packv4: nth_packed_object_sha1() packv4 support
[git/packv4.git] / diffcore-break.c
blobae8a7d03e2d103739897ac37fe1f9966956b9ec8
1 /*
2 * Copyright (C) 2005 Junio C Hamano
3 */
4 #include "cache.h"
5 #include "diff.h"
6 #include "diffcore.h"
8 static int should_break(struct diff_filespec *src,
9 struct diff_filespec *dst,
10 int break_score,
11 int *merge_score_p)
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?
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.
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.
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.
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.
45 * The value we return is 1 if we want the pair to be broken,
46 * or 0 if we do not.
48 unsigned long delta_size, base_size, src_copied, literal_added,
49 src_removed;
51 *merge_score_p = 0; /* assume no deletion --- "do not break"
52 * is the default.
55 if (!S_ISREG(src->mode) || !S_ISREG(dst->mode))
56 return 0; /* leave symlink rename alone */
58 if (src->sha1_valid && dst->sha1_valid &&
59 !hashcmp(src->sha1, dst->sha1))
60 return 0; /* they are the same */
62 if (diff_populate_filespec(src, 0) || diff_populate_filespec(dst, 0))
63 return 0; /* error but caught downstream */
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 */
69 if (diffcore_count_changes(src, dst,
70 NULL, NULL,
72 &src_copied, &literal_added))
73 return 0;
75 /* sanity */
76 if (src->size < src_copied)
77 src_copied = src->size;
78 if (dst->size < literal_added + src_copied) {
79 if (src_copied < dst->size)
80 literal_added = dst->size - src_copied;
81 else
82 literal_added = 0;
84 src_removed = src->size - src_copied;
86 /* Compute merge-score, which is "how much is removed
87 * from the source material". The clean-up stage will
88 * merge the surviving pair together if the score is
89 * less than the minimum, after rename/copy runs.
91 *merge_score_p = (int)(src_removed * MAX_SCORE / src->size);
93 /* Extent of damage, which counts both inserts and
94 * deletes.
96 delta_size = src_removed + literal_added;
97 if (delta_size * MAX_SCORE / base_size < break_score)
98 return 0;
100 /* If you removed a lot without adding new material, that is
101 * not really a rewrite.
103 if ((src->size * break_score < src_removed * MAX_SCORE) &&
104 (literal_added * 20 < src_removed) &&
105 (literal_added * 20 < src_copied))
106 return 0;
108 return 1;
111 void diffcore_break(int break_score)
113 struct diff_queue_struct *q = &diff_queued_diff;
114 struct diff_queue_struct outq;
116 /* When the filepair has this much edit (insert and delete),
117 * it is first considered to be a rewrite and broken into a
118 * create and delete filepair. This is to help breaking a
119 * file that had too much new stuff added, possibly from
120 * moving contents from another file, so that rename/copy can
121 * match it with the other file.
123 * int break_score; we reuse incoming parameter for this.
126 /* After a pair is broken according to break_score and
127 * subjected to rename/copy, both of them may survive intact,
128 * due to lack of suitable rename/copy peer. Or, the caller
129 * may be calling us without using rename/copy. When that
130 * happens, we merge the broken pieces back into one
131 * modification together if the pair did not have more than
132 * this much delete. For this computation, we do not take
133 * insert into account at all. If you start from a 100-line
134 * file and delete 97 lines of it, it does not matter if you
135 * add 27 lines to it to make a new 30-line file or if you add
136 * 997 lines to it to make a 1000-line file. Either way what
137 * you did was a rewrite of 97%. On the other hand, if you
138 * delete 3 lines, keeping 97 lines intact, it does not matter
139 * if you add 3 lines to it to make a new 100-line file or if
140 * you add 903 lines to it to make a new 1000-line file.
141 * Either way you did a lot of additions and not a rewrite.
142 * This merge happens to catch the latter case. A merge_score
143 * of 80% would be a good default value (a broken pair that
144 * has score lower than merge_score will be merged back
145 * together).
