]>
Commit | Line | Data |
---|---|---|
1 | #include "cache.h" | |
2 | #include "config.h" | |
3 | #include "commit.h" | |
4 | #include "diff.h" | |
5 | #include "revision.h" | |
6 | #include "refs.h" | |
7 | #include "list-objects.h" | |
8 | #include "quote.h" | |
9 | #include "sha1-lookup.h" | |
10 | #include "run-command.h" | |
11 | #include "log-tree.h" | |
12 | #include "bisect.h" | |
13 | #include "oid-array.h" | |
14 | #include "strvec.h" | |
15 | #include "commit-slab.h" | |
16 | #include "commit-reach.h" | |
17 | #include "object-store.h" | |
18 | #include "dir.h" | |
19 | ||
20 | static struct oid_array good_revs; | |
21 | static struct oid_array skipped_revs; | |
22 | ||
23 | static struct object_id *current_bad_oid; | |
24 | ||
25 | static const char *argv_checkout[] = {"checkout", "-q", NULL, "--", NULL}; | |
26 | static const char *argv_show_branch[] = {"show-branch", NULL, NULL}; | |
27 | ||
28 | static const char *term_bad; | |
29 | static const char *term_good; | |
30 | ||
31 | /* Remember to update object flag allocation in object.h */ | |
32 | #define COUNTED (1u<<16) | |
33 | ||
34 | /* | |
35 | * This is a truly stupid algorithm, but it's only | |
36 | * used for bisection, and we just don't care enough. | |
37 | * | |
38 | * We care just barely enough to avoid recursing for | |
39 | * non-merge entries. | |
40 | */ | |
41 | static int count_distance(struct commit_list *entry) | |
42 | { | |
43 | int nr = 0; | |
44 | ||
45 | while (entry) { | |
46 | struct commit *commit = entry->item; | |
47 | struct commit_list *p; | |
48 | ||
49 | if (commit->object.flags & (UNINTERESTING | COUNTED)) | |
50 | break; | |
51 | if (!(commit->object.flags & TREESAME)) | |
52 | nr++; | |
53 | commit->object.flags |= COUNTED; | |
54 | p = commit->parents; | |
55 | entry = p; | |
56 | if (p) { | |
57 | p = p->next; | |
58 | while (p) { | |
59 | nr += count_distance(p); | |
60 | p = p->next; | |
61 | } | |
62 | } | |
63 | } | |
64 | ||
65 | return nr; | |
66 | } | |
67 | ||
68 | static void clear_distance(struct commit_list *list) | |
69 | { | |
70 | while (list) { | |
71 | struct commit *commit = list->item; | |
72 | commit->object.flags &= ~COUNTED; | |
73 | list = list->next; | |
74 | } | |
75 | } | |
76 | ||
77 | define_commit_slab(commit_weight, int *); | |
78 | static struct commit_weight commit_weight; | |
79 | ||
80 | #define DEBUG_BISECT 0 | |
81 | ||
82 | static inline int weight(struct commit_list *elem) | |
83 | { | |
84 | return **commit_weight_at(&commit_weight, elem->item); | |
85 | } | |
86 | ||
87 | static inline void weight_set(struct commit_list *elem, int weight) | |
88 | { | |
89 | **commit_weight_at(&commit_weight, elem->item) = weight; | |
90 | } | |
91 | ||
92 | static int count_interesting_parents(struct commit *commit, unsigned bisect_flags) | |
93 | { | |
94 | struct commit_list *p; | |
95 | int count; | |
96 | ||
97 | for (count = 0, p = commit->parents; p; p = p->next) { | |
98 | if (!(p->item->object.flags & UNINTERESTING)) | |
99 | count++; | |
100 | if (bisect_flags & FIND_BISECTION_FIRST_PARENT_ONLY) | |
101 | break; | |
102 | } | |
103 | return count; | |
104 | } | |
105 | ||
106 | static inline int halfway(struct commit_list *p, int nr) | |
107 | { | |
108 | /* | |
109 | * Don't short-cut something we are not going to return! | |
110 | */ | |
111 | if (p->item->object.flags & TREESAME) | |
112 | return 0; | |
113 | if (DEBUG_BISECT) | |
114 | return 0; | |
115 | /* | |
116 | * 2 and 3 are halfway of 5. | |
117 | * 3 is halfway of 6 but 2 and 4 are not. | |
118 | */ | |
119 | switch (2 * weight(p) - nr) { | |
120 | case -1: case 0: case 1: | |
121 | return 1; | |
122 | default: | |
123 | return 0; | |
124 | } | |
125 | } | |
126 | ||
127 | static void show_list(const char *debug, int counted, int nr, | |
128 | struct commit_list *list) | |
129 | { | |
130 | struct commit_list *p; | |
131 | ||
132 | if (!DEBUG_BISECT) | |
133 | return; | |
134 | ||
135 | fprintf(stderr, "%s (%d/%d)\n", debug, counted, nr); | |
136 | ||
137 | for (p = list; p; p = p->next) { | |
138 | struct commit_list *pp; | |
139 | struct commit *commit = p->item; | |
140 | unsigned commit_flags = commit->object.flags; | |
141 | enum object_type type; | |
142 | unsigned long size; | |
143 | char *buf = read_object_file(&commit->object.oid, &type, | |
144 | &size); | |
145 | const char *subject_start; | |
146 | int subject_len; | |
147 | ||
148 | fprintf(stderr, "%c%c%c ", | |
149 | (commit_flags & TREESAME) ? ' ' : 'T', | |
150 | (commit_flags & UNINTERESTING) ? 'U' : ' ', | |
151 | (commit_flags & COUNTED) ? 'C' : ' '); | |
152 | if (*commit_weight_at(&commit_weight, p->item)) | |
153 | fprintf(stderr, "%3d", weight(p)); | |
154 | else | |
155 | fprintf(stderr, "---"); | |
156 | fprintf(stderr, " %.*s", 8, oid_to_hex(&commit->object.oid)); | |
157 | for (pp = commit->parents; pp; pp = pp->next) | |
158 | fprintf(stderr, " %.*s", 8, | |
159 | oid_to_hex(&pp->item->object.oid)); | |
160 | ||
