1/* 2 * Copyright (C) 2005 Junio C Hamano 3 */ 4#include "cache.h" 5#include "diff.h" 6#include "diffcore.h" 7 8/* Table of rename/copy destinations */ 9 10static struct diff_rename_dst { 11 struct diff_filespec *two; 12 struct diff_filepair *pair; 13} *rename_dst; 14static int rename_dst_nr, rename_dst_alloc; 15 16static struct diff_rename_dst *locate_rename_dst(struct diff_filespec *two, 17 int insert_ok) 18{ 19 int first, last; 20 21 first = 0; 22 last = rename_dst_nr; 23 while (last > first) { 24 int next = (last + first) >> 1; 25 struct diff_rename_dst *dst = &(rename_dst[next]); 26 int cmp = strcmp(two->path, dst->two->path); 27 if (!cmp) 28 return dst; 29 if (cmp < 0) { 30 last = next; 31 continue; 32 } 33 first = next+1; 34 } 35 /* not found */ 36 if (!insert_ok) 37 return NULL; 38 /* insert to make it at "first" */ 39 if (rename_dst_alloc <= rename_dst_nr) { 40 rename_dst_alloc = alloc_nr(rename_dst_alloc); 41 rename_dst = xrealloc(rename_dst, 42 rename_dst_alloc * sizeof(*rename_dst)); 43 } 44 rename_dst_nr++; 45 if (first < rename_dst_nr) 46 memmove(rename_dst + first + 1, rename_dst + first, 47 (rename_dst_nr - first - 1) * sizeof(*rename_dst)); 48 rename_dst[first].two = alloc_filespec(two->path); 49 fill_filespec(rename_dst[first].two, two->sha1, two->mode); 50 rename_dst[first].pair = NULL; 51 return &(rename_dst[first]); 52} 53 54/* Table of rename/copy src files */ 55static struct diff_rename_src { 56 struct diff_filespec *one; 57 unsigned short score; /* to remember the break score */ 58 unsigned src_path_left : 1; 59} *rename_src; 60static int rename_src_nr, rename_src_alloc; 61 62static struct diff_rename_src *register_rename_src(struct diff_filespec *one, 63 int src_path_left, 64 unsigned short score) 65{ 66 int first, last; 67 68 first = 0; 69 last = rename_src_nr; 70 while (last > first) { 71 int next = (last + first) >> 1; 72 struct diff_rename_src *src = &(rename_src[next]); 73 int cmp = strcmp(one->path, src->one->path); 74 if (!cmp) 75 return src; 76 if (cmp < 0) { 77 last = next; 78 continue; 79 } 80 first = next+1; 81 } 82 83 /* insert to make it at "first" */ 84 if (rename_src_alloc <= rename_src_nr) { 85 rename_src_alloc = alloc_nr(rename_src_alloc); 86 rename_src = xrealloc(rename_src, 87 rename_src_alloc * sizeof(*rename_src)); 88 } 89 rename_src_nr++; 90 if (first < rename_src_nr) 91 memmove(rename_src + first + 1, rename_src + first, 92 (rename_src_nr - first - 1) * sizeof(*rename_src)); 93 rename_src[first].one = one; 94 rename_src[first].score = score; 95 rename_src[first].src_path_left = src_path_left; 96 return &(rename_src[first]); 97} 98 99static int is_exact_match(struct diff_filespec *src, 100 struct diff_filespec *dst, 101 int contents_too) 102{ 103 if (src->sha1_valid && dst->sha1_valid && 104 !hashcmp(src->sha1, dst->sha1)) 105 return 1; 106 if (!contents_too) 107 return 0; 108 if (diff_populate_filespec(src, 1) || diff_populate_filespec(dst, 1)) 109 return 0; 110 if (src->size != dst->size) 111 return 0; 112 if (src->sha1_valid && dst->sha1_valid) 113 return !hashcmp(src->sha1, dst->sha1); 114 if (diff_populate_filespec(src, 0) || diff_populate_filespec(dst, 0)) 115 return 0; 116 if (src->size == dst->size && 117 !memcmp(src->data, dst->data, src->size)) 118 return 1; 119 return 0; 120} 121 122static int basename_same(struct diff_filespec *src, struct diff_filespec *dst) 123{ 124 int src_len = strlen(src->path), dst_len = strlen(dst->path); 125 while (src_len && dst_len) { 126 char c1 = src->path[--src_len]; 127 char c2 = dst->path[--dst_len]; 128 if (c1 != c2) 129 return 0; 130 if (c1 == '/') 131 return 1; 132 } 133 return (!src_len || src->path[src_len - 1] == '/') && 134 (!dst_len || dst->path[dst_len - 1] == '/'); 135} 136 137struct diff_score { 138 int src; /* index in rename_src */ 139 int dst; /* index in rename_dst */ 140 int score; 141 int name_score; 142}; 143 144static int estimate_similarity(struct diff_filespec *src, 145 struct diff_filespec *dst, 146 int minimum_score) 147{ 148 /* src points at a file that existed in the original tree (or 149 * optionally a file in the destination tree) and dst points 150 * at a newly created file. They may be quite similar, in which 151 * case we want to say src is renamed to dst or src is copied into 152 * dst, and then some edit has been applied to dst. 153 * 154 * Compare them and return how similar they are, representing 155 * the score as an integer between 0 and MAX_SCORE. 