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                 strbuf_release(&cmd);
 428                 return error("cannot fork to run external filter '%s'", params->cmd);
 429         }
 430
 431         sigchain_push(SIGPIPE, SIG_IGN);
 432
 433         if (params->src) {
 434                 write_err = (write_in_full(child_process.in,
 435                                            params->src, params->size) < 0);
 436                 if (errno == EPIPE)
 437                         write_err = 0;
 438         } else {
 439                 write_err = copy_fd(params->fd, child_process.in);
 440                 if (write_err == COPY_WRITE_ERROR && errno == EPIPE)
 441                         write_err = 0;
 442         }
 443
 444         if (close(child_process.in))
 445                 write_err = 1;
 446         if (write_err)
 447                 error("cannot feed the input to external filter '%s'", params->cmd);
 448
 449         sigchain_pop(SIGPIPE);
 450
 451         status = finish_command(&child_process);
 452         if (status)
 453                 error("external filter '%s' failed %d", params->cmd, status);
 454
 455         strbuf_release(&cmd);
 456         return (write_err || status);
 457 }
 458
 459 static int apply_single_file_filter(const char *path, const char *src, size_t len, int fd,
 460                         struct strbuf *dst, const char *cmd)
 461 {
 462         /*
 463          * Create a pipeline to have the command filter the buffer's
 464          * contents.
 465          *
 466          * (child --> cmd) --> us
 467          */
 468         int err = 0;
 469         struct strbuf nbuf = STRBUF_INIT;
 470         struct async async;
 471         struct filter_params params;
 472
 473         memset(&async, 0, sizeof(async));
 474         async.proc = filter_buffer_or_fd;
 475         async.data = &params;
 476         async.out = -1;
 477         params.src = src;
 478         params.size = len;
 479         params.fd = fd;
 480         params.cmd = cmd;
 481         params.path = path;
 482
 483         fflush(NULL);
 484         if (start_async(&async))
 485                 return 0;       /* error was already reported */
 486
 487         if (strbuf_read(&nbuf, async.out, len) < 0) {
 488                 err = error("read from external filter '%s' failed", cmd);
 489         }
 490         if (close(async.out)) {
 491                 err = error("read from external filter '%s' failed", cmd);
 492         }
 493         if (finish_async(&async)) {
 494                 err = error("external filter '%s' failed", cmd);
 495         }
 496
 497         if (!err) {
 498                 strbuf_swap(dst, &nbuf);
 499         }
 500         strbuf_release(&nbuf);
 501         return !err;
 502 }
 503
 504 #define CAP_CLEAN    (1u<<0)
 505 #define CAP_SMUDGE   (1u<<1)
 506 #define CAP_DELAY    (1u<<2)
 507
 508 struct cmd2process {
 509         struct subprocess_entry subprocess; /* must be the first member! */
 510         unsigned int supported_capabilities;
 511 };
 512
 513 static int subprocess_map_initialized;
 514 static struct hashmap subprocess_map;
 515
 516 static int start_multi_file_filter_fn(struct subprocess_entry *subprocess)
 517 {
 518         static int versions[] = {2, 0};
 519         static struct subprocess_capability capabilities[] = {
 520                 { "clean",  CAP_CLEAN  },
 521                 { "smudge", CAP_SMUDGE },
 522                 { "delay",  CAP_DELAY  },
 523                 { NULL, 0 }
 524         };
 525         struct cmd2process *entry = (struct cmd2process *)subprocess;
 526         return subprocess_handshake(subprocess, "git-filter", versions, NULL,
 527                                     capabilities,
 528                                     &entry->supported_capabilities);
 529 }
 530
 531 static void handle_filter_error(const struct strbuf *filter_status,
 532                                 struct cmd2process *entry,
 533                                 const unsigned int wanted_capability) {
 534         if (!strcmp(filter_status->buf, "error"))
 535                 ; /* The filter signaled a problem with the file. */
 536         else if (!strcmp(filter_status->buf, "abort") && wanted_capability) {
 537                 /*
 538                  * The filter signaled a permanent problem. Don't try to filter
 539                  * files with the same command for the lifetime of the current
 540                  * Git process.
 541                  */
 542                  entry->supported_capabilities &= ~wanted_capability;
 543         } else {
 544                 /*
 545                  * Something went wrong with the protocol filter.
 546                  * Force shutdown and restart if another blob requires filtering.
