1/* 2 * Recursive Merge algorithm stolen from git-merge-recursive.py by 3 * Fredrik Kuivinen. 4 * The thieves were Alex Riesen and Johannes Schindelin, in June/July 2006 5 */ 6#include "cache.h" 7#include "config.h" 8#include "advice.h" 9#include "lockfile.h" 10#include "cache-tree.h" 11#include "commit.h" 12#include "blob.h" 13#include "builtin.h" 14#include "tree-walk.h" 15#include "diff.h" 16#include "diffcore.h" 17#include "tag.h" 18#include "unpack-trees.h" 19#include "string-list.h" 20#include "xdiff-interface.h" 21#include "ll-merge.h" 22#include "attr.h" 23#include "merge-recursive.h" 24#include "dir.h" 25#include "submodule.h" 26 27struct path_hashmap_entry { 28 struct hashmap_entry e; 29 char path[FLEX_ARRAY]; 30}; 31 32static int path_hashmap_cmp(const void *cmp_data, 33 const void *entry, 34 const void *entry_or_key, 35 const void *keydata) 36{ 37 const struct path_hashmap_entry *a = entry; 38 const struct path_hashmap_entry *b = entry_or_key; 39 const char *key = keydata; 40 41 if (ignore_case) 42 return strcasecmp(a->path, key ? key : b->path); 43 else 44 return strcmp(a->path, key ? key : b->path); 45} 46 47static unsigned int path_hash(const char *path) 48{ 49 return ignore_case ? strihash(path) : strhash(path); 50} 51 52static struct dir_rename_entry *dir_rename_find_entry(struct hashmap *hashmap, 53 char *dir) 54{ 55 struct dir_rename_entry key; 56 57 if (dir == NULL) 58 return NULL; 59 hashmap_entry_init(&key, strhash(dir)); 60 key.dir = dir; 61 return hashmap_get(hashmap, &key, NULL); 62} 63 64static int dir_rename_cmp(const void *unused_cmp_data, 65 const void *entry, 66 const void *entry_or_key, 67 const void *unused_keydata) 68{ 69 const struct dir_rename_entry *e1 = entry; 70 const struct dir_rename_entry *e2 = entry_or_key; 71 72 return strcmp(e1->dir, e2->dir); 73} 74 75static void dir_rename_init(struct hashmap *map) 76{ 77 hashmap_init(map, dir_rename_cmp, NULL, 0); 78} 79 80static void dir_rename_entry_init(struct dir_rename_entry *entry, 81 char *directory) 82{ 83 hashmap_entry_init(entry, strhash(directory)); 84 entry->dir = directory; 85 entry->non_unique_new_dir = 0; 86 strbuf_init(&entry->new_dir, 0); 87 string_list_init(&entry->possible_new_dirs, 0); 88} 89 90static struct collision_entry *collision_find_entry(struct hashmap *hashmap, 91 char *target_file) 92{ 93 struct collision_entry key; 94 95 hashmap_entry_init(&key, strhash(target_file)); 96 key.target_file = target_file; 97 return hashmap_get(hashmap, &key, NULL); 98} 99 100static int collision_cmp(void *unused_cmp_data, 101 const struct collision_entry *e1, 102 const struct collision_entry *e2, 103 const void *unused_keydata) 104{ 105 return strcmp(e1->target_file, e2->target_file); 106} 107 108static void collision_init(struct hashmap *map) 109{ 110 hashmap_init(map, (hashmap_cmp_fn) collision_cmp, NULL, 0); 111} 112 113static void flush_output(struct merge_options *o) 114{ 115 if (o->buffer_output < 2 && o->obuf.len) { 116 fputs(o->obuf.buf, stdout); 117 strbuf_reset(&o->obuf); 118 } 119} 120 121static int err(struct merge_options *o, const char *err, ...) 122{ 123 va_list params; 124 125 if (o->buffer_output < 2) 126 flush_output(o); 127 else { 128 strbuf_complete(&o->obuf, '\n'); 129 strbuf_addstr(&o->obuf, "error: "); 130 } 131 va_start(params, err); 132 strbuf_vaddf(&o->obuf, err, params); 133 va_end(params); 134 if (o->buffer_output > 1) 135 strbuf_addch(&o->obuf, '\n'); 136 else { 137 error("%s", o->obuf.buf); 138 strbuf_reset(&o->obuf); 139 } 140 141 return -1; 142} 143 144static struct tree *shift_tree_object(struct tree *one, struct tree *two, 145 const char *subtree_shift) 146{ 147 struct object_id shifted; 148 149 if (!*subtree_shift) { 150 shift_tree(&one->object.oid, &two->object.oid, &shifted, 0); 151 } else { 152 shift_tree_by(&one->object.oid, &two->object.oid, &shifted, 153 subtree_shift); 154 } 155 if (!oidcmp(&two->object.oid, &shifted)) 156 return two; 157 return lookup_tree(&shifted); 158} 159 160static struct commit *make_virtual_commit(struct tree *tree, const char *comment) 161{ 162 struct commit *commit = alloc_commit_node(); 163 164 set_merge_remote_desc(commit, comment, (struct object *)commit); 165 commit->tree = tree; 166 commit->object.parsed = 1; 167 return commit; 168} 169 170/* 171 * Since we use get_tree_entry(), which does not put the read object into 172 * the object pool, we cannot rely on a == b. 173 */ 174static int oid_eq(const struct object_id *a, const struct object_id *b) 175{ 176 if (!a && !b) 177 return 2; 178 return a && b && oidcmp(a, b) == 0; 179} 180 181enum rename_type { 182 RENAME_NORMAL = 0, 183 RENAME_DIR, 184 RENAME_DELETE, 185 RENAME_ONE_FILE_TO_ONE, 186 RENAME_ONE_FILE_TO_TWO, 187 RENAME_TWO_FILES_TO_ONE 188}; 189 190struct rename_conflict_info { 191 enum rename_type rename_type; 192 struct diff_filepair *pair1; 193 struct diff_filepair *pair2; 194 const char *branch1; 195 const char *branch2; 196 struct stage_data *dst_entry1; 197 struct stage_data *dst_entry2; 198 struct diff_filespec ren1_other; 199 struct diff_filespec ren2_other; 200}; 201 202/* 203 * Since we want to write the index eventually, we cannot reuse the index 204 * for these (temporary) data. 205 */ 206struct stage_data { 207 struct { 208 unsigned mode; 209 struct object_id oid; 210 } stages[4]; 211 struct rename_conflict_info *rename_conflict_info; 212 unsigned processed:1; 213}; 214 215static inline void setup_rename_conflict_info(enum rename_type rename_type, 216 struct diff_filepair *pair1, 217 struct diff_filepair *pair2, 218 const char *branch1, 219 const char *branch2, 220 struct stage_data *dst_entry1, 221 struct stage_data *dst_entry2, 222 struct merge_options *o, 223 struct stage_data *src_entry1, 224 struct stage_data *src_entry2) 225{ 226 struct rename_conflict_info *ci = xcalloc(1, sizeof(struct rename_conflict_info)); 227 ci->rename_type = rename_type; 228 ci->pair1 = pair1; 229 ci->branch1 = branch1; 230 ci->branch2 = branch2; 231 232 ci->dst_entry1 = dst_entry1; 233 dst_entry1->rename_conflict_info = ci; 234 dst_entry1->processed = 0; 235 236 assert(!pair2 == !dst_entry2); 237 if (dst_entry2) { 238 ci->dst_entry2 = dst_entry2; 239 ci->pair2 = pair2; 240 dst_entry2->rename_conflict_info = ci; 241 } 242 243 if (rename_type == RENAME_TWO_FILES_TO_ONE) { 244 /* 245 * For each rename, there could have been 246 * modifications on the side of history where that 247 * file was not renamed. 248 */ 249 int ostage1 = o->branch1 == branch1 ? 3 : 2; 250 int ostage2 = ostage1 ^ 1; 251 252 ci->ren1_other.path = pair1->one->path; 253 oidcpy(&ci->ren1_other.oid, &src_entry1->stages[ostage1].oid); 254 ci->ren1_other.mode = src_entry1->stages[ostage1].mode; 255 256 ci->ren2_other.path = pair2->one->path; 257 oidcpy(&ci->ren2_other.oid, &src_entry2->stages[ostage2].oid); 258 ci->ren2_other.mode = src_entry2->stages[ostage2].mode; 259 } 260} 261 262static int show(struct merge_options *o, int v) 263{ 264 return (!o->call_depth && o->verbosity >= v) || o->verbosity >= 5; 265} 266 267__attribute__((format (printf, 3, 4))) 268static void output(struct merge_options *o, int v, const char *fmt, ...) 269{ 270 va_list ap; 271 272 if (!show(o, v)) 273 return; 274 275 strbuf_addchars(&o->obuf, ' ', o->call_depth * 2); 276 277 va_start(ap, fmt); 278 strbuf_vaddf(&o->obuf, fmt, ap); 279 va_end(ap); 280 281 strbuf_addch(&o->obuf, '\n'); 282 if (!o->buffer_output) 283 flush_output(o); 284} 285 286static void output_commit_title(struct merge_options *o, struct commit *commit) 287{ 288 strbuf_addchars(&o->obuf, ' ', o->call_depth * 2); 289 if (commit->util) 290 strbuf_addf(&o->obuf, "virtual %s\n", 291 merge_remote_util(commit)->name); 292 else { 293 strbuf_add_unique_abbrev(&o->obuf, &commit->object.oid, 294 DEFAULT_ABBREV); 295 strbuf_addch(&o->obuf, ' '); 296 if (parse_commit(commit) != 0) 297 strbuf_addstr(&o->obuf, _("(bad commit)\n")); 298 else { 299 const char *title; 300 const char *msg = get_commit_buffer(commit, NULL); 301 int len = find_commit_subject(msg, &title); 302 if (len) 303 strbuf_addf(&o->obuf, "%.*s\n", len, title); 304 unuse_commit_buffer(commit, msg); 305 } 306 } 307 flush_output(o); 308} 309 310static int add_cacheinfo(struct merge_options *o, 311 unsigned int mode, const struct object_id *oid, 312 const char *path, int stage, int refresh, int options) 313{ 314 struct cache_entry *ce; 315 int ret; 316 317 ce = make_cache_entry(mode, oid ? oid->hash : null_sha1, path, stage, 0); 318 if (!ce) 319 return err(o, _("add_cacheinfo failed for path '%s'; merge aborting."), path); 320 321 ret = add_cache_entry(ce, options); 322 if (refresh) { 323 struct cache_entry *nce; 324 325 nce = refresh_cache_entry(ce, CE_MATCH_REFRESH | CE_MATCH_IGNORE_MISSING); 326 if (!nce) 327 return err(o, _("add_cacheinfo failed to refresh for path '%s'; merge aborting."), path); 328 if (nce != ce) 329 ret = add_cache_entry(nce, options); 330 } 331 return ret; 332} 333 334static void init_tree_desc_from_tree(struct tree_desc *desc, struct tree *tree) 335{ 336 parse_tree(tree); 337 init_tree_desc(desc, tree->buffer, tree->size); 338} 339 340static int git_merge_trees(struct merge_options *o, 341 struct tree *common, 342 struct tree *head, 343 struct tree *merge) 344{ 345 int rc; 346 struct tree_desc t[3]; 347 struct index_state tmp_index = { NULL }; 348 349 memset(&o->unpack_opts, 0, sizeof(o->unpack_opts)); 350 if (o->call_depth) 351 o->unpack_opts.index_only = 1; 352 else 353 o->unpack_opts.update = 1; 354 o->unpack_opts.merge = 1; 355 o->unpack_opts.head_idx = 2; 356 o->unpack_opts.fn = threeway_merge; 357 o->unpack_opts.src_index = &the_index; 358 o->unpack_opts.dst_index = &tmp_index; 359 setup_unpack_trees_porcelain(&o->unpack_opts, "merge"); 360 361 init_tree_desc_from_tree(t+0, common); 362 init_tree_desc_from_tree(t+1, head); 363 init_tree_desc_from_tree(t+2, merge); 364 365 rc = unpack_trees(3, t, &o->unpack_opts); 366 cache_tree_free(&active_cache_tree); 367 368 /* 369 * Update the_index to match the new results, AFTER saving a copy 370 * in o->orig_index. Update src_index to point to the saved copy. 371 * (verify_uptodate() checks src_index, and the original index is 372 * the one that had the necessary modification timestamps.) 373 */ 374 o->orig_index = the_index; 375 the_index = tmp_index; 376 o->unpack_opts.src_index = &o->orig_index; 377 378 return rc; 379} 380 381struct tree *write_tree_from_memory(struct merge_options *o) 382{ 383 struct tree *result = NULL; 384 385 if (unmerged_cache()) { 386 int i; 387 fprintf(stderr, "BUG: There are unmerged index entries:\n"); 388 for (i = 0; i < active_nr; i++) { 389 const struct cache_entry *ce = active_cache[i]; 390 if (ce_stage(ce)) 391 fprintf(stderr, "BUG: %d %.*s\n", ce_stage(ce), 392 (int)ce_namelen(ce), ce->name); 393 } 394 die("BUG: unmerged index entries in merge-recursive.c"); 395 } 396 397 if (!active_cache_tree) 398 active_cache_tree = cache_tree(); 399 400 if (!cache_tree_fully_valid(active_cache_tree) && 401 cache_tree_update(&the_index, 0) < 0) { 402 err(o, _("error building trees")); 403 return NULL; 404 } 405 406 result = lookup_tree(&active_cache_tree->oid); 407 408 return result; 409} 410 411static int save_files_dirs(const struct object_id *oid, 412 struct strbuf *base, const char *path, 413 unsigned int mode, int stage, void *context) 414{ 415 struct path_hashmap_entry *entry; 416 int baselen = base->len; 417 struct merge_options *o = context; 418 419 strbuf_addstr(base, path); 420 421 FLEX_ALLOC_MEM(entry, path, base->buf, base->len); 422 hashmap_entry_init(entry, path_hash(entry->path)); 423 hashmap_add(&o->current_file_dir_set, entry); 424 425 strbuf_setlen(base, baselen); 426 return (S_ISDIR(mode) ? READ_TREE_RECURSIVE : 0); 427} 428 429static void get_files_dirs(struct merge_options *o, struct tree *tree) 430{ 431 struct pathspec match_all; 432 memset(&match_all, 0, sizeof(match_all)); 433 read_tree_recursive(tree, "", 0, 0, &match_all, save_files_dirs, o); 434} 435 436static int get_tree_entry_if_blob(const struct object_id *tree, 437 const char *path, 438 struct object_id *hashy, 439 unsigned int *mode_o) 440{ 441 int ret; 442 443 ret = get_tree_entry(tree, path, hashy, mode_o); 444 if (S_ISDIR(*mode_o)) { 445 oidcpy(hashy, &null_oid); 446 *mode_o = 0; 447 } 448 return ret; 449} 450 451/* 452 * Returns an index_entry instance which doesn't have to correspond to 453 * a real cache entry in Git's index. 454 */ 455static struct stage_data *insert_stage_data(const char *path, 456 struct tree *o, struct tree *a, struct tree *b, 457 struct string_list *entries) 458{ 459 struct string_list_item *item; 460 struct stage_data *e = xcalloc(1, sizeof(struct stage_data)); 461 get_tree_entry_if_blob(&o->object.oid, path, 462 &e->stages[1].oid, &e->stages[1].mode); 463 get_tree_entry_if_blob(&a->object.oid, path, 464 &e->stages[2].oid, &e->stages[2].mode); 465 get_tree_entry_if_blob(&b->object.oid, path, 466 &e->stages[3].oid, &e->stages[3].mode); 467 item = string_list_insert(entries, path); 468 item->util = e; 469 return e; 470} 471 472/* 473 * Create a dictionary mapping file names to stage_data objects. The 474 * dictionary contains one entry for every path with a non-zero stage entry. 