path, dst->buf, dst->len, dst, ca.crlf_action, checksafe);
1159         ident_to_git(path, dst->buf, dst->len, dst, ca.ident);
1160 }
1161
1162 static int convert_to_working_tree_internal(const char *path, const char *src,
1163                                             size_t len, struct strbuf *dst,
1164                                             int normalizing, struct delayed_checkout *dco)
1165 {
1166         int ret = 0, ret_filter = 0;
1167         struct conv_attrs ca;
1168
1169         convert_attrs(&ca, path);
1170
1171         ret |= ident_to_worktree(path, src, len, dst, ca.ident);
1172         if (ret) {
1173                 src = dst->buf;
1174                 len = dst->len;
1175         }
1176         /*
1177          * CRLF conversion can be skipped if normalizing, unless there
1178          * is a smudge or process filter (even if the process filter doesn't
1179          * support smudge).  The filters might expect CRLFs.
1180          */
1181         if ((ca.drv && (ca.drv->smudge || ca.drv->process)) || !normalizing) {
1182                 ret |= crlf_to_worktree(path, src, len, dst, ca.crlf_action);
1183                 if (ret) {
1184                         src = dst->buf;
1185                         len = dst->len;
1186                 }
1187         }
1188
1189         ret_filter = apply_filter(
1190                 path, src, len, -1, dst, ca.drv, CAP_SMUDGE, dco);
1191         if (!ret_filter && ca.drv && ca.drv->required)
1192                 die("%s: smudge filter %s failed", path, ca.drv->name);
1193
1194         return ret | ret_filter;
1195 }
1196
1197 int async_convert_to_working_tree(const char *path, const char *src,
1198                                   size_t len, struct strbuf *dst,
1199                                   void *dco)
1200 {
1201         return convert_to_working_tree_internal(path, src, len, dst, 0, dco);
1202 }
1203
1204 int convert_to_working_tree(const char *path, const char *src, size_t len, struct strbuf *dst)
1205 {
1206         return convert_to_working_tree_internal(path, src, len, dst, 0, NULL);
1207 }
1208
1209 int renormalize_buffer(const struct index_state *istate, const char *path,
1210                        const char *src, size_t len, struct strbuf *dst)
1211 {
1212         int ret = convert_to_working_tree_internal(path, src, len, dst, 1, NULL);
1213         if (ret) {
1214                 src = dst->buf;
1215                 len = dst->len;
1216         }
1217         return ret | convert_to_git(istate, path, src, len, dst, SAFE_CRLF_RENORMALIZE);
1218 }
1219
1220 /*****************************************************************
1221  *
1222  * Streaming conversion support
1223  *
1224  *****************************************************************/
1225
1226 typedef int (*filter_fn)(struct stream_filter *,
1227                          const char *input, size_t *isize_p,
1228                          char *output, size_t *osize_p);
1229 typedef void (*free_fn)(struct stream_filter *);
1230
1231 struct stream_filter_vtbl {
1232         filter_fn filter;
1233         free_fn free;
1234 };
1235
1236 struct stream_filter {
1237         struct stream_filter_vtbl *vtbl;
1238 };
1239
1240 static int null_filter_fn(struct stream_filter *filter,
1241                           const char *input, size_t *isize_p,
1242                           char *output, size_t *osize_p)
1243 {
1244         size_t count;
1245
1246         if (!input)
1247                 return 0; /* we do not keep any states */
1248         count = *isize_p;
1249         if (*osize_p < count)
1250                 count = *osize_p;
1251         if (count) {
1252                 memmove(output, input, count);
1253                 *isize_p -= count;
1254                 *osize_p -= count;
1255         }
1256         return 0;
1257 }
1258
1259 static void null_free_fn(struct stream_filter *filter)
1260 {
1261         ; /* nothing -- null instances are shared */
1262 }
1263
1264 static struct stream_filter_vtbl null_vtbl = {
1265         null_filter_fn,
1266         null_free_fn,
1267 };
1268
1269 static struct stream_filter null_filter_singleton = {
1270         &null_vtbl,
1271 };
1272
1273 int is_null_stream_filter(struct stream_filter *filter)
1274 {
1275         return filter == &null_filter_singleton;
1276 }
1277
1278
1279 /*
1280  * LF-to-CRLF filter
1281  */
1282
1283 struct lf_to_crlf_filter {
1284         struct stream_filter filter;
1285         unsigned has_held:1;
1286         char held;
1287 };
1288
1289 static int lf_to_crlf_filter_fn(struct stream_filter *filter,
1290                                 const char *input, size_t *isize_p,
1291                                 char *output, size_t *osize_p)
1292 {
1293         size_t count, o = 0;
1294         struct lf_to_crlf_filter *lf_to_crlf = (struct lf_to_crlf_filter *)filter;
1295
1296         /*
1297          * We may be holding onto the CR to see if it is followed by a
1298          * LF, in which case we would need to go to the main loop.
