convert.c

   1 #include "cache.h"
   2 #include "config.h"
   3 #include "object-store.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 #include "utf8.h"
  11 #include "ll-merge.h"
  12
  13 /*
  14  * convert.c - convert a file when checking it out and checking it in.
  15  *
  16  * This should use the pathname to decide on whether it wants to do some
  17  * more interesting conversions (automatic gzip/unzip, general format
  18  * conversions etc etc), but by default it just does automatic CRLF<->LF
  19  * translation when the "text" attribute or "auto_crlf" option is set.
  20  */
  21
  22 /* Stat bits: When BIN is set, the txt bits are unset */
  23 #define CONVERT_STAT_BITS_TXT_LF    0x1
  24 #define CONVERT_STAT_BITS_TXT_CRLF  0x2
  25 #define CONVERT_STAT_BITS_BIN       0x4
  26
  27 enum crlf_action {
  28         CRLF_UNDEFINED,
  29         CRLF_BINARY,
  30         CRLF_TEXT,
  31         CRLF_TEXT_INPUT,
  32         CRLF_TEXT_CRLF,
  33         CRLF_AUTO,
  34         CRLF_AUTO_INPUT,
  35         CRLF_AUTO_CRLF
  36 };
  37
  38 struct text_stat {
  39         /* NUL, CR, LF and CRLF counts */
  40         unsigned nul, lonecr, lonelf, crlf;
  41
  42         /* These are just approximations! */
  43         unsigned printable, nonprintable;
  44 };
  45
  46 static void gather_stats(const char *buf, unsigned long size, struct text_stat *stats)
  47 {
  48         unsigned long i;
  49
  50         memset(stats, 0, sizeof(*stats));
  51
  52         for (i = 0; i < size; i++) {
  53                 unsigned char c = buf[i];
  54                 if (c == '\r') {
  55                         if (i+1 < size && buf[i+1] == '\n') {
  56                                 stats->crlf++;
  57                                 i++;
  58                         } else
  59                                 stats->lonecr++;
  60                         continue;
  61                 }
  62                 if (c == '\n') {
  63                         stats->lonelf++;
  64                         continue;
  65                 }
  66                 if (c == 127)
  67                         /* DEL */
  68                         stats->nonprintable++;
  69                 else if (c < 32) {
  70                         switch (c) {
  71                                 /* BS, HT, ESC and FF */
  72                         case '\b': case '\t': case '\033': case '\014':
  73                                 stats->printable++;
  74                                 break;
  75                         case 0:
  76                                 stats->nul++;
  77                                 /* fall through */
  78                         default:
  79                                 stats->nonprintable++;
  80                         }
  81                 }
  82                 else
  83                         stats->printable++;
  84         }
  85
  86         /* If file ends with EOF then don't count this EOF as non-printable. */
  87         if (size >= 1 && buf[size-1] == '\032')
  88                 stats->nonprintable--;
  89 }
  90
  91 /*
  92  * The same heuristics as diff.c::mmfile_is_binary()
  93  * We treat files with bare CR as binary
  94  */
  95 static int convert_is_binary(const struct text_stat *stats)
  96 {
  97         if (stats->lonecr)
  98                 return 1;
  99         if (stats->nul)
 100                 return 1;
 101         if ((stats->printable >> 7) < stats->nonprintable)
 102                 return 1;
 103         return 0;
 104 }
 105
 106 static unsigned int gather_convert_stats(const char *data, unsigned long size)
 107 {
 108         struct text_stat stats;
 109         int ret = 0;
 110         if (!data || !size)
 111                 return 0;
 112         gather_stats(data, size, &stats);
 113         if (convert_is_binary(&stats))
 114                 ret |= CONVERT_STAT_BITS_BIN;
 115         if (stats.crlf)
 116                 ret |= CONVERT_STAT_BITS_TXT_CRLF;
 117         if (stats.lonelf)
 118                 ret |=  CONVERT_STAT_BITS_TXT_LF;
 119
 120         return ret;
 121 }
 122
 123 static const char *gather_convert_stats_ascii(const char *data, unsigned long size)
 124 {
 125         unsigned int convert_stats = gather_convert_stats(data, size);
 126
 127         if (convert_stats & CONVERT_STAT_BITS_BIN)
 128                 return "-text";
 129         switch (convert_stats) {
 130         case CONVERT_STAT_BITS_TXT_LF:
 131                 return "lf";
 132         case CONVERT_STAT_BITS_TXT_CRLF:
 133                 return "crlf";
 134         case CONVERT_STAT_BITS_TXT_LF | CONVERT_STAT_BITS_TXT_CRLF:
 135                 return "mixed";
 136         default:
 137                 return "none";
 138         }
 139 }
 140
 141 const char *get_cached_convert_stats_ascii(const struct index_state *istate,
 142                                            const char *path)
 143 {
 144         const char *ret;
 145         unsigned long sz;
 146         void *data = read_blob_data_from_index(istate, path, &sz);
 147         ret = gather_convert_stats_ascii(data, sz);
 148         free(data);
 149         return ret;
 150 }
 151
 152 const char *get_wt_convert_stats_ascii(const char *path)
 153 {
 154         const char *ret = "";
 155         struct strbuf sb = STRBUF_INIT;
 156         if (strbuf_read_file(&sb, path, 0) >= 0)
 157                 ret = gather_convert_stats_ascii(sb.buf, sb.len);
 158         strbuf_release(&sb);
 159         return ret;
 160 }
 161
 162 static int text_eol_is_crlf(void)
 163 {
 164         if (auto_crlf == AUTO_CRLF_TRUE)
 165                 return 1;
 166         else if (auto_crlf == AUTO_CRLF_INPUT)
 167                 return 0;
 168         if (core_eol == EOL_CRLF)
 169                 return 1;
 170         if (core_eol == EOL_UNSET && EOL_NATIVE == EOL_CRLF)
 171                 return 1;
 172         return 0;
 173 }
 174
 175 static enum eol output_eol(enum crlf_action crlf_action)
 176 {
 177         switch (crlf_action) {
 178         case CRLF_BINARY:
 179                 return EOL_UNSET;
 180         case CRLF_TEXT_CRLF:
 181                 return EOL_CRLF;
 182         case CRLF_TEXT_INPUT:
 183                 return EOL_LF;
 184         case CRLF_UNDEFINED:
 185         case CRLF_AUTO_CRLF:
 186                 return EOL_CRLF;
 187         case CRLF_AUTO_INPUT:
 188                 return EOL_LF;
 189         case CRLF_TEXT:
 190         case CRLF_AUTO:
 191                 /* fall through */
 192                 return text_eol_is_crlf() ? EOL_CRLF : EOL_LF;
 193         }
 194         warning(_("illegal crlf_action %d"), (int)crlf_action);
 195         return core_eol;
 196 }
 197
 198 static void check_global_conv_flags_eol(const char *path, enum crlf_action crlf_action,
 199                             struct text_stat *old_stats, struct text_stat *new_stats,
 200                             int conv_flags)
 201 {
 202         if (old_stats->crlf && !new_stats->crlf ) {
 203                 /*
 204                  * CRLFs would not be restored by checkout
 205                  */
 206                 if (conv_flags & CONV_EOL_RNDTRP_DIE)
 207                         die(_("CRLF would be replaced by LF in %s"), path);
 208                 else if (conv_flags & CONV_EOL_RNDTRP_WARN)
 209                         warning(_("CRLF will be replaced by LF in %s.\n"
 210                                   "The file will have its original line"
 211                                   " endings in your working directory"), path);
 212         } else if (old_stats->lonelf && !new_stats->lonelf ) {
 213                 /*
 214                  * CRLFs would be added by checkout
 215                  */
 216                 if (conv_flags & CONV_EOL_RNDTRP_DIE)
 217                         die(_("LF would be replaced by CRLF in %s"), path);
 218                 else if (conv_flags & CONV_EOL_RNDTRP_WARN)
 219                         warning(_("LF will be replaced by CRLF in %s.\n"
 220                                   "The file will have its original line"
 221                                   " endings in your working directory"), path);
 222         }
 223 }
 224
 225 static int has_crlf_in_index(const struct index_state *istate, const char *path)
 226 {
 227         unsigned long sz;
 228         void *data;
 229         const char *crp;
 230         int has_crlf = 0;
 231
 232         data = read_blob_data_from_index(istate, path, &sz);
 233         if (!data)
 234                 return 0;
 235
 236         crp = memchr(data, '\r', sz);
 237         if (crp) {
 238                 unsigned int ret_stats;
 239                 ret_stats = gather_convert_stats(data, sz);
 240                 if (!(ret_stats & CONVERT_STAT_BITS_BIN) &&
 241                     (ret_stats & CONVERT_STAT_BITS_TXT_CRLF))
 242                         has_crlf = 1;
 243         }
 244         free(data);
 245         return has_crlf;
 246 }
 247
 248 static int will_convert_lf_to_crlf(struct text_stat *stats,
 249                                    enum crlf_action crlf_action)
 250 {
 251         if (output_eol(crlf_action) != EOL_CRLF)
 252                 return 0;
 253         /* No "naked" LF? Nothing to convert, regardless. */
 254         if (!stats->lonelf)
 255                 return 0;
 256
 257         if (crlf_action == CRLF_AUTO || crlf_action == CRLF_AUTO_INPUT || crlf_action == CRLF_AUTO_CRLF) {
 258                 /* If we have any CR or CRLF line endings, we do not touch it */
 259                 /* This is the new safer autocrlf-handling */
 260                 if (stats->lonecr || stats->crlf)
 261                         return 0;
 262
 263                 if (convert_is_binary(stats))
 264                         return 0;
 265         }
 266         return 1;
 267
 268 }
 269
 270 static int validate_encoding(const char *path, const char *enc,
 271                       const char *data, size_t len, int die_on_error)
 272 {
 273         const char *stripped;
 274
 275         /* We only check for UTF here as UTF?? can be an alias for UTF-?? */
 276         if (skip_iprefix(enc, "UTF", &stripped)) {
 277                 skip_prefix(stripped, "-", &stripped);
 278
 279                 /*
 280                  * Check for detectable errors in UTF encodings
 281                  */
 282                 if (has_prohibited_utf_bom(enc, data, len)) {
 283                         const char *error_msg = _(
 284                                 "BOM is prohibited in '%s' if encoded as %s");
 285                         /*
 286                          * This advice is shown for UTF-??BE and UTF-??LE encodings.
