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 int 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 static int gitdiff_oldname(const char *line)
 192 {
 193         if (!old_name && !is_new)
 194                 old_name = find_name(line, NULL, 1, 0);
 195         return 0;
 196 }
 197
 198 static int gitdiff_newname(const char *line)
 199 {
 200         if (!new_name && !is_delete)
 201                 new_name = find_name(line, NULL, 1, 0);
 202         return 0;
 203 }
 204
 205 static int gitdiff_oldmode(const char *line)
 206 {
 207         old_mode = strtoul(line, NULL, 8);
 208         return 0;
 209 }
 210
 211 static int gitdiff_newmode(const char *line)
 212 {
 213         new_mode = strtoul(line, NULL, 8);
 214         return 0;
 215 }
 216
 217 static int gitdiff_delete(const char *line)
 218 {
 219         is_delete = 1;
 220         return gitdiff_oldmode(line);
 221 }
 222
 223 static int gitdiff_newfile(const char *line)
 224 {
 225         is_new = 1;
 226         return gitdiff_newmode(line);
 227 }
 228
 229 static int gitdiff_copysrc(const char *line)
 230 {
 231         is_copy = 1;
 232         old_name = find_name(line, NULL, 0, 0);
 233         return 0;
 234 }
 235
 236 static int gitdiff_copydst(const char *line)
 237 {
 238         is_copy = 1;
 239         new_name = find_name(line, NULL, 0, 0);
 240         return 0;
 241 }
 242
 243 static int gitdiff_renamesrc(const char *line)
 244 {
 245         is_rename = 1;
 246         old_name = find_name(line, NULL, 0, 0);
 247         return 0;
 248 }
 249
 250 static int gitdiff_renamedst(const char *line)
 251 {
 252         is_rename = 1;
 253         new_name = find_name(line, NULL, 0, 0);
 254         return 0;
 255 }
 256
 257 static int gitdiff_similarity(const char *line)
 258 {
 259         return 0;
 260 }
 261
 262 /*
 263  * This is normal for a diff that doesn't change anything: we'll fall through
 264  * into the next diff. Tell the parser to break out.
 265  */
 266 static int gitdiff_unrecognized(const char *line)
 267 {
 268         return -1;
 269 }
 270
 271 /* Verify that we recognize the lines following a git header */
 272 static int parse_git_header(char *line, int len, unsigned int size)
 273 {
 274         unsigned long offset;
 275
 276         /* A git diff has explicit new/delete information, so we don't guess */
 277         is_new = 0;
 278         is_delete = 0;
 279
 280         line += len;
 281         size -= len;
 282         linenr++;
 283         for (offset = len ; size > 0 ; offset += len, size -= len, line += len, linenr++) {
 284                 static const struct opentry {
 285                         const char *str;
 286                         int (*fn)(const char *);
 287                 } optable[] = {
 288                         { "@@ -", gitdiff_hdrend },
 289                         { "--- ", gitdiff_oldname },
 290                         { "+++ ", gitdiff_newname },
 291                         { "old mode ", gitdiff_oldmode },
 292                         { "new mode ", gitdiff_newmode },
 293                         { "deleted file mode ", gitdiff_delete },
 294                         { "new file mode ", gitdiff_newfile },
 295                         { "copy from ", gitdiff_copysrc },
 296                         { "copy to ", gitdiff_copydst },
 297                         { "rename from ", gitdiff_renamesrc },
 298                         { "rename to ", gitdiff_renamedst },
 299                         { "similarity index ", gitdiff_similarity },
 300                         { "", gitdiff_unrecognized },
 301                 };
 302                 int i;
 303
 304                 len = linelen(line, size);
 305                 if (!len || line[len-1] != '\n')
 306                         break;
 307                 for (i = 0; i < sizeof(optable) / sizeof(optable[0]); i++) {
 308                         const struct opentry *p = optable + i;
 309                         int oplen = strlen(p->str);
 310                         if (len < oplen || memcmp(p->str, line, oplen))
 311                                 continue;
 312                         if (p->fn(line + oplen) < 0)
 313                                 return offset;
 314                         break;
 315                 }
 316         }
 317
 318         return offset;
 319 }
 320
 321 static int parse_num(const char *line, int len, int offset, const char *expect, unsigned long *p)
 322 {
