convert.c

   1 #include "cache.h"
   2 #include "attr.h"
   3 #include "run-command.h"
   4
   5 /*
   6  * convert.c - convert a file when checking it out and checking it in.
   7  *
   8  * This should use the pathname to decide on whether it wants to do some
   9  * more interesting conversions (automatic gzip/unzip, general format
  10  * conversions etc etc), but by default it just does automatic CRLF<->LF
  11  * translation when the "auto_crlf" option is set.
  12  */
  13
  14 #define CRLF_GUESS      (-1)
  15 #define CRLF_BINARY     0
  16 #define CRLF_TEXT       1
  17 #define CRLF_INPUT      2
  18
  19 struct text_stat {
  20         /* NUL, CR, LF and CRLF counts */
  21         unsigned nul, cr, lf, crlf;
  22
  23         /* These are just approximations! */
  24         unsigned printable, nonprintable;
  25 };
  26
  27 static void gather_stats(const char *buf, unsigned long size, struct text_stat *stats)
  28 {
  29         unsigned long i;
  30
  31         memset(stats, 0, sizeof(*stats));
  32
  33         for (i = 0; i < size; i++) {
  34                 unsigned char c = buf[i];
  35                 if (c == '\r') {
  36                         stats->cr++;
  37                         if (i+1 < size && buf[i+1] == '\n')
  38                                 stats->crlf++;
  39                         continue;
  40                 }
  41                 if (c == '\n') {
  42                         stats->lf++;
  43                         continue;
  44                 }
  45                 if (c == 127)
  46                         /* DEL */
  47                         stats->nonprintable++;
  48                 else if (c < 32) {
  49                         switch (c) {
  50                                 /* BS, HT, ESC and FF */
  51                         case '\b': case '\t': case '\033': case '\014':
  52                                 stats->printable++;
  53                                 break;
  54                         case 0:
  55                                 stats->nul++;
  56                                 /* fall through */
  57                         default:
  58                                 stats->nonprintable++;
  59                         }
  60                 }
  61                 else
  62                         stats->printable++;
  63         }
  64
  65         /* If file ends with EOF then don't count this EOF as non-printable. */
  66         if (size >= 1 && buf[size-1] == '\032')
  67                 stats->nonprintable--;
  68 }
  69
  70 /*
  71  * The same heuristics as diff.c::mmfile_is_binary()
  72  */
  73 static int is_binary(unsigned long size, struct text_stat *stats)
  74 {
  75
  76         if (stats->nul)
  77                 return 1;
  78         if ((stats->printable >> 7) < stats->nonprintable)
  79                 return 1;
  80         /*
  81          * Other heuristics? Average line length might be relevant,
  82          * as might LF vs CR vs CRLF counts..
  83          *
  84          * NOTE! It might be normal to have a low ratio of CRLF to LF
  85          * (somebody starts with a LF-only file and edits it with an editor
  86          * that adds CRLF only to lines that are added..). But do  we
  87          * want to support CR-only? Probably not.
  88          */
  89         return 0;
  90 }
  91
  92 static void check_safe_crlf(const char *path, int action,
  93                             struct text_stat *stats, enum safe_crlf checksafe)
  94 {
  95         if (!checksafe)
  96                 return;
  97
  98         if (action == CRLF_INPUT || auto_crlf <= 0) {
  99                 /*
 100                  * CRLFs would not be restored by checkout:
 101                  * check if we'd remove CRLFs
 102                  */
 103                 if (stats->crlf) {
 104                         if (checksafe == SAFE_CRLF_WARN)
 105                                 warning("CRLF will be replaced by LF in %s.", path);
 106                         else /* i.e. SAFE_CRLF_FAIL */
 107                                 die("CRLF would be replaced by LF in %s.", path);
 108                 }
 109         } else if (auto_crlf > 0) {
 110                 /*
 111                  * CRLFs would be added by checkout:
 112                  * check if we have "naked" LFs
 113                  */
 114                 if (stats->lf != stats->crlf) {
 115                         if (checksafe == SAFE_CRLF_WARN)
 116                                 warning("LF will be replaced by CRLF in %s", path);
 117                         else /* i.e. SAFE_CRLF_FAIL */
 118                                 die("LF would be replaced by CRLF in %s", path);
 119                 }
 120         }
 121 }
 122
 123 static int has_cr_in_index(const char *path)
 124 {
 125         int pos, len;
 126         unsigned long sz;
 127         enum object_type type;
 128         void *data;
 129         int has_cr;
 130         struct index_state *istate = &the_index;
 131
 132         len = strlen(path);
 133         pos = index_name_pos(istate, path, len);
 134         if (pos < 0) {
 135                 /*
 136                  * We might be in the middle of a merge, in which
 137                  * case we would read stage #2 (ours).