475 */ 476static struct string_list *get_unmerged(void) 477{ 478 struct string_list *unmerged = xcalloc(1, sizeof(struct string_list)); 479 int i; 480 481 unmerged->strdup_strings = 1; 482 483 for (i = 0; i < active_nr; i++) { 484 struct string_list_item *item; 485 struct stage_data *e; 486 const struct cache_entry *ce = active_cache[i]; 487 if (!ce_stage(ce)) 488 continue; 489 490 item = string_list_lookup(unmerged, ce->name); 491 if (!item) { 492 item = string_list_insert(unmerged, ce->name); 493 item->util = xcalloc(1, sizeof(struct stage_data)); 494 } 495 e = item->util; 496 e->stages[ce_stage(ce)].mode = ce->ce_mode; 497 oidcpy(&e->stages[ce_stage(ce)].oid, &ce->oid); 498 } 499 500 return unmerged; 501} 502 503static int string_list_df_name_compare(const char *one, const char *two) 504{ 505 int onelen = strlen(one); 506 int twolen = strlen(two); 507 /* 508 * Here we only care that entries for D/F conflicts are 509 * adjacent, in particular with the file of the D/F conflict 510 * appearing before files below the corresponding directory. 511 * The order of the rest of the list is irrelevant for us. 512 * 513 * To achieve this, we sort with df_name_compare and provide 514 * the mode S_IFDIR so that D/F conflicts will sort correctly. 515 * We use the mode S_IFDIR for everything else for simplicity, 516 * since in other cases any changes in their order due to 517 * sorting cause no problems for us. 518 */ 519 int cmp = df_name_compare(one, onelen, S_IFDIR, 520 two, twolen, S_IFDIR); 521 /* 522 * Now that 'foo' and 'foo/bar' compare equal, we have to make sure 523 * that 'foo' comes before 'foo/bar'. 524 */ 525 if (cmp) 526 return cmp; 527 return onelen - twolen; 528} 529 530static void record_df_conflict_files(struct merge_options *o, 531 struct string_list *entries) 532{ 533 /* If there is a D/F conflict and the file for such a conflict 534 * currently exist in the working tree, we want to allow it to be 535 * removed to make room for the corresponding directory if needed. 536 * The files underneath the directories of such D/F conflicts will 537 * be processed before the corresponding file involved in the D/F 538 * conflict. If the D/F directory ends up being removed by the 539 * merge, then we won't have to touch the D/F file. If the D/F 540 * directory needs to be written to the working copy, then the D/F 541 * file will simply be removed (in make_room_for_path()) to make 542 * room for the necessary paths. Note that if both the directory 543 * and the file need to be present, then the D/F file will be 544 * reinstated with a new unique name at the time it is processed. 545 */ 546 struct string_list df_sorted_entries = STRING_LIST_INIT_NODUP; 547 const char *last_file = NULL; 548 int last_len = 0; 549 int i; 550 551 /* 552 * If we're merging merge-bases, we don't want to bother with 553 * any working directory changes. 554 */ 555 if (o->call_depth) 556 return; 557 558 /* Ensure D/F conflicts are adjacent in the entries list. */ 559 for (i = 0; i < entries->nr; i++) { 560 struct string_list_item *next = &entries->items[i]; 561 string_list_append(&df_sorted_entries, next->string)->util = 562 next->util; 563 } 564 df_sorted_entries.cmp = string_list_df_name_compare; 565 string_list_sort(&df_sorted_entries); 566 567 string_list_clear(&o->df_conflict_file_set, 1); 568 for (i = 0; i < df_sorted_entries.nr; i++) { 569 const char *path = df_sorted_entries.items[i].string; 570 int len = strlen(path); 571 struct stage_data *e = df_sorted_entries.items[i].util; 572 573 /* 574 * Check if last_file & path correspond to a D/F conflict; 575 * i.e. whether path is last_file+'/'+<something>. 576 * If so, record that it's okay to remove last_file to make 577 * room for path and friends if needed. 578 */ 579 if (last_file && 580 len > last_len && 581 memcmp(path, last_file, last_len) == 0 && 582 path[last_len] == '/') { 583 string_list_insert(&o->df_conflict_file_set, last_file); 584 } 585 586 /* 587 * Determine whether path could exist as a file in the 588 * working directory as a possible D/F conflict. This 589 * will only occur when it exists in stage 2 as a 590 * file. 591 */ 592 if (S_ISREG(e->stages[2].mode) || S_ISLNK(e->stages[2].mode)) { 593 last_file = path; 594 last_len = len; 595 } else { 596 last_file = NULL; 597 } 598 } 599 string_list_clear(&df_sorted_entries, 0); 600} 601 602struct rename { 603 struct diff_filepair *pair; 604 /* 605 * Purpose of src_entry and dst_entry: 606 * 607 * If 'before' is renamed to 'after' then src_entry will contain 608 * the versions of 'before' from the merge_base, HEAD, and MERGE in 609 * stages 1, 2, and 3; dst_entry will contain the respective 610 * versions of 'after' in corresponding locations. Thus, we have a 611 * total of six modes and oids, though some will be null. (Stage 0 612 * is ignored; we're interested in handling conflicts.) 613 * 614 * Since we don't turn on break-rewrites by default, neither 615 * src_entry nor dst_entry can have all three of their stages have 616 * non-null oids, meaning at most four of the six will be non-null. 617 * Also, since this is a rename, both src_entry and dst_entry will 618 * have at least one non-null oid, meaning at least two will be 619 * non-null. Of the six oids, a typical rename will have three be 620 * non-null. Only two implies a rename/delete, and four implies a 621 * rename/add. 622 */ 623 struct stage_data *src_entry; 624 struct stage_data *dst_entry; 625 unsigned add_turned_into_rename:1; 626 unsigned processed:1; 627}; 628 629static int update_stages(struct merge_options *opt, const char *path, 630 const struct diff_filespec *o, 631 const struct diff_filespec *a, 632 const struct diff_filespec *b) 633{ 634 635 /* 636 * NOTE: It is usually a bad idea to call update_stages on a path 637 * before calling update_file on that same path, since it can 638 * sometimes lead to spurious "refusing to lose untracked file..." 639 * messages from update_file (via make_room_for path via 640 * would_lose_untracked). Instead, reverse the order of the calls 641 * (executing update_file first and then update_stages). 642 */ 643 int clear = 1; 644 int options = ADD_CACHE_OK_TO_ADD | ADD_CACHE_SKIP_DFCHECK; 645 if (clear) 646 if (remove_file_from_cache(path)) 647 return -1; 648 if (o) 649 if (add_cacheinfo(opt, o->mode, &o->oid, path, 1, 0, options)) 650 return -1; 651 if (a) 652 if (add_cacheinfo(opt, a->mode, &a->oid, path, 2, 0, options)) 653 return -1; 654 if (b) 655 if (add_cacheinfo(opt, b->mode, &b->oid, path, 3, 0, options)) 656 return -1; 657 return 0; 658} 659 660static int update_stages_for_stage_data(struct merge_options *opt, 661 const char *path, 662 const struct stage_data *stage_data) 663{ 664 struct diff_filespec o, a, b; 665 666 o.mode = stage_data->stages[1].mode; 667 oidcpy(&o.oid, &stage_data->stages[1].oid); 668 669 a.mode = stage_data->stages[2].mode; 670 oidcpy(&a.oid, &stage_data->stages[2].oid); 671 672 b.mode = stage_data->stages[3].mode; 673 oidcpy(&b.oid, &stage_data->stages[3].oid); 674 675 return update_stages(opt, path, 676 is_null_oid(&o.oid) ? NULL : &o, 677 is_null_oid(&a.oid) ? NULL : &a, 678 is_null_oid(&b.oid) ? NULL : &b); 679} 680 681static void update_entry(struct stage_data *entry, 682 struct diff_filespec *o, 683 struct diff_filespec *a, 684 struct diff_filespec *b) 685{ 686 entry->processed = 0; 687 entry->stages[1].mode = o->mode; 688 entry->stages[2].mode = a->mode; 689 entry->stages[3].mode = b->mode; 690 oidcpy(&entry->stages[1].oid, &o->oid); 691 oidcpy(&entry->stages[2].oid, &a->oid); 692 oidcpy(&entry->stages[3].oid, &b->oid); 693} 694 695static int remove_file(struct merge_options *o, int clean, 696 const char *path, int no_wd) 697{ 698 int update_cache = o->call_depth || clean; 699 int update_working_directory = !o->call_depth && !no_wd; 700 701 if (update_cache) { 702 if (remove_file_from_cache(path)) 703 return -1; 704 } 705 if (update_working_directory) { 706 if (ignore_case) { 707 struct cache_entry *ce; 708 ce = cache_file_exists(path, strlen(path), ignore_case); 709 if (ce && ce_stage(ce) == 0 && strcmp(path, ce->name)) 710 return 0; 711 } 712 if (remove_path(path)) 713 return -1; 714 } 715 return 0; 716} 717 718/* add a string to a strbuf, but converting "/" to "_" */ 719static void add_flattened_path(struct strbuf *out, const char *s) 720{ 721 size_t i = out->len; 722 strbuf_addstr(out, s); 723 for (; i < out->len; i++) 724 if (out->buf[i] == '/') 725 out->buf[i] = '_'; 726} 727 728static char *unique_path(struct merge_options *o, const char *path, const char *branch) 729{ 730 struct path_hashmap_entry *entry; 731 struct strbuf newpath = STRBUF_INIT; 732 int suffix = 0; 733 size_t base_len; 734 735 strbuf_addf(&newpath, "%s~", path); 736 add_flattened_path(&newpath, branch); 737 738 base_len = newpath.len; 739 while (hashmap_get_from_hash(&o->current_file_dir_set, 740 path_hash(newpath.buf), newpath.buf) || 741 (!o->call_depth && file_exists(newpath.buf))) { 742 strbuf_setlen(&newpath, base_len); 743 strbuf_addf(&newpath, "_%d", suffix++); 744 } 745 746 FLEX_ALLOC_MEM(entry, path, newpath.buf, newpath.len); 747 hashmap_entry_init(entry, path_hash(entry->path)); 748 hashmap_add(&o->current_file_dir_set, entry); 749 return strbuf_detach(&newpath, NULL); 750} 751 752/** 753 * Check whether a directory in the index is in the way of an incoming 754 * file. Return 1 if so. If check_working_copy is non-zero, also 755 * check the working directory. If empty_ok is non-zero, also return 756 * 0 in the case where the working-tree dir exists but is empty. 757 */ 758static int dir_in_way(const char *path, int check_working_copy, int empty_ok) 759{ 760 int pos; 761 struct strbuf dirpath = STRBUF_INIT; 762 struct stat st; 763 764 strbuf_addstr(&dirpath, path); 765 strbuf_addch(&dirpath, '/'); 766 767 pos = cache_name_pos(dirpath.buf, dirpath.len); 768 769 if (pos < 0) 770 pos = -1 - pos; 771 if (pos < active_nr && 772 !strncmp(dirpath.buf, active_cache[pos]->name, dirpath.len)) { 773 strbuf_release(&dirpath); 774 return 1; 775 } 776 777 strbuf_release(&dirpath); 778 return check_working_copy && !lstat(path, &st) && S_ISDIR(st.st_mode) && 779 !(empty_ok && is_empty_dir(path)); 780} 781 782/* 783 * Returns whether path was tracked in the index before the merge started, 784 * and its oid and mode match the specified values 785 */ 786static int was_tracked_and_matches(struct merge_options *o, const char *path, 787 const struct object_id *oid, unsigned mode) 788{ 789 int pos = index_name_pos(&o->orig_index, path, strlen(path)); 790 struct cache_entry *ce; 791 792 if (0 > pos) 793 /* we were not tracking this path before the merge */ 794 return 0; 795 796 /* See if the file we were tracking before matches */ 797 ce = o->orig_index.cache[pos]; 798 return (oid_eq(&ce->oid, oid) && ce->ce_mode == mode); 799} 800 801/* 802 * Returns whether path was tracked in the index before the merge started 803 */ 804static int was_tracked(struct merge_options *o, const char *path) 805{ 806 int pos = index_name_pos(&o->orig_index, path, strlen(path)); 807 808 if (0 <= pos) 809 /* we were tracking this path before the merge */ 810 return 1; 811 812 return 0; 813} 814 815static int would_lose_untracked(const char *path) 816{ 817 /* 818 * This may look like it can be simplified to: 819 * return !was_tracked(o, path) && file_exists(path) 820 * but it can't. This function needs to know whether path was in 821 * the working tree due to EITHER having been tracked in the index 822 * before the merge OR having been put into the working copy and 823 * index by unpack_trees(). Due to that either-or requirement, we 824 * check the current index instead of the original one. 825 * 826 * Note that we do not need to worry about merge-recursive itself 827 * updating the index after unpack_trees() and before calling this 828 * function, because we strictly require all code paths in 829 * merge-recursive to update the working tree first and the index 830 * second. Doing otherwise would break 831 * update_file()/would_lose_untracked(); see every comment in this 832 * file which mentions "update_stages". 833 */ 834 int pos = cache_name_pos(path, strlen(path)); 835 836 if (pos < 0) 837 pos = -1 - pos; 838 while (pos < active_nr && 839 !strcmp(path, active_cache[pos]->name)) { 840 /* 841 * If stage #0, it is definitely tracked. 842 * If it has stage #2 then it was tracked 843 * before this merge started. All other 844 * cases the path was not tracked. 