1299          * Otherwise, just emit it to the output stream.
1300          */
1301         if (lf_to_crlf->has_held && (lf_to_crlf->held != '\r' || !input)) {
1302                 output[o++] = lf_to_crlf->held;
1303                 lf_to_crlf->has_held = 0;
1304         }
1305
1306         /* We are told to drain */
1307         if (!input) {
1308                 *osize_p -= o;
1309                 return 0;
1310         }
1311
1312         count = *isize_p;
1313         if (count || lf_to_crlf->has_held) {
1314                 size_t i;
1315                 int was_cr = 0;
1316
1317                 if (lf_to_crlf->has_held) {
1318                         was_cr = 1;
1319                         lf_to_crlf->has_held = 0;
1320                 }
1321
1322                 for (i = 0; o < *osize_p && i < count; i++) {
1323                         char ch = input[i];
1324
1325                         if (ch == '\n') {
1326                                 output[o++] = '\r';
1327                         } else if (was_cr) {
1328                                 /*
1329                                  * Previous round saw CR and it is not followed
1330                                  * by a LF; emit the CR before processing the
1331                                  * current character.
1332                                  */
1333                                 output[o++] = '\r';
1334                         }
1335
1336                         /*
1337                          * We may have consumed the last output slot,
1338                          * in which case we need to break out of this
1339                          * loop; hold the current character before
1340                          * returning.
1341                          */
1342                         if (*osize_p <= o) {
1343                                 lf_to_crlf->has_held = 1;
1344                                 lf_to_crlf->held = ch;
1345                                 continue; /* break but increment i */
1346                         }
1347
1348                         if (ch == '\r') {
1349                                 was_cr = 1;
1350                                 continue;
1351                         }
1352
1353                         was_cr = 0;
1354                         output[o++] = ch;
1355                 }
1356
1357                 *osize_p -= o;
1358                 *isize_p -= i;
1359
1360                 if (!lf_to_crlf->has_held && was_cr) {
1361                         lf_to_crlf->has_held = 1;
1362                         lf_to_crlf->held = '\r';
1363                 }
1364         }
1365         return 0;
1366 }
1367
1368 static void lf_to_crlf_free_fn(struct stream_filter *filter)
1369 {
1370         free(filter);
1371 }
1372
1373 static struct stream_filter_vtbl lf_to_crlf_vtbl = {
1374         lf_to_crlf_filter_fn,
1375         lf_to_crlf_free_fn,
1376 };
1377
1378 static struct stream_filter *lf_to_crlf_filter(void)
1379 {
1380         struct lf_to_crlf_filter *lf_to_crlf = xcalloc(1, sizeof(*lf_to_crlf));
1381
1382         lf_to_crlf->filter.vtbl = &lf_to_crlf_vtbl;
1383         return (struct stream_filter *)lf_to_crlf;
1384 }
1385
1386 /*
1387  * Cascade filter
1388  */
1389 #define FILTER_BUFFER 1024
1390 struct cascade_filter {
1391         struct stream_filter filter;
1392         struct stream_filter *one;
1393         struct stream_filter *two;
1394         char buf[FILTER_BUFFER];
1395         int end, ptr;
1396 };
1397
1398 static int cascade_filter_fn(struct stream_filter *filter,
1399                              const char *input, size_t *isize_p,
1400                              char *output, size_t *osize_p)
1401 {
1402         struct cascade_filter *cas = (struct cascade_filter *) filter;
1403         size_t filled = 0;
1404         size_t sz = *osize_p;
1405         size_t to_feed, remaining;
1406
1407         /*
1408          * input -- (one) --> buf -- (two) --> output
1409          */
1410         while (filled < sz) {
1411                 remaining = sz - filled;
1412
1413                 /* do we already have something to feed two with? */
1414                 if (cas->ptr < cas->end) {
1415                         to_feed = cas->end - cas->ptr;
1416                         if (stream_filter(cas->two,
1417                                           cas->buf + cas->ptr, &to_feed,
1418                                           output + filled, &remaining))
1419                                 return -1;
1420                         cas->ptr += (cas->end - cas->ptr) - to_feed;
1421                         filled = sz - remaining;
1422                         continue;
1423                 }
1424
1425                 /* feed one from upstream and have it emit into our buffer */
1426                 to_feed = input ? *isize_p : 0;
1427                 if (input && !to_feed)
1428                         break;
1429                 remaining = sizeof(cas->buf);
1430                 if (stream_filter(cas->one,
