convert.c

   1 #include "cache.h"
   2 #include "attr.h"
   3 #include "run-command.h"
   4 #include "quote.h"
   5 #include "sigchain.h"
   6
   7 /*
   8  * convert.c - convert a file when checking it out and checking it in.
   9  *
  10  * This should use the pathname to decide on whether it wants to do some
  11  * more interesting conversions (automatic gzip/unzip, general format
  12  * conversions etc etc), but by default it just does automatic CRLF<->LF
  13  * translation when the "text" attribute or "auto_crlf" option is set.
  14  */
  15
  16 /* Stat bits: When BIN is set, the txt bits are unset */
  17 #define CONVERT_STAT_BITS_TXT_LF    0x1
  18 #define CONVERT_STAT_BITS_TXT_CRLF  0x2
  19 #define CONVERT_STAT_BITS_BIN       0x4
  20
  21 enum crlf_action {
  22         CRLF_UNDEFINED,
  23         CRLF_BINARY,
  24         CRLF_TEXT,
  25         CRLF_TEXT_INPUT,
  26         CRLF_TEXT_CRLF,
  27         CRLF_AUTO,
  28         CRLF_AUTO_INPUT,
  29         CRLF_AUTO_CRLF
  30 };
  31
  32 struct text_stat {
  33         /* NUL, CR, LF and CRLF counts */
  34         unsigned nul, lonecr, lonelf, crlf;
  35
  36         /* These are just approximations! */
  37         unsigned printable, nonprintable;
  38 };
  39
  40 static void gather_stats(const char *buf, unsigned long size, struct text_stat *stats)
  41 {
  42         unsigned long i;
  43
  44         memset(stats, 0, sizeof(*stats));
  45
  46         for (i = 0; i < size; i++) {
  47                 unsigned char c = buf[i];
  48                 if (c == '\r') {
  49                         if (i+1 < size && buf[i+1] == '\n') {
  50                                 stats->crlf++;
  51                                 i++;
  52                         } else
  53                                 stats->lonecr++;
  54                         continue;
  55                 }
  56                 if (c == '\n') {
  57                         stats->lonelf++;
  58                         continue;
  59                 }
  60                 if (c == 127)
  61                         /* DEL */
  62                         stats->nonprintable++;
  63                 else if (c < 32) {
  64                         switch (c) {
  65                                 /* BS, HT, ESC and FF */
  66                         case '\b': case '\t': case '\033': case '\014':
  67                                 stats->printable++;
  68                                 break;
  69                         case 0:
  70                                 stats->nul++;
  71                                 /* fall through */
  72                         default:
  73                                 stats->nonprintable++;
  74                         }
  75                 }
  76                 else
  77                         stats->printable++;
  78         }
  79
  80         /* If file ends with EOF then don't count this EOF as non-printable. */
  81         if (size >= 1 && buf[size-1] == '\032')
  82                 stats->nonprintable--;
  83 }
  84
  85 /*
  86  * The same heuristics as diff.c::mmfile_is_binary()
  87  * We treat files with bare CR as binary
  88  */
  89 static int convert_is_binary(unsigned long size, const struct text_stat *stats)
  90 {
  91         if (stats->lonecr)
  92                 return 1;
  93         if (stats->nul)
  94                 return 1;
  95         if ((stats->printable >> 7) < stats->nonprintable)
  96                 return 1;
  97         return 0;
  98 }
  99
 100 static unsigned int gather_convert_stats(const char *data, unsigned long size)
 101 {
 102         struct text_stat stats;
 103         int ret = 0;
 104         if (!data || !size)
 105                 return 0;
 106         gather_stats(data, size, &stats);
 107         if (convert_is_binary(size, &stats))
 108                 ret |= CONVERT_STAT_BITS_BIN;
 109         if (stats.crlf)
 110                 ret |= CONVERT_STAT_BITS_TXT_CRLF;
 111         if (stats.lonelf)
 112                 ret |=  CONVERT_STAT_BITS_TXT_LF;
 113
 114         return ret;
 115 }
 116
 117 static const char *gather_convert_stats_ascii(const char *data, unsigned long size)
 118 {
 119         unsigned int convert_stats = gather_convert_stats(data, size);
 120
 121         if (convert_stats & CONVERT_STAT_BITS_BIN)
 122                 return "-text";
 123         switch (convert_stats) {
 124         case CONVERT_STAT_BITS_TXT_LF:
 125                 return "lf";
 126         case CONVERT_STAT_BITS_TXT_CRLF:
 127                 return "crlf";
 128         case CONVERT_STAT_BITS_TXT_LF | CONVERT_STAT_BITS_TXT_CRLF:
 129                 return "mixed";
 130         default:
 131                 return "none";
 132         }
 133 }
 134
 135 const char *get_cached_convert_stats_ascii(const char *path)
 136 {
 137         const char *ret;
 138         unsigned long sz;
 139         void *data = read_blob_data_from_cache(path, &sz);
 140         ret = gather_convert_stats_ascii(data, sz);
 141         free(data);
 142         return ret;
 143 }
 144
 145 const char *get_wt_convert_stats_ascii(const char *path)
 146 {
 147         const char *ret = "";
 148         struct strbuf sb = STRBUF_INIT;
 149         if (strbuf_read_file(&sb, path, 0) >= 0)
 150                 ret = gather_convert_stats_ascii(sb.buf, sb.len);
 151         strbuf_release(&sb);
 152         return ret;
 153 }
 154
 155 static int text_eol_is_crlf(void)
 156 {
 157         if (auto_crlf == AUTO_CRLF_TRUE)
 158                 return 1;
 159         else if (auto_crlf == AUTO_CRLF_INPUT)
 160                 return 0;
 161         if (core_eol == EOL_CRLF)
 162                 return 1;
 163         if (core_eol == EOL_UNSET && EOL_NATIVE == EOL_CRLF)
 164                 return 1;
 165         return 0;
 166 }
 167
 168 static enum eol output_eol(enum crlf_action crlf_action)
 169 {
 170         switch (crlf_action) {
 171         case CRLF_BINARY:
 172                 return EOL_UNSET;
 173         case CRLF_TEXT_CRLF:
 174                 return EOL_CRLF;
 175         case CRLF_TEXT_INPUT:
 176                 return EOL_LF;
 177         case CRLF_UNDEFINED:
 178         case CRLF_AUTO_CRLF:
 179         case CRLF_AUTO_INPUT:
 180         case CRLF_TEXT:
 181         case CRLF_AUTO:
 182                 /* fall through */
 183                 return text_eol_is_crlf() ? EOL_CRLF : EOL_LF;
 184         }
 185         warning("Illegal crlf_action %d\n", (int)crlf_action);
 186         return core_eol;
 187 }
 188
 189 static void check_safe_crlf(const char *path, enum crlf_action crlf_action,
 190                             struct text_stat *stats, enum safe_crlf checksafe)
 191 {
