Merge commit 'git/master'
[git/git-bigfiles.git] / builtin-rev-list.c
blob40d5fcb6b0b26c76c271624408b531cc01e15f7b
1 #include "cache.h"
2 #include "refs.h"
3 #include "tag.h"
4 #include "commit.h"
5 #include "tree.h"
6 #include "blob.h"
7 #include "tree-walk.h"
8 #include "diff.h"
9 #include "revision.h"
10 #include "list-objects.h"
11 #include "builtin.h"
12 #include "log-tree.h"
13 #include "graph.h"
15 /* bits #0-15 in revision.h */
17 #define COUNTED (1u<<16)
19 static const char rev_list_usage[] =
20 "git rev-list [OPTION] <commit-id>... [ -- paths... ]\n"
21 " limiting output:\n"
22 " --max-count=nr\n"
23 " --max-age=epoch\n"
24 " --min-age=epoch\n"
25 " --sparse\n"
26 " --no-merges\n"
27 " --remove-empty\n"
28 " --all\n"
29 " --branches\n"
30 " --tags\n"
31 " --remotes\n"
32 " --stdin\n"
33 " --quiet\n"
34 " ordering output:\n"
35 " --topo-order\n"
36 " --date-order\n"
37 " --reverse\n"
38 " formatting output:\n"
39 " --parents\n"
40 " --children\n"
41 " --objects | --objects-edge\n"
42 " --unpacked\n"
43 " --header | --pretty\n"
44 " --abbrev=nr | --no-abbrev\n"
45 " --abbrev-commit\n"
46 " --left-right\n"
47 " special purpose:\n"
48 " --bisect\n"
49 " --bisect-vars\n"
50 " --bisect-all"
53 static struct rev_info revs;
55 static int bisect_list;
56 static int show_timestamp;
57 static int hdr_termination;
58 static const char *header_prefix;
60 static void finish_commit(struct commit *commit);
61 static void show_commit(struct commit *commit)
63 graph_show_commit(revs.graph);
65 if (show_timestamp)
66 printf("%lu ", commit->date);
67 if (header_prefix)
68 fputs(header_prefix, stdout);
70 if (!revs.graph) {
71 if (commit->object.flags & BOUNDARY)
72 putchar('-');
73 else if (commit->object.flags & UNINTERESTING)
74 putchar('^');
75 else if (revs.left_right) {
76 if (commit->object.flags & SYMMETRIC_LEFT)
77 putchar('<');
78 else
79 putchar('>');
82 if (revs.abbrev_commit && revs.abbrev)
83 fputs(find_unique_abbrev(commit->object.sha1, revs.abbrev),
84 stdout);
85 else
86 fputs(sha1_to_hex(commit->object.sha1), stdout);
87 if (revs.print_parents) {
88 struct commit_list *parents = commit->parents;
89 while (parents) {
90 printf(" %s", sha1_to_hex(parents->item->object.sha1));
91 parents = parents->next;
94 if (revs.children.name) {
95 struct commit_list *children;
97 children = lookup_decoration(&revs.children, &commit->object);
98 while (children) {
99 printf(" %s", sha1_to_hex(children->item->object.sha1));
100 children = children->next;
103 show_decorations(&revs, commit);
104 if (revs.commit_format == CMIT_FMT_ONELINE)
105 putchar(' ');
106 else
107 putchar('\n');
109 if (revs.verbose_header && commit->buffer) {
110 struct strbuf buf = STRBUF_INIT;
111 pretty_print_commit(revs.commit_format, commit,
112 &buf, revs.abbrev, NULL, NULL,
113 revs.date_mode, 0);
114 if (revs.graph) {
115 if (buf.len) {
116 if (revs.commit_format != CMIT_FMT_ONELINE)
117 graph_show_oneline(revs.graph);
119 graph_show_commit_msg(revs.graph, &buf);
122 * Add a newline after the commit message.
124 * Usually, this newline produces a blank
125 * padding line between entries, in which case
126 * we need to add graph padding on this line.
128 * However, the commit message may not end in a
129 * newline. In this case the newline simply
130 * ends the last line of the commit message,
131 * and we don't need any graph output. (This
132 * always happens with CMIT_FMT_ONELINE, and it
133 * happens with CMIT_FMT_USERFORMAT when the
134 * format doesn't explicitly end in a newline.)
