convert.c

   1 #include "cache.h"
   2 #include "attr.h"
   3 #include "run-command.h"
   4 #include "quote.h"
   5 #include "sigchain.h"
   6 #include "pkt-line.h"
   7 #include "sub-process.h"
   8
   9 /*
  10  * convert.c - convert a file when checking it out and checking it in.
  11  *
  12  * This should use the pathname to decide on whether it wants to do some
  13  * more interesting conversions (automatic gzip/unzip, general format
  14  * conversions etc etc), but by default it just does automatic CRLF<->LF
  15  * translation when the "text" attribute or "auto_crlf" option is set.
  16  */
  17
  18 /* Stat bits: When BIN is set, the txt bits are unset */
  19 #define CONVERT_STAT_BITS_TXT_LF    0x1
  20 #define CONVERT_STAT_BITS_TXT_CRLF  0x2
  21 #define CONVERT_STAT_BITS_BIN       0x4
  22
  23 enum crlf_action {
  24         CRLF_UNDEFINED,
  25         CRLF_BINARY,
  26         CRLF_TEXT,
  27         CRLF_TEXT_INPUT,
  28         CRLF_TEXT_CRLF,
  29         CRLF_AUTO,
  30         CRLF_AUTO_INPUT,
  31         CRLF_AUTO_CRLF
  32 };
  33
  34 struct text_stat {
  35         /* NUL, CR, LF and CRLF counts */
  36         unsigned nul, lonecr, lonelf, crlf;
  37
  38         /* These are just approximations! */
  39         unsigned printable, nonprintable;
  40 };
  41
  42 static void gather_stats(const char *buf, unsigned long size, struct text_stat *stats)
  43 {
  44         unsigned long i;
  45
  46         memset(stats, 0, sizeof(*stats));
  47
  48         for (i = 0; i < size; i++) {
  49                 unsigned char c = buf[i];
  50                 if (c == '\r') {
  51                         if (i+1 < size && buf[i+1] == '\n') {
  52                                 stats->crlf++;
  53                                 i++;
  54                         } else
  55                                 stats->lonecr++;
  56                         continue;
  57                 }
  58                 if (c == '\n') {
  59                         stats->lonelf++;
  60                         continue;
  61                 }
  62                 if (c == 127)
  63                         /* DEL */
  64                         stats->nonprintable++;
  65                 else if (c < 32) {
  66                         switch (c) {
  67                                 /* BS, HT, ESC and FF */
  68                         case '\b': case '\t': case '\033': case '\014':
  69                                 stats->printable++;
  70                                 break;
  71                         case 0:
  72                                 stats->nul++;
  73                                 /* fall through */
  74                         default:
  75                                 stats->nonprintable++;
  76                         }
  77                 }
  78                 else
  79                         stats->printable++;
  80         }
  81
  82         /* If file ends with EOF then don't count this EOF as non-printable. */
  83         if (size >= 1 && buf[size-1] == '\032')
  84                 stats->nonprintable--;
  85 }
  86
  87 /*
  88  * The same heuristics as diff.c::mmfile_is_binary()
  89  * We treat files with bare CR as binary
  90  */
  91 static int convert_is_binary(unsigned long size, const struct text_stat *stats)
  92 {
  93         if (stats->lonecr)
  94                 return 1;
  95         if (stats->nul)
  96                 return 1;
  97         if ((stats->printable >> 7) < stats->nonprintable)
  98                 return 1;
  99         return 0;
 100 }
 101
 102 static unsigned int gather_convert_stats(const char *data, unsigned long size)
 103 {
 104         struct text_stat stats;
 105         int ret = 0;
 106         if (!data || !size)
 107                 return 0;
 108         gather_stats(data, size, &stats);
 109         if (convert_is_binary(size, &stats))
 110                 ret |= CONVERT_STAT_BITS_BIN;
 111         if (stats.crlf)
 112                 ret |= CONVERT_STAT_BITS_TXT_CRLF;
 113         if (stats.lonelf)
 114                 ret |=  CONVERT_STAT_BITS_TXT_LF;
 115
 116         return ret;
 117 }
 118
 119 static const char *gather_convert_stats_ascii(const char *data, unsigned long size)
 120 {
 121         unsigned int convert_stats = gather_convert_stats(data, size);
 122
 123         if (convert_stats & CONVERT_STAT_BITS_BIN)
 124                 return "-text";
 125         switch (convert_stats) {
 126         case CONVERT_STAT_BITS_TXT_LF:
 127                 return "lf";
 128         case CONVERT_STAT_BITS_TXT_CRLF:
 129                 return "crlf";
 130         case CONVERT_STAT_BITS_TXT_LF | CONVERT_STAT_BITS_TXT_CRLF:
 131                 return "mixed";
 132         default:
 133                 return "none";
 134         }
 135 }
 136
 137 const char *get_cached_convert_stats_ascii(const char *path)
 138 {
 139         const char *ret;
 140         unsigned long sz;
 141         void *data = read_blob_data_from_cache(path, &sz);
 142         ret = gather_convert_stats_ascii(data, sz);
 143         free(data);
 144         return ret;
 145 }
 146
 147 const char *get_wt_convert_stats_ascii(const char *path)
 148 {
 149         const char *ret = "";
 150         struct strbuf sb = STRBUF_INIT;
 151         if (strbuf_read_file(&sb, path, 0) >= 0)
 152                 ret = gather_convert_stats_ascii(sb.buf, sb.len);
 153         strbuf_release(&sb);
 154         return ret;
 155 }
 156
 157 static int text_eol_is_crlf(void)
 158 {
 159         if (auto_crlf == AUTO_CRLF_TRUE)
 160                 return 1;
 161         else if (auto_crlf == AUTO_CRLF_INPUT)
 162                 return 0;
 163         if (core_eol == EOL_CRLF)
 164                 return 1;
 165         if (core_eol == EOL_UNSET && EOL_NATIVE == EOL_CRLF)
 166                 return 1;
 167         return 0;
 168 }
 169
 170 static enum eol output_eol(enum crlf_action crlf_action)
 171 {
 172         switch (crlf_action) {
 173         case CRLF_BINARY:
 174                 return EOL_UNSET;
 175         case CRLF_TEXT_CRLF:
 176                 return EOL_CRLF;
 177         case CRLF_TEXT_INPUT:
 178                 return EOL_LF;
 179         case CRLF_UNDEFINED:
 180         case CRLF_AUTO_CRLF:
 181                 return EOL_CRLF;
 182         case CRLF_AUTO_INPUT:
 183                 return EOL_LF;
 184         case CRLF_TEXT:
 185         case CRLF_AUTO:
 186                 /* fall through */
 187                 return text_eol_is_crlf() ? EOL_CRLF : EOL_LF;
 188         }
 189         warning("Illegal crlf_action %d\n", (int)crlf_action);
 190         return core_eol;
 191 }
 192
 193 static void check_safe_crlf(const char *path, enum crlf_action crlf_action,
 194                             struct text_stat *old_stats, struct text_stat *new_stats,
 195                             enum safe_crlf checksafe)
 196 {
 197         if (old_stats->crlf && !new_stats->crlf ) {
 198                 /*
 199                  * CRLFs would not be restored by checkout
 200                  */
 201                 if (checksafe == SAFE_CRLF_WARN)
 202                         warning(_("CRLF will be replaced by LF in %s.\n"
 203                                   "The file will have its original line"
 204                                   " endings in your working directory."), path);
 205                 else /* i.e. SAFE_CRLF_FAIL */
 206                         die(_("CRLF would be replaced by LF in %s."), path);
 207         } else if (old_stats->lonelf && !new_stats->lonelf ) {
 208                 /*
 209                  * CRLFs would be added by checkout
 210                  */
 211                 if (checksafe == SAFE_CRLF_WARN)
 212                         warning(_("LF will be replaced by CRLF in %s.\n"
 213                                   "The file will have its original line"
 214                                   " endings in your working directory."), path);
 215                 else /* i.e. SAFE_CRLF_FAIL */
 216                         die(_("LF would be replaced by CRLF in %s"), path);
 217         }
 218 }
 219
 220 static int has_cr_in_index(const char *path)
 221 {
 222         unsigned long sz;
 223         void *data;
 224         int has_cr;
 225
 226         data = read_blob_data_from_cache(path, &sz);
 227         if (!data)
 228                 return 0;
 229         has_cr = memchr(data, '\r', sz) != NULL;
 230         free(data);
 231         return has_cr;
 232 }
 233
 234 static int will_convert_lf_to_crlf(size_t len, struct text_stat *stats,
 235                                    enum crlf_action crlf_action)
 236 {
 237         if (output_eol(crlf_action) != EOL_CRLF)
 238                 return 0;
 239         /* No "naked" LF? Nothing to convert, regardless. */
 240         if (!stats->lonelf)
 241                 return 0;
 242
 243         if (crlf_action == CRLF_AUTO || crlf_action == CRLF_AUTO_INPUT || crlf_action == CRLF_AUTO_CRLF) {
 244                 /* If we have any CR or CRLF line endings, we do not touch it */
 245                 /* This is the new safer autocrlf-handling */
 246                 if (stats->lonecr || stats->crlf)
 247                         return 0;
 248
 249                 if (convert_is_binary(len, stats))
 250                         return 0;
 251         }
 252         return 1;
 253
 254 }
 255
 256 static int crlf_to_git(const char *path, const char *src, size_t len,
 257                        struct strbuf *buf,
 258                        enum crlf_action crlf_action, enum safe_crlf checksafe)
 259 {
 260         struct text_stat stats;
 261         char *dst;
 262         int convert_crlf_into_lf;
 263
 264         if (crlf_action == CRLF_BINARY ||
 265             (src && !len))
 266                 return 0;
 267
 268         /*
 269          * If we are doing a dry-run and have no source buffer, there is
 270          * nothing to analyze; we must assume we would convert.
