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 crlf_to_git(const char *path, const char *src, size_t len,
 124                        struct strbuf *buf, int action, enum safe_crlf checksafe)
 125 {
 126         struct text_stat stats;
 127         char *dst;
 128
 129         if ((action == CRLF_BINARY) || !auto_crlf || !len)
 130                 return 0;
 131
 132         gather_stats(src, len, &stats);
 133
 134         if (action == CRLF_GUESS) {
 135                 /*
 136                  * We're currently not going to even try to convert stuff
 137                  * that has bare CR characters. Does anybody do that crazy
 138                  * stuff?
 139                  */
 140                 if (stats.cr != stats.crlf)
 141                         return 0;
 142
 143                 /*
 144                  * And add some heuristics for binary vs text, of course...
 145                  */
 146                 if (is_binary(len, &stats))
 147                         return 0;
 148         }
 149
 150         check_safe_crlf(path, action, &stats, checksafe);
 151
 152         /* Optimization: No CR? Nothing to convert, regardless. */
 153         if (!stats.cr)
 154                 return 0;
 155
 156         /* only grow if not in place */
 157         if (strbuf_avail(buf) + buf->len < len)
 158                 strbuf_grow(buf, len - buf->len);
 159         dst = buf->buf;
 160         if (action == CRLF_GUESS) {
 161                 /*
 162                  * If we guessed, we already know we rejected a file with
 163                  * lone CR, and we can strip a CR without looking at what
 164                  * follow it.
 165                  */
 166                 do {
 167                         unsigned char c = *src++;
 168                         if (c != '\r')
 169                                 *dst++ = c;
 170                 } while (--len);
 171         } else {
 172                 do {
 173                         unsigned char c = *src++;
 174                         if (! (c == '\r' && (1 < len && *src == '\n')))
 175                                 *dst++ = c;
 176                 } while (--len);
 177         }
 178         strbuf_setlen(buf, dst - buf->buf);
 179         return 1;
 180 }
 181
 182 static int crlf_to_worktree(const char *path, const char *src, size_t len,
 183                             struct strbuf *buf, int action)
 184 {
 185         char *to_free = NULL;
 186         struct text_stat stats;
 187
 188         if ((action == CRLF_BINARY) || (action == CRLF_INPUT) ||
 189             auto_crlf <= 0)
 190                 return 0;
 191
 192         if (!len)
 193                 return 0;
 194
 195         gather_stats(src, len, &stats);
 196
 197         /* No LF? Nothing to convert, regardless. */
 198         if (!stats.lf)
 199                 return 0;
 200
 201         /* Was it already in CRLF format? */
 202         if (stats.lf == stats.crlf)
 203                 return 0;
 204
 205         if (action == CRLF_GUESS) {
 206                 /* If we have any bare CR characters, we're not going to touch it */
 207                 if (stats.cr != stats.crlf)
 208                         return 0;
 209
 210                 if (is_binary(len, &stats))
 211                         return 0;
 212         }
 213
 214         /* are we "faking" in place editing ? */
 215         if (src == buf->buf)
 216                 to_free = strbuf_detach(buf, NULL);
 217
 218         strbuf_grow(buf, len + stats.lf - stats.crlf);
 219         for (;;) {
 220                 const char *nl = memchr(src, '\n', len);
 221                 if (!nl)
 222                         break;
 223                 if (nl > src && nl[-1] == '\r') {
 224                         strbuf_add(buf, src, nl + 1 - src);
 225                 } else {
 226                         strbuf_add(buf, src, nl - src);
 227                         strbuf_addstr(buf, "\r\n");
 228                 }
 229                 len -= nl + 1 - src;
 230                 src  = nl + 1;
 231         }
 232         strbuf_add(buf, src, len);
 233
 234         free(to_free);
 235         return 1;
 236 }
 237
 238 struct filter_params {
 239         const char *src;
 240         unsigned long size;
 241         const char *cmd;
 242 };
 243
 244 static int filter_buffer(int in, int out, void *data)
 245 {
 246         /*
 247          * Spawn cmd and feed the buffer contents through its stdin.
