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 10static int should_break(struct diff_filespec *src, 11 struct diff_filespec *dst, 12 int break_score, 13 int *merge_score_p) 14{ 15 /* dst is recorded as a modification of src. Are they so 16 * different that we are better off recording this as a pair 17 * of delete and create? 18 * 19 * There are two criteria used in this algorithm. For the 20 * purposes of helping later rename/copy, we take both delete 21 * and insert into account and estimate the amount of "edit". 22 * If the edit is very large, we break this pair so that 23 * rename/copy can pick the pieces up to match with other 24 * files. 25 * 26 * On the other hand, we would want to ignore inserts for the 27 * pure "complete rewrite" detection. As long as most of the 28 * existing contents were removed from the file, it is a 29 * complete rewrite, and if sizable chunk from the original 30 * still remains in the result, it is not a rewrite. It does 31 * not matter how much or how little new material is added to 32 * the file. 33 * 34 * The score we leave for such a broken filepair uses the 35 * latter definition so that later clean-up stage can find the 36 * pieces that should not have been broken according to the 37 * latter definition after rename/copy runs, and merge the 38 * broken pair that have a score lower than given criteria 39 * back together. The break operation itself happens 40 * according to the former definition. 41 * 42 * The minimum_edit parameter tells us when to break (the 43 * amount of "edit" required for us to consider breaking the 44 * pair). We leave the amount of deletion in *merge_score_p 45 * when we return. 46 * 47 * The value we return is 1 if we want the pair to be broken, 48 * or 0 if we do not. 49 */ 50 void *delta; 51 unsigned long delta_size, base_size, src_copied, literal_added; 52 int to_break = 0; 53 54 *merge_score_p = 0; /* assume no deletion --- "do not break" 55 * is the default. 56 */ 57 58 if (!S_ISREG(src->mode) || !S_ISREG(dst->mode)) 59 return 0; /* leave symlink rename alone */ 60 61 if (diff_populate_filespec(src, 0) || diff_populate_filespec(dst, 0)) 62 return 0; /* error but caught downstream */ 63 64 base_size = ((src->size < dst->size) ? src->size : dst->size); 65 66 delta = diff_delta(src->data, src->size, 67 dst->data, dst->size, 68 &delta_size, ~0UL); 69 70 /* Estimate the edit size by interpreting delta. */ 71 if (count_delta(delta, delta_size, 72 &src_copied, &literal_added)) { 73 free(delta); 74 return 0; /* we cannot tell */ 75 } 76 free(delta); 77 78 /* Compute merge-score, which is "how much is removed 79 * from the source material". The clean-up stage will 80 * merge the surviving pair together if the score is 81 * less than the minimum, after rename/copy runs. 82 */ 83 if (src->size <= src_copied) 84 ; /* all copied, nothing removed */ 85 else { 86 delta_size = src->size - src_copied; 87 *merge_score_p = delta_size * MAX_SCORE / src->size; 88 } 89 90 /* Extent of damage, which counts both inserts and 91 * deletes. 92 */ 93 if (src->size + literal_added <= src_copied) 94 delta_size = 0; /* avoid wrapping around */ 95 else 96 delta_size = (src->size - src_copied) + literal_added; 97 98 /* We break if the edit exceeds the minimum. 99 * i.e. (break_score / MAX_SCORE < delta_size / base_size) 100 */ 101 if (break_score * base_size < delta_size * MAX_SCORE) 102 to_break = 1; 103 104 return to_break; 105} 106 107void diffcore_break(int break_score) 108{ 109 struct diff_queue_struct *q = &diff_queued_diff; 110 struct diff_queue_struct outq; 111 112 /* When the filepair has this much edit (insert and delete), 113 * it is first considered to be a rewrite and broken into a 114 * create and delete filepair. This is to help breaking a 115 * file that had too much new stuff added, possibly from 116 * moving contents from another file, so that rename/copy can 117 * match it with the other file. 118 * 119 * int break_score; we reuse incoming parameter for this. 120 */ 121 122 /* After a pair is broken according to break_score and 123 * subjected to rename/copy, both of them may survive intact, 124 * due to lack of suitable rename/copy peer. Or, the caller 125 * may be calling us without using rename/copy. When that 126 * happens, we merge the broken pieces back into one 127 * modification together if the pair did not have more than 128 * this much delete. For this computation, we do not take 129 * insert into account at all. If you start from a 100-line 130 * file and delete 97 lines of it, it does not matter if you 131 * add 27 lines to it to make a new 30-line file or if you add 132 * 997 lines to it to make a 1000-line file. Either way what 133 * you did was a rewrite of 97%. On the other hand, if you 134 * delete 3 lines, keeping 97 lines intact, it does not matter 135 * if you add 3 lines to it to make a new 100-line file or if 136 * you add 903 lines to it to make a new 1000-line file. 137 * Either way you did a lot of additions and not a rewrite. 