161 | subject_len = find_commit_subject(buf, &subject_start); | |
162 | if (subject_len) | |
163 | fprintf(stderr, " %.*s", subject_len, subject_start); | |
164 | fprintf(stderr, "\n"); | |
165 | } | |
166 | } | |
167 | ||
168 | static struct commit_list *best_bisection(struct commit_list *list, int nr) | |
169 | { | |
170 | struct commit_list *p, *best; | |
171 | int best_distance = -1; | |
172 | ||
173 | best = list; | |
174 | for (p = list; p; p = p->next) { | |
175 | int distance; | |
176 | unsigned commit_flags = p->item->object.flags; | |
177 | ||
178 | if (commit_flags & TREESAME) | |
179 | continue; | |
180 | distance = weight(p); | |
181 | if (nr - distance < distance) | |
182 | distance = nr - distance; | |
183 | if (distance > best_distance) { | |
184 | best = p; | |
185 | best_distance = distance; | |
186 | } | |
187 | } | |
188 | ||
189 | return best; | |
190 | } | |
191 | ||
192 | struct commit_dist { | |
193 | struct commit *commit; | |
194 | int distance; | |
195 | }; | |
196 | ||
197 | static int compare_commit_dist(const void *a_, const void *b_) | |
198 | { | |
199 | struct commit_dist *a, *b; | |
200 | ||
201 | a = (struct commit_dist *)a_; | |
202 | b = (struct commit_dist *)b_; | |
203 | if (a->distance != b->distance) | |
204 | return b->distance - a->distance; /* desc sort */ | |
205 | return oidcmp(&a->commit->object.oid, &b->commit->object.oid); | |
206 | } | |
207 | ||
208 | static struct commit_list *best_bisection_sorted(struct commit_list *list, int nr) | |
209 | { | |
210 | struct commit_list *p; | |
211 | struct commit_dist *array = xcalloc(nr, sizeof(*array)); | |
212 | struct strbuf buf = STRBUF_INIT; | |
213 | int cnt, i; | |
214 | ||
215 | for (p = list, cnt = 0; p; p = p->next) { | |
216 | int distance; | |
217 | unsigned commit_flags = p->item->object.flags; | |
218 | ||
219 | if (commit_flags & TREESAME) | |
220 | continue; | |
221 | distance = weight(p); | |
222 | if (nr - distance < distance) | |
223 | distance = nr - distance; | |
224 | array[cnt].commit = p->item; | |
225 | array[cnt].distance = distance; | |
226 | cnt++; | |
227 | } | |
228 | QSORT(array, cnt, compare_commit_dist); | |
229 | for (p = list, i = 0; i < cnt; i++) { | |
230 | struct object *obj = &(array[i].commit->object); | |
231 | ||
232 | strbuf_reset(&buf); | |
233 | strbuf_addf(&buf, "dist=%d", array[i].distance); | |
234 | add_name_decoration(DECORATION_NONE, buf.buf, obj); | |
235 | ||
236 | p->item = array[i].commit; | |
237 | if (i < cnt - 1) | |
238 | p = p->next; | |
239 | } | |
240 | if (p) { | |
241 | free_commit_list(p->next); | |
242 | p->next = NULL; | |
243 | } | |
244 | strbuf_release(&buf); | |
245 | free(array); | |
246 | return list; | |
247 | } | |
248 | ||
249 | /* | |
250 | * zero or positive weight is the number of interesting commits it can | |
251 | * reach, including itself. Especially, weight = 0 means it does not | |
252 | * reach any tree-changing commits (e.g. just above uninteresting one | |
253 | * but traversal is with pathspec). | |
254 | * | |
255 | * weight = -1 means it has one parent and its distance is yet to | |
256 | * be computed. | |
257 | * | |
258 | * weight = -2 means it has more than one parent and its distance is | |
259 | * unknown. After running count_distance() first, they will get zero | |
260 | * or positive distance. | |
261 | */ | |
262 | static struct commit_list *do_find_bisection(struct commit_list *list, | |
263 | int nr, int *weights, | |
264 | unsigned bisect_flags) | |
265 | { | |
266 | int n, counted; | |
267 | struct commit_list *p; | |
268 | ||
269 | counted = 0; | |
270 | ||
271 | for (n = 0, p = list; p; p = p->next) { | |
272 | struct commit *commit = p->item; | |
273 | unsigned commit_flags = commit->object.flags; | |
274 | ||
275 | *commit_weight_at(&commit_weight, p->item) = &weights[n++]; | |
276 | switch (count_interesting_parents(commit, bisect_flags)) { | |
277 | case 0: | |
278 | if (!(commit_flags & TREESAME)) { | |
279 | weight_set(p, 1); | |
280 | counted++; | |
281 | show_list("bisection 2 count one", | |
282 | counted, nr, list); | |
283 | } | |
284 | /* | |
285 | * otherwise, it is known not to reach any | |
286 | * tree-changing commit and gets weight 0. | |
287 | */ | |
288 | break; | |
289 | case 1: | |
290 | weight_set(p, -1); | |
291 | break; | |
292 | default: | |
293 | weight_set(p, -2); | |
294 | break; | |
295 | } | |
296 | } | |
297 | ||
298 | show_list("bisection 2 initialize", counted, nr, list); | |
299 | ||
300 | /* | |
301 | * If you have only one parent in the resulting set | |
302 | * then you can reach one commit more than that parent | |
303 | * can reach. So we do not have to run the expensive | |
304 | * count_distance() for single strand of pearls. | |
305 | * | |
306 | * However, if you have more than one parents, you cannot | |
307 | * just add their distance and one for yourself, since | |
308 | * they usually reach the same ancestor and you would | |