 547                  */
 548                 error("external filter '%s' failed", entry->subprocess.cmd);
 549                 subprocess_stop(&subprocess_map, &entry->subprocess);
 550                 free(entry);
 551         }
 552 }
 553
 554 static int apply_multi_file_filter(const char *path, const char *src, size_t len,
 555                                    int fd, struct strbuf *dst, const char *cmd,
 556                                    const unsigned int wanted_capability,
 557                                    struct delayed_checkout *dco)
 558 {
 559         int err;
 560         int can_delay = 0;
 561         struct cmd2process *entry;
 562         struct child_process *process;
 563         struct strbuf nbuf = STRBUF_INIT;
 564         struct strbuf filter_status = STRBUF_INIT;
 565         const char *filter_type;
 566
 567         if (!subprocess_map_initialized) {
 568                 subprocess_map_initialized = 1;
 569                 hashmap_init(&subprocess_map, cmd2process_cmp, NULL, 0);
 570                 entry = NULL;
 571         } else {
 572                 entry = (struct cmd2process *)subprocess_find_entry(&subprocess_map, cmd);
 573         }
 574
 575         fflush(NULL);
 576
 577         if (!entry) {
 578                 entry = xmalloc(sizeof(*entry));
 579                 entry->supported_capabilities = 0;
 580
 581                 if (subprocess_start(&subprocess_map, &entry->subprocess, cmd, start_multi_file_filter_fn)) {
 582                         free(entry);
 583                         return 0;
 584                 }
 585         }
 586         process = &entry->subprocess.process;
 587
 588         if (!(entry->supported_capabilities & wanted_capability))
 589                 return 0;
 590
 591         if (wanted_capability & CAP_CLEAN)
 592                 filter_type = "clean";
 593         else if (wanted_capability & CAP_SMUDGE)
 594                 filter_type = "smudge";
 595         else
 596                 die("unexpected filter type");
 597
 598         sigchain_push(SIGPIPE, SIG_IGN);
 599
 600         assert(strlen(filter_type) < LARGE_PACKET_DATA_MAX - strlen("command=\n"));
 601         err = packet_write_fmt_gently(process->in, "command=%s\n", filter_type);
 602         if (err)
 603                 goto done;
 604
 605         err = strlen(path) > LARGE_PACKET_DATA_MAX - strlen("pathname=\n");
 606         if (err) {
 607                 error("path name too long for external filter");
 608                 goto done;
 609         }
 610
 611         err = packet_write_fmt_gently(process->in, "pathname=%s\n", path);
 612         if (err)
 613                 goto done;
 614
 615         if ((entry->supported_capabilities & CAP_DELAY) &&
 616             dco && dco->state == CE_CAN_DELAY) {
 617                 can_delay = 1;
 618                 err = packet_write_fmt_gently(process->in, "can-delay=1\n");
 619                 if (err)
 620                         goto done;
 621         }
 622
 623         err = packet_flush_gently(process->in);
 624         if (err)
 625                 goto done;
 626
 627         if (fd >= 0)
 628                 err = write_packetized_from_fd(fd, process->in);
 629         else
 630                 err = write_packetized_from_buf(src, len, process->in);
 631         if (err)
 632                 goto done;
 633
 634         err = subprocess_read_status(process->out, &filter_status);
 635         if (err)
 636                 goto done;
 637
 638         if (can_delay && !strcmp(filter_status.buf, "delayed")) {
 639                 string_list_insert(&dco->filters, cmd);
 640                 string_list_insert(&dco->paths, path);
 641         } else {
 642                 /* The filter got the blob and wants to send us a response. */
 643                 err = strcmp(filter_status.buf, "success");
 644                 if (err)
 645                         goto done;
 646
 647                 err = read_packetized_to_strbuf(process->out, &nbuf) < 0;
 648                 if (err)
 649                         goto done;
 650
 651                 err = subprocess_read_status(process->out, &filter_status);
 652                 if (err)
 653                         goto done;
 654
 655                 err = strcmp(filter_status.buf, "success");
 656         }
 657
 658 done:
 659         sigchain_pop(SIGPIPE);
 660
 661         if (err)
 662                 handle_filter_error(&filter_status, entry, wanted_capability);
 663         else
 664                 strbuf_swap(dst, &nbuf);
 665         strbuf_release(&nbuf);
 666         return !err;
 667 }
 668
 669
 670 int async_query_available_blobs(const char *cmd, struct string_list *available_paths)
 671 {
 672         int err;
 673         char *line;
 674         struct cmd2process *entry;
 675         struct child_process *process;
 676         struct strbuf filter_status = STRBUF_INIT;
 677
 678         assert(subprocess_map_initialized);
 679         entry = (struct cmd2process *)subprocess_find_entry(&subprocess_map, cmd);
 680         if (!entry) {
 681                 error("external filter '%s' is not available anymore although "
 682                       "not all paths have been filtered", cmd);
 683                 return 0;
 684         }
 685         process = &entry->subprocess.process;
 686         sigchain_push(SIGPIPE, SIG_IGN);
 687
 688         err = packet_write_fmt_gently(
 689                 process->in, "command=list_available_blobs\n");
 690         if (err)
 691                 goto done;
 692
 693         err = packet_flush_gently(process->in);
 694         if (err)
 695                 goto done;
 696
 697         while ((line = packet_read_line(process->out, NULL))) {
 698                 const char *path;
 699                 if (skip_prefix(line, "pathname=", &path))
 700                         string_list_insert(available_paths, xstrdup(path));
 701                 else
 702                         ; /* ignore unknown keys */
 703         }
 704
 705         err = subprocess_read_status(process->out, &filter_status);
 706         if (err)
 707                 goto done;
 708
 709         err = strcmp(filter_status.buf, "success");
 710
 711 done:
 712         sigchain_pop(SIGPIPE);
 713
 714         if (err)
 715                 handle_filter_error(&filter_status, entry, 0);
 716         return !err;
 717 }
 718
 719 static struct convert_driver {
 720         const char *name;
 721         struct convert_driver *next;
 722         const char *smudge;
 723         const char *clean;
 724         const char *process;
 725         int required;
 726 } *user_convert, **user_convert_tail;
 727
 728 static int apply_filter(const char *path, const char *src, size_t len,
 729                         int fd, struct strbuf *dst, struct convert_driver *drv,
 730                         const unsigned int wanted_capability,
 731                         struct delayed_checkout *dco)
 732 {
 733         const char *cmd = NULL;
 734
 735         if (!drv)
 736                 return 0;
 737
 738         if (!dst)
 739                 return 1;
 740
 741         if ((wanted_capability & CAP_CLEAN) && !drv->process && drv->clean)
 742                 cmd = drv->clean;
 743         else if ((wanted_capability & CAP_SMUDGE) && !drv->process && drv->smudge)
 744                 cmd = drv->smudge;
 745
 746         if (cmd && *cmd)
 747                 return apply_single_file_filter(path, src, len, fd, dst, cmd);
 748         else if (drv->process && *drv->process)
 749                 return apply_multi_file_filter(path, src, len, fd, dst,
 750                         drv->process, wanted_capability, dco);
 751
 752         return 0;
 753 }
 754
 755 static int read_convert_config(const char *var, const char *value, void *cb)
 756 {
 757         const char *key, *name;
 758         int namelen;
 759         struct convert_driver *drv;
 760
 761         /*
 762          * External conversion drivers are configured using
 763          * "filter.<name>.variable".
