1/* 2 * Copyright (C) 2005 Junio C Hamano 3 */ 4#include"cache.h" 5#include"diff.h" 6#include"diffcore.h" 7#include"delta.h" 8#include"count-delta.h" 9 10/* Table of rename/copy destinations */ 11 12static struct diff_rename_dst { 13struct diff_filespec *two; 14struct diff_filepair *pair; 15} *rename_dst; 16static int rename_dst_nr, rename_dst_alloc; 17 18static struct diff_rename_dst *locate_rename_dst(struct diff_filespec *two, 19int insert_ok) 20{ 21int first, last; 22 23 first =0; 24 last = rename_dst_nr; 25while(last > first) { 26int next = (last + first) >>1; 27struct diff_rename_dst *dst = &(rename_dst[next]); 28int cmp =strcmp(two->path, dst->two->path); 29if(!cmp) 30return dst; 31if(cmp <0) { 32 last = next; 33continue; 34} 35 first = next+1; 36} 37/* not found */ 38if(!insert_ok) 39return NULL; 40/* insert to make it at "first" */ 41if(rename_dst_alloc <= rename_dst_nr) { 42 rename_dst_alloc =alloc_nr(rename_dst_alloc); 43 rename_dst =xrealloc(rename_dst, 44 rename_dst_alloc *sizeof(*rename_dst)); 45} 46 rename_dst_nr++; 47if(first < rename_dst_nr) 48memmove(rename_dst + first +1, rename_dst + first, 49(rename_dst_nr - first -1) *sizeof(*rename_dst)); 50 rename_dst[first].two = two; 51 rename_dst[first].pair = NULL; 52return&(rename_dst[first]); 53} 54 55/* Table of rename/copy src files */ 56static struct diff_rename_src { 57struct diff_filespec *one; 58unsigned 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, 63int src_path_left) 64{ 65int first, last; 66 67 first =0; 68 last = rename_src_nr; 69while(last > first) { 70int next = (last + first) >>1; 71struct diff_rename_src *src = &(rename_src[next]); 72int cmp =strcmp(one->path, src->one->path); 73if(!cmp) 74return src; 75if(cmp <0) { 76 last = next; 77continue; 78} 79 first = next+1; 80} 81 82/* insert to make it at "first" */ 83if(rename_src_alloc <= rename_src_nr) { 84 rename_src_alloc =alloc_nr(rename_src_alloc); 85 rename_src =xrealloc(rename_src, 86 rename_src_alloc *sizeof(*rename_src)); 87} 88 rename_src_nr++; 89if(first < rename_src_nr) 90memmove(rename_src + first +1, rename_src + first, 91(rename_src_nr - first -1) *sizeof(*rename_src)); 92 rename_src[first].one = one; 93 rename_src[first].src_path_left = src_path_left; 94return&(rename_src[first]); 95} 96 97static intis_exact_match(struct diff_filespec *src,struct diff_filespec *dst) 98{ 99if(src->sha1_valid && dst->sha1_valid && 100!memcmp(src->sha1, dst->sha1,20)) 101return1; 102if(diff_populate_filespec(src,1) ||diff_populate_filespec(dst,1)) 103return0; 104if(src->size != dst->size) 105return0; 106if(diff_populate_filespec(src,0) ||diff_populate_filespec(dst,0)) 107return0; 108if(src->size == dst->size && 109!memcmp(src->data, dst->data, src->size)) 110return1; 111return0; 112} 113 114struct diff_score { 115int src;/* index in rename_src */ 116int dst;/* index in rename_dst */ 117int score; 118}; 119 120static intestimate_similarity(struct diff_filespec *src, 121struct diff_filespec *dst, 122int minimum_score) 123{ 124/* src points at a file that existed in the original tree (or 125 * optionally a file in the destination tree) and dst points 126 * at a newly created file. They may be quite similar, in which 127 * case we want to say src is renamed to dst or src is copied into 128 * dst, and then some edit has been applied to dst. 129 * 130 * Compare them and return how similar they are, representing 131 * the score as an integer between 0 and MAX_SCORE. 132 * 133 * When there is an exact match, it is considered a better 134 * match than anything else; the destination does not even 135 * call into this function in that case. 136 */ 137void*delta; 138unsigned long delta_size, base_size, src_copied, literal_added; 139unsigned long delta_limit; 140int score; 141 142/* We deal only with regular files. Symlink renames are handled 143 * only when they are exact matches --- in other words, no edits 144 * after renaming. 145 */ 146if(!S_ISREG(src->mode) || !S_ISREG(dst->mode)) 147return0; 148 149 delta_size = ((src->size < dst->size) ? 150(dst->size - src->size) : (src->size - dst->size)); 151 base_size = ((src->size < dst->size) ? src->size : dst->size); 152 153/* We would not consider edits that change the file size so 154 * drastically. delta_size must be smaller than 155 * (MAX_SCORE-minimum_score)/MAX_SCORE * min(src->size, dst->size). 