309 | * end up counting them twice that way. | |
310 | * | |
311 | * So we will first count distance of merges the usual | |
312 | * way, and then fill the blanks using cheaper algorithm. | |
313 | */ | |
314 | for (p = list; p; p = p->next) { | |
315 | if (p->item->object.flags & UNINTERESTING) | |
316 | continue; | |
317 | if (weight(p) != -2) | |
318 | continue; | |
319 | if (bisect_flags & FIND_BISECTION_FIRST_PARENT_ONLY) | |
320 | BUG("shouldn't be calling count-distance in fp mode"); | |
321 | weight_set(p, count_distance(p)); | |
322 | clear_distance(list); | |
323 | ||
324 | /* Does it happen to be at exactly half-way? */ | |
325 | if (!(bisect_flags & FIND_BISECTION_ALL) && halfway(p, nr)) | |
326 | return p; | |
327 | counted++; | |
328 | } | |
329 | ||
330 | show_list("bisection 2 count_distance", counted, nr, list); | |
331 | ||
332 | while (counted < nr) { | |
333 | for (p = list; p; p = p->next) { | |
334 | struct commit_list *q; | |
335 | unsigned commit_flags = p->item->object.flags; | |
336 | ||
337 | if (0 <= weight(p)) | |
338 | continue; | |
339 | ||
340 | for (q = p->item->parents; | |
341 | q; | |
342 | q = bisect_flags & FIND_BISECTION_FIRST_PARENT_ONLY ? NULL : q->next) { | |
343 | if (q->item->object.flags & UNINTERESTING) | |
344 | continue; | |
345 | if (0 <= weight(q)) | |
346 | break; | |
347 | } | |
348 | if (!q) | |
349 | continue; | |
350 | ||
351 | /* | |
352 | * weight for p is unknown but q is known. | |
353 | * add one for p itself if p is to be counted, | |
354 | * otherwise inherit it from q directly. | |
355 | */ | |
356 | if (!(commit_flags & TREESAME)) { | |
357 | weight_set(p, weight(q)+1); | |
358 | counted++; | |
359 | show_list("bisection 2 count one", | |
360 | counted, nr, list); | |
361 | } | |
362 | else | |
363 | weight_set(p, weight(q)); | |
364 | ||
365 | /* Does it happen to be at exactly half-way? */ | |
366 | if (!(bisect_flags & FIND_BISECTION_ALL) && halfway(p, nr)) | |
367 | return p; | |
368 | } | |
369 | } | |
370 | ||
371 | show_list("bisection 2 counted all", counted, nr, list); | |
372 | ||
373 | if (!(bisect_flags & FIND_BISECTION_ALL)) | |
374 | return best_bisection(list, nr); | |
375 | else | |
376 | return best_bisection_sorted(list, nr); | |
377 | } | |
378 | ||
379 | void find_bisection(struct commit_list **commit_list, int *reaches, | |
380 | int *all, unsigned bisect_flags) | |
381 | { | |
382 | int nr, on_list; | |
383 | struct commit_list *list, *p, *best, *next, *last; | |
384 | int *weights; | |
385 | ||
386 | show_list("bisection 2 entry", 0, 0, *commit_list); | |
387 | init_commit_weight(&commit_weight); | |
388 | ||
389 | /* | |
390 | * Count the number of total and tree-changing items on the | |
391 | * list, while reversing the list. | |
392 | */ | |
393 | for (nr = on_list = 0, last = NULL, p = *commit_list; | |
394 | p; | |
395 | p = next) { | |
396 | unsigned commit_flags = p->item->object.flags; | |
397 | ||
398 | next = p->next; | |
399 | if (commit_flags & UNINTERESTING) { | |
400 | free(p); | |
401 | continue; | |
402 | } | |
403 | p->next = last; | |
404 | last = p; | |
405 | if (!(commit_flags & TREESAME)) | |
406 | nr++; | |
407 | on_list++; | |
408 | } | |
409 | list = last; | |
410 | show_list("bisection 2 sorted", 0, nr, list); | |
411 | ||
412 | *all = nr; | |
413 | weights = xcalloc(on_list, sizeof(*weights)); | |
414 | ||
415 | /* Do the real work of finding bisection commit. */ | |
416 | best = do_find_bisection(list, nr, weights, bisect_flags); | |
417 | if (best) { | |
418 | if (!(bisect_flags & FIND_BISECTION_ALL)) { | |
419 | list->item = best->item; | |
420 | free_commit_list(list->next); | |
421 | best = list; | |
422 | best->next = NULL; | |
423 | } | |
424 | *reaches = weight(best); | |
425 | } | |
426 | free(weights); | |
427 | *commit_list = best; | |
428 | clear_commit_weight(&commit_weight); | |
429 | } | |
430 | ||
431 | static int register_ref(const char *refname, const struct object_id *oid, | |
432 | int flags, void *cb_data) | |
433 | { | |
434 | struct strbuf good_prefix = STRBUF_INIT; | |
435 | strbuf_addstr(&good_prefix, term_good); | |
436 | strbuf_addstr(&good_prefix, "-"); | |
437 | ||
438 | if (!strcmp(refname, term_bad)) { | |
439 | current_bad_oid = xmalloc(sizeof(*current_bad_oid)); | |
440 | oidcpy(current_bad_oid, oid); | |
441 | } else if (starts_with(refname, good_prefix.buf)) { | |
442 | oid_array_append(&good_revs, oid); | |
443 | } else if (starts_with(refname, "skip-")) { | |
444 | oid_array_append(&skipped_revs, oid); | |
445 | } | |
446 | ||
447 | strbuf_release(&good_prefix); | |
448 | ||
449 | return 0; | |
450 | } | |
451 | ||
452 | static int read_bisect_refs(void) | |
453 | { | |
454 | return for_each_ref_in("refs/bisect/", register_ref, NULL); | |
455 | } | |
456 | ||
457 | static GIT_PATH_FUNC(git_path_bisect_names, "BISECT_NAMES") | |
458 | static GIT_PATH_FUNC(git_path_bisect_expected_rev, "BISECT_EXPECTED_REV") | |
459 | static GIT_PATH_FUNC(git_path_bisect_ancestors_ok, "BISECT_ANCESTORS_OK") | |