1431                                   input, &to_feed,
1432                                   cas->buf, &remaining))
1433                         return -1;
1434                 cas->end = sizeof(cas->buf) - remaining;
1435                 cas->ptr = 0;
1436                 if (input) {
1437                         size_t fed = *isize_p - to_feed;
1438                         *isize_p -= fed;
1439                         input += fed;
1440                 }
1441
1442                 /* do we know that we drained one completely? */
1443                 if (input || cas->end)
1444                         continue;
1445
1446                 /* tell two to drain; we have nothing more to give it */
1447                 to_feed = 0;
1448                 remaining = sz - filled;
1449                 if (stream_filter(cas->two,
1450                                   NULL, &to_feed,
1451                                   output + filled, &remaining))
1452                         return -1;
1453                 if (remaining == (sz - filled))
1454                         break; /* completely drained two */
1455                 filled = sz - remaining;
1456         }
1457         *osize_p -= filled;
1458         return 0;
1459 }
1460
1461 static void cascade_free_fn(struct stream_filter *filter)
1462 {
1463         struct cascade_filter *cas = (struct cascade_filter *)filter;
1464         free_stream_filter(cas->one);
1465         free_stream_filter(cas->two);
1466         free(filter);
1467 }
1468
1469 static struct stream_filter_vtbl cascade_vtbl = {
1470         cascade_filter_fn,
1471         cascade_free_fn,
1472 };
1473
1474 static struct stream_filter *cascade_filter(struct stream_filter *one,
1475                                             struct stream_filter *two)
1476 {
1477         struct cascade_filter *cascade;
1478
1479         if (!one || is_null_stream_filter(one))
1480                 return two;
1481         if (!two || is_null_stream_filter(two))
1482                 return one;
1483
1484         cascade = xmalloc(sizeof(*cascade));
1485         cascade->one = one;
1486         cascade->two = two;
1487         cascade->end = cascade->ptr = 0;
1488         cascade->filter.vtbl = &cascade_vtbl;
1489         return (struct stream_filter *)cascade;
1490 }
1491
1492 /*
1493  * ident filter
1494  */
1495 #define IDENT_DRAINING (-1)
1496 #define IDENT_SKIPPING (-2)
1497 struct ident_filter {
1498         struct stream_filter filter;
1499         struct strbuf left;
1500         int state;
1501         char ident[45]; /* ": x40 $" */
1502 };
1503
1504 static int is_foreign_ident(const char *str)
1505 {
1506         int i;
1507
1508         if (!skip_prefix(str, "$Id: ", &str))
1509                 return 0;
1510         for (i = 0; str[i]; i++) {
1511                 if (isspace(str[i]) && str[i+1] != '$')
1512                         return 1;
1513         }
1514         return 0;
1515 }
1516
1517 static void ident_drain(struct ident_filter *ident, char **output_p, size_t *osize_p)
1518 {
1519         size_t to_drain = ident->left.len;
1520
1521         if (*osize_p < to_drain)
1522                 to_drain = *osize_p;
1523         if (to_drain) {
1524                 memcpy(*output_p, ident->left.buf, to_drain);
1525                 strbuf_remove(&ident->left, 0, to_drain);
1526                 *output_p += to_drain;
1527                 *osize_p -= to_drain;
1528         }
1529         if (!ident->left.len)
1530                 ident->state = 0;
1531 }
1532
1533 static int ident_filter_fn(struct stream_filter *filter,
1534                            const char *input, size_t *isize_p,
1535                            char *output, size_t *osize_p)
1536 {
1537         struct ident_filter *ident = (struct ident_filter *)filter;
1538         static const char head[] = "$Id";
1539
1540         if (!input) {
1541                 /* drain upon eof */
1542                 switch (ident->state) {
1543                 default:
1544                         strbuf_add(&ident->left, head, ident->state);
1545                 case IDENT_SKIPPING:
1546                         /* fallthru */
1547                 case IDENT_DRAINING:
1548                         ident_drain(ident, &output, osize_p);
1549                 }
1550                 return 0;
1551         }
1552
1553         while (*isize_p || (ident->state == IDENT_DRAINING)) {
1554                 int ch;
1555
1556                 if (ident->state == IDENT_DRAINING) {
1557                         ident_drain(ident, &output, osize_p);
1558                         if (!*osize_p)
1559                                 break;
1560                         continue;
1561                 }
1562
1563                 ch = *(input++);
1564                 (*isize_p)--;
1565
1566                 if (ident->state == IDENT_SKIPPING) {
1567                         /*
1568                          * Skipping until '$' or LF, but keeping them
1569                          * in case it is a foreign ident.
