]>
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 | #include "promisor-remote.h" | |
8 | ||
9 | static int should_break(struct repository *r, | |
10 | 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 | unsigned long delta_size, max_size; | |
51 | unsigned long src_copied, literal_added, src_removed; | |
52 | ||
53 | struct diff_populate_filespec_options options = { 0 }; | |
54 | ||
55 | *merge_score_p = 0; /* assume no deletion --- "do not break" | |
56 | * is the default. | |
57 | */ | |
58 | ||
59 | if (S_ISREG(src->mode) != S_ISREG(dst->mode)) { | |
60 | *merge_score_p = (int)MAX_SCORE; | |
61 | return 1; /* even their types are different */ | |
62 | } | |
63 | ||
64 | if (src->oid_valid && dst->oid_valid && | |
65 | oideq(&src->oid, &dst->oid)) | |
66 | return 0; /* they are the same */ | |
67 | ||
68 | if (r == the_repository && has_promisor_remote()) { | |
69 | options.missing_object_cb = diff_queued_diff_prefetch; | |
70 | options.missing_object_data = r; | |
71 | } | |
72 | ||
73 | if (diff_populate_filespec(r, src, &options) || | |
74 | diff_populate_filespec(r, dst, &options)) | |
75 | return 0; /* error but caught downstream */ | |
76 | ||
77 | max_size = ((src->size > dst->size) ? src->size : dst->size); | |
78 | if (max_size < MINIMUM_BREAK_SIZE) | |
79 | return 0; /* we do not break too small filepair */ | |
80 | ||
81 | if (!src->size) | |
82 | return 0; /* we do not let empty files get renamed */ | |
83 | ||
84 | if (diffcore_count_changes(r, src, dst, | |
85 | &src->cnt_data, &dst->cnt_data, | |
86 | &src_copied, &literal_added)) | |
87 | return 0; | |
88 | ||
89 | /* sanity */ | |
90 | if (src->size < src_copied) | |
91 | src_copied = src->size; | |
92 | if (dst->size < literal_added + src_copied) { | |
93 | if (src_copied < dst->size) | |
94 | literal_added = dst->size - src_copied; | |
95 | else | |
96 | literal_added = 0; | |
97 | } | |
98 | src_removed = src->size - src_copied; | |
99 | ||
100 | /* Compute merge-score, which is "how much is removed | |
101 | * from the source material". The clean-up stage will | |
102 | * merge the surviving pair together if the score is | |
103 | * less than the minimum, after rename/copy runs. | |
104 | */ | |
105 | *merge_score_p = (int)(src_removed * MAX_SCORE / src->size); | |
106 | if (*merge_score_p > break_score) | |
107 | return 1; | |
108 | ||
109 | /* Extent of damage, which counts both inserts and | |
110 | * deletes. | |
111 | */ | |
112 | delta_size = src_removed + literal_added; | |
113 | if (delta_size * MAX_SCORE / max_size < break_score) | |
114 | return 0; | |
115 | ||
116 | /* If you removed a lot without adding new material, that is | |
117 | * not really a rewrite. | |
118 | */ | |
119 | if ((src->size * break_score < src_removed * MAX_SCORE) && | |
120 | (literal_added * 20 < src_removed) && | |
121 | (literal_added * 20 < src_copied)) | |
122 | return 0; | |
123 | ||
124 | return 1; | |
125 | } | |
126 | ||
127 | void diffcore_break(struct repository *r, int break_score) | |
128 | { | |
129 | struct diff_queue_struct *q = &diff_queued_diff; | |
130 | struct diff_queue_struct outq; | |
131 | ||
132 | /* When the filepair has this much edit (insert and delete), | |
133 | * it is first considered to be a rewrite and broken into a | |
134 | * create and delete filepair. This is to help breaking a | |
135 | * file that had too much new stuff added, possibly from | |
136 | * moving contents from another file, so that rename/copy can | |
137 | * match it with the other file. | |
138 | * | |
139 | * int break_score; we reuse incoming parameter for this. | |
140 | */ | |
141 | ||
142 | /* After a pair is broken according to break_score and | |
143 | * subjected to rename/copy, both of them may survive intact, | |
144 | * due to lack of suitable rename/copy peer. Or, the caller | |
145 | * may be calling us without using rename/copy. When that | |
146 | * happens, we merge the broken pieces back into one | |
147 | * modification together if the pair did not have more than | |
148 | * this much delete. For this computation, we do not take | |
149 | * insert into account at all. If you start from a 100-line | |
150 | * file and delete 97 lines of it, it does not matter if you | |
151 | * add 27 lines to it to make a new 30-line file or if you add | |
152 | * 997 lines to it to make a 1000-line file. Either way what | |
153 | * you did was a rewrite of 97%. On the other hand, if you | |
154 | * delete 3 lines, keeping 97 lines intact, it does not matter | |
155 | * if you add 3 lines to it to make a new 100-line file or if | |
156 | * you add 903 lines to it to make a new 1000-line file. | |
157 | * Either way you did a lot of additions and not a rewrite. | |
158 | * This merge happens to catch the latter case. A merge_score | |
159 | * of 80% would be a good default value (a broken pair that | |
160 | * has score lower than merge_score will be merged back | |
