]>
Commit | Line | Data |
---|---|---|
a2ad79ce CC |
1 | #include "cache.h" |
2 | #include "commit.h" | |
3 | #include "diff.h" | |
4 | #include "revision.h" | |
4eb5b646 | 5 | #include "sha1-lookup.h" |
a2ad79ce CC |
6 | #include "bisect.h" |
7 | ||
95188648 CC |
8 | static unsigned char (*skipped_sha1)[20]; |
9 | static int skipped_sha1_nr; | |
10 | ||
a2ad79ce CC |
11 | /* bits #0-15 in revision.h */ |
12 | ||
13 | #define COUNTED (1u<<16) | |
14 | ||
15 | /* | |
16 | * This is a truly stupid algorithm, but it's only | |
17 | * used for bisection, and we just don't care enough. | |
18 | * | |
19 | * We care just barely enough to avoid recursing for | |
20 | * non-merge entries. | |
21 | */ | |
22 | static int count_distance(struct commit_list *entry) | |
23 | { | |
24 | int nr = 0; | |
25 | ||
26 | while (entry) { | |
27 | struct commit *commit = entry->item; | |
28 | struct commit_list *p; | |
29 | ||
30 | if (commit->object.flags & (UNINTERESTING | COUNTED)) | |
31 | break; | |
32 | if (!(commit->object.flags & TREESAME)) | |
33 | nr++; | |
34 | commit->object.flags |= COUNTED; | |
35 | p = commit->parents; | |
36 | entry = p; | |
37 | if (p) { | |
38 | p = p->next; | |
39 | while (p) { | |
40 | nr += count_distance(p); | |
41 | p = p->next; | |
42 | } | |
43 | } | |
44 | } | |
45 | ||
46 | return nr; | |
47 | } | |
48 | ||
49 | static void clear_distance(struct commit_list *list) | |
50 | { | |
51 | while (list) { | |
52 | struct commit *commit = list->item; | |
53 | commit->object.flags &= ~COUNTED; | |
54 | list = list->next; | |
55 | } | |
56 | } | |
57 | ||
58 | #define DEBUG_BISECT 0 | |
59 | ||
60 | static inline int weight(struct commit_list *elem) | |
61 | { | |
62 | return *((int*)(elem->item->util)); | |
63 | } | |
64 | ||
65 | static inline void weight_set(struct commit_list *elem, int weight) | |
66 | { | |
67 | *((int*)(elem->item->util)) = weight; | |
68 | } | |
69 | ||
70 | static int count_interesting_parents(struct commit *commit) | |
71 | { | |
72 | struct commit_list *p; | |
73 | int count; | |
74 | ||
75 | for (count = 0, p = commit->parents; p; p = p->next) { | |
76 | if (p->item->object.flags & UNINTERESTING) | |
77 | continue; | |
78 | count++; | |
79 | } | |
80 | return count; | |
81 | } | |
82 | ||
83 | static inline int halfway(struct commit_list *p, int nr) | |
84 | { | |
85 | /* | |
86 | * Don't short-cut something we are not going to return! | |
87 | */ | |
88 | if (p->item->object.flags & TREESAME) | |
89 | return 0; | |
90 | if (DEBUG_BISECT) | |
91 | return 0; | |
92 | /* | |
93 | * 2 and 3 are halfway of 5. | |
94 | * 3 is halfway of 6 but 2 and 4 are not. | |
95 | */ | |
96 | switch (2 * weight(p) - nr) { | |
97 | case -1: case 0: case 1: | |
98 | return 1; | |
99 | default: | |
100 | return 0; | |
101 | } | |
102 | } | |
103 | ||
104 | #if !DEBUG_BISECT | |
105 | #define show_list(a,b,c,d) do { ; } while (0) | |
106 | #else | |
107 | static void show_list(const char *debug, int counted, int nr, | |
108 | struct commit_list *list) | |
109 | { | |
110 | struct commit_list *p; | |
111 | ||
112 | fprintf(stderr, "%s (%d/%d)\n", debug, counted, nr); | |
113 | ||
114 | for (p = list; p; p = p->next) { | |
115 | struct commit_list *pp; | |
116 | struct commit *commit = p->item; | |
117 | unsigned flags = commit->object.flags; | |
118 | enum object_type type; | |
119 | unsigned long size; | |
120 | char *buf = read_sha1_file(commit->object.sha1, &type, &size); | |