845 */ 846 switch (ce_stage(active_cache[pos])) { 847 case 0: 848 case 2: 849 return 0; 850 } 851 pos++; 852 } 853 return file_exists(path); 854} 855 856static int was_dirty(struct merge_options *o, const char *path) 857{ 858 struct cache_entry *ce; 859 int dirty = 1; 860 861 if (o->call_depth || !was_tracked(o, path)) 862 return !dirty; 863 864 ce = index_file_exists(o->unpack_opts.src_index, 865 path, strlen(path), ignore_case); 866 dirty = verify_uptodate(ce, &o->unpack_opts) != 0; 867 return dirty; 868} 869 870static int make_room_for_path(struct merge_options *o, const char *path) 871{ 872 int status, i; 873 const char *msg = _("failed to create path '%s'%s"); 874 875 /* Unlink any D/F conflict files that are in the way */ 876 for (i = 0; i < o->df_conflict_file_set.nr; i++) { 877 const char *df_path = o->df_conflict_file_set.items[i].string; 878 size_t pathlen = strlen(path); 879 size_t df_pathlen = strlen(df_path); 880 if (df_pathlen < pathlen && 881 path[df_pathlen] == '/' && 882 strncmp(path, df_path, df_pathlen) == 0) { 883 output(o, 3, 884 _("Removing %s to make room for subdirectory\n"), 885 df_path); 886 unlink(df_path); 887 unsorted_string_list_delete_item(&o->df_conflict_file_set, 888 i, 0); 889 break; 890 } 891 } 892 893 /* Make sure leading directories are created */ 894 status = safe_create_leading_directories_const(path); 895 if (status) { 896 if (status == SCLD_EXISTS) 897 /* something else exists */ 898 return err(o, msg, path, _(": perhaps a D/F conflict?")); 899 return err(o, msg, path, ""); 900 } 901 902 /* 903 * Do not unlink a file in the work tree if we are not 904 * tracking it. 905 */ 906 if (would_lose_untracked(path)) 907 return err(o, _("refusing to lose untracked file at '%s'"), 908 path); 909 910 /* Successful unlink is good.. */ 911 if (!unlink(path)) 912 return 0; 913 /* .. and so is no existing file */ 914 if (errno == ENOENT) 915 return 0; 916 /* .. but not some other error (who really cares what?) */ 917 return err(o, msg, path, _(": perhaps a D/F conflict?")); 918} 919 920static int update_file_flags(struct merge_options *o, 921 const struct object_id *oid, 922 unsigned mode, 923 const char *path, 924 int update_cache, 925 int update_wd) 926{ 927 int ret = 0; 928 929 if (o->call_depth) 930 update_wd = 0; 931 932 if (update_wd) { 933 enum object_type type; 934 void *buf; 935 unsigned long size; 936 937 if (S_ISGITLINK(mode)) { 938 /* 939 * We may later decide to recursively descend into 940 * the submodule directory and update its index 941 * and/or work tree, but we do not do that now. 942 */ 943 update_wd = 0; 944 goto update_index; 945 } 946 947 buf = read_object_file(oid, &type, &size); 948 if (!buf) 949 return err(o, _("cannot read object %s '%s'"), oid_to_hex(oid), path); 950 if (type != OBJ_BLOB) { 951 ret = err(o, _("blob expected for %s '%s'"), oid_to_hex(oid), path); 952 goto free_buf; 953 } 954 if (S_ISREG(mode)) { 955 struct strbuf strbuf = STRBUF_INIT; 956 if (convert_to_working_tree(path, buf, size, &strbuf)) { 957 free(buf); 958 size = strbuf.len; 959 buf = strbuf_detach(&strbuf, NULL); 960 } 961 } 962 963 if (make_room_for_path(o, path) < 0) { 964 update_wd = 0; 965 goto free_buf; 966 } 967 if (S_ISREG(mode) || (!has_symlinks && S_ISLNK(mode))) { 968 int fd; 969 if (mode & 0100) 970 mode = 0777; 971 else 972 mode = 0666; 973 fd = open(path, O_WRONLY | O_TRUNC | O_CREAT, mode); 974 if (fd < 0) { 975 ret = err(o, _("failed to open '%s': %s"), 976 path, strerror(errno)); 977 goto free_buf; 978 } 979 write_in_full(fd, buf, size); 980 close(fd); 981 } else if (S_ISLNK(mode)) { 982 char *lnk = xmemdupz(buf, size); 983 safe_create_leading_directories_const(path); 984 unlink(path); 985 if (symlink(lnk, path)) 986 ret = err(o, _("failed to symlink '%s': %s"), 987 path, strerror(errno)); 988 free(lnk); 989 } else 990 ret = err(o, 991 _("do not know what to do with %06o %s '%s'"), 992 mode, oid_to_hex(oid), path); 993 free_buf: 994 free(buf); 995 } 996 update_index: 997 if (!ret && update_cache) 998 if (add_cacheinfo(o, mode, oid, path, 0, update_wd, 999 ADD_CACHE_OK_TO_ADD))1000 return -1;1001 return ret;1002}10031004static int update_file(struct merge_options *o,1005 int clean,1006 const struct object_id *oid,1007 unsigned mode,1008 const char *path)1009{1010 return update_file_flags(o, oid, mode, path, o->call_depth || clean, !o->call_depth);1011}10121013/* Low level file merging, update and removal */10141015struct merge_file_info {1016 struct object_id oid;1017 unsigned mode;1018 unsigned clean:1,1019 merge:1;1020};10211022static int merge_3way(struct merge_options *o,1023 mmbuffer_t *result_buf,1024 const struct diff_filespec *one,1025 const struct diff_filespec *a,1026 const struct diff_filespec *b,1027 const char *branch1,1028 const char *branch2)1029{1030 mmfile_t orig, src1, src2;1031 struct ll_merge_options ll_opts = {0};1032 char *base_name, *name1, *name2;1033 int merge_status;10341035 ll_opts.renormalize = o->renormalize;1036 ll_opts.xdl_opts = o->xdl_opts;10371038 if (o->call_depth) {1039 ll_opts.virtual_ancestor = 1;1040 ll_opts.variant = 0;1041 } else {1042 switch (o->recursive_variant) {1043 case MERGE_RECURSIVE_OURS:1044 ll_opts.variant = XDL_MERGE_FAVOR_OURS;1045 break;1046 case MERGE_RECURSIVE_THEIRS:1047 ll_opts.variant = XDL_MERGE_FAVOR_THEIRS;1048 break;1049 default:1050 ll_opts.variant = 0;1051 break;1052 }1053 }10541055 if (strcmp(a->path, b->path) ||1056 (o->ancestor != NULL && strcmp(a->path, one->path) != 0)) {1057 base_name = o->ancestor == NULL ? NULL :1058 mkpathdup("%s:%s", o->ancestor, one->path);1059 name1 = mkpathdup("%s:%s", branch1, a->path);1060 name2 = mkpathdup("%s:%s", branch2, b->path);1061 } else {1062 base_name = o->ancestor == NULL ? NULL :1063 mkpathdup("%s", o->ancestor);1064 name1 = mkpathdup("%s", branch1);1065 name2 = mkpathdup("%s", branch2);1066 }10671068 read_mmblob(&orig, &one->oid);1069 read_mmblob(&src1, &a->oid);1070 read_mmblob(&src2, &b->oid);10711072 merge_status = ll_merge(result_buf, a->path, &orig, base_name,1073 &src1, name1, &src2, name2, &ll_opts);10741075 free(base_name);1076 free(name1);1077 free(name2);1078 free(orig.ptr);1079 free(src1.ptr);1080 free(src2.ptr);1081 return merge_status;1082}10831084static int merge_file_1(struct merge_options *o,1085 const struct diff_filespec *one,1086 const struct diff_filespec *a,1087 const struct diff_filespec *b,1088 const char *filename,1089 const char *branch1,1090 const char *branch2,1091 struct merge_file_info *result)1092{1093 result->merge = 0;1094 result->clean = 1;10951096 if ((S_IFMT & a->mode) != (S_IFMT & b->mode)) {1097 result->clean = 0;1098 if (S_ISREG(a->mode)) {1099 result->mode = a->mode;1100 oidcpy(&result->oid, &a->oid);1101 } else {1102 result->mode = b->mode;1103 oidcpy(&result->oid, &b->oid);1104 }1105 } else {1106 if (!oid_eq(&a->oid, &one->oid) && !oid_eq(&b->oid, &one->oid))1107 result->merge = 1;11081109 /*1110 * Merge modes1111 */1112 if (a->mode == b->mode || a->mode == one->mode)1113 result->mode = b->mode;1114 else {1115 result->mode = a->mode;1116 if (b->mode != one->mode) {1117 result->clean = 0;1118 result->merge = 1;1119 }1120 }11211122 if (oid_eq(&a->oid, &b->oid) || oid_eq(&a->oid, &one->oid))1123 oidcpy(&result->oid, &b->oid);1124 else if (oid_eq(&b->oid, &one->oid))1125 oidcpy(&result->oid, &a->oid);1126 else if (S_ISREG(a->mode)) {1127 mmbuffer_t result_buf;1128 int ret = 0, merge_status;11291130 merge_status = merge_3way(o, &result_buf, one, a, b,1131 branch1, branch2);11321133 if ((merge_status < 0) || !result_buf.ptr)1134 ret = err(o, _("Failed to execute internal merge"));11351136 if (!ret &&1137 write_object_file(result_buf.ptr, result_buf.size,1138 blob_type, &result->oid))1139 ret = err(o, _("Unable to add %s to database"),1140 a->path);11411142 free(result_buf.ptr);1143 if (ret)1144 return ret;1145 result->clean = (merge_status == 0);1146 } else if (S_ISGITLINK(a->mode)) {1147 result->clean = merge_submodule(&result->oid,1148 one->path,1149 &one->oid,1150 &a->oid,1151 &b->oid,1152 !o->call_depth);1153 } else if (S_ISLNK(a->mode)) {1154 switch (o->recursive_variant) {1155 case MERGE_RECURSIVE_NORMAL:1156 oidcpy(&result->oid, &a->oid);1157 if (!oid_eq(&a->oid, &b->oid))1158 result->clean = 0;1159 break;1160 case MERGE_RECURSIVE_OURS:1161 oidcpy(&result->oid, &a->oid);1162 break;1163 case MERGE_RECURSIVE_THEIRS:1164 oidcpy(&result->oid, &b->oid);1165 break;1166 }1167 } else1168 die("BUG: unsupported object type in the tree");1169 }11701171 if (result->merge)1172 output(o, 2, _("Auto-merging %s"), filename);11731174 return 0;1175}11761177static int merge_file_special_markers(struct merge_options *o,1178 const struct diff_filespec *one,1179 const struct diff_filespec *a,1180 const struct diff_filespec *b,1181 const char *target_filename,1182 const char *branch1,1183 const char *filename1,1184 const char *branch2,1185 const char *filename2,1186 struct merge_file_info *mfi)1187{1188 char *side1 = NULL;1189 char *side2 = NULL;1190 int ret;11911192 if (filename1)1193 side1 = xstrfmt("%s:%s", branch1, filename1);1194 if (filename2)1195 side2 = xstrfmt("%s:%s", branch2, filename2);11961197 ret = merge_file_1(o, one, a, b, target_filename,1198 side1 ? side1 : branch1,1199 side2 ? side2 : branch2, mfi);12001201 free(side1);1202 free(side2);1203 return ret;1204}12051206static int merge_file_one(struct merge_options *o,1207 const char *path,1208 const struct object_id *o_oid, int o_mode,1209 const struct object_id *a_oid, int a_mode,1210 const struct object_id *b_oid, int b_mode,1211 const char *branch1,1212 const char *branch2,1213 struct merge_file_info *mfi)1214{1215 struct diff_filespec one, a, b;12161217 one.path = a.path = b.path = (char *)path;1218 oidcpy(&one.oid, o_oid);1219 one.mode = o_mode;1220 oidcpy(&a.oid, a_oid);1221 a.mode = a_mode;1222 oidcpy(&b.oid, b_oid);1223 b.mode = b_mode;1224 return merge_file_1(o, &one, &a, &b, path, branch1, branch2, mfi);1225}12261227static int conflict_rename_dir(struct merge_options *o,1228 struct diff_filepair *pair,1229 const char *rename_branch,1230 const char *other_branch)1231{1232 const struct diff_filespec *dest = pair->two;12331234 if (!o->call_depth && would_lose_untracked(dest->path)) {1235 char *alt_path = unique_path(o, dest->path, rename_branch);12361237 output(o, 1, _("Error: Refusing to lose untracked file at %s; "1238 "writing to %s instead."),1239 dest->path, alt_path);1240 /*1241 * Write the file in worktree at alt_path, but not in the1242 * index. Instead, write to dest->path for the index but1243 * only at the higher appropriate stage.1244 */1245 if (update_file(o, 0, &dest->oid, dest->mode, alt_path))1246 return -1;1247 free(alt_path);1248 return update_stages(o, dest->path, NULL,1249 rename_branch == o->branch1 ? dest : NULL,1250 rename_branch == o->branch1 ? NULL : dest);1251 }12521253 /* Update dest->path both in index and in worktree */1254 if (update_file(o, 1, &dest->oid, dest->mode, dest->path))1255 return -1;1256 return 0;1257}12581259static int handle_change_delete(struct merge_options *o,1260 const char *path, const char *old_path,1261 const struct object_id *o_oid, int o_mode,1262 const struct object_id *changed_oid,1263 int changed_mode,1264 const char *change_branch,1265 const char *delete_branch,1266 const char *change, const char *change_past)1267{1268 char *alt_path = NULL;1269 const char *update_path = path;1270 int ret = 0;12711272 if (dir_in_way(path, !o->call_depth, 0) ||1273 (!o->call_depth && would_lose_untracked(path))) {1274 update_path = alt_path = unique_path(o, path, change_branch);1275 }12761277 if (o->call_depth) {1278 /*1279 * We cannot arbitrarily accept either a_sha or b_sha as1280 * correct; since there is no true "middle point" between1281 * them, simply reuse the base version for virtual merge base.1282 */1283 ret = remove_file_from_cache(path);1284 if (!ret)1285 ret = update_file(o, 0, o_oid, o_mode, update_path);1286 } else {1287 if (!alt_path) {1288 if (!old_path) {1289 output(o, 1, _("CONFLICT (%s/delete): %s deleted in %s "1290 "and %s in %s. Version %s of %s left in tree."),1291 change, path, delete_branch, change_past,1292 change_branch, change_branch, path);1293 } else {1294 output(o, 1, _("CONFLICT (%s/delete): %s deleted in %s "1295 "and %s to %s in %s. Version %s of %s left in tree."),1296 change, old_path, delete_branch, change_past, path,1297 change_branch, change_branch, path);1298 }1299 } else {1300 if (!old_path) {1301 output(o, 1, _("CONFLICT (%s/delete): %s deleted in %s "1302 "and %s in %s. Version %s of %s left in tree at %s."),1303 change, path, delete_branch, change_past,1304 change_branch, change_branch, path, alt_path);1305 } else {1306 output(o, 1, _("CONFLICT (%s/delete): %s deleted in %s "1307 "and %s to %s in %s. Version %s of %s left in tree at %s."),1308 change, old_path, delete_branch, change_past, path,1309 change_branch, change_branch, path, alt_path);1310 }1311 }1312 /*1313 * No need to call update_file() on path when change_branch ==1314 * o->branch1 && !alt_path, since that would needlessly touch1315 * path. We could call update_file_flags() with update_cache=01316 * and update_wd=0, but that's a no-op.1317 */1318 if (change_branch != o->branch1 || alt_path)1319 ret = update_file(o, 0, changed_oid, changed_mode, update_path);1320 }1321 free(alt_path);13221323 return ret;1324}13251326static int conflict_rename_delete(struct merge_options *o,1327 struct diff_filepair *pair,1328 const char *rename_branch,1329 const char *delete_branch)1330{1331 const struct diff_filespec *orig = pair->one;1332 const struct diff_filespec *dest = pair->two;13331334 if (handle_change_delete(o,1335 o->call_depth ? orig->path : dest->path,1336 o->call_depth ? NULL : orig->path,1337 &orig->oid, orig->mode,1338 &dest->oid, dest->mode,1339 rename_branch, delete_branch,1340 _("rename"), _("renamed")))1341 return -1;13421343 if (o->call_depth)1344 