use commit_list_count() to count the members of commit_lists
[git.git] / builtin / blame.c
blob75b3a028a729f2fa447d8c0d81e21c08ba25977d
1 /*
2 * Blame
4 * Copyright (c) 2006, 2014 by its authors
5 * See COPYING for licensing conditions
6 */
8 #include "cache.h"
9 #include "builtin.h"
10 #include "blob.h"
11 #include "commit.h"
12 #include "tag.h"
13 #include "tree-walk.h"
14 #include "diff.h"
15 #include "diffcore.h"
16 #include "revision.h"
17 #include "quote.h"
18 #include "xdiff-interface.h"
19 #include "cache-tree.h"
20 #include "string-list.h"
21 #include "mailmap.h"
22 #include "mergesort.h"
23 #include "parse-options.h"
24 #include "prio-queue.h"
25 #include "utf8.h"
26 #include "userdiff.h"
27 #include "line-range.h"
28 #include "line-log.h"
30 static char blame_usage[] = N_("git blame [options] [rev-opts] [rev] [--] file");
32 static const char *blame_opt_usage[] = {
33 blame_usage,
34 "",
35 N_("[rev-opts] are documented in git-rev-list(1)"),
36 NULL
39 static int longest_file;
40 static int longest_author;
41 static int max_orig_digits;
42 static int max_digits;
43 static int max_score_digits;
44 static int show_root;
45 static int reverse;
46 static int blank_boundary;
47 static int incremental;
48 static int xdl_opts;
49 static int abbrev = -1;
50 static int no_whole_file_rename;
52 static enum date_mode blame_date_mode = DATE_ISO8601;
53 static size_t blame_date_width;
55 static struct string_list mailmap;
57 #ifndef DEBUG
58 #define DEBUG 0
59 #endif
61 /* stats */
62 static int num_read_blob;
63 static int num_get_patch;
64 static int num_commits;
66 #define PICKAXE_BLAME_MOVE 01
67 #define PICKAXE_BLAME_COPY 02
68 #define PICKAXE_BLAME_COPY_HARDER 04
69 #define PICKAXE_BLAME_COPY_HARDEST 010
72 * blame for a blame_entry with score lower than these thresholds
73 * is not passed to the parent using move/copy logic.
75 static unsigned blame_move_score;
76 static unsigned blame_copy_score;
77 #define BLAME_DEFAULT_MOVE_SCORE 20
78 #define BLAME_DEFAULT_COPY_SCORE 40
80 /* Remember to update object flag allocation in object.h */
81 #define METAINFO_SHOWN (1u<<12)
82 #define MORE_THAN_ONE_PATH (1u<<13)
85 * One blob in a commit that is being suspected
87 struct origin {
88 int refcnt;
89 /* Record preceding blame record for this blob */
90 struct origin *previous;
91 /* origins are put in a list linked via `next' hanging off the
92 * corresponding commit's util field in order to make finding
93 * them fast. The presence in this chain does not count
94 * towards the origin's reference count. It is tempting to
95 * let it count as long as the commit is pending examination,
96 * but even under circumstances where the commit will be
97 * present multiple times in the priority queue of unexamined
98 * commits, processing the first instance will not leave any
99 * work requiring the origin data for the second instance. An
100 * interspersed commit changing that would have to be
101 * preexisting with a different ancestry and with the same
102 * commit date in order to wedge itself between two instances
103 * of the same commit in the priority queue _and_ produce
104 * blame entries relevant for it. While we don't want to let
105 * us get tripped up by this case, it certainly does not seem
106 * worth optimizing for.
108 struct origin *next;
109 struct commit *commit;
110 /* `suspects' contains blame entries that may be attributed to
111 * this origin's commit or to parent commits. When a commit
112 * is being processed, all suspects will be moved, either by
113 * assigning them to an origin in a different commit, or by
114 * shipping them to the scoreboard's ent list because they
115 * cannot be attributed to a different commit.
117 struct blame_entry *suspects;
118 mmfile_t file;
119 unsigned char blob_sha1[20];
120 unsigned mode;
121 /* guilty gets set when shipping any suspects to the final
122 * blame list instead of other commits
124 char guilty;
125 char path[FLEX_ARRAY];
128 static int diff_hunks(mmfile_t *file_a, mmfile_t *file_b, long ctxlen,
129 xdl_emit_hunk_consume_func_t hunk_func, void *cb_data)
131 xpparam_t xpp = {0};
132 xdemitconf_t xecfg = {0};
133 xdemitcb_t ecb = {NULL};
135 xpp.flags = xdl_opts;
136 xecfg.ctxlen = ctxlen;
137 xecfg.hunk_func = hunk_func;
138 ecb.priv = cb_data;
139 return xdi_diff(file_a, file_b, &xpp, &xecfg, &ecb);
143 * Prepare diff_filespec and convert it using diff textconv API
144 * if the textconv driver exists.
145 * Return 1 if the conversion succeeds, 0 otherwise.
147 int textconv_object(const char *path,
148 unsigned mode,
149 const unsigned char *sha1,
150 int sha1_valid,
151 char **buf,
152 unsigned long *buf_size)
154 struct diff_filespec *df;
155 struct userdiff_driver *textconv;
157 df = alloc_filespec(path);
158 fill_filespec(df, sha1, sha1_valid, mode);
159 textconv = get_textconv(df);
160 if (!textconv) {
161 free_filespec(df);
162 return 0;
165 *buf_size = fill_textconv(textconv, df, buf);
166 free_filespec(df);
167 return 1;
171 * Given an origin, prepare mmfile_t structure to be used by the
172 * diff machinery
174 static void fill_origin_blob(struct diff_options *opt,
175 struct origin *o, mmfile_t *file)
177 if (!o->file.ptr) {
178 enum object_type type;
179 unsigned long file_size;
181 num_read_blob++;
182 if (DIFF_OPT_TST(opt, ALLOW_TEXTCONV) &&
183 textconv_object(o->path, o->mode, o->blob_sha1, 1, &file->ptr, &file_size))
185 else
186 file->ptr = read_sha1_file(o->blob_sha1, &type, &file_size);
187 file->size = file_size;
189 if (!file->ptr)
190 die("Cannot read blob %s for path %s",
191 sha1_to_hex(o->blob_sha1),
192 o->path);
193 o->file = *file;
195 else
196 *file = o->file;
200 * Origin is refcounted and usually we keep the blob contents to be
201 * reused.