 287                          * We cut off the last two characters of the encoding name
 288                          * to generate the encoding name suitable for BOMs.
 289                          */
 290                         const char *advise_msg = _(
 291                                 "The file '%s' contains a byte order "
 292                                 "mark (BOM). Please use UTF-%.*s as "
 293                                 "working-tree-encoding.");
 294                         int stripped_len = strlen(stripped) - strlen("BE");
 295                         advise(advise_msg, path, stripped_len, stripped);
 296                         if (die_on_error)
 297                                 die(error_msg, path, enc);
 298                         else {
 299                                 return error(error_msg, path, enc);
 300                         }
 301
 302                 } else if (is_missing_required_utf_bom(enc, data, len)) {
 303                         const char *error_msg = _(
 304                                 "BOM is required in '%s' if encoded as %s");
 305                         const char *advise_msg = _(
 306                                 "The file '%s' is missing a byte order "
 307                                 "mark (BOM). Please use UTF-%sBE or UTF-%sLE "
 308                                 "(depending on the byte order) as "
 309                                 "working-tree-encoding.");
 310                         advise(advise_msg, path, stripped, stripped);
 311                         if (die_on_error)
 312                                 die(error_msg, path, enc);
 313                         else {
 314                                 return error(error_msg, path, enc);
 315                         }
 316                 }
 317
 318         }
 319         return 0;
 320 }
 321
 322 static void trace_encoding(const char *context, const char *path,
 323                            const char *encoding, const char *buf, size_t len)
 324 {
 325         static struct trace_key coe = TRACE_KEY_INIT(WORKING_TREE_ENCODING);
 326         struct strbuf trace = STRBUF_INIT;
 327         int i;
 328
 329         strbuf_addf(&trace, "%s (%s, considered %s):\n", context, path, encoding);
 330         for (i = 0; i < len && buf; ++i) {
 331                 strbuf_addf(
 332                         &trace, "| \033[2m%2i:\033[0m %2x \033[2m%c\033[0m%c",
 333                         i,
 334                         (unsigned char) buf[i],
 335                         (buf[i] > 32 && buf[i] < 127 ? buf[i] : ' '),
 336                         ((i+1) % 8 && (i+1) < len ? ' ' : '\n')
 337                 );
 338         }
 339         strbuf_addchars(&trace, '\n', 1);
 340
 341         trace_strbuf(&coe, &trace);
 342         strbuf_release(&trace);
 343 }
 344
 345 static int check_roundtrip(const char *enc_name)
 346 {
 347         /*
 348          * check_roundtrip_encoding contains a string of comma and/or
 349          * space separated encodings (eg. "UTF-16, ASCII, CP1125").
 350          * Search for the given encoding in that string.
 351          */
 352         const char *found = strcasestr(check_roundtrip_encoding, enc_name);
 353         const char *next;
 354         int len;
 355         if (!found)
 356                 return 0;
 357         next = found + strlen(enc_name);
 358         len = strlen(check_roundtrip_encoding);
 359         return (found && (
 360                         /*
 361                          * check that the found encoding is at the
 362                          * beginning of check_roundtrip_encoding or
 363                          * that it is prefixed with a space or comma
 364                          */
 365                         found == check_roundtrip_encoding || (
 366                                 (isspace(found[-1]) || found[-1] == ',')
 367                         )
 368                 ) && (
 369                         /*
 370                          * check that the found encoding is at the
 371                          * end of check_roundtrip_encoding or
 372                          * that it is suffixed with a space or comma
 373                          */
 374                         next == check_roundtrip_encoding + len || (
 375                                 next < check_roundtrip_encoding + len &&
 376                                 (isspace(next[0]) || next[0] == ',')
 377                         )
 378                 ));
 379 }
 380
 381 static const char *default_encoding = "UTF-8";
 382
 383 static int encode_to_git(const char *path, const char *src, size_t src_len,
 384                          struct strbuf *buf, const char *enc, int conv_flags)
 385 {
 386         char *dst;
 387         size_t dst_len;
 388         int die_on_error = conv_flags & CONV_WRITE_OBJECT;
 389
 390         /*
 391          * No encoding is specified or there is nothing to encode.
 392          * Tell the caller that the content was not modified.
 393          */
 394         if (!enc || (src && !src_len))
 395                 return 0;
 396
 397         /*
 398          * Looks like we got called from "would_convert_to_git()".
 399          * This means Git wants to know if it would encode (= modify!)
 400          * the content. Let's answer with "yes", since an encoding was
 401          * specified.
 402          */
 403         if (!buf && !src)
 404                 return 1;
 405
 406         if (validate_encoding(path, enc, src, src_len, die_on_error))
 407                 return 0;
 408
 409         trace_encoding("source", path, enc, src, src_len);
 410         dst = reencode_string_len(src, src_len, default_encoding, enc,
 411                                   &dst_len);
 412         if (!dst) {
 413                 /*
 414                  * We could add the blob "as-is" to Git. However, on checkout
 415                  * we would try to re-encode to the original encoding. This
 416                  * would fail and we would leave the user with a messed-up
 417                  * working tree. Let's try to avoid this by screaming loud.
 418                  */
 419                 const char* msg = _("failed to encode '%s' from %s to %s");
 420                 if (die_on_error)
 421                         die(msg, path, enc, default_encoding);
 422                 else {
 423                         error(msg, path, enc, default_encoding);
 424                         return 0;
 425                 }
 426         }
 427         trace_encoding("destination", path, default_encoding, dst, dst_len);
 428
 429         /*
 430          * UTF supports lossless conversion round tripping [1] and conversions
 431          * between UTF and other encodings are mostly round trip safe as
 432          * Unicode aims to be a superset of all other character encodings.
 433          * However, certain encodings (e.g. SHIFT-JIS) are known to have round
 434          * trip issues [2]. Check the round trip conversion for all encodings
 435          * listed in core.checkRoundtripEncoding.
 436          *
 437          * The round trip check is only performed if content is written to Git.
 438          * This ensures that no information is lost during conversion to/from
 439          * the internal UTF-8 representation.
 440          *
 441          * Please note, the code below is not tested because I was not able to
 442          * generate a faulty round trip without an iconv error. Iconv errors
 443          * are already caught above.
 444          *
 445          * [1] http://unicode.org/faq/utf_bom.html#gen2
 446          * [2] https://support.microsoft.com/en-us/help/170559/prb-conversion-problem-between-shift-jis-and-unicode
 447          */
 448         if (die_on_error && check_roundtrip(enc)) {
 449                 char *re_src;
 450                 size_t re_src_len;
 451
 452                 re_src = reencode_string_len(dst, dst_len,
 453                                              enc, default_encoding,
 454                                              &re_src_len);
 455
 456                 trace_printf("Checking roundtrip encoding for %s...\n", enc);
 457                 trace_encoding("reencoded source", path, enc,
 458                                re_src, re_src_len);
 459
 460                 if (!re_src || src_len != re_src_len ||
 461                     memcmp(src, re_src, src_len)) {
 462                         const char* msg = _("encoding '%s' from %s to %s and "
 463                                             "back is not the same");
 464                         die(msg, path, enc, default_encoding);
 465                 }
 466
 467                 free(re_src);
 468         }
 469
 470         strbuf_attach(buf, dst, dst_len, dst_len + 1);
 471         return 1;
 472 }
 473
 474 static int encode_to_worktree(const char *path, const char *src, size_t src_len,
 475                               struct strbuf *buf, const char *enc)
 476 {
 477         char *dst;
 478         size_t dst_len;
 479
 480         /*
 481          * No encoding is specified or there is nothing to encode.
 482          * Tell the caller that the content was not modified.
