apply.c

   1 /*
   2  * apply.c
   3  *
   4  * Copyright (C) Linus Torvalds, 2005
   5  *
   6  * This applies patches on top of some (arbitrary) version of the SCM.
   7  *
   8  * NOTE! It does all its work in the index file, and only cares about
   9  * the files in the working directory if you tell it to "merge" the
  10  * patch apply.
  11  *
  12  * Even when merging it always takes the source from the index, and
  13  * uses the working tree as a "branch" for a 3-way merge.
  14  */
  15 #include <ctype.h>
  16
  17 #include "cache.h"
  18
  19 // We default to the merge behaviour, since that's what most people would
  20 // expect
  21 static int merge_patch = 1;
  22 static const char apply_usage[] = "git-apply <patch>";
  23
  24 /*
  25  * Various "current state", notably line numbers and what
  26  * file (and how) we're patching right now.. The "is_xxxx"
  27  * things are flags, where -1 means "don't know yet".
  28  */
  29 static int linenr = 1;
  30 static int old_mode, new_mode;
  31 static char *old_name, *new_name, *def_name;
  32 static int is_rename, is_copy, is_new, is_delete;
  33
  34 #define CHUNKSIZE (8192)
  35 #define SLOP (16)
  36
  37 static void *read_patch_file(int fd, unsigned long *sizep)
  38 {
  39         unsigned long size = 0, alloc = CHUNKSIZE;
  40         void *buffer = xmalloc(alloc);
  41
  42         for (;;) {
  43                 int nr = alloc - size;
  44                 if (nr < 1024) {
  45                         alloc += CHUNKSIZE;
  46                         buffer = xrealloc(buffer, alloc);
  47                         nr = alloc - size;
  48                 }
  49                 nr = read(fd, buffer + size, nr);
  50                 if (!nr)
  51                         break;
  52                 if (nr < 0) {
  53                         if (errno == EAGAIN)
  54                                 continue;
  55                         die("git-apply: read returned %s", strerror(errno));
  56                 }
  57                 size += nr;
  58         }
  59         *sizep = size;
  60
  61         /*
  62          * Make sure that we have some slop in the buffer
  63          * so that we can do speculative "memcmp" etc, and
  64          * see to it that it is NUL-filled.
  65          */
  66         if (alloc < size + SLOP)
  67                 buffer = xrealloc(buffer, size + SLOP);
  68         memset(buffer + size, 0, SLOP);
  69         return buffer;
  70 }
  71
  72 static unsigned long linelen(char *buffer, unsigned long size)
  73 {
  74         unsigned long len = 0;
  75         while (size--) {
  76                 len++;
  77                 if (*buffer++ == '\n')
  78                         break;
  79         }
  80         return len;
  81 }
  82
  83 static int is_dev_null(const char *str)
  84 {
  85         return !memcmp("/dev/null", str, 9) && isspace(str[9]);
  86 }
  87
  88 #define TERM_EXIST      1
  89 #define TERM_SPACE      2
  90 #define TERM_TAB        4
  91
  92 static int name_terminate(const char *name, int namelen, int c, int terminate)
  93 {
  94         if (c == ' ' && !(terminate & TERM_SPACE))
  95                 return 0;
  96         if (c == '\t' && !(terminate & TERM_TAB))
  97                 return 0;
  98
  99         /*
 100          * Do we want an existing name? Return false and
 101          * continue if it's not there.
 102          */
 103         if (terminate & TERM_EXIST)
 104                 return cache_name_pos(name, namelen) >= 0;
 105
 106         return 1;
 107 }
 108
 109 static char * find_name(const char *line, char *def, int p_value, int terminate)
 110 {
 111         int len;
 112         const char *start = line;
 113         char *name;
 114
 115         for (;;) {
 116                 char c = *line;
 117
 118                 if (isspace(c)) {
 119                         if (c == '\n')
 120                                 break;
 121                         if (name_terminate(start, line-start, c, terminate))
 122                                 break;
 123                 }
 124                 line++;
 125                 if (c == '/' && !--p_value)
 126                         start = line;
 127         }
 128         if (!start)
 129                 return def;
 130         len = line - start;
 131         if (!len)
 132                 return def;
 133
 134         /*
 135          * Generally we prefer the shorter name, especially
 136          * if the other one is just a variation of that with
 137          * something else tacked on to the end (ie "file.orig"
 138          * or "file~").