 271          */
 272         if (!buf && !src)
 273                 return 1;
 274
 275         gather_stats(src, len, &stats);
 276         /* Optimization: No CRLF? Nothing to convert, regardless. */
 277         convert_crlf_into_lf = !!stats.crlf;
 278
 279         if (crlf_action == CRLF_AUTO || crlf_action == CRLF_AUTO_INPUT || crlf_action == CRLF_AUTO_CRLF) {
 280                 if (convert_is_binary(len, &stats))
 281                         return 0;
 282                 /*
 283                  * If the file in the index has any CR in it, do not
 284                  * convert.  This is the new safer autocrlf handling,
 285                  * unless we want to renormalize in a merge or
 286                  * cherry-pick.
 287                  */
 288                 if ((checksafe != SAFE_CRLF_RENORMALIZE) && has_cr_in_index(path))
 289                         convert_crlf_into_lf = 0;
 290         }
 291         if ((checksafe == SAFE_CRLF_WARN ||
 292             (checksafe == SAFE_CRLF_FAIL)) && len) {
 293                 struct text_stat new_stats;
 294                 memcpy(&new_stats, &stats, sizeof(new_stats));
 295                 /* simulate "git add" */
 296                 if (convert_crlf_into_lf) {
 297                         new_stats.lonelf += new_stats.crlf;
 298                         new_stats.crlf = 0;
 299                 }
 300                 /* simulate "git checkout" */
 301                 if (will_convert_lf_to_crlf(len, &new_stats, crlf_action)) {
 302                         new_stats.crlf += new_stats.lonelf;
 303                         new_stats.lonelf = 0;
 304                 }
 305                 check_safe_crlf(path, crlf_action, &stats, &new_stats, checksafe);
 306         }
 307         if (!convert_crlf_into_lf)
 308                 return 0;
 309
 310         /*
 311          * At this point all of our source analysis is done, and we are sure we
 312          * would convert. If we are in dry-run mode, we can give an answer.
 313          */
 314         if (!buf)
 315                 return 1;
 316
 317         /* only grow if not in place */
 318         if (strbuf_avail(buf) + buf->len < len)
 319                 strbuf_grow(buf, len - buf->len);
 320         dst = buf->buf;
 321         if (crlf_action == CRLF_AUTO || crlf_action == CRLF_AUTO_INPUT || crlf_action == CRLF_AUTO_CRLF) {
 322                 /*
 323                  * If we guessed, we already know we rejected a file with
 324                  * lone CR, and we can strip a CR without looking at what
 325                  * follow it.
 326                  */
 327                 do {
 328                         unsigned char c = *src++;
 329                         if (c != '\r')
 330                                 *dst++ = c;
 331                 } while (--len);
 332         } else {
 333                 do {
 334                         unsigned char c = *src++;
 335                         if (! (c == '\r' && (1 < len && *src == '\n')))
 336                                 *dst++ = c;
 337                 } while (--len);
 338         }
 339         strbuf_setlen(buf, dst - buf->buf);
 340         return 1;
 341 }
 342
 343 static int crlf_to_worktree(const char *path, const char *src, size_t len,
 344                             struct strbuf *buf, enum crlf_action crlf_action)
 345 {
 346         char *to_free = NULL;
 347         struct text_stat stats;
 348
 349         if (!len || output_eol(crlf_action) != EOL_CRLF)
 350                 return 0;
 351
 352         gather_stats(src, len, &stats);
 353         if (!will_convert_lf_to_crlf(len, &stats, crlf_action))
 354                 return 0;
 355
 356         /* are we "faking" in place editing ? */
 357         if (src == buf->buf)
 358                 to_free = strbuf_detach(buf, NULL);
 359
 360         strbuf_grow(buf, len + stats.lonelf);
 361         for (;;) {
 362                 const char *nl = memchr(src, '\n', len);
 363                 if (!nl)
 364                         break;
 365                 if (nl > src && nl[-1] == '\r') {
 366                         strbuf_add(buf, src, nl + 1 - src);
 367                 } else {
 368                         strbuf_add(buf, src, nl - src);
 369                         strbuf_addstr(buf, "\r\n");
 370                 }
 371                 len -= nl + 1 - src;
 372                 src  = nl + 1;
 373         }
 374         strbuf_add(buf, src, len);
 375
 376         free(to_free);
 377         return 1;
 378 }
 379
 380 struct filter_params {
 381         const char *src;
 382         unsigned long size;
 383         int fd;
 384         const char *cmd;
 385         const char *path;
 386 };
 387
 388 static int filter_buffer_or_fd(int in, int out, void *data)
 389 {
 390         /*
 391          * Spawn cmd and feed the buffer contents through its stdin.
