bisect.c 23.5 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 unsigned char *current_bad_sha1;
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 27 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
/* bits #0-15 in revision.h */

#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);
134 135
		const char *subject_start;
		int subject_len;
136 137 138 139 140 141 142 143 144 145 146 147 148 149

		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));

150 151 152
		subject_len = find_commit_subject(buf, &subject_start);
		if (subject_len)
			fprintf(stderr, " %.*s", subject_len, subject_start);
153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402
		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++) {
		struct name_decoration *r = xmalloc(sizeof(*r) + 100);
		struct object *obj = &(array[i].commit->object);

		sprintf(r->name, "dist=%d", array[i].distance);
		r->next = add_decoration(&name_decoration, obj, r);
		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;
}

403 404 405 406
static int register_ref(const char *refname, const unsigned char *sha1,
			int flags, void *cb_data)
{
	if (!strcmp(refname, "bad")) {
407 408
		current_bad_sha1 = xmalloc(20);
		hashcpy(current_bad_sha1, sha1);
409
	} else if (starts_with(refname, "good-")) {
410
		sha1_array_append(&good_revs, sha1);
411
	} else if (starts_with(refname, "skip-")) {
412
		sha1_array_append(&skipped_revs, sha1);
413 414 415 416 417 418 419 420 421 422
	}

	return 0;
}

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

423
static void read_bisect_paths(struct argv_array *array)
424 425 426 427 428 429
{
	struct strbuf str = STRBUF_INIT;
	const char *filename = git_path("BISECT_NAMES");
	FILE *fp = fopen(filename, "r");

	if (!fp)
430
		die_errno("Could not open file '%s'", filename);
431 432 433

	while (strbuf_getline(&str, fp, '\n') != EOF) {
		strbuf_trim(&str);
Jeff King's avatar
Jeff King committed
434
		if (sq_dequote_to_argv_array(str.buf, array))
435
			die("Badly quoted content in file '%s': %s",
Jeff King's avatar
Jeff King committed
436
			    filename, str.buf);
437 438 439 440 441 442
	}

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

443 444 445 446 447
static char *join_sha1_array_hex(struct sha1_array *array, char delim)
{
	struct strbuf joined_hexs = STRBUF_INIT;
	int i;

448
	for (i = 0; i < array->nr; i++) {
449
		strbuf_addstr(&joined_hexs, sha1_to_hex(array->sha1[i]));
450
		if (i + 1 < array->nr)
451 452 453 454 455 456
			strbuf_addch(&joined_hexs, delim);
	}

	return strbuf_detach(&joined_hexs, NULL);
}

457 458 459 460 461 462 463 464 465 466 467 468 469
/*
 * 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.
 */
470 471
struct commit_list *filter_skipped(struct commit_list *list,
				   struct commit_list **tried,
472 473 474
				   int show_all,
				   int *count,
				   int *skipped_first)
475 476 477 478 479
{
	struct commit_list *filtered = NULL, **f = &filtered;

	*tried = NULL;

480 481 482 483 484
	if (skipped_first)
		*skipped_first = 0;
	if (count)
		*count = 0;

485
	if (!skipped_revs.nr)
486 487 488 489 490
		return list;

	while (list) {
		struct commit_list *next = list->next;
		list->next = NULL;
491
		if (0 <= sha1_array_lookup(&skipped_revs,
492
					   list->item->object.sha1)) {
493 494
			if (skipped_first && !*skipped_first)
				*skipped_first = 1;
495 496 497 498
			/* Move current to tried list */
			*tried = list;
			tried = &list->next;
		} else {
499 500 501 502 503 504 505
			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;
			}
506 507 508
			/* Move current to filtered list */
			*f = list;
			f = &list->next;
509 510
			if (count)
				(*count)++;
511 512 513 514
		}
		list = next;
	}

515 516 517
	if (skipped_first && *skipped_first == -1)
		*skipped_first = 0;

518 519
	return filtered;
}
520

521 522 523 524 525 526 527 528
#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.
 */
529
static unsigned get_prn(unsigned count) {
530
	count = count * 1103515245 + 12345;
531
	return (count/65536) % PRN_MODULO;
532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554
}

/*
 * 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)
555 556
{
	struct commit_list *cur, *previous;
557 558 559 560
	int prn, index, i;

	prn = get_prn(count);
	index = (count * prn / PRN_MODULO) * sqrti(prn) / sqrti(PRN_MODULO);
561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585

	cur = list;
	previous = NULL;

	for (i = 0; cur; cur = cur->next, i++) {
		if (i == index) {
			if (hashcmp(cur->item->object.sha1, current_bad_sha1))
				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;

586
	if (!skipped_revs.nr)
587 588 589 590 591 592 593
		return list;

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

	if (!skipped_first)
		return list;