460 | static GIT_PATH_FUNC(git_path_bisect_run, "BISECT_RUN") | |
461 | static GIT_PATH_FUNC(git_path_bisect_start, "BISECT_START") | |
462 | static GIT_PATH_FUNC(git_path_bisect_log, "BISECT_LOG") | |
463 | static GIT_PATH_FUNC(git_path_bisect_terms, "BISECT_TERMS") | |
464 | static GIT_PATH_FUNC(git_path_bisect_first_parent, "BISECT_FIRST_PARENT") | |
465 | static GIT_PATH_FUNC(git_path_head_name, "head-name") | |
466 | ||
467 | static void read_bisect_paths(struct strvec *array) | |
468 | { | |
469 | struct strbuf str = STRBUF_INIT; | |
470 | const char *filename = git_path_bisect_names(); | |
471 | FILE *fp = xfopen(filename, "r"); | |
472 | ||
473 | while (strbuf_getline_lf(&str, fp) != EOF) { | |
474 | strbuf_trim(&str); | |
475 | if (sq_dequote_to_strvec(str.buf, array)) | |
476 | die(_("Badly quoted content in file '%s': %s"), | |
477 | filename, str.buf); | |
478 | } | |
479 | ||
480 | strbuf_release(&str); | |
481 | fclose(fp); | |
482 | } | |
483 | ||
484 | static char *join_oid_array_hex(struct oid_array *array, char delim) | |
485 | { | |
486 | struct strbuf joined_hexs = STRBUF_INIT; | |
487 | int i; | |
488 | ||
489 | for (i = 0; i < array->nr; i++) { | |
490 | strbuf_addstr(&joined_hexs, oid_to_hex(array->oid + i)); | |
491 | if (i + 1 < array->nr) | |
492 | strbuf_addch(&joined_hexs, delim); | |
493 | } | |
494 | ||
495 | return strbuf_detach(&joined_hexs, NULL); | |
496 | } | |
497 | ||
498 | /* | |
499 | * In this function, passing a not NULL skipped_first is very special. | |
500 | * It means that we want to know if the first commit in the list is | |
501 | * skipped because we will want to test a commit away from it if it is | |
502 | * indeed skipped. | |
503 | * So if the first commit is skipped, we cannot take the shortcut to | |
504 | * just "return list" when we find the first non skipped commit, we | |
505 | * have to return a fully filtered list. | |
506 | * | |
507 | * We use (*skipped_first == -1) to mean "it has been found that the | |
508 | * first commit is not skipped". In this case *skipped_first is set back | |
509 | * to 0 just before the function returns. | |
510 | */ | |
511 | struct commit_list *filter_skipped(struct commit_list *list, | |
512 | struct commit_list **tried, | |
513 | int show_all, | |
514 | int *count, | |
515 | int *skipped_first) | |
516 | { | |
517 | struct commit_list *filtered = NULL, **f = &filtered; | |
518 | ||
519 | *tried = NULL; | |
520 | ||
521 | if (skipped_first) | |
522 | *skipped_first = 0; | |
523 | if (count) | |
524 | *count = 0; | |
525 | ||
526 | if (!skipped_revs.nr) | |
527 | return list; | |
528 | ||
529 | while (list) { | |
530 | struct commit_list *next = list->next; | |
531 | list->next = NULL; | |
532 | if (0 <= oid_array_lookup(&skipped_revs, &list->item->object.oid)) { | |
533 | if (skipped_first && !*skipped_first) | |
534 | *skipped_first = 1; | |
535 | /* Move current to tried list */ | |
536 | *tried = list; | |
537 | tried = &list->next; | |
538 | } else { | |
539 | if (!show_all) { | |
540 | if (!skipped_first || !*skipped_first) | |
541 | return list; | |
542 | } else if (skipped_first && !*skipped_first) { | |
543 | /* This means we know it's not skipped */ | |
544 | *skipped_first = -1; | |
545 | } | |
546 | /* Move current to filtered list */ | |
547 | *f = list; | |
548 | f = &list->next; | |
549 | if (count) | |
550 | (*count)++; | |
551 | } | |
552 | list = next; | |
553 | } | |
554 | ||
555 | if (skipped_first && *skipped_first == -1) | |
556 | *skipped_first = 0; | |
557 | ||
558 | return filtered; | |
559 | } | |
560 | ||
561 | #define PRN_MODULO 32768 | |
562 | ||
563 | /* | |
564 | * This is a pseudo random number generator based on "man 3 rand". | |
565 | * It is not used properly because the seed is the argument and it | |
566 | * is increased by one between each call, but that should not matter | |
567 | * for this application. | |
568 | */ | |
569 | static unsigned get_prn(unsigned count) | |
570 | { | |
571 | count = count * 1103515245 + 12345; | |
572 | return (count/65536) % PRN_MODULO; | |
573 | } | |
574 | ||
575 | /* | |
576 | * Custom integer square root from | |
577 | * https://en.wikipedia.org/wiki/Integer_square_root | |
578 | */ | |
579 | static int sqrti(int val) | |
580 | { | |
581 | float d, x = val; | |
582 | ||
583 | if (!val) | |
584 | return 0; | |
585 | ||
586 | do { | |
587 | float y = (x + (float)val / x) / 2; | |
588 | d = (y > x) ? y - x : x - y; | |
589 | x = y; | |
590 | } while (d >= 0.5); | |
591 | ||
592 | return (int)x; | |
593 | } | |
594 | ||
595 | static struct commit_list *skip_away(struct commit_list *list, int count) | |
596 | { | |
597 | struct commit_list *cur, *previous; | |
598 | int prn, index, i; | |
599 | ||
600 | prn = get_prn(count); | |
601 | index = (count * prn / PRN_MODULO) * sqrti(prn) / sqrti(PRN_MODULO); | |
602 | ||
603 | cur = list; | |
604 | previous = NULL; | |
605 | ||