 392          */
 393         struct child_process child_process = CHILD_PROCESS_INIT;
 394         struct filter_params *params = (struct filter_params *)data;
 395         int write_err, status;
 396         const char *argv[] = { NULL, NULL };
 397
 398         /* apply % substitution to cmd */
 399         struct strbuf cmd = STRBUF_INIT;
 400         struct strbuf path = STRBUF_INIT;
 401         struct strbuf_expand_dict_entry dict[] = {
 402                 { "f", NULL, },
 403                 { NULL, NULL, },
 404         };
 405
 406         /* quote the path to preserve spaces, etc. */
 407         sq_quote_buf(&path, params->path);
 408         dict[0].value = path.buf;
 409
 410         /* expand all %f with the quoted path */
 411         strbuf_expand(&cmd, params->cmd, strbuf_expand_dict_cb, &dict);
 412         strbuf_release(&path);
 413
 414         argv[0] = cmd.buf;
 415
 416         child_process.argv = argv;
 417         child_process.use_shell = 1;
 418         child_process.in = -1;
 419         child_process.out = out;
 420
 421         if (start_command(&child_process))
 422                 return error("cannot fork to run external filter '%s'", params->cmd);
 423
 424         sigchain_push(SIGPIPE, SIG_IGN);
 425
 426         if (params->src) {
 427                 write_err = (write_in_full(child_process.in,
 428                                            params->src, params->size) < 0);
 429                 if (errno == EPIPE)
 430                         write_err = 0;
 431         } else {
 432                 write_err = copy_fd(params->fd, child_process.in);
 433                 if (write_err == COPY_WRITE_ERROR && errno == EPIPE)
 434                         write_err = 0;
 435         }
 436
 437         if (close(child_process.in))
 438                 write_err = 1;
 439         if (write_err)
 440                 error("cannot feed the input to external filter '%s'", params->cmd);
 441
 442         sigchain_pop(SIGPIPE);
 443
 444         status = finish_command(&child_process);
 445         if (status)
 446                 error("external filter '%s' failed %d", params->cmd, status);
 447
 448         strbuf_release(&cmd);
 449         return (write_err || status);
 450 }
 451
 452 static int apply_single_file_filter(const char *path, const char *src, size_t len, int fd,
 453                         struct strbuf *dst, const char *cmd)
 454 {
 455         /*
 456          * Create a pipeline to have the command filter the buffer's
 457          * contents.
 458          *
 459          * (child --> cmd) --> us
 460          */
 461         int err = 0;
 462         struct strbuf nbuf = STRBUF_INIT;
 463         struct async async;
 464         struct filter_params params;
 465
 466         memset(&async, 0, sizeof(async));
 467         async.proc = filter_buffer_or_fd;
 468         async.data = &params;
 469         async.out = -1;
 470         params.src = src;
 471         params.size = len;
 472         params.fd = fd;
 473         params.cmd = cmd;
 474         params.path = path;
 475
 476         fflush(NULL);
 477         if (start_async(&async))
 478                 return 0;       /* error was already reported */
 479
 480         if (strbuf_read(&nbuf, async.out, len) < 0) {
 481                 err = error("read from external filter '%s' failed", cmd);
 482         }
 483         if (close(async.out)) {
 484                 err = error("read from external filter '%s' failed", cmd);
 485         }
 486         if (finish_async(&async)) {
 487                 err = error("external filter '%s' failed", cmd);
 488         }
 489
 490         if (!err) {
 491                 strbuf_swap(dst, &nbuf);
 492         }
 493         strbuf_release(&nbuf);
 494         return !err;
 495 }
 496
 497 #define CAP_CLEAN    (1u<<0)
 498 #define CAP_SMUDGE   (1u<<1)
 499 #define CAP_DELAY    (1u<<2)
 500
 501 struct cmd2process {
 502         struct subprocess_entry subprocess; /* must be the first member! */
 503         unsigned int supported_capabilities;
 504 };
 505
 506 static int subprocess_map_initialized;
 507 static struct hashmap subprocess_map;
 508
 509 static int start_multi_file_filter_fn(struct subprocess_entry *subprocess)
 510 {
 511         int err, i;
 512         struct cmd2process *entry = (struct cmd2process *)subprocess;
 513         struct string_list cap_list = STRING_LIST_INIT_NODUP;
 514         char *cap_buf;
 515         const char *cap_name;
 516         struct child_process *process = &subprocess->process;
 517         const char *cmd = subprocess->cmd;
 518
 519         static const struct {
 520                 const char *name;
 521                 unsigned int cap;
 522         } known_caps[] = {
 523                 { "clean",  CAP_CLEAN  },
 524                 { "smudge", CAP_SMUDGE },
 525                 { "delay",  CAP_DELAY  },
 526         };
 527
 528         sigchain_push(SIGPIPE, SIG_IGN);
 529
 530         err = packet_writel(process->in, "git-filter-client", "version=2", NULL);
 531         if (err)
 532                 goto done;
 533
 534         err = strcmp(packet_read_line(process->out, NULL), "git-filter-server");
 535         if (err) {
 536                 error("external filter '%s' does not support filter protocol version 2", cmd);
 537                 goto done;
 538         }
 539         err = strcmp(packet_read_line(process->out, NULL), "version=2");
 540         if (err)
 541                 goto done;
 542         err = packet_read_line(process->out, NULL) != NULL;
 543         if (err)
 544                 goto done;
 545
 546         for (i = 0; i < ARRAY_SIZE(known_caps); ++i) {
 547                 err = packet_write_fmt_gently(
 548                         process->in, "capability=%s\n", known_caps[i].name);
 549                 if (err)
 550                         goto done;
 551         }
 552         err = packet_flush_gently(process->in);
 553         if (err)
 554                 goto done;
 555
 556         for (;;) {
 557                 cap_buf = packet_read_line(process->out, NULL);
 558                 if (!cap_buf)
 559                         break;
 560                 string_list_split_in_place(&cap_list, cap_buf, '=', 1);
 561
 562                 if (cap_list.nr != 2 || strcmp(cap_list.items[0].string, "capability"))
 563                         continue;
 564
 565                 cap_name = cap_list.items[1].string;
 566                 i = ARRAY_SIZE(known_caps) - 1;
 567                 while (i >= 0 && strcmp(cap_name, known_caps[i].name))
 568                         i--;
 569
 570                 if (i >= 0)
 571                         entry->supported_capabilities |= known_caps[i].cap;
 572                 else
 573                         warning("external filter '%s' requested unsupported filter capability '%s'",
 574                         cmd, cap_name);
 575
 576                 string_list_clear(&cap_list, 0);
 577         }
 578
 579 done:
 580         sigchain_pop(SIGPIPE);
 581
 582         return err;
 583 }
 584
 585 static void handle_filter_error(const struct strbuf *filter_status,
 586                                 struct cmd2process *entry,
 587                                 const unsigned int wanted_capability) {
 588         if (!strcmp(filter_status->buf, "error"))
 589                 ; /* The filter signaled a problem with the file. */
 590         else if (!strcmp(filter_status->buf, "abort") && wanted_capability) {
 591                 /*
 592                  * The filter signaled a permanent problem. Don't try to filter
 593                  * files with the same command for the lifetime of the current
 594                  * Git process.
 595                  */
 596                  entry->supported_capabilities &= ~wanted_capability;
 597         } else {
 598                 /*
 599                  * Something went wrong with the protocol filter.
 600                  * Force shutdown and restart if another blob requires filtering.