 138                  */
 139                 int i;
 140                 for (i = -pos - 1;
 141                      (pos < 0 && i < istate->cache_nr &&
 142                       !strcmp(istate->cache[i]->name, path));
 143                      i++)
 144                         if (ce_stage(istate->cache[i]) == 2)
 145                                 pos = i;
 146         }
 147         if (pos < 0)
 148                 return 0;
 149         data = read_sha1_file(istate->cache[pos]->sha1, &type, &sz);
 150         if (!data || type != OBJ_BLOB) {
 151                 free(data);
 152                 return 0;
 153         }
 154
 155         has_cr = memchr(data, '\r', sz) != NULL;
 156         free(data);
 157         return has_cr;
 158 }
 159
 160 static int crlf_to_git(const char *path, const char *src, size_t len,
 161                        struct strbuf *buf, int action, enum safe_crlf checksafe)
 162 {
 163         struct text_stat stats;
 164         char *dst;
 165
 166         if ((action == CRLF_BINARY) || !auto_crlf || !len)
 167                 return 0;
 168
 169         gather_stats(src, len, &stats);
 170
 171         if (action == CRLF_GUESS) {
 172                 /*
 173                  * We're currently not going to even try to convert stuff
 174                  * that has bare CR characters. Does anybody do that crazy
 175                  * stuff?
 176                  */
 177                 if (stats.cr != stats.crlf)
 178                         return 0;
 179
 180                 /*
 181                  * And add some heuristics for binary vs text, of course...
 182                  */
 183                 if (is_binary(len, &stats))
 184                         return 0;
 185
 186                 /*
 187                  * If the file in the index has any CR in it, do not convert.
 188                  * This is the new safer autocrlf handling.
 189                  */
 190                 if (has_cr_in_index(path))
 191                         return 0;
 192         }
 193
 194         check_safe_crlf(path, action, &stats, checksafe);
 195
 196         /* Optimization: No CR? Nothing to convert, regardless. */
 197         if (!stats.cr)
 198                 return 0;
 199
 200         /* only grow if not in place */
 201         if (strbuf_avail(buf) + buf->len < len)
 202                 strbuf_grow(buf, len - buf->len);
 203         dst = buf->buf;
 204         if (action == CRLF_GUESS) {
 205                 /*
 206                  * If we guessed, we already know we rejected a file with
 207                  * lone CR, and we can strip a CR without looking at what
 208                  * follow it.