 192         if (!checksafe)
 193                 return;
 194
 195         if (output_eol(crlf_action) == EOL_LF) {
 196                 /*
 197                  * CRLFs would not be restored by checkout:
 198                  * check if we'd remove CRLFs
 199                  */
 200                 if (stats->crlf) {
 201                         if (checksafe == SAFE_CRLF_WARN)
 202                                 warning("CRLF will be replaced by LF in %s.\nThe file will have its original line endings in your working directory.", path);
 203                         else /* i.e. SAFE_CRLF_FAIL */
 204                                 die("CRLF would be replaced by LF in %s.", path);
 205                 }
 206         } else if (output_eol(crlf_action) == EOL_CRLF) {
 207                 /*
 208                  * CRLFs would be added by checkout:
 209                  * check if we have "naked" LFs
 210                  */
 211                 if (stats->lonelf) {
 212                         if (checksafe == SAFE_CRLF_WARN)
 213                                 warning("LF will be replaced by CRLF in %s.\nThe file will have its original line endings in your working directory.", path);
 214                         else /* i.e. SAFE_CRLF_FAIL */
 215                                 die("LF would be replaced by CRLF in %s", path);
 216                 }
 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 crlf_to_git(const char *path, const char *src, size_t len,
 235                        struct strbuf *buf,
 236                        enum crlf_action crlf_action, enum safe_crlf checksafe)
 237 {
 238         struct text_stat stats;
 239         char *dst;
 240
 241         if (crlf_action == CRLF_BINARY ||
 242             (src && !len))
 243                 return 0;
 244
 245         /*
 246          * If we are doing a dry-run and have no source buffer, there is
 247          * nothing to analyze; we must assume we would convert.
 248          */
 249         if (!buf && !src)
 250                 return 1;
 251
 252         gather_stats(src, len, &stats);
 253
 254         if (crlf_action == CRLF_AUTO || crlf_action == CRLF_AUTO_INPUT || crlf_action == CRLF_AUTO_CRLF) {
 255                 if (convert_is_binary(len, &stats))
 256                         return 0;
 257
 258                 if (crlf_action == CRLF_AUTO_INPUT || crlf_action == CRLF_AUTO_CRLF) {
 259                         /*
 260                          * If the file in the index has any CR in it, do not convert.
 261                          * This is the new safer autocrlf handling.
 262                          */
 263                         if (has_cr_in_index(path))
 264                                 return 0;
 265                 }
 266         }
 267
 268         check_safe_crlf(path, crlf_action, &stats, checksafe);
 269
 270         /* Optimization: No CRLF? Nothing to convert, regardless. */
 271         if (!stats.crlf)
 272                 return 0;
 273
 274         /*
 275          * At this point all of our source analysis is done, and we are sure we
 276          * would convert. If we are in dry-run mode, we can give an answer.
 277          */
 278         if (!buf)
 279                 return 1;
 280
 281         /* only grow if not in place */
 282         if (strbuf_avail(buf) + buf->len < len)
 283                 strbuf_grow(buf, len - buf->len);
 284         dst = buf->buf;
 285         if (crlf_action == CRLF_AUTO || crlf_action == CRLF_AUTO_INPUT || crlf_action == CRLF_AUTO_CRLF) {
 286                 /*
 287                  * If we guessed, we already know we rejected a file with
 288                  * lone CR, and we can strip a CR without looking at what
 289                  * follow it.
 290                  */
 291                 do {
 292                         unsigned char c = *src++;
 293                         if (c != '\r')
 294                                 *dst++ = c;
 295                 } while (--len);
 296         } else {
 297                 do {
 298                         unsigned char c = *src++;
 299                         if (! (c == '\r' && (1 < len && *src == '\n')))
 300                                 *dst++ = c;
 301                 } while (--len);
 302         }
 303         strbuf_setlen(buf, dst - buf->buf);
 304         return 1;
 305 }
 306
 307 static int crlf_to_worktree(const char *path, const char *src, size_t len,
 308                             struct strbuf *buf, enum crlf_action crlf_action)
 309 {
 310         char *to_free = NULL;
 311         struct text_stat stats;
 312
 313         if (!len || output_eol(crlf_action) != EOL_CRLF)
 314                 return 0;
 315
 316         gather_stats(src, len, &stats);
 317
 318         /* No "naked" LF? Nothing to convert, regardless. */
 319         if (!stats.lonelf)
 320                 return 0;
 321
 322         if (crlf_action == CRLF_AUTO || crlf_action == CRLF_AUTO_INPUT || crlf_action == CRLF_AUTO_CRLF) {
 323                 if (crlf_action == CRLF_AUTO_INPUT || crlf_action == CRLF_AUTO_CRLF) {
 324                         /* If we have any CR or CRLF line endings, we do not touch it */
 325                         /* This is the new safer autocrlf-handling */
 326                         if (stats.lonecr || stats.crlf )
 327                                 return 0;
 328                 }
 329
 330                 if (convert_is_binary(len, &stats))
 331                         return 0;
 332         }
 333
 334         /* are we "faking" in place editing ? */
 335         if (src == buf->buf)
 336                 to_free = strbuf_detach(buf, NULL);
 337
 338         strbuf_grow(buf, len + stats.lonelf);
 339         for (;;) {
 340                 const char *nl = memchr(src, '\n', len);
 341                 if (!nl)
 342                         break;
 343                 if (nl > src && nl[-1] == '\r') {
 344                         strbuf_add(buf, src, nl + 1 - src);
 345                 } else {
 346                         strbuf_add(buf, src, nl - src);
 347                         strbuf_addstr(buf, "\r\n");
 348                 }
 349                 len -= nl + 1 - src;
 350                 src  = nl + 1;
 351         }
 352         strbuf_add(buf, src, len);
 353
 354         free(to_free);
 355         return 1;
 356 }
 357
 358 struct filter_params {
 359         const char *src;
 360         unsigned long size;
 361         int fd;
 362         const char *cmd;
 363         const char *path;
 364 };
 365
 366 static int filter_buffer_or_fd(int in, int out, void *data)
 367 {
 368         /*
 369          * Spawn cmd and feed the buffer contents through its stdin.