 601                  */
 602                 error("external filter '%s' failed", entry->subprocess.cmd);
 603                 subprocess_stop(&subprocess_map, &entry->subprocess);
 604                 free(entry);
 605         }
 606 }
 607
 608 static int apply_multi_file_filter(const char *path, const char *src, size_t len,
 609                                    int fd, struct strbuf *dst, const char *cmd,
 610                                    const unsigned int wanted_capability,
 611                                    struct delayed_checkout *dco)
 612 {
 613         int err;
 614         int can_delay = 0;
 615         struct cmd2process *entry;
 616         struct child_process *process;
 617         struct strbuf nbuf = STRBUF_INIT;
 618         struct strbuf filter_status = STRBUF_INIT;
 619         const char *filter_type;
 620
 621         if (!subprocess_map_initialized) {
 622                 subprocess_map_initialized = 1;
 623                 hashmap_init(&subprocess_map, (hashmap_cmp_fn) cmd2process_cmp, 0);
 624                 entry = NULL;
 625         } else {
 626                 entry = (struct cmd2process *)subprocess_find_entry(&subprocess_map, cmd);
 627         }
 628
 629         fflush(NULL);
 630
 631         if (!entry) {
 632                 entry = xmalloc(sizeof(*entry));
 633                 entry->supported_capabilities = 0;
 634
 635                 if (subprocess_start(&subprocess_map, &entry->subprocess, cmd, start_multi_file_filter_fn)) {
 636                         free(entry);
 637                         return 0;
 638                 }
 639         }
 640         process = &entry->subprocess.process;
 641
 642         if (!(entry->supported_capabilities & wanted_capability))
 643                 return 0;
 644
 645         if (wanted_capability & CAP_CLEAN)
 646                 filter_type = "clean";
 647         else if (wanted_capability & CAP_SMUDGE)
 648                 filter_type = "smudge";
 649         else
 650                 die("unexpected filter type");
 651
 652         sigchain_push(SIGPIPE, SIG_IGN);
 653
 654         assert(strlen(filter_type) < LARGE_PACKET_DATA_MAX - strlen("command=\n"));
 655         err = packet_write_fmt_gently(process->in, "command=%s\n", filter_type);
 656         if (err)
 657                 goto done;
 658
 659         err = strlen(path) > LARGE_PACKET_DATA_MAX - strlen("pathname=\n");
 660         if (err) {
 661                 error("path name too long for external filter");
 662                 goto done;
 663         }
 664
 665         err = packet_write_fmt_gently(process->in, "pathname=%s\n", path);
 666         if (err)
 667                 goto done;
 668
 669         if ((entry->supported_capabilities & CAP_DELAY) &&
 670             dco && dco->state == CE_CAN_DELAY) {
 671                 can_delay = 1;
 672                 err = packet_write_fmt_gently(process->in, "can-delay=1\n");
 673                 if (err)
 674                         goto done;
 675         }
 676
 677         err = packet_flush_gently(process->in);
 678         if (err)
 679                 goto done;
 680
 681         if (fd >= 0)
 682                 err = write_packetized_from_fd(fd, process->in);
 683         else
 684                 err = write_packetized_from_buf(src, len, process->in);
 685         if (err)
 686                 goto done;
 687
 688         err = subprocess_read_status(process->out, &filter_status);
 689         if (err)
 690                 goto done;
 691
 692         if (can_delay && !strcmp(filter_status.buf, "delayed")) {
 693                 string_list_insert(&dco->filters, cmd);
 694                 string_list_insert(&dco->paths, path);
 695         } else {
 696                 /* The filter got the blob and wants to send us a response. */
 697                 err = strcmp(filter_status.buf, "success");
 698                 if (err)
 699                         goto done;
 700
 701                 err = read_packetized_to_strbuf(process->out, &nbuf) < 0;
 702                 if (err)
 703                         goto done;
 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
 712 done:
 713         sigchain_pop(SIGPIPE);
 714
 715         if (err)
 716                 handle_filter_error(&filter_status, entry, wanted_capability);
 717         else
 718                 strbuf_swap(dst, &nbuf);
 719         strbuf_release(&nbuf);
 720         return !err;
 721 }
 722
 723
 724 int async_query_available_blobs(const char *cmd, struct string_list *available_paths)
 725 {
 726         int err;
 727         char *line;
 728         struct cmd2process *entry;
 729         struct child_process *process;
 730         struct strbuf filter_status = STRBUF_INIT;
 731
 732         assert(subprocess_map_initialized);
 733         entry = (struct cmd2process *)subprocess_find_entry(&subprocess_map, cmd);
 734         if (!entry) {
 735                 error("external filter '%s' is not available anymore although "
 736                       "not all paths have been filtered", cmd);
 737                 return 0;
 738         }
 739         process = &entry->subprocess.process;
 740         sigchain_push(SIGPIPE, SIG_IGN);
 741
 742         err = packet_write_fmt_gently(
 743                 process->in, "command=list_available_blobs\n");
 744         if (err)
 745                 goto done;
 746
 747         err = packet_flush_gently(process->in);
 748         if (err)
 749                 goto done;
 750
 751         while ((line = packet_read_line(process->out, NULL))) {
 752                 const char *path;
 753                 if (skip_prefix(line, "pathname=", &path))
 754                         string_list_insert(available_paths, xstrdup(path));
 755                 else
 756                         ; /* ignore unknown keys */
 757         }
 758
 759         err = subprocess_read_status(process->out, &filter_status);
 760         if (err)
 761                 goto done;
 762
 763         err = strcmp(filter_status.buf, "success");
 764
 765 done:
 766         sigchain_pop(SIGPIPE);
 767
 768         if (err)
 769                 handle_filter_error(&filter_status, entry, 0);
 770         return !err;
 771 }
 772
 773 static struct convert_driver {
 774         const char *name;
 775         struct convert_driver *next;
 776         const char *smudge;
 777         const char *clean;
 778         const char *process;
 779         int required;
 780 } *user_convert, **user_convert_tail;
 781
 782 static int apply_filter(const char *path, const char *src, size_t len,
 783                         int fd, struct strbuf *dst, struct convert_driver *drv,
 784                         const unsigned int wanted_capability,
 785                         struct delayed_checkout *dco)
 786 {
 787         const char *cmd = NULL;
 788
 789         if (!drv)
 790                 return 0;
 791
 792         if (!dst)
 793                 return 1;
 794
 795         if ((wanted_capability & CAP_CLEAN) && !drv->process && drv->clean)
 796                 cmd = drv->clean;
 797         else if ((wanted_capability & CAP_SMUDGE) && !drv->process && drv->smudge)
 798                 cmd = drv->smudge;
 799
 800         if (cmd && *cmd)
 801                 return apply_single_file_filter(path, src, len, fd, dst, cmd);
 802         else if (drv->process && *drv->process)
 803                 return apply_multi_file_filter(path, src, len, fd, dst,
 804                         drv->process, wanted_capability, dco);
 805
 806         return 0;
 807 }
 808
 809 static int read_convert_config(const char *var, const char *value, void *cb)
 810 {
 811         const char *key, *name;
 812         int namelen;
 813         struct convert_driver *drv;
 814
 815         /*
 816          * External conversion drivers are configured using
 817          * "filter.<name>.variable".
 818          */
 819         if (parse_config_key(var, "filter", &name, &namelen, &key) < 0 || !name)
 820                 return 0;
 821         for (drv = user_convert; drv; drv = drv->next)
 822                 if (!strncmp(drv->name, name, namelen) && !drv->name[namelen])
 823                         break;
 824         if (!drv) {
 825                 drv = xcalloc(1, sizeof(struct convert_driver));
 826                 drv->name = xmemdupz(name, namelen);
 827                 *user_convert_tail = drv;
 828                 user_convert_tail = &(drv->next);
 829         }
 830
 831         /*
 832          * filter.<name>.smudge and filter.<name>.clean specifies
 833          * the command line:
 834          *
 835          *      command-line
 836          *
 837          * The command-line will not be interpolated in any way.
