bisect.c 24.3 KB
Newer Older
1 2 3 4
#include "cache.h"
#include "commit.h"
#include "diff.h"
#include "revision.h"
5 6
#include "refs.h"
#include "list-objects.h"
7
#include "quote.h"
8
#include "sha1-lookup.h"
9
#include "run-command.h"
10
#include "log-tree.h"
11
#include "bisect.h"
12
#include "sha1-array.h"
Jeff King's avatar
Jeff King committed
13
#include "argv-array.h"
14

15
static struct sha1_array good_revs;
16
static struct sha1_array skipped_revs;
17

18
static struct object_id *current_bad_oid;
19 20 21

static const char *argv_checkout[] = {"checkout", "-q", NULL, "--", NULL};
static const char *argv_show_branch[] = {"show-branch", NULL, NULL};
22
static const char *argv_update_ref[] = {"update-ref", "--no-deref", "BISECT_HEAD", NULL, NULL};
23

24 25 26
static const char *term_bad;
static const char *term_good;

27
/* Remember to update object flag allocation in object.h */
28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135
#define COUNTED		(1u<<16)

/*
 * This is a truly stupid algorithm, but it's only
 * used for bisection, and we just don't care enough.
 *
 * We care just barely enough to avoid recursing for
 * non-merge entries.
 */
static int count_distance(struct commit_list *entry)
{
	int nr = 0;

	while (entry) {
		struct commit *commit = entry->item;
		struct commit_list *p;

		if (commit->object.flags & (UNINTERESTING | COUNTED))
			break;
		if (!(commit->object.flags & TREESAME))
			nr++;
		commit->object.flags |= COUNTED;
		p = commit->parents;
		entry = p;
		if (p) {
			p = p->next;
			while (p) {
				nr += count_distance(p);
				p = p->next;
			}
		}
	}

	return nr;
}

static void clear_distance(struct commit_list *list)
{
	while (list) {
		struct commit *commit = list->item;
		commit->object.flags &= ~COUNTED;
		list = list->next;
	}
}

#define DEBUG_BISECT 0

static inline int weight(struct commit_list *elem)
{
	return *((int*)(elem->item->util));
}

static inline void weight_set(struct commit_list *elem, int weight)
{
	*((int*)(elem->item->util)) = weight;
}

static int count_interesting_parents(struct commit *commit)
{
	struct commit_list *p;
	int count;

	for (count = 0, p = commit->parents; p; p = p->next) {
		if (p->item->object.flags & UNINTERESTING)
			continue;
		count++;
	}
	return count;
}

static inline int halfway(struct commit_list *p, int nr)
{
	/*
	 * Don't short-cut something we are not going to return!
	 */
	if (p->item->object.flags & TREESAME)
		return 0;
	if (DEBUG_BISECT)
		return 0;
	/*
	 * 2 and 3 are halfway of 5.
	 * 3 is halfway of 6 but 2 and 4 are not.
	 */
	switch (2 * weight(p) - nr) {
	case -1: case 0: case 1:
		return 1;
	default:
		return 0;
	}
}

#if !DEBUG_BISECT
#define show_list(a,b,c,d) do { ; } while (0)
#else
static void show_list(const char *debug, int counted, int nr,
		      struct commit_list *list)
{
	struct commit_list *p;

	fprintf(stderr, "%s (%d/%d)\n", debug, counted, nr);

	for (p = list; p; p = p->next) {
		struct commit_list *pp;
		struct commit *commit = p->item;
		unsigned flags = commit->object.flags;
		enum object_type type;
		unsigned long size;
		char *buf = read_sha1_file(commit->object.sha1, &type, &size);
136 137
		const char *subject_start;
		int subject_len;
138 139 140 141 142 143 144 145 146 147 148 149 150 151

		fprintf(stderr, "%c%c%c ",
			(flags & TREESAME) ? ' ' : 'T',
			(flags & UNINTERESTING) ? 'U' : ' ',
			(flags & COUNTED) ? 'C' : ' ');
		if (commit->util)
			fprintf(stderr, "%3d", weight(p));
		else
			fprintf(stderr, "---");
		fprintf(stderr, " %.*s", 8, sha1_to_hex(commit->object.sha1));
		for (pp = commit->parents; pp; pp = pp->next)
			fprintf(stderr, " %.*s", 8,
				sha1_to_hex(pp->item->object.sha1));