 370          */
 371         struct child_process child_process = CHILD_PROCESS_INIT;
 372         struct filter_params *params = (struct filter_params *)data;
 373         int write_err, status;
 374         const char *argv[] = { NULL, NULL };
 375
 376         /* apply % substitution to cmd */
 377         struct strbuf cmd = STRBUF_INIT;
 378         struct strbuf path = STRBUF_INIT;
 379         struct strbuf_expand_dict_entry dict[] = {
 380                 { "f", NULL, },
 381                 { NULL, NULL, },
 382         };
 383
 384         /* quote the path to preserve spaces, etc. */
 385         sq_quote_buf(&path, params->path);
 386         dict[0].value = path.buf;
 387
 388         /* expand all %f with the quoted path */
 389         strbuf_expand(&cmd, params->cmd, strbuf_expand_dict_cb, &dict);
 390         strbuf_release(&path);
 391
 392         argv[0] = cmd.buf;
 393
 394         child_process.argv = argv;
 395         child_process.use_shell = 1;
 396         child_process.in = -1;
 397         child_process.out = out;
 398
 399         if (start_command(&child_process))
 400                 return error("cannot fork to run external filter %s", params->cmd);
 401
 402         sigchain_push(SIGPIPE, SIG_IGN);
 403
 404         if (params->src) {
 405                 write_err = (write_in_full(child_process.in,
 406                                            params->src, params->size) < 0);
 407                 if (errno == EPIPE)
 408                         write_err = 0;
 409         } else {
 410                 write_err = copy_fd(params->fd, child_process.in);
 411                 if (write_err == COPY_WRITE_ERROR && errno == EPIPE)
 412                         write_err = 0;
 413         }
 414
 415         if (close(child_process.in))
 416                 write_err = 1;
 417         if (write_err)
 418                 error("cannot feed the input to external filter %s", params->cmd);
 419
 420         sigchain_pop(SIGPIPE);
 421
 422         status = finish_command(&child_process);
 423         if (status)
 424                 error("external filter %s failed %d", params->cmd, status);
 425
 426         strbuf_release(&cmd);
 427         return (write_err || status);
 428 }
 429
 430 static int apply_filter(const char *path, const char *src, size_t len, int fd,
 431                         struct strbuf *dst, const char *cmd)
 432 {
 433         /*
 434          * Create a pipeline to have the command filter the buffer's
 435          * contents.
 436          *
 437          * (child --> cmd) --> us
 438          */
 439         int ret = 1;
 440         struct strbuf nbuf = STRBUF_INIT;
 441         struct async async;
 442         struct filter_params params;
 443
 444         if (!cmd || !*cmd)
 445                 return 0;
 446
 447         if (!dst)
 448                 return 1;
 449
 450         memset(&async, 0, sizeof(async));
 451         async.proc = filter_buffer_or_fd;
 452         async.data = &params;
 453         async.out = -1;
 454         params.src = src;
 455         params.size = len;
 456         params.fd = fd;
 457         params.cmd = cmd;
 458         params.path = path;
 459
 460         fflush(NULL);
 461         if (start_async(&async))
 462                 return 0;       /* error was already reported */
 463
 464         if (strbuf_read(&nbuf, async.out, len) < 0) {
 465                 error("read from external filter %s failed", cmd);
 466                 ret = 0;
 467         }
 468         if (close(async.out)) {
 469                 error("read from external filter %s failed", cmd);
 470                 ret = 0;
 471         }
 472         if (finish_async(&async)) {
 473                 error("external filter %s failed", cmd);
 474                 ret = 0;
 475         }
 476
 477         if (ret) {
 478                 strbuf_swap(dst, &nbuf);
 479         }
 480         strbuf_release(&nbuf);
 481         return ret;
 482 }
 483
 484 static struct convert_driver {
 485         const char *name;
 486         struct convert_driver *next;
 487         const char *smudge;
 488         const char *clean;
 489         int required;
 490 } *user_convert, **user_convert_tail;
 491
 492 static int read_convert_config(const char *var, const char *value, void *cb)
 493 {
 494         const char *key, *name;
 495         int namelen;
 496         struct convert_driver *drv;
 497
 498         /*
 499          * External conversion drivers are configured using
 500          * "filter.<name>.variable".
 501          */
 502         if (parse_config_key(var, "filter", &name, &namelen, &key) < 0 || !name)
 503                 return 0;
 504         for (drv = user_convert; drv; drv = drv->next)
 505                 if (!strncmp(drv->name, name, namelen) && !drv->name[namelen])
 506                         break;
 507         if (!drv) {
 508                 drv = xcalloc(1, sizeof(struct convert_driver));
 509                 drv->name = xmemdupz(name, namelen);
 510                 *user_convert_tail = drv;
 511                 user_convert_tail = &(drv->next);
 512         }
 513
 514         /*
 515          * filter.<name>.smudge and filter.<name>.clean specifies
 516          * the command line:
 517          *
 518          *      command-line
 519          *
 520          * The command-line will not be interpolated in any way.