 139          */
 140         if (def) {
 141                 int deflen = strlen(def);
 142                 if (deflen < len && !strncmp(start, def, deflen))
 143                         return def;
 144         }
 145
 146         name = xmalloc(len + 1);
 147         memcpy(name, start, len);
 148         name[len] = 0;
 149         free(def);
 150         return name;
 151 }
 152
 153 /*
 154  * Get the name etc info from the --/+++ lines of a traditional patch header
 155  *
 156  * NOTE! This hardcodes "-p1" behaviour in filename detection.
 157  *
 158  * FIXME! The end-of-filename heuristics are kind of screwy. For existing
 159  * files, we can happily check the index for a match, but for creating a
 160  * new file we should try to match whatever "patch" does. I have no idea.
 161  */
 162 static void parse_traditional_patch(const char *first, const char *second)
 163 {
 164         int p_value = 1;
 165         char *name;
 166
 167         first += 4;     // skip "--- "
 168         second += 4;    // skip "+++ "
 169         if (is_dev_null(first)) {
 170                 is_new = 1;
 171                 name = find_name(second, def_name, p_value, TERM_SPACE | TERM_TAB);
 172                 new_name = name;
 173         } else if (is_dev_null(second)) {
 174                 is_delete = 1;
 175                 name = find_name(first, def_name, p_value, TERM_EXIST | TERM_SPACE | TERM_TAB);
 176                 old_name = name;
 177         } else {
 178                 name = find_name(first, def_name, p_value, TERM_EXIST | TERM_SPACE | TERM_TAB);
 179                 name = find_name(second, name, p_value, TERM_EXIST | TERM_SPACE | TERM_TAB);
 180                 old_name = new_name = name;
 181         }
 182         if (!name)
 183                 die("unable to find filename in patch at line %d", linenr);
 184 }
 185
 186 static int gitdiff_hdrend(const char *line)
 187 {
 188         return -1;
 189 }
 190
 191 /*
 192  * We're anal about diff header consistency, to make
 193  * sure that we don't end up having strange ambiguous
 194  * patches floating around.
 195  *
 196  * As a result, gitdiff_{old|new}name() will check
 197  * their names against any previous information, just
 198  * to make sure..
 199  */
 200 static char *gitdiff_verify_name(const char *line, int isnull, char *orig_name, const char *oldnew)
 201 {
 202         int len;
 203         const char *name;
 204
 205         if (!orig_name && !isnull)
 206                 return find_name(line, NULL, 1, 0);
 207
 208         name = "/dev/null";
 209         len = 9;
 210         if (orig_name) {
 211                 name = orig_name;
 212                 len = strlen(name);
 213                 if (isnull)
 214                         die("git-apply: bad git-diff - expected /dev/null, got %s on line %d", name, linenr);
 215         }
 216
 217         if (*name == '/')
 218                 goto absolute_path;
 219
 220         for (;;) {
 221                 char c = *line++;
 222                 if (c == '\n')
 223                         break;
 224                 if (c != '/')
 225                         continue;
 226 absolute_path:
 227                 if (memcmp(line, name, len) || line[len] != '\n')
 228                         break;
 229                 return orig_name;
 230         }
 231         die("git-apply: bad git-diff - inconsistent %s filename on line %d", oldnew, linenr);
 232         return NULL;
 233 }
 234
 235 static int gitdiff_oldname(const char *line)
 236 {
 237         old_name = gitdiff_verify_name(line, is_new, old_name, "old");
 238         return 0;
 239 }
 240
 241 static int gitdiff_newname(const char *line)
 242 {
 243         new_name = gitdiff_verify_name(line, is_delete, new_name, "new");
 244         return 0;
 245 }
 246
 247 static int gitdiff_oldmode(const char *line)
 248 {
 249         old_mode = strtoul(line, NULL, 8);
 250         return 0;
 251 }
 252
 253 static int gitdiff_newmode(const char *line)
 254 {
 255         new_mode = strtoul(line, NULL, 8);
 256         return 0;
 257 }
 258