203 static inline struct origin *origin_incref(struct origin *o)
205 if (o)
206 o->refcnt++;
207 return o;
210 static void origin_decref(struct origin *o)
212 if (o && --o->refcnt <= 0) {
213 struct origin *p, *l = NULL;
214 if (o->previous)
215 origin_decref(o->previous);
216 free(o->file.ptr);
217 /* Should be present exactly once in commit chain */
218 for (p = o->commit->util; p; l = p, p = p->next) {
219 if (p == o) {
220 if (l)
221 l->next = p->next;
222 else
223 o->commit->util = p->next;
224 free(o);
225 return;
228 die("internal error in blame::origin_decref");
232 static void drop_origin_blob(struct origin *o)
234 if (o->file.ptr) {
235 free(o->file.ptr);
236 o->file.ptr = NULL;
241 * Each group of lines is described by a blame_entry; it can be split
242 * as we pass blame to the parents. They are arranged in linked lists
243 * kept as `suspects' of some unprocessed origin, or entered (when the
244 * blame origin has been finalized) into the scoreboard structure.
245 * While the scoreboard structure is only sorted at the end of
246 * processing (according to final image line number), the lists
247 * attached to an origin are sorted by the target line number.
249 struct blame_entry {
250 struct blame_entry *next;
252 /* the first line of this group in the final image;
253 * internally all line numbers are 0 based.
255 int lno;
257 /* how many lines this group has */
258 int num_lines;
260 /* the commit that introduced this group into the final image */
261 struct origin *suspect;
263 /* the line number of the first line of this group in the
264 * suspect's file; internally all line numbers are 0 based.
266 int s_lno;
268 /* how significant this entry is -- cached to avoid
269 * scanning the lines over and over.
271 unsigned score;
275 * Any merge of blames happens on lists of blames that arrived via
276 * different parents in a single suspect. In this case, we want to
277 * sort according to the suspect line numbers as opposed to the final
278 * image line numbers. The function body is somewhat longish because
279 * it avoids unnecessary writes.
282 static struct blame_entry *blame_merge(struct blame_entry *list1,
283 struct blame_entry *list2)
285 struct blame_entry *p1 = list1, *p2 = list2,
286 **tail = &list1;
288 if (!p1)
289 return p2;
290 if (!p2)
291 return p1;
293 if (p1->s_lno <= p2->s_lno) {
294 do {
295 tail = &p1->next;
296 if ((p1 = *tail) == NULL) {
297 *tail = p2;
298 return list1;
300 } while (p1->s_lno <= p2->s_lno);
302 for (;;) {
303 *tail = p2;
304 do {
305 tail = &p2->next;
306 if ((p2 = *tail) == NULL) {
307 *tail = p1;
308 return list1;
310 } while (p1->s_lno > p2->s_lno);
311 *tail = p1;
312 do {
313 tail = &p1->next;
314 if ((p1 = *tail) == NULL) {
315 *tail = p2;
316 return list1;
318 } while (p1->s_lno <= p2->s_lno);
322 static void *get_next_blame(const void *p)
324 return ((struct blame_entry *)p)->next;
327 static void set_next_blame(void *p1, void *p2)
329 ((struct blame_entry *)p1)->next = p2;
333 * Final image line numbers are all different, so we don't need a
334 * three-way comparison here.
337 static int compare_blame_final(const void *p1, const void *p2)
339 return ((struct blame_entry *)p1)->lno > ((struct blame_entry *)p2)->lno
340 ? 1 : -1;
343 static int compare_blame_suspect(const void *p1, const void *p2)
345 const struct blame_entry *s1 = p1, *s2 = p2;
347 * to allow for collating suspects, we sort according to the
348 * respective pointer value as the primary sorting criterion.
349 * The actual relation is pretty unimportant as long as it
350 * establishes a total order. Comparing as integers gives us
351 * that.
353 if (s1->suspect != s2->suspect)
354 return (intptr_t)s1->suspect > (intptr_t)s2->suspect ? 1 : -1;
355 if (s1->s_lno == s2->s_lno)
356 return 0;
357 return s1->s_lno > s2->s_lno ? 1 : -1;
360 static struct blame_entry *blame_sort(struct blame_entry *head,
361 int (*compare_fn)(const void *, const void *))
363 return llist_mergesort (head, get_next_blame, set_next_blame, compare_fn);
366 static int compare_commits_by_reverse_commit_date(const void *a,
367 const void *b,
368 void *c)
370 return -compare_commits_by_commit_date(a, b, c);
374 * The current state of the blame assignment.
376 struct scoreboard {
377 /* the final commit (i.e. where we started digging from) */
378 struct commit *final;
379 /* Priority queue for commits with unassigned blame records */
380 struct prio_queue commits;
381 struct rev_info *revs;
382 const char *path;
385 * The contents in the final image.
386 * Used by many functions to obtain contents of the nth line,
387 * indexed with scoreboard.lineno[blame_entry.lno].
389 const char *final_buf;
390 unsigned long final_buf_size;
392 /* linked list of blames */
393 struct blame_entry *ent;
395 /* look-up a line in the final buffer */
396 int num_lines;
397 int *lineno;
400 static void sanity_check_refcnt(struct scoreboard *);
403 * If two blame entries that are next to each other came from
404 * contiguous lines in the same origin (i.e. <commit, path> pair),
405 * merge them together.
407 static void coalesce(struct scoreboard *sb)
409 struct blame_entry *ent, *next;
411 for (ent = sb->ent; ent && (next = ent->next); ent = next) {
412 if (ent->suspect == next->suspect &&
413 ent->s_lno + ent->num_lines == next->s_lno) {
414 ent->num_lines += next->num_lines;
415 ent->next = next->next;
416 origin_decref(next->suspect);
417 free(next);
418 ent->score = 0;
419 next = ent; /* again */
423 if (DEBUG) /* sanity */
424 sanity_check_refcnt(sb);
428 * Merge the given sorted list of blames into a preexisting origin.
429 * If there were no previous blames to that commit, it is entered into
430 * the commit priority queue of the score board.
433 static void queue_blames(struct scoreboard *sb, struct origin *porigin,
434 struct blame_entry *sorted)
436 if (porigin->suspects)
437 porigin->suspects = blame_merge(porigin->suspects, sorted);
438 else {
439 struct origin *o;
440 for (o = porigin->commit->util; o; o = o->next) {
441 if (o->suspects) {
442 porigin->suspects = sorted;
443 return;
446 porigin->suspects = sorted;
447 prio_queue_put(&sb->commits, porigin->commit);
452 * Given a commit and a path in it, create a new origin structure.
453 * The callers that add blame to the scoreboard should use
454 * get_origin() to obtain shared, refcounted copy instead of calling
455 * this function directly.
457 static struct origin *make_origin(struct commit *commit, const char *path)
459 struct origin *o;
460 o = xcalloc(1, sizeof(*o) + strlen(path) + 1);
461 o->commit = commit;
462 o->refcnt = 1;
463 o->next = commit->util;
464 commit->util = o;
465 strcpy(o->path, path);
466 return o;
470 * Locate an existing origin or create a new one.
471 * This moves the origin to front position in the commit util list.