 521          */
 522
 523         if (!strcmp("smudge", key))
 524                 return git_config_string(&drv->smudge, var, value);
 525
 526         if (!strcmp("clean", key))
 527                 return git_config_string(&drv->clean, var, value);
 528
 529         if (!strcmp("required", key)) {
 530                 drv->required = git_config_bool(var, value);
 531                 return 0;
 532         }
 533
 534         return 0;
 535 }
 536
 537 static int count_ident(const char *cp, unsigned long size)
 538 {
 539         /*
 540          * "$Id: 0000000000000000000000000000000000000000 $" <=> "$Id$"
 541          */
 542         int cnt = 0;
 543         char ch;
 544
 545         while (size) {
 546                 ch = *cp++;
 547                 size--;
 548                 if (ch != '$')
 549                         continue;
 550                 if (size < 3)
 551                         break;
 552                 if (memcmp("Id", cp, 2))
 553                         continue;
 554                 ch = cp[2];
 555                 cp += 3;
 556                 size -= 3;
 557                 if (ch == '$')
 558                         cnt++; /* $Id$ */
 559                 if (ch != ':')
 560                         continue;
 561
 562                 /*
 563                  * "$Id: ... "; scan up to the closing dollar sign and discard.
 564                  */
 565                 while (size) {
 566                         ch = *cp++;
 567                         size--;
 568                         if (ch == '$') {
 569                                 cnt++;
 570                                 break;
 571                         }
 572                         if (ch == '\n')
 573                                 break;
 574                 }
 575         }
 576         return cnt;
 577 }
 578
 579 static int ident_to_git(const char *path, const char *src, size_t len,
 580                         struct strbuf *buf, int ident)
 581 {
 582         char *dst, *dollar;
 583
 584         if (!ident || (src && !count_ident(src, len)))
 585                 return 0;
 586
 587         if (!buf)
 588                 return 1;
 589
 590         /* only grow if not in place */
 591         if (strbuf_avail(buf) + buf->len < len)
 592                 strbuf_grow(buf, len - buf->len);
 593         dst = buf->buf;
 594         for (;;) {
 595                 dollar = memchr(src, '$', len);
 596                 if (!dollar)
 597                         break;
 598                 memmove(dst, src, dollar + 1 - src);
 599                 dst += dollar + 1 - src;
 600                 len -= dollar + 1 - src;
 601                 src  = dollar + 1;
 602
 603                 if (len > 3 && !memcmp(src, "Id:", 3)) {
 604                         dollar = memchr(src + 3, '$', len - 3);
 605                         if (!dollar)
 606                                 break;
 607                         if (memchr(src + 3, '\n', dollar - src - 3)) {
 608                                 /* Line break before the next dollar. */
 609                                 continue;
 610                         }
 611
 612                         memcpy(dst, "Id$", 3);
 613                         dst += 3;
 614                         len -= dollar + 1 - src;
 615                         src  = dollar + 1;
 616                 }
 617         }
 618         memmove(dst, src, len);
 619         strbuf_setlen(buf, dst + len - buf->buf);
 620         return 1;
 621 }
 622
 623 static int ident_to_worktree(const char *path, const char *src, size_t len,
 624                              struct strbuf *buf, int ident)
 625 {
 626         unsigned char sha1[20];
 627         char *to_free = NULL, *dollar, *spc;
 628         int cnt;
 629
 630         if (!ident)
 631                 return 0;
 632
 633         cnt = count_ident(src, len);
 634         if (!cnt)
 635                 return 0;
 636
 637         /* are we "faking" in place editing ? */
 638         if (src == buf->buf)
 639                 to_free = strbuf_detach(buf, NULL);
 640         hash_sha1_file(src, len, "blob", sha1);
 641
 642         strbuf_grow(buf, len + cnt * 43);
 643         for (;;) {
 644                 /* step 1: run to the next '$' */
 645                 dollar = memchr(src, '$', len);
 646                 if (!dollar)
 647                         break;
 648                 strbuf_add(buf, src, dollar + 1 - src);
 649                 len -= dollar + 1 - src;
 650                 src  = dollar + 1;
 651
 652                 /* step 2: does it looks like a bit like Id:xxx$ or Id$ ? */
 653                 if (len < 3 || memcmp("Id", src, 2))
 654                         continue;
 655
 656                 /* step 3: skip over Id$ or Id:xxxxx$ */
 657                 if (src[2] == '$') {
 658                         src += 3;
 659                         len -= 3;
 660                 } else if (src[2] == ':') {
 661                         /*
 662                          * It's possible that an expanded Id has crept its way into the
 663                          * repository, we cope with that by stripping the expansion out.
 664                          * This is probably not a good idea, since it will cause changes
 665                          * on checkout, which won't go away by stash, but let's keep it
 666                          * for git-style ids.
 667                          */
 668                         dollar = memchr(src + 3, '$', len - 3);
 669                         if (!dollar) {
 670                                 /* incomplete keyword, no more '$', so just quit the loop */
 671                                 break;
 672                         }
 673
 674                         if (memchr(src + 3, '\n', dollar - src - 3)) {
 675                                 /* Line break before the next dollar. */
 676                                 continue;
 677                         }
 678
 679                         spc = memchr(src + 4, ' ', dollar - src - 4);
 680                         if (spc && spc < dollar-1) {
 681                                 /* There are spaces in unexpected places.
 682                                  * This is probably an id from some other
 683                                  * versioning system. Keep it for now.