 259 static int gitdiff_delete(const char *line)
 260 {
 261         is_delete = 1;
 262         return gitdiff_oldmode(line);
 263 }
 264
 265 static int gitdiff_newfile(const char *line)
 266 {
 267         is_new = 1;
 268         return gitdiff_newmode(line);
 269 }
 270
 271 static int gitdiff_copysrc(const char *line)
 272 {
 273         is_copy = 1;
 274         old_name = find_name(line, NULL, 0, 0);
 275         return 0;
 276 }
 277
 278 static int gitdiff_copydst(const char *line)
 279 {
 280         is_copy = 1;
 281         new_name = find_name(line, NULL, 0, 0);
 282         return 0;
 283 }
 284
 285 static int gitdiff_renamesrc(const char *line)
 286 {
 287         is_rename = 1;
 288         old_name = find_name(line, NULL, 0, 0);
 289         return 0;
 290 }
 291
 292 static int gitdiff_renamedst(const char *line)
 293 {
 294         is_rename = 1;
 295         new_name = find_name(line, NULL, 0, 0);
 296         return 0;
 297 }
 298
 299 static int gitdiff_similarity(const char *line)
 300 {
 301         return 0;
 302 }
 303
 304 /*
 305  * This is normal for a diff that doesn't change anything: we'll fall through
 306  * into the next diff. Tell the parser to break out.
 307  */
 308 static int gitdiff_unrecognized(const char *line)
 309 {
 310         return -1;
 311 }
 312
 313 /* Verify that we recognize the lines following a git header */
 314 static int parse_git_header(char *line, int len, unsigned int size)
 315 {
 316         unsigned long offset;
 317
 318         /* A git diff has explicit new/delete information, so we don't guess */
 319         is_new = 0;
 320         is_delete = 0;
 321
 322         line += len;
 323         size -= len;
 324         linenr++;
 325         for (offset = len ; size > 0 ; offset += len, size -= len, line += len, linenr++) {
 326                 static const struct opentry {
 327                         const char *str;
 328                         int (*fn)(const char *);
 329                 } optable[] = {
 330                         { "@@ -", gitdiff_hdrend },
 331                         { "--- ", gitdiff_oldname },
 332                         { "+++ ", gitdiff_newname },
 333                         { "old mode ", gitdiff_oldmode },
 334                         { "new mode ", gitdiff_newmode },
 335                         { "deleted file mode ", gitdiff_delete },
 336                         { "new file mode ", gitdiff_newfile },
 337                         { "copy from ", gitdiff_copysrc },
 338                         { "copy to ", gitdiff_copydst },
 339                         { "rename from ", gitdiff_renamesrc },
 340                         { "rename to ", gitdiff_renamedst },
 341                         { "similarity index ", gitdiff_similarity },
 342                         { "", gitdiff_unrecognized },
 343                 };
 344                 int i;
 345
 346                 len = linelen(line, size);
 347                 if (!len || line[len-1] != '\n')
 348                         break;
 349                 for (i = 0; i < sizeof(optable) / sizeof(optable[0]); i++) {
 350                         const struct opentry *p = optable + i;
 351                         int oplen = strlen(p->str);
 352                         if (len < oplen || memcmp(p->str, line, oplen))
 353                                 continue;
 354                         if (p->fn(line + oplen) < 0)
 355                                 return offset;
 356                         break;
 357                 }
 358         }
 359
 360         return offset;
 361 }
 362
 363 static int parse_num(const char *line, int len, int offset, const char *expect, unsigned long *p)
 364 {
 365         char *ptr;
 366         int digits, ex;
 367
 368         if (offset < 0 || offset >= len)
 369                 return -1;
 370         line += offset;
 371         len -= offset;
 372
 373         if (!isdigit(*line))
 374                 return -1;
 375         *p = strtoul(line, &ptr, 10);
 376
 377         digits = ptr - line;
 378
 379         offset += digits;
 380         line += digits;
