]>
Commit | Line | Data |
---|---|---|
1 | /* | |
2 | * Copyright (C) 2005 Junio C Hamano | |
3 | */ | |
4 | #include "cache.h" | |
5 | #include "diff.h" | |
6 | #include "diffcore.h" | |
7 | ||
8 | static int should_break(struct diff_filespec *src, | |
9 | struct diff_filespec *dst, | |
10 | int break_score, | |
11 | int *merge_score_p) | |
12 | { | |
13 | /* dst is recorded as a modification of src. Are they so | |
14 | * different that we are better off recording this as a pair | |
15 | * of delete and create? | |
16 | * | |
17 | * There are two criteria used in this algorithm. For the | |
18 | * purposes of helping later rename/copy, we take both delete | |
19 | * and insert into account and estimate the amount of "edit". | |
20 | * If the edit is very large, we break this pair so that | |
21 | * rename/copy can pick the pieces up to match with other | |
22 | * files. | |
23 | * | |
24 | * On the other hand, we would want to ignore inserts for the | |
25 | * pure "complete rewrite" detection. As long as most of the | |
26 | * existing contents were removed from the file, it is a | |
27 | * complete rewrite, and if sizable chunk from the original | |
28 | * still remains in the result, it is not a rewrite. It does | |
29 | * not matter how much or how little new material is added to | |
30 | * the file. | |
31 | * | |
32 | * The score we leave for such a broken filepair uses the | |
33 | * latter definition so that later clean-up stage can find the | |
34 | * pieces that should not have been broken according to the | |
35 | * latter definition after rename/copy runs, and merge the | |
36 | * broken pair that have a score lower than given criteria | |
37 | * back together. The break operation itself happens | |
38 | * according to the former definition. | |
39 | * | |
40 | * The minimum_edit parameter tells us when to break (the | |
41 | * amount of "edit" required for us to consider breaking the | |
42 | * pair). We leave the amount of deletion in *merge_score_p | |
43 | * when we return. | |
44 | * | |
45 | * The value we return is 1 if we want the pair to be broken, | |
46 | * or 0 if we do not. | |
47 | */ | |
48 | unsigned long delta_size, max_size; | |
49 | unsigned long src_copied, literal_added, src_removed; | |
50 | ||
51 | *merge_score_p = 0; /* assume no deletion --- "do not break" | |
52 | * is the default. | |
53 | */ | |
54 | ||
55 | if (S_ISREG(src->mode) != S_ISREG(dst->mode)) { | |
56 | *merge_score_p = (int)MAX_SCORE; | |
57 | return 1; /* even their types are different */ | |
58 | } | |
59 | ||
60 | if (src->sha1_valid && dst->sha1_valid && | |
61 | !hashcmp(src->sha1, dst->sha1)) | |
62 | return 0; /* they are the same */ | |
63 | ||
64 | if (diff_populate_filespec(src, 0) || diff_populate_filespec(dst, 0)) | |
65 | return 0; /* error but caught downstream */ | |
66 | ||
67 | max_size = ((src->size > dst->size) ? src->size : dst->size); | |
68 | if (max_size < MINIMUM_BREAK_SIZE) | |
69 | return 0; /* we do not break too small filepair */ | |
70 | ||
71 | if (diffcore_count_changes(src, dst, | |
72 | &src->cnt_data, &dst->cnt_data, | |
73 | 0, | |
74 | &src_copied, &literal_added)) | |
75 | return 0; | |
76 | ||
77 | /* sanity */ | |
78 | if (src->size < src_copied) | |
79 | src_copied = src->size; | |
80 | if (dst->size < literal_added + src_copied) { | |
81 | if (src_copied < dst->size) | |
82 | literal_added = dst->size - src_copied; | |
83 | else | |
84 | literal_added = 0; | |
85 | } | |
86 | src_removed = src->size - src_copied; | |
87 | ||
88 | /* Compute merge-score, which is "how much is removed | |
89 | * from the source material". The clean-up stage will | |
90 | * merge the surviving pair together if the score is | |