121 | char *ep, *sp; | |
122 | ||
123 | fprintf(stderr, "%c%c%c ", | |
124 | (flags & TREESAME) ? ' ' : 'T', | |
125 | (flags & UNINTERESTING) ? 'U' : ' ', | |
126 | (flags & COUNTED) ? 'C' : ' '); | |
127 | if (commit->util) | |
128 | fprintf(stderr, "%3d", weight(p)); | |
129 | else | |
130 | fprintf(stderr, "---"); | |
131 | fprintf(stderr, " %.*s", 8, sha1_to_hex(commit->object.sha1)); | |
132 | for (pp = commit->parents; pp; pp = pp->next) | |
133 | fprintf(stderr, " %.*s", 8, | |
134 | sha1_to_hex(pp->item->object.sha1)); | |
135 | ||
136 | sp = strstr(buf, "\n\n"); | |
137 | if (sp) { | |
138 | sp += 2; | |
139 | for (ep = sp; *ep && *ep != '\n'; ep++) | |
140 | ; | |
141 | fprintf(stderr, " %.*s", (int)(ep - sp), sp); | |
142 | } | |
143 | fprintf(stderr, "\n"); | |
144 | } | |
145 | } | |
146 | #endif /* DEBUG_BISECT */ | |
147 | ||
148 | static struct commit_list *best_bisection(struct commit_list *list, int nr) | |
149 | { | |
150 | struct commit_list *p, *best; | |
151 | int best_distance = -1; | |
152 | ||
153 | best = list; | |
154 | for (p = list; p; p = p->next) { | |
155 | int distance; | |
156 | unsigned flags = p->item->object.flags; | |
157 | ||
158 | if (flags & TREESAME) | |
159 | continue; | |
160 | distance = weight(p); | |
161 | if (nr - distance < distance) | |
162 | distance = nr - distance; | |
163 | if (distance > best_distance) { | |
164 | best = p; | |
165 | best_distance = distance; | |
166 | } | |
167 | } | |
168 | ||
169 | return best; | |
170 | } | |
171 | ||
172 | struct commit_dist { | |
173 | struct commit *commit; | |
174 | int distance; | |
175 | }; | |
176 | ||
177 | static int compare_commit_dist(const void *a_, const void *b_) | |
178 | { | |
179 | struct commit_dist *a, *b; | |
180 | ||
181 | a = (struct commit_dist *)a_; | |
182 | b = (struct commit_dist *)b_; | |
183 | if (a->distance != b->distance) | |
184 | return b->distance - a->distance; /* desc sort */ | |
185 | return hashcmp(a->commit->object.sha1, b->commit->object.sha1); | |
186 | } | |
187 | ||
188 | static struct commit_list *best_bisection_sorted(struct commit_list *list, int nr) | |
189 | { | |
190 | struct commit_list *p; | |
191 | struct commit_dist *array = xcalloc(nr, sizeof(*array)); | |
192 | int cnt, i; | |
193 | ||
194 | for (p = list, cnt = 0; p; p = p->next) { | |
195 | int distance; | |
196 | unsigned flags = p->item->object.flags; | |
197 | ||
198 | if (flags & TREESAME) | |
199 | continue; | |
200 | distance = weight(p); | |
201 | if (nr - distance < distance) | |
202 | distance = nr - distance; | |
203 | array[cnt].commit = p->item; | |
204 | array[cnt].distance = distance; | |
205 | cnt++; | |
206 | } | |
207 | qsort(array, cnt, sizeof(*array), compare_commit_dist); | |
208 | for (p = list, i = 0; i < cnt; i++) { | |
209 | struct name_decoration *r = xmalloc(sizeof(*r) + 100); | |
210 | struct object *obj = &(array[i].commit->object); | |
211 | ||
212 | sprintf(r->name, "dist=%d", array[i].distance); | |
213 | r->next = add_decoration(&name_decoration, obj, r); | |
214 | p->item = array[i].commit; | |
215 | p = p->next; | |
216 | } | |
217 | if (p) | |
218 | p->next = NULL; | |
219 | free(array); | |
220 | return list; | |
221 | } | |
222 | ||
223 | /* | |
224 | * zero or positive weight is the number of interesting commits it can | |
225 | * reach, including itself. Especially, weight = 0 means it does not | |