 838          */
 839
 840         if (!strcmp("smudge", key))
 841                 return git_config_string(&drv->smudge, var, value);
 842
 843         if (!strcmp("clean", key))
 844                 return git_config_string(&drv->clean, var, value);
 845
 846         if (!strcmp("process", key))
 847                 return git_config_string(&drv->process, var, value);
 848
 849         if (!strcmp("required", key)) {
 850                 drv->required = git_config_bool(var, value);
 851                 return 0;
 852         }
 853
 854         return 0;
 855 }
 856
 857 static int count_ident(const char *cp, unsigned long size)
 858 {
 859         /*
 860          * "$Id: 0000000000000000000000000000000000000000 $" <=> "$Id$"
 861          */
 862         int cnt = 0;
 863         char ch;
 864
 865         while (size) {
 866                 ch = *cp++;
 867                 size--;
 868                 if (ch != '$')
 869                         continue;
 870                 if (size < 3)
 871                         break;
 872                 if (memcmp("Id", cp, 2))
 873                         continue;
 874                 ch = cp[2];
 875                 cp += 3;
 876                 size -= 3;
 877                 if (ch == '$')
 878                         cnt++; /* $Id$ */
 879                 if (ch != ':')
 880                         continue;
 881
 882                 /*
 883                  * "$Id: ... "; scan up to the closing dollar sign and discard.
 884                  */
 885                 while (size) {
 886                         ch = *cp++;
 887                         size--;
 888                         if (ch == '$') {
 889                                 cnt++;
 890                                 break;
 891                         }
 892                         if (ch == '\n')
 893                                 break;
 894                 }
 895         }
 896         return cnt;
 897 }
 898
 899 static int ident_to_git(const char *path, const char *src, size_t len,
 900                         struct strbuf *buf, int ident)
 901 {
 902         char *dst, *dollar;
 903
 904         if (!ident || (src && !count_ident(src, len)))
 905                 return 0;
 906
 907         if (!buf)
 908                 return 1;
 909
 910         /* only grow if not in place */
 911         if (strbuf_avail(buf) + buf->len < len)
 912                 strbuf_grow(buf, len - buf->len);
 913         dst = buf->buf;
 914         for (;;) {
 915                 dollar = memchr(src, '$', len);
 916                 if (!dollar)
 917                         break;
 918                 memmove(dst, src, dollar + 1 - src);
 919                 dst += dollar + 1 - src;
 920                 len -= dollar + 1 - src;
 921                 src  = dollar + 1;
 922
 923                 if (len > 3 && !memcmp(src, "Id:", 3)) {
 924                         dollar = memchr(src + 3, '$', len - 3);
 925                         if (!dollar)
 926                                 break;
 927                         if (memchr(src + 3, '\n', dollar - src - 3)) {
 928                                 /* Line break before the next dollar. */
 929                                 continue;
 930                         }
 931
 932                         memcpy(dst, "Id$", 3);
 933                         dst += 3;
 934                         len -= dollar + 1 - src;
 935                         src  = dollar + 1;
 936                 }
 937         }
 938         memmove(dst, src, len);
 939         strbuf_setlen(buf, dst + len - buf->buf);
 940         return 1;
 941 }
 942
 943 static int ident_to_worktree(const char *path, const char *src, size_t len,
 944                              struct strbuf *buf, int ident)
 945 {
 946         unsigned char sha1[20];
 947         char *to_free = NULL, *dollar, *spc;
 948         int cnt;
 949
 950         if (!ident)
 951                 return 0;
 952
 953         cnt = count_ident(src, len);
 954         if (!cnt)
 955                 return 0;
 956
 957         /* are we "faking" in place editing ? */
 958         if (src == buf->buf)
 959                 to_free = strbuf_detach(buf, NULL);
 960         hash_sha1_file(src, len, "blob", sha1);
 961
 962         strbuf_grow(buf, len + cnt * 43);
 963         for (;;) {
 964                 /* step 1: run to the next '$' */
 965                 dollar = memchr(src, '$', len);
 966                 if (!dollar)
 967                         break;
 968                 strbuf_add(buf, src, dollar + 1 - src);
 969                 len -= dollar + 1 - src;
 970                 src  = dollar + 1;
 971
 972                 /* step 2: does it looks like a bit like Id:xxx$ or Id$ ? */
 973                 if (len < 3 || memcmp("Id", src, 2))
 974                         continue;
 975
 976                 /* step 3: skip over Id$ or Id:xxxxx$ */
 977                 if (src[2] == '$') {
 978                         src += 3;
 979                         len -= 3;
 980                 } else if (src[2] == ':') {
 981                         /*
 982                          * It's possible that an expanded Id has crept its way into the
 983                          * repository, we cope with that by stripping the expansion out.
 984                          * This is probably not a good idea, since it will cause changes
 985                          * on checkout, which won't go away by stash, but let's keep it
 986                          * for git-style ids.
 987                          */
 988                         dollar = memchr(src + 3, '$', len - 3);
 989                         if (!dollar) {
 990                                 /* incomplete keyword, no more '$', so just quit the loop */
 991                                 break;
 992                         }
 993
 994                         if (memchr(src + 3, '\n', dollar - src - 3)) {
 995                                 /* Line break before the next dollar. */
 996                                 continue;
 997                         }
 998
 999                         spc = memchr(src + 4, ' ', dollar - src - 4);
1000                         if (spc && spc < dollar-1) {
1001                                 /* There are spaces in unexpected places.
1002                                  * This is probably an id from some other
1003                                  * versioning system. Keep it for now.
1004                                  */
1005                                 continue;
1006                         }
1007
1008                         len -= dollar + 1 - src;
1009                         src  = dollar + 1;
1010                 } else {
1011                         /* it wasn't a "Id$" or "Id:xxxx$" */
1012                         continue;
1013                 }
1014
1015                 /* step 4: substitute */
1016                 strbuf_addstr(buf, "Id: ");
1017                 strbuf_add(buf, sha1_to_hex(sha1), 40);
1018                 strbuf_addstr(buf, " $");
1019         }
1020         strbuf_add(buf, src, len);
1021
1022         free(to_free);
1023         return 1;
1024 }
1025
1026 static enum crlf_action git_path_check_crlf(struct attr_check_item *check)
1027 {
1028         const char *value = check->value;
1029
1030         if (ATTR_TRUE(value))
1031                 return CRLF_TEXT;
1032         else if (ATTR_FALSE(value))
1033                 return CRLF_BINARY;
1034         else if (ATTR_UNSET(value))
1035                 ;
1036         else if (!strcmp(value, "input"))
1037                 return CRLF_TEXT_INPUT;
1038         else if (!strcmp(value, "auto"))