91 | * less than the minimum, after rename/copy runs. | |
92 | */ | |
93 | *merge_score_p = (int)(src_removed * MAX_SCORE / src->size); | |
94 | if (*merge_score_p > break_score) | |
95 | return 1; | |
96 | ||
97 | /* Extent of damage, which counts both inserts and | |
98 | * deletes. | |
99 | */ | |
100 | delta_size = src_removed + literal_added; | |
101 | if (delta_size * MAX_SCORE / max_size < break_score) | |
102 | return 0; | |
103 | ||
104 | /* If you removed a lot without adding new material, that is | |
105 | * not really a rewrite. | |
106 | */ | |
107 | if ((src->size * break_score < src_removed * MAX_SCORE) && | |
108 | (literal_added * 20 < src_removed) && | |
109 | (literal_added * 20 < src_copied)) | |
110 | return 0; | |
111 | ||
112 | return 1; | |
113 | } | |
114 | ||
115 | void diffcore_break(int break_score) | |
116 | { | |
117 | struct diff_queue_struct *q = &diff_queued_diff; | |
118 | struct diff_queue_struct outq; | |
119 | ||
120 | /* When the filepair has this much edit (insert and delete), | |
121 | * it is first considered to be a rewrite and broken into a | |
122 | * create and delete filepair. This is to help breaking a | |
123 | * file that had too much new stuff added, possibly from | |
124 | * moving contents from another file, so that rename/copy can | |
125 | * match it with the other file. | |
126 | * | |
127 | * int break_score; we reuse incoming parameter for this. | |
128 | */ | |
129 | ||
130 | /* After a pair is broken according to break_score and | |
131 | * subjected to rename/copy, both of them may survive intact, | |
132 | * due to lack of suitable rename/copy peer. Or, the caller | |
133 | * may be calling us without using rename/copy. When that | |
134 | * happens, we merge the broken pieces back into one | |
135 | * modification together if the pair did not have more than | |
136 | * this much delete. For this computation, we do not take | |
137 | * insert into account at all. If you start from a 100-line | |
138 | * file and delete 97 lines of it, it does not matter if you | |
139 | * add 27 lines to it to make a new 30-line file or if you add | |
140 | * 997 lines to it to make a 1000-line file. Either way what | |
141 | * you did was a rewrite of 97%. On the other hand, if you | |
142 | * delete 3 lines, keeping 97 lines intact, it does not matter | |
143 | * if you add 3 lines to it to make a new 100-line file or if | |
144 | * you add 903 lines to it to make a new 1000-line file. | |
145 | * Either way you did a lot of additions and not a rewrite. | |
146 | * This merge happens to catch the latter case. A merge_score | |
147 | * of 80% would be a good default value (a broken pair that | |
148 | * has score lower than merge_score will be merged back | |
149 | * together). | |
150 | */ | |
151 | int merge_score; | |
152 | int i; | |
153 | ||
154 | /* See comment on DEFAULT_BREAK_SCORE and | |
155 | * DEFAULT_MERGE_SCORE in diffcore.h | |
156 | */ | |
157 | merge_score = (break_score >> 16) & 0xFFFF; | |
158 | break_score = (break_score & 0xFFFF); | |
159 | ||
160 | if (!break_score) | |
161 | break_score = DEFAULT_BREAK_SCORE; | |
162 | if (!merge_score) | |
163 | merge_score = DEFAULT_MERGE_SCORE; | |
164 | ||
165 | DIFF_QUEUE_CLEAR(&outq); | |
166 | ||
167 | for (i = 0; i < q->nr; i++) { | |
168 | struct diff_filepair *p = q->queue[i]; | |
169 | int score; | |
170 | ||
171 | /* | |
172 | * We deal only with in-place edit of blobs. | |
173 | * We do not break anything else. | |
174 | */ | |
175 | if (DIFF_FILE_VALID(p->one) && DIFF_FILE_VALID(p->two) && | |
176 | object_type(p->one->mode) == OBJ_BLOB && | |
177 | object_type(p->two->mode) == OBJ_BLOB && | |
