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_GUESS = -1,
  23         CRLF_BINARY = 0,
  24         CRLF_TEXT,
  25         CRLF_INPUT,
  26         CRLF_CRLF,
  27         CRLF_AUTO
  28 };
  29
  30 struct text_stat {
  31         /* NUL, CR, LF and CRLF counts */
  32         unsigned nul, cr, lf, crlf;
  33
  34         /* These are just approximations! */
  35         unsigned printable, nonprintable;
  36 };
  37
  38 static void gather_stats(const char *buf, unsigned long size, struct text_stat *stats)
  39 {
  40         unsigned long i;
  41
  42         memset(stats, 0, sizeof(*stats));
  43
  44         for (i = 0; i < size; i++) {
  45                 unsigned char c = buf[i];
  46                 if (c == '\r') {
  47                         stats->cr++;
  48                         if (i+1 < size && buf[i+1] == '\n')
  49                                 stats->crlf++;
  50                         continue;
  51                 }
  52                 if (c == '\n') {
  53                         stats->lf++;
  54                         continue;
  55                 }
  56                 if (c == 127)
  57                         /* DEL */
  58                         stats->nonprintable++;
  59                 else if (c < 32) {
  60                         switch (c) {
  61                                 /* BS, HT, ESC and FF */
  62                         case '\b': case '\t': case '\033': case '\014':
  63                                 stats->printable++;
  64                                 break;
  65                         case 0:
  66                                 stats->nul++;
  67                                 /* fall through */
  68                         default:
  69                                 stats->nonprintable++;
  70                         }
  71                 }
  72                 else
  73                         stats->printable++;
  74         }
  75
  76         /* If file ends with EOF then don't count this EOF as non-printable. */
  77         if (size >= 1 && buf[size-1] == '\032')
  78                 stats->nonprintable--;
  79 }
  80
  81 /*
  82  * The same heuristics as diff.c::mmfile_is_binary()
  83  * We treat files with bare CR as binary
  84  */
  85 static int convert_is_binary(unsigned long size, const struct text_stat *stats)
  86 {
  87         if (stats->cr != stats->crlf)
  88                 return 1;
  89         if (stats->nul)
  90                 return 1;
  91         if ((stats->printable >> 7) < stats->nonprintable)
  92                 return 1;
  93         return 0;
  94 }
  95
  96 static unsigned int gather_convert_stats(const char *data, unsigned long size)
  97 {
  98         struct text_stat stats;
  99         if (!data || !size)
 100                 return 0;
 101         gather_stats(data, size, &stats);
 102         if (convert_is_binary(size, &stats))
 103                 return CONVERT_STAT_BITS_BIN;
 104         else if (stats.crlf && stats.crlf == stats.lf)
 105                 return CONVERT_STAT_BITS_TXT_CRLF;
 106         else if (stats.crlf && stats.lf)
 107                 return CONVERT_STAT_BITS_TXT_CRLF | CONVERT_STAT_BITS_TXT_LF;
 108         else if (stats.lf)
 109                 return CONVERT_STAT_BITS_TXT_LF;
 110         else
 111                 return 0;
 112 }
 113
 114 static const char *gather_convert_stats_ascii(const char *data, unsigned long size)
 115 {
 116         unsigned int convert_stats = gather_convert_stats(data, size);
 117
 118         if (convert_stats & CONVERT_STAT_BITS_BIN)
 119                 return "-text";
 120         switch (convert_stats) {
 121         case CONVERT_STAT_BITS_TXT_LF:
 122                 return "lf";
 123         case CONVERT_STAT_BITS_TXT_CRLF:
 124                 return "crlf";
 125         case CONVERT_STAT_BITS_TXT_LF | CONVERT_STAT_BITS_TXT_CRLF:
 126                 return "mixed";
 127         default:
 128                 return "none";
 129         }
 130 }
 131
 132 const char *get_cached_convert_stats_ascii(const char *path)
 133 {
 134         const char *ret;
 135         unsigned long sz;
 136         void *data = read_blob_data_from_cache(path, &sz);
 137         ret = gather_convert_stats_ascii(data, sz);
 138         free(data);
 139         return ret;
 140 }
 141
 142 const char *get_wt_convert_stats_ascii(const char *path)
 143 {
 144         const char *ret = "";
 145         struct strbuf sb = STRBUF_INIT;
 146         if (strbuf_read_file(&sb, path, 0) >= 0)
 147                 ret = gather_convert_stats_ascii(sb.buf, sb.len);
 148         strbuf_release(&sb);
 149         return ret;
 150 }
 151
 152 static enum eol output_eol(enum crlf_action crlf_action)
 153 {
 154         switch (crlf_action) {
 155         case CRLF_BINARY:
 156                 return EOL_UNSET;
 157         case CRLF_CRLF:
 158                 return EOL_CRLF;
 159         case CRLF_INPUT:
 160                 return EOL_LF;
 161         case CRLF_GUESS:
 162                 if (!auto_crlf)
 163                         return EOL_UNSET;
 164                 /* fall through */
 165         case CRLF_TEXT:
 166         case CRLF_AUTO:
 167                 if (auto_crlf == AUTO_CRLF_TRUE)
 168                         return EOL_CRLF;
 169                 else if (auto_crlf == AUTO_CRLF_INPUT)
 170                         return EOL_LF;
 171                 else if (core_eol == EOL_UNSET)
 172                         return EOL_NATIVE;
 173         }
 174         return core_eol;
 175 }
 176
 177 static void check_safe_crlf(const char *path, enum crlf_action crlf_action,
 178                             struct text_stat *stats, enum safe_crlf checksafe)
 179 {
 180         if (!checksafe)
 181                 return;
 182
 183         if (output_eol(crlf_action) == EOL_LF) {
 184                 /*
 185                  * CRLFs would not be restored by checkout:
 186                  * check if we'd remove CRLFs
 187                  */
 188                 if (stats->crlf) {
 189                         if (checksafe == SAFE_CRLF_WARN)
 190                                 warning("CRLF will be replaced by LF in %s.\nThe file will have its original line endings in your working directory.", path);
 191                         else /* i.e. SAFE_CRLF_FAIL */
 192                                 die("CRLF would be replaced by LF in %s.", path);