226 | * reach any tree-changing commits (e.g. just above uninteresting one | |
227 | * but traversal is with pathspec). | |
228 | * | |
229 | * weight = -1 means it has one parent and its distance is yet to | |
230 | * be computed. | |
231 | * | |
232 | * weight = -2 means it has more than one parent and its distance is | |
233 | * unknown. After running count_distance() first, they will get zero | |
234 | * or positive distance. | |
235 | */ | |
236 | static struct commit_list *do_find_bisection(struct commit_list *list, | |
237 | int nr, int *weights, | |
238 | int find_all) | |
239 | { | |
240 | int n, counted; | |
241 | struct commit_list *p; | |
242 | ||
243 | counted = 0; | |
244 | ||
245 | for (n = 0, p = list; p; p = p->next) { | |
246 | struct commit *commit = p->item; | |
247 | unsigned flags = commit->object.flags; | |
248 | ||
249 | p->item->util = &weights[n++]; | |
250 | switch (count_interesting_parents(commit)) { | |
251 | case 0: | |
252 | if (!(flags & TREESAME)) { | |
253 | weight_set(p, 1); | |
254 | counted++; | |
255 | show_list("bisection 2 count one", | |
256 | counted, nr, list); | |
257 | } | |
258 | /* | |
259 | * otherwise, it is known not to reach any | |
260 | * tree-changing commit and gets weight 0. | |
261 | */ | |
262 | break; | |
263 | case 1: | |
264 | weight_set(p, -1); | |
265 | break; | |
266 | default: | |
267 | weight_set(p, -2); | |
268 | break; | |
269 | } | |
270 | } | |
271 | ||
272 | show_list("bisection 2 initialize", counted, nr, list); | |
273 | ||
274 | /* | |
275 | * If you have only one parent in the resulting set | |
276 | * then you can reach one commit more than that parent | |
277 | * can reach. So we do not have to run the expensive | |
278 | * count_distance() for single strand of pearls. | |
279 | * | |
280 | * However, if you have more than one parents, you cannot | |
281 | * just add their distance and one for yourself, since | |
282 | * they usually reach the same ancestor and you would | |
283 | * end up counting them twice that way. | |
284 | * | |
285 | * So we will first count distance of merges the usual | |
286 | * way, and then fill the blanks using cheaper algorithm. | |
287 | */ | |
288 | for (p = list; p; p = p->next) { | |
289 | if (p->item->object.flags & UNINTERESTING) | |
290 | continue; | |
291 | if (weight(p) != -2) | |
292 | continue; | |
293 | weight_set(p, count_distance(p)); | |
294 | clear_distance(list); | |
295 | ||
296 | /* Does it happen to be at exactly half-way? */ | |
297 | if (!find_all && halfway(p, nr)) | |
298 | return p; | |
299 | counted++; | |
300 | } | |
301 | ||
302 | show_list("bisection 2 count_distance", counted, nr, list); | |
303 | ||
304 | while (counted < nr) { | |
305 | for (p = list; p; p = p->next) { | |
306 | struct commit_list *q; | |
307 | unsigned flags = p->item->object.flags; | |
308 | ||
309 | if (0 <= weight(p)) | |
310 | continue; | |
311 | for (q = p->item->parents; q; q = q->next) { | |
312 | if (q->item->object.flags & UNINTERESTING) | |
313 | continue; | |
314 | if (0 <= weight(q)) | |
315 | break; | |
316 | } | |
317 | if (!q) | |
318 | continue; | |
319 | ||
320 | /* | |
321 | * weight for p is unknown but q is known. | |
322 | * add one for p itself if p is to be counted, | |
323 | * otherwise inherit it from q directly. | |
324 | */ | |
325 | if (!(flags & TREESAME)) { | |
326 | weight_set(p, weight(q)+1); | |
327 | counted++; | |