606 | for (i = 0; cur; cur = cur->next, i++) { | |
607 | if (i == index) { | |
608 | if (!oideq(&cur->item->object.oid, current_bad_oid)) | |
609 | return cur; | |
610 | if (previous) | |
611 | return previous; | |
612 | return list; | |
613 | } | |
614 | previous = cur; | |
615 | } | |
616 | ||
617 | return list; | |
618 | } | |
619 | ||
620 | static struct commit_list *managed_skipped(struct commit_list *list, | |
621 | struct commit_list **tried) | |
622 | { | |
623 | int count, skipped_first; | |
624 | ||
625 | *tried = NULL; | |
626 | ||
627 | if (!skipped_revs.nr) | |
628 | return list; | |
629 | ||
630 | list = filter_skipped(list, tried, 0, &count, &skipped_first); | |
631 | ||
632 | if (!skipped_first) | |
633 | return list; | |
634 | ||
635 | return skip_away(list, count); | |
636 | } | |
637 | ||
638 | static void bisect_rev_setup(struct repository *r, struct rev_info *revs, | |
639 | const char *prefix, | |
640 | const char *bad_format, const char *good_format, | |
641 | int read_paths) | |
642 | { | |
643 | struct strvec rev_argv = STRVEC_INIT; | |
644 | int i; | |
645 | ||
646 | repo_init_revisions(r, revs, prefix); | |
647 | revs->abbrev = 0; | |
648 | revs->commit_format = CMIT_FMT_UNSPECIFIED; | |
649 | ||
650 | /* rev_argv.argv[0] will be ignored by setup_revisions */ | |
651 | strvec_push(&rev_argv, "bisect_rev_setup"); | |
652 | strvec_pushf(&rev_argv, bad_format, oid_to_hex(current_bad_oid)); | |
653 | for (i = 0; i < good_revs.nr; i++) | |
654 | strvec_pushf(&rev_argv, good_format, | |
655 | oid_to_hex(good_revs.oid + i)); | |
656 | strvec_push(&rev_argv, "--"); | |
657 | if (read_paths) | |
658 | read_bisect_paths(&rev_argv); | |
659 | ||
660 | setup_revisions(rev_argv.nr, rev_argv.v, revs, NULL); | |
661 | /* XXX leak rev_argv, as "revs" may still be pointing to it */ | |
662 | } | |
663 | ||
664 | static void bisect_common(struct rev_info *revs) | |
665 | { | |
666 | if (prepare_revision_walk(revs)) | |
667 | die("revision walk setup failed"); | |
668 | if (revs->tree_objects) | |
669 | mark_edges_uninteresting(revs, NULL, 0); | |
670 | } | |
671 | ||
672 | static enum bisect_error error_if_skipped_commits(struct commit_list *tried, | |
673 | const struct object_id *bad) | |
674 | { | |
675 | if (!tried) | |
676 | return BISECT_OK; | |
677 | ||
678 | printf("There are only 'skip'ped commits left to test.\n" | |
679 | "The first %s commit could be any of:\n", term_bad); | |
680 | ||
681 | for ( ; tried; tried = tried->next) | |
682 | printf("%s\n", oid_to_hex(&tried->item->object.oid)); | |
683 | ||
684 | if (bad) | |
685 | printf("%s\n", oid_to_hex(bad)); | |
686 | printf(_("We cannot bisect more!\n")); | |
687 | ||
688 | return BISECT_ONLY_SKIPPED_LEFT; | |
689 | } | |
690 | ||
691 | static int is_expected_rev(const struct object_id *oid) | |
692 | { | |
693 | const char *filename = git_path_bisect_expected_rev(); | |
694 | struct stat st; | |
695 | struct strbuf str = STRBUF_INIT; | |
696 | FILE *fp; | |
697 | int res = 0; | |
698 | ||
699 | if (stat(filename, &st) || !S_ISREG(st.st_mode)) | |
700 | return 0; | |
701 | ||
702 | fp = fopen_or_warn(filename, "r"); | |
703 | if (!fp) | |
704 | return 0; | |
705 | ||
706 | if (strbuf_getline_lf(&str, fp) != EOF) | |
707 | res = !strcmp(str.buf, oid_to_hex(oid)); | |
708 | ||
709 | strbuf_release(&str); | |
710 | fclose(fp); | |
711 | ||
712 | return res; | |
713 | } | |
714 | ||
715 | static enum bisect_error bisect_checkout(const struct object_id *bisect_rev, int no_checkout) | |
716 | { | |
717 | char bisect_rev_hex[GIT_MAX_HEXSZ + 1]; | |
718 | enum bisect_error res = BISECT_OK; | |
719 | ||
720 | oid_to_hex_r(bisect_rev_hex, bisect_rev); | |
721 | update_ref(NULL, "BISECT_EXPECTED_REV", bisect_rev, NULL, 0, UPDATE_REFS_DIE_ON_ERR); | |
722 | ||
723 | argv_checkout[2] = bisect_rev_hex; | |
724 | if (no_checkout) { | |
725 | update_ref(NULL, "BISECT_HEAD", bisect_rev, NULL, 0, | |
726 | UPDATE_REFS_DIE_ON_ERR); | |
727 | } else { | |
728 | res = run_command_v_opt(argv_checkout, RUN_GIT_CMD); | |
729 | if (res) | |
730 | /* | |
731 | * Errors in `run_command()` itself, signaled by res < 0, | |
732 | * and errors in the child process, signaled by res > 0 | |
733 | * can both be treated as regular BISECT_FAILURE (-1). | |
734 | */ | |
735 | return -abs(res); | |
736 | } | |
737 | ||
738 | argv_show_branch[1] = bisect_rev_hex; | |
739 | res = run_command_v_opt(argv_show_branch, RUN_GIT_CMD); | |
740 | /* | |
741 | * Errors in `run_command()` itself, signaled by res < 0, | |
742 | * and errors in the child process, signaled by res > 0 | |
743 | * can both be treated as regular BISECT_FAILURE (-1). | |
744 | */ | |
745 | return -abs(res); | |
746 | } | |
747 | ||
748 | static struct commit *get_commit_reference(struct repository *r, | |
749 | const struct object_id *oid) | |
750 | { | |
751 | struct commit *c = lookup_commit_reference(r, oid); | |
752 | if (!c) | |
753 | die(_("Not a valid commit name %s"), oid_to_hex(oid)); | |
754 | return c; | |