138 * This merge happens to catch the latter case. A merge_score 139 * of 80% would be a good default value (a broken pair that 140 * has score lower than merge_score will be merged back 141 * together). 142 */ 143 int merge_score; 144 int i; 145 146 /* See comment on DEFAULT_BREAK_SCORE and 147 * DEFAULT_MERGE_SCORE in diffcore.h 148 */ 149 merge_score = (break_score >> 16) & 0xFFFF; 150 break_score = (break_score & 0xFFFF); 151 152 if (!break_score) 153 break_score = DEFAULT_BREAK_SCORE; 154 if (!merge_score) 155 merge_score = DEFAULT_MERGE_SCORE; 156 157 outq.nr = outq.alloc = 0; 158 outq.queue = NULL; 159 160 for (i = 0; i < q->nr; i++) { 161 struct diff_filepair *p = q->queue[i]; 162 int score; 163 164 /* We deal only with in-place edit of non directory. 165 * We do not break anything else. 166 */ 167 if (DIFF_FILE_VALID(p->one) && DIFF_FILE_VALID(p->two) && 168 !S_ISDIR(p->one->mode) && !S_ISDIR(p->two->mode) && 169 !strcmp(p->one->path, p->two->path)) { 170 if (should_break(p->one, p->two, 171 break_score, &score) && 172 MINIMUM_BREAK_SIZE <= p->one->size) { 173 /* Split this into delete and create */ 174 struct diff_filespec *null_one, *null_two; 175 struct diff_filepair *dp; 176 177 /* Set score to 0 for the pair that 178 * needs to be merged back together 179 * should they survive rename/copy. 180 * Also we do not want to break very 181 * small files. 182 */ 183 if (score < merge_score) 184 score = 0; 185 186 /* deletion of one */ 187 null_one = alloc_filespec(p->one->path); 188 dp = diff_queue(&outq, p->one, null_one); 189 dp->score = score; 190 dp->broken_pair = 1; 191 192 /* creation of two */ 193 null_two = alloc_filespec(p->two->path); 194 dp = diff_queue(&outq, null_two, p->two); 195 dp->score = score; 196 dp->broken_pair = 1; 197 198 free(p); /* not diff_free_filepair(), we are 199 * reusing one and two here. 200 */ 201 continue; 202 } 203 } 204 diff_q(&outq, p); 205 } 206 free(q->queue); 207 *q = outq; 208 209 return; 210} 211 212static void merge_broken(struct diff_filepair *p, 213 struct diff_filepair *pp, 214 struct diff_queue_struct *outq) 215{ 216 /* p and pp are broken pairs we want to merge */ 217 struct diff_filepair *c = p, *d = pp, *dp; 218 if (DIFF_FILE_VALID(p->one)) { 219 /* this must be a delete half */ 220 d = p; c = pp; 221 } 222 /* Sanity check */ 223 if (!DIFF_FILE_VALID(d->one)) 224 die("internal error in merge #1"); 225 if (DIFF_FILE_VALID(d->two)) 226 die("internal error in merge #2"); 227 if (DIFF_FILE_VALID(c->one)) 228 die("internal error in merge #3"); 229 if (!DIFF_FILE_VALID(c->two)) 230 die("internal error in merge #4"); 231 232 dp = diff_queue(outq, d->one, c->two); 233 dp->score = p->score; 234 diff_free_filespec_data(d->two); 235 diff_free_filespec_data(c->one); 236 free(d); 237 free(c); 238} 239 240void diffcore_merge_broken(void) 241{ 242 struct diff_queue_struct *q = &diff_queued_diff; 243 struct diff_queue_struct outq; 244 int i, j; 245 246 outq.nr = outq.alloc = 0; 247 outq.queue = NULL; 248 249 for (i = 0; i < q->nr; i++) { 250 struct diff_filepair *p = q->queue[i]; 251 if (!p) 252 /* we already merged this with its peer */ 253 continue; 254 else if (p->broken_pair && 255 !strcmp(p->one->path, p->two->path)) { 256 /* If the peer also survived rename/copy, then 257 * we merge them back together. 258 */ 259 for (j = i + 1; j < q->nr; j++) { 260 struct diff_filepair *pp = q->queue[j]; 261 if (pp->broken_pair && 262 !strcmp(pp->one->path, pp->two->path) && 263 !strcmp(p->one->path, pp->two->path)) { 264 /* Peer survived. Merge them */ 265 merge_broken(p, pp, &outq); 266 q->queue[j] = NULL; 267 break; 268 } 269 } 270 if (q->nr <= j) 271 /* The peer did not survive, so we keep 272 * it in the output. 273 */ 274 diff_q(&outq, p); 275 } 276 else 277 diff_q(&outq, p); 278 } 279 free(q->queue); 280 *q = outq; 281 282 return; 283}