594
	return skip_away(list, count);
595 596
}

597 598 599
static void bisect_rev_setup(struct rev_info *revs, const char *prefix,
			     const char *bad_format, const char *good_format,
			     int read_paths)
600
{
Jeff King's avatar
Jeff King committed
601
	struct argv_array rev_argv = ARGV_ARRAY_INIT;
602 603
	int i;

604 605 606
	init_revisions(revs, prefix);
	revs->abbrev = 0;
	revs->commit_format = CMIT_FMT_UNSPECIFIED;
607

608
	/* rev_argv.argv[0] will be ignored by setup_revisions */
Jeff King's avatar
Jeff King committed
609 610
	argv_array_push(&rev_argv, "bisect_rev_setup");
	argv_array_pushf(&rev_argv, bad_format, sha1_to_hex(current_bad_sha1));
611
	for (i = 0; i < good_revs.nr; i++)
Jeff King's avatar
Jeff King committed
612 613 614
		argv_array_pushf(&rev_argv, good_format,
				 sha1_to_hex(good_revs.sha1[i]));
	argv_array_push(&rev_argv, "--");
615 616
	if (read_paths)
		read_bisect_paths(&rev_argv);
617

Jeff King's avatar
Jeff King committed
618 619
	setup_revisions(rev_argv.argc, rev_argv.argv, revs, NULL);
	/* XXX leak rev_argv, as "revs" may still be pointing to it */
620 621
}

622
static void bisect_common(struct rev_info *revs)
623 624 625 626
{
	if (prepare_revision_walk(revs))
		die("revision walk setup failed");
	if (revs->tree_objects)
627
		mark_edges_uninteresting(revs, NULL);
628 629
}

630 631 632 633 634 635 636 637 638 639 640 641 642 643 644
static void exit_if_skipped_commits(struct commit_list *tried,
				    const unsigned char *bad)
{
	if (!tried)
		return;

	printf("There are only 'skip'ped commits left to test.\n"
	       "The first bad commit could be any of:\n");
	print_commit_list(tried, "%s\n", "%s\n");
	if (bad)
		printf("%s\n", sha1_to_hex(bad));
	printf("We cannot bisect more!\n");
	exit(2);
}

645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668
static int is_expected_rev(const unsigned char *sha1)
{
	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)
		res = !strcmp(str.buf, sha1_to_hex(sha1));

	strbuf_release(&str);
	fclose(fp);

	return res;
}

669 670 671 672 673 674 675
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)
676
		die_errno("could not create file '%s'", filename);
677 678 679 680 681 682 683 684 685

	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));
}

686
static int bisect_checkout(char *bisect_rev_hex, int no_checkout)
687 688 689 690 691 692
{
	int res;

	mark_expected_rev(bisect_rev_hex);

	argv_checkout[2] = bisect_rev_hex;
693 694 695 696 697 698 699 700 701 702
	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 {
		res = run_command_v_opt(argv_checkout, RUN_GIT_CMD);
		if (res)
			exit(res);
	}
703 704 705 706 707

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

708 709 710 711 712 713 714 715 716 717
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)
{
718
	int len = 1 + good_revs.nr;
719 720 721 722
	struct commit **rev = xmalloc(len * sizeof(*rev));
	int i, n = 0;

	rev[n++] = get_commit_reference(current_bad_sha1);
723
	for (i = 0; i < good_revs.nr; i++)
724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749
		rev[n++] = get_commit_reference(good_revs.sha1[i]);
	*rev_nr = n;

	return rev;
}

static void handle_bad_merge_base(void)
{
	if (is_expected_rev(current_bad_sha1)) {
		char *bad_hex = sha1_to_hex(current_bad_sha1);
		char *good_hex = join_sha1_array_hex(&good_revs, ' ');

		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);

		exit(3);
	}

	fprintf(stderr, "Some good revs are not ancestor of the bad rev.\n"
		"git bisect cannot work properly in this case.\n"
		"Maybe you mistake good and bad revs?\n");
	exit(1);
}

750
static void handle_skipped_merge_base(const unsigned char *mb)
751 752 753 754 755
{
	char *mb_hex = sha1_to_hex(mb);
	char *bad_hex = sha1_to_hex(current_bad_sha1);
	char *good_hex = join_sha1_array_hex(&good_revs, ' ');