 248          */
 249         struct child_process child_process;
 250         struct filter_params *params = (struct filter_params *)data;
 251         int write_err, status;
 252         const char *argv[] = { NULL, NULL };
 253
 254         argv[0] = params->cmd;
 255
 256         memset(&child_process, 0, sizeof(child_process));
 257         child_process.argv = argv;
 258         child_process.use_shell = 1;
 259         child_process.in = -1;
 260         child_process.out = out;
 261
 262         if (start_command(&child_process))
 263                 return error("cannot fork to run external filter %s", params->cmd);
 264
 265         write_err = (write_in_full(child_process.in, params->src, params->size) < 0);
 266         if (close(child_process.in))
 267                 write_err = 1;
 268         if (write_err)
 269                 error("cannot feed the input to external filter %s", params->cmd);
 270
 271         status = finish_command(&child_process);
 272         if (status)
 273                 error("external filter %s failed %d", params->cmd, status);
 274         return (write_err || status);
 275 }
 276
 277 static int apply_filter(const char *path, const char *src, size_t len,
 278                         struct strbuf *dst, const char *cmd)
 279 {
 280         /*
 281          * Create a pipeline to have the command filter the buffer's
 282          * contents.
 283          *
 284          * (child --> cmd) --> us
 285          */
 286         int ret = 1;
 287         struct strbuf nbuf = STRBUF_INIT;
 288         struct async async;
 289         struct filter_params params;
 290
 291         if (!cmd)
 292                 return 0;
 293
 294         memset(&async, 0, sizeof(async));
 295         async.proc = filter_buffer;
 296         async.data = &params;
 297         async.out = -1;
 298         params.src = src;
 299         params.size = len;
 300         params.cmd = cmd;
 301
 302         fflush(NULL);
 303         if (start_async(&async))
 304                 return 0;       /* error was already reported */
 305
 306         if (strbuf_read(&nbuf, async.out, len) < 0) {
 307                 error("read from external filter %s failed", cmd);
 308                 ret = 0;
 309         }
 310         if (close(async.out)) {
 311                 error("read from external filter %s failed", cmd);
 312                 ret = 0;
 313         }
 314         if (finish_async(&async)) {
 315                 error("external filter %s failed", cmd);
 316                 ret = 0;
 317         }
 318
 319         if (ret) {
 320                 strbuf_swap(dst, &nbuf);
 321         }
 322         strbuf_release(&nbuf);
 323         return ret;
 324 }
 325
 326 static struct convert_driver {
 327         const char *name;
 328         struct convert_driver *next;
 329         const char *smudge;
 330         const char *clean;
 331 } *user_convert, **user_convert_tail;
 332
 333 static int read_convert_config(const char *var, const char *value, void *cb)
 334 {
 335         const char *ep, *name;
 336         int namelen;
 337         struct convert_driver *drv;
 338
 339         /*
 340          * External conversion drivers are configured using
 341          * "filter.<name>.variable".
 342          */
 343         if (prefixcmp(var, "filter.") || (ep = strrchr(var, '.')) == var + 6)
 344                 return 0;
 345         name = var + 7;
 346         namelen = ep - name;
 347         for (drv = user_convert; drv; drv = drv->next)
 348                 if (!strncmp(drv->name, name, namelen) && !drv->name[namelen])
 349                         break;
 350         if (!drv) {
 351                 drv = xcalloc(1, sizeof(struct convert_driver));
 352                 drv->name = xmemdupz(name, namelen);
 353                 *user_convert_tail = drv;
 354                 user_convert_tail = &(drv->next);
 355         }
 356
 357         ep++;
 358
 359         /*
 360          * filter.<name>.smudge and filter.<name>.clean specifies
 361          * the command line:
 362          *
 363          *      command-line
 364          *
 365          * The command-line will not be interpolated in any way.
 366          */
 367
 368         if (!strcmp("smudge", ep))
 369                 return git_config_string(&drv->smudge, var, value);
 370
 371         if (!strcmp("clean", ep))
 372                 return git_config_string(&drv->clean, var, value);
 373
 374         return 0;
 375 }
 376
 377 static void setup_convert_check(struct git_attr_check *check)
 378 {
 379         static struct git_attr *attr_crlf;
 380         static struct git_attr *attr_ident;
 381         static struct git_attr *attr_filter;
 382
 383         if (!attr_crlf) {
 384                 attr_crlf = git_attr("crlf");
 385                 attr_ident = git_attr("ident");
 386                 attr_filter = git_attr("filter");
 387                 user_convert_tail = &user_convert;
 388                 git_config(read_convert_config, NULL);
 389         }
 390         check[0].attr = attr_crlf;
 391         check[1].attr = attr_ident;
 392         check[2].attr = attr_filter;
 393 }
 394
 395 static int count_ident(const char *cp, unsigned long size)
 396 {
 397         /*
 398          * "$Id: 0000000000000000000000000000000000000000 $" <=> "$Id$"
 399          */
 400         int cnt = 0;
 401         char ch;
 402
 403         while (size) {
 404                 ch = *cp++;
 405                 size--;
 406                 if (ch != '$')
 407                         continue;
 408                 if (size < 3)
 409                         break;
 410                 if (memcmp("Id", cp, 2))
 411                         continue;
 412                 ch = cp[2];
 413                 cp += 3;
 414                 size -= 3;
 415                 if (ch == '$')
 416                         cnt++; /* $Id$ */
 417                 if (ch != ':')
 418                         continue;
 419
 420                 /*
 421                  * "$Id: ... "; scan up to the closing dollar sign and discard.