473 static struct origin *get_origin(struct scoreboard *sb,
474 struct commit *commit,
475 const char *path)
477 struct origin *o, *l;
479 for (o = commit->util, l = NULL; o; l = o, o = o->next) {
480 if (!strcmp(o->path, path)) {
481 /* bump to front */
482 if (l) {
483 l->next = o->next;
484 o->next = commit->util;
485 commit->util = o;
487 return origin_incref(o);
490 return make_origin(commit, path);
494 * Fill the blob_sha1 field of an origin if it hasn't, so that later
495 * call to fill_origin_blob() can use it to locate the data. blob_sha1
496 * for an origin is also used to pass the blame for the entire file to
497 * the parent to detect the case where a child's blob is identical to
498 * that of its parent's.
500 * This also fills origin->mode for corresponding tree path.
502 static int fill_blob_sha1_and_mode(struct origin *origin)
504 if (!is_null_sha1(origin->blob_sha1))
505 return 0;
506 if (get_tree_entry(origin->commit->object.sha1,
507 origin->path,
508 origin->blob_sha1, &origin->mode))
509 goto error_out;
510 if (sha1_object_info(origin->blob_sha1, NULL) != OBJ_BLOB)
511 goto error_out;
512 return 0;
513 error_out:
514 hashclr(origin->blob_sha1);
515 origin->mode = S_IFINVALID;
516 return -1;
520 * We have an origin -- check if the same path exists in the
521 * parent and return an origin structure to represent it.
523 static struct origin *find_origin(struct scoreboard *sb,
524 struct commit *parent,
525 struct origin *origin)
527 struct origin *porigin;
528 struct diff_options diff_opts;
529 const char *paths[2];
531 /* First check any existing origins */
532 for (porigin = parent->util; porigin; porigin = porigin->next)
533 if (!strcmp(porigin->path, origin->path)) {
535 * The same path between origin and its parent
536 * without renaming -- the most common case.
538 return origin_incref (porigin);
541 /* See if the origin->path is different between parent
542 * and origin first. Most of the time they are the
543 * same and diff-tree is fairly efficient about this.
545 diff_setup(&diff_opts);
546 DIFF_OPT_SET(&diff_opts, RECURSIVE);
547 diff_opts.detect_rename = 0;
548 diff_opts.output_format = DIFF_FORMAT_NO_OUTPUT;
549 paths[0] = origin->path;
550 paths[1] = NULL;
552 parse_pathspec(&diff_opts.pathspec,
553 PATHSPEC_ALL_MAGIC & ~PATHSPEC_LITERAL,
554 PATHSPEC_LITERAL_PATH, "", paths);
555 diff_setup_done(&diff_opts);
557 if (is_null_sha1(origin->commit->object.sha1))
558 do_diff_cache(parent->tree->object.sha1, &diff_opts);
559 else
560 diff_tree_sha1(parent->tree->object.sha1,
561 origin->commit->tree->object.sha1,
562 "", &diff_opts);
563 diffcore_std(&diff_opts);
565 if (!diff_queued_diff.nr) {
566 /* The path is the same as parent */
567 porigin = get_origin(sb, parent, origin->path);
568 hashcpy(porigin->blob_sha1, origin->blob_sha1);
569 porigin->mode = origin->mode;
570 } else {
572 * Since origin->path is a pathspec, if the parent
573 * commit had it as a directory, we will see a whole
574 * bunch of deletion of files in the directory that we
575 * do not care about.
577 int i;
578 struct diff_filepair *p = NULL;
579 for (i = 0; i < diff_queued_diff.nr; i++) {
580 const char *name;
581 p = diff_queued_diff.queue[i];
582 name = p->one->path ? p->one->path : p->two->path;
583 if (!strcmp(name, origin->path))
584 break;
586 if (!p)
587 die("internal error in blame::find_origin");
588 switch (p->status) {
589 default:
590 die("internal error in blame::find_origin (%c)",
591 p->status);
592 case 'M':
593 porigin = get_origin(sb, parent, origin->path);
594 hashcpy(porigin->blob_sha1, p->one->sha1);
595 porigin->mode = p->one->mode;
596 break;
597 case 'A':
598 case 'T':
599 /* Did not exist in parent, or type changed */
600 break;
603 diff_flush(&diff_opts);
604 free_pathspec(&diff_opts.pathspec);
605 return porigin;
609 * We have an origin -- find the path that corresponds to it in its
610 * parent and return an origin structure to represent it.
612 static struct origin *find_rename(struct scoreboard *sb,
613 struct commit *parent,
614 struct origin *origin)
616 struct origin *porigin = NULL;
617 struct diff_options diff_opts;
618 int i;
620 diff_setup(&diff_opts);
621 DIFF_OPT_SET(&diff_opts, RECURSIVE);
622 diff_opts.detect_rename = DIFF_DETECT_RENAME;
623 diff_opts.output_format = DIFF_FORMAT_NO_OUTPUT;
624 diff_opts.single_follow = origin->path;
625 diff_setup_done(&diff_opts);
627 if (is_null_sha1(origin->commit->object.sha1))
628 do_diff_cache(parent->tree->object.sha1, &diff_opts);
629 else
630 diff_tree_sha1(parent->tree->object.sha1,
631 origin->commit->tree->object.sha1,
632 "", &diff_opts);
633 diffcore_std(&diff_opts);
635 for (i = 0; i < diff_queued_diff.nr; i++) {
636 struct diff_filepair *p = diff_queued_diff.queue[i];
637 if ((p->status == 'R' || p->status == 'C') &&
638 !strcmp(p->two->path, origin->path)) {
639 porigin = get_origin(sb, parent, p->one->path);
640 hashcpy(porigin->blob_sha1, p->one->sha1);
641 porigin->mode = p->one->mode;
642 break;
645 diff_flush(&diff_opts);
646 free_pathspec(&diff_opts.pathspec);
647 return porigin;
651 * Append a new blame entry to a given output queue.
653 static void add_blame_entry(struct blame_entry ***queue, struct blame_entry *e)
655 origin_incref(e->suspect);
657 e->next = **queue;
658 **queue = e;
659 *queue = &e->next;
663 * src typically is on-stack; we want to copy the information in it to
664 * a malloced blame_entry that gets added to the given queue. The
665 * origin of dst loses a refcnt.
667 static void dup_entry(struct blame_entry ***queue,
668 struct blame_entry *dst, struct blame_entry *src)
670 origin_incref(src->suspect);
671 origin_decref(dst->suspect);
672 memcpy(dst, src, sizeof(*src));
673 dst->next = **queue;
674 **queue = dst;
675 *queue = &dst->next;
678 static const char *nth_line(struct scoreboard *sb, long lno)
680 return sb->final_buf + sb->lineno[lno];
683 static const char *nth_line_cb(void *data, long lno)
685 return nth_line((struct scoreboard *)data, lno);
689 * It is known that lines between tlno to same came from parent, and e
690 * has an overlap with that range. it also is known that parent's
691 * line plno corresponds to e's line tlno.