 483          */
 484         if (!enc || (src && !src_len))
 485                 return 0;
 486
 487         dst = reencode_string_len(src, src_len, enc, default_encoding,
 488                                   &dst_len);
 489         if (!dst) {
 490                 error(_("failed to encode '%s' from %s to %s"),
 491                       path, default_encoding, enc);
 492                 return 0;
 493         }
 494
 495         strbuf_attach(buf, dst, dst_len, dst_len + 1);
 496         return 1;
 497 }
 498
 499 static int crlf_to_git(const struct index_state *istate,
 500                        const char *path, const char *src, size_t len,
 501                        struct strbuf *buf,
 502                        enum crlf_action crlf_action, int conv_flags)
 503 {
 504         struct text_stat stats;
 505         char *dst;
 506         int convert_crlf_into_lf;
 507
 508         if (crlf_action == CRLF_BINARY ||
 509             (src && !len))
 510                 return 0;
 511
 512         /*
 513          * If we are doing a dry-run and have no source buffer, there is
 514          * nothing to analyze; we must assume we would convert.
 515          */
 516         if (!buf && !src)
 517                 return 1;
 518
 519         gather_stats(src, len, &stats);
 520         /* Optimization: No CRLF? Nothing to convert, regardless. */
 521         convert_crlf_into_lf = !!stats.crlf;
 522
 523         if (crlf_action == CRLF_AUTO || crlf_action == CRLF_AUTO_INPUT || crlf_action == CRLF_AUTO_CRLF) {
 524                 if (convert_is_binary(&stats))
 525                         return 0;
 526                 /*
 527                  * If the file in the index has any CR in it, do not
 528                  * convert.  This is the new safer autocrlf handling,
 529                  * unless we want to renormalize in a merge or
 530                  * cherry-pick.
 531                  */
 532                 if ((!(conv_flags & CONV_EOL_RENORMALIZE)) &&
 533                     has_crlf_in_index(istate, path))
 534                         convert_crlf_into_lf = 0;
 535         }
 536         if (((conv_flags & CONV_EOL_RNDTRP_WARN) ||
 537              ((conv_flags & CONV_EOL_RNDTRP_DIE) && len))) {
 538                 struct text_stat new_stats;
 539                 memcpy(&new_stats, &stats, sizeof(new_stats));
 540                 /* simulate "git add" */
 541                 if (convert_crlf_into_lf) {
 542                         new_stats.lonelf += new_stats.crlf;
 543                         new_stats.crlf = 0;
 544                 }
 545                 /* simulate "git checkout" */
 546                 if (will_convert_lf_to_crlf(&new_stats, crlf_action)) {
 547                         new_stats.crlf += new_stats.lonelf;
 548                         new_stats.lonelf = 0;
 549                 }
 550                 check_global_conv_flags_eol(path, crlf_action, &stats, &new_stats, conv_flags);
 551         }
 552         if (!convert_crlf_into_lf)
 553                 return 0;
 554
 555         /*
 556          * At this point all of our source analysis is done, and we are sure we
 557          * would convert. If we are in dry-run mode, we can give an answer.
 558          */
 559         if (!buf)
 560                 return 1;
 561
 562         /* only grow if not in place */
 563         if (strbuf_avail(buf) + buf->len < len)
 564                 strbuf_grow(buf, len - buf->len);
 565         dst = buf->buf;
 566         if (crlf_action == CRLF_AUTO || crlf_action == CRLF_AUTO_INPUT || crlf_action == CRLF_AUTO_CRLF) {
 567                 /*
 568                  * If we guessed, we already know we rejected a file with
 569                  * lone CR, and we can strip a CR without looking at what
 570                  * follow it.
 571                  */
 572                 do {
 573                         unsigned char c = *src++;
 574                         if (c != '\r')
 575                                 *dst++ = c;
 576                 } while (--len);
 577         } else {
 578                 do {
 579                         unsigned char c = *src++;
 580                         if (! (c == '\r' && (1 < len && *src == '\n')))
 581                                 *dst++ = c;
 582                 } while (--len);
 583         }
 584         strbuf_setlen(buf, dst - buf->buf);
 585         return 1;
 586 }
 587
 588 static int crlf_to_worktree(const char *src, size_t len,
 589                             struct strbuf *buf, enum crlf_action crlf_action)
 590 {
 591         char *to_free = NULL;
 592         struct text_stat stats;
 593
 594         if (!len || output_eol(crlf_action) != EOL_CRLF)
 595                 return 0;
 596
 597         gather_stats(src, len, &stats);
 598         if (!will_convert_lf_to_crlf(&stats, crlf_action))
 599                 return 0;
 600
 601         /* are we "faking" in place editing ? */
 602         if (src == buf->buf)
 603                 to_free = strbuf_detach(buf, NULL);
 604
 605         strbuf_grow(buf, len + stats.lonelf);
 606         for (;;) {
 607                 const char *nl = memchr(src, '\n', len);
 608                 if (!nl)
 609                         break;
 610                 if (nl > src && nl[-1] == '\r') {
 611                         strbuf_add(buf, src, nl + 1 - src);
 612                 } else {
 613                         strbuf_add(buf, src, nl - src);
 614                         strbuf_addstr(buf, "\r\n");
 615                 }
 616                 len -= nl + 1 - src;
 617                 src  = nl + 1;
 618         }
 619         strbuf_add(buf, src, len);
 620
 621         free(to_free);
 622         return 1;
 623 }
 624
 625 struct filter_params {
 626         const char *src;
 627         unsigned long size;
 628         int fd;
 629         const char *cmd;
 630         const char *path;
 631 };
 632
 633 static int filter_buffer_or_fd(int in, int out, void *data)
 634 {
 635         /*
 636          * Spawn cmd and feed the buffer contents through its stdin.
 637          */
 638         struct child_process child_process = CHILD_PROCESS_INIT;
 639         struct filter_params *params = (struct filter_params *)data;
 640         int write_err, status;
 641         const char *argv[] = { NULL, NULL };
 642
 643         /* apply % substitution to cmd */
 644         struct strbuf cmd = STRBUF_INIT;
 645         struct strbuf path = STRBUF_INIT;
 646         struct strbuf_expand_dict_entry dict[] = {
 647                 { "f", NULL, },
 648                 { NULL, NULL, },
 649         };
 650
 651         /* quote the path to preserve spaces, etc. */
 652         sq_quote_buf(&path, params->path);
 653         dict[0].value = path.buf;
 654
 655         /* expand all %f with the quoted path */
 656         strbuf_expand(&cmd, params->cmd, strbuf_expand_dict_cb, &dict);
 657         strbuf_release(&path);
 658
 659         argv[0] = cmd.buf;
 660
 661         child_process.argv = argv;
 662         child_process.use_shell = 1;
 663         child_process.in = -1;
 664         child_process.out = out;
 665
 666         if (start_command(&child_process)) {
 667                 strbuf_release(&cmd);
 668                 return error(_("cannot fork to run external filter '%s'"),
 669                              params->cmd);
 670         }
 671
 672         sigchain_push(SIGPIPE, SIG_IGN);
 673
 674         if (params->src) {
 675                 write_err = (write_in_full(child_process.in,
 676                                            params->src, params->size) < 0);
 677                 if (errno == EPIPE)
 678                         write_err = 0;
 679         } else {
 680                 write_err = copy_fd(params->fd, child_process.in);
 681                 if (write_err == COPY_WRITE_ERROR && errno == EPIPE)
 682                         write_err = 0;
 683         }
 684
 685         if (close(child_process.in))
 686                 write_err = 1;
 687         if (write_err)
 688                 error(_("cannot feed the input to external filter '%s'"),
 689                       params->cmd);
 690
 691         sigchain_pop(SIGPIPE);
 692
 693         status = finish_command(&child_process);
 694         if (status)
 695                 error(_("external filter '%s' failed %d"), params->cmd, status);
 696
 697         strbuf_release(&cmd);
 698         return (write_err || status);
 699 }
 700
 701 static int apply_single_file_filter(const char *path, const char *src, size_t len, int fd,
 702                                     struct strbuf *dst, const char *cmd)
 703 {
 704         /*
 705          * Create a pipeline to have the command filter the buffer's
 706          * contents.