328 | show_list("bisection 2 count one", | |
329 | counted, nr, list); | |
330 | } | |
331 | else | |
332 | weight_set(p, weight(q)); | |
333 | ||
334 | /* Does it happen to be at exactly half-way? */ | |
335 | if (!find_all && halfway(p, nr)) | |
336 | return p; | |
337 | } | |
338 | } | |
339 | ||
340 | show_list("bisection 2 counted all", counted, nr, list); | |
341 | ||
342 | if (!find_all) | |
343 | return best_bisection(list, nr); | |
344 | else | |
345 | return best_bisection_sorted(list, nr); | |
346 | } | |
347 | ||
348 | struct commit_list *find_bisection(struct commit_list *list, | |
349 | int *reaches, int *all, | |
350 | int find_all) | |
351 | { | |
352 | int nr, on_list; | |
353 | struct commit_list *p, *best, *next, *last; | |
354 | int *weights; | |
355 | ||
356 | show_list("bisection 2 entry", 0, 0, list); | |
357 | ||
358 | /* | |
359 | * Count the number of total and tree-changing items on the | |
360 | * list, while reversing the list. | |
361 | */ | |
362 | for (nr = on_list = 0, last = NULL, p = list; | |
363 | p; | |
364 | p = next) { | |
365 | unsigned flags = p->item->object.flags; | |
366 | ||
367 | next = p->next; | |
368 | if (flags & UNINTERESTING) | |
369 | continue; | |
370 | p->next = last; | |
371 | last = p; | |
372 | if (!(flags & TREESAME)) | |
373 | nr++; | |
374 | on_list++; | |
375 | } | |
376 | list = last; | |
377 | show_list("bisection 2 sorted", 0, nr, list); | |
378 | ||
379 | *all = nr; | |
380 | weights = xcalloc(on_list, sizeof(*weights)); | |
381 | ||
382 | /* Do the real work of finding bisection commit. */ | |
383 | best = do_find_bisection(list, nr, weights, find_all); | |
384 | if (best) { | |
385 | if (!find_all) | |
386 | best->next = NULL; | |
387 | *reaches = weight(best); | |
388 | } | |
389 | free(weights); | |
390 | return best; | |
391 | } | |
392 | ||
95188648 CC |
393 | static int skipcmp(const void *a, const void *b) |
394 | { | |
395 | return hashcmp(a, b); | |
396 | } | |
397 | ||
398 | static void prepare_skipped(void) | |
399 | { | |
400 | qsort(skipped_sha1, skipped_sha1_nr, sizeof(*skipped_sha1), skipcmp); | |
401 | } | |
402 | ||
4eb5b646 CC |
403 | static const unsigned char *skipped_sha1_access(size_t index, void *table) |
404 | { | |
405 | unsigned char (*skipped)[20] = table; | |
406 | return skipped[index]; | |
407 | } | |
408 | ||
95188648 CC |
409 | static int lookup_skipped(unsigned char *sha1) |
410 | { | |
4eb5b646 CC |
411 | return sha1_pos(sha1, skipped_sha1, skipped_sha1_nr, |
412 | skipped_sha1_access); | |
95188648 CC |
413 | } |
414 | ||
415 | struct commit_list *filter_skipped(struct commit_list *list, | |
416 | struct commit_list **tried, | |
417 | int show_all) | |
418 | { | |
419 | struct commit_list *filtered = NULL, **f = &filtered; | |
420 | ||
421 | *tried = NULL; | |
422 | ||
423 | if (!skipped_sha1_nr) | |
424 | return list; | |
425 | ||
426 | prepare_skipped(); | |
427 | ||
428 | while (list) { | |
429 | struct commit_list *next = list->next; | |
430 | list->next = NULL; | |
431 | if (0 <= lookup_skipped(list->item->object.sha1)) { | |
432 | /* Move current to tried list */ | |
433 | *tried = list; | |
434 | tried = &list->next; | |
435 | } else { | |
436 | if (!show_all) | |
437 | return list; | |
438 | /* Move current to filtered list */ | |
439 | *f = list; | |
440 | f = &list->next; | |
441 | } | |
442 | list = next; | |
443 | } | |
444 | ||
445 | return filtered; | |
446 | } |