 209                  */
 210                 do {
 211                         unsigned char c = *src++;
 212                         if (c != '\r')
 213                                 *dst++ = c;
 214                 } while (--len);
 215         } else {
 216                 do {
 217                         unsigned char c = *src++;
 218                         if (! (c == '\r' && (1 < len && *src == '\n')))
 219                                 *dst++ = c;
 220                 } while (--len);
 221         }
 222         strbuf_setlen(buf, dst - buf->buf);
 223         return 1;
 224 }
 225
 226 static int crlf_to_worktree(const char *path, const char *src, size_t len,
 227                             struct strbuf *buf, int action)
 228 {
 229         char *to_free = NULL;
 230         struct text_stat stats;
 231
 232         if ((action == CRLF_BINARY) || (action == CRLF_INPUT) ||
 233             auto_crlf <= 0)
 234                 return 0;
 235
 236         if (!len)
 237                 return 0;
 238
 239         gather_stats(src, len, &stats);
 240
 241         /* No LF? Nothing to convert, regardless. */
 242         if (!stats.lf)
 243                 return 0;
 244
 245         /* Was it already in CRLF format? */
 246         if (stats.lf == stats.crlf)
 247                 return 0;
 248
 249         if (action == CRLF_GUESS) {
 250                 /* If we have any CR or CRLF line endings, we do not touch it */
 251                 /* This is the new safer autocrlf-handling */
 252                 if (stats.cr > 0 || stats.crlf > 0)
 253                         return 0;
 254
 255                 /* If we have any bare CR characters, we're not going to touch it */
 256                 if (stats.cr != stats.crlf)
 257                         return 0;
 258
 259                 if (is_binary(len, &stats))
 260                         return 0;
 261         }
 262
 263         /* are we "faking" in place editing ? */
 264         if (src == buf->buf)
 265                 to_free = strbuf_detach(buf, NULL);
 266
 267         strbuf_grow(buf, len + stats.lf - stats.crlf);
 268         for (;;) {
 269                 const char *nl = memchr(src, '\n', len);
 270                 if (!nl)
 271                         break;
 272                 if (nl > src && nl[-1] == '\r') {
 273                         strbuf_add(buf, src, nl + 1 - src);
 274                 } else {
 275                         strbuf_add(buf, src, nl - src);
 276                         strbuf_addstr(buf, "\r\n");
 277                 }
 278                 len -= nl + 1 - src;
 279                 src  = nl + 1;
 280         }
 281         strbuf_add(buf, src, len);
 282
 283         free(to_free);
 284         return 1;
 285 }
 286
 287 struct filter_params {
 288         const char *src;
 289         unsigned long size;
 290         const char *cmd;
 291 };
 292
 293 static int filter_buffer(int in, int out, void *data)
 294 {
 295         /*
 296          * Spawn cmd and feed the buffer contents through its stdin.
 297          */
 298         struct child_process child_process;
 299         struct filter_params *params = (struct filter_params *)data;
 300         int write_err, status;
 301         const char *argv[] = { NULL, NULL };
 302
 303         argv[0] = params->cmd;
 304
 305         memset(&child_process, 0, sizeof(child_process));
 306         child_process.argv = argv;
 307         child_process.use_shell = 1;
 308         child_process.in = -1;
 309         child_process.out = out;
 310
 311         if (start_command(&child_process))
 312                 return error("cannot fork to run external filter %s", params->cmd);
 313
 314         write_err = (write_in_full(child_process.in, params->src, params->size) < 0);
 315         if (close(child_process.in))
 316                 write_err = 1;
 317         if (write_err)
 318                 error("cannot feed the input to external filter %s", params->cmd);
 319
 320         status = finish_command(&child_process);
 321         if (status)
 322                 error("external filter %s failed %d", params->cmd, status);
 323         return (write_err || status);
 324 }
 325
 326 static int apply_filter(const char *path, const char *src, size_t len,
 327                         struct strbuf *dst, const char *cmd)
 328 {
 329         /*
 330          * Create a pipeline to have the command filter the buffer's
 331          * contents.