156 * 157 * When there is an exact match, it is considered a better 158 * match than anything else; the destination does not even 159 * call into this function in that case. 160 */ 161 unsigned long max_size, delta_size, base_size, src_copied, literal_added; 162 unsigned long delta_limit; 163 int score; 164 165 /* We deal only with regular files. Symlink renames are handled 166 * only when they are exact matches --- in other words, no edits 167 * after renaming. 168 */ 169 if (!S_ISREG(src->mode) || !S_ISREG(dst->mode)) 170 return 0; 171 172 max_size = ((src->size > dst->size) ? src->size : dst->size); 173 base_size = ((src->size < dst->size) ? src->size : dst->size); 174 delta_size = max_size - base_size; 175 176 /* We would not consider edits that change the file size so 177 * drastically. delta_size must be smaller than 178 * (MAX_SCORE-minimum_score)/MAX_SCORE * min(src->size, dst->size). 179 * 180 * Note that base_size == 0 case is handled here already 181 * and the final score computation below would not have a 182 * divide-by-zero issue. 183 */ 184 if (base_size * (MAX_SCORE-minimum_score) < delta_size * MAX_SCORE) 185 return 0; 186 187 if ((!src->cnt_data && diff_populate_filespec(src, 0)) 188 || (!dst->cnt_data && diff_populate_filespec(dst, 0))) 189 return 0; /* error but caught downstream */ 190 191 192 delta_limit = (unsigned long) 193 (base_size * (MAX_SCORE-minimum_score) / MAX_SCORE); 194 if (diffcore_count_changes(src, dst, 195 &src->cnt_data, &dst->cnt_data, 196 delta_limit, 197 &src_copied, &literal_added)) 198 return 0; 199 200 /* How similar are they? 201 * what percentage of material in dst are from source? 202 */ 203 if (!dst->size) 204 score = 0; /* should not happen */ 205 else 206 score = (int)(src_copied * MAX_SCORE / max_size); 207 return score; 208} 209 210static void record_rename_pair(int dst_index, int src_index, int score) 211{ 212 struct diff_filespec *one, *two, *src, *dst; 213 struct diff_filepair *dp; 214 215 if (rename_dst[dst_index].pair) 216 die("internal error: dst already matched."); 217 218 src = rename_src[src_index].one; 219 one = alloc_filespec(src->path); 220 fill_filespec(one, src->sha1, src->mode); 221 222 dst = rename_dst[dst_index].two; 223 two = alloc_filespec(dst->path); 224 fill_filespec(two, dst->sha1, dst->mode); 225 226 dp = diff_queue(NULL, one, two); 227 dp->renamed_pair = 1; 228 if (!strcmp(src->path, dst->path)) 229 dp->score = rename_src[src_index].score; 230 else 231 dp->score = score; 232 dp->source_stays = rename_src[src_index].src_path_left; 233 rename_dst[dst_index].pair = dp; 234} 235 236/* 237 * We sort the rename similarity matrix with the score, in descending 238 * order (the most similar first). 239 */ 240static int score_compare(const void *a_, const void *b_) 241{ 242 const struct diff_score *a = a_, *b = b_; 243 244 if (a->score == b->score) 245 return b->name_score - a->name_score; 246 247 return b->score - a->score; 248} 249 250static int compute_stays(struct diff_queue_struct *q, 251 struct diff_filespec *one) 252{ 253 int i; 254 for (i = 0; i < q->nr; i++) { 255 struct diff_filepair *p = q->queue[i]; 256 if (strcmp(one->path, p->two->path)) 257 continue; 258 if (DIFF_PAIR_RENAME(p)) { 259 return 0; /* something else is renamed into this */ 260 } 261 } 262 return 1; 263} 264 265/* 266 * Find exact renames first. 