136 if (buf.len && buf.buf[buf.len - 1] == '\n')
137 graph_show_padding(revs.graph);
138 putchar('\n');
139 } else {
141 * If the message buffer is empty, just show
142 * the rest of the graph output for this
143 * commit.
145 if (graph_show_remainder(revs.graph))
146 putchar('\n');
148 } else {
149 if (buf.len)
150 printf("%s%c", buf.buf, hdr_termination);
152 strbuf_release(&buf);
153 } else {
154 if (graph_show_remainder(revs.graph))
155 putchar('\n');
157 maybe_flush_or_die(stdout, "stdout");
158 finish_commit(commit);
161 static void finish_commit(struct commit *commit)
163 if (commit->parents) {
164 free_commit_list(commit->parents);
165 commit->parents = NULL;
167 free(commit->buffer);
168 commit->buffer = NULL;
171 static void finish_object(struct object_array_entry *p)
173 if (p->item->type == OBJ_BLOB && !has_sha1_file(p->item->sha1))
174 die("missing blob object '%s'", sha1_to_hex(p->item->sha1));
177 static void show_object(struct object_array_entry *p)
179 /* An object with name "foo\n0000000..." can be used to
180 * confuse downstream "git pack-objects" very badly.
182 const char *ep = strchr(p->name, '\n');
184 finish_object(p);
185 if (ep) {
186 printf("%s %.*s\n", sha1_to_hex(p->item->sha1),
187 (int) (ep - p->name),
188 p->name);
190 else
191 printf("%s %s\n", sha1_to_hex(p->item->sha1), p->name);
194 static void show_edge(struct commit *commit)
196 printf("-%s\n", sha1_to_hex(commit->object.sha1));
200 * This is a truly stupid algorithm, but it's only
201 * used for bisection, and we just don't care enough.
203 * We care just barely enough to avoid recursing for
204 * non-merge entries.
206 static int count_distance(struct commit_list *entry)
208 int nr = 0;
210 while (entry) {
211 struct commit *commit = entry->item;
212 struct commit_list *p;
214 if (commit->object.flags & (UNINTERESTING | COUNTED))
215 break;
216 if (!(commit->object.flags & TREESAME))
217 nr++;
218 commit->object.flags |= COUNTED;
219 p = commit->parents;
220 entry = p;
221 if (p) {
222 p = p->next;
223 while (p) {
224 nr += count_distance(p);
225 p = p->next;
230 return nr;
233 static void clear_distance(struct commit_list *list)
235 while (list) {
236 struct commit *commit = list->item;
237 commit->object.flags &= ~COUNTED;
238 list = list->next;
242 #define DEBUG_BISECT 0
244 static inline int weight(struct commit_list *elem)
246 return *((int*)(elem->item->util));
249 static inline void weight_set(struct commit_list *elem, int weight)
251 *((int*)(elem->item->util)) = weight;
254 static int count_interesting_parents(struct commit *commit)
256 struct commit_list *p;
257 int count;
259 for (count = 0, p = commit->parents; p; p = p->next) {
260 if (p->item->object.flags & UNINTERESTING)
261 continue;
262 count++;
264 return count;
267 static inline int halfway(struct commit_list *p, int nr)
270 * Don't short-cut something we are not going to return!
272 if (p->item->object.flags & TREESAME)
273 return 0;
274 if (DEBUG_BISECT)
275 return 0;
277 * 2 and 3 are halfway of 5.
278 * 3 is halfway of 6 but 2 and 4 are not.