 764          */
 765         if (parse_config_key(var, "filter", &name, &namelen, &key) < 0 || !name)
 766                 return 0;
 767         for (drv = user_convert; drv; drv = drv->next)
 768                 if (!strncmp(drv->name, name, namelen) && !drv->name[namelen])
 769                         break;
 770         if (!drv) {
 771                 drv = xcalloc(1, sizeof(struct convert_driver));
 772                 drv->name = xmemdupz(name, namelen);
 773                 *user_convert_tail = drv;
 774                 user_convert_tail = &(drv->next);
 775         }
 776
 777         /*
 778          * filter.<name>.smudge and filter.<name>.clean specifies
 779          * the command line:
 780          *
 781          *      command-line
 782          *
 783          * The command-line will not be interpolated in any way.
 784          */
 785
 786         if (!strcmp("smudge", key))
 787                 return git_config_string(&drv->smudge, var, value);
 788
 789         if (!strcmp("clean", key))
 790                 return git_config_string(&drv->clean, var, value);
 791
 792         if (!strcmp("process", key))
 793                 return git_config_string(&drv->process, var, value);
 794
 795         if (!strcmp("required", key)) {
 796                 drv->required = git_config_bool(var, value);
 797                 return 0;
 798         }
 799
 800         return 0;
 801 }
 802
 803 static int count_ident(const char *cp, unsigned long size)
 804 {
 805         /*
 806          * "$Id: 0000000000000000000000000000000000000000 $" <=> "$Id$"
 807          */
 808         int cnt = 0;
 809         char ch;
 810
 811         while (size) {
 812                 ch = *cp++;
 813                 size--;
 814                 if (ch != '$')
 815                         continue;
 816                 if (size < 3)
 817                         break;
 818                 if (memcmp("Id", cp, 2))
 819                         continue;
 820                 ch = cp[2];
 821                 cp += 3;
 822                 size -= 3;
 823                 if (ch == '$')
 824                         cnt++; /* $Id$ */
 825                 if (ch != ':')
 826                         continue;
 827
 828                 /*
 829                  * "$Id: ... "; scan up to the closing dollar sign and discard.
 830                  */
 831                 while (size) {
 832                         ch = *cp++;
 833                         size--;
 834                         if (ch == '$') {
 835                                 cnt++;
 836                                 break;
 837                         }
 838                         if (ch == '\n')
 839                                 break;
 840                 }
 841         }
 842         return cnt;
 843 }
 844
 845 static int ident_to_git(const char *path, const char *src, size_t len,
 846                         struct strbuf *buf, int ident)
 847 {
 848         char *dst, *dollar;
 849
 850         if (!ident || (src && !count_ident(src, len)))
 851                 return 0;
 852
 853         if (!buf)
 854                 return 1;
 855
 856         /* only grow if not in place */
 857         if (strbuf_avail(buf) + buf->len < len)
 858                 strbuf_grow(buf, len - buf->len);
 859         dst = buf->buf;
 860         for (;;) {
 861                 dollar = memchr(src, '$', len);
 862                 if (!dollar)
 863                         break;
 864                 memmove(dst, src, dollar + 1 - src);
 865                 dst += dollar + 1 - src;
 866                 len -= dollar + 1 - src;
 867                 src  = dollar + 1;
 868
 869                 if (len > 3 && !memcmp(src, "Id:", 3)) {
 870                         dollar = memchr(src + 3, '$', len - 3);
 871                         if (!dollar)
 872                                 break;
 873                         if (memchr(src + 3, '\n', dollar - src - 3)) {
 874                                 /* Line break before the next dollar. */
 875                                 continue;
 876                         }
 877
 878                         memcpy(dst, "Id$", 3);
 879                         dst += 3;
 880                         len -= dollar + 1 - src;
 881                         src  = dollar + 1;
 882                 }
 883         }
 884         memmove(dst, src, len);
 885         strbuf_setlen(buf, dst + len - buf->buf);
 886         return 1;
 887 }
 888
 889 static int ident_to_worktree(const char *path, const char *src, size_t len,
 890                              struct strbuf *buf, int ident)
 891 {
 892         unsigned char sha1[20];
 893         char *to_free = NULL, *dollar, *spc;
 894         int cnt;
 895
 896         if (!ident)
 897                 return 0;
 898
 899         cnt = count_ident(src, len);
 900         if (!cnt)
 901                 return 0;
 902
 903         /* are we "faking" in place editing ? */
 904         if (src == buf->buf)
 905                 to_free = strbuf_detach(buf, NULL);
 906         hash_sha1_file(src, len, "blob", sha1);
 907
 908         strbuf_grow(buf, len + cnt * 43);
 909         for (;;) {
 910                 /* step 1: run to the next '$' */
 911                 dollar = memchr(src, '$', len);
 912                 if (!dollar)
 913                         break;
 914                 strbuf_add(buf, src, dollar + 1 - src);
 915                 len -= dollar + 1 - src;
 916                 src  = dollar + 1;
 917
 918                 /* step 2: does it looks like a bit like Id:xxx$ or Id$ ? */
 919                 if (len < 3 || memcmp("Id", src, 2))
 920                         continue;
 921
 922                 /* step 3: skip over Id$ or Id:xxxxx$ */
 923                 if (src[2] == '$') {
 924                         src += 3;
 925                         len -= 3;
 926                 } else if (src[2] == ':') {
 927                         /*
 928                          * It's possible that an expanded Id has crept its way into the
 929                          * repository, we cope with that by stripping the expansion out.