152 153 154
		subject_len = find_commit_subject(buf, &subject_start);
		if (subject_len)
			fprintf(stderr, " %.*s", subject_len, subject_start);
155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220
		fprintf(stderr, "\n");
	}
}
#endif /* DEBUG_BISECT */

static struct commit_list *best_bisection(struct commit_list *list, int nr)
{
	struct commit_list *p, *best;
	int best_distance = -1;

	best = list;
	for (p = list; p; p = p->next) {
		int distance;
		unsigned flags = p->item->object.flags;

		if (flags & TREESAME)
			continue;
		distance = weight(p);
		if (nr - distance < distance)
			distance = nr - distance;
		if (distance > best_distance) {
			best = p;
			best_distance = distance;
		}
	}

	return best;
}

struct commit_dist {
	struct commit *commit;
	int distance;
};

static int compare_commit_dist(const void *a_, const void *b_)
{
	struct commit_dist *a, *b;

	a = (struct commit_dist *)a_;
	b = (struct commit_dist *)b_;
	if (a->distance != b->distance)
		return b->distance - a->distance; /* desc sort */
	return hashcmp(a->commit->object.sha1, b->commit->object.sha1);
}

static struct commit_list *best_bisection_sorted(struct commit_list *list, int nr)
{
	struct commit_list *p;
	struct commit_dist *array = xcalloc(nr, sizeof(*array));
	int cnt, i;

	for (p = list, cnt = 0; p; p = p->next) {
		int distance;
		unsigned flags = p->item->object.flags;

		if (flags & TREESAME)
			continue;
		distance = weight(p);
		if (nr - distance < distance)
			distance = nr - distance;
		array[cnt].commit = p->item;
		array[cnt].distance = distance;
		cnt++;
	}
	qsort(array, cnt, sizeof(*array), compare_commit_dist);
	for (p = list, i = 0; i < cnt; i++) {
221
		char buf[100]; /* enough for dist=%d */
222 223
		struct object *obj = &(array[i].commit->object);

224 225 226
		snprintf(buf, sizeof(buf), "dist=%d", array[i].distance);
		add_name_decoration(DECORATION_NONE, buf, obj);

227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405
		p->item = array[i].commit;
		p = p->next;
	}
	if (p)
		p->next = NULL;
	free(array);
	return list;
}

/*
 * zero or positive weight is the number of interesting commits it can
 * reach, including itself.  Especially, weight = 0 means it does not
 * reach any tree-changing commits (e.g. just above uninteresting one
 * but traversal is with pathspec).
 *
 * weight = -1 means it has one parent and its distance is yet to
 * be computed.
 *
 * weight = -2 means it has more than one parent and its distance is
 * unknown.  After running count_distance() first, they will get zero
 * or positive distance.
 */
static struct commit_list *do_find_bisection(struct commit_list *list,
					     int nr, int *weights,
					     int find_all)
{
	int n, counted;
	struct commit_list *p;

	counted = 0;

	for (n = 0, p = list; p; p = p->next) {
		struct commit *commit = p->item;
		unsigned flags = commit->object.flags;

		p->item->util = &weights[n++];
		switch (count_interesting_parents(commit)) {
		case 0:
			if (!(flags & TREESAME)) {
				weight_set(p, 1);
				counted++;
				show_list("bisection 2 count one",
					  counted, nr, list);
			}
			/*
			 * otherwise, it is known not to reach any
			 * tree-changing commit and gets weight 0.
			 */
			break;
		case 1:
			weight_set(p, -1);
			break;
		default:
			weight_set(p, -2);
			break;
		}
	}

	show_list("bisection 2 initialize", counted, nr, list);