 381         len -= digits;
 382
 383         ex = strlen(expect);
 384         if (ex > len)
 385                 return -1;
 386         if (memcmp(line, expect, ex))
 387                 return -1;
 388
 389         return offset + ex;
 390 }
 391
 392 /*
 393  * Parse a unified diff fragment header of the
 394  * form "@@ -a,b +c,d @@"
 395  */
 396 static int parse_fragment_header(char *line, int len, unsigned long *pos)
 397 {
 398         int offset;
 399
 400         if (!len || line[len-1] != '\n')
 401                 return -1;
 402
 403         /* Figure out the number of lines in a fragment */
 404         offset = parse_num(line, len, 4, ",", pos);
 405         offset = parse_num(line, len, offset, " +", pos+1);
 406         offset = parse_num(line, len, offset, ",", pos+2);
 407         offset = parse_num(line, len, offset, " @@", pos+3);
 408
 409         return offset;
 410 }
 411
 412 static int find_header(char *line, unsigned long size, int *hdrsize)
 413 {
 414         unsigned long offset, len;
 415
 416         is_rename = is_copy = 0;
 417         is_new = is_delete = -1;
 418         old_mode = new_mode = 0;
 419         def_name = old_name = new_name = NULL;
 420         for (offset = 0; size > 0; offset += len, size -= len, line += len, linenr++) {
 421                 unsigned long nextlen;
 422
 423                 len = linelen(line, size);
 424                 if (!len)
 425                         break;
 426
 427                 /* Testing this early allows us to take a few shortcuts.. */
 428                 if (len < 6)
 429                         continue;
 430
 431                 /*
 432                  * Make sure we don't find any unconnected patch fragmants.
 433                  * That's a sign that we didn't find a header, and that a
 434                  * patch has become corrupted/broken up.
 435                  */
 436                 if (!memcmp("@@ -", line, 4)) {
 437                         unsigned long pos[4];
 438                         if (parse_fragment_header(line, len, pos) < 0)
 439                                 continue;
 440                         error("patch fragment without header at line %d: %.*s", linenr, len-1, line);
 441                 }
 442
 443                 if (size < len + 6)
 444                         break;
 445
 446                 /*
 447                  * Git patch? It might not have a real patch, just a rename
 448                  * or mode change, so we handle that specially
 449                  */
 450                 if (!memcmp("diff --git ", line, 11)) {
 451                         int git_hdr_len = parse_git_header(line, len, size);
 452                         if (git_hdr_len < 0)
 453                                 continue;
 454
 455                         *hdrsize = git_hdr_len;
 456                         return offset;
 457                 }
 458
 459                 /** --- followed by +++ ? */
 460                 if (memcmp("--- ", line,  4) || memcmp("+++ ", line + len, 4))
 461                         continue;
 462
 463                 /*
 464                  * We only accept unified patches, so we want it to
 465                  * at least have "@@ -a,b +c,d @@\n", which is 14 chars
 466                  * minimum
 467                  */
 468                 nextlen = linelen(line + len, size - len);
 469                 if (size < nextlen + 14 || memcmp("@@ -", line + len + nextlen, 4))
 470                         continue;
 471
 472                 /* Ok, we'll consider it a patch */
 473                 parse_traditional_patch(line, line+len);
 474                 *hdrsize = len + nextlen;
 475                 linenr += 2;
 476                 return offset;
 477         }
 478         return -1;
 479 }
 480
 481 /*
 482  * Parse a unified diff. Note that this really needs
 483  * to parse each fragment separately, since the only
 484  * way to know the difference between a "---" that is
 485  * part of a patch, and a "---" that starts the next
 486  * patch is to look at the line counts..