 684                                  */
 685                                 continue;
 686                         }
 687
 688                         len -= dollar + 1 - src;
 689                         src  = dollar + 1;
 690                 } else {
 691                         /* it wasn't a "Id$" or "Id:xxxx$" */
 692                         continue;
 693                 }
 694
 695                 /* step 4: substitute */
 696                 strbuf_addstr(buf, "Id: ");
 697                 strbuf_add(buf, sha1_to_hex(sha1), 40);
 698                 strbuf_addstr(buf, " $");
 699         }
 700         strbuf_add(buf, src, len);
 701
 702         free(to_free);
 703         return 1;
 704 }
 705
 706 static enum crlf_action git_path_check_crlf(struct git_attr_check *check)
 707 {
 708         const char *value = check->value;
 709
 710         if (ATTR_TRUE(value))
 711                 return CRLF_TEXT;
 712         else if (ATTR_FALSE(value))
 713                 return CRLF_BINARY;
 714         else if (ATTR_UNSET(value))
 715                 ;
 716         else if (!strcmp(value, "input"))
 717                 return CRLF_TEXT_INPUT;
 718         else if (!strcmp(value, "auto"))
 719                 return CRLF_AUTO;
 720         return CRLF_UNDEFINED;
 721 }
 722
 723 static enum eol git_path_check_eol(struct git_attr_check *check)
 724 {
 725         const char *value = check->value;
 726
 727         if (ATTR_UNSET(value))
 728                 ;
 729         else if (!strcmp(value, "lf"))
 730                 return EOL_LF;
 731         else if (!strcmp(value, "crlf"))
 732                 return EOL_CRLF;
 733         return EOL_UNSET;
 734 }
 735
 736 static struct convert_driver *git_path_check_convert(struct git_attr_check *check)
 737 {
 738         const char *value = check->value;
 739         struct convert_driver *drv;
 740
 741         if (ATTR_TRUE(value) || ATTR_FALSE(value) || ATTR_UNSET(value))
 742                 return NULL;
 743         for (drv = user_convert; drv; drv = drv->next)
 744                 if (!strcmp(value, drv->name))
 745                         return drv;
 746         return NULL;
 747 }
 748
 749 static int git_path_check_ident(struct git_attr_check *check)
 750 {
 751         const char *value = check->value;
 752
 753         return !!ATTR_TRUE(value);
 754 }
 755
 756 struct conv_attrs {
 757         struct convert_driver *drv;
 758         enum crlf_action attr_action; /* What attr says */
 759         enum crlf_action crlf_action; /* When no attr is set, use core.autocrlf */
 760         int ident;
 761 };
 762
 763 static const char *conv_attr_name[] = {
 764         "crlf", "ident", "filter", "eol", "text",
 765 };
 766 #define NUM_CONV_ATTRS ARRAY_SIZE(conv_attr_name)
 767
 768 static void convert_attrs(struct conv_attrs *ca, const char *path)
 769 {
 770         int i;
 771         static struct git_attr_check ccheck[NUM_CONV_ATTRS];
 772
 773         if (!ccheck[0].attr) {
 774                 for (i = 0; i < NUM_CONV_ATTRS; i++)
 775                         ccheck[i].attr = git_attr(conv_attr_name[i]);
 776                 user_convert_tail = &user_convert;
 777                 git_config(read_convert_config, NULL);
 778         }
 779
 780         if (!git_check_attr(path, NUM_CONV_ATTRS, ccheck)) {
 781                 ca->crlf_action = git_path_check_crlf(ccheck + 4);
 782                 if (ca->crlf_action == CRLF_UNDEFINED)
 783                         ca->crlf_action = git_path_check_crlf(ccheck + 0);
 784                 ca->attr_action = ca->crlf_action;
 785                 ca->ident = git_path_check_ident(ccheck + 1);
 786                 ca->drv = git_path_check_convert(ccheck + 2);
 787                 if (ca->crlf_action != CRLF_BINARY) {
 788                         enum eol eol_attr = git_path_check_eol(ccheck + 3);
 789                         if (eol_attr == EOL_LF)
 790                                 ca->crlf_action = CRLF_TEXT_INPUT;
 791                         else if (eol_attr == EOL_CRLF)
 792                                 ca->crlf_action = CRLF_TEXT_CRLF;
 793                 }
 794                 ca->attr_action = ca->crlf_action;
 795         } else {
 796                 ca->drv = NULL;
 797                 ca->crlf_action = CRLF_UNDEFINED;
 798                 ca->ident = 0;
 799         }
 800         if (ca->crlf_action == CRLF_TEXT)
 801                 ca->crlf_action = text_eol_is_crlf() ? CRLF_TEXT_CRLF : CRLF_TEXT_INPUT;
 802         if (ca->crlf_action == CRLF_UNDEFINED && auto_crlf == AUTO_CRLF_FALSE)
 803                 ca->crlf_action = CRLF_BINARY;
 804         if (ca->crlf_action == CRLF_UNDEFINED && auto_crlf == AUTO_CRLF_TRUE)
 805                 ca->crlf_action = CRLF_AUTO_CRLF;
 806         if (ca->crlf_action == CRLF_UNDEFINED && auto_crlf == AUTO_CRLF_INPUT)
 807                 ca->crlf_action = CRLF_AUTO_INPUT;
 808 }
 809
 810 int would_convert_to_git_filter_fd(const char *path)
 811 {
 812         struct conv_attrs ca;
 813
 814         convert_attrs(&ca, path);
 815         if (!ca.drv)
 816                 return 0;
 817
 818         /*
 819          * Apply a filter to an fd only if the filter is required to succeed.
 820          * We must die if the filter fails, because the original data before
 821          * filtering is not available.