 422                  */
 423                 while (size) {
 424                         ch = *cp++;
 425                         size--;
 426                         if (ch == '$') {
 427                                 cnt++;
 428                                 break;
 429                         }
 430                         if (ch == '\n')
 431                                 break;
 432                 }
 433         }
 434         return cnt;
 435 }
 436
 437 static int ident_to_git(const char *path, const char *src, size_t len,
 438                         struct strbuf *buf, int ident)
 439 {
 440         char *dst, *dollar;
 441
 442         if (!ident || !count_ident(src, len))
 443                 return 0;
 444
 445         /* only grow if not in place */
 446         if (strbuf_avail(buf) + buf->len < len)
 447                 strbuf_grow(buf, len - buf->len);
 448         dst = buf->buf;
 449         for (;;) {
 450                 dollar = memchr(src, '$', len);
 451                 if (!dollar)
 452                         break;
 453                 memcpy(dst, src, dollar + 1 - src);
 454                 dst += dollar + 1 - src;
 455                 len -= dollar + 1 - src;
 456                 src  = dollar + 1;
 457
 458                 if (len > 3 && !memcmp(src, "Id:", 3)) {
 459                         dollar = memchr(src + 3, '$', len - 3);
 460                         if (!dollar)
 461                                 break;
 462                         if (memchr(src + 3, '\n', dollar - src - 3)) {
 463                                 /* Line break before the next dollar. */
 464                                 continue;
 465                         }
 466
 467                         memcpy(dst, "Id$", 3);
 468                         dst += 3;
 469                         len -= dollar + 1 - src;
 470                         src  = dollar + 1;
 471                 }
 472         }
 473         memcpy(dst, src, len);
 474         strbuf_setlen(buf, dst + len - buf->buf);
 475         return 1;
 476 }
 477
 478 static int ident_to_worktree(const char *path, const char *src, size_t len,
 479                              struct strbuf *buf, int ident)
 480 {
 481         unsigned char sha1[20];
 482         char *to_free = NULL, *dollar, *spc;
 483         int cnt;
 484
 485         if (!ident)
 486                 return 0;
 487
 488         cnt = count_ident(src, len);
 489         if (!cnt)
 490                 return 0;
 491
 492         /* are we "faking" in place editing ? */
 493         if (src == buf->buf)
 494                 to_free = strbuf_detach(buf, NULL);
 495         hash_sha1_file(src, len, "blob", sha1);
 496
 497         strbuf_grow(buf, len + cnt * 43);
 498         for (;;) {
 499                 /* step 1: run to the next '$' */
 500                 dollar = memchr(src, '$', len);
 501                 if (!dollar)
 502                         break;
 503                 strbuf_add(buf, src, dollar + 1 - src);
 504                 len -= dollar + 1 - src;
 505                 src  = dollar + 1;
 506
 507                 /* step 2: does it looks like a bit like Id:xxx$ or Id$ ? */
 508                 if (len < 3 || memcmp("Id", src, 2))
 509                         continue;
 510
 511                 /* step 3: skip over Id$ or Id:xxxxx$ */
 512                 if (src[2] == '$') {
 513                         src += 3;
 514                         len -= 3;
 515                 } else if (src[2] == ':') {
 516                         /*
 517                          * It's possible that an expanded Id has crept its way into the
 518                          * repository, we cope with that by stripping the expansion out.
 519                          * This is probably not a good idea, since it will cause changes
 520                          * on checkout, which won't go away by stash, but let's keep it
 521                          * for git-style ids.
 522                          */
 523                         dollar = memchr(src + 3, '$', len - 3);
 524                         if (!dollar) {
 525                                 /* incomplete keyword, no more '$', so just quit the loop */
 526                                 break;
 527                         }
 528
 529                         if (memchr(src + 3, '\n', dollar - src - 3)) {
 530                                 /* Line break before the next dollar. */
 531                                 continue;
 532                         }
 533
 534                         spc = memchr(src + 4, ' ', dollar - src - 4);
 535                         if (spc && spc < dollar-1) {
 536                                 /* There are spaces in unexpected places.