178 | !strcmp(p->one->path, p->two->path)) { | |
179 | if (should_break(p->one, p->two, | |
180 | break_score, &score)) { | |
181 | /* Split this into delete and create */ | |
182 | struct diff_filespec *null_one, *null_two; | |
183 | struct diff_filepair *dp; | |
184 | ||
185 | /* Set score to 0 for the pair that | |
186 | * needs to be merged back together | |
187 | * should they survive rename/copy. | |
188 | * Also we do not want to break very | |
189 | * small files. | |
190 | */ | |
191 | if (score < merge_score) | |
192 | score = 0; | |
193 | ||
194 | /* deletion of one */ | |
195 | null_one = alloc_filespec(p->one->path); | |
196 | dp = diff_queue(&outq, p->one, null_one); | |
197 | dp->score = score; | |
198 | dp->broken_pair = 1; | |
199 | ||
200 | /* creation of two */ | |
201 | null_two = alloc_filespec(p->two->path); | |
202 | dp = diff_queue(&outq, null_two, p->two); | |
203 | dp->score = score; | |
204 | dp->broken_pair = 1; | |
205 | ||
206 | diff_free_filespec_blob(p->one); | |
207 | diff_free_filespec_blob(p->two); | |
208 | free(p); /* not diff_free_filepair(), we are | |
209 | * reusing one and two here. | |
210 | */ | |
211 | continue; | |
212 | } | |
213 | } | |
214 | diff_free_filespec_data(p->one); | |
215 | diff_free_filespec_data(p->two); | |
216 | diff_q(&outq, p); | |
217 | } | |
218 | free(q->queue); | |
219 | *q = outq; | |
220 | ||
221 | return; | |
222 | } | |
223 | ||
224 | static void merge_broken(struct diff_filepair *p, | |
225 | struct diff_filepair *pp, | |
226 | struct diff_queue_struct *outq) | |
227 | { | |
228 | /* p and pp are broken pairs we want to merge */ | |
229 | struct diff_filepair *c = p, *d = pp, *dp; | |
230 | if (DIFF_FILE_VALID(p->one)) { | |
231 | /* this must be a delete half */ | |
232 | d = p; c = pp; | |
233 | } | |
234 | /* Sanity check */ | |
235 | if (!DIFF_FILE_VALID(d->one)) | |
236 | die("internal error in merge #1"); | |
237 | if (DIFF_FILE_VALID(d->two)) | |
238 | die("internal error in merge #2"); | |
239 | if (DIFF_FILE_VALID(c->one)) | |
240 | die("internal error in merge #3"); | |
241 | if (!DIFF_FILE_VALID(c->two)) | |
242 | die("internal error in merge #4"); | |
243 | ||
244 | dp = diff_queue(outq, d->one, c->two); | |
245 | dp->score = p->score; | |
246 | diff_free_filespec_data(d->two); | |
247 | diff_free_filespec_data(c->one); | |
248 | free(d); | |
249 | free(c); | |
250 | } | |
251 | ||
252 | void diffcore_merge_broken(void) | |
253 | { | |
254 | struct diff_queue_struct *q = &diff_queued_diff; | |
255 | struct diff_queue_struct outq; | |
256 | int i, j; | |
257 | ||
258 | DIFF_QUEUE_CLEAR(&outq); | |
259 | ||
260 | for (i = 0; i < q->nr; i++) { | |
261 | struct diff_filepair *p = q->queue[i]; | |
262 | if (!p) | |
263 | /* we already merged this with its peer */ | |
264 | continue; | |
265 | else if (p->broken_pair && | |
266 | !strcmp(p->one->path, p->two->path)) { | |
267 | /* If the peer also survived rename/copy, then | |
268 | * we merge them back together. | |
269 | */ | |
270 | for (j = i + 1; j < q->nr; j++) { | |
271 | struct diff_filepair *pp = q->queue[j]; | |
272 | if (pp->broken_pair && | |
273 | !strcmp(pp->one->path, pp->two->path) && | |
274 | !strcmp(p->one->path, pp->two->path)) { | |
275 | /* Peer survived. Merge them */ | |
276 | merge_broken(p, pp, &outq); | |
277 | q->queue[j] = NULL; | |
278 | break; | |
279 | } | |
280 | } | |
281 | if (q->nr <= j) | |
282 | /* The peer did not survive, so we keep | |
283 | * it in the output. | |
284 | */ | |
285 | diff_q(&outq, p); | |
286 | } | |
287 | else | |
288 | diff_q(&outq, p); | |
289 | } | |
290 | free(q->queue); | |
291 | *q = outq; | |
292 | ||
293 | return; | |
294 | } |