 323         char *ptr;
 324         int digits, ex;
 325
 326         if (offset < 0 || offset >= len)
 327                 return -1;
 328         line += offset;
 329         len -= offset;
 330
 331         if (!isdigit(*line))
 332                 return -1;
 333         *p = strtoul(line, &ptr, 10);
 334
 335         digits = ptr - line;
 336
 337         offset += digits;
 338         line += digits;
 339         len -= digits;
 340
 341         ex = strlen(expect);
 342         if (ex > len)
 343                 return -1;
 344         if (memcmp(line, expect, ex))
 345                 return -1;
 346
 347         return offset + ex;
 348 }
 349
 350 /*
 351  * Parse a unified diff fragment header of the
 352  * form "@@ -a,b +c,d @@"
 353  */
 354 static int parse_fragment_header(char *line, int len, unsigned long *pos)
 355 {
 356         int offset;
 357
 358         if (!len || line[len-1] != '\n')
 359                 return -1;
 360
 361         /* Figure out the number of lines in a fragment */
 362         offset = parse_num(line, len, 4, ",", pos);
 363         offset = parse_num(line, len, offset, " +", pos+1);
 364         offset = parse_num(line, len, offset, ",", pos+2);
 365         offset = parse_num(line, len, offset, " @@", pos+3);
 366
 367         return offset;
 368 }
 369
 370 static int find_header(char *line, unsigned long size, int *hdrsize)
 371 {
 372         unsigned long offset, len;
 373
 374         is_rename = is_copy = 0;
 375         is_new = is_delete = -1;
 376         old_mode = new_mode = 0;
 377         def_name = old_name = new_name = NULL;
 378         for (offset = 0; size > 0; offset += len, size -= len, line += len, linenr++) {
 379                 unsigned long nextlen;
 380
 381                 len = linelen(line, size);
 382                 if (!len)
 383                         break;
 384
 385                 /* Testing this early allows us to take a few shortcuts.. */
 386                 if (len < 6)
 387                         continue;
 388
 389                 /*
 390                  * Make sure we don't find any unconnected patch fragmants.
 391                  * That's a sign that we didn't find a header, and that a
 392                  * patch has become corrupted/broken up.
 393                  */
 394                 if (!memcmp("@@ -", line, 4)) {
 395                         unsigned long pos[4];
 396                         if (parse_fragment_header(line, len, pos) < 0)
 397                                 continue;
 398                         error("patch fragment without header at line %d: %.*s", linenr, len-1, line);
 399                 }
 400
 401                 if (size < len + 6)
 402                         break;
 403
 404                 /*
 405                  * Git patch? It might not have a real patch, just a rename
 406                  * or mode change, so we handle that specially
 407                  */
 408                 if (!memcmp("diff --git ", line, 11)) {
 409                         int git_hdr_len = parse_git_header(line, len, size);
 410                         if (git_hdr_len < 0)
 411                                 continue;
 412
 413                         *hdrsize = git_hdr_len;
 414                         return offset;
 415                 }
 416
 417                 /** --- followed by +++ ? */
 418                 if (memcmp("--- ", line,  4) || memcmp("+++ ", line + len, 4))
 419                         continue;
 420
 421                 /*
 422                  * We only accept unified patches, so we want it to
 423                  * at least have "@@ -a,b +c,d @@\n", which is 14 chars
 424                  * minimum
 425                  */
 426                 nextlen = linelen(line + len, size - len);
 427                 if (size < nextlen + 14 || memcmp("@@ -", line + len + nextlen, 4))
 428                         continue;
 429
 430                 /* Ok, we'll consider it a patch */
 431                 parse_traditional_patch(line, line+len);
 432                 *hdrsize = len + nextlen;
 433                 linenr += 2;
 434                 return offset;
 435         }
 436         return -1;
 437 }
 438
 439 /*
 440  * Parse a unified diff. Note that this really needs
 441  * to parse each fragment separately, since the only
 442  * way to know the difference between a "---" that is
 443  * part of a patch, and a "---" that starts the next
 444  * patch is to look at the line counts..