156 * 157 * Note that base_size == 0 case is handled here already 158 * and the final score computation below would not have a 159 * divide-by-zero issue. 160 */ 161if(base_size * (MAX_SCORE-minimum_score) < delta_size * MAX_SCORE) 162return0; 163 164if(diff_populate_filespec(src,0) ||diff_populate_filespec(dst,0)) 165return0;/* error but caught downstream */ 166 167 delta_limit = base_size * (MAX_SCORE-minimum_score) / MAX_SCORE; 168 delta =diff_delta(src->data, src->size, 169 dst->data, dst->size, 170&delta_size, delta_limit); 171if(!delta) 172/* If delta_limit is exceeded, we have too much differences */ 173return0; 174 175/* A delta that has a lot of literal additions would have 176 * big delta_size no matter what else it does. 177 */ 178if(base_size * (MAX_SCORE-minimum_score) < delta_size * MAX_SCORE) 179return0; 180 181/* Estimate the edit size by interpreting delta. */ 182if(count_delta(delta, delta_size, &src_copied, &literal_added)) { 183free(delta); 184return0; 185} 186free(delta); 187 188/* Extent of damage */ 189if(src->size + literal_added < src_copied) 190 delta_size =0; 191else 192 delta_size = (src->size - src_copied) + literal_added; 193 194/* 195 * Now we will give some score to it. 100% edit gets 0 points 196 * and 0% edit gets MAX_SCORE points. 197 */ 198 score = MAX_SCORE - (MAX_SCORE * delta_size / base_size); 199if(score <0)return0; 200if(MAX_SCORE < score)return MAX_SCORE; 201return score; 202} 203 204static voidrecord_rename_pair(struct diff_queue_struct *renq, 205int dst_index,int src_index,int score) 206{ 207struct diff_filespec *one, *two, *src, *dst; 208struct diff_filepair *dp; 209 210if(rename_dst[dst_index].pair) 211die("internal error: dst already matched."); 212 213 src = rename_src[src_index].one; 214 one =alloc_filespec(src->path); 215fill_filespec(one, src->sha1, src->mode); 216 217 dst = rename_dst[dst_index].two; 218 two =alloc_filespec(dst->path); 219fill_filespec(two, dst->sha1, dst->mode); 220 221 dp =diff_queue(renq, one, two); 222 dp->score = score; 223 dp->source_stays = rename_src[src_index].src_path_left; 224 rename_dst[dst_index].pair = dp; 225} 226 227/* 228 * We sort the rename similarity matrix with the score, in descending 229 * order (the most similar first). 230 */ 231static intscore_compare(const void*a_,const void*b_) 232{ 233const struct diff_score *a = a_, *b = b_; 234return b->score - a->score; 235} 236 237static intcompute_stays(struct diff_queue_struct *q, 238struct diff_filespec *one) 239{ 240int i; 241for(i =0; i < q->nr; i++) { 242struct diff_filepair *p = q->queue[i]; 243if(strcmp(one->path, p->two->path)) 244continue; 245if(DIFF_PAIR_RENAME(p)) { 246return0;/* something else is renamed into this */ 247} 248} 249return1; 250} 251 252voiddiffcore_rename(int detect_rename,int minimum_score) 253{ 254struct diff_queue_struct *q = &diff_queued_diff; 255struct diff_queue_struct renq, outq; 256struct diff_score *mx; 257int i, j; 258int num_create, num_src, dst_cnt; 259 260if(!minimum_score) 261 minimum_score = DEFAULT_RENAME_SCORE; 262 renq.queue = NULL; 263 renq.nr = renq.alloc =0; 264 265for(i =0; i < q->nr; i++) { 266struct diff_filepair *p = q->queue[i]; 267if(!DIFF_FILE_VALID(p->one)) 268if(!DIFF_FILE_VALID(p->two)) 269continue;/* unmerged */ 270else 271locate_rename_dst(p->two,1); 272else if(!DIFF_FILE_VALID(p->two)) { 273/* If the source is a broken "delete", and 274 * they did not really want to get broken, 275 * that means the source actually stays. 276 */ 277int stays = (p->broken_pair && !p->score); 278register_rename_src(p->one, stays); 279} 280else if(detect_rename == DIFF_DETECT_COPY) 281register_rename_src(p->one,1); 282} 283if(rename_dst_nr ==0) 284goto cleanup;/* nothing to do */ 285 286/* We really want to cull the candidates list early 287 * with cheap tests in order to avoid doing deltas. 288 */ 289for(i =0; i < rename_dst_nr; i++) { 290struct diff_filespec *two = rename_dst[i].two; 291for(j =0; j < rename_src_nr; j++) { 292struct diff_filespec *one = rename_src[j].one; 293if(!is_exact_match(one, two)) 294continue; 295record_rename_pair(&renq, i, j, MAX_SCORE); 296break;/* we are done with this entry */ 297} 298} 299diff_debug_queue("done detecting exact", &renq); 300 301/* Have we run out the created file pool? If so we can avoid 302 * doing the delta matrix altogether. 