755 | } | |
756 | ||
757 | static struct commit **get_bad_and_good_commits(struct repository *r, | |
758 | int *rev_nr) | |
759 | { | |
760 | struct commit **rev; | |
761 | int i, n = 0; | |
762 | ||
763 | ALLOC_ARRAY(rev, 1 + good_revs.nr); | |
764 | rev[n++] = get_commit_reference(r, current_bad_oid); | |
765 | for (i = 0; i < good_revs.nr; i++) | |
766 | rev[n++] = get_commit_reference(r, good_revs.oid + i); | |
767 | *rev_nr = n; | |
768 | ||
769 | return rev; | |
770 | } | |
771 | ||
772 | static enum bisect_error handle_bad_merge_base(void) | |
773 | { | |
774 | if (is_expected_rev(current_bad_oid)) { | |
775 | char *bad_hex = oid_to_hex(current_bad_oid); | |
776 | char *good_hex = join_oid_array_hex(&good_revs, ' '); | |
777 | if (!strcmp(term_bad, "bad") && !strcmp(term_good, "good")) { | |
778 | fprintf(stderr, _("The merge base %s is bad.\n" | |
779 | "This means the bug has been fixed " | |
780 | "between %s and [%s].\n"), | |
781 | bad_hex, bad_hex, good_hex); | |
782 | } else if (!strcmp(term_bad, "new") && !strcmp(term_good, "old")) { | |
783 | fprintf(stderr, _("The merge base %s is new.\n" | |
784 | "The property has changed " | |
785 | "between %s and [%s].\n"), | |
786 | bad_hex, bad_hex, good_hex); | |
787 | } else { | |
788 | fprintf(stderr, _("The merge base %s is %s.\n" | |
789 | "This means the first '%s' commit is " | |
790 | "between %s and [%s].\n"), | |
791 | bad_hex, term_bad, term_good, bad_hex, good_hex); | |
792 | } | |
793 | return BISECT_MERGE_BASE_CHECK; | |
794 | } | |
795 | ||
796 | fprintf(stderr, _("Some %s revs are not ancestors of the %s rev.\n" | |
797 | "git bisect cannot work properly in this case.\n" | |
798 | "Maybe you mistook %s and %s revs?\n"), | |
799 | term_good, term_bad, term_good, term_bad); | |
800 | return BISECT_FAILED; | |
801 | } | |
802 | ||
803 | static void handle_skipped_merge_base(const struct object_id *mb) | |
804 | { | |
805 | char *mb_hex = oid_to_hex(mb); | |
806 | char *bad_hex = oid_to_hex(current_bad_oid); | |
807 | char *good_hex = join_oid_array_hex(&good_revs, ' '); | |
808 | ||
809 | warning(_("the merge base between %s and [%s] " | |
810 | "must be skipped.\n" | |
811 | "So we cannot be sure the first %s commit is " | |
812 | "between %s and %s.\n" | |
813 | "We continue anyway."), | |
814 | bad_hex, good_hex, term_bad, mb_hex, bad_hex); | |
815 | free(good_hex); | |
816 | } | |
817 | ||
818 | /* | |
819 | * "check_merge_bases" checks that merge bases are not "bad" (or "new"). | |
820 | * | |
821 | * - If one is "bad" (or "new"), it means the user assumed something wrong | |
822 | * and we must return error with a non 0 error code. | |
823 | * - If one is "good" (or "old"), that's good, we have nothing to do. | |
824 | * - If one is "skipped", we can't know but we should warn. | |
825 | * - If we don't know, we should check it out and ask the user to test. | |
826 | * - If a merge base must be tested, on success return | |
827 | * BISECT_INTERNAL_SUCCESS_MERGE_BASE (-11) a special condition | |
828 | * for early success, this will be converted back to 0 in | |
829 | * check_good_are_ancestors_of_bad(). | |
830 | */ | |
831 | static enum bisect_error check_merge_bases(int rev_nr, struct commit **rev, int no_checkout) | |
832 | { | |
833 | enum bisect_error res = BISECT_OK; | |
834 | struct commit_list *result; | |
835 | ||
836 | result = get_merge_bases_many(rev[0], rev_nr - 1, rev + 1); | |
837 | ||
838 | for (; result; result = result->next) { | |
839 | const struct object_id *mb = &result->item->object.oid; | |
840 | if (oideq(mb, current_bad_oid)) { | |
841 | res = handle_bad_merge_base(); | |
842 | break; | |
843 | } else if (0 <= oid_array_lookup(&good_revs, mb)) { | |
844 | continue; | |
845 | } else if (0 <= oid_array_lookup(&skipped_revs, mb)) { | |
846 | handle_skipped_merge_base(mb); | |
847 | } else { | |
848 | printf(_("Bisecting: a merge base must be tested\n")); | |
849 | res = bisect_checkout(mb, no_checkout); | |
850 | if (!res) | |
851 | /* indicate early success */ | |
852 | res = BISECT_INTERNAL_SUCCESS_MERGE_BASE; | |
853 | break; | |
854 | } | |
855 | } | |
856 | ||
857 | free_commit_list(result); | |
858 | return res; | |
859 | } | |
860 | ||
861 | static int check_ancestors(struct repository *r, int rev_nr, | |
862 | struct commit **rev, const char *prefix) | |
863 | { | |
864 | struct rev_info revs; | |
865 | int res; | |
866 | ||
867 | bisect_rev_setup(r, &revs, prefix, "^%s", "%s", 0); | |
868 | ||
869 | bisect_common(&revs); | |
870 | res = (revs.commits != NULL); | |
871 | ||
872 | /* Clean up objects used, as they will be reused. */ | |
873 | clear_commit_marks_many(rev_nr, rev, ALL_REV_FLAGS); | |
874 | ||
875 | return res; | |
876 | } | |
877 | ||
878 | /* | |
879 | * "check_good_are_ancestors_of_bad" checks that all "good" revs are | |
880 | * ancestor of the "bad" rev. | |
881 | * | |
882 | * If that's not the case, we need to check the merge bases. | |