1039                 return CRLF_AUTO;
1040         return CRLF_UNDEFINED;
1041 }
1042
1043 static enum eol git_path_check_eol(struct attr_check_item *check)
1044 {
1045         const char *value = check->value;
1046
1047         if (ATTR_UNSET(value))
1048                 ;
1049         else if (!strcmp(value, "lf"))
1050                 return EOL_LF;
1051         else if (!strcmp(value, "crlf"))
1052                 return EOL_CRLF;
1053         return EOL_UNSET;
1054 }
1055
1056 static struct convert_driver *git_path_check_convert(struct attr_check_item *check)
1057 {
1058         const char *value = check->value;
1059         struct convert_driver *drv;
1060
1061         if (ATTR_TRUE(value) || ATTR_FALSE(value) || ATTR_UNSET(value))
1062                 return NULL;
1063         for (drv = user_convert; drv; drv = drv->next)
1064                 if (!strcmp(value, drv->name))
1065                         return drv;
1066         return NULL;
1067 }
1068
1069 static int git_path_check_ident(struct attr_check_item *check)
1070 {
1071         const char *value = check->value;
1072
1073         return !!ATTR_TRUE(value);
1074 }
1075
1076 struct conv_attrs {
1077         struct convert_driver *drv;
1078         enum crlf_action attr_action; /* What attr says */
1079         enum crlf_action crlf_action; /* When no attr is set, use core.autocrlf */
1080         int ident;
1081 };
1082
1083 static void convert_attrs(struct conv_attrs *ca, const char *path)
1084 {
1085         static struct attr_check *check;
1086
1087         if (!check) {
1088                 check = attr_check_initl("crlf", "ident", "filter",
1089                                          "eol", "text", NULL);
1090                 user_convert_tail = &user_convert;
1091                 git_config(read_convert_config, NULL);
1092         }
1093
1094         if (!git_check_attr(path, check)) {
1095                 struct attr_check_item *ccheck = check->items;
1096                 ca->crlf_action = git_path_check_crlf(ccheck + 4);
1097                 if (ca->crlf_action == CRLF_UNDEFINED)
1098                         ca->crlf_action = git_path_check_crlf(ccheck + 0);
1099                 ca->attr_action = ca->crlf_action;
1100                 ca->ident = git_path_check_ident(ccheck + 1);
1101                 ca->drv = git_path_check_convert(ccheck + 2);
1102                 if (ca->crlf_action != CRLF_BINARY) {
1103                         enum eol eol_attr = git_path_check_eol(ccheck + 3);
1104                         if (ca->crlf_action == CRLF_AUTO && eol_attr == EOL_LF)
1105                                 ca->crlf_action = CRLF_AUTO_INPUT;
1106                         else if (ca->crlf_action == CRLF_AUTO && eol_attr == EOL_CRLF)
1107                                 ca->crlf_action = CRLF_AUTO_CRLF;
1108                         else if (eol_attr == EOL_LF)
1109                                 ca->crlf_action = CRLF_TEXT_INPUT;
1110                         else if (eol_attr == EOL_CRLF)
1111                                 ca->crlf_action = CRLF_TEXT_CRLF;
1112                 }
1113                 ca->attr_action = ca->crlf_action;
1114         } else {
1115                 ca->drv = NULL;
1116                 ca->crlf_action = CRLF_UNDEFINED;
1117                 ca->ident = 0;
1118         }
1119         if (ca->crlf_action == CRLF_TEXT)
1120                 ca->crlf_action = text_eol_is_crlf() ? CRLF_TEXT_CRLF : CRLF_TEXT_INPUT;
1121         if (ca->crlf_action == CRLF_UNDEFINED && auto_crlf == AUTO_CRLF_FALSE)
1122                 ca->crlf_action = CRLF_BINARY;
1123         if (ca->crlf_action == CRLF_UNDEFINED && auto_crlf == AUTO_CRLF_TRUE)
1124                 ca->crlf_action = CRLF_AUTO_CRLF;
1125         if (ca->crlf_action == CRLF_UNDEFINED && auto_crlf == AUTO_CRLF_INPUT)
1126                 ca->crlf_action = CRLF_AUTO_INPUT;
1127 }
1128
1129 int would_convert_to_git_filter_fd(const char *path)
1130 {
1131         struct conv_attrs ca;
1132
1133         convert_attrs(&ca, path);
1134         if (!ca.drv)
1135                 return 0;
1136
1137         /*
1138          * Apply a filter to an fd only if the filter is required to succeed.
1139          * We must die if the filter fails, because the original data before
1140          * filtering is not available.
1141          */
1142         if (!ca.drv->required)
1143                 return 0;
1144
1145         return apply_filter(path, NULL, 0, -1, NULL, ca.drv, CAP_CLEAN, NULL);
1146 }
1147
1148 const char *get_convert_attr_ascii(const char *path)
1149 {
1150         struct conv_attrs ca;
1151
1152         convert_attrs(&ca, path);
1153         switch (ca.attr_action) {
1154         case CRLF_UNDEFINED:
1155                 return "";
1156         case CRLF_BINARY:
1157                 return "-text";
1158         case CRLF_TEXT:
1159                 return "text";
1160         case CRLF_TEXT_INPUT:
1161                 return "text eol=lf";
1162         case CRLF_TEXT_CRLF:
1163                 return "text eol=crlf";
1164         case CRLF_AUTO:
1165                 return "text=auto";
1166         case CRLF_AUTO_CRLF:
1167                 return "text=auto eol=crlf";
1168         case CRLF_AUTO_INPUT:
1169                 return "text=auto eol=lf";
1170         }
1171         return "";
1172 }
1173
1174 int convert_to_git(const char *path, const char *src, size_t len,
1175                    struct strbuf *dst, enum safe_crlf checksafe)
1176 {
1177         int ret = 0;
1178         struct conv_attrs ca;
1179
1180         convert_attrs(&ca, path);
1181
1182         ret |= apply_filter(path, src, len, -1, dst, ca.drv, CAP_CLEAN, NULL);
1183         if (!ret && ca.drv && ca.drv->required)
1184                 die("%s: clean filter '%s' failed", path, ca.drv->name);
1185
1186         if (ret && dst) {
1187                 src = dst->buf;
1188                 len = dst->len;
1189         }
1190         ret |= crlf_to_git(path, src, len, dst, ca.crlf_action, checksafe);
1191         if (ret && dst) {
1192                 src = dst->buf;
1193                 len = dst->len;
1194         }
1195         return ret | ident_to_git(path, src, len, dst, ca.ident);
1196 }
1197
1198 void convert_to_git_filter_fd(const char *path, int fd, struct strbuf *dst,
1199                               enum safe_crlf checksafe)
1200 {
1201         struct conv_attrs ca;
1202         convert_attrs(&ca, path);
1203
1204         assert(ca.drv);
1205         assert(ca.drv->clean || ca.drv->process);
1206
1207         if (!apply_filter(path, NULL, 0, fd, dst, ca.drv, CAP_CLEAN, NULL))
1208                 die("%s: clean filter '%s' failed", path, ca.drv->name);
1209
1210         crlf_to_git(path, dst->buf, dst->len, dst, ca.crlf_action, checksafe);
1211         ident_to_git(path, dst->buf, dst->len, dst, ca.ident);
1212 }
1213
1214 static int convert_to_working_tree_internal(const char *path, const char *src,
1215                                             size_t len, struct strbuf *dst,
1216                                             int normalizing, struct delayed_checkout *dco)
1217 {
1218         int ret = 0, ret_filter = 0;
1219         struct conv_attrs ca;
1220
1221         convert_attrs(&ca, path);
1222
1223         ret |= ident_to_worktree(path, src, len, dst, ca.ident);
1224         if (ret) {
1225                 src = dst->buf;
1226                 len = dst->len;
1227         }
1228         /*
1229          * CRLF conversion can be skipped if normalizing, unless there
1230          * is a smudge or process filter (even if the process filter doesn't
1231          * support smudge).  The filters might expect CRLFs.