 707          *
 708          * (child --> cmd) --> us
 709          */
 710         int err = 0;
 711         struct strbuf nbuf = STRBUF_INIT;
 712         struct async async;
 713         struct filter_params params;
 714
 715         memset(&async, 0, sizeof(async));
 716         async.proc = filter_buffer_or_fd;
 717         async.data = &params;
 718         async.out = -1;
 719         params.src = src;
 720         params.size = len;
 721         params.fd = fd;
 722         params.cmd = cmd;
 723         params.path = path;
 724
 725         fflush(NULL);
 726         if (start_async(&async))
 727                 return 0;       /* error was already reported */
 728
 729         if (strbuf_read(&nbuf, async.out, 0) < 0) {
 730                 err = error(_("read from external filter '%s' failed"), cmd);
 731         }
 732         if (close(async.out)) {
 733                 err = error(_("read from external filter '%s' failed"), cmd);
 734         }
 735         if (finish_async(&async)) {
 736                 err = error(_("external filter '%s' failed"), cmd);
 737         }
 738
 739         if (!err) {
 740                 strbuf_swap(dst, &nbuf);
 741         }
 742         strbuf_release(&nbuf);
 743         return !err;
 744 }
 745
 746 #define CAP_CLEAN    (1u<<0)
 747 #define CAP_SMUDGE   (1u<<1)
 748 #define CAP_DELAY    (1u<<2)
 749
 750 struct cmd2process {
 751         struct subprocess_entry subprocess; /* must be the first member! */
 752         unsigned int supported_capabilities;
 753 };
 754
 755 static int subprocess_map_initialized;
 756 static struct hashmap subprocess_map;
 757
 758 static int start_multi_file_filter_fn(struct subprocess_entry *subprocess)
 759 {
 760         static int versions[] = {2, 0};
 761         static struct subprocess_capability capabilities[] = {
 762                 { "clean",  CAP_CLEAN  },
 763                 { "smudge", CAP_SMUDGE },
 764                 { "delay",  CAP_DELAY  },
 765                 { NULL, 0 }
 766         };
 767         struct cmd2process *entry = (struct cmd2process *)subprocess;
 768         return subprocess_handshake(subprocess, "git-filter", versions, NULL,
 769                                     capabilities,
 770                                     &entry->supported_capabilities);
 771 }
 772
 773 static void handle_filter_error(const struct strbuf *filter_status,
 774                                 struct cmd2process *entry,
 775                                 const unsigned int wanted_capability)
 776 {
 777         if (!strcmp(filter_status->buf, "error"))
 778                 ; /* The filter signaled a problem with the file. */
 779         else if (!strcmp(filter_status->buf, "abort") && wanted_capability) {
 780                 /*
 781                  * The filter signaled a permanent problem. Don't try to filter
 782                  * files with the same command for the lifetime of the current
 783                  * Git process.
 784                  */
 785                  entry->supported_capabilities &= ~wanted_capability;
 786         } else {
 787                 /*
 788                  * Something went wrong with the protocol filter.
 789                  * Force shutdown and restart if another blob requires filtering.
 790                  */
 791                 error(_("external filter '%s' failed"), entry->subprocess.cmd);
 792                 subprocess_stop(&subprocess_map, &entry->subprocess);
 793                 free(entry);
 794         }
 795 }
 796
 797 static int apply_multi_file_filter(const char *path, const char *src, size_t len,
 798                                    int fd, struct strbuf *dst, const char *cmd,
 799                                    const unsigned int wanted_capability,
 800                                    const struct checkout_metadata *meta,
 801                                    struct delayed_checkout *dco)
 802 {
 803         int err;
 804         int can_delay = 0;
 805         struct cmd2process *entry;
 806         struct child_process *process;
 807         struct strbuf nbuf = STRBUF_INIT;
 808         struct strbuf filter_status = STRBUF_INIT;
 809         const char *filter_type;
 810
 811         if (!subprocess_map_initialized) {
 812                 subprocess_map_initialized = 1;
 813                 hashmap_init(&subprocess_map, cmd2process_cmp, NULL, 0);
 814                 entry = NULL;
 815         } else {
 816                 entry = (struct cmd2process *)subprocess_find_entry(&subprocess_map, cmd);
 817         }
 818
 819         fflush(NULL);
 820
 821         if (!entry) {
 822                 entry = xmalloc(sizeof(*entry));
 823                 entry->supported_capabilities = 0;
 824
 825                 if (subprocess_start(&subprocess_map, &entry->subprocess, cmd, start_multi_file_filter_fn)) {
 826                         free(entry);
 827                         return 0;
 828                 }
 829         }
 830         process = &entry->subprocess.process;
 831
 832         if (!(entry->supported_capabilities & wanted_capability))
 833                 return 0;
 834
 835         if (wanted_capability & CAP_CLEAN)
 836                 filter_type = "clean";
 837         else if (wanted_capability & CAP_SMUDGE)
 838                 filter_type = "smudge";
 839         else
 840                 die(_("unexpected filter type"));
 841
 842         sigchain_push(SIGPIPE, SIG_IGN);
 843
 844         assert(strlen(filter_type) < LARGE_PACKET_DATA_MAX - strlen("command=\n"));
 845         err = packet_write_fmt_gently(process->in, "command=%s\n", filter_type);
 846         if (err)
 847                 goto done;
 848
 849         err = strlen(path) > LARGE_PACKET_DATA_MAX - strlen("pathname=\n");
 850         if (err) {
 851                 error(_("path name too long for external filter"));
 852                 goto done;
 853         }
 854
 855         err = packet_write_fmt_gently(process->in, "pathname=%s\n", path);
 856         if (err)
 857                 goto done;
 858
 859         if (meta && meta->refname) {
 860                 err = packet_write_fmt_gently(process->in, "ref=%s\n", meta->refname);
 861                 if (err)
 862                         goto done;
 863         }
 864
 865         if (meta && !is_null_oid(&meta->treeish)) {
 866                 err = packet_write_fmt_gently(process->in, "treeish=%s\n", oid_to_hex(&meta->treeish));
 867                 if (err)
 868                         goto done;
 869         }
 870
 871         if (meta && !is_null_oid(&meta->blob)) {
 872                 err = packet_write_fmt_gently(process->in, "blob=%s\n", oid_to_hex(&meta->blob));
 873                 if (err)
 874                         goto done;
 875         }
 876
 877         if ((entry->supported_capabilities & CAP_DELAY) &&
 878             dco && dco->state == CE_CAN_DELAY) {
 879                 can_delay = 1;
 880                 err = packet_write_fmt_gently(process->in, "can-delay=1\n");
 881                 if (err)
 882                         goto done;
 883         }
 884
 885         err = packet_flush_gently(process->in);
 886         if (err)
 887                 goto done;
 888
 889         if (fd >= 0)
 890                 err = write_packetized_from_fd(fd, process->in);
 891         else
 892                 err = write_packetized_from_buf(src, len, process->in);
 893         if (err)
 894                 goto done;
 895
 896         err = subprocess_read_status(process->out, &filter_status);
 897         if (err)
 898                 goto done;
 899
 900         if (can_delay && !strcmp(filter_status.buf, "delayed")) {
 901                 string_list_insert(&dco->filters, cmd);
 902                 string_list_insert(&dco->paths, path);
 903         } else {
 904                 /* The filter got the blob and wants to send us a response. */
 905                 err = strcmp(filter_status.buf, "success");
 906                 if (err)
 907                         goto done;
 908
 909                 err = read_packetized_to_strbuf(process->out, &nbuf) < 0;
 910                 if (err)
 911                         goto done;
 912
 913                 err = subprocess_read_status(process->out, &filter_status);
 914                 if (err)
 915                         goto done;
 916
 917                 err = strcmp(filter_status.buf, "success");
 918         }
 919
 920 done:
 921         sigchain_pop(SIGPIPE);
 922
 923         if (err)
 924                 handle_filter_error(&filter_status, entry, wanted_capability);
 925         else
 926                 strbuf_swap(dst, &nbuf);
 927         strbuf_release(&nbuf);
 928         return !err;
 929 }
 930
 931
 932 int async_query_available_blobs(const char *cmd, struct string_list *available_paths)
 933 {
 934         int err;
 935         char *line;
 936         struct cmd2process *entry;
 937         struct child_process *process;
 938         struct strbuf filter_status = STRBUF_INIT;
 939
 940         assert(subprocess_map_initialized);
 941         entry = (struct cmd2process *)subprocess_find_entry(&subprocess_map, cmd);
 942         if (!entry) {
 943                 error(_("external filter '%s' is not available anymore although "
 944                         "not all paths have been filtered"), cmd);
 945                 return 0;
 946         }
 947         process = &entry->subprocess.process;
 948         sigchain_push(SIGPIPE, SIG_IGN);
 949
 950         err = packet_write_fmt_gently(
 951                 process->in, "command=list_available_blobs\n");
 952         if (err)
 953                 goto done;
 954
 955         err = packet_flush_gently(process->in);
 956         if (err)
 957                 goto done;
 958
 959         while ((line = packet_read_line(process->out, NULL))) {
 960                 const char *path;
 961                 if (skip_prefix(line, "pathname=", &path))
 962                         string_list_insert(available_paths, xstrdup(path));
 963                 else
 964                         ; /* ignore unknown keys */
 965         }
 966
 967         err = subprocess_read_status(process->out, &filter_status);
 968         if (err)
 969                 goto done;
 970
 971         err = strcmp(filter_status.buf, "success");
 972
 973 done:
 974         sigchain_pop(SIGPIPE);
 975
 976         if (err)
 977                 handle_filter_error(&filter_status, entry, 0);
 978         return !err;
 979 }
 980
 981 static struct convert_driver {
 982         const char *name;
 983         struct convert_driver *next;
 984         const char *smudge;
 985         const char *clean;
 986         const char *process;
 987         int required;
 988 } *user_convert, **user_convert_tail;
 989
 990 static int apply_filter(const char *path, const char *src, size_t len,
 991                         int fd, struct strbuf *dst, struct convert_driver *drv,
 992                         const unsigned int wanted_capability,
 993                         const struct checkout_metadata *meta,
 994                         struct delayed_checkout *dco)
 995 {
 996         const char *cmd = NULL;
 997
 998         if (!drv)
 999                 return 0;
1000
1001         if (!dst)
1002                 return 1;
1003
1004         if ((wanted_capability & CAP_CLEAN) && !drv->process && drv->clean)
1005                 cmd = drv->clean;
1006         else if ((wanted_capability & CAP_SMUDGE) && !drv->process && drv->smudge)
1007                 cmd = drv->smudge;
1008
1009         if (cmd && *cmd)
1010                 return apply_single_file_filter(path, src, len, fd, dst, cmd);
1011         else if (drv->process && *drv->process)
1012                 return apply_multi_file_filter(path, src, len, fd, dst,
1013                         drv->process, wanted_capability, meta, dco);
1014
1015         return 0;
1016 }
1017
1018 static int read_convert_config(const char *var, const char *value, void *cb)
1019 {
1020         const char *key, *name;
1021         size_t namelen;
1022         struct convert_driver *drv;
1023
1024         /*
1025          * External conversion drivers are configured using
1026          * "filter.<name>.variable".