 332          *
 333          * (child --> cmd) --> us
 334          */
 335         int ret = 1;
 336         struct strbuf nbuf = STRBUF_INIT;
 337         struct async async;
 338         struct filter_params params;
 339
 340         if (!cmd)
 341                 return 0;
 342
 343         memset(&async, 0, sizeof(async));
 344         async.proc = filter_buffer;
 345         async.data = &params;
 346         async.out = -1;
 347         params.src = src;
 348         params.size = len;
 349         params.cmd = cmd;
 350
 351         fflush(NULL);
 352         if (start_async(&async))
 353                 return 0;       /* error was already reported */
 354
 355         if (strbuf_read(&nbuf, async.out, len) < 0) {
 356                 error("read from external filter %s failed", cmd);
 357                 ret = 0;
 358         }
 359         if (close(async.out)) {
 360                 error("read from external filter %s failed", cmd);
 361                 ret = 0;
 362         }
 363         if (finish_async(&async)) {
 364                 error("external filter %s failed", cmd);
 365                 ret = 0;
 366         }
 367
 368         if (ret) {
 369                 strbuf_swap(dst, &nbuf);
 370         }
 371         strbuf_release(&nbuf);
 372         return ret;
 373 }
 374
 375 static struct convert_driver {
 376         const char *name;
 377         struct convert_driver *next;
 378         const char *smudge;
 379         const char *clean;
 380 } *user_convert, **user_convert_tail;
 381
 382 static int read_convert_config(const char *var, const char *value, void *cb)
 383 {
 384         const char *ep, *name;
 385         int namelen;
 386         struct convert_driver *drv;
 387
 388         /*
 389          * External conversion drivers are configured using
 390          * "filter.<name>.variable".
 391          */
 392         if (prefixcmp(var, "filter.") || (ep = strrchr(var, '.')) == var + 6)
 393                 return 0;
 394         name = var + 7;
 395         namelen = ep - name;
 396         for (drv = user_convert; drv; drv = drv->next)
 397                 if (!strncmp(drv->name, name, namelen) && !drv->name[namelen])
 398                         break;
 399         if (!drv) {
 400                 drv = xcalloc(1, sizeof(struct convert_driver));
 401                 drv->name = xmemdupz(name, namelen);
 402                 *user_convert_tail = drv;
 403                 user_convert_tail = &(drv->next);
 404         }
 405
 406         ep++;
 407
 408         /*
 409          * filter.<name>.smudge and filter.<name>.clean specifies
 410          * the command line:
 411          *
 412          *      command-line
 413          *
 414          * The command-line will not be interpolated in any way.
 415          */
 416
 417         if (!strcmp("smudge", ep))
 418                 return git_config_string(&drv->smudge, var, value);
 419
 420         if (!strcmp("clean", ep))
 421                 return git_config_string(&drv->clean, var, value);
 422
 423         return 0;
 424 }
 425
 426 static void setup_convert_check(struct git_attr_check *check)
 427 {
 428         static struct git_attr *attr_crlf;
 429         static struct git_attr *attr_ident;
 430         static struct git_attr *attr_filter;
 431
 432         if (!attr_crlf) {
 433                 attr_crlf = git_attr("crlf");
 434                 attr_ident = git_attr("ident");
 435                 attr_filter = git_attr("filter");
 436                 user_convert_tail = &user_convert;
 437                 git_config(read_convert_config, NULL);
 438         }
 439         check[0].attr = attr_crlf;
 440         check[1].attr = attr_ident;
 441         check[2].attr = attr_filter;
 442 }
 443
 444 static int count_ident(const char *cp, unsigned long size)
 445 {
 446         /*
 447          * "$Id: 0000000000000000000000000000000000000000 $" <=> "$Id$"
 448          */
 449         int cnt = 0;
 450         char ch;
 451
 452         while (size) {
 453                 ch = *cp++;
 454                 size--;
 455                 if (ch != '$')
 456                         continue;
 457                 if (size < 3)
 458                         break;
 459                 if (memcmp("Id", cp, 2))
 460                         continue;
 461                 ch = cp[2];
 462                 cp += 3;
 463                 size -= 3;
 464                 if (ch == '$')
 465                         cnt++; /* $Id$ */
 466                 if (ch != ':')
 467                         continue;
 468
 469                 /*
 470                  * "$Id: ... "; scan up to the closing dollar sign and discard.