161 | * together). | |
162 | */ | |
163 | int merge_score; | |
164 | int i; | |
165 | ||
166 | /* See comment on DEFAULT_BREAK_SCORE and | |
167 | * DEFAULT_MERGE_SCORE in diffcore.h | |
168 | */ | |
169 | merge_score = (break_score >> 16) & 0xFFFF; | |
170 | break_score = (break_score & 0xFFFF); | |
171 | ||
172 | if (!break_score) | |
173 | break_score = DEFAULT_BREAK_SCORE; | |
174 | if (!merge_score) | |
175 | merge_score = DEFAULT_MERGE_SCORE; | |
176 | ||
177 | DIFF_QUEUE_CLEAR(&outq); | |
178 | ||
179 | for (i = 0; i < q->nr; i++) { | |
180 | struct diff_filepair *p = q->queue[i]; | |
181 | int score; | |
182 | ||
183 | /* | |
184 | * We deal only with in-place edit of blobs. | |
185 | * We do not break anything else. | |
186 | */ | |
187 | if (DIFF_FILE_VALID(p->one) && DIFF_FILE_VALID(p->two) && | |
188 | object_type(p->one->mode) == OBJ_BLOB && | |
189 | object_type(p->two->mode) == OBJ_BLOB && | |
190 | !strcmp(p->one->path, p->two->path)) { | |
191 | if (should_break(r, p->one, p->two, | |
192 | break_score, &score)) { | |
193 | /* Split this into delete and create */ | |
194 | struct diff_filespec *null_one, *null_two; | |
195 | struct diff_filepair *dp; | |
196 | ||
197 | /* Set score to 0 for the pair that | |
198 | * needs to be merged back together | |
199 | * should they survive rename/copy. | |
200 | * Also we do not want to break very | |
201 | * small files. | |
202 | */ | |
203 | if (score < merge_score) | |
204 | score = 0; | |
205 | ||
206 | /* deletion of one */ | |
207 | null_one = alloc_filespec(p->one->path); | |
208 | dp = diff_queue(&outq, p->one, null_one); | |
209 | dp->score = score; | |
210 | dp->broken_pair = 1; | |
211 | ||
212 | /* creation of two */ | |
213 | null_two = alloc_filespec(p->two->path); | |
214 | dp = diff_queue(&outq, null_two, p->two); | |
215 | dp->score = score; | |
216 | dp->broken_pair = 1; | |
217 | ||
218 | diff_free_filespec_blob(p->one); | |
219 | diff_free_filespec_blob(p->two); | |
220 | free(p); /* not diff_free_filepair(), we are | |
221 | * reusing one and two here. | |
222 | */ | |
223 | continue; | |
224 | } | |
225 | } | |
226 | diff_free_filespec_data(p->one); | |
227 | diff_free_filespec_data(p->two); | |
228 | diff_q(&outq, p); | |
229 | } | |
230 | free(q->queue); | |
231 | *q = outq; | |
232 | ||
233 | return; | |
234 | } | |
235 | ||
236 | static void merge_broken(struct diff_filepair *p, | |
237 | struct diff_filepair *pp, | |
238 | struct diff_queue_struct *outq) | |
239 | { | |
240 | /* p and pp are broken pairs we want to merge */ | |
241 | struct diff_filepair *c = p, *d = pp, *dp; | |
242 | if (DIFF_FILE_VALID(p->one)) { | |
243 | /* this must be a delete half */ | |
244 | d = p; c = pp; | |
245 | } | |
246 | /* Sanity check */ | |
247 | if (!DIFF_FILE_VALID(d->one)) | |
248 | die("internal error in merge #1"); | |
249 | if (DIFF_FILE_VALID(d->two)) | |
250 | die("internal error in merge #2"); | |
251 | if (DIFF_FILE_VALID(c->one)) | |
252 | die("internal error in merge #3"); | |
253 | if (!DIFF_FILE_VALID(c->two)) | |
254 | die("internal error in merge #4"); | |
255 | ||
256 | dp = diff_queue(outq, d->one, c->two); | |
257 | dp->score = p->score; | |
258 | /* | |
259 | * We will be one extra user of the same src side of the | |
260 | * broken pair, if it was used as the rename source for other | |
261 | * paths elsewhere. Increment to mark that the path stays | |
262 | * in the resulting tree. | |
263 | */ | |
264 | d->one->rename_used++; | |
265 | diff_free_filespec_data(d->two); | |
266 | diff_free_filespec_data(c->one); | |
267 | free(d); | |
268 | free(c); | |
269 | } | |
270 | ||
271 | void diffcore_merge_broken(void) | |
272 | { | |
273 | struct diff_queue_struct *q = &diff_queued_diff; | |
274 | struct diff_queue_struct outq; | |
275 | int i, j; | |
276 | ||
277 | DIFF_QUEUE_CLEAR(&outq); | |
278 | ||
279 | for (i = 0; i < q->nr; i++) { | |
280 | struct diff_filepair *p = q->queue[i]; | |
281 | if (!p) | |
282 | /* we already merged this with its peer */ | |
283 | continue; | |
284 | else if (p->broken_pair && | |
285 | !strcmp(p->one->path, p->two->path)) { | |
286 | /* If the peer also survived rename/copy, then | |
287 | * we merge them back together. | |
288 | */ | |
289 | for (j = i + 1; j < q->nr; j++) { | |
290 | struct diff_filepair *pp = q->queue[j]; | |
291 | if (pp->broken_pair && | |
292 | !strcmp(pp->one->path, pp->two->path) && | |
293 | !strcmp(p->one->path, pp->two->path)) { | |
294 | /* Peer survived. Merge them */ | |
295 | merge_broken(p, pp, &outq); | |
296 | q->queue[j] = NULL; | |
297 | goto next; | |
298 | } | |
299 | } | |
300 | /* The peer did not survive, so we keep | |
301 | * it in the output. | |
302 | */ | |
303 | diff_q(&outq, p); | |
304 | } | |
305 | else | |
306 | diff_q(&outq, p); | |
307 | next:; | |
308 | } | |
309 | free(q->queue); | |
310 | *q = outq; | |
311 | ||
312 | return; | |
313 | } |