1027          */
1028         if (parse_config_key(var, "filter", &name, &namelen, &key) < 0 || !name)
1029                 return 0;
1030         for (drv = user_convert; drv; drv = drv->next)
1031                 if (!strncmp(drv->name, name, namelen) && !drv->name[namelen])
1032                         break;
1033         if (!drv) {
1034                 drv = xcalloc(1, sizeof(struct convert_driver));
1035                 drv->name = xmemdupz(name, namelen);
1036                 *user_convert_tail = drv;
1037                 user_convert_tail = &(drv->next);
1038         }
1039
1040         /*
1041          * filter.<name>.smudge and filter.<name>.clean specifies
1042          * the command line:
1043          *
1044          *      command-line
1045          *
1046          * The command-line will not be interpolated in any way.
1047          */
1048
1049         if (!strcmp("smudge", key))
1050                 return git_config_string(&drv->smudge, var, value);
1051
1052         if (!strcmp("clean", key))
1053                 return git_config_string(&drv->clean, var, value);
1054
1055         if (!strcmp("process", key))
1056                 return git_config_string(&drv->process, var, value);
1057
1058         if (!strcmp("required", key)) {
1059                 drv->required = git_config_bool(var, value);
1060                 return 0;
1061         }
1062
1063         return 0;
1064 }
1065
1066 static int count_ident(const char *cp, unsigned long size)
1067 {
1068         /*
1069          * "$Id: 0000000000000000000000000000000000000000 $" <=> "$Id$"
1070          */
1071         int cnt = 0;
1072         char ch;
1073
1074         while (size) {
1075                 ch = *cp++;
1076                 size--;
1077                 if (ch != '$')
1078                         continue;
1079                 if (size < 3)
1080                         break;
1081                 if (memcmp("Id", cp, 2))
1082                         continue;
1083                 ch = cp[2];
1084                 cp += 3;
1085                 size -= 3;
1086                 if (ch == '$')
1087                         cnt++; /* $Id$ */
1088                 if (ch != ':')
1089                         continue;
1090
1091                 /*
1092                  * "$Id: ... "; scan up to the closing dollar sign and discard.
1093                  */
1094                 while (size) {
1095                         ch = *cp++;
1096                         size--;
1097                         if (ch == '$') {
1098                                 cnt++;
1099                                 break;
1100                         }
1101                         if (ch == '\n')
1102                                 break;
1103                 }
1104         }
1105         return cnt;
1106 }
1107
1108 static int ident_to_git(const char *src, size_t len,
1109                         struct strbuf *buf, int ident)
1110 {
1111         char *dst, *dollar;
1112
1113         if (!ident || (src && !count_ident(src, len)))
1114                 return 0;
1115
1116         if (!buf)
1117                 return 1;
1118
1119         /* only grow if not in place */
1120         if (strbuf_avail(buf) + buf->len < len)
1121                 strbuf_grow(buf, len - buf->len);
1122         dst = buf->buf;
1123         for (;;) {
1124                 dollar = memchr(src, '$', len);
1125                 if (!dollar)
1126                         break;
1127                 memmove(dst, src, dollar + 1 - src);
1128                 dst += dollar + 1 - src;
1129                 len -= dollar + 1 - src;
1130                 src  = dollar + 1;
1131
1132                 if (len > 3 && !memcmp(src, "Id:", 3)) {
1133                         dollar = memchr(src + 3, '$', len - 3);
1134                         if (!dollar)
1135                                 break;
1136                         if (memchr(src + 3, '\n', dollar - src - 3)) {
1137                                 /* Line break before the next dollar. */
1138                                 continue;
1139                         }
1140
1141                         memcpy(dst, "Id$", 3);
1142                         dst += 3;
1143                         len -= dollar + 1 - src;
1144                         src  = dollar + 1;
1145                 }
1146         }
1147         memmove(dst, src, len);
1148         strbuf_setlen(buf, dst + len - buf->buf);
1149         return 1;
1150 }
1151
1152 static int ident_to_worktree(const char *src, size_t len,
1153                              struct strbuf *buf, int ident)
1154 {
1155         struct object_id oid;
1156         char *to_free = NULL, *dollar, *spc;
1157         int cnt;
1158
1159         if (!ident)
1160                 return 0;
1161
1162         cnt = count_ident(src, len);
1163         if (!cnt)
1164                 return 0;
1165
1166         /* are we "faking" in place editing ? */
1167         if (src == buf->buf)
1168                 to_free = strbuf_detach(buf, NULL);
1169         hash_object_file(the_hash_algo, src, len, "blob", &oid);
1170
1171         strbuf_grow(buf, len + cnt * (the_hash_algo->hexsz + 3));
1172         for (;;) {
1173                 /* step 1: run to the next '$' */
1174                 dollar = memchr(src, '$', len);
1175                 if (!dollar)
1176                         break;
1177                 strbuf_add(buf, src, dollar + 1 - src);
1178                 len -= dollar + 1 - src;
1179                 src  = dollar + 1;
1180
1181                 /* step 2: does it looks like a bit like Id:xxx$ or Id$ ? */
1182                 if (len < 3 || memcmp("Id", src, 2))
1183                         continue;
1184
1185                 /* step 3: skip over Id$ or Id:xxxxx$ */
1186                 if (src[2] == '$') {
1187                         src += 3;
1188                         len -= 3;
1189                 } else if (src[2] == ':') {
1190                         /*
1191                          * It's possible that an expanded Id has crept its way into the
1192                          * repository, we cope with that by stripping the expansion out.
1193                          * This is probably not a good idea, since it will cause changes
1194                          * on checkout, which won't go away by stash, but let's keep it
1195                          * for git-style ids.
1196                          */
1197                         dollar = memchr(src + 3, '$', len - 3);
1198                         if (!dollar) {
1199                                 /* incomplete keyword, no more '$', so just quit the loop */
1200                                 break;
1201                         }
1202
1203                         if (memchr(src + 3, '\n', dollar - src - 3)) {
1204                                 /* Line break before the next dollar. */
1205                                 continue;
1206                         }
1207
1208                         spc = memchr(src + 4, ' ', dollar - src - 4);
1209                         if (spc && spc < dollar-1) {
1210                                 /* There are spaces in unexpected places.
1211                                  * This is probably an id from some other
1212                                  * versioning system. Keep it for now.