	/*
	 * If you have only one parent in the resulting set
	 * then you can reach one commit more than that parent
	 * can reach.  So we do not have to run the expensive
	 * count_distance() for single strand of pearls.
	 *
	 * However, if you have more than one parents, you cannot
	 * just add their distance and one for yourself, since
	 * they usually reach the same ancestor and you would
	 * end up counting them twice that way.
	 *
	 * So we will first count distance of merges the usual
	 * way, and then fill the blanks using cheaper algorithm.
	 */
	for (p = list; p; p = p->next) {
		if (p->item->object.flags & UNINTERESTING)
			continue;
		if (weight(p) != -2)
			continue;
		weight_set(p, count_distance(p));
		clear_distance(list);

		/* Does it happen to be at exactly half-way? */
		if (!find_all && halfway(p, nr))
			return p;
		counted++;
	}

	show_list("bisection 2 count_distance", counted, nr, list);

	while (counted < nr) {
		for (p = list; p; p = p->next) {
			struct commit_list *q;
			unsigned flags = p->item->object.flags;

			if (0 <= weight(p))
				continue;
			for (q = p->item->parents; q; q = q->next) {
				if (q->item->object.flags & UNINTERESTING)
					continue;
				if (0 <= weight(q))
					break;
			}
			if (!q)
				continue;

			/*
			 * weight for p is unknown but q is known.
			 * add one for p itself if p is to be counted,
			 * otherwise inherit it from q directly.
			 */
			if (!(flags & TREESAME)) {
				weight_set(p, weight(q)+1);
				counted++;
				show_list("bisection 2 count one",
					  counted, nr, list);
			}
			else
				weight_set(p, weight(q));

			/* Does it happen to be at exactly half-way? */
			if (!find_all && halfway(p, nr))
				return p;
		}
	}

	show_list("bisection 2 counted all", counted, nr, list);

	if (!find_all)
		return best_bisection(list, nr);
	else
		return best_bisection_sorted(list, nr);
}

struct commit_list *find_bisection(struct commit_list *list,
					  int *reaches, int *all,
					  int find_all)
{
	int nr, on_list;
	struct commit_list *p, *best, *next, *last;
	int *weights;

	show_list("bisection 2 entry", 0, 0, list);

	/*
	 * Count the number of total and tree-changing items on the
	 * list, while reversing the list.
	 */
	for (nr = on_list = 0, last = NULL, p = list;
	     p;
	     p = next) {
		unsigned flags = p->item->object.flags;

		next = p->next;
		if (flags & UNINTERESTING)
			continue;
		p->next = last;
		last = p;
		if (!(flags & TREESAME))
			nr++;
		on_list++;
	}
	list = last;
	show_list("bisection 2 sorted", 0, nr, list);

	*all = nr;
	weights = xcalloc(on_list, sizeof(*weights));

	/* Do the real work of finding bisection commit. */
	best = do_find_bisection(list, nr, weights, find_all);
	if (best) {
		if (!find_all)
			best->next = NULL;
		*reaches = weight(best);
	}
	free(weights);
	return best;
}

406
static int register_ref(const char *refname, const struct object_id *oid,
407 408
			int flags, void *cb_data)
{
409 410 411 412 413
	struct strbuf good_prefix = STRBUF_INIT;
	strbuf_addstr(&good_prefix, term_good);
	strbuf_addstr(&good_prefix, "-");

	if (!strcmp(refname, term_bad)) {
414
		current_bad_oid = xmalloc(sizeof(*current_bad_oid));
415
		oidcpy(current_bad_oid, oid);
416
	} else if (starts_with(refname, good_prefix.buf)) {
417
		sha1_array_append(&good_revs, oid->hash);
418
	} else if (starts_with(refname, "skip-")) {
419
		sha1_array_append(&skipped_revs, oid->hash);
420 421
	}

422 423
	strbuf_release(&good_prefix);

424 425 426 427 428
	return 0;
}

static int read_bisect_refs(void)
{
429
	return for_each_ref_in("refs/bisect/", register_ref, NULL);
430 431
}