 930                          * This is probably not a good idea, since it will cause changes
 931                          * on checkout, which won't go away by stash, but let's keep it
 932                          * for git-style ids.
 933                          */
 934                         dollar = memchr(src + 3, '$', len - 3);
 935                         if (!dollar) {
 936                                 /* incomplete keyword, no more '$', so just quit the loop */
 937                                 break;
 938                         }
 939
 940                         if (memchr(src + 3, '\n', dollar - src - 3)) {
 941                                 /* Line break before the next dollar. */
 942                                 continue;
 943                         }
 944
 945                         spc = memchr(src + 4, ' ', dollar - src - 4);
 946                         if (spc && spc < dollar-1) {
 947                                 /* There are spaces in unexpected places.
 948                                  * This is probably an id from some other
 949                                  * versioning system. Keep it for now.
 950                                  */
 951                                 continue;
 952                         }
 953
 954                         len -= dollar + 1 - src;
 955                         src  = dollar + 1;
 956                 } else {
 957                         /* it wasn't a "Id$" or "Id:xxxx$" */
 958                         continue;
 959                 }
 960
 961                 /* step 4: substitute */
 962                 strbuf_addstr(buf, "Id: ");
 963                 strbuf_add(buf, sha1_to_hex(sha1), 40);
 964                 strbuf_addstr(buf, " $");
 965         }
 966         strbuf_add(buf, src, len);
 967
 968         free(to_free);
 969         return 1;
 970 }
 971
 972 static enum crlf_action git_path_check_crlf(struct attr_check_item *check)
 973 {
 974         const char *value = check->value;
 975
 976         if (ATTR_TRUE(value))
 977                 return CRLF_TEXT;
 978         else if (ATTR_FALSE(value))
 979                 return CRLF_BINARY;
 980         else if (ATTR_UNSET(value))
 981                 ;
 982         else if (!strcmp(value, "input"))
 983                 return CRLF_TEXT_INPUT;
 984         else if (!strcmp(value, "auto"))
 985                 return CRLF_AUTO;
 986         return CRLF_UNDEFINED;
 987 }
 988
 989 static enum eol git_path_check_eol(struct attr_check_item *check)
 990 {
 991         const char *value = check->value;
 992
 993         if (ATTR_UNSET(value))
 994                 ;
 995         else if (!strcmp(value, "lf"))
 996                 return EOL_LF;
 997         else if (!strcmp(value, "crlf"))
 998                 return EOL_CRLF;
 999         return EOL_UNSET;
1000 }
1001
1002 static struct convert_driver *git_path_check_convert(struct attr_check_item *check)
1003 {
1004         const char *value = check->value;
1005         struct convert_driver *drv;
1006
1007         if (ATTR_TRUE(value) || ATTR_FALSE(value) || ATTR_UNSET(value))
1008                 return NULL;
1009         for (drv = user_convert; drv; drv = drv->next)
1010                 if (!strcmp(value, drv->name))
1011                         return drv;
1012         return NULL;
1013 }
1014
1015 static int git_path_check_ident(struct attr_check_item *check)
1016 {
1017         const char *value = check->value;
1018
1019         return !!ATTR_TRUE(value);
1020 }
1021
1022 struct conv_attrs {
1023         struct convert_driver *drv;
1024         enum crlf_action attr_action; /* What attr says */
1025         enum crlf_action crlf_action; /* When no attr is set, use core.autocrlf */
1026         int ident;
1027 };
1028
1029 static void convert_attrs(struct conv_attrs *ca, const char *path)
1030 {
1031         static struct attr_check *check;
1032
1033         if (!check) {
1034                 check = attr_check_initl("crlf", "ident", "filter",
1035                                          "eol", "text", NULL);
1036                 user_convert_tail = &user_convert;
1037                 git_config(read_convert_config, NULL);
1038         }
1039
1040         if (!git_check_attr(path, check)) {
1041                 struct attr_check_item *ccheck = check->items;
1042                 ca->crlf_action = git_path_check_crlf(ccheck + 4);
1043                 if (ca->crlf_action == CRLF_UNDEFINED)
1044                         ca->crlf_action = git_path_check_crlf(ccheck + 0);
1045                 ca->attr_action = ca->crlf_action;
1046                 ca->ident = git_path_check_ident(ccheck + 1);
1047                 ca->drv = git_path_check_convert(ccheck + 2);
1048                 if (ca->crlf_action != CRLF_BINARY) {
1049                         enum eol eol_attr = git_path_check_eol(ccheck + 3);
1050                         if (ca->crlf_action == CRLF_AUTO && eol_attr == EOL_LF)
1051                                 ca->crlf_action = CRLF_AUTO_INPUT;
1052                         else if (ca->crlf_action == CRLF_AUTO && eol_attr == EOL_CRLF)
1053                                 ca->crlf_action = CRLF_AUTO_CRLF;
1054                         else if (eol_attr == EOL_LF)
1055                                 ca->crlf_action = CRLF_TEXT_INPUT;
1056                         else if (eol_attr == EOL_CRLF)
1057                                 ca->crlf_action = CRLF_TEXT_CRLF;
1058                 }
1059                 ca->attr_action = ca->crlf_action;
1060         } else {
1061                 ca->drv = NULL;
1062                 ca->crlf_action = CRLF_UNDEFINED;
1063                 ca->ident = 0;
1064         }
1065         if (ca->crlf_action == CRLF_TEXT)
1066                 ca->crlf_action = text_eol_is_crlf() ? CRLF_TEXT_CRLF : CRLF_TEXT_INPUT;
1067         if (ca->crlf_action == CRLF_UNDEFINED && auto_crlf == AUTO_CRLF_FALSE)
1068                 ca->crlf_action = CRLF_BINARY;
1069         if (ca->crlf_action == CRLF_UNDEFINED && auto_crlf == AUTO_CRLF_TRUE)
1070                 ca->crlf_action = CRLF_AUTO_CRLF;
1071         if (ca->crlf_action == CRLF_UNDEFINED && auto_crlf == AUTO_CRLF_INPUT)
1072                 ca->crlf_action = CRLF_AUTO_INPUT;
1073 }
1074
1075 int would_convert_to_git_filter_fd(const char *path)
1076 {
1077         struct conv_attrs ca;
1078
1079         convert_attrs(&ca, path);
1080         if (!ca.drv)
1081                 return 0;
1082
1083         /*
1084          * Apply a filter to an fd only if the filter is required to succeed.