1213                                  */
1214                                 continue;
1215                         }
1216
1217                         len -= dollar + 1 - src;
1218                         src  = dollar + 1;
1219                 } else {
1220                         /* it wasn't a "Id$" or "Id:xxxx$" */
1221                         continue;
1222                 }
1223
1224                 /* step 4: substitute */
1225                 strbuf_addstr(buf, "Id: ");
1226                 strbuf_addstr(buf, oid_to_hex(&oid));
1227                 strbuf_addstr(buf, " $");
1228         }
1229         strbuf_add(buf, src, len);
1230
1231         free(to_free);
1232         return 1;
1233 }
1234
1235 static const char *git_path_check_encoding(struct attr_check_item *check)
1236 {
1237         const char *value = check->value;
1238
1239         if (ATTR_UNSET(value) || !strlen(value))
1240                 return NULL;
1241
1242         if (ATTR_TRUE(value) || ATTR_FALSE(value)) {
1243                 die(_("true/false are no valid working-tree-encodings"));
1244         }
1245
1246         /* Don't encode to the default encoding */
1247         if (same_encoding(value, default_encoding))
1248                 return NULL;
1249
1250         return value;
1251 }
1252
1253 static enum crlf_action git_path_check_crlf(struct attr_check_item *check)
1254 {
1255         const char *value = check->value;
1256
1257         if (ATTR_TRUE(value))
1258                 return CRLF_TEXT;
1259         else if (ATTR_FALSE(value))
1260                 return CRLF_BINARY;
1261         else if (ATTR_UNSET(value))
1262                 ;
1263         else if (!strcmp(value, "input"))
1264                 return CRLF_TEXT_INPUT;
1265         else if (!strcmp(value, "auto"))
1266                 return CRLF_AUTO;
1267         return CRLF_UNDEFINED;
1268 }
1269
1270 static enum eol git_path_check_eol(struct attr_check_item *check)
1271 {
1272         const char *value = check->value;
1273
1274         if (ATTR_UNSET(value))
1275                 ;
1276         else if (!strcmp(value, "lf"))
1277                 return EOL_LF;
1278         else if (!strcmp(value, "crlf"))
1279                 return EOL_CRLF;
1280         return EOL_UNSET;
1281 }
1282
1283 static struct convert_driver *git_path_check_convert(struct attr_check_item *check)
1284 {
1285         const char *value = check->value;
1286         struct convert_driver *drv;
1287
1288         if (ATTR_TRUE(value) || ATTR_FALSE(value) || ATTR_UNSET(value))
1289                 return NULL;
1290         for (drv = user_convert; drv; drv = drv->next)
1291                 if (!strcmp(value, drv->name))
1292                         return drv;
1293         return NULL;
1294 }
1295
1296 static int git_path_check_ident(struct attr_check_item *check)
1297 {
1298         const char *value = check->value;
1299
1300         return !!ATTR_TRUE(value);
1301 }
1302
1303 struct conv_attrs {
1304         struct convert_driver *drv;
1305         enum crlf_action attr_action; /* What attr says */
1306         enum crlf_action crlf_action; /* When no attr is set, use core.autocrlf */
1307         int ident;
1308         const char *working_tree_encoding; /* Supported encoding or default encoding if NULL */
1309 };
1310
1311 static struct attr_check *check;
1312
1313 static void convert_attrs(const struct index_state *istate,
1314                           struct conv_attrs *ca, const char *path)
1315 {
1316         struct attr_check_item *ccheck = NULL;
1317
1318         if (!check) {
1319                 check = attr_check_initl("crlf", "ident", "filter",
1320                                          "eol", "text", "working-tree-encoding",
1321                                          NULL);
1322                 user_convert_tail = &user_convert;
1323                 git_config(read_convert_config, NULL);
1324         }
1325
1326         git_check_attr(istate, path, check);
1327         ccheck = check->items;
1328         ca->crlf_action = git_path_check_crlf(ccheck + 4);
1329         if (ca->crlf_action == CRLF_UNDEFINED)
1330                 ca->crlf_action = git_path_check_crlf(ccheck + 0);
1331         ca->ident = git_path_check_ident(ccheck + 1);
1332         ca->drv = git_path_check_convert(ccheck + 2);
1333         if (ca->crlf_action != CRLF_BINARY) {
1334                 enum eol eol_attr = git_path_check_eol(ccheck + 3);
1335                 if (ca->crlf_action == CRLF_AUTO && eol_attr == EOL_LF)
1336                         ca->crlf_action = CRLF_AUTO_INPUT;
1337                 else if (ca->crlf_action == CRLF_AUTO && eol_attr == EOL_CRLF)
1338                         ca->crlf_action = CRLF_AUTO_CRLF;
1339                 else if (eol_attr == EOL_LF)
1340                         ca->crlf_action = CRLF_TEXT_INPUT;
1341                 else if (eol_attr == EOL_CRLF)
1342                         ca->crlf_action = CRLF_TEXT_CRLF;
1343         }
1344         ca->working_tree_encoding = git_path_check_encoding(ccheck + 5);
1345
1346         /* Save attr and make a decision for action */
1347         ca->attr_action = ca->crlf_action;
1348         if (ca->crlf_action == CRLF_TEXT)
1349                 ca->crlf_action = text_eol_is_crlf() ? CRLF_TEXT_CRLF : CRLF_TEXT_INPUT;
1350         if (ca->crlf_action == CRLF_UNDEFINED && auto_crlf == AUTO_CRLF_FALSE)
1351                 ca->crlf_action = CRLF_BINARY;
1352         if (ca->crlf_action == CRLF_UNDEFINED && auto_crlf == AUTO_CRLF_TRUE)
1353                 ca->crlf_action = CRLF_AUTO_CRLF;
1354         if (ca->crlf_action == CRLF_UNDEFINED && auto_crlf == AUTO_CRLF_INPUT)
1355                 ca->crlf_action = CRLF_AUTO_INPUT;
1356 }
1357
1358 void reset_parsed_attributes(void)
1359 {
1360         struct convert_driver *drv, *next;
1361
1362         attr_check_free(check);
1363         check = NULL;
1364         reset_merge_attributes();
1365
1366         for (drv = user_convert; drv; drv = next) {
1367                 next = drv->next;
1368                 free((void *)drv->name);
1369                 free(drv);
1370         }
1371         user_convert = NULL;
1372         user_convert_tail = NULL;
1373 }
1374
1375 int would_convert_to_git_filter_fd(const struct index_state *istate, const char *path)
1376 {
1377         struct conv_attrs ca;
1378
1379         convert_attrs(istate, &ca, path);
1380         if (!ca.drv)
1381                 return 0;
1382
1383         /*
1384          * Apply a filter to an fd only if the filter is required to succeed.
1385          * We must die if the filter fails, because the original data before
1386          * filtering is not available.
1387          */
1388         if (!ca.drv->required)
1389                 return 0;
1390
1391         return apply_filter(path, NULL, 0, -1, NULL, ca.drv, CAP_CLEAN, NULL, NULL);
1392 }
1393
1394 const char *get_convert_attr_ascii(const struct index_state *istate, const char *path)
1395 {
1396         struct conv_attrs ca;
1397
1398         convert_attrs(istate, &ca, path);
1399         switch (ca.attr_action) {
1400         case CRLF_UNDEFINED:
1401                 return "";
1402         case CRLF_BINARY:
1403                 return "-text";
1404         case CRLF_TEXT:
1405                 return "text";
1406         case CRLF_TEXT_INPUT:
1407                 return "text eol=lf";
1408         case CRLF_TEXT_CRLF:
1409                 return "text eol=crlf";
1410         case CRLF_AUTO:
1411                 return "text=auto";
1412         case CRLF_AUTO_CRLF:
1413                 return "text=auto eol=crlf";
1414         case CRLF_AUTO_INPUT:
1415                 return "text=auto eol=lf";
1416         }
1417         return "";
1418 }
1419
1420 int convert_to_git(const struct index_state *istate,
1421                    const char *path, const char *src, size_t len,
1422                    struct strbuf *dst, int conv_flags)
1423 {
1424         int ret = 0;
1425         struct conv_attrs ca;
1426
1427         convert_attrs(istate, &ca, path);
1428
1429         ret |= apply_filter(path, src, len, -1, dst, ca.drv, CAP_CLEAN, NULL, NULL);
1430         if (!ret && ca.drv && ca.drv->required)
1431                 die(_("%s: clean filter '%s' failed"), path, ca.drv->name);
1432
1433         if (ret && dst) {
1434                 src = dst->buf;
1435                 len = dst->len;
1436         }
1437
1438         ret |= encode_to_git(path, src, len, dst, ca.working_tree_encoding, conv_flags);
1439         if (ret && dst) {
1440                 src = dst->buf;
1441                 len = dst->len;
1442         }
1443
1444         if (!(conv_flags & CONV_EOL_KEEP_CRLF)) {
1445                 ret |= crlf_to_git(istate, path, src, len, dst, ca.crlf_action, conv_flags);
1446                 if (ret && dst) {
1447                         src = dst->buf;
1448                         len = dst->len;
1449                 }
1450         }
1451         return ret | ident_to_git(src, len, dst, ca.ident);
1452 }
1453
1454 void convert_to_git_filter_fd(const struct index_state *istate,
1455                               const char *path, int fd, struct strbuf *dst,
1456                               int conv_flags)
1457 {
1458         struct conv_attrs ca;
1459         convert_attrs(istate, &ca, path);
1460
1461         assert(ca.drv);
1462         assert(ca.drv->clean || ca.drv->process);
1463
1464         if (!apply_filter(path, NULL, 0, fd, dst, ca.drv, CAP_CLEAN, NULL, NULL))
1465                 die(_("%s: clean filter '%s' failed"), path, ca.drv->name);
1466
1467         encode_to_git(path, dst->buf, dst->len, dst, ca.working_tree_encoding, conv_flags);
1468         crlf_to_git(istate, path, dst->buf, dst->len, dst, ca.crlf_action, conv_flags);
1469         ident_to_git(dst->buf, dst->len, dst, ca.ident);
1470 }
1471
1472 static int convert_to_working_tree_internal(const struct index_state *istate,
1473                                             const char *path, const char *src,
1474                                             size_t len, struct strbuf *dst,
1475                                             int normalizing,
1476                                             const struct checkout_metadata *meta,
1477                                             struct delayed_checkout *dco)
1478 {
1479         int ret = 0, ret_filter = 0;
1480         struct conv_attrs ca;
1481
1482         convert_attrs(istate, &ca, path);
1483
1484         ret |= ident_to_worktree(src, len, dst, ca.ident);
1485         if (ret) {
1486                 src = dst->buf;
1487                 len = dst->len;
1488         }
1489         /*
1490          * CRLF conversion can be skipped if normalizing, unless there
1491          * is a smudge or process filter (even if the process filter doesn't
1492          * support smudge).  The filters might expect CRLFs.