 445  */
 446 static int apply_fragment(char *line, unsigned long size)
 447 {
 448         int len = linelen(line, size), offset;
 449         unsigned long pos[4], oldlines, newlines;
 450
 451         offset = parse_fragment_header(line, len, pos);
 452         if (offset < 0)
 453                 return -1;
 454         oldlines = pos[1];
 455         newlines = pos[3];
 456
 457         if (is_new < 0 && (pos[0] || oldlines))
 458                 is_new = 0;
 459         if (is_delete < 0 && (pos[1] || newlines))
 460                 is_delete = 0;
 461
 462         /* Parse the thing.. */
 463         line += len;
 464         size -= len;
 465         linenr++;
 466         for (offset = len; size > 0; offset += len, size -= len, line += len, linenr++) {
 467                 if (!oldlines && !newlines)
 468                         break;
 469                 len = linelen(line, size);
 470                 if (!len || line[len-1] != '\n')
 471                         return -1;
 472                 switch (*line) {
 473                 default:
 474                         return -1;
 475                 case ' ':
 476                         oldlines--;
 477                         newlines--;
 478                         break;
 479                 case '-':
 480                         oldlines--;
 481                         break;
 482                 case '+':
 483                         newlines--;
 484                         break;
 485                 }
 486         }
 487         return offset;
 488 }
 489
 490 static int apply_single_patch(char *line, unsigned long size)
 491 {
 492         unsigned long offset = 0;
 493
 494         while (size > 4 && !memcmp(line, "@@ -", 4)) {
 495                 int len = apply_fragment(line, size);
 496                 if (len <= 0)
 497                         die("corrupt patch at line %d", linenr);
 498
 499 printf("applying fragment:\n%.*s\n\n", len, line);
 500
 501                 offset += len;
 502                 line += len;
 503                 size -= len;
 504         }
 505         return offset;
 506 }
 507
 508 static int apply_chunk(char *buffer, unsigned long size)
 509 {
 510         int hdrsize, patchsize;
 511         int offset = find_header(buffer, size, &hdrsize);
 512         char *header, *patch;
 513
 514         if (offset < 0)
 515                 return offset;
 516         header = buffer + offset;
 517
 518 printf("Found header:\n%.*s\n\n", hdrsize, header);
 519 printf("Rename: %d\n", is_rename);
 520 printf("Copy:   %d\n", is_copy);
 521 printf("New:    %d\n", is_new);
 522 printf("Delete: %d\n", is_delete);
 523 printf("Mode:   %o:%o\n", old_mode, new_mode);
 524 printf("Name:   '%s':'%s'\n", old_name, new_name);
 525
 526         if (old_name && cache_name_pos(old_name, strlen(old_name)) < 0)
 527                 die("file %s does not exist", old_name);
 528         if (new_name && (is_new | is_rename | is_copy)) {
 529                 if (cache_name_pos(new_name, strlen(new_name)) >= 0)
 530                         die("file %s already exists", new_name);
 531         }
 532
 533         patch = header + hdrsize;
 534         patchsize = apply_single_patch(patch, size - offset - hdrsize);
 535
 536         return offset + hdrsize + patchsize;
 537 }
 538
 539 static int apply_patch(int fd)
 540 {
 541         unsigned long offset, size;
 542         char *buffer = read_patch_file(fd, &size);
 543
 544         if (!buffer)
 545                 return -1;
 546         offset = 0;
 547         while (size > 0) {
 548                 int nr = apply_chunk(buffer + offset, size);
 549                 if (nr < 0)
 550                         break;
 551                 offset += nr;
 552                 size -= nr;
 553         }
 554         free(buffer);
 555         return 0;
 556 }
 557
 558 int main(int argc, char **argv)
 559 {
 560         int i;
 561         int read_stdin = 1;
 562
 563         if (read_cache() < 0)
 564                 die("unable to read index file");
 565
 566         for (i = 1; i < argc; i++) {
 567                 const char *arg = argv[i];
 568                 int fd;
 569
 570                 if (!strcmp(arg, "-")) {
 571                         apply_patch(0);
 572                         read_stdin = 0;
 573                         continue;
 574                 }
 575                 if (!strcmp(arg, "--no-merge")) {
 576                         merge_patch = 0;
 577                         continue;
 578                 }
 579                 fd = open(arg, O_RDONLY);
 580                 if (fd < 0)
 581                         usage(apply_usage);
 582                 read_stdin = 0;
 583                 apply_patch(fd);
 584                 close(fd);
 585         }
 586         if (read_stdin)
 587                 apply_patch(0);
 588         return 0;
 589 }