1085          * We must die if the filter fails, because the original data before
1086          * filtering is not available.
1087          */
1088         if (!ca.drv->required)
1089                 return 0;
1090
1091         return apply_filter(path, NULL, 0, -1, NULL, ca.drv, CAP_CLEAN, NULL);
1092 }
1093
1094 const char *get_convert_attr_ascii(const char *path)
1095 {
1096         struct conv_attrs ca;
1097
1098         convert_attrs(&ca, path);
1099         switch (ca.attr_action) {
1100         case CRLF_UNDEFINED:
1101                 return "";
1102         case CRLF_BINARY:
1103                 return "-text";
1104         case CRLF_TEXT:
1105                 return "text";
1106         case CRLF_TEXT_INPUT:
1107                 return "text eol=lf";
1108         case CRLF_TEXT_CRLF:
1109                 return "text eol=crlf";
1110         case CRLF_AUTO:
1111                 return "text=auto";
1112         case CRLF_AUTO_CRLF:
1113                 return "text=auto eol=crlf";
1114         case CRLF_AUTO_INPUT:
1115                 return "text=auto eol=lf";
1116         }
1117         return "";
1118 }
1119
1120 int convert_to_git(const struct index_state *istate,
1121                    const char *path, const char *src, size_t len,
1122                    struct strbuf *dst, enum safe_crlf checksafe)
1123 {
1124         int ret = 0;
1125         struct conv_attrs ca;
1126
1127         convert_attrs(&ca, path);
1128
1129         ret |= apply_filter(path, src, len, -1, dst, ca.drv, CAP_CLEAN, NULL);
1130         if (!ret && ca.drv && ca.drv->required)
1131                 die("%s: clean filter '%s' failed", path, ca.drv->name);
1132
1133         if (ret && dst) {
1134                 src = dst->buf;
1135                 len = dst->len;
1136         }
1137         if (checksafe != SAFE_CRLF_KEEP_CRLF) {
1138                 ret |= crlf_to_git(istate, path, src, len, dst, ca.crlf_action, checksafe);
1139                 if (ret && dst) {
1140                         src = dst->buf;
1141                         len = dst->len;
1142                 }
1143         }
1144         return ret | ident_to_git(path, src, len, dst, ca.ident);
1145 }
1146
1147 void convert_to_git_filter_fd(const struct index_state *istate,
1148                               const char *path, int fd, struct strbuf *dst,
1149                               enum safe_crlf checksafe)
1150 {
1151         struct conv_attrs ca;
1152         convert_attrs(&ca, path);
1153
1154         assert(ca.drv);
1155         assert(ca.drv->clean || ca.drv->process);
1156
1157         if (!apply_filter(path, NULL, 0, fd, dst, ca.drv, CAP_CLEAN, NULL))
1158                 die("%s: clean filter '%s' failed", path, ca.drv->name);
1159
1160         crlf_to_git(istate, path, dst->buf, dst->len, dst, ca.crlf_action, checksafe);
1161         ident_to_git(path, dst->buf, dst->len, dst, ca.ident);
1162 }
1163
1164 static int convert_to_working_tree_internal(const char *path, const char *src,
1165                                             size_t len, struct strbuf *dst,
1166                                             int normalizing, struct delayed_checkout *dco)
1167 {
1168         int ret = 0, ret_filter = 0;
1169         struct conv_attrs ca;
1170
1171         convert_attrs(&ca, path);
1172
1173         ret |= ident_to_worktree(path, src, len, dst, ca.ident);
1174         if (ret) {
1175                 src = dst->buf;
1176                 len = dst->len;
1177         }
1178         /*
1179          * CRLF conversion can be skipped if normalizing, unless there
1180          * is a smudge or process filter (even if the process filter doesn't
1181          * support smudge).  The filters might expect CRLFs.