1493          */
1494         if ((ca.drv && (ca.drv->smudge || ca.drv->process)) || !normalizing) {
1495                 ret |= crlf_to_worktree(src, len, dst, ca.crlf_action);
1496                 if (ret) {
1497                         src = dst->buf;
1498                         len = dst->len;
1499                 }
1500         }
1501
1502         ret |= encode_to_worktree(path, src, len, dst, ca.working_tree_encoding);
1503         if (ret) {
1504                 src = dst->buf;
1505                 len = dst->len;
1506         }
1507
1508         ret_filter = apply_filter(
1509                 path, src, len, -1, dst, ca.drv, CAP_SMUDGE, meta, dco);
1510         if (!ret_filter && ca.drv && ca.drv->required)
1511                 die(_("%s: smudge filter %s failed"), path, ca.drv->name);
1512
1513         return ret | ret_filter;
1514 }
1515
1516 int async_convert_to_working_tree(const struct index_state *istate,
1517                                   const char *path, const char *src,
1518                                   size_t len, struct strbuf *dst,
1519                                   const struct checkout_metadata *meta,
1520                                   void *dco)
1521 {
1522         return convert_to_working_tree_internal(istate, path, src, len, dst, 0, meta, dco);
1523 }
1524
1525 int convert_to_working_tree(const struct index_state *istate,
1526                             const char *path, const char *src,
1527                             size_t len, struct strbuf *dst,
1528                             const struct checkout_metadata *meta)
1529 {
1530         return convert_to_working_tree_internal(istate, path, src, len, dst, 0, meta, NULL);
1531 }
1532
1533 int renormalize_buffer(const struct index_state *istate, const char *path,
1534                        const char *src, size_t len, struct strbuf *dst)
1535 {
1536         int ret = convert_to_working_tree_internal(istate, path, src, len, dst, 1, NULL, NULL);
1537         if (ret) {
1538                 src = dst->buf;
1539                 len = dst->len;
1540         }
1541         return ret | convert_to_git(istate, path, src, len, dst, CONV_EOL_RENORMALIZE);
1542 }
1543
1544 /*****************************************************************
1545  *
1546  * Streaming conversion support
1547  *
1548  *****************************************************************/
1549
1550 typedef int (*filter_fn)(struct stream_filter *,
1551                          const char *input, size_t *isize_p,
1552                          char *output, size_t *osize_p);
1553 typedef void (*free_fn)(struct stream_filter *);
1554
1555 struct stream_filter_vtbl {
1556         filter_fn filter;
1557         free_fn free;
1558 };
1559
1560 struct stream_filter {
1561         struct stream_filter_vtbl *vtbl;
1562 };
1563
1564 static int null_filter_fn(struct stream_filter *filter,
1565                           const char *input, size_t *isize_p,
1566                           char *output, size_t *osize_p)
1567 {
1568         size_t count;
1569
1570         if (!input)
1571                 return 0; /* we do not keep any states */
1572         count = *isize_p;
1573         if (*osize_p < count)
1574                 count = *osize_p;
1575         if (count) {
1576                 memmove(output, input, count);
1577                 *isize_p -= count;
1578                 *osize_p -= count;
1579         }
1580         return 0;
1581 }
1582
1583 static void null_free_fn(struct stream_filter *filter)
1584 {
1585         ; /* nothing -- null instances are shared */
1586 }
1587
1588 static struct stream_filter_vtbl null_vtbl = {
1589         null_filter_fn,
1590         null_free_fn,
1591 };
1592
1593 static struct stream_filter null_filter_singleton = {
1594         &null_vtbl,
1595 };
1596
1597 int is_null_stream_filter(struct stream_filter *filter)
1598 {
1599         return filter == &null_filter_singleton;
1600 }
1601
1602
1603 /*
1604  * LF-to-CRLF filter
1605  */
1606
1607 struct lf_to_crlf_filter {
1608         struct stream_filter filter;
1609         unsigned has_held:1;
1610         char held;
1611 };
1612
1613 static int lf_to_crlf_filter_fn(struct stream_filter *filter,
1614                                 const char *input, size_t *isize_p,
1615                                 char *output, size_t *osize_p)
1616 {
1617         size_t count, o = 0;
1618         struct lf_to_crlf_filter *lf_to_crlf = (struct lf_to_crlf_filter *)filter;
1619
1620         /*
1621          * We may be holding onto the CR to see if it is followed by a
1622          * LF, in which case we would need to go to the main loop.
1623          * Otherwise, just emit it to the output stream.
1624          */
1625         if (lf_to_crlf->has_held && (lf_to_crlf->held != '\r' || !input)) {
1626                 output[o++] = lf_to_crlf->held;
1627                 lf_to_crlf->has_held = 0;
1628         }
1629
1630         /* We are told to drain */
1631         if (!input) {
1632                 *osize_p -= o;
1633                 return 0;
1634         }
1635
1636         count = *isize_p;
1637         if (count || lf_to_crlf->has_held) {
1638                 size_t i;
1639                 int was_cr = 0;
1640
1641                 if (lf_to_crlf->has_held) {
1642                         was_cr = 1;
1643                         lf_to_crlf->has_held = 0;
1644                 }
1645
1646                 for (i = 0; o < *osize_p && i < count; i++) {
1647                         char ch = input[i];
1648
1649                         if (ch == '\n') {
1650                                 output[o++] = '\r';
1651                         } else if (was_cr) {
1652                                 /*
1653                                  * Previous round saw CR and it is not followed
1654                                  * by a LF; emit the CR before processing the
1655                                  * current character.
1656                                  */
1657                                 output[o++] = '\r';
1658                         }
1659
1660                         /*
1661                          * We may have consumed the last output slot,
1662                          * in which case we need to break out of this
1663                          * loop; hold the current character before
1664                          * returning.
1665                          */
1666                         if (*osize_p <= o) {
1667                                 lf_to_crlf->has_held = 1;
1668                                 lf_to_crlf->held = ch;
1669                                 continue; /* break but increment i */
1670                         }
1671
1672                         if (ch == '\r') {
1673                                 was_cr = 1;
1674                                 continue;
1675                         }
1676
1677                         was_cr = 0;
1678                         output[o++] = ch;
1679                 }
1680
1681                 *osize_p -= o;
1682                 *isize_p -= i;
1683
1684                 if (!lf_to_crlf->has_held && was_cr) {
1685                         lf_to_crlf->has_held = 1;
1686                         lf_to_crlf->held = '\r';
1687                 }
1688         }
1689         return 0;
1690 }
1691
1692 static void lf_to_crlf_free_fn(struct stream_filter *filter)
1693 {
1694         free(filter);
1695 }
1696
1697 static struct stream_filter_vtbl lf_to_crlf_vtbl = {
1698         lf_to_crlf_filter_fn,
1699         lf_to_crlf_free_fn,
1700 };
1701
1702 static struct stream_filter *lf_to_crlf_filter(void)
1703 {
1704         struct lf_to_crlf_filter *lf_to_crlf = xcalloc(1, sizeof(*lf_to_crlf));
1705
1706         lf_to_crlf->filter.vtbl = &lf_to_crlf_vtbl;
1707         return (struct stream_filter *)lf_to_crlf;
1708 }
1709
1710 /*
1711  * Cascade filter
1712  */
1713 #define FILTER_BUFFER 1024
1714 struct cascade_filter {
1715         struct stream_filter filter;
1716         struct stream_filter *one;
1717         struct stream_filter *two;
1718         char buf[FILTER_BUFFER];
1719         int end, ptr;
1720 };
1721
1722 static int cascade_filter_fn(struct stream_filter *filter,
1723                              const char *input, size_t *isize_p,
1724                              char *output, size_t *osize_p)
1725 {
1726         struct cascade_filter *cas = (struct cascade_filter *) filter;
1727         size_t filled = 0;
1728         size_t sz = *osize_p;
1729         size_t to_feed, remaining;
1730
1731         /*
1732          * input -- (one) --> buf -- (two) --> output
1733          */
1734         while (filled < sz) {
1735                 remaining = sz - filled;
1736
1737                 /* do we already have something to feed two with? */
1738                 if (cas->ptr < cas->end) {
1739                         to_feed = cas->end - cas->ptr;
1740                         if (stream_filter(cas->two,
1741                                           cas->buf + cas->ptr, &to_feed,
1742                                           output + filled, &remaining))
1743                                 return -1;
1744                         cas->ptr += (cas->end - cas->ptr) - to_feed;
1745                         filled = sz - remaining;
1746                         continue;
1747                 }
1748
1749                 /* feed one from upstream and have it emit into our buffer */
1750                 to_feed = input ? *isize_p : 0;
1751                 if (input && !to_feed)
1752                         break;
1753                 remaining = sizeof(cas->buf);
1754                 if (stream_filter(cas->one,
1755                                   input, &to_feed,
1756                                   cas->buf, &remaining))
1757                         return -1;
1758                 cas->end = sizeof(cas->buf) - remaining;
1759                 cas->ptr = 0;
1760                 if (input) {
1761                         size_t fed = *isize_p - to_feed;