280 switch (2 * weight(p) - nr) {
281 case -1: case 0: case 1:
282 return 1;
283 default:
284 return 0;
288 #if !DEBUG_BISECT
289 #define show_list(a,b,c,d) do { ; } while (0)
290 #else
291 static void show_list(const char *debug, int counted, int nr,
292 struct commit_list *list)
294 struct commit_list *p;
296 fprintf(stderr, "%s (%d/%d)\n", debug, counted, nr);
298 for (p = list; p; p = p->next) {
299 struct commit_list *pp;
300 struct commit *commit = p->item;
301 unsigned flags = commit->object.flags;
302 enum object_type type;
303 unsigned long size;
304 char *buf = read_sha1_file(commit->object.sha1, &type, &size);
305 char *ep, *sp;
307 fprintf(stderr, "%c%c%c ",
308 (flags & TREESAME) ? ' ' : 'T',
309 (flags & UNINTERESTING) ? 'U' : ' ',
310 (flags & COUNTED) ? 'C' : ' ');
311 if (commit->util)
312 fprintf(stderr, "%3d", weight(p));
313 else
314 fprintf(stderr, "---");
315 fprintf(stderr, " %.*s", 8, sha1_to_hex(commit->object.sha1));
316 for (pp = commit->parents; pp; pp = pp->next)
317 fprintf(stderr, " %.*s", 8,
318 sha1_to_hex(pp->item->object.sha1));
320 sp = strstr(buf, "\n\n");
321 if (sp) {
322 sp += 2;
323 for (ep = sp; *ep && *ep != '\n'; ep++)
325 fprintf(stderr, " %.*s", (int)(ep - sp), sp);
327 fprintf(stderr, "\n");
330 #endif /* DEBUG_BISECT */
332 static struct commit_list *best_bisection(struct commit_list *list, int nr)
334 struct commit_list *p, *best;
335 int best_distance = -1;
337 best = list;
338 for (p = list; p; p = p->next) {
339 int distance;
340 unsigned flags = p->item->object.flags;
342 if (flags & TREESAME)
343 continue;
344 distance = weight(p);
345 if (nr - distance < distance)
346 distance = nr - distance;
347 if (distance > best_distance) {
348 best = p;
349 best_distance = distance;
353 return best;
356 struct commit_dist {
357 struct commit *commit;
358 int distance;
361 static int compare_commit_dist(const void *a_, const void *b_)
363 struct commit_dist *a, *b;
365 a = (struct commit_dist *)a_;
366 b = (struct commit_dist *)b_;
367 if (a->distance != b->distance)
368 return b->distance - a->distance; /* desc sort */
369 return hashcmp(a->commit->object.sha1, b->commit->object.sha1);
372 static struct commit_list *best_bisection_sorted(struct commit_list *list, int nr)
374 struct commit_list *p;
375 struct commit_dist *array = xcalloc(nr, sizeof(*array));
376 int cnt, i;
378 for (p = list, cnt = 0; p; p = p->next) {
379 int distance;
380 unsigned flags = p->item->object.flags;
382 if (flags & TREESAME)
383 continue;
384 distance = weight(p);
385 if (nr - distance < distance)
386 distance = nr - distance;
387 array[cnt].commit = p->item;
388 array[cnt].distance = distance;
389 cnt++;
391 qsort(array, cnt, sizeof(*array), compare_commit_dist);
392 for (p = list, i = 0; i < cnt; i++) {
393 struct name_decoration *r = xmalloc(sizeof(*r) + 100);
394 struct object *obj = &(array[i].commit->object);
396 sprintf(r->name, "dist=%d", array[i].distance);
397 r->next = add_decoration(&name_decoration, obj, r);
398 p->item = array[i].commit;
399 p = p->next;
401 if (p)
402 p->next = NULL;
403 free(array);
404 return list;
408 * zero or positive weight is the number of interesting commits it can
409 * reach, including itself. Especially, weight = 0 means it does not
410 * reach any tree-changing commits (e.g. just above uninteresting one
411 * but traversal is with pathspec).
413 * weight = -1 means it has one parent and its distance is yet to
414 * be computed.
416 * weight = -2 means it has more than one parent and its distance is
417 * unknown. After running count_distance() first, they will get zero
418 * or positive distance.
420 static struct commit_list *do_find_bisection(struct commit_list *list,
421 int nr, int *weights,
422 int find_all)
424 int n, counted;
425 struct commit_list *p;
427 counted = 0;
429 for (n = 0, p = list; p; p = p->next) {
430 struct commit *commit = p->item;
431 unsigned flags = commit->object.flags;
433 p->item->util = &weights[n++];
434 switch (count_interesting_parents(commit)) {
435 case 0:
436 if (!(flags & TREESAME)) {
437 weight_set(p, 1);
438 counted++;
439 show_list("bisection 2 count one",
440 counted, nr, list);
443 * otherwise, it is known not to reach any
444 * tree-changing commit and gets weight 0.