267 * 268 * The first round matches up the up-to-date entries, 269 * and then during the second round we try to match 270 * cache-dirty entries as well. 271 * 272 * Note: the rest of the rename logic depends on this 273 * phase also populating all the filespecs for any 274 * entry that isn't matched up with an exact rename, 275 * see "is_exact_match()". 276 */ 277static int find_exact_renames(void) 278{ 279 int rename_count = 0; 280 int contents_too; 281 282 for (contents_too = 0; contents_too < 2; contents_too++) { 283 int i; 284 285 for (i = 0; i < rename_dst_nr; i++) { 286 struct diff_filespec *two = rename_dst[i].two; 287 int j; 288 289 if (rename_dst[i].pair) 290 continue; /* dealt with an earlier round */ 291 for (j = 0; j < rename_src_nr; j++) { 292 int k; 293 struct diff_filespec *one = rename_src[j].one; 294 if (!is_exact_match(one, two, contents_too)) 295 continue; 296 297 /* see if there is a basename match, too */ 298 for (k = j; k < rename_src_nr; k++) { 299 one = rename_src[k].one; 300 if (basename_same(one, two) && 301 is_exact_match(one, two, 302 contents_too)) { 303 j = k; 304 break; 305 } 306 } 307 308 record_rename_pair(i, j, (int)MAX_SCORE); 309 rename_count++; 310 break; /* we are done with this entry */ 311 } 312 } 313 } 314 return rename_count; 315} 316 317void diffcore_rename(struct diff_options *options) 318{ 319 int detect_rename = options->detect_rename; 320 int minimum_score = options->rename_score; 321 int rename_limit = options->rename_limit; 322 struct diff_queue_struct *q = &diff_queued_diff; 323 struct diff_queue_struct outq; 324 struct diff_score *mx; 325 int i, j, rename_count; 326 int num_create, num_src, dst_cnt; 327 328 if (!minimum_score) 329 minimum_score = DEFAULT_RENAME_SCORE; 330 331 for (i = 0; i < q->nr; i++) { 332 struct diff_filepair *p = q->queue[i]; 333 if (!DIFF_FILE_VALID(p->one)) { 334 if (!DIFF_FILE_VALID(p->two)) 335 continue; /* unmerged */ 336 else if (options->single_follow && 337 strcmp(options->single_follow, p->two->path)) 338 continue; /* not interested */ 339 else 340 locate_rename_dst(p->two, 1); 341 } 342 else if (!DIFF_FILE_VALID(p->two)) { 343 /* If the source is a broken "delete", and 344 * they did not really want to get broken, 345 * that means the source actually stays. 346 */ 347 int stays = (p->broken_pair && !p->score); 348 register_rename_src(p->one, stays, p->score); 349 } 350 else if (detect_rename == DIFF_DETECT_COPY) 351 register_rename_src(p->one, 1, p->score); 352 } 353 if (rename_dst_nr == 0 || rename_src_nr == 0) 354 goto cleanup; /* nothing to do */ 355 356 /* 357 * This basically does a test for the rename matrix not 358 * growing larger than a "rename_limit" square matrix, ie: 359 * 360 * rename_dst_nr * rename_src_nr > rename_limit * rename_limit 361 * 362 * but handles the potential overflow case specially (and we 363 * assume at least 32-bit integers) 364 */ 365 if (rename_limit <= 0 || rename_limit > 32767) 366 rename_limit = 32767; 367 if (rename_dst_nr > rename_limit && rename_src_nr > rename_limit) 368 goto cleanup; 369 if (rename_dst_nr * rename_src_nr > rename_limit * rename_limit) 370 goto cleanup; 371 372 /* 373 * We really want to cull the candidates list early 374 * with cheap tests in order to avoid doing deltas. 375 */ 376 rename_count = find_exact_renames(); 377 378 /* Have we run out the created file pool? If so we can avoid 379 * doing the delta matrix altogether. 380 */ 381 if (rename_count == rename_dst_nr) 382 goto cleanup; 383 384 if (minimum_score == MAX_SCORE) 385 goto cleanup; 386 387 num_create = (rename_dst_nr - rename_count); 388 num_src = rename_src_nr; 389 mx = xmalloc(sizeof(*mx) * num_create * num_src); 390 for (dst_cnt = i = 0; i < rename_dst_nr; i++) { 391 int base = dst_cnt * num_src; 392 struct diff_filespec *two = rename_dst[i].two; 393 if (rename_dst[i].pair) 394 continue; /* dealt with exact match already. */ 395 for (j = 0; j < rename_src_nr; j++) { 396 struct diff_filespec *one = rename_src[j].one; 397 struct diff_score *m = &mx[base+j]; 398 m->src = j; 399 m->dst = i; 400 m->score = estimate_similarity(one, two, 