1762                         *isize_p -= fed;
1763                         input += fed;
1764                 }
1765
1766                 /* do we know that we drained one completely? */
1767                 if (input || cas->end)
1768                         continue;
1769
1770                 /* tell two to drain; we have nothing more to give it */
1771                 to_feed = 0;
1772                 remaining = sz - filled;
1773                 if (stream_filter(cas->two,
1774                                   NULL, &to_feed,
1775                                   output + filled, &remaining))
1776                         return -1;
1777                 if (remaining == (sz - filled))
1778                         break; /* completely drained two */
1779                 filled = sz - remaining;
1780         }
1781         *osize_p -= filled;
1782         return 0;
1783 }
1784
1785 static void cascade_free_fn(struct stream_filter *filter)
1786 {
1787         struct cascade_filter *cas = (struct cascade_filter *)filter;
1788         free_stream_filter(cas->one);
1789         free_stream_filter(cas->two);
1790         free(filter);
1791 }
1792
1793 static struct stream_filter_vtbl cascade_vtbl = {
1794         cascade_filter_fn,
1795         cascade_free_fn,
1796 };
1797
1798 static struct stream_filter *cascade_filter(struct stream_filter *one,
1799                                             struct stream_filter *two)
1800 {
1801         struct cascade_filter *cascade;
1802
1803         if (!one || is_null_stream_filter(one))
1804                 return two;
1805         if (!two || is_null_stream_filter(two))
1806                 return one;
1807
1808         cascade = xmalloc(sizeof(*cascade));
1809         cascade->one = one;
1810         cascade->two = two;
1811         cascade->end = cascade->ptr = 0;
1812         cascade->filter.vtbl = &cascade_vtbl;
1813         return (struct stream_filter *)cascade;
1814 }
1815
1816 /*
1817  * ident filter
1818  */
1819 #define IDENT_DRAINING (-1)
1820 #define IDENT_SKIPPING (-2)
1821 struct ident_filter {
1822         struct stream_filter filter;
1823         struct strbuf left;
1824         int state;
1825         char ident[GIT_MAX_HEXSZ + 5]; /* ": x40 $" */
1826 };
1827
1828 static int is_foreign_ident(const char *str)
1829 {
1830         int i;
1831
1832         if (!skip_prefix(str, "$Id: ", &str))
1833                 return 0;
1834         for (i = 0; str[i]; i++) {
1835                 if (isspace(str[i]) && str[i+1] != '$')
1836                         return 1;
1837         }
1838         return 0;
1839 }
1840
1841 static void ident_drain(struct ident_filter *ident, char **output_p, size_t *osize_p)
1842 {
1843         size_t to_drain = ident->left.len;
1844
1845         if (*osize_p < to_drain)
1846                 to_drain = *osize_p;
1847         if (to_drain) {
1848                 memcpy(*output_p, ident->left.buf, to_drain);
1849                 strbuf_remove(&ident->left, 0, to_drain);
1850                 *output_p += to_drain;
1851                 *osize_p -= to_drain;
1852         }
1853         if (!ident->left.len)
1854                 ident->state = 0;
1855 }
1856
1857 static int ident_filter_fn(struct stream_filter *filter,
1858                            const char *input, size_t *isize_p,
1859                            char *output, size_t *osize_p)
1860 {
1861         struct ident_filter *ident = (struct ident_filter *)filter;
1862         static const char head[] = "$Id";
1863
1864         if (!input) {
1865                 /* drain upon eof */
1866                 switch (ident->state) {
1867                 default:
1868                         strbuf_add(&ident->left, head, ident->state);
1869                         /* fallthrough */
1870                 case IDENT_SKIPPING:
1871                         /* fallthrough */
1872                 case IDENT_DRAINING:
1873                         ident_drain(ident, &output, osize_p);
1874                 }
1875                 return 0;
1876         }
1877
1878         while (*isize_p || (ident->state == IDENT_DRAINING)) {
1879                 int ch;
1880
1881                 if (ident->state == IDENT_DRAINING) {
1882                         ident_drain(ident, &output, osize_p);
1883                         if (!*osize_p)
1884                                 break;
1885                         continue;
1886                 }
1887
1888                 ch = *(input++);
1889                 (*isize_p)--;
1890
1891                 if (ident->state == IDENT_SKIPPING) {
1892                         /*
1893                          * Skipping until '$' or LF, but keeping them
1894                          * in case it is a foreign ident.
1895                          */
1896                         strbuf_addch(&ident->left, ch);
1897                         if (ch != '\n' && ch != '$')
1898                                 continue;
1899                         if (ch == '$' && !is_foreign_ident(ident->left.buf)) {
1900                                 strbuf_setlen(&ident->left, sizeof(head) - 1);
1901                                 strbuf_addstr(&ident->left, ident->ident);
1902                         }
1903                         ident->state = IDENT_DRAINING;
1904                         continue;
1905                 }
1906
1907                 if (ident->state < sizeof(head) &&
1908                     head[ident->state] == ch) {
1909                         ident->state++;
1910                         continue;
1911                 }
1912
1913                 if (ident->state)
1914                         strbuf_add(&ident->left, head, ident->state);
1915                 if (ident->state == sizeof(head) - 1) {
1916                         if (ch != ':' && ch != '$') {
1917                                 strbuf_addch(&ident->left, ch);
1918                                 ident->state = 0;
1919                                 continue;
1920                         }
1921
1922                         if (ch == ':') {
1923                                 strbuf_addch(&ident->left, ch);
1924                                 ident->state = IDENT_SKIPPING;
1925                         } else {
1926                                 strbuf_addstr(&ident->left, ident->ident);
1927                                 ident->state = IDENT_DRAINING;
1928                         }
1929                         continue;
1930                 }
1931
1932                 strbuf_addch(&ident->left, ch);
1933                 ident->state = IDENT_DRAINING;
1934         }
1935         return 0;
1936 }
1937
1938 static void ident_free_fn(struct stream_filter *filter)
1939 {
1940         struct ident_filter *ident = (struct ident_filter *)filter;
1941         strbuf_release(&ident->left);
1942         free(filter);
1943 }
1944
1945 static struct stream_filter_vtbl ident_vtbl = {
1946         ident_filter_fn,
1947         ident_free_fn,
1948 };
1949
1950 static struct stream_filter *ident_filter(const struct object_id *oid)
1951 {
1952         struct ident_filter *ident = xmalloc(sizeof(*ident));
1953
1954         xsnprintf(ident->ident, sizeof(ident->ident),
1955                   ": %s $", oid_to_hex(oid));
1956         strbuf_init(&ident->left, 0);
1957         ident->filter.vtbl = &ident_vtbl;
1958         ident->state = 0;
1959         return (struct stream_filter *)ident;
1960 }
1961
1962 /*
1963  * Return an appropriately constructed filter for the path, or NULL if
1964  * the contents cannot be filtered without reading the whole thing
1965  * in-core.
1966  *
1967  * Note that you would be crazy to set CRLF, smudge/clean or ident to a
1968  * large binary blob you would want us not to slurp into the memory!
1969  */
1970 struct stream_filter *get_stream_filter(const struct index_state *istate,
1971                                         const char *path,
1972                                         const struct object_id *oid)
1973 {
1974         struct conv_attrs ca;
1975         struct stream_filter *filter = NULL;
1976
1977         convert_attrs(istate, &ca, path);
1978         if (ca.drv && (ca.drv->process || ca.drv->smudge || ca.drv->clean))
1979                 return NULL;
1980
1981         if (ca.working_tree_encoding)
1982                 return NULL;
1983
1984         if (ca.crlf_action == CRLF_AUTO || ca.crlf_action == CRLF_AUTO_CRLF)
1985                 return NULL;
1986
1987         if (ca.ident)
1988                 filter = ident_filter(oid);
1989
1990         if (output_eol(ca.crlf_action) == EOL_CRLF)
1991                 filter = cascade_filter(filter, lf_to_crlf_filter());
1992         else
1993                 filter = cascade_filter(filter, &null_filter_singleton);
1994
1995         return filter;
1996 }
1997
1998 void free_stream_filter(struct stream_filter *filter)
1999 {
2000         filter->vtbl->free(filter);
2001 }
2002
2003 int stream_filter(struct stream_filter *filter,
2004                   const char *input, size_t *isize_p,
2005                   char *output, size_t *osize_p)
2006 {
2007         return filter->vtbl->filter(filter, input, isize_p, output, osize_p);
2008 }
2009
2010 void init_checkout_metadata(struct checkout_metadata *meta, const char *refname,
2011                             const struct object_id *treeish,
2012                             const struct object_id *blob)
2013 {
2014         memset(meta, 0, sizeof(*meta));
2015         if (refname)
2016                 meta->refname = refname;
2017         if (treeish)
2018                 oidcpy(&meta->treeish, treeish);
2019         if (blob)
2020                 oidcpy(&meta->blob, blob);
2021 }
2022
2023 void clone_checkout_metadata(struct checkout_metadata *dst,
2024                              const struct checkout_metadata *src,
2025                              const struct object_id *blob)
2026 {
2027         memcpy(dst, src, sizeof(*dst));
2028         if (blob)
2029                 oidcpy(&dst->blob, blob);
2030 }