 193                 }
 194         } else if (output_eol(crlf_action) == EOL_CRLF) {
 195                 /*
 196                  * CRLFs would be added by checkout:
 197                  * check if we have "naked" LFs
 198                  */
 199                 if (stats->lf != stats->crlf) {
 200                         if (checksafe == SAFE_CRLF_WARN)
 201                                 warning("LF will be replaced by CRLF in %s.\nThe file will have its original line endings in your working directory.", path);
 202                         else /* i.e. SAFE_CRLF_FAIL */
 203                                 die("LF would be replaced by CRLF in %s", path);
 204                 }
 205         }
 206 }
 207
 208 static int has_cr_in_index(const char *path)
 209 {
 210         unsigned long sz;
 211         void *data;
 212         int has_cr;
 213
 214         data = read_blob_data_from_cache(path, &sz);
 215         if (!data)
 216                 return 0;
 217         has_cr = memchr(data, '\r', sz) != NULL;
 218         free(data);
 219         return has_cr;
 220 }
 221
 222 static int crlf_to_git(const char *path, const char *src, size_t len,
 223                        struct strbuf *buf,
 224                        enum crlf_action crlf_action, enum safe_crlf checksafe)
 225 {
 226         struct text_stat stats;
 227         char *dst;
 228
 229         if (crlf_action == CRLF_BINARY ||
 230             (crlf_action == CRLF_GUESS && auto_crlf == AUTO_CRLF_FALSE) ||
 231             (src && !len))
 232                 return 0;
 233
 234         /*
 235          * If we are doing a dry-run and have no source buffer, there is
 236          * nothing to analyze; we must assume we would convert.
 237          */
 238         if (!buf && !src)
 239                 return 1;
 240
 241         gather_stats(src, len, &stats);
 242
 243         if (crlf_action == CRLF_AUTO || crlf_action == CRLF_GUESS) {
 244                 if (convert_is_binary(len, &stats))
 245                         return 0;
 246
 247                 if (crlf_action == CRLF_GUESS) {
 248                         /*
 249                          * If the file in the index has any CR in it, do not convert.
 250                          * This is the new safer autocrlf handling.
 251                          */
 252                         if (has_cr_in_index(path))
 253                                 return 0;
 254                 }
 255         }
 256
 257         check_safe_crlf(path, crlf_action, &stats, checksafe);
 258
 259         /* Optimization: No CR? Nothing to convert, regardless. */
 260         if (!stats.cr)
 261                 return 0;
 262
 263         /*
 264          * At this point all of our source analysis is done, and we are sure we
 265          * would convert. If we are in dry-run mode, we can give an answer.
 266          */
 267         if (!buf)
 268                 return 1;
 269
 270         /* only grow if not in place */
 271         if (strbuf_avail(buf) + buf->len < len)
 272                 strbuf_grow(buf, len - buf->len);
 273         dst = buf->buf;
 274         if (crlf_action == CRLF_AUTO || crlf_action == CRLF_GUESS) {
 275                 /*
 276                  * If we guessed, we already know we rejected a file with
 277                  * lone CR, and we can strip a CR without looking at what
 278                  * follow it.
 279                  */
 280                 do {
 281                         unsigned char c = *src++;
 282                         if (c != '\r')
 283                                 *dst++ = c;
 284                 } while (--len);
 285         } else {
 286                 do {
 287                         unsigned char c = *src++;
 288                         if (! (c == '\r' && (1 < len && *src == '\n')))
 289                                 *dst++ = c;
 290                 } while (--len);
 291         }
 292         strbuf_setlen(buf, dst - buf->buf);
 293         return 1;
 294 }
 295
 296 static int crlf_to_worktree(const char *path, const char *src, size_t len,
 297                             struct strbuf *buf, enum crlf_action crlf_action)
 298 {
 299         char *to_free = NULL;
 300         struct text_stat stats;
 301
 302         if (!len || output_eol(crlf_action) != EOL_CRLF)
 303                 return 0;
 304
 305         gather_stats(src, len, &stats);
 306
 307         /* No LF? Nothing to convert, regardless. */
 308         if (!stats.lf)
 309                 return 0;
 310
 311         /* Was it already in CRLF format? */
 312         if (stats.lf == stats.crlf)
 313                 return 0;
 314
 315         if (crlf_action == CRLF_AUTO || crlf_action == CRLF_GUESS) {
 316                 if (crlf_action == CRLF_GUESS) {
 317                         /* If we have any CR or CRLF line endings, we do not touch it */
 318                         /* This is the new safer autocrlf-handling */
 319                         if (stats.cr > 0 || stats.crlf > 0)
 320                                 return 0;
 321                 }
 322
 323                 if (convert_is_binary(len, &stats))
 324                         return 0;
 325         }
 326
 327         /* are we "faking" in place editing ? */
 328         if (src == buf->buf)
 329                 to_free = strbuf_detach(buf, NULL);
 330
 331         strbuf_grow(buf, len + stats.lf - stats.crlf);
 332         for (;;) {
 333                 const char *nl = memchr(src, '\n', len);
 334                 if (!nl)
 335                         break;
 336                 if (nl > src && nl[-1] == '\r') {
 337                         strbuf_add(buf, src, nl + 1 - src);
 338                 } else {
 339                         strbuf_add(buf, src, nl - src);
 340                         strbuf_addstr(buf, "\r\n");
 341                 }
 342                 len -= nl + 1 - src;
 343                 src  = nl + 1;
 344         }
 345         strbuf_add(buf, src, len);
 346
 347         free(to_free);
 348         return 1;
 349 }
 350
 351 struct filter_params {
 352         const char *src;
 353         unsigned long size;
 354         int fd;
 355         const char *cmd;
 356         const char *path;
 357 };
 358
 359 static int filter_buffer_or_fd(int in, int out, void *data)
 360 {
 361         /*
 362          * Spawn cmd and feed the buffer contents through its stdin.