303 */ 304if(renq.nr == rename_dst_nr) 305goto cleanup; 306 307 num_create = (rename_dst_nr - renq.nr); 308 num_src = rename_src_nr; 309 mx =xmalloc(sizeof(*mx) * num_create * num_src); 310for(dst_cnt = i =0; i < rename_dst_nr; i++) { 311int base = dst_cnt * num_src; 312struct diff_filespec *two = rename_dst[i].two; 313if(rename_dst[i].pair) 314continue;/* dealt with exact match already. */ 315for(j =0; j < rename_src_nr; j++) { 316struct diff_filespec *one = rename_src[j].one; 317struct diff_score *m = &mx[base+j]; 318 m->src = j; 319 m->dst = i; 320 m->score =estimate_similarity(one, two, 321 minimum_score); 322} 323 dst_cnt++; 324} 325/* cost matrix sorted by most to least similar pair */ 326qsort(mx, num_create * num_src,sizeof(*mx), score_compare); 327for(i =0; i < num_create * num_src; i++) { 328struct diff_rename_dst *dst = &rename_dst[mx[i].dst]; 329if(dst->pair) 330continue;/* already done, either exact or fuzzy. */ 331if(mx[i].score < minimum_score) 332break;/* there is no more usable pair. */ 333record_rename_pair(&renq, mx[i].dst, mx[i].src, mx[i].score); 334} 335free(mx); 336diff_debug_queue("done detecting fuzzy", &renq); 337 338 cleanup: 339/* At this point, we have found some renames and copies and they 340 * are kept in renq. The original list is still in *q. 341 */ 342 outq.queue = NULL; 343 outq.nr = outq.alloc =0; 344for(i =0; i < q->nr; i++) { 345struct diff_filepair *p = q->queue[i]; 346struct diff_filepair *pair_to_free = NULL; 347 348if(!DIFF_FILE_VALID(p->one) &&DIFF_FILE_VALID(p->two)) { 349/* 350 * Creation 351 * 352 * We would output this create record if it has 353 * not been turned into a rename/copy already. 354 */ 355struct diff_rename_dst *dst = 356locate_rename_dst(p->two,0); 357if(dst && dst->pair) { 358diff_q(&outq, dst->pair); 359 pair_to_free = p; 360} 361else 362/* no matching rename/copy source, so 363 * record this as a creation. 364 */ 365diff_q(&outq, p); 366} 367else if(DIFF_FILE_VALID(p->one) && !DIFF_FILE_VALID(p->two)) { 368/* 369 * Deletion 370 * 371 * We would output this delete record if: 372 * 373 * (1) this is a broken delete and the counterpart 374 * broken create remains in the output; or 375 * (2) this is not a broken delete, and renq does 376 * not have a rename/copy to move p->one->path 377 * out. 378 * 379 * Otherwise, the counterpart broken create 380 * has been turned into a rename-edit; or 381 * delete did not have a matching create to 382 * begin with. 383 */ 384if(DIFF_PAIR_BROKEN(p)) { 385/* broken delete */ 386struct diff_rename_dst *dst = 387locate_rename_dst(p->one,0); 388if(dst && dst->pair) 389/* counterpart is now rename/copy */ 390 pair_to_free = p; 391} 392else{ 393for(j =0; j < renq.nr; j++) 394if(!strcmp(renq.queue[j]->one->path, 395 p->one->path)) 396break; 397if(j < renq.nr) 398/* this path remains */ 399 pair_to_free = p; 400} 401 402if(pair_to_free) 403; 404else 405diff_q(&outq, p); 406} 407else if(!diff_unmodified_pair(p)) 408/* all the usual ones need to be kept */ 409diff_q(&outq, p); 410else 411/* no need to keep unmodified pairs */ 412 pair_to_free = p; 413 414if(pair_to_free) 415diff_free_filepair(pair_to_free); 416} 417diff_debug_queue("done copying original", &outq); 418 419free(renq.queue); 420free(q->queue); 421*q = outq; 422diff_debug_queue("done collapsing", q); 423 424/* We need to see which rename source really stays here; 425 * earlier we only checked if the path is left in the result, 426 * but even if a path remains in the result, if that is coming 427 * from copying something else on top of it, then the original 428 * source is lost and does not stay. 429 */ 430for(i =0; i < q->nr; i++) { 431struct diff_filepair *p = q->queue[i]; 432if(DIFF_PAIR_RENAME(p) && p->source_stays) { 433/* If one appears as the target of a rename-copy, 434 * then mark p->source_stays = 0; otherwise 435 * leave it as is. 436 */ 437 p->source_stays =compute_stays(q, p->one); 438} 439} 440 441free(rename_dst); 442 rename_dst = NULL; 443 rename_dst_nr = rename_dst_alloc =0; 444free(rename_src); 445 rename_src = NULL; 446 rename_src_nr = rename_src_alloc =0; 447return; 448}