 822          */
 823         if (!ca.drv->required)
 824                 return 0;
 825
 826         return apply_filter(path, NULL, 0, -1, NULL, ca.drv->clean);
 827 }
 828
 829 const char *get_convert_attr_ascii(const char *path)
 830 {
 831         struct conv_attrs ca;
 832
 833         convert_attrs(&ca, path);
 834         switch (ca.attr_action) {
 835         case CRLF_UNDEFINED:
 836                 return "";
 837         case CRLF_BINARY:
 838                 return "-text";
 839         case CRLF_TEXT:
 840                 return "text";
 841         case CRLF_TEXT_INPUT:
 842                 return "text eol=lf";
 843         case CRLF_TEXT_CRLF:
 844                 return "text eol=crlf";
 845         case CRLF_AUTO:
 846                 return "text=auto";
 847         case CRLF_AUTO_CRLF:
 848                 return "text=auto eol=crlf"; /* This is not supported yet */
 849         case CRLF_AUTO_INPUT:
 850                 return "text=auto eol=lf"; /* This is not supported yet */
 851         }
 852         return "";
 853 }
 854
 855 int convert_to_git(const char *path, const char *src, size_t len,
 856                    struct strbuf *dst, enum safe_crlf checksafe)
 857 {
 858         int ret = 0;
 859         const char *filter = NULL;
 860         int required = 0;
 861         struct conv_attrs ca;
 862
 863         convert_attrs(&ca, path);
 864         if (ca.drv) {
 865                 filter = ca.drv->clean;
 866                 required = ca.drv->required;
 867         }
 868
 869         ret |= apply_filter(path, src, len, -1, dst, filter);
 870         if (!ret && required)
 871                 die("%s: clean filter '%s' failed", path, ca.drv->name);
 872
 873         if (ret && dst) {
 874                 src = dst->buf;
 875                 len = dst->len;
 876         }
 877         ret |= crlf_to_git(path, src, len, dst, ca.crlf_action, checksafe);
 878         if (ret && dst) {
 879                 src = dst->buf;
 880                 len = dst->len;
 881         }
 882         return ret | ident_to_git(path, src, len, dst, ca.ident);
 883 }
 884
 885 void convert_to_git_filter_fd(const char *path, int fd, struct strbuf *dst,
 886                               enum safe_crlf checksafe)
 887 {
 888         struct conv_attrs ca;
 889         convert_attrs(&ca, path);
 890
 891         assert(ca.drv);
 892         assert(ca.drv->clean);
 893
 894         if (!apply_filter(path, NULL, 0, fd, dst, ca.drv->clean))
 895                 die("%s: clean filter '%s' failed", path, ca.drv->name);
 896
 897         crlf_to_git(path, dst->buf, dst->len, dst, ca.crlf_action, checksafe);
 898         ident_to_git(path, dst->buf, dst->len, dst, ca.ident);
 899 }
 900
 901 static int convert_to_working_tree_internal(const char *path, const char *src,
 902                                             size_t len, struct strbuf *dst,
 903                                             int normalizing)
 904 {
 905         int ret = 0, ret_filter = 0;
 906         const char *filter = NULL;
 907         int required = 0;
 908         struct conv_attrs ca;
 909
 910         convert_attrs(&ca, path);
 911         if (ca.drv) {
 912                 filter = ca.drv->smudge;
 913                 required = ca.drv->required;
 914         }
 915
 916         ret |= ident_to_worktree(path, src, len, dst, ca.ident);
 917         if (ret) {
 918                 src = dst->buf;
 919                 len = dst->len;
 920         }
 921         /*
 922          * CRLF conversion can be skipped if normalizing, unless there
 923          * is a smudge filter.  The filter might expect CRLFs.
 924          */
 925         if (filter || !normalizing) {
 926                 ret |= crlf_to_worktree(path, src, len, dst, ca.crlf_action);
 927                 if (ret) {
 928                         src = dst->buf;
 929                         len = dst->len;
 930                 }
 931         }
 932
 933         ret_filter = apply_filter(path, src, len, -1, dst, filter);
 934         if (!ret_filter && required)
 935                 die("%s: smudge filter %s failed", path, ca.drv->name);
 936
 937         return ret | ret_filter;
 938 }
 939
 940 int convert_to_working_tree(const char *path, const char *src, size_t len, struct strbuf *dst)
 941 {
 942         return convert_to_working_tree_internal(path, src, len, dst, 0);
 943 }
 944
 945 int renormalize_buffer(const char *path, const char *src, size_t len, struct strbuf *dst)
 946 {
 947         int ret = convert_to_working_tree_internal(path, src, len, dst, 1);
 948         if (ret) {
 949                 src = dst->buf;
 950                 len = dst->len;
 951         }
 952         return ret | convert_to_git(path, src, len, dst, SAFE_CRLF_FALSE);
 953 }
 954
 955 /*****************************************************************
 956  *
 957  * Streaming conversion support
 958  *
 959  *****************************************************************/
 960
 961 typedef int (*filter_fn)(struct stream_filter *,
 962                          const char *input, size_t *isize_p,
 963                          char *output, size_t *osize_p);
 964 typedef void (*free_fn)(struct stream_filter *);
 965
 966 struct stream_filter_vtbl {
 967         filter_fn filter;
 968         free_fn free;
 969 };
 970
 971 struct stream_filter {
 972         struct stream_filter_vtbl *vtbl;
 973 };
 974
 975 static int null_filter_fn(struct stream_filter *filter,
 976                           const char *input, size_t *isize_p,
 977                           char *output, size_t *osize_p)
 978 {
 979         size_t count;
 980
 981         if (!input)
 982                 return 0; /* we do not keep any states */
 983         count = *isize_p;
 984         if (*osize_p < count)
 985                 count = *osize_p;
 986         if (count) {
 987                 memmove(output, input, count);
 988                 *isize_p -= count;
 989                 *osize_p -= count;
 990         }
 991         return 0;
 992 }
 993
 994 static void null_free_fn(struct stream_filter *filter)
 995 {
 996         ; /* nothing -- null instances are shared */
 997 }
 998
 999 static struct stream_filter_vtbl null_vtbl = {
1000         null_filter_fn,
1001         null_free_fn,
1002 };
1003
1004 static struct stream_filter null_filter_singleton = {
1005         &null_vtbl,
1006 };
1007
1008 int is_null_stream_filter(struct stream_filter *filter)
1009 {
1010         return filter == &null_filter_singleton;
1011 }
1012
1013
1014 /*
1015  * LF-to-CRLF filter
1016  */
1017
1018 struct lf_to_crlf_filter {
1019         struct stream_filter filter;
1020         unsigned has_held:1;
1021         char held;
1022 };
1023
1024 static int lf_to_crlf_filter_fn(struct stream_filter *filter,
1025                                 const char *input, size_t *isize_p,
1026                                 char *output, size_t *osize_p)
1027 {
1028         size_t count, o = 0;
1029         struct lf_to_crlf_filter *lf_to_crlf = (struct lf_to_crlf_filter *)filter;
1030
1031         /*
1032          * We may be holding onto the CR to see if it is followed by a
1033          * LF, in which case we would need to go to the main loop.
1034          * Otherwise, just emit it to the output stream.
1035          */
1036         if (lf_to_crlf->has_held && (lf_to_crlf->held != '\r' || !input)) {
1037                 output[o++] = lf_to_crlf->held;
1038                 lf_to_crlf->has_held = 0;
1039         }
1040
1041         /* We are told to drain */
1042         if (!input) {
1043                 *osize_p -= o;
1044                 return 0;
1045         }
1046
1047         count = *isize_p;
1048         if (count || lf_to_crlf->has_held) {
1049                 size_t i;
1050                 int was_cr = 0;
1051
1052                 if (lf_to_crlf->has_held) {
1053                         was_cr = 1;
1054                         lf_to_crlf->has_held = 0;
1055                 }
1056
1057                 for (i = 0; o < *osize_p && i < count; i++) {
1058                         char ch = input[i];
1059
1060                         if (ch == '\n') {
1061                                 output[o++] = '\r';
1062                         } else if (was_cr) {
1063                                 /*
1064                                  * Previous round saw CR and it is not followed
1065                                  * by a LF; emit the CR before processing the
1066                                  * current character.