 363          */
 364         struct child_process child_process = CHILD_PROCESS_INIT;
 365         struct filter_params *params = (struct filter_params *)data;
 366         int write_err, status;
 367         const char *argv[] = { NULL, NULL };
 368
 369         /* apply % substitution to cmd */
 370         struct strbuf cmd = STRBUF_INIT;
 371         struct strbuf path = STRBUF_INIT;
 372         struct strbuf_expand_dict_entry dict[] = {
 373                 { "f", NULL, },
 374                 { NULL, NULL, },
 375         };
 376
 377         /* quote the path to preserve spaces, etc. */
 378         sq_quote_buf(&path, params->path);
 379         dict[0].value = path.buf;
 380
 381         /* expand all %f with the quoted path */
 382         strbuf_expand(&cmd, params->cmd, strbuf_expand_dict_cb, &dict);
 383         strbuf_release(&path);
 384
 385         argv[0] = cmd.buf;
 386
 387         child_process.argv = argv;
 388         child_process.use_shell = 1;
 389         child_process.in = -1;
 390         child_process.out = out;
 391
 392         if (start_command(&child_process))
 393                 return error("cannot fork to run external filter %s", params->cmd);
 394
 395         sigchain_push(SIGPIPE, SIG_IGN);
 396
 397         if (params->src) {
 398                 write_err = (write_in_full(child_process.in,
 399                                            params->src, params->size) < 0);
 400                 if (errno == EPIPE)
 401                         write_err = 0;
 402         } else {
 403                 write_err = copy_fd(params->fd, child_process.in);
 404                 if (write_err == COPY_WRITE_ERROR && errno == EPIPE)
 405                         write_err = 0;
 406         }
 407
 408         if (close(child_process.in))
 409                 write_err = 1;
 410         if (write_err)
 411                 error("cannot feed the input to external filter %s", params->cmd);
 412
 413         sigchain_pop(SIGPIPE);
 414
 415         status = finish_command(&child_process);
 416         if (status)
 417                 error("external filter %s failed %d", params->cmd, status);
 418
 419         strbuf_release(&cmd);
 420         return (write_err || status);
 421 }
 422
 423 static int apply_filter(const char *path, const char *src, size_t len, int fd,
 424                         struct strbuf *dst, const char *cmd)
 425 {
 426         /*
 427          * Create a pipeline to have the command filter the buffer's
 428          * contents.
 429          *
 430          * (child --> cmd) --> us
 431          */
 432         int ret = 1;
 433         struct strbuf nbuf = STRBUF_INIT;
 434         struct async async;
 435         struct filter_params params;
 436
 437         if (!cmd)
 438                 return 0;
 439
 440         if (!dst)
 441                 return 1;
 442
 443         memset(&async, 0, sizeof(async));
 444         async.proc = filter_buffer_or_fd;
 445         async.data = &params;
 446         async.out = -1;
 447         params.src = src;
 448         params.size = len;
 449         params.fd = fd;
 450         params.cmd = cmd;
 451         params.path = path;
 452
 453         fflush(NULL);
 454         if (start_async(&async))
 455                 return 0;       /* error was already reported */
 456
 457         if (strbuf_read(&nbuf, async.out, len) < 0) {
 458                 error("read from external filter %s failed", cmd);
 459                 ret = 0;
 460         }
 461         if (close(async.out)) {
 462                 error("read from external filter %s failed", cmd);
 463                 ret = 0;
 464         }
 465         if (finish_async(&async)) {
 466                 error("external filter %s failed", cmd);
 467                 ret = 0;
 468         }
 469
 470         if (ret) {
 471                 strbuf_swap(dst, &nbuf);
 472         }
 473         strbuf_release(&nbuf);
 474         return ret;
 475 }
 476
 477 static struct convert_driver {
 478         const char *name;
 479         struct convert_driver *next;
 480         const char *smudge;
 481         const char *clean;
 482         int required;
 483 } *user_convert, **user_convert_tail;
 484
 485 static int read_convert_config(const char *var, const char *value, void *cb)
 486 {
 487         const char *key, *name;
 488         int namelen;
 489         struct convert_driver *drv;
 490
 491         /*
 492          * External conversion drivers are configured using
 493          * "filter.<name>.variable".
 494          */
 495         if (parse_config_key(var, "filter", &name, &namelen, &key) < 0 || !name)
 496                 return 0;
 497         for (drv = user_convert; drv; drv = drv->next)
 498                 if (!strncmp(drv->name, name, namelen) && !drv->name[namelen])
 499                         break;
 500         if (!drv) {
 501                 drv = xcalloc(1, sizeof(struct convert_driver));
 502                 drv->name = xmemdupz(name, namelen);
 503                 *user_convert_tail = drv;
 504                 user_convert_tail = &(drv->next);
 505         }
 506
 507         /*
 508          * filter.<name>.smudge and filter.<name>.clean specifies
 509          * the command line:
 510          *
 511          *      command-line
 512          *
 513          * The command-line will not be interpolated in any way.
 514          */
 515
 516         if (!strcmp("smudge", key))
 517                 return git_config_string(&drv->smudge, var, value);
 518
 519         if (!strcmp("clean", key))
 520                 return git_config_string(&drv->clean, var, value);
 521
 522         if (!strcmp("required", key)) {
 523                 drv->required = git_config_bool(var, value);
 524                 return 0;
 525         }
 526
 527         return 0;
 528 }
 529
 530 static int count_ident(const char *cp, unsigned long size)
 531 {
 532         /*
 533          * "$Id: 0000000000000000000000000000000000000000 $" <=> "$Id$"
 534          */
 535         int cnt = 0;
 536         char ch;
 537
 538         while (size) {
 539                 ch = *cp++;
 540                 size--;
 541                 if (ch != '$')
 542                         continue;
 543                 if (size < 3)
 544                         break;
 545                 if (memcmp("Id", cp, 2))
 546                         continue;
 547                 ch = cp[2];
 548                 cp += 3;
 549                 size -= 3;
 550                 if (ch == '$')
 551                         cnt++; /* $Id$ */
 552                 if (ch != ':')
 553                         continue;
 554
 555                 /*
 556                  * "$Id: ... "; scan up to the closing dollar sign and discard.