1182          */
1183         if ((ca.drv && (ca.drv->smudge || ca.drv->process)) || !normalizing) {
1184                 ret |= crlf_to_worktree(path, src, len, dst, ca.crlf_action);
1185                 if (ret) {
1186                         src = dst->buf;
1187                         len = dst->len;
1188                 }
1189         }
1190
1191         ret_filter = apply_filter(
1192                 path, src, len, -1, dst, ca.drv, CAP_SMUDGE, dco);
1193         if (!ret_filter && ca.drv && ca.drv->required)
1194                 die("%s: smudge filter %s failed", path, ca.drv->name);
1195
1196         return ret | ret_filter;
1197 }
1198
1199 int async_convert_to_working_tree(const char *path, const char *src,
1200                                   size_t len, struct strbuf *dst,
1201                                   void *dco)
1202 {
1203         return convert_to_working_tree_internal(path, src, len, dst, 0, dco);
1204 }
1205
1206 int convert_to_working_tree(const char *path, const char *src, size_t len, struct strbuf *dst)
1207 {
1208         return convert_to_working_tree_internal(path, src, len, dst, 0, NULL);
1209 }
1210
1211 int renormalize_buffer(const struct index_state *istate, const char *path,
1212                        const char *src, size_t len, struct strbuf *dst)
1213 {
1214         int ret = convert_to_working_tree_internal(path, src, len, dst, 1, NULL);
1215         if (ret) {
1216                 src = dst->buf;
1217                 len = dst->len;
1218         }
1219         return ret | convert_to_git(istate, path, src, len, dst, SAFE_CRLF_RENORMALIZE);
1220 }
1221
1222 /*****************************************************************
1223  *
1224  * Streaming conversion support
1225  *
1226  *****************************************************************/
1227
1228 typedef int (*filter_fn)(struct stream_filter *,
1229                          const char *input, size_t *isize_p,
1230                          char *output, size_t *osize_p);
1231 typedef void (*free_fn)(struct stream_filter *);
1232
1233 struct stream_filter_vtbl {
1234         filter_fn filter;
1235         free_fn free;
1236 };
1237
1238 struct stream_filter {
1239         struct stream_filter_vtbl *vtbl;
1240 };
1241
1242 static int null_filter_fn(struct stream_filter *filter,
1243                           const char *input, size_t *isize_p,
1244                           char *output, size_t *osize_p)
1245 {
1246         size_t count;
1247
1248         if (!input)
1249                 return 0; /* we do not keep any states */
1250         count = *isize_p;
1251         if (*osize_p < count)
1252                 count = *osize_p;
1253         if (count) {
1254                 memmove(output, input, count);
1255                 *isize_p -= count;
1256                 *osize_p -= count;
1257         }
1258         return 0;
1259 }
1260
1261 static void null_free_fn(struct stream_filter *filter)
1262 {
1263         ; /* nothing -- null instances are shared */
1264 }
1265
1266 static struct stream_filter_vtbl null_vtbl = {
1267         null_filter_fn,
1268         null_free_fn,
1269 };
1270
1271 static struct stream_filter null_filter_singleton = {
1272         &null_vtbl,
1273 };
1274
1275 int is_null_stream_filter(struct stream_filter *filter)
1276 {
1277         return filter == &null_filter_singleton;
1278 }
1279
1280
1281 /*
1282  * LF-to-CRLF filter
1283  */
1284
1285 struct lf_to_crlf_filter {
1286         struct stream_filter filter;
1287         unsigned has_held:1;
1288         char held;
1289 };
1290
1291 static int lf_to_crlf_filter_fn(struct stream_filter *filter,
1292                                 const char *input, size_t *isize_p,
1293                                 char *output, size_t *osize_p)
1294 {
1295         size_t count, o = 0;
1296         struct lf_to_crlf_filter *lf_to_crlf = (struct lf_to_crlf_filter *)filter;
1297
1298         /*
1299          * We may be holding onto the CR to see if it is followed by a
1300          * LF, in which case we would need to go to the main loop.
1301          * Otherwise, just emit it to the output stream.
1302          */
1303         if (lf_to_crlf->has_held && (lf_to_crlf->held != '\r' || !input)) {
1304                 output[o++] = lf_to_crlf->held;
1305                 lf_to_crlf->has_held = 0;
1306         }
1307
1308         /* We are told to drain */
1309         if (!input) {
1310                 *osize_p -= o;
1311                 return 0;
1312         }
1313
1314         count = *isize_p;
1315         if (count || lf_to_crlf->has_held) {
1316                 size_t i;
1317                 int was_cr = 0;
1318
1319                 if (lf_to_crlf->has_held) {
1320                         was_cr = 1;
1321                         lf_to_crlf->has_held = 0;
1322                 }
1323
1324                 for (i = 0; o < *osize_p && i < count; i++) {
1325                         char ch = input[i];
1326
1327                         if (ch == '\n') {
1328                                 output[o++] = '\r';
1329                         } else if (was_cr) {
1330                                 /*
1331                                  * Previous round saw CR and it is not followed
1332                                  * by a LF; emit the CR before processing the
1333                                  * current character.
1334                                  */
1335                                 output[o++] = '\r';
1336                         }
1337
1338                         /*
1339                          * We may have consumed the last output slot,
1340                          * in which case we need to break out of this
1341                          * loop; hold the current character before
1342                          * returning.