432
static void read_bisect_paths(struct argv_array *array)
433 434 435 436 437 438
{
	struct strbuf str = STRBUF_INIT;
	const char *filename = git_path("BISECT_NAMES");
	FILE *fp = fopen(filename, "r");

	if (!fp)
439
		die_errno("Could not open file '%s'", filename);
440 441 442

	while (strbuf_getline(&str, fp, '\n') != EOF) {
		strbuf_trim(&str);
Jeff King's avatar
Jeff King committed
443
		if (sq_dequote_to_argv_array(str.buf, array))
444
			die("Badly quoted content in file '%s': %s",
Jeff King's avatar
Jeff King committed
445
			    filename, str.buf);
446 447 448 449 450 451
	}

	strbuf_release(&str);
	fclose(fp);
}

452 453 454 455 456
static char *join_sha1_array_hex(struct sha1_array *array, char delim)
{
	struct strbuf joined_hexs = STRBUF_INIT;
	int i;

457
	for (i = 0; i < array->nr; i++) {
458
		strbuf_addstr(&joined_hexs, sha1_to_hex(array->sha1[i]));
459
		if (i + 1 < array->nr)
460 461 462 463 464 465
			strbuf_addch(&joined_hexs, delim);
	}

	return strbuf_detach(&joined_hexs, NULL);
}

466 467 468 469 470 471 472 473 474 475 476 477 478
/*
 * In this function, passing a not NULL skipped_first is very special.
 * It means that we want to know if the first commit in the list is
 * skipped because we will want to test a commit away from it if it is
 * indeed skipped.
 * So if the first commit is skipped, we cannot take the shortcut to
 * just "return list" when we find the first non skipped commit, we
 * have to return a fully filtered list.
 *
 * We use (*skipped_first == -1) to mean "it has been found that the
 * first commit is not skipped". In this case *skipped_first is set back
 * to 0 just before the function returns.
 */
479 480
struct commit_list *filter_skipped(struct commit_list *list,
				   struct commit_list **tried,
481 482 483
				   int show_all,
				   int *count,
				   int *skipped_first)
484 485 486 487 488
{
	struct commit_list *filtered = NULL, **f = &filtered;

	*tried = NULL;

489 490 491 492 493
	if (skipped_first)
		*skipped_first = 0;
	if (count)
		*count = 0;

494
	if (!skipped_revs.nr)
495 496 497 498 499
		return list;

	while (list) {
		struct commit_list *next = list->next;
		list->next = NULL;
500
		if (0 <= sha1_array_lookup(&skipped_revs,
501
					   list->item->object.sha1)) {
502 503
			if (skipped_first && !*skipped_first)
				*skipped_first = 1;
504 505 506 507
			/* Move current to tried list */
			*tried = list;
			tried = &list->next;
		} else {
508 509 510 511 512 513 514
			if (!show_all) {
				if (!skipped_first || !*skipped_first)
					return list;
			} else if (skipped_first && !*skipped_first) {
				/* This means we know it's not skipped */
				*skipped_first = -1;
			}
515 516 517
			/* Move current to filtered list */
			*f = list;
			f = &list->next;
518 519
			if (count)
				(*count)++;
520 521 522 523
		}
		list = next;
	}

524 525 526
	if (skipped_first && *skipped_first == -1)
		*skipped_first = 0;

527 528
	return filtered;
}
529

530 531 532 533 534 535 536 537
#define PRN_MODULO 32768

/*
 * This is a pseudo random number generator based on "man 3 rand".
 * It is not used properly because the seed is the argument and it
 * is increased by one between each call, but that should not matter
 * for this application.
 */
538
static unsigned get_prn(unsigned count) {
539
	count = count * 1103515245 + 12345;
540
	return (count/65536) % PRN_MODULO;
541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563
}