1067                                  */
1068                                 output[o++] = '\r';
1069                         }
1070
1071                         /*
1072                          * We may have consumed the last output slot,
1073                          * in which case we need to break out of this
1074                          * loop; hold the current character before
1075                          * returning.
1076                          */
1077                         if (*osize_p <= o) {
1078                                 lf_to_crlf->has_held = 1;
1079                                 lf_to_crlf->held = ch;
1080                                 continue; /* break but increment i */
1081                         }
1082
1083                         if (ch == '\r') {
1084                                 was_cr = 1;
1085                                 continue;
1086                         }
1087
1088                         was_cr = 0;
1089                         output[o++] = ch;
1090                 }
1091
1092                 *osize_p -= o;
1093                 *isize_p -= i;
1094
1095                 if (!lf_to_crlf->has_held && was_cr) {
1096                         lf_to_crlf->has_held = 1;
1097                         lf_to_crlf->held = '\r';
1098                 }
1099         }
1100         return 0;
1101 }
1102
1103 static void lf_to_crlf_free_fn(struct stream_filter *filter)
1104 {
1105         free(filter);
1106 }
1107
1108 static struct stream_filter_vtbl lf_to_crlf_vtbl = {
1109         lf_to_crlf_filter_fn,
1110         lf_to_crlf_free_fn,
1111 };
1112
1113 static struct stream_filter *lf_to_crlf_filter(void)
1114 {
1115         struct lf_to_crlf_filter *lf_to_crlf = xcalloc(1, sizeof(*lf_to_crlf));
1116
1117         lf_to_crlf->filter.vtbl = &lf_to_crlf_vtbl;
1118         return (struct stream_filter *)lf_to_crlf;
1119 }
1120
1121 /*
1122  * Cascade filter
1123  */
1124 #define FILTER_BUFFER 1024
1125 struct cascade_filter {
1126         struct stream_filter filter;
1127         struct stream_filter *one;
1128         struct stream_filter *two;
1129         char buf[FILTER_BUFFER];
1130         int end, ptr;
1131 };
1132
1133 static int cascade_filter_fn(struct stream_filter *filter,
1134                              const char *input, size_t *isize_p,
1135                              char *output, size_t *osize_p)
1136 {
1137         struct cascade_filter *cas = (struct cascade_filter *) filter;
1138         size_t filled = 0;
1139         size_t sz = *osize_p;
1140         size_t to_feed, remaining;
1141
1142         /*
1143          * input -- (one) --> buf -- (two) --> output
1144          */
1145         while (filled < sz) {
1146                 remaining = sz - filled;
1147
1148                 /* do we already have something to feed two with? */
1149                 if (cas->ptr < cas->end) {
1150                         to_feed = cas->end - cas->ptr;
1151                         if (stream_filter(cas->two,
1152                                           cas->buf + cas->ptr, &to_feed,
1153                                           output + filled, &remaining))
1154                                 return -1;
1155                         cas->ptr += (cas->end - cas->ptr) - to_feed;
1156                         filled = sz - remaining;
1157                         continue;
1158                 }
1159
1160                 /* feed one from upstream and have it emit into our buffer */
1161                 to_feed = input ? *isize_p : 0;
1162                 if (input && !to_feed)
1163                         break;
1164                 remaining = sizeof(cas->buf);
1165                 if (stream_filter(cas->one,
1166                                   input, &to_feed,
1167                                   cas->buf, &remaining))
1168                         return -1;
1169                 cas->end = sizeof(cas->buf) - remaining;
1170                 cas->ptr = 0;
1171                 if (input) {
1172                         size_t fed = *isize_p - to_feed;
1173                         *isize_p -= fed;
1174                         input += fed;
1175                 }
1176
1177                 /* do we know that we drained one completely? */
1178                 if (input || cas->end)
1179                         continue;
1180
1181                 /* tell two to drain; we have nothing more to give it */
1182                 to_feed = 0;
1183                 remaining = sz - filled;
1184                 if (stream_filter(cas->two,
1185                                   NULL, &to_feed,
1186                                   output + filled, &remaining))
1187                         return -1;
1188                 if (remaining == (sz - filled))
1189                         break; /* completely drained two */
1190                 filled = sz - remaining;
1191         }
1192         *osize_p -= filled;
1193         return 0;
1194 }
1195
1196 static void cascade_free_fn(struct stream_filter *filter)
1197 {
1198         struct cascade_filter *cas = (struct cascade_filter *)filter;
1199         free_stream_filter(cas->one);
1200         free_stream_filter(cas->two);
1201         free(filter);
1202 }
1203
1204 static struct stream_filter_vtbl cascade_vtbl = {
1205         cascade_filter_fn,
1206         cascade_free_fn,
1207 };
1208
1209 static struct stream_filter *cascade_filter(struct stream_filter *one,
1210                                             struct stream_filter *two)
1211 {
1212         struct cascade_filter *cascade;
1213
1214         if (!one || is_null_stream_filter(one))
1215                 return two;
1216         if (!two || is_null_stream_filter(two))
1217                 return one;
1218
1219         cascade = xmalloc(sizeof(*cascade));
1220         cascade->one = one;
1221         cascade->two = two;
1222         cascade->end = cascade->ptr = 0;
1223         cascade->filter.vtbl = &cascade_vtbl;
1224         return (struct stream_filter *)cascade;
1225 }
1226
1227 /*
1228  * ident filter
1229  */
1230 #define IDENT_DRAINING (-1)
1231 #define IDENT_SKIPPING (-2)
1232 struct ident_filter {
1233         struct stream_filter filter;
1234         struct strbuf left;
1235         int state;
1236         char ident[45]; /* ": x40 $" */
1237 };
1238
1239 static int is_foreign_ident(const char *str)
1240 {
1241         int i;
1242
1243         if (!skip_prefix(str, "$Id: ", &str))
1244                 return 0;
1245         for (i = 0; str[i]; i++) {
1246                 if (isspace(str[i]) && str[i+1] != '$')