 471                  */
 472                 while (size) {
 473                         ch = *cp++;
 474                         size--;
 475                         if (ch == '$') {
 476                                 cnt++;
 477                                 break;
 478                         }
 479                         if (ch == '\n')
 480                                 break;
 481                 }
 482         }
 483         return cnt;
 484 }
 485
 486 static int ident_to_git(const char *path, const char *src, size_t len,
 487                         struct strbuf *buf, int ident)
 488 {
 489         char *dst, *dollar;
 490
 491         if (!ident || !count_ident(src, len))
 492                 return 0;
 493
 494         /* only grow if not in place */
 495         if (strbuf_avail(buf) + buf->len < len)
 496                 strbuf_grow(buf, len - buf->len);
 497         dst = buf->buf;
 498         for (;;) {
 499                 dollar = memchr(src, '$', len);
 500                 if (!dollar)
 501                         break;
 502                 memcpy(dst, src, dollar + 1 - src);
 503                 dst += dollar + 1 - src;
 504                 len -= dollar + 1 - src;
 505                 src  = dollar + 1;
 506
 507                 if (len > 3 && !memcmp(src, "Id:", 3)) {
 508                         dollar = memchr(src + 3, '$', len - 3);
 509                         if (!dollar)
 510                                 break;
 511                         if (memchr(src + 3, '\n', dollar - src - 3)) {
 512                                 /* Line break before the next dollar. */
 513                                 continue;
 514                         }
 515
 516                         memcpy(dst, "Id$", 3);
 517                         dst += 3;
 518                         len -= dollar + 1 - src;
 519                         src  = dollar + 1;
 520                 }
 521         }
 522         memcpy(dst, src, len);
 523         strbuf_setlen(buf, dst + len - buf->buf);
 524         return 1;
 525 }
 526
 527 static int ident_to_worktree(const char *path, const char *src, size_t len,
 528                              struct strbuf *buf, int ident)
 529 {
 530         unsigned char sha1[20];
 531         char *to_free = NULL, *dollar, *spc;
 532         int cnt;
 533
 534         if (!ident)
 535                 return 0;
 536
 537         cnt = count_ident(src, len);
 538         if (!cnt)
 539                 return 0;
 540
 541         /* are we "faking" in place editing ? */
 542         if (src == buf->buf)
 543                 to_free = strbuf_detach(buf, NULL);
 544         hash_sha1_file(src, len, "blob", sha1);
 545
 546         strbuf_grow(buf, len + cnt * 43);
 547         for (;;) {
 548                 /* step 1: run to the next '$' */
 549                 dollar = memchr(src, '$', len);
 550                 if (!dollar)
 551                         break;
 552                 strbuf_add(buf, src, dollar + 1 - src);
 553                 len -= dollar + 1 - src;
 554                 src  = dollar + 1;
 555
 556                 /* step 2: does it looks like a bit like Id:xxx$ or Id$ ? */
 557                 if (len < 3 || memcmp("Id", src, 2))
 558                         continue;
 559
 560                 /* step 3: skip over Id$ or Id:xxxxx$ */
 561                 if (src[2] == '$') {
 562                         src += 3;
 563                         len -= 3;
 564                 } else if (src[2] == ':') {
 565                         /*
 566                          * It's possible that an expanded Id has crept its way into the
 567                          * repository, we cope with that by stripping the expansion out.
 568                          * This is probably not a good idea, since it will cause changes
 569                          * on checkout, which won't go away by stash, but let's keep it
 570                          * for git-style ids.
 571                          */
 572                         dollar = memchr(src + 3, '$', len - 3);
 573                         if (!dollar) {
 574                                 /* incomplete keyword, no more '$', so just quit the loop */
 575                                 break;
 576                         }
 577
 578                         if (memchr(src + 3, '\n', dollar - src - 3)) {
 579                                 /* Line break before the next dollar. */
 580                                 continue;
 581                         }
 582
 583                         spc = memchr(src + 4, ' ', dollar - src - 4);
 584                         if (spc && spc < dollar-1) {
 585                                 /* There are spaces in unexpected places.
 586                                  * This is probably an id from some other
 587                                  * versioning system. Keep it for now.