 557                  */
 558                 while (size) {
 559                         ch = *cp++;
 560                         size--;
 561                         if (ch == '$') {
 562                                 cnt++;
 563                                 break;
 564                         }
 565                         if (ch == '\n')
 566                                 break;
 567                 }
 568         }
 569         return cnt;
 570 }
 571
 572 static int ident_to_git(const char *path, const char *src, size_t len,
 573                         struct strbuf *buf, int ident)
 574 {
 575         char *dst, *dollar;
 576
 577         if (!ident || (src && !count_ident(src, len)))
 578                 return 0;
 579
 580         if (!buf)
 581                 return 1;
 582
 583         /* only grow if not in place */
 584         if (strbuf_avail(buf) + buf->len < len)
 585                 strbuf_grow(buf, len - buf->len);
 586         dst = buf->buf;
 587         for (;;) {
 588                 dollar = memchr(src, '$', len);
 589                 if (!dollar)
 590                         break;
 591                 memmove(dst, src, dollar + 1 - src);
 592                 dst += dollar + 1 - src;
 593                 len -= dollar + 1 - src;
 594                 src  = dollar + 1;
 595
 596                 if (len > 3 && !memcmp(src, "Id:", 3)) {
 597                         dollar = memchr(src + 3, '$', len - 3);
 598                         if (!dollar)
 599                                 break;
 600                         if (memchr(src + 3, '\n', dollar - src - 3)) {
 601                                 /* Line break before the next dollar. */
 602                                 continue;
 603                         }
 604
 605                         memcpy(dst, "Id$", 3);
 606                         dst += 3;
 607                         len -= dollar + 1 - src;
 608                         src  = dollar + 1;
 609                 }
 610         }
 611         memmove(dst, src, len);
 612         strbuf_setlen(buf, dst + len - buf->buf);
 613         return 1;
 614 }
 615
 616 static int ident_to_worktree(const char *path, const char *src, size_t len,
 617                              struct strbuf *buf, int ident)
 618 {
 619         unsigned char sha1[20];
 620         char *to_free = NULL, *dollar, *spc;
 621         int cnt;
 622
 623         if (!ident)
 624                 return 0;
 625
 626         cnt = count_ident(src, len);
 627         if (!cnt)
 628                 return 0;
 629
 630         /* are we "faking" in place editing ? */
 631         if (src == buf->buf)
 632                 to_free = strbuf_detach(buf, NULL);
 633         hash_sha1_file(src, len, "blob", sha1);
 634
 635         strbuf_grow(buf, len + cnt * 43);
 636         for (;;) {
 637                 /* step 1: run to the next '$' */
 638                 dollar = memchr(src, '$', len);
 639                 if (!dollar)
 640                         break;
 641                 strbuf_add(buf, src, dollar + 1 - src);
 642                 len -= dollar + 1 - src;
 643                 src  = dollar + 1;
 644
 645                 /* step 2: does it looks like a bit like Id:xxx$ or Id$ ? */
 646                 if (len < 3 || memcmp("Id", src, 2))
 647                         continue;
 648
 649                 /* step 3: skip over Id$ or Id:xxxxx$ */
 650                 if (src[2] == '$') {
 651                         src += 3;
 652                         len -= 3;
 653                 } else if (src[2] == ':') {
 654                         /*
 655                          * It's possible that an expanded Id has crept its way into the
 656                          * repository, we cope with that by stripping the expansion out.
 657                          * This is probably not a good idea, since it will cause changes
 658                          * on checkout, which won't go away by stash, but let's keep it
 659                          * for git-style ids.
 660                          */
 661                         dollar = memchr(src + 3, '$', len - 3);
 662                         if (!dollar) {
 663                                 /* incomplete keyword, no more '$', so just quit the loop */
 664                                 break;
 665                         }
 666
 667                         if (memchr(src + 3, '\n', dollar - src - 3)) {
 668                                 /* Line break before the next dollar. */
 669                                 continue;
 670                         }
 671
 672                         spc = memchr(src + 4, ' ', dollar - src - 4);
 673                         if (spc && spc < dollar-1) {
 674                                 /* There are spaces in unexpected places.
 675                                  * This is probably an id from some other
 676                                  * versioning system. Keep it for now.