/*
 * Custom integer square root from
 * http://en.wikipedia.org/wiki/Integer_square_root
 */
static int sqrti(int val)
{
	float d, x = val;

	if (val == 0)
		return 0;

	do {
		float y = (x + (float)val / x) / 2;
		d = (y > x) ? y - x : x - y;
		x = y;
	} while (d >= 0.5);

	return (int)x;
}

static struct commit_list *skip_away(struct commit_list *list, int count)
564 565
{
	struct commit_list *cur, *previous;
566 567 568 569
	int prn, index, i;

	prn = get_prn(count);
	index = (count * prn / PRN_MODULO) * sqrti(prn) / sqrti(PRN_MODULO);
570 571 572 573 574 575

	cur = list;
	previous = NULL;

	for (i = 0; cur; cur = cur->next, i++) {
		if (i == index) {
576
			if (hashcmp(cur->item->object.sha1, current_bad_oid->hash))
577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594
				return cur;
			if (previous)
				return previous;
			return list;
		}
		previous = cur;
	}

	return list;
}

static struct commit_list *managed_skipped(struct commit_list *list,
					   struct commit_list **tried)
{
	int count, skipped_first;

	*tried = NULL;

595
	if (!skipped_revs.nr)
596 597 598 599 600 601 602
		return list;

	list = filter_skipped(list, tried, 0, &count, &skipped_first);

	if (!skipped_first)
		return list;

603
	return skip_away(list, count);
604 605
}

606 607 608
static void bisect_rev_setup(struct rev_info *revs, const char *prefix,
			     const char *bad_format, const char *good_format,
			     int read_paths)
609
{
Jeff King's avatar
Jeff King committed
610
	struct argv_array rev_argv = ARGV_ARRAY_INIT;
611 612
	int i;

613 614 615
	init_revisions(revs, prefix);
	revs->abbrev = 0;
	revs->commit_format = CMIT_FMT_UNSPECIFIED;
616

617
	/* rev_argv.argv[0] will be ignored by setup_revisions */
Jeff King's avatar
Jeff King committed
618
	argv_array_push(&rev_argv, "bisect_rev_setup");
619
	argv_array_pushf(&rev_argv, bad_format, oid_to_hex(current_bad_oid));
620
	for (i = 0; i < good_revs.nr; i++)
Jeff King's avatar
Jeff King committed
621 622 623
		argv_array_pushf(&rev_argv, good_format,
				 sha1_to_hex(good_revs.sha1[i]));
	argv_array_push(&rev_argv, "--");
624 625
	if (read_paths)
		read_bisect_paths(&rev_argv);
626

Jeff King's avatar
Jeff King committed
627 628
	setup_revisions(rev_argv.argc, rev_argv.argv, revs, NULL);
	/* XXX leak rev_argv, as "revs" may still be pointing to it */
629 630
}

631
static void bisect_common(struct rev_info *revs)
632 633 634 635
{
	if (prepare_revision_walk(revs))
		die("revision walk setup failed");
	if (revs->tree_objects)
636
		mark_edges_uninteresting(revs, NULL);
637 638
}

639
static void exit_if_skipped_commits(struct commit_list *tried,
640
				    const struct object_id *bad)
641 642 643 644 645
{
	if (!tried)
		return;

	printf("There are only 'skip'ped commits left to test.\n"
646
	       "The first %s commit could be any of:\n", term_bad);
647 648
	print_commit_list(tried, "%s\n", "%s\n");
	if (bad)
649
		printf("%s\n", oid_to_hex(bad));
650 651 652 653
	printf("We cannot bisect more!\n");
	exit(2);
}

654
static int is_expected_rev(const struct object_id *oid)
655 656 657 658 659 660 661 662 663 664 665 666 667 668 669
{
	const char *filename = git_path("BISECT_EXPECTED_REV");
	struct stat st;
	struct strbuf str = STRBUF_INIT;
	FILE *fp;
	int res = 0;

	if (stat(filename, &st) || !S_ISREG(st.st_mode))
		return 0;

	fp = fopen(filename, "r");
	if (!fp)
		return 0;

	if (strbuf_getline(&str, fp, '\n') != EOF)
670
		res = !strcmp(str.buf, oid_to_hex(oid));
671 672 673 674 675 676 677

	strbuf_release(&str);
	fclose(fp);