return remove_file_from_cache(dest->path);1345 else1346 return update_stages(o, dest->path, NULL,1347 rename_branch == o->branch1 ? dest : NULL,1348 rename_branch == o->branch1 ? NULL : dest);1349}13501351static struct diff_filespec *filespec_from_entry(struct diff_filespec *target,1352 struct stage_data *entry,1353 int stage)1354{1355 struct object_id *oid = &entry->stages[stage].oid;1356 unsigned mode = entry->stages[stage].mode;1357 if (mode == 0 || is_null_oid(oid))1358 return NULL;1359 oidcpy(&target->oid, oid);1360 target->mode = mode;1361 return target;1362}13631364static int handle_file(struct merge_options *o,1365 struct diff_filespec *rename,1366 int stage,1367 struct rename_conflict_info *ci)1368{1369 char *dst_name = rename->path;1370 struct stage_data *dst_entry;1371 const char *cur_branch, *other_branch;1372 struct diff_filespec other;1373 struct diff_filespec *add;1374 int ret;13751376 if (stage == 2) {1377 dst_entry = ci->dst_entry1;1378 cur_branch = ci->branch1;1379 other_branch = ci->branch2;1380 } else {1381 dst_entry = ci->dst_entry2;1382 cur_branch = ci->branch2;1383 other_branch = ci->branch1;1384 }13851386 add = filespec_from_entry(&other, dst_entry, stage ^ 1);1387 if (add) {1388 int ren_src_was_dirty = was_dirty(o, rename->path);1389 char *add_name = unique_path(o, rename->path, other_branch);1390 if (update_file(o, 0, &add->oid, add->mode, add_name))1391 return -1;13921393 if (ren_src_was_dirty) {1394 output(o, 1, _("Refusing to lose dirty file at %s"),1395 rename->path);1396 }1397 /*1398 * Because the double negatives somehow keep confusing me...1399 * 1) update_wd iff !ren_src_was_dirty.1400 * 2) no_wd iff !update_wd1401 * 3) so, no_wd == !!ren_src_was_dirty == ren_src_was_dirty1402 */1403 remove_file(o, 0, rename->path, ren_src_was_dirty);1404 dst_name = unique_path(o, rename->path, cur_branch);1405 } else {1406 if (dir_in_way(rename->path, !o->call_depth, 0)) {1407 dst_name = unique_path(o, rename->path, cur_branch);1408 output(o, 1, _("%s is a directory in %s adding as %s instead"),1409 rename->path, other_branch, dst_name);1410 } else if (!o->call_depth &&1411 would_lose_untracked(rename->path)) {1412 dst_name = unique_path(o, rename->path, cur_branch);1413 output(o, 1, _("Refusing to lose untracked file at %s; "1414 "adding as %s instead"),1415 rename->path, dst_name);1416 }1417 }1418 if ((ret = update_file(o, 0, &rename->oid, rename->mode, dst_name)))1419 ; /* fall through, do allow dst_name to be released */1420 else if (stage == 2)1421 ret = update_stages(o, rename->path, NULL, rename, add);1422 else1423 ret = update_stages(o, rename->path, NULL, add, rename);14241425 if (dst_name != rename->path)1426 free(dst_name);14271428 return ret;1429}14301431static int conflict_rename_rename_1to2(struct merge_options *o,1432 struct rename_conflict_info *ci)1433{1434 /* One file was renamed in both branches, but to different names. */1435 struct diff_filespec *one = ci->pair1->one;1436 struct diff_filespec *a = ci->pair1->two;1437 struct diff_filespec *b = ci->pair2->two;14381439 output(o, 1, _("CONFLICT (rename/rename): "1440 "Rename \"%s\"->\"%s\" in branch \"%s\" "1441 "rename \"%s\"->\"%s\" in \"%s\"%s"),1442 one->path, a->path, ci->branch1,1443 one->path, b->path, ci->branch2,1444 o->call_depth ? _(" (left unresolved)") : "");1445 if (o->call_depth) {1446 struct merge_file_info mfi;1447 struct diff_filespec other;1448 struct diff_filespec *add;1449 if (merge_file_one(o, one->path,1450 &one->oid, one->mode,1451 &a->oid, a->mode,1452 &b->oid, b->mode,1453 ci->branch1, ci->branch2, &mfi))1454 return -1;14551456 /*1457 * FIXME: For rename/add-source conflicts (if we could detect1458 * such), this is wrong. We should instead find a unique1459 * pathname and then either rename the add-source file to that1460 * unique path, or use that unique path instead of src here.1461 */1462 if (update_file(o, 0, &mfi.oid, mfi.mode, one->path))1463 return -1;14641465 /*1466 * Above, we put the merged content at the merge-base's1467 * path. Now we usually need to delete both a->path and1468 * b->path. However, the rename on each side of the merge1469 * could also be involved in a rename/add conflict. In1470 * such cases, we should keep the added file around,1471 * resolving the conflict at that path in its favor.1472 */1473 add = filespec_from_entry(&other, ci->dst_entry1, 2 ^ 1);1474 if (add) {1475 if (update_file(o, 0, &add->oid, add->mode, a->path))1476 return -1;1477 }1478 else1479 remove_file_from_cache(a->path);1480 add = filespec_from_entry(&other, ci->dst_entry2, 3 ^ 1);1481 if (add) {1482 if (update_file(o, 0, &add->oid, add->mode, b->path))1483 return -1;1484 }1485 else1486 remove_file_from_cache(b->path);1487 } else if (handle_file(o, a, 2, ci) || handle_file(o, b, 3, ci))1488 return -1;14891490 return 0;1491}14921493static int conflict_rename_rename_2to1(struct merge_options *o,1494 struct rename_conflict_info *ci)1495{1496 /* Two files, a & b, were renamed to the same thing, c. */1497 struct diff_filespec *a = ci->pair1->one;1498 struct diff_filespec *b = ci->pair2->one;1499 struct diff_filespec *c1 = ci->pair1->two;1500 struct diff_filespec *c2 = ci->pair2->two;1501 char *path = c1->path; /* == c2->path */1502 char *path_side_1_desc;1503 char *path_side_2_desc;1504 struct merge_file_info mfi_c1;1505 struct merge_file_info mfi_c2;1506 int ret;15071508 output(o, 1, _("CONFLICT (rename/rename): "1509 "Rename %s->%s in %s. "1510 "Rename %s->%s in %s"),1511 a->path, c1->path, ci->branch1,1512 b->path, c2->path, ci->branch2);15131514 remove_file(o, 1, a->path, o->call_depth || would_lose_untracked(a->path));1515 remove_file(o, 1, b->path, o->call_depth || would_lose_untracked(b->path));15161517 path_side_1_desc = xstrfmt("%s (was %s)", path, a->path);1518 path_side_2_desc = xstrfmt("%s (was %s)", path, b->path);1519 if (merge_file_special_markers(o, a, c1, &ci->ren1_other,1520 path_side_1_desc,1521 o->branch1, c1->path,1522 o->branch2, ci->ren1_other.path, &mfi_c1) ||1523 merge_file_special_markers(o, b, &ci->ren2_other, c2,1524 path_side_2_desc,1525 o->branch1, ci->ren2_other.path,1526 o->branch2, c2->path, &mfi_c2))1527 return -1;1528 free(path_side_1_desc);1529 free(path_side_2_desc);15301531 if (o->call_depth) {1532 /*1533 * If mfi_c1.clean && mfi_c2.clean, then it might make1534 * sense to do a two-way merge of those results. But, I1535 * think in all cases, it makes sense to have the virtual1536 * merge base just undo the renames; they can be detected1537 * again later for the non-recursive merge.1538 */1539 remove_file(o, 0, path, 0);1540 ret = update_file(o, 0, &mfi_c1.oid, mfi_c1.mode, a->path);1541 if (!ret)1542 ret = update_file(o, 0, &mfi_c2.oid, mfi_c2.mode,1543 b->path);1544 } else {1545 char *new_path1 = unique_path(o, path, ci->branch1);1546 char *new_path2 = unique_path(o, path, ci->branch2);1547 output(o, 1, _("Renaming %s to %s and %s to %s instead"),1548 a->path, new_path1, b->path, new_path2);1549 if (was_dirty(o, path))1550 output(o, 1, _("Refusing to lose dirty file at %s"),1551 path);1552 else if (would_lose_untracked(path))1553 /*1554 * Only way we get here is if both renames were from1555 * a directory rename AND user had an untracked file1556 * at the location where both files end up after the1557 * two directory renames. See testcase 10d of t6043.1558 */1559 output(o, 1, _("Refusing to lose untracked file at "1560 "%s, even though it's in the way."),1561 path);1562 else1563 remove_file(o, 0, path, 0);1564 ret = update_file(o, 0, &mfi_c1.oid, mfi_c1.mode, new_path1);1565 if (!ret)1566 ret = update_file(o, 0, &mfi_c2.oid, mfi_c2.mode,1567 new_path2);1568 /*1569 * unpack_trees() actually populates the index for us for1570 * "normal" rename/rename(2to1) situtations so that the1571 * correct entries are at the higher stages, which would1572 * make the call below to update_stages_for_stage_data1573 * unnecessary. However, if either of the renames came1574 * from a directory rename, then unpack_trees() will not1575 * have gotten the right data loaded into the index, so we1576 * need to do so now. (While it'd be tempting to move this1577 * call to update_stages_for_stage_data() to1578 * apply_directory_rename_modifications(), that would break1579 * our intermediate calls to would_lose_untracked() since1580 * those rely on the current in-memory index. See also the1581 * big "NOTE" in update_stages()).1582 */1583 if (update_stages_for_stage_data(o, path, ci->dst_entry1))1584 ret = -1;15851586 free(new_path2);1587 free(new_path1);1588 }15891590 return ret;1591}15921593/*1594 * Get the diff_filepairs changed between o_tree and tree.1595 */1596static struct diff_queue_struct *get_diffpairs(struct merge_options *o,1597 struct tree *o_tree,1598 struct tree *tree)1599{1600 struct diff_queue_struct *ret;1601 struct diff_options opts;16021603 diff_setup(&opts);1604 opts.flags.recursive = 1;1605 opts.flags.rename_empty = 0;1606 opts.detect_rename = merge_detect_rename(o);1607 /*1608 * We do not have logic to handle the detection of copies. In1609 * fact, it may not even make sense to add such logic: would we1610 * really want a change to a base file to be propagated through1611 * multiple other files by a merge?1612 */1613 if (opts.detect_rename > DIFF_DETECT_RENAME)1614 opts.detect_rename = DIFF_DETECT_RENAME;1615 opts.rename_limit = o->merge_rename_limit >= 0 ? o->merge_rename_limit :1616 o->diff_rename_limit >= 0 ? o->diff_rename_limit :1617 1000;1618 opts.rename_score = o->rename_score;1619 opts.show_rename_progress = o->show_rename_progress;1620 opts.output_format = DIFF_FORMAT_NO_OUTPUT;1621 diff_setup_done(&opts);1622 diff_tree_oid(&o_tree->object.oid, &tree->object.oid, "", &opts);1623 diffcore_std(&opts);1624 if (opts.needed_rename_limit > o->needed_rename_limit)1625 o->needed_rename_limit = opts.needed_rename_limit;16261627 ret = xmalloc(sizeof(*ret));1628 *ret = diff_queued_diff;16291630 opts.output_format = DIFF_FORMAT_NO_OUTPUT;1631 diff_queued_diff.nr = 0;1632 diff_queued_diff.queue = NULL;1633 diff_flush(&opts);1634 return ret;1635}16361637static int tree_has_path(struct tree *tree, const char *path)1638{1639 struct object_id hashy;1640 unsigned int mode_o;16411642 return !get_tree_entry(&tree->object.oid, path,1643 &hashy, &mode_o);1644}16451646/*1647 * Return a new string that replaces the beginning portion (which matches1648 * entry->dir), with entry->new_dir. In perl-speak:1649 * new_path_name = (old_path =~ s/entry->dir/entry->new_dir/);1650 * NOTE:1651 * Caller must ensure that old_path starts with entry->dir + '/'.1652 */1653static char *apply_dir_rename(struct dir_rename_entry *entry,1654 const char *old_path)1655{1656 struct strbuf new_path = STRBUF_INIT;1657 int oldlen, newlen;16581659 if (entry->non_unique_new_dir)1660 return NULL;16611662 oldlen = strlen(entry->dir);1663 newlen = entry->new_dir.len + (strlen(old_path) - oldlen) + 1;1664 strbuf_grow(&new_path, newlen);1665 strbuf_addbuf(&new_path, &entry->new_dir);1666 strbuf_addstr(&new_path, &old_path[oldlen]);16671668 return strbuf_detach(&new_path, NULL);1669}16701671static void get_renamed_dir_portion(const char *old_path, const char *new_path,1672 char **old_dir, char **new_dir)1673{1674 char *end_of_old, *end_of_new;1675 int old_len, new_len;16761677 *old_dir = NULL;1678 *new_dir = NULL;16791680 /*1681 * For1682 * "a/b/c/d/e/foo.c" -> "a/b/some/thing/else/e/foo.c"1683 * the "e/foo.c" part is the same, we just want to know that1684 * "a/b/c/d" was renamed to "a/b/some/thing/else"1685 * so, for this example, this function returns "a/b/c/d" in1686 * *old_dir and "a/b/some/thing/else" in *new_dir.1687 *1688 * Also, if the basename of the file changed, we don't care. We1689 * want to know which portion of the directory, if any, changed.1690 */1691 end_of_old = strrchr(old_path, '/');1692 end_of_new = strrchr(new_path, '/');16931694 if (end_of_old == NULL || end_of_new == NULL)1695 return;1696 while (*--end_of_new == *--end_of_old &&1697 end_of_old != old_path &&1698 end_of_new != new_path)1699 ; /* Do nothing; all in the while loop */1700 /*1701 * We've found the first non-matching character in the directory1702 * paths. That means the current directory we were comparing1703 * represents the rename. Move end_of_old and end_of_new back1704 * to the full directory name.1705 */1706 if (*end_of_old == '/')1707 end_of_old++;1708 if (*end_of_old != '/')1709 end_of_new++;1710 end_of_old = strchr(end_of_old, '/');1711 end_of_new = strchr(end_of_new, '/');17121713 /*1714 * It may have been the case that old_path and new_path were the same1715 * directory all along. Don't claim a rename if they're the same.1716 */1717 old_len = end_of_old - old_path;1718 new_len = end_of_new - new_path;17191720 if (old_len != new_len || strncmp(old_path, new_path, old_len)) {1721 *old_dir = xstrndup(old_path, old_len);1722 *new_dir = xstrndup(new_path, new_len);1723 }1724}17251726static void remove_hashmap_entries(struct hashmap *dir_renames,1727 struct string_list *items_to_remove)1728{1729 int i;1730 struct dir_rename_entry *entry;17311732 for (i = 0; i < items_to_remove->nr; i++) {1733 entry = items_to_remove->items[i].util;1734 hashmap_remove(dir_renames, entry, NULL);1735 }1736 string_list_clear(items_to_remove, 0);1737}17381739/*1740 * See if there is a directory rename for path, and if there are any file1741 * level conflicts for the renamed location. If there is a rename and1742 * there are no conflicts, return the new name. Otherwise, return NULL.1743 */1744static char *handle_path_level_conflicts(struct merge_options *o,1745 const char *path,1746 struct dir_rename_entry *entry,1747 struct hashmap *collisions,1748 struct tree *tree)1749{1750 char *new_path = NULL;1751 struct collision_entry *collision_ent;1752 int clean = 1;1753 struct strbuf collision_paths = STRBUF_INIT;17541755 /*1756 * entry has the mapping of old directory name to new directory name1757 * that we want to apply to path.1758 */1759 new_path = apply_dir_rename(entry, path);17601761 if (!new_path) {1762 /* This should only happen when entry->non_unique_new_dir set */1763 if (!entry->non_unique_new_dir)1764 BUG("entry->non_unqiue_dir not set and !new_path");1765 output(o, 1, _("CONFLICT (directory rename split): "1766 "Unclear where to place %s because directory "1767 "%s was renamed to multiple other directories, "1768 "with no destination getting a majority of the "1769 "files."),1770 path, entry->dir);1771 clean = 0;1772 return NULL;1773 }17741775 /*1776 * The caller needs to have ensured that it has pre-populated1777 * collisions with all paths that map to new_path. Do a quick check1778 * to ensure that's the case.1779 */1780 collision_ent = collision_find_entry(collisions, new_path);1781 if (collision_ent == NULL)1782 BUG("collision_ent is NULL");17831784 /*1785 * Check for one-sided add/add/.../add conflicts, i.e.1786 * where implicit renames from the other side doing1787 * directory rename(s) can affect this side of history1788 * to put multiple paths into the same location. Warn1789 * and bail on directory renames for such paths.1790 */1791 if (collision_ent->reported_already) {1792 clean = 0;1793 } else if (tree_has_path(tree, new_path)) {1794 collision_ent->reported_already = 1;1795 strbuf_add_separated_string_list(&collision_paths, ", ",1796 &collision_ent->source_files);1797 output(o, 1, _("CONFLICT (implicit dir rename): Existing "1798 "file/dir at %s in the way of implicit "1799 "directory rename(s) putting the following "1800 "path(s) there: %s."),1801 new_path, collision_paths.buf);1802 clean = 0;1803 } else if (collision_ent->source_files.nr > 1) {1804 collision_ent->reported_already = 1;1805 strbuf_add_separated_string_list(&collision_paths, ", ",1806 &collision_ent->source_files);1807 output(o, 1, _("CONFLICT (implicit dir rename): Cannot map "1808 "more than one path to %s; implicit directory "1809 "renames tried to put these paths there: %s"),1810 new_path, collision_paths.buf);1811 clean = 0;1812 }18131814 /* Free memory we no longer need */1815 strbuf_release(&collision_paths);1816 if (!clean && new_path) {1817 free(new_path);1818 return NULL;1819 }18201821 return new_path;1822}18231824/*1825 * There are a couple things we want to do at the directory level:1826 * 1. Check for both sides renaming to the same thing, in order to avoid1827 * implicit renaming of files that should be left in place. (See1828 * testcase 6b in t6043 for details.)1829 * 2. Prune directory renames if there are still files left in the1830 * the original directory. These represent a partial directory rename,1831 * i.e. a rename where only some of the files within the directory1832 * were renamed elsewhere. (Technically, this could be done earlier1833 * in get_directory_renames(), except that would prevent us from1834 * doing the previous check and thus failing testcase 6b.)1835 * 3. Check for rename/rename(1to2) conflicts (at the directory level).1836 * In the future, we could potentially record this info as well and1837 * omit reporting rename/rename(1to2) conflicts for each path within1838 * the affected directories, thus cleaning up the merge output.1839 * NOTE: We do NOT check for rename/rename(2to1) conflicts at the1840 * directory level, because merging directories is fine. If it1841 * causes conflicts for files within those merged directories, then1842 * that should be detected at the individual path level.1843 */1844static void handle_directory_level_conflicts(struct merge_options *o,1845 struct hashmap *dir_re_head,1846 struct tree *head,1847 struct hashmap *dir_re_merge,1848 struct tree *merge)1849{1850 struct hashmap_iter iter;1851 struct dir_rename_entry *head_ent;1852 struct dir_rename_entry *merge_ent;18531854 struct string_list remove_from_head = STRING_LIST_INIT_NODUP;1855 struct string_list remove_from_merge = STRING_LIST_INIT_NODUP;18561857 hashmap_iter_init(dir_re_head, &iter);1858 while ((head_ent = hashmap_iter_next(&iter))) {1859 merge_ent = dir_rename_find_entry(dir_re_merge, head_ent->dir);1860 if (merge_ent &&1861 !head_ent->non_unique_new_dir &&1862 !merge_ent->non_unique_new_dir &&1863 !strbuf_cmp(&head_ent->new_dir, &merge_ent->new_dir)) {1864 /* 1. Renamed identically; remove it from both sides */1865 string_list_append(&remove_from_head,1866 head_ent->dir)->util = head_ent;1867 strbuf_release(&head_ent->new_dir);1868 string_list_append(&remove_from_merge,1869 merge_ent->dir)->util = merge_ent;1870 strbuf_release(&merge_ent->new_dir);1871 } else if (tree_has_path(head, head_ent->dir)) {1872 /* 2. This wasn't a directory rename after all */1873 string_list_append(&remove_from_head,1874 head_ent->dir)->util = head_ent;1875 strbuf_release(&head_ent->new_dir);1876 }1877 }18781879 remove_hashmap_entries(dir_re_head, &remove_from_head);1880 remove_hashmap_entries(dir_re_merge, &remove_from_merge);18811882 hashmap_iter_init(dir_re_merge, &iter);1883 while ((merge_ent = hashmap_iter_next(&iter))) {1884 head_ent = dir_rename_find_entry(dir_re_head, merge_ent->dir);1885 if (tree_has_path(merge, merge_ent->dir)) {1886 /* 2. This wasn't a directory rename after all */1887 string_list_append(&remove_from_merge,1888 merge_ent->dir)->util = merge_ent;1889 } else if (head_ent &&1890 !head_ent->non_unique_new_dir &&1891 !merge_ent->non_unique_new_dir) {1892 /* 3. rename/rename(1to2) */1893 /*1894 * We can assume it's not rename/rename(1to1) because1895 * that was case (1), already checked above. So we1896 * know that head_ent->new_dir and merge_ent->new_dir1897 * are different strings.1898 */1899 output(o, 1, _("CONFLICT (rename/rename): "1900 "Rename directory %s->%s in %s. "1901 "Rename directory %s->%s in %s"),1902 head_ent->dir, head_ent->new_dir.buf, o->branch1,1903 head_ent->dir, merge_ent->new_dir.buf, o->branch2);1904 string_list_append(&remove_from_head,1905 head_ent->dir)->util = head_ent;1906 strbuf_release(&head_ent->new_dir);1907 string_list_append(&remove_from_merge,1908 merge_ent->dir)->util = merge_ent;1909 strbuf_release(&merge_ent->new_dir);1910 }1911 }19121913 remove_hashmap_entries(dir_re_head, &remove_from_head);1914 remove_hashmap_entries(dir_re_merge, &remove_from_merge);1915}19161917static struct hashmap *get_directory_renames(struct diff_queue_struct *pairs,1918 struct tree *tree)1919{1920 struct hashmap *dir_renames;1921 struct hashmap_iter iter;1922 struct dir_rename_entry *entry;1923 int i;19241925 /*1926 * Typically, we think of a directory rename as all files from a1927 * certain directory being moved to a target directory. However,1928 * what if someone first moved two files from the original1929 * directory in one commit, and then renamed the directory1930 * somewhere else in a later commit? At merge time, we just know1931 * that files from the original directory went to two different1932 * places, and that the bulk of them ended up in the same place.1933 * We want each directory rename to represent where the bulk of the1934 * files from that directory end up; this function exists to find1935 * where the bulk of the files went.1936 *1937 * The first loop below simply iterates through the list of file1938 * renames, finding out how often each directory rename pair1939 * possibility occurs.1940 */1941 dir_renames = xmalloc(sizeof(*dir_renames));1942 dir_rename_init(dir_renames);1943 for (i = 0; i < pairs->nr; ++i) {1944 struct string_list_item *item;1945 int *count;1946 struct diff_filepair *pair = pairs->queue[i];1947 char *old_dir, *new_dir;19481949 /* File not part of directory rename if it wasn't renamed */1950 if (pair->status != 'R')1951 continue;19521953 get_renamed_dir_portion(pair->one->path, pair->two->path,1954 &old_dir, &new_dir);1955 if (!old_dir)1956 /* Directory didn't change at all; ignore this one. */1957 continue;19581959 entry = dir_rename_find_entry(dir_renames, old_dir);1960 if (!entry) {1961 entry = xmalloc(sizeof(*entry));1962 dir_rename_entry_init(entry, old_dir);1963 hashmap_put(dir_renames, entry);1964 } else {1965 free(old_dir);1966 }1967 item = string_list_lookup(&entry->possible_new_dirs, new_dir);1968 if (!item) {1969 item = string_list_insert(&entry->possible_new_dirs,1970 new_dir);1971 item->util = xcalloc(1, sizeof(int));1972 } else {1973 free(new_dir);1974 }1975 count = item->util;1976 *count += 1;1977 }19781979 /*1980 * For each directory with files moved out of it, we find out which1981 * target directory received the most files so we can declare it to1982 * be the "winning" target location for the directory rename. This1983 * winner gets recorded in new_dir. If there is no winner1984 * (multiple target directories received the same number of files),1985 * we set non_unique_new_dir. Once we've determined the winner (or1986 * that there is no winner), we no longer need possible_new_dirs.1987 */1988 hashmap_iter_init(dir_renames, &iter);1989 while ((entry = hashmap_iter_next(&iter))) {1990 int max = 0;1991 int bad_max = 0;1992 char *best = NULL;19931994 for (i = 0; i < entry->possible_new_dirs.nr; i++) {1995 int *count = entry->possible_new_dirs.items[i].util;19961997 if (*count == max)1998 bad_max = max;1999 else if (*count > max) {2000 max = *count;2001 best = entry->possible_new_dirs.items[i].string;2002 }2003 }2004 if (bad_max == max)2005 entry->non_unique_new_dir = 1;2006 else {2007 assert(entry->new_dir.len == 0);2008 strbuf_addstr(&entry->new_dir, best);2009 }2010 /*2011 * The relevant directory sub-portion of the original full2012 * filepaths were xstrndup'ed before inserting into2013 * possible_new_dirs, and instead of manually iterating the2014 * list and free'ing each, just lie and tell2015 * possible_new_dirs that it did the strdup'ing so that it2016 * will free them for us.2017 */2018 entry->possible_new_dirs.strdup_strings = 1;2019 string_list_clear(&entry->possible_new_dirs, 1);2020 }20212022 return dir_renames;2023}20242025static struct dir_rename_entry *check_dir_renamed(const char *path,2026 struct hashmap *dir_renames)2027{2028 char temp[PATH_MAX];2029 char *end;2030 struct dir_rename_entry *entry;20312032 strcpy(temp, path);2033 while ((end = strrchr(temp, '/'))) {2034 *end = '\0';2035 entry = dir_rename_find_entry(dir_renames, temp);2036 if (entry)2037 return entry;2038 }2039 return NULL;2040}20412042static void compute_collisions(struct hashmap *collisions,2043 struct hashmap *dir_renames,2044 struct diff_queue_struct *pairs)2045{2046 int i;20472048 /*2049 * Multiple files can be mapped to the same path due to directory2050 * renames done by the other side of history. Since that other2051 * side of history could have merged multiple directories into one,2052 * if our side of history added the same file basename to each of2053 * those directories, then all N of them would get implicitly2054 * renamed by the directory rename detection into the same path,2055 * and we'd get an add/add/.../add conflict, and all those adds2056 * from *this* side of history. This is not representable in the2057 * index, and users aren't going to easily be able to make sense of2058 * it. So we need to provide a good warning about what's2059 * happening, and fall back to no-directory-rename detection2060 * behavior for those paths.2061 *2062 * See testcases 9e and all of section 5 from t6043 for examples.2063 */2064 collision_init(collisions);20652066 for (i = 0; i < pairs->nr; ++i) {2067 struct dir_rename_entry *dir_rename_ent;2068 struct collision_entry *collision_ent;2069 char *new_path;2070 struct diff_filepair *pair = pairs->queue[i];20712072 if (pair->status != 'A' && pair->status != 'R')2073 continue;2074 dir_rename_ent = check_dir_renamed(pair->two->path,2075 dir_renames);2076 if (!dir_rename_ent)2077 continue;20782079 new_path = apply_dir_rename(dir_rename_ent, pair->two->path);2080 if (!new_path)2081 /*2082 * dir_rename_ent->non_unique_new_path is true, which2083 * means there is no directory rename for us to use,2084 * which means it won't cause us any additional2085 * collisions.2086 */2087 continue;2088 collision_ent = collision_find_entry(collisions, new_path);2089 if (!collision_ent) {2090 collision_ent = xcalloc(1,2091 sizeof(struct collision_entry));2092 hashmap_entry_init(collision_ent, strhash(new_path));2093 hashmap_put(collisions, collision_ent);2094 collision_ent->target_file = new_path;2095 } else {2096 free(new_path);2097 }2098 string_list_insert(&collision_ent->source_files,2099 pair->two->path);2100 }2101}21022103static char *check_for_directory_rename(struct merge_options *o,2104 const char *path,2105 struct tree *tree,2106 struct hashmap *dir_renames,2107 struct hashmap *dir_rename_exclusions,2108 struct hashmap *collisions,2109 int *clean_merge)2110{2111 char *new_path = NULL;2112 struct dir_rename_entry *entry = check_dir_renamed(path, dir_renames);2113 struct dir_rename_entry *oentry = NULL;21142115 if (!entry)2116 return new_path;21172118 /*2119 * This next part is a little weird. We do not want to do an2120 * implicit rename into a directory we renamed on our side, because2121 * that will result in a spurious rename/rename(1to2) conflict. An2122 * example:2123 * Base commit: dumbdir/afile, otherdir/bfile2124 * Side 1: smrtdir/afile, otherdir/bfile2125 * Side 2: dumbdir/afile, dumbdir/bfile2126 * Here, while working on Side 1, we could notice that otherdir was2127 * renamed/merged to dumbdir, and change the diff_filepair for2128 * otherdir/bfile into a rename into dumbdir/bfile. However, Side2129 * 2 will notice the rename from dumbdir to smrtdir, and do the2130 * transitive rename to move it from dumbdir/bfile to2131 * smrtdir/bfile. That gives us bfile in dumbdir vs being in2132 * smrtdir, a rename/rename(1to2) conflict. We really just want2133 * the file to end up in smrtdir. And the way to achieve that is2134 * to not let Side1 do the rename to dumbdir, since we know that is2135 * the source of one of our directory renames.2136 *2137 * That's why oentry and dir_rename_exclusions is here.2138 *2139 * As it turns out, this also prevents N-way transient rename2140 * confusion; See testcases 9c and 9d of t6043.2141 */2142 oentry = dir_rename_find_entry(dir_rename_exclusions, entry->new_dir.buf);2143 if (oentry) {2144 output(o, 1, _("WARNING: Avoiding applying %s -> %s rename "2145 "to %s, because %s itself was renamed."),2146 entry->dir, entry->new_dir.buf, path, entry->new_dir.buf);2147 } else {2148 new_path = handle_path_level_conflicts(o, path, entry,2149 collisions, tree);2150 *clean_merge &= (new_path != NULL);2151 }21522153 return new_path;2154}21552156static void apply_directory_rename_modifications(struct merge_options *o,2157 struct diff_filepair *pair,2158 char *new_path,2159 struct rename *re,2160 struct tree *tree,2161 struct tree *o_tree,2162 struct tree *a_tree,2163 struct tree *b_tree,2164 struct string_list *entries,2165 int *clean)2166{2167 struct string_list_item *item;2168 int stage = (tree == a_tree ? 2 : 3);2169 int update_wd;21702171 /*2172 * In all cases where we can do directory rename detection,2173 * unpack_trees() will have read pair->two->path into the2174 * index and the working copy. We need to remove it so that2175 * we can instead place it at new_path. It is guaranteed to2176 * not be untracked (unpack_trees() would have errored out2177 * saying the file would have been overwritten), but it might2178 * be dirty, though.2179 */2180 update_wd = !was_dirty(o, pair->two->path);2181 if (!update_wd)2182 output(o, 1, _("Refusing to lose dirty file at %s"),2183 pair->two->path);2184 remove_file(o, 1, pair->two->path, !update_wd);21852186 /* Find or create a new re->dst_entry */2187 item = string_list_lookup(entries, new_path);2188 if (item) {2189 /*2190 * Since we're renaming on this side of history, and it's2191 * due to a directory rename on the other side of history2192 * (which we only allow when the directory in question no2193 * longer exists on the other side of history), the2194 * original entry for re->dst_entry is no longer2195 * necessary...2196 */2197 re->dst_entry->processed = 1;21982199 /*2200 * ...because we'll be using this new one.2201 */2202 re->dst_entry = item->util;2203 } else {2204 /*2205 * re->dst_entry is for the before-dir-rename path, and we2206 * need it to hold information for the after-dir-rename2207 * path. Before creating a new entry, we need to mark the2208 * old one as unnecessary (...unless it is shared by2209 * src_entry, i.e. this didn't use to be a rename, in which2210 * case we can just allow the normal processing to happen2211 * for it).2212 */2213 if (pair->status == 'R')2214 re->dst_entry->processed = 1;22152216 re->dst_entry = insert_stage_data(new_path,2217 o_tree, a_tree, b_tree,2218 entries);2219 item = string_list_insert(entries, new_path);2220 item->util = re->dst_entry;2221 }22222223 /*2224 * Update the stage_data with the information about the path we are2225 * moving into place. That slot will be empty and available for us2226 * to write to because of the collision checks in2227 * handle_path_level_conflicts(). In other words,2228 * re->dst_entry->stages[stage].oid will be the null_oid, so it's2229 * open for us to write to.2230 *2231 * It may be tempting to actually update the index at this point as2232 * well, using update_stages_for_stage_data(), but as per the big2233 * "NOTE" in update_stages(), doing so will modify the current2234 * in-memory index which will break calls to would_lose_untracked()2235 * that we need to make. Instead, we need to just make sure that2236 * the various conflict_rename_*() functions update the index2237 * explicitly rather than relying on unpack_trees() to have done it.2238 */2239 get_tree_entry(&tree->object.oid,2240 pair->two->path,2241 &re->dst_entry->stages[stage].oid,2242 &re->dst_entry->stages[stage].mode);22432244 /* Update pair status */2245 if (pair->status == 'A') {2246 /*2247 * Recording rename information for this add makes it look2248 * like a rename/delete conflict. Make sure we can2249 * correctly handle this as an add that was moved to a new2250 * directory instead of reporting a rename/delete conflict.2251 */2252 re->add_turned_into_rename = 1;2253 }2254 /*2255 * We don't actually look at pair->status again, but it seems2256 * pedagogically correct to adjust it.2257 */2258 pair->status = 'R';22592260 /*2261 * Finally, record the new location.2262 */2263 pair->two->path = new_path;2264}22652266/*2267 * Get information of all renames which occurred in 'pairs', making use of2268 * any implicit directory renames inferred from the other side of history.2269 * We need the three trees in the merge ('o_tree', 'a_tree' and 'b_tree')2270 * to be able to associate the correct cache entries with the rename2271 * information; tree is always equal to either a_tree or b_tree.2272 */2273static struct string_list *get_renames(struct merge_options *o,2274 struct diff_queue_struct *pairs,2275 struct hashmap *dir_renames,2276 struct hashmap *dir_rename_exclusions,2277 struct tree *tree,2278 struct tree *o_tree,2279 struct tree *a_tree,2280 struct tree *b_tree,2281 struct string_list *entries,2282 int *clean_merge)2283{2284 int i;2285 struct hashmap collisions;2286 struct hashmap_iter iter;2287 struct collision_entry *e;2288 struct string_list *renames;22892290 compute_collisions(&collisions, dir_renames, pairs);2291 renames = xcalloc(1, sizeof(struct string_list));22922293 for (i = 0; i < pairs->nr; ++i) {2294 struct string_list_item *item;2295 struct rename *re;2296 struct diff_filepair *pair = pairs->queue[i];2297 char *new_path; /* non-NULL only with directory renames */22982299 if (pair->status != 'A' && pair->status != 'R') {2300 diff_free_filepair(pair);2301 continue;2302 }2303 new_path = check_for_directory_rename(o, pair->two->path, tree,2304 dir_renames,2305 dir_rename_exclusions,2306 &collisions,2307 clean_merge);2308 if (pair->status != 'R' && !new_path) {2309 diff_free_filepair(pair);2310 continue;2311 }23122313 re = xmalloc(sizeof(*re));2314 re->processed = 0;2315 re->add_turned_into_rename = 0;2316 re->pair = pair;2317 item = string_list_lookup(entries, re->pair->one->path);2318 if (!item)2319 re->src_entry = insert_stage_data(re->pair->one->path,2320 o_tree, a_tree, b_tree, entries);2321 else2322 re->src_entry = item->util;23232324 item = string_list_lookup(entries, re->pair->two->path);2325 if (!item)2326 re->dst_entry = insert_stage_data(re->pair->two->path,2327 o_tree, a_tree, b_tree, entries);2328 else2329 re->dst_entry = item->util;2330 item = string_list_insert(renames, pair->one->path);2331 item->util = re;2332 if (new_path)2333 apply_directory_rename_modifications(o, pair, new_path,2334 re, tree, o_tree,2335 a_tree, b_tree,2336 entries,2337 clean_merge);2338 }23392340 hashmap_iter_init(&collisions, &iter);2341 while ((e = hashmap_iter_next(&iter))) {2342 free(e->target_file);2343 string_list_clear(&e->source_files, 0);2344 }2345 hashmap_free(&collisions, 1);2346 return renames;2347}23482349static int process_renames(struct merge_options *o,2350 struct string_list *a_renames,2351 struct string_list *b_renames)2352{2353 int clean_merge = 1, i, j;2354 struct string_list a_by_dst = STRING_LIST_INIT_NODUP;2355 struct string_list b_by_dst = STRING_LIST_INIT_NODUP;2356 const struct rename *sre;23572358 for (i = 0; i < a_renames->nr; i++) {2359 sre = a_renames->items[i].util;2360 string_list_insert(&a_by_dst, sre->pair->two->path)->util2361 = (void *)sre;2362 }2363 for (i = 0; i < b_renames->nr; i++) {2364 sre = b_renames->items[i].util;2365 string_list_insert(&b_by_dst, sre->pair->two->path)->util2366 = (void *)sre;2367 }23682369 for (i = 0, j = 0; i < a_renames->nr || j < b_renames->nr;) {2370 struct string_list *renames1, *renames2Dst;2371 struct rename *ren1 = NULL, *ren2 = NULL;2372 const char *branch1, *branch2;2373 const char *ren1_src, *ren1_dst;2374 struct string_list_item *lookup;23752376 if (i >= a_renames->nr) {2377 ren2 = b_renames->items[j++].util;2378 } else if (j >= b_renames->nr) {2379 ren1 = a_renames->items[i++].util;2380 } else {2381 int compare = strcmp(a_renames->items[i].string,2382 b_renames->items[j].string);2383 if (compare <= 0)2384 ren1 = a_renames->items[i++].util;2385 if (compare >= 0)2386 ren2 = b_renames->items[j++].util;2387 }23882389 /* TODO: refactor, so that 1/2 are not needed */2390 if (ren1) {2391 renames1 = a_renames;2392 renames2Dst = &b_by_dst;2393 branch1 = o->branch1;2394 branch2 = o->branch2;2395 } else {2396 renames1 = b_renames;2397 renames2Dst = &a_by_dst;2398 branch1 = o->branch2;2399 branch2 = o->branch1;2400 SWAP(ren2, ren1);2401 }24022403 if (ren1->processed)2404 continue;2405 ren1->processed = 1;2406 ren1->dst_entry->processed = 1;2407 /* BUG: We should only mark src_entry as processed if we2408 * are not dealing with a rename + add-source case.2409 */2410 ren1->src_entry->processed = 1;24112412 ren1_src = ren1->pair->one->path;2413 ren1_dst = ren1->pair->two->path;24142415 if (ren2) {2416 /* One file renamed on both sides */2417 const char *ren2_src = ren2->pair->one->path;2418 const char *ren2_dst = ren2->pair->two->path;2419 enum rename_type rename_type;2420 if (strcmp(ren1_src, ren2_src) != 0)2421 die("BUG: ren1_src != ren2_src");2422 ren2->dst_entry->processed = 1;2423 ren2->processed = 1;2424 if (strcmp(ren1_dst, ren2_dst) != 0) {2425 rename_type = RENAME_ONE_FILE_TO_TWO;2426 clean_merge = 0;2427 } else {2428 rename_type = RENAME_ONE_FILE_TO_ONE;2429 /* BUG: We should only remove ren1_src in2430 * the base stage (think of rename +2431 * add-source cases).2432 */2433 remove_file(o, 1, ren1_src, 1);2434 update_entry(ren1->dst_entry,2435 ren1->pair->one,2436 ren1->pair->two,2437 ren2->pair->two);2438 }2439 setup_rename_conflict_info(rename_type,2440 ren1->pair,2441 ren2->pair,2442 branch1,2443 branch2,2444 ren1->dst_entry,2445 ren2->dst_entry,2446 o,2447 NULL,2448 NULL);2449 } else if ((lookup = string_list_lookup(renames2Dst, ren1_dst))) {2450 /* Two different files renamed to the same thing */2451 char *ren2_dst;2452 ren2 = lookup->util;2453 ren2_dst = ren2->pair->two->path;2454 if (strcmp(ren1_dst, ren2_dst) != 0)2455 die("BUG: ren1_dst != ren2_dst");24562457 clean_merge = 0;2458 ren2->processed = 1;2459 /*2460 * BUG: We should only mark src_entry as processed2461 * if we are not dealing with a rename + add-source2462 * case.2463 */2464 ren2->src_entry->processed = 1;24652466 setup_rename_conflict_info(RENAME_TWO_FILES_TO_ONE,2467 ren1->pair,2468 ren2->pair,2469 branch1,2470 branch2,2471 ren1->dst_entry,2472 ren2->dst_entry,2473 o,2474 ren1->src_entry,2475 ren2->src_entry);24762477 } else {2478 /* Renamed in 1, maybe changed in 2 */2479 /* we only use sha1 and mode of these */2480 struct diff_filespec src_other, dst_other;2481 int try_merge;24822483 /*2484 * unpack_trees loads entries from common-commit2485 * into stage 1, from head-commit into stage 2, and2486 * from merge-commit into stage 3. We keep track2487 * of which side corresponds to the rename.2488 */2489 int renamed_stage = a_renames == renames1 ? 2 : 3;2490 int other_stage = a_renames == renames1 ? 