 677                                  */
 678                                 continue;
 679                         }
 680
 681                         len -= dollar + 1 - src;
 682                         src  = dollar + 1;
 683                 } else {
 684                         /* it wasn't a "Id$" or "Id:xxxx$" */
 685                         continue;
 686                 }
 687
 688                 /* step 4: substitute */
 689                 strbuf_addstr(buf, "Id: ");
 690                 strbuf_add(buf, sha1_to_hex(sha1), 40);
 691                 strbuf_addstr(buf, " $");
 692         }
 693         strbuf_add(buf, src, len);
 694
 695         free(to_free);
 696         return 1;
 697 }
 698
 699 static enum crlf_action git_path_check_crlf(struct git_attr_check *check)
 700 {
 701         const char *value = check->value;
 702
 703         if (ATTR_TRUE(value))
 704                 return CRLF_TEXT;
 705         else if (ATTR_FALSE(value))
 706                 return CRLF_BINARY;
 707         else if (ATTR_UNSET(value))
 708                 ;
 709         else if (!strcmp(value, "input"))
 710                 return CRLF_INPUT;
 711         else if (!strcmp(value, "auto"))
 712                 return CRLF_AUTO;
 713         return CRLF_GUESS;
 714 }
 715
 716 static enum eol git_path_check_eol(struct git_attr_check *check)
 717 {
 718         const char *value = check->value;
 719
 720         if (ATTR_UNSET(value))
 721                 ;
 722         else if (!strcmp(value, "lf"))
 723                 return EOL_LF;
 724         else if (!strcmp(value, "crlf"))
 725                 return EOL_CRLF;
 726         return EOL_UNSET;
 727 }
 728
 729 static struct convert_driver *git_path_check_convert(struct git_attr_check *check)
 730 {
 731         const char *value = check->value;
 732         struct convert_driver *drv;
 733
 734         if (ATTR_TRUE(value) || ATTR_FALSE(value) || ATTR_UNSET(value))
 735                 return NULL;
 736         for (drv = user_convert; drv; drv = drv->next)
 737                 if (!strcmp(value, drv->name))
 738                         return drv;
 739         return NULL;
 740 }
 741
 742 static int git_path_check_ident(struct git_attr_check *check)
 743 {
 744         const char *value = check->value;
 745
 746         return !!ATTR_TRUE(value);
 747 }
 748
 749 static enum crlf_action input_crlf_action(enum crlf_action text_attr, enum eol eol_attr)
 750 {
 751         if (text_attr == CRLF_BINARY)
 752                 return CRLF_BINARY;
 753         if (eol_attr == EOL_LF)
 754                 return CRLF_INPUT;
 755         if (eol_attr == EOL_CRLF)
 756                 return CRLF_CRLF;
 757         return text_attr;
 758 }
 759
 760 struct conv_attrs {
 761         struct convert_driver *drv;
 762         enum crlf_action crlf_action;
 763         enum eol eol_attr;
 764         int ident;
 765 };
 766
 767 static const char *conv_attr_name[] = {
 768         "crlf", "ident", "filter", "eol", "text",
 769 };
 770 #define NUM_CONV_ATTRS ARRAY_SIZE(conv_attr_name)
 771
 772 static void convert_attrs(struct conv_attrs *ca, const char *path)
 773 {
 774         int i;
 775         static struct git_attr_check ccheck[NUM_CONV_ATTRS];
 776
 777         if (!ccheck[0].attr) {
 778                 for (i = 0; i < NUM_CONV_ATTRS; i++)
 779                         ccheck[i].attr = git_attr(conv_attr_name[i]);
 780                 user_convert_tail = &user_convert;
 781                 git_config(read_convert_config, NULL);
 782         }
 783
 784         if (!git_check_attr(path, NUM_CONV_ATTRS, ccheck)) {
 785                 ca->crlf_action = git_path_check_crlf( ccheck + 4);
 786                 if (ca->crlf_action == CRLF_GUESS)
 787                         ca->crlf_action = git_path_check_crlf(ccheck + 0);
 788                 ca->ident = git_path_check_ident(ccheck + 1);
 789                 ca->drv = git_path_check_convert(ccheck + 2);
 790                 ca->eol_attr = git_path_check_eol(ccheck + 3);
 791         } else {
 792                 ca->drv = NULL;
 793                 ca->crlf_action = CRLF_GUESS;
 794                 ca->eol_attr = EOL_UNSET;
 795                 ca->ident = 0;
 796         }
 797 }
 798
 799 int would_convert_to_git_filter_fd(const char *path)
 800 {
 801         struct conv_attrs ca;
 802
 803         convert_attrs(&ca, path);
 804         if (!ca.drv)
 805                 return 0;
 806
 807         /*
 808          * Apply a filter to an fd only if the filter is required to succeed.
 809          * We must die if the filter fails, because the original data before
 810          * filtering is not available.
 811          */
 812         if (!ca.drv->required)
 813                 return 0;
 814
 815         return apply_filter(path, NULL, 0, -1, NULL, ca.drv->clean);
 816 }
 817
 818 const char *get_convert_attr_ascii(const char *path)
 819 {
 820         struct conv_attrs ca;
 821         enum crlf_action crlf_action;
 822
 823         convert_attrs(&ca, path);
 824         crlf_action = input_crlf_action(ca.crlf_action, ca.eol_attr);
 825         switch (crlf_action) {
 826         case CRLF_GUESS:
 827                 return "";
 828         case CRLF_BINARY:
 829                 return "-text";
 830         case CRLF_TEXT:
 831                 return "text";
 832         case CRLF_INPUT:
 833                 return "text eol=lf";
 834         case CRLF_CRLF:
 835                 return "text=auto eol=crlf";
 836         case CRLF_AUTO:
 837                 return "text=auto";
 838         }
 839         return "";
 840 }
 841
 842 int convert_to_git(const char *path, const char *src, size_t len,
 843                    struct strbuf *dst, enum safe_crlf checksafe)
 844 {
 845         int ret = 0;
 846         const char *filter = NULL;
 847         int required = 0;
 848         struct conv_attrs ca;
 849
 850         convert_attrs(&ca, path);
 851         if (ca.drv) {
 852                 filter = ca.drv->clean;
 853                 required = ca.drv->required;
 854         }
 855
 856         ret |= apply_filter(path, src, len, -1, dst, filter);
 857         if (!ret && required)
 858                 die("%s: clean filter '%s' failed", path, ca.drv->name);
 859
 860         if (ret && dst) {
 861                 src = dst->buf;
 862                 len = dst->len;
 863         }
 864         ca.crlf_action = input_crlf_action(ca.crlf_action, ca.eol_attr);
 865         ret |= crlf_to_git(path, src, len, dst, ca.crlf_action, checksafe);
 866         if (ret && dst) {
 867                 src = dst->buf;
 868                 len = dst->len;
 869         }
 870         return ret | ident_to_git(path, src, len, dst, ca.ident);
 871 }
 872
 873 void convert_to_git_filter_fd(const char *path, int fd, struct strbuf *dst,
 874                               enum safe_crlf checksafe)
 875 {
 876         struct conv_attrs ca;
 877         convert_attrs(&ca, path);
 878
 879         assert(ca.drv);
 880         assert(ca.drv->clean);
 881
 882         if (!apply_filter(path, NULL, 0, fd, dst, ca.drv->clean))
 883                 die("%s: clean filter '%s' failed", path, ca.drv->name);
 884
 885         ca.crlf_action = input_crlf_action(ca.crlf_action, ca.eol_attr);
 886         crlf_to_git(path, dst->buf, dst->len, dst, ca.crlf_action, checksafe);
 887         ident_to_git(path, dst->buf, dst->len, dst, ca.ident);
 888 }
 889
 890 static int convert_to_working_tree_internal(const char *path, const char *src,
 891                                             size_t len, struct strbuf *dst,
 892                                             int normalizing)
 893 {
 894         int ret = 0, ret_filter = 0;
 895         const char *filter = NULL;
 896         int required = 0;
 897         struct conv_attrs ca;
 898
 899         convert_attrs(&ca, path);
 900         if (ca.drv) {
 901                 filter = ca.drv->smudge;
 902                 required = ca.drv->required;
 903         }
 904
 905         ret |= ident_to_worktree(path, src, len, dst, ca.ident);
 906         if (ret) {
 907                 src = dst->buf;
 908                 len = dst->len;
 909         }
 910         /*
 911          * CRLF conversion can be skipped if normalizing, unless there
 912          * is a smudge filter.  The filter might expect CRLFs.