	return res;
}

678 679 680 681 682 683 684
static void mark_expected_rev(char *bisect_rev_hex)
{
	int len = strlen(bisect_rev_hex);
	const char *filename = git_path("BISECT_EXPECTED_REV");
	int fd = open(filename, O_CREAT | O_TRUNC | O_WRONLY, 0600);

	if (fd < 0)
685
		die_errno("could not create file '%s'", filename);
686 687 688 689 690 691 692 693 694

	bisect_rev_hex[len] = '\n';
	write_or_die(fd, bisect_rev_hex, len + 1);
	bisect_rev_hex[len] = '\0';

	if (close(fd) < 0)
		die("closing file %s: %s", filename, strerror(errno));
}

695
static int bisect_checkout(char *bisect_rev_hex, int no_checkout)
696 697 698 699 700
{

	mark_expected_rev(bisect_rev_hex);

	argv_checkout[2] = bisect_rev_hex;
701 702 703 704 705 706
	if (no_checkout) {
		argv_update_ref[3] = bisect_rev_hex;
		if (run_command_v_opt(argv_update_ref, RUN_GIT_CMD))
			die("update-ref --no-deref HEAD failed on %s",
			    bisect_rev_hex);
	} else {
707
		int res;
708 709 710 711
		res = run_command_v_opt(argv_checkout, RUN_GIT_CMD);
		if (res)
			exit(res);
	}
712 713 714 715 716

	argv_show_branch[1] = bisect_rev_hex;
	return run_command_v_opt(argv_show_branch, RUN_GIT_CMD);
}

717 718 719 720 721 722 723 724 725 726
static struct commit *get_commit_reference(const unsigned char *sha1)
{
	struct commit *r = lookup_commit_reference(sha1);
	if (!r)
		die("Not a valid commit name %s", sha1_to_hex(sha1));
	return r;
}

static struct commit **get_bad_and_good_commits(int *rev_nr)
{
727
	int len = 1 + good_revs.nr;
728 729 730
	struct commit **rev = xmalloc(len * sizeof(*rev));
	int i, n = 0;

731
	rev[n++] = get_commit_reference(current_bad_oid->hash);
732
	for (i = 0; i < good_revs.nr; i++)
733 734 735 736 737 738 739 740
		rev[n++] = get_commit_reference(good_revs.sha1[i]);
	*rev_nr = n;

	return rev;
}

static void handle_bad_merge_base(void)
{
741 742
	if (is_expected_rev(current_bad_oid)) {
		char *bad_hex = oid_to_hex(current_bad_oid);
743
		char *good_hex = join_sha1_array_hex(&good_revs, ' ');
744 745 746 747 748 749 750 751 752 753 754
		if (!strcmp(term_bad, "bad") && !strcmp(term_good, "good")) {
			fprintf(stderr, "The merge base %s is bad.\n"
				"This means the bug has been fixed "
				"between %s and [%s].\n",
				bad_hex, bad_hex, good_hex);
		} else {
			fprintf(stderr, "The merge base %s is %s.\n"
				"This means the first '%s' commit is "
				"between %s and [%s].\n",
				bad_hex, term_bad, term_good, bad_hex, good_hex);
		}
755 756 757
		exit(3);
	}

758
	fprintf(stderr, "Some %s revs are not ancestor of the %s rev.\n"
759
		"git bisect cannot work properly in this case.\n"
760 761
		"Maybe you mistook %s and %s revs?\n",
		term_good, term_bad, term_good, term_bad);
762 763 764
	exit(1);
}

765
static void handle_skipped_merge_base(const unsigned char *mb)
766 767
{
	char *mb_hex = sha1_to_hex(mb);
768
	char *bad_hex = sha1_to_hex(current_bad_oid->hash);
769 770
	char *good_hex = join_sha1_array_hex(&good_revs, ' ');

771
	warning("the merge base between %s and [%s] "
772
		"must be skipped.\n"
773
		"So we cannot be sure the first %s commit is "
774
		"between %s and %s.\n"
775
		"We continue anyway.",
776
		bad_hex, good_hex, term_bad, mb_hex, bad_hex);
777 778 779 780 781 782 783 784 785 786 787 788
	free(good_hex);
}

