diffcore-rename: somewhat optimized.
[git/git.git] / diffcore-break.c
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, base_size, src_copied, literal_added,
49 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 return 0; /* leave symlink rename alone */
57
58 if (src->sha1_valid && dst->sha1_valid &&
59 !memcmp(src->sha1, dst->sha1, 20))
60 return 0; /* they are the same */
61
62 if (diff_populate_filespec(src, 0) || diff_populate_filespec(dst, 0))
63 return 0; /* error but caught downstream */
64
65 base_size = ((src->size < dst->size) ? src->size : dst->size);
66 if (base_size < MINIMUM_BREAK_SIZE)
67 return 0; /* we do not break too small filepair */
68
69 if (diffcore_count_changes(src->data, src->size,
70 dst->data, dst->size,
71 NULL, NULL,
72 0,
73 &src_copied, &literal_added))
74 return 0;
75
76 /* sanity */
77 if (src->size < src_copied)
78 src_copied = src->size;
79 if (dst->size < literal_added + src_copied) {
80 if (src_copied < dst->size)
81 literal_added = dst->size - src_copied;
82 else
83 literal_added = 0;
84 }
85 src_removed = src->size - src_copied;
86
87 /* Compute merge-score, which is "how much is removed
88 * from the source material". The clean-up stage will
89 * merge the surviving pair together if the score is
90 * less than the minimum, after rename/copy runs.
91 */
92 *merge_score_p = src_removed * MAX_SCORE / src->size;
93
94 /* Extent of damage, which counts both inserts and
95 * deletes.
96 */
97 delta_size = src_removed + literal_added;
98 if (delta_size * MAX_SCORE / base_size < break_score)
99 return 0;
100
101 /* If you removed a lot without adding new material, that is
102 * not really a rewrite.
103 */
104 if ((src->size * break_score < src_removed * MAX_SCORE) &&
105 (literal_added * 20 < src_removed) &&
106 (literal_added * 20 < src_copied))
107 return 0;
108
109 return 1;
110 }
111
112 void diffcore_break(int break_score)
113 {
114 struct diff_queue_struct *q = &diff_queued_diff;
115 struct diff_queue_struct outq;
116
117 /* When the filepair has this much edit (insert and delete),
118 * it is first considered to be a rewrite and broken into a
119 * create and delete filepair. This is to help breaking a
120 * file that had too much new stuff added, possibly from
121 * moving contents from another file, so that rename/copy can
122 * match it with the other file.
123 *
124 * int break_score; we reuse incoming parameter for this.
125 */
126
127 /* After a pair is broken according to break_score and
128 * subjected to rename/copy, both of them may survive intact,
129 * due to lack of suitable rename/copy peer. Or, the caller
130 * may be calling us without using rename/copy. When that
131 * happens, we merge the broken pieces back into one
132 * modification together if the pair did not have more than
133 * this much delete. For this computation, we do not take
134 * insert into account at all. If you start from a 100-line
135 * file and delete 97 lines of it, it does not matter if you
136 * add 27 lines to it to make a new 30-line file or if you add
137 * 997 lines to it to make a 1000-line file. Either way what
138 * you did was a rewrite of 97%. On the other hand, if you
139 * delete 3 lines, keeping 97 lines intact, it does not matter
140 * if you add 3 lines to it to make a new 100-line file or if
141 * you add 903 lines to it to make a new 1000-line file.
142 * Either way you did a lot of additions and not a rewrite.
143 * This merge happens to catch the latter case. A merge_score
144 * of 80% would be a good default value (a broken pair that
145 * has score lower than merge_score will be merged back
146 * together).
147 */
148 int merge_score;
149 int i;
150
151 /* See comment on DEFAULT_BREAK_SCORE and
152 * DEFAULT_MERGE_SCORE in diffcore.h
153 */
154 merge_score = (break_score >> 16) & 0xFFFF;
155 break_score = (break_score & 0xFFFF);
156
157 if (!break_score)
158 break_score = DEFAULT_BREAK_SCORE;
159 if (!merge_score)
160 merge_score = DEFAULT_MERGE_SCORE;
161
162 outq.nr = outq.alloc = 0;
163 outq.queue = NULL;
164
165 for (i = 0; i < q->nr; i++) {
166 struct diff_filepair *p = q->queue[i];
167 int score;
168
169 /* We deal only with in-place edit of non directory.