147 int merge_score;
148 int i;
150 /* See comment on DEFAULT_BREAK_SCORE and
151 * DEFAULT_MERGE_SCORE in diffcore.h
153 merge_score = (break_score >> 16) & 0xFFFF;
154 break_score = (break_score & 0xFFFF);
156 if (!break_score)
157 break_score = DEFAULT_BREAK_SCORE;
158 if (!merge_score)
159 merge_score = DEFAULT_MERGE_SCORE;
161 outq.nr = outq.alloc = 0;
162 outq.queue = NULL;
164 for (i = 0; i < q->nr; i++) {
165 struct diff_filepair *p = q->queue[i];
166 int score;
168 /* We deal only with in-place edit of non directory.
169 * We do not break anything else.
171 if (DIFF_FILE_VALID(p->one) && DIFF_FILE_VALID(p->two) &&
172 !S_ISDIR(p->one->mode) && !S_ISDIR(p->two->mode) &&
173 !strcmp(p->one->path, p->two->path)) {
174 if (should_break(p->one, p->two,
175 break_score, &score)) {
176 /* Split this into delete and create */
177 struct diff_filespec *null_one, *null_two;
178 struct diff_filepair *dp;
180 /* Set score to 0 for the pair that
181 * needs to be merged back together
182 * should they survive rename/copy.
183 * Also we do not want to break very
184 * small files.
186 if (score < merge_score)
187 score = 0;
189 /* deletion of one */
190 null_one = alloc_filespec(p->one->path);
191 dp = diff_queue(&outq, p->one, null_one);
192 dp->score = score;
193 dp->broken_pair = 1;
195 /* creation of two */
196 null_two = alloc_filespec(p->two->path);
197 dp = diff_queue(&outq, null_two, p->two);
198 dp->score = score;
199 dp->broken_pair = 1;
201 free(p); /* not diff_free_filepair(), we are
202 * reusing one and two here.
204 continue;
207 diff_q(&outq, p);
209 free(q->queue);
210 *q = outq;
212 return;
215 static void merge_broken(struct diff_filepair *p,
216 struct diff_filepair *pp,
217 struct diff_queue_struct *outq)
219 /* p and pp are broken pairs we want to merge */
220 struct diff_filepair *c = p, *d = pp, *dp;
221 if (DIFF_FILE_VALID(p->one)) {
222 /* this must be a delete half */
223 d = p; c = pp;
225 /* Sanity check */
226 if (!DIFF_FILE_VALID(d->one))
227 die("internal error in merge #1");
228 if (DIFF_FILE_VALID(d->two))
229 die("internal error in merge #2");
230 if (DIFF_FILE_VALID(c->one))
231 die("internal error in merge #3");
232 if (!DIFF_FILE_VALID(c->two))
233 die("internal error in merge #4");
235 dp = diff_queue(outq, d->one, c->two);
236 dp->score = p->score;
237 diff_free_filespec_data(d->two);
238 diff_free_filespec_data(c->one);
239 free(d);
240 free(c);
243 void diffcore_merge_broken(void)
245 struct diff_queue_struct *q = &diff_queued_diff;
246 struct diff_queue_struct outq;
247 int i, j;
249 outq.nr = outq.alloc = 0;
250 outq.queue = NULL;
252 for (i = 0; i < q->nr; i++) {
253 struct diff_filepair *p = q->queue[i];
254 if (!p)
255 /* we already merged this with its peer */
256 continue;
257 else if (p->broken_pair &&
258 !strcmp(p->one->path, p->two->path)) {
259 /* If the peer also survived rename/copy, then
260 * we merge them back together.
262 for (j = i + 1; j < q->nr; j++) {
263 struct diff_filepair *pp = q->queue[j];
264 if (pp->broken_pair &&
265 !strcmp(pp->one->path, pp->two->path) &&
266 !strcmp(p->one->path, pp->two->path)) {
267 /* Peer survived. Merge them */
268 merge_broken(p, pp, &outq);
269 q->queue[j] = NULL;
270 break;
273 if (q->nr <= j)
274 /* The peer did not survive, so we keep
275 * it in the output.
277 diff_q(&outq, p);
279 else
280 diff_q(&outq, p);
282 free(q->queue);
283 *q = outq;
285 return;