446 break;
447 case 1:
448 weight_set(p, -1);
449 break;
450 default:
451 weight_set(p, -2);
452 break;
456 show_list("bisection 2 initialize", counted, nr, list);
459 * If you have only one parent in the resulting set
460 * then you can reach one commit more than that parent
461 * can reach. So we do not have to run the expensive
462 * count_distance() for single strand of pearls.
464 * However, if you have more than one parents, you cannot
465 * just add their distance and one for yourself, since
466 * they usually reach the same ancestor and you would
467 * end up counting them twice that way.
469 * So we will first count distance of merges the usual
470 * way, and then fill the blanks using cheaper algorithm.
472 for (p = list; p; p = p->next) {
473 if (p->item->object.flags & UNINTERESTING)
474 continue;
475 if (weight(p) != -2)
476 continue;
477 weight_set(p, count_distance(p));
478 clear_distance(list);
480 /* Does it happen to be at exactly half-way? */
481 if (!find_all && halfway(p, nr))
482 return p;
483 counted++;
486 show_list("bisection 2 count_distance", counted, nr, list);
488 while (counted < nr) {
489 for (p = list; p; p = p->next) {
490 struct commit_list *q;
491 unsigned flags = p->item->object.flags;
493 if (0 <= weight(p))
494 continue;
495 for (q = p->item->parents; q; q = q->next) {
496 if (q->item->object.flags & UNINTERESTING)
497 continue;
498 if (0 <= weight(q))
499 break;
501 if (!q)
502 continue;
505 * weight for p is unknown but q is known.
506 * add one for p itself if p is to be counted,
507 * otherwise inherit it from q directly.
509 if (!(flags & TREESAME)) {
510 weight_set(p, weight(q)+1);
511 counted++;
512 show_list("bisection 2 count one",
513 counted, nr, list);
515 else
516 weight_set(p, weight(q));
518 /* Does it happen to be at exactly half-way? */
519 if (!find_all && halfway(p, nr))
520 return p;
524 show_list("bisection 2 counted all", counted, nr, list);
526 if (!find_all)
527 return best_bisection(list, nr);
528 else
529 return best_bisection_sorted(list, nr);
532 static struct commit_list *find_bisection(struct commit_list *list,
533 int *reaches, int *all,
534 int find_all)
536 int nr, on_list;
537 struct commit_list *p, *best, *next, *last;
538 int *weights;
540 show_list("bisection 2 entry", 0, 0, list);
543 * Count the number of total and tree-changing items on the
544 * list, while reversing the list.
546 for (nr = on_list = 0, last = NULL, p = list;
548 p = next) {
549 unsigned flags = p->item->object.flags;
551 next = p->next;
552 if (flags & UNINTERESTING)
553 continue;
554 p->next = last;
555 last = p;
556 if (!(flags & TREESAME))
557 nr++;
558 on_list++;
560 list = last;
561 show_list("bisection 2 sorted", 0, nr, list);
563 *all = nr;
564 weights = xcalloc(on_list, sizeof(*weights));
566 /* Do the real work of finding bisection commit. */
567 best = do_find_bisection(list, nr, weights, find_all);
568 if (best) {
569 if (!find_all)
570 best->next = NULL;
571 *reaches = weight(best);
573 free(weights);
574 return best;
577 static inline int log2i(int n)
579 int log2 = 0;
581 for (; n > 1; n >>= 1)
582 log2++;
584 return log2;
587 static inline int exp2i(int n)
589 return 1 << n;
593 * Estimate the number of bisect steps left (after the current step)
595 * For any x between 0 included and 2^n excluded, the probability for
596 * n - 1 steps left looks like:
598 * P(2^n + x) == (2^n - x) / (2^n + x)