1570                          */
1571                         strbuf_addch(&ident->left, ch);
1572                         if (ch != '\n' && ch != '$')
1573                                 continue;
1574                         if (ch == '$' && !is_foreign_ident(ident->left.buf)) {
1575                                 strbuf_setlen(&ident->left, sizeof(head) - 1);
1576                                 strbuf_addstr(&ident->left, ident->ident);
1577                         }
1578                         ident->state = IDENT_DRAINING;
1579                         continue;
1580                 }
1581
1582                 if (ident->state < sizeof(head) &&
1583                     head[ident->state] == ch) {
1584                         ident->state++;
1585                         continue;
1586                 }
1587
1588                 if (ident->state)
1589                         strbuf_add(&ident->left, head, ident->state);
1590                 if (ident->state == sizeof(head) - 1) {
1591                         if (ch != ':' && ch != '$') {
1592                                 strbuf_addch(&ident->left, ch);
1593                                 ident->state = 0;
1594                                 continue;
1595                         }
1596
1597                         if (ch == ':') {
1598                                 strbuf_addch(&ident->left, ch);
1599                                 ident->state = IDENT_SKIPPING;
1600                         } else {
1601                                 strbuf_addstr(&ident->left, ident->ident);
1602                                 ident->state = IDENT_DRAINING;
1603                         }
1604                         continue;
1605                 }
1606
1607                 strbuf_addch(&ident->left, ch);
1608                 ident->state = IDENT_DRAINING;
1609         }
1610         return 0;
1611 }
1612
1613 static void ident_free_fn(struct stream_filter *filter)
1614 {
1615         struct ident_filter *ident = (struct ident_filter *)filter;
1616         strbuf_release(&ident->left);
1617         free(filter);
1618 }
1619
1620 static struct stream_filter_vtbl ident_vtbl = {
1621         ident_filter_fn,
1622         ident_free_fn,
1623 };
1624
1625 static struct stream_filter *ident_filter(const unsigned char *sha1)
1626 {
1627         struct ident_filter *ident = xmalloc(sizeof(*ident));
1628
1629         xsnprintf(ident->ident, sizeof(ident->ident),
1630                   ": %s $", sha1_to_hex(sha1));
1631         strbuf_init(&ident->left, 0);
1632         ident->filter.vtbl = &ident_vtbl;
1633         ident->state = 0;
1634         return (struct stream_filter *)ident;
1635 }
1636
1637 /*
1638  * Return an appropriately constructed filter for the path, or NULL if
1639  * the contents cannot be filtered without reading the whole thing
1640  * in-core.
1641  *
1642  * Note that you would be crazy to set CRLF, smuge/clean or ident to a
1643  * large binary blob you would want us not to slurp into the memory!
1644  */
1645 struct stream_filter *get_stream_filter(const char *path, const unsigned char *sha1)
1646 {
1647         struct conv_attrs ca;
1648         struct stream_filter *filter = NULL;
1649
1650         convert_attrs(&ca, path);
1651         if (ca.drv && (ca.drv->process || ca.drv->smudge || ca.drv->clean))
1652                 return NULL;
1653
1654         if (ca.crlf_action == CRLF_AUTO || ca.crlf_action == CRLF_AUTO_CRLF)
1655                 return NULL;
1656
1657         if (ca.ident)
1658                 filter = ident_filter(sha1);
1659
1660         if (output_eol(ca.crlf_action) == EOL_CRLF)
1661                 filter = cascade_filter(filter, lf_to_crlf_filter());
1662         else
1663                 filter = cascade_filter(filter, &null_filter_singleton);
1664
1665         return filter;
1666 }
1667
1668 void free_stream_filter(struct stream_filter *filter)
1669 {
1670         filter->vtbl->free(filter);
1671 }
1672
1673 int stream_filter(struct stream_filter *filter,
1674                   const char *input, size_t *isize_p,
1675                   char *output, size_t *osize_p)
1676 {
1677         return filter->vtbl->filter(filter, input, isize_p, output, osize_p);
1678 }