693 * <---- e ----->
694 * <------>
695 * <------------>
696 * <------------>
697 * <------------------>
699 * Split e into potentially three parts; before this chunk, the chunk
700 * to be blamed for the parent, and after that portion.
702 static void split_overlap(struct blame_entry *split,
703 struct blame_entry *e,
704 int tlno, int plno, int same,
705 struct origin *parent)
707 int chunk_end_lno;
708 memset(split, 0, sizeof(struct blame_entry [3]));
710 if (e->s_lno < tlno) {
711 /* there is a pre-chunk part not blamed on parent */
712 split[0].suspect = origin_incref(e->suspect);
713 split[0].lno = e->lno;
714 split[0].s_lno = e->s_lno;
715 split[0].num_lines = tlno - e->s_lno;
716 split[1].lno = e->lno + tlno - e->s_lno;
717 split[1].s_lno = plno;
719 else {
720 split[1].lno = e->lno;
721 split[1].s_lno = plno + (e->s_lno - tlno);
724 if (same < e->s_lno + e->num_lines) {
725 /* there is a post-chunk part not blamed on parent */
726 split[2].suspect = origin_incref(e->suspect);
727 split[2].lno = e->lno + (same - e->s_lno);
728 split[2].s_lno = e->s_lno + (same - e->s_lno);
729 split[2].num_lines = e->s_lno + e->num_lines - same;
730 chunk_end_lno = split[2].lno;
732 else
733 chunk_end_lno = e->lno + e->num_lines;
734 split[1].num_lines = chunk_end_lno - split[1].lno;
737 * if it turns out there is nothing to blame the parent for,
738 * forget about the splitting. !split[1].suspect signals this.
740 if (split[1].num_lines < 1)
741 return;
742 split[1].suspect = origin_incref(parent);
746 * split_overlap() divided an existing blame e into up to three parts
747 * in split. Any assigned blame is moved to queue to
748 * reflect the split.
750 static void split_blame(struct blame_entry ***blamed,
751 struct blame_entry ***unblamed,
752 struct blame_entry *split,
753 struct blame_entry *e)
755 struct blame_entry *new_entry;
757 if (split[0].suspect && split[2].suspect) {
758 /* The first part (reuse storage for the existing entry e) */
759 dup_entry(unblamed, e, &split[0]);
761 /* The last part -- me */
762 new_entry = xmalloc(sizeof(*new_entry));
763 memcpy(new_entry, &(split[2]), sizeof(struct blame_entry));
764 add_blame_entry(unblamed, new_entry);
766 /* ... and the middle part -- parent */
767 new_entry = xmalloc(sizeof(*new_entry));
768 memcpy(new_entry, &(split[1]), sizeof(struct blame_entry));
769 add_blame_entry(blamed, new_entry);
771 else if (!split[0].suspect && !split[2].suspect)
773 * The parent covers the entire area; reuse storage for
774 * e and replace it with the parent.
776 dup_entry(blamed, e, &split[1]);
777 else if (split[0].suspect) {
778 /* me and then parent */
779 dup_entry(unblamed, e, &split[0]);
781 new_entry = xmalloc(sizeof(*new_entry));
782 memcpy(new_entry, &(split[1]), sizeof(struct blame_entry));
783 add_blame_entry(blamed, new_entry);
785 else {
786 /* parent and then me */
787 dup_entry(blamed, e, &split[1]);
789 new_entry = xmalloc(sizeof(*new_entry));
790 memcpy(new_entry, &(split[2]), sizeof(struct blame_entry));
791 add_blame_entry(unblamed, new_entry);
796 * After splitting the blame, the origins used by the
797 * on-stack blame_entry should lose one refcnt each.
799 static void decref_split(struct blame_entry *split)
801 int i;
803 for (i = 0; i < 3; i++)
804 origin_decref(split[i].suspect);
808 * reverse_blame reverses the list given in head, appending tail.
809 * That allows us to build lists in reverse order, then reverse them
810 * afterwards. This can be faster than building the list in proper
811 * order right away. The reason is that building in proper order
812 * requires writing a link in the _previous_ element, while building
813 * in reverse order just requires placing the list head into the
814 * _current_ element.
817 static struct blame_entry *reverse_blame(struct blame_entry *head,
818 struct blame_entry *tail)
820 while (head) {
821 struct blame_entry *next = head->next;
822 head->next = tail;
823 tail = head;
824 head = next;
826 return tail;
830 * Process one hunk from the patch between the current suspect for
831 * blame_entry e and its parent. This first blames any unfinished
832 * entries before the chunk (which is where target and parent start
833 * differing) on the parent, and then splits blame entries at the
834 * start and at the end of the difference region. Since use of -M and
835 * -C options may lead to overlapping/duplicate source line number
836 * ranges, all we can rely on from sorting/merging is the order of the
837 * first suspect line number.
839 static void blame_chunk(struct blame_entry ***dstq, struct blame_entry ***srcq,
840 int tlno, int offset, int same,
841 struct origin *parent)
843 struct blame_entry *e = **srcq;
844 struct blame_entry *samep = NULL, *diffp = NULL;
846 while (e && e->s_lno < tlno) {
847 struct blame_entry *next = e->next;
849 * current record starts before differing portion. If
850 * it reaches into it, we need to split it up and
851 * examine the second part separately.
853 if (e->s_lno + e->num_lines > tlno) {
854 /* Move second half to a new record */
855 int len = tlno - e->s_lno;
856 struct blame_entry *n = xcalloc(1, sizeof (struct blame_entry));
857 n->suspect = e->suspect;
858 n->lno = e->lno + len;
859 n->s_lno = e->s_lno + len;
860 n->num_lines = e->num_lines - len;
861 e->num_lines = len;
862 e->score = 0;
863 /* Push new record to diffp */
864 n->next = diffp;
865 diffp = n;
866 } else
867 origin_decref(e->suspect);
868 /* Pass blame for everything before the differing
869 * chunk to the parent */
870 e->suspect = origin_incref(parent);
871 e->s_lno += offset;
872 e->next = samep;
873 samep = e;
874 e = next;
877 * As we don't know how much of a common stretch after this
878 * diff will occur, the currently blamed parts are all that we
879 * can assign to the parent for now.
882 if (samep) {
883 **dstq = reverse_blame(samep, **dstq);
884 *dstq = &samep->next;
887 * Prepend the split off portions: everything after e starts
888 * after the blameable portion.
890 e = reverse_blame(diffp, e);
893 * Now retain records on the target while parts are different
894 * from the parent.
896 samep = NULL;
897 diffp = NULL;
898 while (e && e->s_lno < same) {
899 struct blame_entry *next = e->next;
902 * If current record extends into sameness, need to split.