1232          */
1233         if ((ca.drv && (ca.drv->smudge || ca.drv->process)) || !normalizing) {
1234                 ret |= crlf_to_worktree(path, src, len, dst, ca.crlf_action);
1235                 if (ret) {
1236                         src = dst->buf;
1237                         len = dst->len;
1238                 }
1239         }
1240
1241         ret_filter = apply_filter(
1242                 path, src, len, -1, dst, ca.drv, CAP_SMUDGE, dco);
1243         if (!ret_filter && ca.drv && ca.drv->required)
1244                 die("%s: smudge filter %s failed", path, ca.drv->name);
1245
1246         return ret | ret_filter;
1247 }
1248
1249 int async_convert_to_working_tree(const char *path, const char *src,
1250                                   size_t len, struct strbuf *dst,
1251                                   void *dco)
1252 {
1253         return convert_to_working_tree_internal(path, src, len, dst, 0, dco);
1254 }
1255
1256 int convert_to_working_tree(const char *path, const char *src, size_t len, struct strbuf *dst)
1257 {
1258         return convert_to_working_tree_internal(path, src, len, dst, 0, NULL);
1259 }
1260
1261 int renormalize_buffer(const char *path, const char *src, size_t len, struct strbuf *dst)
1262 {
1263         int ret = convert_to_working_tree_internal(path, src, len, dst, 1, NULL);
1264         if (ret) {
1265                 src = dst->buf;
1266                 len = dst->len;
1267         }
1268         return ret | convert_to_git(path, src, len, dst, SAFE_CRLF_RENORMALIZE);
1269 }
1270
1271 /*****************************************************************
1272  *
1273  * Streaming conversion support
1274  *
1275  *****************************************************************/
1276
1277 typedef int (*filter_fn)(struct stream_filter *,
1278                          const char *input, size_t *isize_p,
1279                          char *output, size_t *osize_p);
1280 typedef void (*free_fn)(struct stream_filter *);
1281
1282 struct stream_filter_vtbl {
1283         filter_fn filter;
1284         free_fn free;
1285 };
1286
1287 struct stream_filter {
1288         struct stream_filter_vtbl *vtbl;
1289 };
1290
1291 static int null_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;
1296
1297         if (!input)
1298                 return 0; /* we do not keep any states */
1299         count = *isize_p;
1300         if (*osize_p < count)
1301                 count = *osize_p;
1302         if (count) {
1303                 memmove(output, input, count);
1304                 *isize_p -= count;
1305                 *osize_p -= count;
1306         }
1307         return 0;
1308 }
1309
1310 static void null_free_fn(struct stream_filter *filter)
1311 {
1312         ; /* nothing -- null instances are shared */
1313 }
1314
1315 static struct stream_filter_vtbl null_vtbl = {
1316         null_filter_fn,
1317         null_free_fn,
1318 };
1319
1320 static struct stream_filter null_filter_singleton = {
1321         &null_vtbl,
1322 };
1323
1324 int is_null_stream_filter(struct stream_filter *filter)
1325 {
1326         return filter == &null_filter_singleton;
1327 }
1328
1329
1330 /*
1331  * LF-to-CRLF filter
1332  */
1333
1334 struct lf_to_crlf_filter {
1335         struct stream_filter filter;
1336         unsigned has_held:1;
1337         char held;
1338 };
1339
1340 static int lf_to_crlf_filter_fn(struct stream_filter *filter,
1341                                 const char *input, size_t *isize_p,
1342                                 char *output, size_t *osize_p)
1343 {
1344         size_t count, o = 0;
1345         struct lf_to_crlf_filter *lf_to_crlf = (struct lf_to_crlf_filter *)filter;
1346
1347         /*
1348          * We may be holding onto the CR to see if it is followed by a
1349          * LF, in which case we would need to go to the main loop.
1350          * Otherwise, just emit it to the output stream.
1351          */
1352         if (lf_to_crlf->has_held && (lf_to_crlf->held != '\r' || !input)) {
1353                 output[o++] = lf_to_crlf->held;
1354                 lf_to_crlf->has_held = 0;
1355         }
1356
1357         /* We are told to drain */
1358         if (!input) {
1359                 *osize_p -= o;
1360                 return 0;
1361         }
1362
1363         count = *isize_p;
1364         if (count || lf_to_crlf->has_held) {
1365                 size_t i;
1366                 int was_cr = 0;
1367
1368                 if (lf_to_crlf->has_held) {
1369                         was_cr = 1;
1370                         lf_to_crlf->has_held = 0;
1371                 }
1372
1373                 for (i = 0; o < *osize_p && i < count; i++) {
1374                         char ch = input[i];
1375
1376                         if (ch == '\n') {
1377                                 output[o++] = '\r';
1378                         } else if (was_cr) {
1379                                 /*
1380                                  * Previous round saw CR and it is not followed
1381                                  * by a LF; emit the CR before processing the
1382                                  * current character.
1383                                  */
1384                                 output[o++] = '\r';
1385                         }
1386
1387                         /*
1388                          * We may have consumed the last output slot,
1389                          * in which case we need to break out of this
1390                          * loop; hold the current character before
1391                          * returning.
1392                          */
1393                         if (*osize_p <= o) {
1394                                 lf_to_crlf->has_held = 1;
1395                                 lf_to_crlf->held = ch;
1396                                 continue; /* break but increment i */
1397                         }
1398
1399                         if (ch == '\r') {
1400                                 was_cr = 1;
1401                                 continue;
1402                         }
1403
1404                         was_cr = 0;
1405                         output[o++] = ch;
1406                 }
1407
1408                 *osize_p -= o;
1409                 *isize_p -= i;
1410
1411                 if (!lf_to_crlf->has_held && was_cr) {
1412                         lf_to_crlf->has_held = 1;
1413                         lf_to_crlf->held = '\r';
1414                 }
1415         }
1416         return 0;
1417 }
1418
1419 static void lf_to_crlf_free_fn(struct stream_filter *filter)
1420 {
1421         free(filter);
1422 }
1423
1424 static struct stream_filter_vtbl lf_to_crlf_vtbl = {
1425         lf_to_crlf_filter_fn,
1426         lf_to_crlf_free_fn,
1427 };
1428
1429 static struct stream_filter *lf_to_crlf_filter(void)
1430 {
1431         struct lf_to_crlf_filter *lf_to_crlf = xcalloc(1, sizeof(*lf_to_crlf));
1432
1433         lf_to_crlf->filter.vtbl = &lf_to_crlf_vtbl;
1434         return (struct stream_filter *)lf_to_crlf;
1435 }
1436
1437 /*
1438  * Cascade filter
1439  */
1440 #define FILTER_BUFFER 1024
1441 struct cascade_filter {
1442         struct stream_filter filter;
1443         struct stream_filter *one;
1444         struct stream_filter *two;
1445         char buf[FILTER_BUFFER];
1446         int end, ptr;
1447 };
1448
1449 static int cascade_filter_fn(struct stream_filter *filter,
1450                              const char *input, size_t *isize_p,
1451                              char *output, size_t *osize_p)
1452 {
1453         struct cascade_filter *cas = (struct cascade_filter *) filter;
1454         size_t filled = 0;
1455         size_t sz = *osize_p;
1456         size_t to_feed, remaining;
1457
1458         /*
1459          * input -- (one) --> buf -- (two) --> output
1460          */
1461         while (filled < sz) {
1462                 remaining = sz - filled;
1463
1464                 /* do we already have something to feed two with? */
1465                 if (cas->ptr < cas->end) {
1466                         to_feed = cas->end - cas->ptr;
1467                         if (stream_filter(cas->two,
1468                                           cas->buf + cas->ptr, &to_feed,
1469                                           output + filled, &remaining))
1470                                 return -1;
1471                         cas->ptr += (cas->end - cas->ptr) - to_feed;
1472                         filled = sz - remaining;
1473                         continue;
1474                 }
1475
1476                 /* feed one from upstream and have it emit into our buffer */
1477                 to_feed = input ? *isize_p : 0;
1478                 if (input && !to_feed)
1479                         break;
1480                 remaining = sizeof(cas->buf);
1481                 if (stream_filter(cas->one,
1482                                   input, &to_feed,
1483                                   cas->buf, &remaining))
1484                         return -1;