 913          */
 914         if (filter || !normalizing) {
 915                 ca.crlf_action = input_crlf_action(ca.crlf_action, ca.eol_attr);
 916                 ret |= crlf_to_worktree(path, src, len, dst, ca.crlf_action);
 917                 if (ret) {
 918                         src = dst->buf;
 919                         len = dst->len;
 920                 }
 921         }
 922
 923         ret_filter = apply_filter(path, src, len, -1, dst, filter);
 924         if (!ret_filter && required)
 925                 die("%s: smudge filter %s failed", path, ca.drv->name);
 926
 927         return ret | ret_filter;
 928 }
 929
 930 int convert_to_working_tree(const char *path, const char *src, size_t len, struct strbuf *dst)
 931 {
 932         return convert_to_working_tree_internal(path, src, len, dst, 0);
 933 }
 934
 935 int renormalize_buffer(const char *path, const char *src, size_t len, struct strbuf *dst)
 936 {
 937         int ret = convert_to_working_tree_internal(path, src, len, dst, 1);
 938         if (ret) {
 939                 src = dst->buf;
 940                 len = dst->len;
 941         }
 942         return ret | convert_to_git(path, src, len, dst, SAFE_CRLF_FALSE);
 943 }
 944
 945 /*****************************************************************
 946  *
 947  * Streaming conversion support
 948  *
 949  *****************************************************************/
 950
 951 typedef int (*filter_fn)(struct stream_filter *,
 952                          const char *input, size_t *isize_p,
 953                          char *output, size_t *osize_p);
 954 typedef void (*free_fn)(struct stream_filter *);
 955
 956 struct stream_filter_vtbl {
 957         filter_fn filter;
 958         free_fn free;
 959 };
 960
 961 struct stream_filter {
 962         struct stream_filter_vtbl *vtbl;
 963 };
 964
 965 static int null_filter_fn(struct stream_filter *filter,
 966                           const char *input, size_t *isize_p,
 967                           char *output, size_t *osize_p)
 968 {
 969         size_t count;
 970
 971         if (!input)
 972                 return 0; /* we do not keep any states */
 973         count = *isize_p;
 974         if (*osize_p < count)
 975                 count = *osize_p;
 976         if (count) {
 977                 memmove(output, input, count);
 978                 *isize_p -= count;
 979                 *osize_p -= count;
 980         }
 981         return 0;
 982 }
 983
 984 static void null_free_fn(struct stream_filter *filter)
 985 {
 986         ; /* nothing -- null instances are shared */
 987 }
 988
 989 static struct stream_filter_vtbl null_vtbl = {
 990         null_filter_fn,
 991         null_free_fn,
 992 };
 993
 994 static struct stream_filter null_filter_singleton = {
 995         &null_vtbl,
 996 };
 997
 998 int is_null_stream_filter(struct stream_filter *filter)
 999 {
1000         return filter == &null_filter_singleton;
1001 }
1002
1003
1004 /*
1005  * LF-to-CRLF filter
1006  */
1007
1008 struct lf_to_crlf_filter {
1009         struct stream_filter filter;
1010         unsigned has_held:1;
1011         char held;
1012 };
1013
1014 static int lf_to_crlf_filter_fn(struct stream_filter *filter,
1015                                 const char *input, size_t *isize_p,
1016                                 char *output, size_t *osize_p)
1017 {
1018         size_t count, o = 0;
1019         struct lf_to_crlf_filter *lf_to_crlf = (struct lf_to_crlf_filter *)filter;
1020
1021         /*
1022          * We may be holding onto the CR to see if it is followed by a
1023          * LF, in which case we would need to go to the main loop.
1024          * Otherwise, just emit it to the output stream.
1025          */
1026         if (lf_to_crlf->has_held && (lf_to_crlf->held != '\r' || !input)) {
1027                 output[o++] = lf_to_crlf->held;
1028                 lf_to_crlf->has_held = 0;
1029         }
1030
1031         /* We are told to drain */
1032         if (!input) {
1033                 *osize_p -= o;
1034                 return 0;
1035         }
1036
1037         count = *isize_p;
1038         if (count || lf_to_crlf->has_held) {
1039                 size_t i;
1040                 int was_cr = 0;
1041
1042                 if (lf_to_crlf->has_held) {
1043                         was_cr = 1;
1044                         lf_to_crlf->has_held = 0;
1045                 }
1046
1047                 for (i = 0; o < *osize_p && i < count; i++) {
1048                         char ch = input[i];
1049
1050                         if (ch == '\n') {
1051                                 output[o++] = '\r';
1052                         } else if (was_cr) {
1053                                 /*
1054                                  * Previous round saw CR and it is not followed
1055                                  * by a LF; emit the CR before processing the
1056                                  * current character.
1057                                  */
1058                                 output[o++] = '\r';
1059                         }
1060
1061                         /*
1062                          * We may have consumed the last output slot,
1063                          * in which case we need to break out of this
1064                          * loop; hold the current character before
1065                          * returning.