904 if (e->s_lno + e->num_lines > same) {
906 * Move second half to a new record to be
907 * processed by later chunks
909 int len = same - e->s_lno;
910 struct blame_entry *n = xcalloc(1, sizeof (struct blame_entry));
911 n->suspect = origin_incref(e->suspect);
912 n->lno = e->lno + len;
913 n->s_lno = e->s_lno + len;
914 n->num_lines = e->num_lines - len;
915 e->num_lines = len;
916 e->score = 0;
917 /* Push new record to samep */
918 n->next = samep;
919 samep = n;
921 e->next = diffp;
922 diffp = e;
923 e = next;
925 **srcq = reverse_blame(diffp, reverse_blame(samep, e));
926 /* Move across elements that are in the unblamable portion */
927 if (diffp)
928 *srcq = &diffp->next;
931 struct blame_chunk_cb_data {
932 struct origin *parent;
933 long offset;
934 struct blame_entry **dstq;
935 struct blame_entry **srcq;
938 /* diff chunks are from parent to target */
939 static int blame_chunk_cb(long start_a, long count_a,
940 long start_b, long count_b, void *data)
942 struct blame_chunk_cb_data *d = data;
943 if (start_a - start_b != d->offset)
944 die("internal error in blame::blame_chunk_cb");
945 blame_chunk(&d->dstq, &d->srcq, start_b, start_a - start_b,
946 start_b + count_b, d->parent);
947 d->offset = start_a + count_a - (start_b + count_b);
948 return 0;
952 * We are looking at the origin 'target' and aiming to pass blame
953 * for the lines it is suspected to its parent. Run diff to find
954 * which lines came from parent and pass blame for them.
956 static void pass_blame_to_parent(struct scoreboard *sb,
957 struct origin *target,
958 struct origin *parent)
960 mmfile_t file_p, file_o;
961 struct blame_chunk_cb_data d;
962 struct blame_entry *newdest = NULL;
964 if (!target->suspects)
965 return; /* nothing remains for this target */
967 d.parent = parent;
968 d.offset = 0;
969 d.dstq = &newdest; d.srcq = &target->suspects;
971 fill_origin_blob(&sb->revs->diffopt, parent, &file_p);
972 fill_origin_blob(&sb->revs->diffopt, target, &file_o);
973 num_get_patch++;
975 diff_hunks(&file_p, &file_o, 0, blame_chunk_cb, &d);
976 /* The rest are the same as the parent */
977 blame_chunk(&d.dstq, &d.srcq, INT_MAX, d.offset, INT_MAX, parent);
978 *d.dstq = NULL;
979 queue_blames(sb, parent, newdest);
981 return;
985 * The lines in blame_entry after splitting blames many times can become
986 * very small and trivial, and at some point it becomes pointless to
987 * blame the parents. E.g. "\t\t}\n\t}\n\n" appears everywhere in any
988 * ordinary C program, and it is not worth to say it was copied from
989 * totally unrelated file in the parent.
991 * Compute how trivial the lines in the blame_entry are.
993 static unsigned ent_score(struct scoreboard *sb, struct blame_entry *e)
995 unsigned score;
996 const char *cp, *ep;
998 if (e->score)
999 return e->score;
1001 score = 1;
1002 cp = nth_line(sb, e->lno);
1003 ep = nth_line(sb, e->lno + e->num_lines);
1004 while (cp < ep) {
1005 unsigned ch = *((unsigned char *)cp);
1006 if (isalnum(ch))
1007 score++;
1008 cp++;
1010 e->score = score;
1011 return score;
1015 * best_so_far[] and this[] are both a split of an existing blame_entry
1016 * that passes blame to the parent. Maintain best_so_far the best split
1017 * so far, by comparing this and best_so_far and copying this into
1018 * bst_so_far as needed.
1020 static void copy_split_if_better(struct scoreboard *sb,
1021 struct blame_entry *best_so_far,
1022 struct blame_entry *this)
1024 int i;
1026 if (!this[1].suspect)
1027 return;
1028 if (best_so_far[1].suspect) {
1029 if (ent_score(sb, &this[1]) < ent_score(sb, &best_so_far[1]))
1030 return;
1033 for (i = 0; i < 3; i++)
1034 origin_incref(this[i].suspect);
1035 decref_split(best_so_far);
1036 memcpy(best_so_far, this, sizeof(struct blame_entry [3]));
1040 * We are looking at a part of the final image represented by
1041 * ent (tlno and same are offset by ent->s_lno).
1042 * tlno is where we are looking at in the final image.
1043 * up to (but not including) same match preimage.
1044 * plno is where we are looking at in the preimage.
1046 * <-------------- final image ---------------------->
1047 * <------ent------>
1048 * ^tlno ^same
1049 * <---------preimage----->
1050 * ^plno
1052 * All line numbers are 0-based.
1054 static void handle_split(struct scoreboard *sb,
1055 struct blame_entry *ent,
1056 int tlno, int plno, int same,
1057 struct origin *parent,
1058 struct blame_entry *split)
1060 if (ent->num_lines <= tlno)
1061 return;
1062 if (tlno < same) {
1063 struct blame_entry this[3];
1064 tlno += ent->s_lno;
1065 same += ent->s_lno;
1066 split_overlap(this, ent, tlno, plno, same, parent);
1067 copy_split_if_better(sb, split, this);
1068 decref_split(this);
1072 struct handle_split_cb_data {
1073 struct scoreboard *sb;
1074 struct blame_entry *ent;
1075 struct origin *parent;
1076 struct blame_entry *split;
1077 long plno;
1078 long tlno;
1081 static int handle_split_cb(long start_a, long count_a,
1082 long start_b, long count_b, void *data)
1084 struct handle_split_cb_data *d = data;
1085 handle_split(d->sb, d->ent, d->tlno, d->plno, start_b, d->parent,
1086 d->split);
1087 d->plno = start_a + count_a;
1088 d->tlno = start_b + count_b;
1089 return 0;
1093 * Find the lines from parent that are the same as ent so that
1094 * we can pass blames to it. file_p has the blob contents for
1095 * the parent.
1097 static void find_copy_in_blob(struct scoreboard *sb,
1098 struct blame_entry *ent,
1099 struct origin *parent,
1100 struct blame_entry *split,
1101 mmfile_t *file_p)
1103 const char *cp;
1104 mmfile_t file_o;
1105 struct handle_split_cb_data d;
1107 memset(&d, 0, sizeof(d));
1108 d.sb = sb; d.ent = ent; d.parent = parent; d.split = split;
1110 * Prepare mmfile that contains only the lines in ent.
1112 cp = nth_line(sb, ent->lno);
1113 file_o.ptr = (char *) cp;
1114 file_o.size = nth_line(sb, ent->lno + ent->num_lines) - cp;
1117 * file_o is a part of final image we are annotating.