883 | * If a merge base must be tested by the user, its source code will be | |
884 | * checked out to be tested by the user and we will return. | |
885 | */ | |
886 | ||
887 | static enum bisect_error check_good_are_ancestors_of_bad(struct repository *r, | |
888 | const char *prefix, | |
889 | int no_checkout) | |
890 | { | |
891 | char *filename; | |
892 | struct stat st; | |
893 | int fd, rev_nr; | |
894 | enum bisect_error res = BISECT_OK; | |
895 | struct commit **rev; | |
896 | ||
897 | if (!current_bad_oid) | |
898 | return error(_("a %s revision is needed"), term_bad); | |
899 | ||
900 | filename = git_pathdup("BISECT_ANCESTORS_OK"); | |
901 | ||
902 | /* Check if file BISECT_ANCESTORS_OK exists. */ | |
903 | if (!stat(filename, &st) && S_ISREG(st.st_mode)) | |
904 | goto done; | |
905 | ||
906 | /* Bisecting with no good rev is ok. */ | |
907 | if (!good_revs.nr) | |
908 | goto done; | |
909 | ||
910 | /* Check if all good revs are ancestor of the bad rev. */ | |
911 | ||
912 | rev = get_bad_and_good_commits(r, &rev_nr); | |
913 | if (check_ancestors(r, rev_nr, rev, prefix)) | |
914 | res = check_merge_bases(rev_nr, rev, no_checkout); | |
915 | free(rev); | |
916 | ||
917 | if (!res) { | |
918 | /* Create file BISECT_ANCESTORS_OK. */ | |
919 | fd = open(filename, O_CREAT | O_TRUNC | O_WRONLY, 0600); | |
920 | if (fd < 0) | |
921 | /* | |
922 | * BISECT_ANCESTORS_OK file is not absolutely necessary, | |
923 | * the bisection process will continue at the next | |
924 | * bisection step. | |
925 | * So, just signal with a warning that something | |
926 | * might be wrong. | |
927 | */ | |
928 | warning_errno(_("could not create file '%s'"), | |
929 | filename); | |
930 | else | |
931 | close(fd); | |
932 | } | |
933 | done: | |
934 | free(filename); | |
935 | return res; | |
936 | } | |
937 | ||
938 | /* | |
939 | * This does "git diff-tree --pretty COMMIT" without one fork+exec. | |
940 | */ | |
941 | static void show_diff_tree(struct repository *r, | |
942 | const char *prefix, | |
943 | struct commit *commit) | |
944 | { | |
945 | const char *argv[] = { | |
946 | "diff-tree", "--pretty", "--stat", "--summary", "--cc", NULL | |
947 | }; | |
948 | struct rev_info opt; | |
949 | ||
950 | git_config(git_diff_ui_config, NULL); | |
951 | repo_init_revisions(r, &opt, prefix); | |
952 | ||
953 | setup_revisions(ARRAY_SIZE(argv) - 1, argv, &opt, NULL); | |
954 | log_tree_commit(&opt, commit); | |
955 | } | |
956 | ||
957 | /* | |
958 | * The terms used for this bisect session are stored in BISECT_TERMS. | |
959 | * We read them and store them to adapt the messages accordingly. | |
960 | * Default is bad/good. | |
961 | */ | |
962 | void read_bisect_terms(const char **read_bad, const char **read_good) | |
963 | { | |
964 | struct strbuf str = STRBUF_INIT; | |
965 | const char *filename = git_path_bisect_terms(); | |
966 | FILE *fp = fopen(filename, "r"); | |
967 | ||
968 | if (!fp) { | |
969 | if (errno == ENOENT) { | |
970 | *read_bad = "bad"; | |
971 | *read_good = "good"; | |
972 | return; | |
973 | } else { | |
974 | die_errno(_("could not read file '%s'"), filename); | |
975 | } | |
976 | } else { | |
977 | strbuf_getline_lf(&str, fp); | |
978 | *read_bad = strbuf_detach(&str, NULL); | |
979 | strbuf_getline_lf(&str, fp); | |
980 | *read_good = strbuf_detach(&str, NULL); | |
981 | } | |
982 | strbuf_release(&str); | |
983 | fclose(fp); | |
984 | } | |
985 | ||
986 | /* | |
987 | * We use the convention that return BISECT_INTERNAL_SUCCESS_1ST_BAD_FOUND (-10) means | |
988 | * the bisection process finished successfully. | |
989 | * In this case the calling function or command should not turn a | |
990 | * BISECT_INTERNAL_SUCCESS_1ST_BAD_FOUND return code into an error or a non zero exit code. | |
991 | * If no_checkout is non-zero, the bisection process does not | |
992 | * checkout the trial commit but instead simply updates BISECT_HEAD. | |
993 | */ | |
994 | enum bisect_error bisect_next_all(struct repository *r, const char *prefix) | |
995 | { | |
996 | struct rev_info revs; | |
997 | struct commit_list *tried; | |
998 | int reaches = 0, all = 0, nr, steps; | |
999 | enum bisect_error res = BISECT_OK; | |
1000 | struct object_id *bisect_rev; | |
1001 | char *steps_msg; | |
1002 | int no_checkout = ref_exists("BISECT_HEAD"); | |
1003 | unsigned bisect_flags = 0; | |
1004 | ||
1005 | read_bisect_terms(&term_bad, &term_good); | |
1006 | if (read_bisect_refs()) | |
1007 | die(_("reading bisect refs failed")); | |
1008 | ||
1009 | if (file_exists(git_path_bisect_first_parent())) | |
1010 | bisect_flags |= FIND_BISECTION_FIRST_PARENT_ONLY; | |
1011 | ||
1012 | if (skipped_revs.nr) | |
1013 | bisect_flags |= FIND_BISECTION_ALL; | |
1014 | ||
1015 | res = check_good_are_ancestors_of_bad(r, prefix, no_checkout); | |
1016 | if (res) | |
1017 | return res; | |
1018 | ||
1019 | bisect_rev_setup(r, &revs, prefix, "%s", "^%s", 1); | |
1020 | ||
1021 | revs.first_parent_only = !!(bisect_flags & FIND_BISECTION_FIRST_PARENT_ONLY); | |