756
	warning("the merge base between %s and [%s] "
757 758 759
		"must be skipped.\n"
		"So we cannot be sure the first bad commit is "
		"between %s and %s.\n"
760
		"We continue anyway.",
761 762 763 764 765 766 767 768 769 770 771 772 773
		bad_hex, good_hex, mb_hex, bad_hex);
	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.
 */
774
static void check_merge_bases(int no_checkout)
775 776 777 778 779 780 781 782 783 784 785
{
	struct commit_list *result;
	int rev_nr;
	struct commit **rev = get_bad_and_good_commits(&rev_nr);

	result = get_merge_bases_many(rev[0], rev_nr - 1, rev + 1, 0);

	for (; result; result = result->next) {
		const unsigned char *mb = result->item->object.sha1;
		if (!hashcmp(mb, current_bad_sha1)) {
			handle_bad_merge_base();
786
		} else if (0 <= sha1_array_lookup(&good_revs, mb)) {
787
			continue;
788
		} else if (0 <= sha1_array_lookup(&skipped_revs, mb)) {
789 790 791
			handle_skipped_merge_base(mb);
		} else {
			printf("Bisecting: a merge base must be tested\n");
792
			exit(bisect_checkout(sha1_to_hex(mb), no_checkout));
793 794 795 796 797 798 799
		}
	}

	free(rev);
	free_commit_list(result);
}

800
static int check_ancestors(const char *prefix)
801
{
802 803
	struct rev_info revs;
	struct object_array pending_copy;
804
	int res;
805

806
	bisect_rev_setup(&revs, prefix, "^%s", "%s", 0);
807

808
	/* Save pending objects, so they can be cleaned up later. */
809 810
	pending_copy = revs.pending;
	revs.leak_pending = 1;
811

812 813 814 815 816
	/*
	 * 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.
	 */
817 818 819 820
	bisect_common(&revs);
	res = (revs.commits != NULL);

	/* Clean up objects used, as they will be reused. */
821
	clear_commit_marks_for_object_array(&pending_copy, ALL_REV_FLAGS);
822
	free(pending_copy.objects);
823

824
	return res;
825 826 827 828 829 830 831 832 833 834
}

/*
 * "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.
 */
835
static void check_good_are_ancestors_of_bad(const char *prefix, int no_checkout)
836
{
837
	char *filename = git_pathdup("BISECT_ANCESTORS_OK");
838 839 840 841 842 843 844 845
	struct stat st;
	int fd;

	if (!current_bad_sha1)
		die("a bad revision is needed");

	/* Check if file BISECT_ANCESTORS_OK exists. */
	if (!stat(filename, &st) && S_ISREG(st.st_mode))
846
		goto done;
847 848

	/* Bisecting with no good rev is ok. */
849
	if (good_revs.nr == 0)
850
		goto done;
851

852 853
	/* Check if all good revs are ancestor of the bad rev. */
	if (check_ancestors(prefix))
854
		check_merge_bases(no_checkout);
855 856 857 858 859 860 861 862

	/* 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);
863 864
 done:
	free(filename);
865 866
}

867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891
/*
 * 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);
}

892 893 894 895
/*
 * 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.
896 897 898
 *
 * If no_checkout is non-zero, the bisection process does not
 * checkout the trial commit but instead simply updates BISECT_HEAD.
899
 */
900
int bisect_next_all(const char *prefix, int no_checkout)
901 902 903
{
	struct rev_info revs;
	struct commit_list *tried;
904
	int reaches = 0, all = 0, nr, steps;
905 906 907
	const unsigned char *bisect_rev;
	char bisect_rev_hex[41];

908 909 910
	if (read_bisect_refs())
		die("reading bisect refs failed");

911
	check_good_are_ancestors_of_bad(prefix, no_checkout);
912

913 914
	bisect_rev_setup(&revs, prefix, "%s", "^%s", 1);
	revs.limited = 1;
915

916
	bisect_common(&revs);
917

918
	revs.commits = find_bisection(revs.commits, &reaches, &all,
919
				       !!skipped_revs.nr);
920
	revs.commits = managed_skipped(revs.commits, &tried);
921 922 923 924 925 926 927 928 929 930 931 932 933

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

		printf("%s was both good and bad\n",
		       sha1_to_hex(current_bad_sha1));
		exit(1);
	}

934 935 936 937 938 939
	if (!all) {
		fprintf(stderr, "No testable commit found.\n"
			"Maybe you started with bad path parameters?\n");
		exit(4);
	}

940 941 942 943 944
	bisect_rev = revs.commits->item->object.sha1;
	memcpy(bisect_rev_hex, sha1_to_hex(bisect_rev), 41);

	if (!hashcmp(bisect_rev, current_bad_sha1)) {
		exit_if_skipped_commits(tried, current_bad_sha1);
945
		printf("%s is the first bad commit\n", bisect_rev_hex);
946
		show_diff_tree(prefix, revs.commits->item);
947 948 949 950 951
		/* This means the bisection process succeeded. */
		exit(10);
	}

	nr = all - reaches - 1;
952 953 954 955
	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"));
956

957
	return bisect_checkout(bisect_rev_hex, no_checkout);
958 959
}

960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997
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;
}