/*
 * "check_merge_bases" checks that merge bases are not "bad".
 *
 * - If one is "bad", it means the user assumed something wrong
 * and we must exit with a non 0 error code.
 * - If one is "good", that's good, we have nothing to do.
 * - If one is "skipped", we can't know but we should warn.
 * - If we don't know, we should check it out and ask the user to test.
 */
789
static void check_merge_bases(int no_checkout)
790 791 792 793 794
{
	struct commit_list *result;
	int rev_nr;
	struct commit **rev = get_bad_and_good_commits(&rev_nr);

795
	result = get_merge_bases_many(rev[0], rev_nr - 1, rev + 1);
796 797 798

	for (; result; result = result->next) {
		const unsigned char *mb = result->item->object.sha1;
799
		if (!hashcmp(mb, current_bad_oid->hash)) {
800
			handle_bad_merge_base();
801
		} else if (0 <= sha1_array_lookup(&good_revs, mb)) {
802
			continue;
803
		} else if (0 <= sha1_array_lookup(&skipped_revs, mb)) {
804 805 806
			handle_skipped_merge_base(mb);
		} else {
			printf("Bisecting: a merge base must be tested\n");
807
			exit(bisect_checkout(sha1_to_hex(mb), no_checkout));
808 809 810 811 812 813 814
		}
	}

	free(rev);
	free_commit_list(result);
}

815
static int check_ancestors(const char *prefix)
816
{
817 818
	struct rev_info revs;
	struct object_array pending_copy;
819
	int res;
820

821
	bisect_rev_setup(&revs, prefix, "^%s", "%s", 0);
822

823
	/* Save pending objects, so they can be cleaned up later. */
824 825
	pending_copy = revs.pending;
	revs.leak_pending = 1;
826

827 828 829 830 831
	/*
	 * bisect_common calls prepare_revision_walk right away, which
	 * (together with .leak_pending = 1) makes us the sole owner of
	 * the list of pending objects.
	 */
832 833 834 835
	bisect_common(&revs);
	res = (revs.commits != NULL);

	/* Clean up objects used, as they will be reused. */
836
	clear_commit_marks_for_object_array(&pending_copy, ALL_REV_FLAGS);
837
	free(pending_copy.objects);
838

839
	return res;
840 841 842 843 844 845 846 847 848 849
}

/*
 * "check_good_are_ancestors_of_bad" checks that all "good" revs are
 * ancestor of the "bad" rev.
 *
 * If that's not the case, we need to check the merge bases.
 * If a merge base must be tested by the user, its source code will be
 * checked out to be tested by the user and we will exit.
 */
850
static void check_good_are_ancestors_of_bad(const char *prefix, int no_checkout)
851
{
852
	char *filename = git_pathdup("BISECT_ANCESTORS_OK");
853 854 855
	struct stat st;
	int fd;

856
	if (!current_bad_oid)
857
		die("a %s revision is needed", term_bad);
858 859 860

	/* Check if file BISECT_ANCESTORS_OK exists. */
	if (!stat(filename, &st) && S_ISREG(st.st_mode))
861
		goto done;
862 863

	/* Bisecting with no good rev is ok. */
864
	if (good_revs.nr == 0)
865
		goto done;
866

867 868
	/* Check if all good revs are ancestor of the bad rev. */
	if (check_ancestors(prefix))
869
		check_merge_bases(no_checkout);
870 871 872 873 874 875 876 877

	/* Create file BISECT_ANCESTORS_OK. */
	fd = open(filename, O_CREAT | O_TRUNC | O_WRONLY, 0600);
	if (fd < 0)
		warning("could not create file '%s': %s",
			filename, strerror(errno));
	else
		close(fd);
878 879
 done:
	free(filename);
880 881
}

882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906
/*
 * This does "git diff-tree --pretty COMMIT" without one fork+exec.
 */
static void show_diff_tree(const char *prefix, struct commit *commit)
{
	struct rev_info opt;

	/* diff-tree init */
	init_revisions(&opt, prefix);
	git_config(git_diff_basic_config, NULL); /* no "diff" UI options */
	opt.abbrev = 0;
	opt.diff = 1;