170 * We do not break anything else.
171 */
172 if (DIFF_FILE_VALID(p->one) && DIFF_FILE_VALID(p->two) &&
173 !S_ISDIR(p->one->mode) && !S_ISDIR(p->two->mode) &&
174 !strcmp(p->one->path, p->two->path)) {
175 if (should_break(p->one, p->two,
176 break_score, &score)) {
177 /* Split this into delete and create */
178 struct diff_filespec *null_one, *null_two;
179 struct diff_filepair *dp;
180
181 /* Set score to 0 for the pair that
182 * needs to be merged back together
183 * should they survive rename/copy.
184 * Also we do not want to break very
185 * small files.
186 */
187 if (score < merge_score)
188 score = 0;
189
190 /* deletion of one */
191 null_one = alloc_filespec(p->one->path);
192 dp = diff_queue(&outq, p->one, null_one);
193 dp->score = score;
194 dp->broken_pair = 1;
195
196 /* creation of two */
197 null_two = alloc_filespec(p->two->path);
198 dp = diff_queue(&outq, null_two, p->two);
199 dp->score = score;
200 dp->broken_pair = 1;
201
202 free(p); /* not diff_free_filepair(), we are
203 * reusing one and two here.
204 */
205 continue;
206 }
207 }
208 diff_q(&outq, p);
209 }
210 free(q->queue);
211 *q = outq;
212
213 return;
214 }
215
216 static void merge_broken(struct diff_filepair *p,
217 struct diff_filepair *pp,
218 struct diff_queue_struct *outq)
219 {
220 /* p and pp are broken pairs we want to merge */
221 struct diff_filepair *c = p, *d = pp, *dp;
222 if (DIFF_FILE_VALID(p->one)) {
223 /* this must be a delete half */
224 d = p; c = pp;
225 }
226 /* Sanity check */
227 if (!DIFF_FILE_VALID(d->one))
228 die("internal error in merge #1");
229 if (DIFF_FILE_VALID(d->two))
230 die("internal error in merge #2");
231 if (DIFF_FILE_VALID(c->one))
232 die("internal error in merge #3");
233 if (!DIFF_FILE_VALID(c->two))
234 die("internal error in merge #4");
235
236 dp = diff_queue(outq, d->one, c->two);
237 dp->score = p->score;
238 diff_free_filespec_data(d->two);
239 diff_free_filespec_data(c->one);
240 free(d);
241 free(c);
242 }
243
244 void diffcore_merge_broken(void)
245 {
246 struct diff_queue_struct *q = &diff_queued_diff;
247 struct diff_queue_struct outq;
248 int i, j;
249
250 outq.nr = outq.alloc = 0;
251 outq.queue = NULL;
252
253 for (i = 0; i < q->nr; i++) {
254 struct diff_filepair *p = q->queue[i];
255 if (!p)
256 /* we already merged this with its peer */
257 continue;
258 else if (p->broken_pair &&
259 !strcmp(p->one->path, p->two->path)) {
260 /* If the peer also survived rename/copy, then
261 * we merge them back together.
262 */
263 for (j = i + 1; j < q->nr; j++) {
264 struct diff_filepair *pp = q->queue[j];
265 if (pp->broken_pair &&
266 !strcmp(pp->one->path, pp->two->path) &&
267 !strcmp(p->one->path, pp->two->path)) {
268 /* Peer survived. Merge them */
269 merge_broken(p, pp, &outq);
270 q->queue[j] = NULL;
271 break;
272 }
273 }
274 if (q->nr <= j)
275 /* The peer did not survive, so we keep
276 * it in the output.
277 */
278 diff_q(&outq, p);
279 }
280 else
281 diff_q(&outq, p);
282 }
283 free(q->queue);
284 *q = outq;
285
286 return;
287 }