 537                                  * This is probably an id from some other
 538                                  * versioning system. Keep it for now.
 539                                  */
 540                                 continue;
 541                         }
 542
 543                         len -= dollar + 1 - src;
 544                         src  = dollar + 1;
 545                 } else {
 546                         /* it wasn't a "Id$" or "Id:xxxx$" */
 547                         continue;
 548                 }
 549
 550                 /* step 4: substitute */
 551                 strbuf_addstr(buf, "Id: ");
 552                 strbuf_add(buf, sha1_to_hex(sha1), 40);
 553                 strbuf_addstr(buf, " $");
 554         }
 555         strbuf_add(buf, src, len);
 556
 557         free(to_free);
 558         return 1;
 559 }
 560
 561 static int git_path_check_crlf(const char *path, struct git_attr_check *check)
 562 {
 563         const char *value = check->value;
 564
 565         if (ATTR_TRUE(value))
 566                 return CRLF_TEXT;
 567         else if (ATTR_FALSE(value))
 568                 return CRLF_BINARY;
 569         else if (ATTR_UNSET(value))
 570                 ;
 571         else if (!strcmp(value, "input"))
 572                 return CRLF_INPUT;
 573         return CRLF_GUESS;
 574 }
 575
 576 static struct convert_driver *git_path_check_convert(const char *path,
 577                                              struct git_attr_check *check)
 578 {
 579         const char *value = check->value;
 580         struct convert_driver *drv;
 581
 582         if (ATTR_TRUE(value) || ATTR_FALSE(value) || ATTR_UNSET(value))
 583                 return NULL;
 584         for (drv = user_convert; drv; drv = drv->next)
 585                 if (!strcmp(value, drv->name))
 586                         return drv;
 587         return NULL;
 588 }
 589
 590 static int git_path_check_ident(const char *path, struct git_attr_check *check)
 591 {
 592         const char *value = check->value;
 593
 594         return !!ATTR_TRUE(value);
 595 }
 596
 597 int convert_to_git(const char *path, const char *src, size_t len,
 598                    struct strbuf *dst, enum safe_crlf checksafe)
 599 {
 600         struct git_attr_check check[3];
 601         int crlf = CRLF_GUESS;
 602         int ident = 0, ret = 0;
 603         const char *filter = NULL;
 604
 605         setup_convert_check(check);
 606         if (!git_checkattr(path, ARRAY_SIZE(check), check)) {
 607                 struct convert_driver *drv;
 608                 crlf = git_path_check_crlf(path, check + 0);
 609                 ident = git_path_check_ident(path, check + 1);
 610                 drv = git_path_check_convert(path, check + 2);
 611                 if (drv && drv->clean)
 612                         filter = drv->clean;
 613         }
 614
 615         ret |= apply_filter(path, src, len, dst, filter);
 616         if (ret) {
 617                 src = dst->buf;
 618                 len = dst->len;
 619         }
 620         ret |= crlf_to_git(path, src, len, dst, crlf, checksafe);
 621         if (ret) {
 622                 src = dst->buf;
 623                 len = dst->len;
 624         }
 625         return ret | ident_to_git(path, src, len, dst, ident);
 626 }
 627
 628 int convert_to_working_tree(const char *path, const char *src, size_t len, struct strbuf *dst)
 629 {
 630         struct git_attr_check check[3];
 631         int crlf = CRLF_GUESS;
 632         int ident = 0, ret = 0;
 633         const char *filter = NULL;
 634
 635         setup_convert_check(check);
 636         if (!git_checkattr(path, ARRAY_SIZE(check), check)) {
 637                 struct convert_driver *drv;
 638                 crlf = git_path_check_crlf(path, check + 0);
 639                 ident = git_path_check_ident(path, check + 1);
 640                 drv = git_path_check_convert(path, check + 2);
 641                 if (drv && drv->smudge)
 642                         filter = drv->smudge;
 643         }
 644
 645         ret |= ident_to_worktree(path, src, len, dst, ident);
 646         if (ret) {
 647                 src = dst->buf;
 648                 len = dst->len;
 649         }
 650         ret |= crlf_to_worktree(path, src, len, dst, crlf);
 651         if (ret) {
 652                 src = dst->buf;
 653                 len = dst->len;
 654         }
 655         return ret | apply_filter(path, src, len, dst, filter);
 656 }