 487  */
 488 static int apply_fragment(char *line, unsigned long size)
 489 {
 490         int len = linelen(line, size), offset;
 491         unsigned long pos[4], oldlines, newlines;
 492
 493         offset = parse_fragment_header(line, len, pos);
 494         if (offset < 0)
 495                 return -1;
 496         oldlines = pos[1];
 497         newlines = pos[3];
 498
 499         if (is_new < 0 && (pos[0] || oldlines))
 500                 is_new = 0;
 501         if (is_delete < 0 && (pos[1] || newlines))
 502                 is_delete = 0;
 503
 504         /* Parse the thing.. */
 505         line += len;
 506         size -= len;
 507         linenr++;
 508         for (offset = len; size > 0; offset += len, size -= len, line += len, linenr++) {
 509                 if (!oldlines && !newlines)
 510                         break;
 511                 len = linelen(line, size);
 512                 if (!len || line[len-1] != '\n')
 513                         return -1;
 514                 switch (*line) {
 515                 default:
 516                         return -1;
 517                 case ' ':
 518                         oldlines--;
 519                         newlines--;
 520                         break;
 521                 case '-':
 522                         oldlines--;
 523                         break;
 524                 case '+':
 525                         newlines--;
 526                         break;
 527                 }
 528         }
 529         return offset;
 530 }
 531
 532 static int apply_single_patch(char *line, unsigned long size)
 533 {
 534         unsigned long offset = 0;
 535
 536         while (size > 4 && !memcmp(line, "@@ -", 4)) {
 537                 int len = apply_fragment(line, size);
 538                 if (len <= 0)
 539                         die("corrupt patch at line %d", linenr);
 540
 541 printf("applying fragment:\n%.*s\n\n", len, line);
 542
 543                 offset += len;
 544                 line += len;
 545                 size -= len;
 546         }
 547         return offset;
 548 }
 549
 550 static int apply_chunk(char *buffer, unsigned long size)
 551 {
 552         int hdrsize, patchsize;
 553         int offset = find_header(buffer, size, &hdrsize);
 554         char *header, *patch;
 555
 556         if (offset < 0)
 557                 return offset;
 558         header = buffer + offset;
 559
 560 printf("Found header:\n%.*s\n\n", hdrsize, header);
 561 printf("Rename: %d\n", is_rename);
 562 printf("Copy:   %d\n", is_copy);
 563 printf("New:    %d\n", is_new);
 564 printf("Delete: %d\n", is_delete);
 565 printf("Mode:   %o:%o\n", old_mode, new_mode);
 566 printf("Name:   '%s':'%s'\n", old_name, new_name);
 567
 568         if (old_name && cache_name_pos(old_name, strlen(old_name)) < 0)
 569                 die("file %s does not exist", old_name);
 570         if (new_name && (is_new | is_rename | is_copy)) {
 571                 if (cache_name_pos(new_name, strlen(new_name)) >= 0)
 572                         die("file %s already exists", new_name);
 573         }
 574
 575         patch = header + hdrsize;
 576         patchsize = apply_single_patch(patch, size - offset - hdrsize);
 577
 578         return offset + hdrsize + patchsize;
 579 }
 580
 581 static int apply_patch(int fd)
 582 {
 583         unsigned long offset, size;
 584         char *buffer = read_patch_file(fd, &size);
 585
 586         if (!buffer)
 587                 return -1;
 588         offset = 0;
 589         while (size > 0) {
 590                 int nr = apply_chunk(buffer + offset, size);
 591                 if (nr < 0)
 592                         break;
 593                 offset += nr;
 594                 size -= nr;
 595         }
 596         free(buffer);
 597         return 0;
 598 }
 599
 600 int main(int argc, char **argv)
 601 {
 602         int i;
 603         int read_stdin = 1;
 604
 605         if (read_cache() < 0)
 606                 die("unable to read index file");
 607
 608         for (i = 1; i < argc; i++) {
 609                 const char *arg = argv[i];
 610                 int fd;
 611
 612                 if (!strcmp(arg, "-")) {
 613                         apply_patch(0);
 614                         read_stdin = 0;
 615                         continue;
 616                 }
 617                 if (!strcmp(arg, "--no-merge")) {
 618                         merge_patch = 0;
 619                         continue;
 620                 }
 621                 fd = open(arg, O_RDONLY);
 622                 if (fd < 0)
 623                         usage(apply_usage);
 624                 read_stdin = 0;
 625                 apply_patch(fd);
 626                 close(fd);
 627         }
 628         if (read_stdin)
 629                 apply_patch(0);
 630         return 0;
 631 }