1343                          */
1344                         if (*osize_p <= o) {
1345                                 lf_to_crlf->has_held = 1;
1346                                 lf_to_crlf->held = ch;
1347                                 continue; /* break but increment i */
1348                         }
1349
1350                         if (ch == '\r') {
1351                                 was_cr = 1;
1352                                 continue;
1353                         }
1354
1355                         was_cr = 0;
1356                         output[o++] = ch;
1357                 }
1358
1359                 *osize_p -= o;
1360                 *isize_p -= i;
1361
1362                 if (!lf_to_crlf->has_held && was_cr) {
1363                         lf_to_crlf->has_held = 1;
1364                         lf_to_crlf->held = '\r';
1365                 }
1366         }
1367         return 0;
1368 }
1369
1370 static void lf_to_crlf_free_fn(struct stream_filter *filter)
1371 {
1372         free(filter);
1373 }
1374
1375 static struct stream_filter_vtbl lf_to_crlf_vtbl = {
1376         lf_to_crlf_filter_fn,
1377         lf_to_crlf_free_fn,
1378 };
1379
1380 static struct stream_filter *lf_to_crlf_filter(void)
1381 {
1382         struct lf_to_crlf_filter *lf_to_crlf = xcalloc(1, sizeof(*lf_to_crlf));
1383
1384         lf_to_crlf->filter.vtbl = &lf_to_crlf_vtbl;
1385         return (struct stream_filter *)lf_to_crlf;
1386 }
1387
1388 /*
1389  * Cascade filter
1390  */
1391 #define FILTER_BUFFER 1024
1392 struct cascade_filter {
1393         struct stream_filter filter;
1394         struct stream_filter *one;
1395         struct stream_filter *two;
1396         char buf[FILTER_BUFFER];
1397         int end, ptr;
1398 };
1399
1400 static int cascade_filter_fn(struct stream_filter *filter,
1401                              const char *input, size_t *isize_p,
1402                              char *output, size_t *osize_p)
1403 {
1404         struct cascade_filter *cas = (struct cascade_filter *) filter;
1405         size_t filled = 0;
1406         size_t sz = *osize_p;
1407         size_t to_feed, remaining;
1408
1409         /*
1410          * input -- (one) --> buf -- (two) --> output
1411          */
1412         while (filled < sz) {
1413                 remaining = sz - filled;
1414
1415                 /* do we already have something to feed two with? */
1416                 if (cas->ptr < cas->end) {
1417                         to_feed = cas->end - cas->ptr;
1418                         if (stream_filter(cas->two,
1419                                           cas->buf + cas->ptr, &to_feed,
1420                                           output + filled, &remaining))
1421                                 return -1;
1422                         cas->ptr += (cas->end - cas->ptr) - to_feed;
1423                         filled = sz - remaining;
1424                         continue;
1425                 }
1426
1427                 /* feed one from upstream and have it emit into our buffer */
1428                 to_feed = input ? *isize_p : 0;
1429                 if (input && !to_feed)
1430                         break;
1431                 remaining = sizeof(cas->buf);
1432                 if (stream_filter(cas->one,
1433                                   input, &to_feed,
1434                                   cas->buf, &remaining))
1435                         return -1;
1436                 cas->end = sizeof(cas->buf) - remaining;
1437                 cas->ptr = 0;
1438                 if (input) {
1439                         size_t fed = *isize_p - to_feed;
1440                         *isize_p -= fed;
1441                         input += fed;
1442                 }
1443
1444                 /* do we know that we drained one completely? */
1445                 if (input || cas->end)
1446                         continue;
1447
1448                 /* tell two to drain; we have nothing more to give it */
1449                 to_feed = 0;
1450                 remaining = sz - filled;
1451                 if (stream_filter(cas->two,
1452                                   NULL, &to_feed,
1453                                   output + filled, &remaining))
1454                         return -1;
1455                 if (remaining == (sz - filled))
1456                         break; /* completely drained two */
1457                 filled = sz - remaining;
1458         }
1459         *osize_p -= filled;
1460         return 0;
1461 }
1462
1463 static void cascade_free_fn(struct stream_filter *filter)
1464 {
1465         struct cascade_filter *cas = (struct cascade_filter *)filter;
1466         free_stream_filter(cas->one);
1467         free_stream_filter(cas->two);
1468         free(filter);
1469 }
1470
1471 static struct stream_filter_vtbl cascade_vtbl = {
1472         cascade_filter_fn,
1473         cascade_free_fn,
1474 };
1475
1476 static struct stream_filter *cascade_filter(struct stream_filter *one,
1477                                             struct stream_filter *two)
1478 {
1479         struct cascade_filter *cascade;
1480
1481         if (!one || is_null_stream_filter(one))
1482                 return two;
1483         if (!two || is_null_stream_filter(two))
1484                 return one;
1485
1486         cascade = xmalloc(sizeof(*cascade));
1487         cascade->one = one;
1488         cascade->two = two;
1489         cascade->end = cascade->ptr = 0;
1490         cascade->filter.vtbl = &cascade_vtbl;
1491         return (struct stream_filter *)cascade;
1492 }
1493
1494 /*
1495  * ident filter
1496  */
1497 #define IDENT_DRAINING (-1)
1498 #define IDENT_SKIPPING (-2)
1499 struct ident_filter {
1500         struct stream_filter filter;
1501         struct strbuf left;
1502         int state;
1503         char ident[45]; /* ": x40 $" */
1504 };
1505
1506 static int is_foreign_ident(const char *str)
1507 {
1508         int i;
1509
1510         if (!skip_prefix(str, "$Id: ", &str))
1511                 return 0;
1512         for (i = 0; str[i]; i++) {
1513                 if (isspace(str[i]) && str[i+1] != '$')
1514                         return 1;
1515         }
1516         return 0;
1517 }
1518
1519 static void ident_drain(struct ident_filter *ident, char **output_p, size_t *osize_p)
1520 {
1521         size_t to_drain = ident->left.len;
1522
1523         if (*osize_p < to_drain)
1524                 to_drain = *osize_p;
1525         if (to_drain) {
1526                 memcpy(*output_p, ident->left.buf, to_drain);
1527                 strbuf_remove(&ident->left, 0, to_drain);
1528                 *output_p += to_drain;
1529                 *osize_p -= to_drain;
1530         }
1531         if (!ident->left.len)
1532                 ident->state = 0;
1533 }
1534
1535 static int ident_filter_fn(struct stream_filter *filter,
1536                            const char *input, size_t *isize_p,
1537                            char *output, size_t *osize_p)
1538 {
1539         struct ident_filter *ident = (struct ident_filter *)filter;
1540         static const char head[] = "$Id";
1541
1542         if (!input) {
1543                 /* drain upon eof */
1544                 switch (ident->state) {
1545                 default:
1546                         strbuf_add(&ident->left, head, ident->state);
1547                 case IDENT_SKIPPING:
1548                         /* fallthru */
1549                 case IDENT_DRAINING:
1550                         ident_drain(ident, &output, osize_p);
1551                 }
1552                 return 0;
1553         }
1554
1555         while (*isize_p || (ident->state == IDENT_DRAINING)) {
1556                 int ch;
1557
1558                 if (ident->state == IDENT_DRAINING) {
1559                         ident_drain(ident, &output, osize_p);
1560                         if (!*osize_p)
1561                                 break;
1562                         continue;
1563                 }
1564
1565                 ch = *(input++);
1566                 (*isize_p)--;
1567
1568                 if (ident->state == IDENT_SKIPPING) {
1569                         /*
1570                          * Skipping until '$' or LF, but keeping them
1571                          * in case it is a foreign ident.