1118 * file_p partially may match that image.
1120 memset(split, 0, sizeof(struct blame_entry [3]));
1121 diff_hunks(file_p, &file_o, 1, handle_split_cb, &d);
1122 /* remainder, if any, all match the preimage */
1123 handle_split(sb, ent, d.tlno, d.plno, ent->num_lines, parent, split);
1126 /* Move all blame entries from list *source that have a score smaller
1127 * than score_min to the front of list *small.
1128 * Returns a pointer to the link pointing to the old head of the small list.
1131 static struct blame_entry **filter_small(struct scoreboard *sb,
1132 struct blame_entry **small,
1133 struct blame_entry **source,
1134 unsigned score_min)
1136 struct blame_entry *p = *source;
1137 struct blame_entry *oldsmall = *small;
1138 while (p) {
1139 if (ent_score(sb, p) <= score_min) {
1140 *small = p;
1141 small = &p->next;
1142 p = *small;
1143 } else {
1144 *source = p;
1145 source = &p->next;
1146 p = *source;
1149 *small = oldsmall;
1150 *source = NULL;
1151 return small;
1155 * See if lines currently target is suspected for can be attributed to
1156 * parent.
1158 static void find_move_in_parent(struct scoreboard *sb,
1159 struct blame_entry ***blamed,
1160 struct blame_entry **toosmall,
1161 struct origin *target,
1162 struct origin *parent)
1164 struct blame_entry *e, split[3];
1165 struct blame_entry *unblamed = target->suspects;
1166 struct blame_entry *leftover = NULL;
1167 mmfile_t file_p;
1169 if (!unblamed)
1170 return; /* nothing remains for this target */
1172 fill_origin_blob(&sb->revs->diffopt, parent, &file_p);
1173 if (!file_p.ptr)
1174 return;
1176 /* At each iteration, unblamed has a NULL-terminated list of
1177 * entries that have not yet been tested for blame. leftover
1178 * contains the reversed list of entries that have been tested
1179 * without being assignable to the parent.
1181 do {
1182 struct blame_entry **unblamedtail = &unblamed;
1183 struct blame_entry *next;
1184 for (e = unblamed; e; e = next) {
1185 next = e->next;
1186 find_copy_in_blob(sb, e, parent, split, &file_p);
1187 if (split[1].suspect &&
1188 blame_move_score < ent_score(sb, &split[1])) {
1189 split_blame(blamed, &unblamedtail, split, e);
1190 } else {
1191 e->next = leftover;
1192 leftover = e;
1194 decref_split(split);
1196 *unblamedtail = NULL;
1197 toosmall = filter_small(sb, toosmall, &unblamed, blame_move_score);
1198 } while (unblamed);
1199 target->suspects = reverse_blame(leftover, NULL);
1202 struct blame_list {
1203 struct blame_entry *ent;
1204 struct blame_entry split[3];
1208 * Count the number of entries the target is suspected for,
1209 * and prepare a list of entry and the best split.
1211 static struct blame_list *setup_blame_list(struct blame_entry *unblamed,
1212 int *num_ents_p)
1214 struct blame_entry *e;
1215 int num_ents, i;
1216 struct blame_list *blame_list = NULL;
1218 for (e = unblamed, num_ents = 0; e; e = e->next)
1219 num_ents++;
1220 if (num_ents) {
1221 blame_list = xcalloc(num_ents, sizeof(struct blame_list));
1222 for (e = unblamed, i = 0; e; e = e->next)
1223 blame_list[i++].ent = e;
1225 *num_ents_p = num_ents;
1226 return blame_list;
1230 * For lines target is suspected for, see if we can find code movement
1231 * across file boundary from the parent commit. porigin is the path
1232 * in the parent we already tried.
1234 static void find_copy_in_parent(struct scoreboard *sb,
1235 struct blame_entry ***blamed,
1236 struct blame_entry **toosmall,
1237 struct origin *target,
1238 struct commit *parent,
1239 struct origin *porigin,
1240 int opt)
1242 struct diff_options diff_opts;
1243 int i, j;
1244 struct blame_list *blame_list;
1245 int num_ents;
1246 struct blame_entry *unblamed = target->suspects;
1247 struct blame_entry *leftover = NULL;
1249 if (!unblamed)
1250 return; /* nothing remains for this target */
1252 diff_setup(&diff_opts);
1253 DIFF_OPT_SET(&diff_opts, RECURSIVE);
1254 diff_opts.output_format = DIFF_FORMAT_NO_OUTPUT;
1256 diff_setup_done(&diff_opts);
1258 /* Try "find copies harder" on new path if requested;
1259 * we do not want to use diffcore_rename() actually to
1260 * match things up; find_copies_harder is set only to
1261 * force diff_tree_sha1() to feed all filepairs to diff_queue,
1262 * and this code needs to be after diff_setup_done(), which
1263 * usually makes find-copies-harder imply copy detection.
1265 if ((opt & PICKAXE_BLAME_COPY_HARDEST)
1266 || ((opt & PICKAXE_BLAME_COPY_HARDER)
1267 && (!porigin || strcmp(target->path, porigin->path))))
1268 DIFF_OPT_SET(&diff_opts, FIND_COPIES_HARDER);
1270 if (is_null_sha1(target->commit->object.sha1))
1271 do_diff_cache(parent->tree->object.sha1, &diff_opts);
1272 else
1273 diff_tree_sha1(parent->tree->object.sha1,
1274 target->commit->tree->object.sha1,
1275 "", &diff_opts);
1277 if (!DIFF_OPT_TST(&diff_opts, FIND_COPIES_HARDER))
1278 diffcore_std(&diff_opts);
1280 do {
1281 struct blame_entry **unblamedtail = &unblamed;
1282 blame_list = setup_blame_list(unblamed, &num_ents);
1284 for (i = 0; i < diff_queued_diff.nr; i++) {
1285 struct diff_filepair *p = diff_queued_diff.queue[i];
1286 struct origin *norigin;
1287 mmfile_t file_p;
1288 struct blame_entry this[3];
1290 if (!DIFF_FILE_VALID(p->one))
1291 continue; /* does not exist in parent */
1292 if (S_ISGITLINK(p->one->mode))
1293 continue; /* ignore git links */
1294 if (porigin && !strcmp(p->one->path, porigin->path))
1295 /* find_move already dealt with this path */
1296 continue;
1298 norigin = get_origin(sb, parent, p->one->path);
1299 hashcpy(norigin->blob_sha1, p->one->sha1);
1300 norigin->mode = p->one->mode;
1301 fill_origin_blob(&sb->revs->diffopt, norigin, &file_p);
1302 if (!file_p.ptr)
1303 continue;
1305 for (j = 0; j < num_ents; j++) {
1306 find_copy_in_blob(sb, blame_list[j].ent,
1307 norigin, this, &file_p);
1308 copy_split_if_better(sb, blame_list[j].split,
1309 this);
1310 decref_split(this);
1312 origin_decref(norigin);
1315 for (j = 0; j < num_ents; j++) {
1316 struct blame_entry *split = blame_list[j].split;
1317 if (split[1].suspect &&
1318 blame_copy_score < ent_score(sb, &split[1])) {
1319 split_blame(blamed, &unblamedtail, split,
1320 blame_list[j].ent);
1321 } else {
1322 blame_list[j].ent->next = leftover;
1323 leftover = blame_list[j].ent;
1325 decref_split(split);
1327 free(blame_list);
1328 *unblamedtail = NULL;
1329 toosmall = filter_small(sb, toosmall, &unblamed, blame_copy_score);
1330 } while (unblamed);
1331 target->suspects = reverse_blame(leftover, NULL);
1332 diff_flush(&diff_opts);
1333 free_pathspec(&diff_opts.pathspec);
1337 * The blobs of origin and porigin exactly match, so everything
1338 * origin is suspected for can be blamed on the parent.