1247                         return 1;
1248         }
1249         return 0;
1250 }
1251
1252 static void ident_drain(struct ident_filter *ident, char **output_p, size_t *osize_p)
1253 {
1254         size_t to_drain = ident->left.len;
1255
1256         if (*osize_p < to_drain)
1257                 to_drain = *osize_p;
1258         if (to_drain) {
1259                 memcpy(*output_p, ident->left.buf, to_drain);
1260                 strbuf_remove(&ident->left, 0, to_drain);
1261                 *output_p += to_drain;
1262                 *osize_p -= to_drain;
1263         }
1264         if (!ident->left.len)
1265                 ident->state = 0;
1266 }
1267
1268 static int ident_filter_fn(struct stream_filter *filter,
1269                            const char *input, size_t *isize_p,
1270                            char *output, size_t *osize_p)
1271 {
1272         struct ident_filter *ident = (struct ident_filter *)filter;
1273         static const char head[] = "$Id";
1274
1275         if (!input) {
1276                 /* drain upon eof */
1277                 switch (ident->state) {
1278                 default:
1279                         strbuf_add(&ident->left, head, ident->state);
1280                 case IDENT_SKIPPING:
1281                         /* fallthru */
1282                 case IDENT_DRAINING:
1283                         ident_drain(ident, &output, osize_p);
1284                 }
1285                 return 0;
1286         }
1287
1288         while (*isize_p || (ident->state == IDENT_DRAINING)) {
1289                 int ch;
1290
1291                 if (ident->state == IDENT_DRAINING) {
1292                         ident_drain(ident, &output, osize_p);
1293                         if (!*osize_p)
1294                                 break;
1295                         continue;
1296                 }
1297
1298                 ch = *(input++);
1299                 (*isize_p)--;
1300
1301                 if (ident->state == IDENT_SKIPPING) {
1302                         /*
1303                          * Skipping until '$' or LF, but keeping them
1304                          * in case it is a foreign ident.
1305                          */
1306                         strbuf_addch(&ident->left, ch);
1307                         if (ch != '\n' && ch != '$')
1308                                 continue;
1309                         if (ch == '$' && !is_foreign_ident(ident->left.buf)) {
1310                                 strbuf_setlen(&ident->left, sizeof(head) - 1);
1311                                 strbuf_addstr(&ident->left, ident->ident);
1312                         }
1313                         ident->state = IDENT_DRAINING;
1314                         continue;
1315                 }
1316
1317                 if (ident->state < sizeof(head) &&
1318                     head[ident->state] == ch) {
1319                         ident->state++;
1320                         continue;
1321                 }
1322
1323                 if (ident->state)
1324                         strbuf_add(&ident->left, head, ident->state);
1325                 if (ident->state == sizeof(head) - 1) {
1326                         if (ch != ':' && ch != '$') {
1327                                 strbuf_addch(&ident->left, ch);
1328                                 ident->state = 0;
1329                                 continue;
1330                         }
1331
1332                         if (ch == ':') {
1333                                 strbuf_addch(&ident->left, ch);
1334                                 ident->state = IDENT_SKIPPING;
1335                         } else {
1336                                 strbuf_addstr(&ident->left, ident->ident);
1337                                 ident->state = IDENT_DRAINING;
1338                         }
1339                         continue;
1340                 }
1341
1342                 strbuf_addch(&ident->left, ch);
1343                 ident->state = IDENT_DRAINING;
1344         }
1345         return 0;
1346 }
1347
1348 static void ident_free_fn(struct stream_filter *filter)
1349 {
1350         struct ident_filter *ident = (struct ident_filter *)filter;
1351         strbuf_release(&ident->left);
1352         free(filter);
1353 }
1354
1355 static struct stream_filter_vtbl ident_vtbl = {
1356         ident_filter_fn,
1357         ident_free_fn,
1358 };
1359
1360 static struct stream_filter *ident_filter(const unsigned char *sha1)
1361 {
1362         struct ident_filter *ident = xmalloc(sizeof(*ident));
1363
1364         xsnprintf(ident->ident, sizeof(ident->ident),
1365                   ": %s $", sha1_to_hex(sha1));
1366         strbuf_init(&ident->left, 0);
1367         ident->filter.vtbl = &ident_vtbl;
1368         ident->state = 0;
1369         return (struct stream_filter *)ident;
1370 }
1371
1372 /*
1373  * Return an appropriately constructed filter for the path, or NULL if
1374  * the contents cannot be filtered without reading the whole thing
1375  * in-core.
1376  *
1377  * Note that you would be crazy to set CRLF, smuge/clean or ident to a
1378  * large binary blob you would want us not to slurp into the memory!
1379  */
1380 struct stream_filter *get_stream_filter(const char *path, const unsigned char *sha1)
1381 {
1382         struct conv_attrs ca;
1383         enum crlf_action crlf_action;
1384         struct stream_filter *filter = NULL;
1385
1386         convert_attrs(&ca, path);
1387
1388         if (ca.drv && (ca.drv->smudge || ca.drv->clean))
1389                 return filter;
1390
1391         if (ca.ident)
1392                 filter = ident_filter(sha1);
1393
1394         crlf_action = ca.crlf_action;
1395
1396         if ((crlf_action == CRLF_BINARY) ||
1397                         crlf_action == CRLF_AUTO_INPUT ||
1398                         (crlf_action == CRLF_TEXT_INPUT))
1399                 filter = cascade_filter(filter, &null_filter_singleton);
1400
1401         else if (output_eol(crlf_action) == EOL_CRLF &&
1402                  !(crlf_action == CRLF_AUTO || crlf_action == CRLF_AUTO_CRLF))
1403                 filter = cascade_filter(filter, lf_to_crlf_filter());
1404
1405         return filter;
1406 }
1407
1408 void free_stream_filter(struct stream_filter *filter)
1409 {
1410         filter->vtbl->free(filter);
1411 }
1412
1413 int stream_filter(struct stream_filter *filter,
1414                   const char *input, size_t *isize_p,
1415                   char *output, size_t *osize_p)
1416 {
1417         return filter->vtbl->filter(filter, input, isize_p, output, osize_p);
1418 }