1485                 cas->end = sizeof(cas->buf) - remaining;
1486                 cas->ptr = 0;
1487                 if (input) {
1488                         size_t fed = *isize_p - to_feed;
1489                         *isize_p -= fed;
1490                         input += fed;
1491                 }
1492
1493                 /* do we know that we drained one completely? */
1494                 if (input || cas->end)
1495                         continue;
1496
1497                 /* tell two to drain; we have nothing more to give it */
1498                 to_feed = 0;
1499                 remaining = sz - filled;
1500                 if (stream_filter(cas->two,
1501                                   NULL, &to_feed,
1502                                   output + filled, &remaining))
1503                         return -1;
1504                 if (remaining == (sz - filled))
1505                         break; /* completely drained two */
1506                 filled = sz - remaining;
1507         }
1508         *osize_p -= filled;
1509         return 0;
1510 }
1511
1512 static void cascade_free_fn(struct stream_filter *filter)
1513 {
1514         struct cascade_filter *cas = (struct cascade_filter *)filter;
1515         free_stream_filter(cas->one);
1516         free_stream_filter(cas->two);
1517         free(filter);
1518 }
1519
1520 static struct stream_filter_vtbl cascade_vtbl = {
1521         cascade_filter_fn,
1522         cascade_free_fn,
1523 };
1524
1525 static struct stream_filter *cascade_filter(struct stream_filter *one,
1526                                             struct stream_filter *two)
1527 {
1528         struct cascade_filter *cascade;
1529
1530         if (!one || is_null_stream_filter(one))
1531                 return two;
1532         if (!two || is_null_stream_filter(two))
1533                 return one;
1534
1535         cascade = xmalloc(sizeof(*cascade));
1536         cascade->one = one;
1537         cascade->two = two;
1538         cascade->end = cascade->ptr = 0;
1539         cascade->filter.vtbl = &cascade_vtbl;
1540         return (struct stream_filter *)cascade;
1541 }
1542
1543 /*
1544  * ident filter
1545  */
1546 #define IDENT_DRAINING (-1)
1547 #define IDENT_SKIPPING (-2)
1548 struct ident_filter {
1549         struct stream_filter filter;
1550         struct strbuf left;
1551         int state;
1552         char ident[45]; /* ": x40 $" */
1553 };
1554
1555 static int is_foreign_ident(const char *str)
1556 {
1557         int i;
1558
1559         if (!skip_prefix(str, "$Id: ", &str))
1560                 return 0;
1561         for (i = 0; str[i]; i++) {
1562                 if (isspace(str[i]) && str[i+1] != '$')
1563                         return 1;
1564         }
1565         return 0;
1566 }
1567
1568 static void ident_drain(struct ident_filter *ident, char **output_p, size_t *osize_p)
1569 {
1570         size_t to_drain = ident->left.len;
1571
1572         if (*osize_p < to_drain)
1573                 to_drain = *osize_p;
1574         if (to_drain) {
1575                 memcpy(*output_p, ident->left.buf, to_drain);
1576                 strbuf_remove(&ident->left, 0, to_drain);
1577                 *output_p += to_drain;
1578                 *osize_p -= to_drain;
1579         }
1580         if (!ident->left.len)
1581                 ident->state = 0;
1582 }
1583
1584 static int ident_filter_fn(struct stream_filter *filter,
1585                            const char *input, size_t *isize_p,
1586                            char *output, size_t *osize_p)
1587 {
1588         struct ident_filter *ident = (struct ident_filter *)filter;
1589         static const char head[] = "$Id";
1590
1591         if (!input) {
1592                 /* drain upon eof */
1593                 switch (ident->state) {
1594                 default:
1595                         strbuf_add(&ident->left, head, ident->state);
1596                 case IDENT_SKIPPING:
1597                         /* fallthru */
1598                 case IDENT_DRAINING:
1599                         ident_drain(ident, &output, osize_p);
1600                 }
1601                 return 0;
1602         }
1603
1604         while (*isize_p || (ident->state == IDENT_DRAINING)) {
1605                 int ch;
1606
1607                 if (ident->state == IDENT_DRAINING) {
1608                         ident_drain(ident, &output, osize_p);
1609                         if (!*osize_p)
1610                                 break;
1611                         continue;
1612                 }
1613
1614                 ch = *(input++);
1615                 (*isize_p)--;
1616
1617                 if (ident->state == IDENT_SKIPPING) {
1618                         /*
1619                          * Skipping until '$' or LF, but keeping them
1620                          * in case it is a foreign ident.
1621                          */
1622                         strbuf_addch(&ident->left, ch);
1623                         if (ch != '\n' && ch != '$')
1624                                 continue;
1625                         if (ch == '$' && !is_foreign_ident(ident->left.buf)) {
1626                                 strbuf_setlen(&ident->left, sizeof(head) - 1);
1627                                 strbuf_addstr(&ident->left, ident->ident);
1628                         }
1629                         ident->state = IDENT_DRAINING;
1630                         continue;
1631                 }
1632
1633                 if (ident->state < sizeof(head) &&
1634                     head[ident->state] == ch) {
1635                         ident->state++;
1636                         continue;
1637                 }
1638
1639                 if (ident->state)
1640                         strbuf_add(&ident->left, head, ident->state);
1641                 if (ident->state == sizeof(head) - 1) {
1642                         if (ch != ':' && ch != '$') {
1643                                 strbuf_addch(&ident->left, ch);
1644                                 ident->state = 0;
1645                                 continue;
1646                         }
1647
1648                         if (ch == ':') {
1649                                 strbuf_addch(&ident->left, ch);
1650                                 ident->state = IDENT_SKIPPING;
1651                         } else {
1652                                 strbuf_addstr(&ident->left, ident->ident);
1653                                 ident->state = IDENT_DRAINING;
1654                         }
1655                         continue;
1656                 }
1657
1658                 strbuf_addch(&ident->left, ch);
1659                 ident->state = IDENT_DRAINING;
1660         }
1661         return 0;
1662 }
1663
1664 static void ident_free_fn(struct stream_filter *filter)
1665 {
1666         struct ident_filter *ident = (struct ident_filter *)filter;
1667         strbuf_release(&ident->left);
1668         free(filter);
1669 }
1670
1671 static struct stream_filter_vtbl ident_vtbl = {
1672         ident_filter_fn,
1673         ident_free_fn,
1674 };
1675
1676 static struct stream_filter *ident_filter(const unsigned char *sha1)
1677 {
1678         struct ident_filter *ident = xmalloc(sizeof(*ident));
1679
1680         xsnprintf(ident->ident, sizeof(ident->ident),
1681                   ": %s $", sha1_to_hex(sha1));
1682         strbuf_init(&ident->left, 0);
1683         ident->filter.vtbl = &ident_vtbl;
1684         ident->state = 0;
1685         return (struct stream_filter *)ident;
1686 }
1687
1688 /*
1689  * Return an appropriately constructed filter for the path, or NULL if
1690  * the contents cannot be filtered without reading the whole thing
1691  * in-core.
1692  *
1693  * Note that you would be crazy to set CRLF, smuge/clean or ident to a
1694  * large binary blob you would want us not to slurp into the memory!
1695  */
1696 struct stream_filter *get_stream_filter(const char *path, const unsigned char *sha1)
1697 {
1698         struct conv_attrs ca;
1699         struct stream_filter *filter = NULL;
1700
1701         convert_attrs(&ca, path);
1702         if (ca.drv && (ca.drv->process || ca.drv->smudge || ca.drv->clean))
1703                 return NULL;
1704
1705         if (ca.crlf_action == CRLF_AUTO || ca.crlf_action == CRLF_AUTO_CRLF)
1706                 return NULL;
1707
1708         if (ca.ident)
1709                 filter = ident_filter(sha1);
1710
1711         if (output_eol(ca.crlf_action) == EOL_CRLF)
1712                 filter = cascade_filter(filter, lf_to_crlf_filter());
1713         else
1714                 filter = cascade_filter(filter, &null_filter_singleton);
1715
1716         return filter;
1717 }
1718
1719 void free_stream_filter(struct stream_filter *filter)
1720 {
1721         filter->vtbl->free(filter);
1722 }
1723
1724 int stream_filter(struct stream_filter *filter,
1725                   const char *input, size_t *isize_p,
1726                   char *output, size_t *osize_p)
1727 {
1728         return filter->vtbl->filter(filter, input, isize_p, output, osize_p);
1729 }