401 minimum_score); 402 m->name_score = basename_same(one, two); 403 diff_free_filespec_blob(one); 404 } 405 /* We do not need the text anymore */ 406 diff_free_filespec_blob(two); 407 dst_cnt++; 408 } 409 /* cost matrix sorted by most to least similar pair */ 410 qsort(mx, num_create * num_src, sizeof(*mx), score_compare); 411 for (i = 0; i < num_create * num_src; i++) { 412 struct diff_rename_dst *dst = &rename_dst[mx[i].dst]; 413 if (dst->pair) 414 continue; /* already done, either exact or fuzzy. */ 415 if (mx[i].score < minimum_score) 416 break; /* there is no more usable pair. */ 417 record_rename_pair(mx[i].dst, mx[i].src, mx[i].score); 418 rename_count++; 419 } 420 free(mx); 421 422 cleanup: 423 /* At this point, we have found some renames and copies and they 424 * are recorded in rename_dst. The original list is still in *q. 425 */ 426 outq.queue = NULL; 427 outq.nr = outq.alloc = 0; 428 for (i = 0; i < q->nr; i++) { 429 struct diff_filepair *p = q->queue[i]; 430 struct diff_filepair *pair_to_free = NULL; 431 432 if (!DIFF_FILE_VALID(p->one) && DIFF_FILE_VALID(p->two)) { 433 /* 434 * Creation 435 * 436 * We would output this create record if it has 437 * not been turned into a rename/copy already. 438 */ 439 struct diff_rename_dst *dst = 440 locate_rename_dst(p->two, 0); 441 if (dst && dst->pair) { 442 diff_q(&outq, dst->pair); 443 pair_to_free = p; 444 } 445 else 446 /* no matching rename/copy source, so 447 * record this as a creation. 448 */ 449 diff_q(&outq, p); 450 } 451 else if (DIFF_FILE_VALID(p->one) && !DIFF_FILE_VALID(p->two)) { 452 /* 453 * Deletion 454 * 455 * We would output this delete record if: 456 * 457 * (1) this is a broken delete and the counterpart 458 * broken create remains in the output; or 459 * (2) this is not a broken delete, and rename_dst 460 * does not have a rename/copy to move p->one->path 461 * out of existence. 462 * 463 * Otherwise, the counterpart broken create 464 * has been turned into a rename-edit; or 465 * delete did not have a matching create to 466 * begin with. 467 */ 468 if (DIFF_PAIR_BROKEN(p)) { 469 /* broken delete */ 470 struct diff_rename_dst *dst = 471 locate_rename_dst(p->one, 0); 472 if (dst && dst->pair) 473 /* counterpart is now rename/copy */ 474 pair_to_free = p; 475 } 476 else { 477 for (j = 0; j < rename_dst_nr; j++) { 478 if (!rename_dst[j].pair) 479 continue; 480 if (strcmp(rename_dst[j].pair-> 481 one->path, 482 p->one->path)) 483 continue; 484 break; 485 } 486 if (j < rename_dst_nr) 487 /* this path remains */ 488 pair_to_free = p; 489 } 490 491 if (pair_to_free) 492 ; 493 else 494 diff_q(&outq, p); 495 } 496 else if (!diff_unmodified_pair(p)) 497 /* all the usual ones need to be kept */ 498 diff_q(&outq, p); 499 else 500 /* no need to keep unmodified pairs */ 501 pair_to_free = p; 502 503 if (pair_to_free) 504 diff_free_filepair(pair_to_free); 505 } 506 diff_debug_queue("done copying original", &outq); 507 508 free(q->queue); 509 *q = outq; 510 diff_debug_queue("done collapsing", q); 511 512 /* We need to see which rename source really stays here; 513 * earlier we only checked if the path is left in the result, 514 * but even if a path remains in the result, if that is coming 515 * from copying something else on top of it, then the original 516 * source is lost and does not stay. 517 */ 518 for (i = 0; i < q->nr; i++) { 519 struct diff_filepair *p = q->queue[i]; 520 if (DIFF_PAIR_RENAME(p) && p->source_stays) { 521 /* If one appears as the target of a rename-copy, 522 * then mark p->source_stays = 0; otherwise 523 * leave it as is. 524 */ 525 p->source_stays = compute_stays(q, p->one); 526 } 527 } 528 529 for (i = 0; i < rename_dst_nr; i++) { 530 diff_free_filespec_data(rename_dst[i].two); 531 free(rename_dst[i].two); 532 } 533 534 free(rename_dst); 535 rename_dst = NULL; 536 rename_dst_nr = rename_dst_alloc = 0; 537 free(rename_src); 538 rename_src = NULL; 539 rename_src_nr = rename_src_alloc = 0; 540 return; 541}