1572                          */
1573                         strbuf_addch(&ident->left, ch);
1574                         if (ch != '\n' && ch != '$')
1575                                 continue;
1576                         if (ch == '$' && !is_foreign_ident(ident->left.buf)) {
1577                                 strbuf_setlen(&ident->left, sizeof(head) - 1);
1578                                 strbuf_addstr(&ident->left, ident->ident);
1579                         }
1580                         ident->state = IDENT_DRAINING;
1581                         continue;
1582                 }
1583
1584                 if (ident->state < sizeof(head) &&
1585                     head[ident->state] == ch) {
1586                         ident->state++;
1587                         continue;
1588                 }
1589
1590                 if (ident->state)
1591                         strbuf_add(&ident->left, head, ident->state);
1592                 if (ident->state == sizeof(head) - 1) {
1593                         if (ch != ':' && ch != '$') {
1594                                 strbuf_addch(&ident->left, ch);
1595                                 ident->state = 0;
1596                                 continue;
1597                         }
1598
1599                         if (ch == ':') {
1600                                 strbuf_addch(&ident->left, ch);
1601                                 ident->state = IDENT_SKIPPING;
1602                         } else {
1603                                 strbuf_addstr(&ident->left, ident->ident);
1604                                 ident->state = IDENT_DRAINING;
1605                         }
1606                         continue;
1607                 }
1608
1609                 strbuf_addch(&ident->left, ch);
1610                 ident->state = IDENT_DRAINING;
1611         }
1612         return 0;
1613 }
1614
1615 static void ident_free_fn(struct stream_filter *filter)
1616 {
1617         struct ident_filter *ident = (struct ident_filter *)filter;
1618         strbuf_release(&ident->left);
1619         free(filter);
1620 }
1621
1622 static struct stream_filter_vtbl ident_vtbl = {
1623         ident_filter_fn,
1624         ident_free_fn,
1625 };
1626
1627 static struct stream_filter *ident_filter(const unsigned char *sha1)
1628 {
1629         struct ident_filter *ident = xmalloc(sizeof(*ident));
1630
1631         xsnprintf(ident->ident, sizeof(ident->ident),
1632                   ": %s $", sha1_to_hex(sha1));
1633         strbuf_init(&ident->left, 0);
1634         ident->filter.vtbl = &ident_vtbl;
1635         ident->state = 0;
1636         return (struct stream_filter *)ident;
1637 }
1638
1639 /*
1640  * Return an appropriately constructed filter for the path, or NULL if
1641  * the contents cannot be filtered without reading the whole thing
1642  * in-core.
1643  *
1644  * Note that you would be crazy to set CRLF, smuge/clean or ident to a
1645  * large binary blob you would want us not to slurp into the memory!
1646  */
1647 struct stream_filter *get_stream_filter(const char *path, const unsigned char *sha1)
1648 {
1649         struct conv_attrs ca;
1650         struct stream_filter *filter = NULL;
1651
1652         convert_attrs(&ca, path);
1653         if (ca.drv && (ca.drv->process || ca.drv->smudge || ca.drv->clean))
1654                 return NULL;
1655
1656         if (ca.crlf_action == CRLF_AUTO || ca.crlf_action == CRLF_AUTO_CRLF)
1657                 return NULL;
1658
1659         if (ca.ident)
1660                 filter = ident_filter(sha1);
1661
1662         if (output_eol(ca.crlf_action) == EOL_CRLF)
1663                 filter = cascade_filter(filter, lf_to_crlf_filter());
1664         else
1665                 filter = cascade_filter(filter, &null_filter_singleton);
1666
1667         return filter;
1668 }
1669
1670 void free_stream_filter(struct stream_filter *filter)
1671 {
1672         filter->vtbl->free(filter);
1673 }
1674
1675 int stream_filter(struct stream_filter *filter,
1676                   const char *input, size_t *isize_p,
1677                   char *output, size_t *osize_p)
1678 {
1679         return filter->vtbl->filter(filter, input, isize_p, output, osize_p);
1680 }