1340 static void pass_whole_blame(struct scoreboard *sb,
1341 struct origin *origin, struct origin *porigin)
1343 struct blame_entry *e, *suspects;
1345 if (!porigin->file.ptr && origin->file.ptr) {
1346 /* Steal its file */
1347 porigin->file = origin->file;
1348 origin->file.ptr = NULL;
1350 suspects = origin->suspects;
1351 origin->suspects = NULL;
1352 for (e = suspects; e; e = e->next) {
1353 origin_incref(porigin);
1354 origin_decref(e->suspect);
1355 e->suspect = porigin;
1357 queue_blames(sb, porigin, suspects);
1361 * We pass blame from the current commit to its parents. We keep saying
1362 * "parent" (and "porigin"), but what we mean is to find scapegoat to
1363 * exonerate ourselves.
1365 static struct commit_list *first_scapegoat(struct rev_info *revs, struct commit *commit)
1367 if (!reverse)
1368 return commit->parents;
1369 return lookup_decoration(&revs->children, &commit->object);
1372 static int num_scapegoats(struct rev_info *revs, struct commit *commit)
1374 struct commit_list *l = first_scapegoat(revs, commit);
1375 return commit_list_count(l);
1378 /* Distribute collected unsorted blames to the respected sorted lists
1379 * in the various origins.
1381 static void distribute_blame(struct scoreboard *sb, struct blame_entry *blamed)
1383 blamed = blame_sort(blamed, compare_blame_suspect);
1384 while (blamed)
1386 struct origin *porigin = blamed->suspect;
1387 struct blame_entry *suspects = NULL;
1388 do {
1389 struct blame_entry *next = blamed->next;
1390 blamed->next = suspects;
1391 suspects = blamed;
1392 blamed = next;
1393 } while (blamed && blamed->suspect == porigin);
1394 suspects = reverse_blame(suspects, NULL);
1395 queue_blames(sb, porigin, suspects);
1399 #define MAXSG 16
1401 static void pass_blame(struct scoreboard *sb, struct origin *origin, int opt)
1403 struct rev_info *revs = sb->revs;
1404 int i, pass, num_sg;
1405 struct commit *commit = origin->commit;
1406 struct commit_list *sg;
1407 struct origin *sg_buf[MAXSG];
1408 struct origin *porigin, **sg_origin = sg_buf;
1409 struct blame_entry *toosmall = NULL;
1410 struct blame_entry *blames, **blametail = &blames;
1412 num_sg = num_scapegoats(revs, commit);
1413 if (!num_sg)
1414 goto finish;
1415 else if (num_sg < ARRAY_SIZE(sg_buf))
1416 memset(sg_buf, 0, sizeof(sg_buf));
1417 else
1418 sg_origin = xcalloc(num_sg, sizeof(*sg_origin));
1421 * The first pass looks for unrenamed path to optimize for
1422 * common cases, then we look for renames in the second pass.
1424 for (pass = 0; pass < 2 - no_whole_file_rename; pass++) {
1425 struct origin *(*find)(struct scoreboard *,
1426 struct commit *, struct origin *);
1427 find = pass ? find_rename : find_origin;
1429 for (i = 0, sg = first_scapegoat(revs, commit);
1430 i < num_sg && sg;
1431 sg = sg->next, i++) {
1432 struct commit *p = sg->item;
1433 int j, same;
1435 if (sg_origin[i])
1436 continue;
1437 if (parse_commit(p))
1438 continue;
1439 porigin = find(sb, p, origin);
1440 if (!porigin)
1441 continue;
1442 if (!hashcmp(porigin->blob_sha1, origin->blob_sha1)) {
1443 pass_whole_blame(sb, origin, porigin);
1444 origin_decref(porigin);
1445 goto finish;
1447 for (j = same = 0; j < i; j++)
1448 if (sg_origin[j] &&
1449 !hashcmp(sg_origin[j]->blob_sha1,
1450 porigin->blob_sha1)) {
1451 same = 1;
1452 break;
1454 if (!same)
1455 sg_origin[i] = porigin;
1456 else
1457 origin_decref(porigin);
1461 num_commits++;
1462 for (i = 0, sg = first_scapegoat(revs, commit);
1463 i < num_sg && sg;
1464 sg = sg->next, i++) {
1465 struct origin *porigin = sg_origin[i];
1466 if (!porigin)
1467 continue;
1468 if (!origin->previous) {
1469 origin_incref(porigin);
1470 origin->previous = porigin;
1472 pass_blame_to_parent(sb, origin, porigin);
1473 if (!origin->suspects)
1474 goto finish;
1478 * Optionally find moves in parents' files.
1480 if (opt & PICKAXE_BLAME_MOVE) {
1481 filter_small(sb, &toosmall, &origin->suspects, blame_move_score);
1482 if (origin->suspects) {
1483 for (i = 0, sg = first_scapegoat(revs, commit);
1484 i < num_sg && sg;
1485 sg = sg->next, i++) {
1486 struct origin *porigin = sg_origin[i];
1487 if (!porigin)
1488 continue;
1489 find_move_in_parent(sb, &blametail, &toosmall, origin, porigin);
1490 if (!origin->suspects)
1491 break;
1497 * Optionally find copies from parents' files.