600 * and P(2^n + x) < 0.5 means 2^n < 3x
602 static int estimate_bisect_steps(int all)
604 int n, x, e;
606 if (all < 3)
607 return 0;
609 n = log2i(all);
610 e = exp2i(n);
611 x = all - e;
613 return (e < 3 * x) ? n : n - 1;
616 int cmd_rev_list(int argc, const char **argv, const char *prefix)
618 struct commit_list *list;
619 int i;
620 int read_from_stdin = 0;
621 int bisect_show_vars = 0;
622 int bisect_find_all = 0;
623 int quiet = 0;
625 git_config(git_default_config, NULL);
626 init_revisions(&revs, prefix);
627 revs.abbrev = 0;
628 revs.commit_format = CMIT_FMT_UNSPECIFIED;
629 argc = setup_revisions(argc, argv, &revs, NULL);
631 quiet = DIFF_OPT_TST(&revs.diffopt, QUIET);
632 for (i = 1 ; i < argc; i++) {
633 const char *arg = argv[i];
635 if (!strcmp(arg, "--header")) {
636 revs.verbose_header = 1;
637 continue;
639 if (!strcmp(arg, "--timestamp")) {
640 show_timestamp = 1;
641 continue;
643 if (!strcmp(arg, "--bisect")) {
644 bisect_list = 1;
645 continue;
647 if (!strcmp(arg, "--bisect-all")) {
648 bisect_list = 1;
649 bisect_find_all = 1;
650 revs.show_decorations = 1;
651 continue;
653 if (!strcmp(arg, "--bisect-vars")) {
654 bisect_list = 1;
655 bisect_show_vars = 1;
656 continue;
658 if (!strcmp(arg, "--stdin")) {
659 if (read_from_stdin++)
660 die("--stdin given twice?");
661 read_revisions_from_stdin(&revs);
662 continue;
664 usage(rev_list_usage);
667 if (revs.commit_format != CMIT_FMT_UNSPECIFIED) {
668 /* The command line has a --pretty */
669 hdr_termination = '\n';
670 if (revs.commit_format == CMIT_FMT_ONELINE)
671 header_prefix = "";
672 else
673 header_prefix = "commit ";
675 else if (revs.verbose_header)
676 /* Only --header was specified */
677 revs.commit_format = CMIT_FMT_RAW;
679 list = revs.commits;
681 if ((!list &&
682 (!(revs.tag_objects||revs.tree_objects||revs.blob_objects) &&
683 !revs.pending.nr)) ||
684 revs.diff)
685 usage(rev_list_usage);
687 save_commit_buffer = revs.verbose_header ||
688 revs.grep_filter.pattern_list;
689 if (bisect_list)
690 revs.limited = 1;
692 if (prepare_revision_walk(&revs))
693 die("revision walk setup failed");
694 if (revs.tree_objects)
695 mark_edges_uninteresting(revs.commits, &revs, show_edge);
697 if (bisect_list) {
698 int reaches = reaches, all = all;
700 revs.commits = find_bisection(revs.commits, &reaches, &all,
701 bisect_find_all);
702 if (bisect_show_vars) {
703 int cnt;
704 char hex[41];
705 if (!revs.commits)
706 return 1;
708 * revs.commits can reach "reaches" commits among
709 * "all" commits. If it is good, then there are
710 * (all-reaches) commits left to be bisected.
711 * On the other hand, if it is bad, then the set
712 * to bisect is "reaches".
713 * A bisect set of size N has (N-1) commits further
714 * to test, as we already know one bad one.
716 cnt = all - reaches;
717 if (cnt < reaches)
718 cnt = reaches;
719 strcpy(hex, sha1_to_hex(revs.commits->item->object.sha1));
721 if (bisect_find_all) {
722 traverse_commit_list(&revs, show_commit, show_object);
723 printf("------\n");
726 printf("bisect_rev=%s\n"
727 "bisect_nr=%d\n"
728 "bisect_good=%d\n"
729 "bisect_bad=%d\n"
730 "bisect_all=%d\n"
731 "bisect_steps=%d\n",
732 hex,
733 cnt - 1,
734 all - reaches - 1,
735 reaches - 1,
736 all,
737 estimate_bisect_steps(all));
738 return 0;
742 traverse_commit_list(&revs,
743 quiet ? finish_commit : show_commit,
744 quiet ? finish_object : show_object);
746 return 0;