1022 | revs.limited = 1; | |
1023 | ||
1024 | bisect_common(&revs); | |
1025 | ||
1026 | find_bisection(&revs.commits, &reaches, &all, bisect_flags); | |
1027 | revs.commits = managed_skipped(revs.commits, &tried); | |
1028 | ||
1029 | if (!revs.commits) { | |
1030 | /* | |
1031 | * We should return error here only if the "bad" | |
1032 | * commit is also a "skip" commit. | |
1033 | */ | |
1034 | res = error_if_skipped_commits(tried, NULL); | |
1035 | if (res < 0) | |
1036 | return res; | |
1037 | printf(_("%s was both %s and %s\n"), | |
1038 | oid_to_hex(current_bad_oid), | |
1039 | term_good, | |
1040 | term_bad); | |
1041 | ||
1042 | return BISECT_FAILED; | |
1043 | } | |
1044 | ||
1045 | if (!all) { | |
1046 | fprintf(stderr, _("No testable commit found.\n" | |
1047 | "Maybe you started with bad path parameters?\n")); | |
1048 | ||
1049 | return BISECT_NO_TESTABLE_COMMIT; | |
1050 | } | |
1051 | ||
1052 | bisect_rev = &revs.commits->item->object.oid; | |
1053 | ||
1054 | if (oideq(bisect_rev, current_bad_oid)) { | |
1055 | res = error_if_skipped_commits(tried, current_bad_oid); | |
1056 | if (res) | |
1057 | return res; | |
1058 | printf("%s is the first %s commit\n", oid_to_hex(bisect_rev), | |
1059 | term_bad); | |
1060 | ||
1061 | show_diff_tree(r, prefix, revs.commits->item); | |
1062 | /* | |
1063 | * This means the bisection process succeeded. | |
1064 | * Using BISECT_INTERNAL_SUCCESS_1ST_BAD_FOUND (-10) | |
1065 | * so that the call chain can simply check | |
1066 | * for negative return values for early returns up | |
1067 | * until the cmd_bisect__helper() caller. | |
1068 | */ | |
1069 | return BISECT_INTERNAL_SUCCESS_1ST_BAD_FOUND; | |
1070 | } | |
1071 | ||
1072 | nr = all - reaches - 1; | |
1073 | steps = estimate_bisect_steps(all); | |
1074 | ||
1075 | steps_msg = xstrfmt(Q_("(roughly %d step)", "(roughly %d steps)", | |
1076 | steps), steps); | |
1077 | /* | |
1078 | * TRANSLATORS: the last %s will be replaced with "(roughly %d | |
1079 | * steps)" translation. | |
1080 | */ | |
1081 | printf(Q_("Bisecting: %d revision left to test after this %s\n", | |
1082 | "Bisecting: %d revisions left to test after this %s\n", | |
1083 | nr), nr, steps_msg); | |
1084 | free(steps_msg); | |
1085 | ||
1086 | return bisect_checkout(bisect_rev, no_checkout); | |
1087 | } | |
1088 | ||
1089 | static inline int log2i(int n) | |
1090 | { | |
1091 | int log2 = 0; | |
1092 | ||
1093 | for (; n > 1; n >>= 1) | |
1094 | log2++; | |
1095 | ||
1096 | return log2; | |
1097 | } | |
1098 | ||
1099 | static inline int exp2i(int n) | |
1100 | { | |
1101 | return 1 << n; | |
1102 | } | |
1103 | ||
1104 | /* | |
1105 | * Estimate the number of bisect steps left (after the current step) | |
1106 | * | |
1107 | * For any x between 0 included and 2^n excluded, the probability for | |
1108 | * n - 1 steps left looks like: | |
1109 | * | |
1110 | * P(2^n + x) == (2^n - x) / (2^n + x) | |
1111 | * | |
1112 | * and P(2^n + x) < 0.5 means 2^n < 3x | |
1113 | */ | |
1114 | int estimate_bisect_steps(int all) | |
1115 | { | |
1116 | int n, x, e; | |
1117 | ||
1118 | if (all < 3) | |
1119 | return 0; | |
1120 | ||
1121 | n = log2i(all); | |
1122 | e = exp2i(n); | |
1123 | x = all - e; | |
1124 | ||
1125 | return (e < 3 * x) ? n : n - 1; | |
1126 | } | |
1127 | ||
1128 | static int mark_for_removal(const char *refname, const struct object_id *oid, | |
1129 | int flag, void *cb_data) | |
1130 | { | |
1131 | struct string_list *refs = cb_data; | |
1132 | char *ref = xstrfmt("refs/bisect%s", refname); | |
1133 | string_list_append(refs, ref); | |
1134 | return 0; | |
1135 | } | |
1136 | ||
1137 | int bisect_clean_state(void) | |
1138 | { | |
1139 | int result = 0; | |
1140 | ||
1141 | /* There may be some refs packed during bisection */ | |
1142 | struct string_list refs_for_removal = STRING_LIST_INIT_NODUP; | |
1143 | for_each_ref_in("refs/bisect", mark_for_removal, (void *) &refs_for_removal); | |
1144 | string_list_append(&refs_for_removal, xstrdup("BISECT_HEAD")); | |
1145 | result = delete_refs("bisect: remove", &refs_for_removal, REF_NO_DEREF); | |
1146 | refs_for_removal.strdup_strings = 1; | |
1147 | string_list_clear(&refs_for_removal, 0); | |
1148 | unlink_or_warn(git_path_bisect_expected_rev()); | |
1149 | unlink_or_warn(git_path_bisect_ancestors_ok()); | |
1150 | unlink_or_warn(git_path_bisect_log()); | |
1151 | unlink_or_warn(git_path_bisect_names()); | |
1152 | unlink_or_warn(git_path_bisect_run()); | |
1153 | unlink_or_warn(git_path_bisect_terms()); | |
1154 | unlink_or_warn(git_path_bisect_first_parent()); | |
1155 | /* Cleanup head-name if it got left by an old version of git-bisect */ | |
1156 | unlink_or_warn(git_path_head_name()); | |
1157 | /* | |
1158 | * Cleanup BISECT_START last to support the --no-checkout option | |
1159 | * introduced in the commit 4796e823a. | |
1160 | */ | |
1161 | unlink_or_warn(git_path_bisect_start()); | |
1162 | ||
1163 | return result; | |
1164 | } |