 588                                  */
 589                                 continue;
 590                         }
 591
 592                         len -= dollar + 1 - src;
 593                         src  = dollar + 1;
 594                 } else {
 595                         /* it wasn't a "Id$" or "Id:xxxx$" */
 596                         continue;
 597                 }
 598
 599                 /* step 4: substitute */
 600                 strbuf_addstr(buf, "Id: ");
 601                 strbuf_add(buf, sha1_to_hex(sha1), 40);
 602                 strbuf_addstr(buf, " $");
 603         }
 604         strbuf_add(buf, src, len);
 605
 606         free(to_free);
 607         return 1;
 608 }
 609
 610 static int git_path_check_crlf(const char *path, struct git_attr_check *check)
 611 {
 612         const char *value = check->value;
 613
 614         if (ATTR_TRUE(value))
 615                 return CRLF_TEXT;
 616         else if (ATTR_FALSE(value))
 617                 return CRLF_BINARY;
 618         else if (ATTR_UNSET(value))
 619                 ;
 620         else if (!strcmp(value, "input"))
 621                 return CRLF_INPUT;
 622         return CRLF_GUESS;
 623 }
 624
 625 static struct convert_driver *git_path_check_convert(const char *path,
 626                                              struct git_attr_check *check)
 627 {
 628         const char *value = check->value;
 629         struct convert_driver *drv;
 630
 631         if (ATTR_TRUE(value) || ATTR_FALSE(value) || ATTR_UNSET(value))
 632                 return NULL;
 633         for (drv = user_convert; drv; drv = drv->next)
 634                 if (!strcmp(value, drv->name))
 635                         return drv;
 636         return NULL;
 637 }
 638
 639 static int git_path_check_ident(const char *path, struct git_attr_check *check)
 640 {
 641         const char *value = check->value;
 642
 643         return !!ATTR_TRUE(value);
 644 }
 645
 646 int convert_to_git(const char *path, const char *src, size_t len,
 647                    struct strbuf *dst, enum safe_crlf checksafe)
 648 {
 649         struct git_attr_check check[3];
 650         int crlf = CRLF_GUESS;
 651         int ident = 0, ret = 0;
 652         const char *filter = NULL;
 653
 654         setup_convert_check(check);
 655         if (!git_checkattr(path, ARRAY_SIZE(check), check)) {
 656                 struct convert_driver *drv;
 657                 crlf = git_path_check_crlf(path, check + 0);
 658                 ident = git_path_check_ident(path, check + 1);
 659                 drv = git_path_check_convert(path, check + 2);
 660                 if (drv && drv->clean)
 661                         filter = drv->clean;
 662         }
 663
 664         ret |= apply_filter(path, src, len, dst, filter);
 665         if (ret) {
 666                 src = dst->buf;
 667                 len = dst->len;
 668         }
 669         ret |= crlf_to_git(path, src, len, dst, crlf, checksafe);
 670         if (ret) {
 671                 src = dst->buf;
 672                 len = dst->len;
 673         }
 674         return ret | ident_to_git(path, src, len, dst, ident);
 675 }
 676
 677 int convert_to_working_tree(const char *path, const char *src, size_t len, struct strbuf *dst)
 678 {
 679         struct git_attr_check check[3];
 680         int crlf = CRLF_GUESS;
 681         int ident = 0, ret = 0;
 682         const char *filter = NULL;
 683
 684         setup_convert_check(check);
 685         if (!git_checkattr(path, ARRAY_SIZE(check), check)) {
 686                 struct convert_driver *drv;
 687                 crlf = git_path_check_crlf(path, check + 0);
 688                 ident = git_path_check_ident(path, check + 1);
 689                 drv = git_path_check_convert(path, check + 2);
 690                 if (drv && drv->smudge)
 691                         filter = drv->smudge;
 692         }
 693
 694         ret |= ident_to_worktree(path, src, len, dst, ident);
 695         if (ret) {
 696                 src = dst->buf;
 697                 len = dst->len;
 698         }
 699         ret |= crlf_to_worktree(path, src, len, dst, crlf);
 700         if (ret) {
 701                 src = dst->buf;
 702                 len = dst->len;
 703         }
 704         return ret | apply_filter(path, src, len, dst, filter);
 705 }