1066                          */
1067                         if (*osize_p <= o) {
1068                                 lf_to_crlf->has_held = 1;
1069                                 lf_to_crlf->held = ch;
1070                                 continue; /* break but increment i */
1071                         }
1072
1073                         if (ch == '\r') {
1074                                 was_cr = 1;
1075                                 continue;
1076                         }
1077
1078                         was_cr = 0;
1079                         output[o++] = ch;
1080                 }
1081
1082                 *osize_p -= o;
1083                 *isize_p -= i;
1084
1085                 if (!lf_to_crlf->has_held && was_cr) {
1086                         lf_to_crlf->has_held = 1;
1087                         lf_to_crlf->held = '\r';
1088                 }
1089         }
1090         return 0;
1091 }
1092
1093 static void lf_to_crlf_free_fn(struct stream_filter *filter)
1094 {
1095         free(filter);
1096 }
1097
1098 static struct stream_filter_vtbl lf_to_crlf_vtbl = {
1099         lf_to_crlf_filter_fn,
1100         lf_to_crlf_free_fn,
1101 };
1102
1103 static struct stream_filter *lf_to_crlf_filter(void)
1104 {
1105         struct lf_to_crlf_filter *lf_to_crlf = xcalloc(1, sizeof(*lf_to_crlf));
1106
1107         lf_to_crlf->filter.vtbl = &lf_to_crlf_vtbl;
1108         return (struct stream_filter *)lf_to_crlf;
1109 }
1110
1111 /*
1112  * Cascade filter
1113  */
1114 #define FILTER_BUFFER 1024
1115 struct cascade_filter {
1116         struct stream_filter filter;
1117         struct stream_filter *one;
1118         struct stream_filter *two;
1119         char buf[FILTER_BUFFER];
1120         int end, ptr;
1121 };
1122
1123 static int cascade_filter_fn(struct stream_filter *filter,
1124                              const char *input, size_t *isize_p,
1125                              char *output, size_t *osize_p)
1126 {
1127         struct cascade_filter *cas = (struct cascade_filter *) filter;
1128         size_t filled = 0;
1129         size_t sz = *osize_p;
1130         size_t to_feed, remaining;
1131
1132         /*
1133          * input -- (one) --> buf -- (two) --> output
1134          */
1135         while (filled < sz) {
1136                 remaining = sz - filled;
1137
1138                 /* do we already have something to feed two with? */
1139                 if (cas->ptr < cas->end) {
1140                         to_feed = cas->end - cas->ptr;
1141                         if (stream_filter(cas->two,
1142                                           cas->buf + cas->ptr, &to_feed,
1143                                           output + filled, &remaining))
1144                                 return -1;
1145                         cas->ptr += (cas->end - cas->ptr) - to_feed;
1146                         filled = sz - remaining;
1147                         continue;
1148                 }
1149
1150                 /* feed one from upstream and have it emit into our buffer */
1151                 to_feed = input ? *isize_p : 0;
1152                 if (input && !to_feed)
1153                         break;
1154                 remaining = sizeof(cas->buf);
1155                 if (stream_filter(cas->one,
1156                                   input, &to_feed,
1157                                   cas->buf, &remaining))
1158                         return -1;
1159                 cas->end = sizeof(cas->buf) - remaining;
1160                 cas->ptr = 0;
1161                 if (input) {
1162                         size_t fed = *isize_p - to_feed;
1163                         *isize_p -= fed;
1164                         input += fed;
1165                 }
1166
1167                 /* do we know that we drained one completely? */
1168                 if (input || cas->end)
1169                         continue;
1170
1171                 /* tell two to drain; we have nothing more to give it */
1172                 to_feed = 0;
1173                 remaining = sz - filled;
1174                 if (stream_filter(cas->two,
1175                                   NULL, &to_feed,
1176                                   output + filled, &remaining))
1177                         return -1;
1178                 if (remaining == (sz - filled))
1179                         break; /* completely drained two */
1180                 filled = sz - remaining;
1181         }
1182         *osize_p -= filled;
1183         return 0;
1184 }
1185
1186 static void cascade_free_fn(struct stream_filter *filter)
1187 {
1188         struct cascade_filter *cas = (struct cascade_filter *)filter;
1189         free_stream_filter(cas->one);
1190         free_stream_filter(cas->two);
1191         free(filter);
1192 }
1193
1194 static struct stream_filter_vtbl cascade_vtbl = {
1195         cascade_filter_fn,
1196         cascade_free_fn,
1197 };
1198
1199 static struct stream_filter *cascade_filter(struct stream_filter *one,
1200                                             struct stream_filter *two)
1201 {
1202         struct cascade_filter *cascade;
1203
1204         if (!one || is_null_stream_filter(one))
1205                 return two;
1206         if (!two || is_null_stream_filter(two))
1207                 return one;
1208
1209         cascade = xmalloc(sizeof(*cascade));
1210         cascade->one = one;
1211         cascade->two = two;
1212         cascade->end = cascade->ptr = 0;
1213         cascade->filter.vtbl = &cascade_vtbl;
1214         return (struct stream_filter *)cascade;
1215 }
1216
1217 /*
1218  * ident filter
1219  */
1220 #define IDENT_DRAINING (-1)
1221 #define IDENT_SKIPPING (-2)
1222 struct ident_filter {
1223         struct stream_filter filter;
1224         struct strbuf left;
1225         int state;
1226         char ident[45]; /* ": x40 $" */
1227 };
1228
1229 static int is_foreign_ident(const char *str)
1230 {
1231         int i;
1232
1233         if (!skip_prefix(str, "$Id: ", &str))
1234                 return 0;
1235         for (i = 0; str[i]; i++) {
1236                 if (isspace(str[i]) && str[i+1] != '$')
1237                         return 1;
1238         }
1239         return 0;
1240 }
1241
1242 static void ident_drain(struct ident_filter *ident, char **output_p, size_t *osize_p)
1243 {
1244         size_t to_drain = ident->left.len;
1245
1246         if (*osize_p < to_drain)
1247                 to_drain = *osize_p;
1248         if (to_drain) {
1249                 memcpy(*output_p, ident->left.buf, to_drain);
1250                 strbuf_remove(&ident->left, 0, to_drain);
1251                 *output_p += to_drain;
1252                 *osize_p -= to_drain;
1253         }
1254         if (!ident->left.len)
1255                 ident->state = 0;
1256 }
1257
1258 static int ident_filter_fn(struct stream_filter *filter,
1259                            const char *input, size_t *isize_p,
1260                            char *output, size_t *osize_p)
1261 {
1262         struct ident_filter *ident = (struct ident_filter *)filter;
1263         static const char head[] = "$Id";
1264
1265         if (!input) {
1266                 /* drain upon eof */
1267                 switch (ident->state) {
1268                 default:
1269                         strbuf_add(&ident->left, head, ident->state);
1270                 case IDENT_SKIPPING:
1271                         /* fallthru */
1272                 case IDENT_DRAINING:
1273                         ident_drain(ident, &output, osize_p);
1274                 }
1275                 return 0;
1276         }
1277
1278         while (*isize_p || (ident->state == IDENT_DRAINING)) {
1279                 int ch;
1280
1281                 if (ident->state == IDENT_DRAINING) {
1282                         ident_drain(ident, &output, osize_p);
1283                         if (!*osize_p)
1284                                 break;
1285                         continue;
1286                 }
1287
1288                 ch = *(input++);
1289                 (*isize_p)--;
1290
1291                 if (ident->state == IDENT_SKIPPING) {
1292                         /*
1293                          * Skipping until '$' or LF, but keeping them
1294                          * in case it is a foreign ident.