3 : 2;24912492 /* BUG: We should only remove ren1_src in the base2493 * stage and in other_stage (think of rename +2494 * add-source case).2495 */2496 remove_file(o, 1, ren1_src,2497 renamed_stage == 2 || !was_tracked(o, ren1_src));24982499 oidcpy(&src_other.oid,2500 &ren1->src_entry->stages[other_stage].oid);2501 src_other.mode = ren1->src_entry->stages[other_stage].mode;2502 oidcpy(&dst_other.oid,2503 &ren1->dst_entry->stages[other_stage].oid);2504 dst_other.mode = ren1->dst_entry->stages[other_stage].mode;2505 try_merge = 0;25062507 if (oid_eq(&src_other.oid, &null_oid) &&2508 ren1->add_turned_into_rename) {2509 setup_rename_conflict_info(RENAME_DIR,2510 ren1->pair,2511 NULL,2512 branch1,2513 branch2,2514 ren1->dst_entry,2515 NULL,2516 o,2517 NULL,2518 NULL);2519 } else if (oid_eq(&src_other.oid, &null_oid)) {2520 setup_rename_conflict_info(RENAME_DELETE,2521 ren1->pair,2522 NULL,2523 branch1,2524 branch2,2525 ren1->dst_entry,2526 NULL,2527 o,2528 NULL,2529 NULL);2530 } else if ((dst_other.mode == ren1->pair->two->mode) &&2531 oid_eq(&dst_other.oid, &ren1->pair->two->oid)) {2532 /*2533 * Added file on the other side identical to2534 * the file being renamed: clean merge.2535 * Also, there is no need to overwrite the2536 * file already in the working copy, so call2537 * update_file_flags() instead of2538 * update_file().2539 */2540 if (update_file_flags(o,2541 &ren1->pair->two->oid,2542 ren1->pair->two->mode,2543 ren1_dst,2544 1, /* update_cache */2545 0 /* update_wd */))2546 clean_merge = -1;2547 } else if (!oid_eq(&dst_other.oid, &null_oid)) {2548 clean_merge = 0;2549 try_merge = 1;2550 output(o, 1, _("CONFLICT (rename/add): Rename %s->%s in %s. "2551 "%s added in %s"),2552 ren1_src, ren1_dst, branch1,2553 ren1_dst, branch2);2554 if (o->call_depth) {2555 struct merge_file_info mfi;2556 if (merge_file_one(o, ren1_dst, &null_oid, 0,2557 &ren1->pair->two->oid,2558 ren1->pair->two->mode,2559 &dst_other.oid,2560 dst_other.mode,2561 branch1, branch2, &mfi)) {2562 clean_merge = -1;2563 goto cleanup_and_return;2564 }2565 output(o, 1, _("Adding merged %s"), ren1_dst);2566 if (update_file(o, 0, &mfi.oid,2567 mfi.mode, ren1_dst))2568 clean_merge = -1;2569 try_merge = 0;2570 } else {2571 char *new_path = unique_path(o, ren1_dst, branch2);2572 output(o, 1, _("Adding as %s instead"), new_path);2573 if (update_file(o, 0, &dst_other.oid,2574 dst_other.mode, new_path))2575 clean_merge = -1;2576 free(new_path);2577 }2578 } else2579 try_merge = 1;25802581 if (clean_merge < 0)2582 goto cleanup_and_return;2583 if (try_merge) {2584 struct diff_filespec *one, *a, *b;2585 src_other.path = (char *)ren1_src;25862587 one = ren1->pair->one;2588 if (a_renames == renames1) {2589 a = ren1->pair->two;2590 b = &src_other;2591 } else {2592 b = ren1->pair->two;2593 a = &src_other;2594 }2595 update_entry(ren1->dst_entry, one, a, b);2596 setup_rename_conflict_info(RENAME_NORMAL,2597 ren1->pair,2598 NULL,2599 branch1,2600 NULL,2601 ren1->dst_entry,2602 NULL,2603 o,2604 NULL,2605 NULL);2606 }2607 }2608 }2609cleanup_and_return:2610 string_list_clear(&a_by_dst, 0);2611 string_list_clear(&b_by_dst, 0);26122613 return clean_merge;2614}26152616struct rename_info {2617 struct string_list *head_renames;2618 struct string_list *merge_renames;2619};26202621static void initial_cleanup_rename(struct diff_queue_struct *pairs,2622 struct hashmap *dir_renames)2623{2624 struct hashmap_iter iter;2625 struct dir_rename_entry *e;26262627 hashmap_iter_init(dir_renames, &iter);2628 while ((e = hashmap_iter_next(&iter))) {2629 free(e->dir);2630 strbuf_release(&e->new_dir);2631 /* possible_new_dirs already cleared in get_directory_renames */2632 }2633 hashmap_free(dir_renames, 1);2634 free(dir_renames);26352636 free(pairs->queue);2637 free(pairs);2638}26392640static int handle_renames(struct merge_options *o,2641 struct tree *common,2642 struct tree *head,2643 struct tree *merge,2644 struct string_list *entries,2645 struct rename_info *ri)2646{2647 struct diff_queue_struct *head_pairs, *merge_pairs;2648 struct hashmap *dir_re_head, *dir_re_merge;2649 int clean = 1;26502651 ri->head_renames = NULL;2652 ri->merge_renames = NULL;26532654 if (!merge_detect_rename(o))2655 return 1;26562657 head_pairs = get_diffpairs(o, common, head);2658 merge_pairs = get_diffpairs(o, common, merge);26592660 dir_re_head = get_directory_renames(head_pairs, head);2661 dir_re_merge = get_directory_renames(merge_pairs, merge);26622663 handle_directory_level_conflicts(o,2664 dir_re_head, head,2665 dir_re_merge, merge);26662667 ri->head_renames = get_renames(o, head_pairs,2668 dir_re_merge, dir_re_head, head,2669 common, head, merge, entries,2670 &clean);2671 if (clean < 0)2672 goto cleanup;2673 ri->merge_renames = get_renames(o, merge_pairs,2674 dir_re_head, dir_re_merge, merge,2675 common, head, merge, entries,2676 &clean);2677 if (clean < 0)2678 goto cleanup;2679 clean &= process_renames(o, ri->head_renames, ri->merge_renames);26802681cleanup:2682 /*2683 * Some cleanup is deferred until cleanup_renames() because the2684 * data structures are still needed and referenced in2685 * process_entry(). But there are a few things we can free now.2686 */2687 initial_cleanup_rename(head_pairs, dir_re_head);2688 initial_cleanup_rename(merge_pairs, dir_re_merge);26892690 return clean;2691}26922693static void final_cleanup_rename(struct string_list *rename)2694{2695 const struct rename *re;2696 int i;26972698 if (rename == NULL)2699 return;27002701 for (i = 0; i < rename->nr; i++) {2702 re = rename->items[i].util;2703 diff_free_filepair(re->pair);2704 }2705 string_list_clear(rename, 1);2706 free(rename);2707}27082709static void final_cleanup_renames(struct rename_info *re_info)2710{2711 final_cleanup_rename(re_info->head_renames);2712 final_cleanup_rename(re_info->merge_renames);2713}27142715static struct object_id *stage_oid(const struct object_id *oid, unsigned mode)2716{2717 return (is_null_oid(oid) || mode == 0) ? NULL: (struct object_id *)oid;2718}27192720static int read_oid_strbuf(struct merge_options *o,2721 const struct object_id *oid, struct strbuf *dst)2722{2723 void *buf;2724 enum object_type type;2725 unsigned long size;2726 buf = read_object_file(oid, &type, &size);2727 if (!buf)2728 return err(o, _("cannot read object %s"), oid_to_hex(oid));2729 if (type != OBJ_BLOB) {2730 free(buf);2731 return err(o, _("object %s is not a blob"), oid_to_hex(oid));2732 }2733 strbuf_attach(dst, buf, size, size + 1);2734 return 0;2735}27362737static int blob_unchanged(struct merge_options *opt,2738 const struct object_id *o_oid,2739 unsigned o_mode,2740 const struct object_id *a_oid,2741 unsigned a_mode,2742 int renormalize, const char *path)2743{2744 struct strbuf o = STRBUF_INIT;2745 struct strbuf a = STRBUF_INIT;2746 int ret = 0; /* assume changed for safety */27472748 if (a_mode != o_mode)2749 return 0;2750 if (oid_eq(o_oid, a_oid))2751 return 1;2752 if (!renormalize)2753 return 0;27542755 assert(o_oid && a_oid);2756 if (read_oid_strbuf(opt, o_oid, &o) || read_oid_strbuf(opt, a_oid, &a))2757 goto error_return;2758 /*2759 * Note: binary | is used so that both renormalizations are2760 * performed. Comparison can be skipped if both files are2761 * unchanged since their sha1s have already been compared.2762 */2763 if (renormalize_buffer(&the_index, path, o.buf, o.len, &o) |2764 renormalize_buffer(&the_index, path, a.buf, a.len, &a))2765 ret = (o.len == a.len && !memcmp(o.buf, a.buf, o.len));27662767error_return:2768 strbuf_release(&o);2769 strbuf_release(&a);2770 return ret;2771}27722773static int handle_modify_delete(struct merge_options *o,2774 const char *path,2775 struct object_id *o_oid, int o_mode,2776 struct object_id *a_oid, int a_mode,2777 struct object_id *b_oid, int b_mode)2778{2779 const char *modify_branch, *delete_branch;2780 struct object_id *changed_oid;2781 int changed_mode;27822783 if (a_oid) {2784 modify_branch = o->branch1;2785 delete_branch = o->branch2;2786 changed_oid = a_oid;2787 changed_mode = a_mode;2788 } else {2789 modify_branch = o->branch2;2790 delete_branch = o->branch1;2791 changed_oid = b_oid;2792 changed_mode = b_mode;2793 }27942795 return handle_change_delete(o,2796 path, NULL,2797 o_oid, o_mode,2798 changed_oid, changed_mode,2799 modify_branch, delete_branch,2800 _("modify"), _("modified"));2801}28022803static int merge_content(struct merge_options *o,2804 const char *path,2805 int is_dirty,2806 struct object_id *o_oid, int o_mode,2807 struct object_id *a_oid, int a_mode,2808 struct object_id *b_oid, int b_mode,2809 struct rename_conflict_info *rename_conflict_info)2810{2811 const char *reason = _("content");2812 const char *path1 = NULL, *path2 = NULL;2813 struct merge_file_info mfi;2814 struct diff_filespec one, a, b;2815 unsigned df_conflict_remains = 0;28162817 if (!o_oid) {2818 reason = _("add/add");2819 o_oid = (struct object_id *)&null_oid;2820 }2821 one.path = a.path = b.path = (char *)path;2822 oidcpy(&one.oid, o_oid);2823 one.mode = o_mode;2824 oidcpy(&a.oid, a_oid);2825 a.mode = a_mode;2826 oidcpy(&b.oid, b_oid);2827 b.mode = b_mode;28282829 if (rename_conflict_info) {2830 struct diff_filepair *pair1 = rename_conflict_info->pair1;28312832 path1 = (o->branch1 == rename_conflict_info->branch1) ?2833 pair1->two->path : pair1->one->path;2834 /* If rename_conflict_info->pair2 != NULL, we are in2835 * RENAME_ONE_FILE_TO_ONE case. Otherwise, we have a2836 * normal rename.2837 */2838 path2 = (rename_conflict_info->pair2 ||2839 o->branch2 == rename_conflict_info->branch1) ?2840 pair1->two->path : pair1->one->path;28412842 if (dir_in_way(path, !o->call_depth,2843 S_ISGITLINK(pair1->two->mode)))2844 df_conflict_remains = 1;2845 }2846 if (merge_file_special_markers(o, &one, &a, &b, path,2847 o->branch1, path1,2848 o->branch2, path2, &mfi))2849 return -1;28502851 /*2852 * We can skip updating the working tree file iff:2853 * a) The merge is clean2854 * b) The merge matches what was in HEAD (content, mode, pathname)2855 * c) The target path is usable (i.e. not involved in D/F conflict)2856 */2857 if (mfi.clean &&2858 was_tracked_and_matches(o, path, &mfi.oid, mfi.mode) &&2859 !df_conflict_remains) {2860 output(o, 3, _("Skipped %s (merged same as existing)"), path);2861 if (add_cacheinfo(o, mfi.mode, &mfi.oid, path,2862 0, (!o->call_depth && !is_dirty), 0))2863 return -1;2864 return mfi.clean;2865 }28662867 if (!mfi.clean) {2868 if (S_ISGITLINK(mfi.mode))2869 reason = _("submodule");2870 output(o, 1, _("CONFLICT (%s): Merge conflict in %s"),2871 reason, path);2872 if (rename_conflict_info && !df_conflict_remains)2873 if (update_stages(o, path, &one, &a, &b))2874 return -1;2875 }28762877 if (df_conflict_remains || is_dirty) {2878 char *new_path;2879 if (o->call_depth) {2880 remove_file_from_cache(path);2881 } else {2882 if (!mfi.clean) {2883 if (update_stages(o, path, &one, &a, &b))2884 return -1;2885 } else {2886 int file_from_stage2 = was_tracked(o, path);2887 struct diff_filespec merged;2888 oidcpy(&merged.oid, &mfi.oid);2889 merged.mode = mfi.mode;28902891 if (update_stages(o, path, NULL,2892 file_from_stage2 ? &merged : NULL,2893 file_from_stage2 ? NULL : &merged))2894 return -1;2895 }28962897 }2898 new_path = unique_path(o, path, rename_conflict_info->branch1);2899 if (is_dirty) {2900 output(o, 1, _("Refusing to lose dirty file at %s"),2901 path);2902 }2903 output(o, 1, _("Adding as %s instead"), new_path);2904 if (update_file(o, 0, &mfi.oid, mfi.mode, new_path)) {2905 free(new_path);2906 return -1;2907 }2908 free(new_path);2909 mfi.clean = 0;2910 } else if (update_file(o, mfi.clean, &mfi.oid, mfi.mode, path))2911 return -1;2912 return !is_dirty && mfi.clean;2913}29142915static int conflict_rename_normal(struct merge_options *o,2916 const char *path,2917 struct object_id *o_oid, unsigned int o_mode,2918 struct object_id *a_oid, unsigned int a_mode,2919 struct object_id *b_oid, unsigned int b_mode,2920 struct rename_conflict_info *ci)2921{2922 /* Merge the content and write it out */2923 return merge_content(o, path, was_dirty(o, path),2924 o_oid, o_mode, a_oid, a_mode, b_oid, b_mode,2925 ci);2926}29272928/* Per entry merge function */2929static int process_entry(struct merge_options *o,2930 const char *path, struct stage_data *entry)2931{2932 int clean_merge = 1;2933 int normalize = o->renormalize;2934 unsigned o_mode = entry->stages[1].mode;2935 unsigned a_mode = entry->stages[2].mode;2936 unsigned b_mode = entry->stages[3].mode;2937 struct object_id *o_oid = stage_oid(&entry->stages[1].oid, o_mode);2938 struct object_id *a_oid = stage_oid(&entry->stages[2].oid, a_mode);2939 struct object_id *b_oid = stage_oid(&entry->stages[3].oid, b_mode);29402941 entry->processed = 1;2942 if (entry->rename_conflict_info) {2943 struct rename_conflict_info *conflict_info = entry->rename_conflict_info;2944 switch (conflict_info->rename_type) {2945 case RENAME_NORMAL:2946 case RENAME_ONE_FILE_TO_ONE:2947 clean_merge = conflict_rename_normal(o,2948 path,2949 o_oid, o_mode,2950 a_oid, a_mode,2951 b_oid, b_mode,2952 conflict_info);2953 break;2954 case RENAME_DIR:2955 clean_merge = 1;2956 if (conflict_rename_dir(o,2957 conflict_info->pair1,2958 conflict_info->branch1,2959 conflict_info->branch2))2960 clean_merge = -1;2961 break;2962 case RENAME_DELETE:2963 clean_merge = 0;2964 if (conflict_rename_delete(o,2965 conflict_info->pair1,2966 conflict_info->branch1,2967 conflict_info->branch2))2968 clean_merge = -1;2969 break;2970 case RENAME_ONE_FILE_TO_TWO:2971 clean_merge = 0;2972 if (conflict_rename_rename_1to2(o, conflict_info))2973 clean_merge = -1;2974 break;2975 case RENAME_TWO_FILES_TO_ONE:2976 clean_merge = 0;2977 if (conflict_rename_rename_2to1(o, conflict_info))2978 clean_merge = -1;2979 break;2980 default:2981 entry->processed = 0;2982 break;2983 }2984 } else if (o_oid && (!a_oid || !b_oid)) {2985 /* Case A: Deleted in one */2986 if ((!a_oid && !b_oid) ||2987 (!b_oid && blob_unchanged(o, o_oid, o_mode, a_oid, a_mode, normalize, path)) ||2988 (!a_oid && blob_unchanged(o, o_oid, o_mode, b_oid, b_mode, normalize, path))) {2989 /* Deleted in both or deleted in one and2990 * unchanged in the other */2991 if (a_oid)2992 output(o, 2, _("Removing %s"), path);2993 /* do not touch working file if it did not exist */2994 remove_file(o, 1, path, !a_oid);2995 } else {2996 /* Modify/delete; deleted side may have put a directory in the way */2997 clean_merge = 0;2998 if (handle_modify_delete(o, path, o_oid, o_mode,2999 a_oid, a_mode, b_oid, b_mode))3000 clean_merge = -1;3001 }3002 } else if ((!o_oid && a_oid && !b_oid) ||3003 (!o_oid && !a_oid && b_oid)) {3004 /* Case B: Added in one. */3005 /* [nothing|directory] -> ([nothing|directory], file) */30063007 const char *add_branch;3008 const char *other_branch;3009 unsigned mode;3010 const struct object_id *oid;3011 const char *conf;30123013 if (a_oid) {3014 add_branch = o->branch1;3015 other_branch = o->branch2;3016 mode = a_mode;3017 oid = a_oid;3018 conf = _("file/directory");3019 } else {3020 add_branch = o->branch2;3021 other_branch = o->branch1;3022 mode = b_mode;3023 oid = b_oid;3024 conf = _("directory/file");3025 }3026 if (dir_in_way(path,3027 !o->call_depth && !S_ISGITLINK(a_mode),3028 0)) {3029 char *new_path = unique_path(o, path, add_branch);3030 clean_merge = 0;3031 output(o, 1, _("CONFLICT (%s): There is a directory with name %s in %s. "3032 "Adding %s as %s"),3033 conf, path, other_branch, path, new_path);3034 if (update_file(o, 0, oid, mode, new_path))3035 clean_merge = -1;3036 else if (o->call_depth)3037 remove_file_from_cache(path);3038 free(new_path);3039 } else {3040 output(o, 2, _("Adding %s"), path);3041 /* do not overwrite file if already present */3042 if (update_file_flags(o, oid, mode, path, 1, !a_oid))3043 clean_merge = -1;3044 }3045 } else if (a_oid && b_oid) {3046 /* Case C: Added in both (check for same permissions) and */3047 /* case D: Modified in both, but differently. */3048 int is_dirty = 0; /* unpack_trees would have bailed if dirty */3049 clean_merge = merge_content(o, path, is_dirty,3050 o_oid, o_mode, a_oid, a_mode, b_oid, b_mode,3051 NULL);3052 } else if (!o_oid && !a_oid && !b_oid) {3053 /*3054 * this entry was deleted altogether. a_mode == 0 means3055 * we had that path and want to actively remove it.3056 */3057 remove_file(o, 1, path, !a_mode);3058 } else3059 die("BUG: fatal merge failure, shouldn't happen.");30603061 return clean_merge;3062}30633064int merge_trees(struct merge_options *o,3065 struct tree *head,3066 struct tree *merge,3067 struct tree *common,3068 struct tree **result)3069{3070 int code, clean;30713072 if (o->subtree_shift) {3073 merge = shift_tree_object(head, merge, o->subtree_shift);3074 common = shift_tree_object(head, common, o->subtree_shift);3075 }30763077 if (oid_eq(&common->object.oid, &merge->object.oid)) {3078 struct strbuf sb = STRBUF_INIT;30793080 if (!o->call_depth && index_has_changes(&sb)) {3081 err(o, _("Dirty index: cannot merge (dirty: %s)"),3082 sb.buf);3083 return 0;3084 }3085 output(o, 0, _("Already up to date!"));3086 *result = head;3087 return 1;3088 }30893090 code = git_merge_trees(o, common, head, merge);30913092 if (code != 0) {3093 if (show(o, 4) || o->call_depth)3094 err(o, _("merging of trees %s and %s failed"),3095 oid_to_hex(&head->object.oid),3096 oid_to_hex(&merge->object.oid));3097 return -1;3098 }30993100 if (unmerged_cache()) {3101 struct string_list *entries;3102 struct rename_info re_info;3103 int i;3104 /*3105 * Only need the hashmap while processing entries, so3106 * initialize it here and free it when we are done running3107 * through the entries. Keeping it in the merge_options as3108 * opposed to decaring a local hashmap is for convenience3109 * so that we don't have to pass it to around.3110 */3111 hashmap_init(&o->current_file_dir_set, path_hashmap_cmp, NULL, 512);3112 get_files_dirs(o, head);3113 get_files_dirs(o, merge);31143115 entries = get_unmerged();3116 clean = handle_renames(o, common, head, merge, entries,3117 &re_info);3118 record_df_conflict_files(o, entries);3119 if (clean < 0)3120 goto cleanup;3121 for (i = entries->nr-1; 0 <= i; i--) {3122 const char *path = entries->items[i].string;3123 struct stage_data *e = entries->items[i].util;3124 if (!e->processed) {3125 int ret = process_entry(o, path, e);3126 if (!ret)3127 clean = 0;3128 else if (ret < 0) {3129 clean = ret;3130 goto cleanup;3131 }3132 }3133 }3134 for (i = 0; i < entries->nr; i++) {3135 struct stage_data *e = entries->items[i].util;3136 if (!e->processed)3137 die("BUG: unprocessed path??? %s",3138 entries->items[i].string);3139 }31403141cleanup:3142 final_cleanup_renames(&re_info);31433144 string_list_clear(entries, 1);3145 free(entries);31463147 hashmap_free(&o->current_file_dir_set, 1);31483149 if (clean < 0)3150 return clean;3151 }3152 else3153 clean = 1;31543155 /* Free the extra index left from git_merge_trees() */3156 /*3157 * FIXME: Need to also free data allocated by3158 * setup_unpack_trees_porcelain() tucked away in o->unpack_opts.msgs,3159 * but the problem is that only half of it refers to dynamically3160 * allocated data, while the other half points at static strings.3161 */3162 discard_index(&o->orig_index);31633164 if (o->call_depth && !(*result = write_tree_from_memory(o)))3165 return -1;31663167 return clean;3168}31693170static struct commit_list *reverse_commit_list(struct commit_list *list)3171{3172 struct commit_list *next = NULL, *current, *backup;3173 for (current = list; current; current = backup) {3174 backup = current->next;3175 current->next = next;3176 next = current;3177 }3178 return next;3179}31803181/*3182 * Merge the commits h1 and h2, return the resulting virtual3183 * commit object and a flag indicating the cleanness of the merge.3184 */3185int merge_recursive(struct merge_options *o,3186 struct commit *h1,3187 struct commit *h2,3188 struct commit_list *ca,3189 struct commit **result)3190{3191 struct commit_list *iter;3192 struct commit *merged_common_ancestors;3193 struct tree *mrtree;3194 int clean;31953196 if (show(o, 4)) {3197 output(o, 4, _("Merging:"));3198 output_commit_title(o, h1);3199 output_commit_title(o, h2);3200 }32013202 if (!ca) {3203 ca = get_merge_bases(h1, h2);3204 ca = reverse_commit_list(ca);3205 }32063207 if (show(o, 5)) {3208 unsigned cnt = commit_list_count(ca);32093210 output(o, 5, Q_("found %u common ancestor:",3211 "found %u common ancestors:", cnt), cnt);3212 for (iter = ca; iter; iter = iter->next)3213 output_commit_title(o, iter->item);3214 }32153216 merged_common_ancestors = pop_commit(&ca);3217 if (merged_common_ancestors == NULL) {3218 /* if there is no common ancestor, use an empty tree */3219 struct tree *tree;32203221 tree = lookup_tree(the_hash_algo->empty_tree);3222 merged_common_ancestors = make_virtual_commit(tree, "ancestor");3223 }32243225 for (iter = ca; iter; iter = iter->next) {3226 const char *saved_b1, *saved_b2;3227 o->call_depth++;3228 /*3229 * When the merge fails, the result contains files3230 * with conflict markers. The cleanness flag is3231 * ignored (unless indicating an error), it was never3232 * actually used, as result of merge_trees has always3233 * overwritten it: the committed "conflicts" were3234 * already resolved.3235 */3236 discard_cache();3237 saved_b1 = o->branch1;3238 saved_b2 = o->branch2;3239 o->branch1 = "Temporary merge branch 1";3240 o->branch2 = "Temporary merge branch 2";3241 if (merge_recursive(o, merged_common_ancestors, iter->item,3242 NULL, &merged_common_ancestors) < 0)3243 return -1;3244 o->branch1 = saved_b1;3245 o->branch2 = saved_b2;3246 o->call_depth--;32473248 if (!merged_common_ancestors)3249 return err(o, _("merge returned no commit"));3250 }32513252 discard_cache();3253 if (!o->call_depth)3254 read_cache();32553256 o->ancestor = "merged common ancestors";3257 clean = merge_trees(o, h1->tree, h2->tree, merged_common_ancestors->tree,3258 &mrtree);3259 if (clean < 0) {3260 flush_output(o);3261 return clean;3262 }32633264 if (o->call_depth) {3265 *result = make_virtual_commit(mrtree, "merged tree");3266 commit_list_insert(h1, &(*result)->parents);3267 commit_list_insert(h2, &(*result)->parents->next);3268 }3269 flush_output(o);3270 if (!o->call_depth && o->buffer_output < 2)3271 strbuf_release(&o->obuf);3272 if (show(o, 2))3273 diff_warn_rename_limit("merge.renamelimit",3274 o->needed_rename_limit, 0);3275 return clean;3276}32773278static struct commit *get_ref(const struct object_id *oid, const char *name)3279{3280 struct object *object;32813282 object = deref_tag(parse_object(oid), name, strlen(name));3283 if (!object)3284 return NULL;3285 if (object->type == OBJ_TREE)3286 return make_virtual_commit((struct tree*)object, name);3287 if (object->type != OBJ_COMMIT)3288 return NULL;3289 if (parse_commit((struct commit *)object))3290 return NULL;3291 return (struct commit *)object;3292}32933294int merge_recursive_generic(struct merge_options *o,3295 const struct object_id *head,3296 const struct object_id *merge,3297 int num_base_list,3298 const struct object_id **base_list,3299 struct commit **result)3300{3301 int clean;3302 struct lock_file lock = LOCK_INIT;3303 struct commit *head_commit = get_ref(head, o->branch1);3304 struct commit *next_commit = get_ref(merge, o->branch2);3305 struct commit_list *ca = NULL;33063307 if (base_list) {3308 int i;3309 for (i = 0; i < num_base_list; ++i) {3310 struct commit *base;3311 if (!(base = get_ref(base_list[i], oid_to_hex(base_list[i]))))3312 return err(o, _("Could not parse object '%s'"),3313 oid_to_hex(base_list[i]));3314 commit_list_insert(base, &ca);3315 }3316 }33173318 hold_locked_index(&lock, LOCK_DIE_ON_ERROR);3319 clean = merge_recursive(o, head_commit, next_commit, ca,3320 result);3321 if (clean < 0) {3322 rollback_lock_file(&lock);3323 return clean;3324 }33253326 if (write_locked_index(&the_index, &lock,3327 COMMIT_LOCK | SKIP_IF_UNCHANGED))3328 return err(o, _("Unable to write index."));33293330 return clean ? 0 : 1;3331}33323333static void merge_recursive_config(struct merge_options *o)3334{3335 char *value = NULL;3336 git_config_get_int("merge.verbosity", &o->verbosity);3337 git_config_get_int("diff.renamelimit", &o->diff_rename_limit);3338 git_config_get_int("merge.renamelimit", &o->merge_rename_limit);3339 if (!git_config_get_string("diff.renames", &value)) {3340 o->diff_detect_rename = git_config_rename("diff.renames", value);3341 free(value);3342 }3343 if (!git_config_get_string("merge.renames", &value)) {3344 o->merge_detect_rename = git_config_rename("merge.renames", value);3345 free(value);3346 }3347 git_config(git_xmerge_config, NULL);3348}33493350void init_merge_options(struct merge_options *o)3351{3352 const char *merge_verbosity;3353 memset(o, 0, sizeof(struct merge_options));3354 o->verbosity = 2;3355 o->buffer_output = 1;3356 o->diff_rename_limit = -1;3357 o->merge_rename_limit = -1;3358 o->renormalize = 0;3359 o->diff_detect_rename = -1;3360 o->merge_detect_rename = -1;3361 merge_recursive_config(o);3362 merge_verbosity = getenv("GIT_MERGE_VERBOSITY");3363 if (merge_verbosity)3364 o->verbosity = strtol(merge_verbosity, NULL, 10);3365 if (o->verbosity >= 5)3366 o->buffer_output = 0;3367 strbuf_init(&o->obuf, 0);3368 string_list_init(&o->df_conflict_file_set, 1);3369}33703371int parse_merge_opt(struct merge_options *o, const char *s)3372{3373 const char *arg;33743375 if (!s || !*s)3376 return -1;3377 if (!strcmp(s, "ours"))3378 o->recursive_variant = MERGE_RECURSIVE_OURS;3379 else if (!strcmp(s, "theirs"))3380 o->recursive_variant = MERGE_RECURSIVE_THEIRS;3381 else if (!strcmp(s, "subtree"))3382 o->subtree_shift = "";3383 else if (skip_prefix(s, "subtree=", &arg))3384 o->subtree_shift = arg;3385 else if (!strcmp(s, "patience"))3386 o->xdl_opts = DIFF_WITH_ALG(o, PATIENCE_DIFF);3387 else if (!strcmp(s, "histogram"))3388 o->xdl_opts = DIFF_WITH_ALG(o, HISTOGRAM_DIFF);3389 else if (skip_prefix(s, "diff-algorithm=", &arg)) {3390 long value = parse_algorithm_value(arg);3391 if (value < 0)3392 return -1;3393 /* clear out previous settings */3394 DIFF_XDL_CLR(o, NEED_MINIMAL);3395 o->xdl_opts &= ~XDF_DIFF_ALGORITHM_MASK;3396 o->xdl_opts |= value;3397 }3398 else if (!strcmp(s, "ignore-space-change"))3399 DIFF_XDL_SET(o, IGNORE_WHITESPACE_CHANGE);3400 else if (!strcmp(s, "ignore-all-space"))3401 DIFF_XDL_SET(o, IGNORE_WHITESPACE);3402 else if (!strcmp(s, "ignore-space-at-eol"))3403 DIFF_XDL_SET(o, IGNORE_WHITESPACE_AT_EOL);3404 else if (!strcmp(s, "ignore-cr-at-eol"))3405 DIFF_XDL_SET(o, IGNORE_CR_AT_EOL);3406 else if (!strcmp(s, "renormalize"))3407 o->renormalize = 1;3408 else if (!strcmp(s, "no-renormalize"))3409 o->renormalize = 0;3410 else if (!strcmp(s, "no-renames"))3411 o->merge_detect_rename = 0;3412 else if (!strcmp(s, "find-renames")) {3413 o->merge_detect_rename = 1;3414 o->rename_score = 0;3415 }3416 else if (skip_prefix(s, "find-renames=", &arg) ||3417 skip_prefix(s, "rename-threshold=", &arg)) {3418 if ((o->rename_score = parse_rename_score(&arg)) == -1 || *arg != 0)3419 return -1;3420 o->merge_detect_rename = 1;3421 }3422 else3423 return -1;3424 return 0;3425}