	/* This is what "--pretty" does */
	opt.verbose_header = 1;
	opt.use_terminator = 0;
	opt.commit_format = CMIT_FMT_DEFAULT;

	/* diff-tree init */
	if (!opt.diffopt.output_format)
		opt.diffopt.output_format = DIFF_FORMAT_RAW;

	log_tree_commit(&opt, commit);
}

907 908 909 910
/*
 * We use the convention that exiting with an exit code 10 means that
 * the bisection process finished successfully.
 * In this case the calling shell script should exit 0.
911 912 913
 *
 * If no_checkout is non-zero, the bisection process does not
 * checkout the trial commit but instead simply updates BISECT_HEAD.
914
 */
915
int bisect_next_all(const char *prefix, int no_checkout)
916 917 918
{
	struct rev_info revs;
	struct commit_list *tried;
919
	int reaches = 0, all = 0, nr, steps;
920
	const unsigned char *bisect_rev;
921
	char bisect_rev_hex[GIT_SHA1_HEXSZ + 1];
922

923 924
	term_bad = "bad";
	term_good = "good";
925 926 927
	if (read_bisect_refs())
		die("reading bisect refs failed");

928
	check_good_are_ancestors_of_bad(prefix, no_checkout);
929

930 931
	bisect_rev_setup(&revs, prefix, "%s", "^%s", 1);
	revs.limited = 1;
932

933
	bisect_common(&revs);
934

935
	revs.commits = find_bisection(revs.commits, &reaches, &all,
936
				       !!skipped_revs.nr);
937
	revs.commits = managed_skipped(revs.commits, &tried);
938 939 940 941 942 943 944 945

	if (!revs.commits) {
		/*
		 * We should exit here only if the "bad"
		 * commit is also a "skip" commit.
		 */
		exit_if_skipped_commits(tried, NULL);

946 947 948 949
		printf("%s was both %s and %s\n",
		       oid_to_hex(current_bad_oid),
		       term_good,
		       term_bad);
950 951 952
		exit(1);
	}

953 954 955 956 957 958
	if (!all) {
		fprintf(stderr, "No testable commit found.\n"
			"Maybe you started with bad path parameters?\n");
		exit(4);
	}

959
	bisect_rev = revs.commits->item->object.sha1;
960
	memcpy(bisect_rev_hex, sha1_to_hex(bisect_rev), GIT_SHA1_HEXSZ + 1);
961

962 963
	if (!hashcmp(bisect_rev, current_bad_oid->hash)) {
		exit_if_skipped_commits(tried, current_bad_oid);
964 965
		printf("%s is the first %s commit\n", bisect_rev_hex,
			term_bad);
966
		show_diff_tree(prefix, revs.commits->item);
967 968 969 970 971
		/* This means the bisection process succeeded. */
		exit(10);
	}

	nr = all - reaches - 1;
972 973 974 975
	steps = estimate_bisect_steps(all);
	printf("Bisecting: %d revision%s left to test after this "
	       "(roughly %d step%s)\n", nr, (nr == 1 ? "" : "s"),
	       steps, (steps == 1 ? "" : "s"));
976

977
	return bisect_checkout(bisect_rev_hex, no_checkout);
978 979
}

980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017
static inline int log2i(int n)
{
	int log2 = 0;

	for (; n > 1; n >>= 1)
		log2++;

	return log2;
}

static inline int exp2i(int n)
{
	return 1 << n;
}

/*
 * Estimate the number of bisect steps left (after the current step)
 *
 * For any x between 0 included and 2^n excluded, the probability for
 * n - 1 steps left looks like:
 *
 * P(2^n + x) == (2^n - x) / (2^n + x)
 *
 * and P(2^n + x) < 0.5 means 2^n < 3x
 */
int estimate_bisect_steps(int all)
{
	int n, x, e;

	if (all < 3)
		return 0;

	n = log2i(all);
	e = exp2i(n);
	x = all - e;

	return (e < 3 * x) ? n : n - 1;
}