1295                          */
1296                         strbuf_addch(&ident->left, ch);
1297                         if (ch != '\n' && ch != '$')
1298                                 continue;
1299                         if (ch == '$' && !is_foreign_ident(ident->left.buf)) {
1300                                 strbuf_setlen(&ident->left, sizeof(head) - 1);
1301                                 strbuf_addstr(&ident->left, ident->ident);
1302                         }
1303                         ident->state = IDENT_DRAINING;
1304                         continue;
1305                 }
1306
1307                 if (ident->state < sizeof(head) &&
1308                     head[ident->state] == ch) {
1309                         ident->state++;
1310                         continue;
1311                 }
1312
1313                 if (ident->state)
1314                         strbuf_add(&ident->left, head, ident->state);
1315                 if (ident->state == sizeof(head) - 1) {
1316                         if (ch != ':' && ch != '$') {
1317                                 strbuf_addch(&ident->left, ch);
1318                                 ident->state = 0;
1319                                 continue;
1320                         }
1321
1322                         if (ch == ':') {
1323                                 strbuf_addch(&ident->left, ch);
1324                                 ident->state = IDENT_SKIPPING;
1325                         } else {
1326                                 strbuf_addstr(&ident->left, ident->ident);
1327                                 ident->state = IDENT_DRAINING;
1328                         }
1329                         continue;
1330                 }
1331
1332                 strbuf_addch(&ident->left, ch);
1333                 ident->state = IDENT_DRAINING;
1334         }
1335         return 0;
1336 }
1337
1338 static void ident_free_fn(struct stream_filter *filter)
1339 {
1340         struct ident_filter *ident = (struct ident_filter *)filter;
1341         strbuf_release(&ident->left);
1342         free(filter);
1343 }
1344
1345 static struct stream_filter_vtbl ident_vtbl = {
1346         ident_filter_fn,
1347         ident_free_fn,
1348 };
1349
1350 static struct stream_filter *ident_filter(const unsigned char *sha1)
1351 {
1352         struct ident_filter *ident = xmalloc(sizeof(*ident));
1353
1354         xsnprintf(ident->ident, sizeof(ident->ident),
1355                   ": %s $", sha1_to_hex(sha1));
1356         strbuf_init(&ident->left, 0);
1357         ident->filter.vtbl = &ident_vtbl;
1358         ident->state = 0;
1359         return (struct stream_filter *)ident;
1360 }
1361
1362 /*
1363  * Return an appropriately constructed filter for the path, or NULL if
1364  * the contents cannot be filtered without reading the whole thing
1365  * in-core.
1366  *
1367  * Note that you would be crazy to set CRLF, smuge/clean or ident to a
1368  * large binary blob you would want us not to slurp into the memory!
1369  */
1370 struct stream_filter *get_stream_filter(const char *path, const unsigned char *sha1)
1371 {
1372         struct conv_attrs ca;
1373         enum crlf_action crlf_action;
1374         struct stream_filter *filter = NULL;
1375
1376         convert_attrs(&ca, path);
1377
1378         if (ca.drv && (ca.drv->smudge || ca.drv->clean))
1379                 return filter;
1380
1381         if (ca.ident)
1382                 filter = ident_filter(sha1);
1383
1384         crlf_action = input_crlf_action(ca.crlf_action, ca.eol_attr);
1385
1386         if ((crlf_action == CRLF_BINARY) || (crlf_action == CRLF_INPUT) ||
1387             (crlf_action == CRLF_GUESS && auto_crlf == AUTO_CRLF_FALSE))
1388                 filter = cascade_filter(filter, &null_filter_singleton);
1389
1390         else if (output_eol(crlf_action) == EOL_CRLF &&
1391                  !(crlf_action == CRLF_AUTO || crlf_action == CRLF_GUESS))
1392                 filter = cascade_filter(filter, lf_to_crlf_filter());
1393
1394         return filter;
1395 }
1396
1397 void free_stream_filter(struct stream_filter *filter)
1398 {
1399         filter->vtbl->free(filter);
1400 }
1401
1402 int stream_filter(struct stream_filter *filter,
1403                   const char *input, size_t *isize_p,
1404                   char *output, size_t *osize_p)
1405 {
1406         return filter->vtbl->filter(filter, input, isize_p, output, osize_p);
1407 }