1499 if (opt & PICKAXE_BLAME_COPY) {
1500 if (blame_copy_score > blame_move_score)
1501 filter_small(sb, &toosmall, &origin->suspects, blame_copy_score);
1502 else if (blame_copy_score < blame_move_score) {
1503 origin->suspects = blame_merge(origin->suspects, toosmall);
1504 toosmall = NULL;
1505 filter_small(sb, &toosmall, &origin->suspects, blame_copy_score);
1507 if (!origin->suspects)
1508 goto finish;
1510 for (i = 0, sg = first_scapegoat(revs, commit);
1511 i < num_sg && sg;
1512 sg = sg->next, i++) {
1513 struct origin *porigin = sg_origin[i];
1514 find_copy_in_parent(sb, &blametail, &toosmall,
1515 origin, sg->item, porigin, opt);
1516 if (!origin->suspects)
1517 goto finish;
1521 finish:
1522 *blametail = NULL;
1523 distribute_blame(sb, blames);
1525 * prepend toosmall to origin->suspects
1527 * There is no point in sorting: this ends up on a big
1528 * unsorted list in the caller anyway.
1530 if (toosmall) {
1531 struct blame_entry **tail = &toosmall;
1532 while (*tail)
1533 tail = &(*tail)->next;
1534 *tail = origin->suspects;
1535 origin->suspects = toosmall;
1537 for (i = 0; i < num_sg; i++) {
1538 if (sg_origin[i]) {
1539 drop_origin_blob(sg_origin[i]);
1540 origin_decref(sg_origin[i]);
1543 drop_origin_blob(origin);
1544 if (sg_buf != sg_origin)
1545 free(sg_origin);
1549 * Information on commits, used for output.
1551 struct commit_info {
1552 struct strbuf author;
1553 struct strbuf author_mail;
1554 unsigned long author_time;
1555 struct strbuf author_tz;
1557 /* filled only when asked for details */
1558 struct strbuf committer;
1559 struct strbuf committer_mail;
1560 unsigned long committer_time;
1561 struct strbuf committer_tz;
1563 struct strbuf summary;
1567 * Parse author/committer line in the commit object buffer
1569 static void get_ac_line(const char *inbuf, const char *what,
1570 struct strbuf *name, struct strbuf *mail,
1571 unsigned long *time, struct strbuf *tz)
1573 struct ident_split ident;
1574 size_t len, maillen, namelen;
1575 char *tmp, *endp;
1576 const char *namebuf, *mailbuf;
1578 tmp = strstr(inbuf, what);
1579 if (!tmp)
1580 goto error_out;
1581 tmp += strlen(what);
1582 endp = strchr(tmp, '\n');
1583 if (!endp)
1584 len = strlen(tmp);
1585 else
1586 len = endp - tmp;
1588 if (split_ident_line(&ident, tmp, len)) {
1589 error_out:
1590 /* Ugh */
1591 tmp = "(unknown)";
1592 strbuf_addstr(name, tmp);
1593 strbuf_addstr(mail, tmp);
1594 strbuf_addstr(tz, tmp);
1595 *time = 0;
1596 return;
1599 namelen = ident.name_end - ident.name_begin;
1600 namebuf = ident.name_begin;
1602 maillen = ident.mail_end - ident.mail_begin;
1603 mailbuf = ident.mail_begin;
1605 if (ident.date_begin && ident.date_end)
1606 *time = strtoul(ident.date_begin, NULL, 10);
1607 else
1608 *time = 0;
1610 if (ident.tz_begin && ident.tz_end)
1611 strbuf_add(tz, ident.tz_begin, ident.tz_end - ident.tz_begin);
1612 else
1613 strbuf_addstr(tz, "(unknown)");
1616 * Now, convert both name and e-mail using mailmap
1618 map_user(&mailmap, &mailbuf, &maillen,
1619 &namebuf, &namelen);
1621 strbuf_addf(mail, "<%.*s>", (int)maillen, mailbuf);
1622 strbuf_add(name, namebuf, namelen);
1625 static void commit_info_init(struct commit_info *ci)
1628 strbuf_init(&ci->author, 0);
1629 strbuf_init(&ci->author_mail, 0);
1630 strbuf_init(&ci->author_tz, 0);
1631 strbuf_init(&ci->committer, 0);
1632 strbuf_init(&ci->committer_mail, 0);
1633 strbuf_init(&ci->committer_tz, 0);
1634 strbuf_init(&ci->summary, 0);
1637 static void commit_info_destroy(struct commit_info *ci)
1640 strbuf_release(&ci->author);
1641 strbuf_release(&ci->author_mail);
1642 strbuf_release(&ci->author_tz);
1643 strbuf_release(&ci->committer);
1644 strbuf_release(&ci->committer_mail);
1645 strbuf_release(&ci->committer_tz);
1646 strbuf_release(&ci->summary);
1649 static void get_commit_info(struct commit *commit,
1650 struct commit_info *ret,
1651 int detailed)
1653 int len;
1654 const char *subject, *encoding;
1655 const char *message;
1657 commit_info_init(ret);
1659 encoding = get_log_output_encoding();
1660 message = logmsg_reencode(commit, NULL, encoding);
1661 get_ac_line(message, "\nauthor ",
1662 &ret->author, &ret->author_mail,
1663 &ret->author_time, &ret->author_tz);
1665 if (!detailed) {
1666 unuse_commit_buffer(commit, message);
1667 return;
1670 get_ac_line(message, "\ncommitter ",
1671 &ret->committer, &ret->committer_mail,
1672 &ret->committer_time, &ret->committer_tz);
1674 len = find_commit_subject(message, &subject);
1675 if (len)
1676 strbuf_add(&ret->summary, subject, len);
1677 else
1678 strbuf_addf(&ret->summary, "(%s)", sha1_to_hex(commit->object.sha1));
1680 unuse_commit_buffer(commit, message);
1684 * To allow LF and other nonportable characters in pathnames,
1685 * they are c-style quoted as needed.
1687 static void write_filename_info(const char *path)
1689 printf("filename ");
1690 write_name_quoted(path, stdout, '\n');
1694 * Porcelain/Incremental format wants to show a lot of details per
1695 * commit. Instead of repeating this every line, emit it only once,
1696 * the first time each commit appears in the output (unless the
1697 * user has specifically asked for us to repeat).
1699 static int emit_one_suspect_detail(struct origin *suspect, int repeat)
1701 struct commit_info ci;
1703 if (!repeat && (suspect->commit->object.flags & METAINFO_SHOWN))
1704 return 0;
1706 suspect->commit->object.flags |= METAINFO_SHOWN;
1707 get_commit_info(suspect->commit, &ci, 1);
1708 printf("author %s\n", ci.author.buf);
1709 printf("author-mail %s\n", ci.author_mail.buf);
1710 printf("author-time %lu\n", ci.author_time);
1711 printf("author-tz %s\n", ci.author_tz.buf);
1712 printf("committer %s\n", ci.committer.buf);
1713 printf("committer-mail %s\n", ci.committer_mail.buf);
1714 printf("committer-time %lu\n", ci.committer_time);
1715 printf("committer-tz %s\n", ci.committer_tz.buf);
1716 printf("summary %s\n", ci.summary.buf);
1717 if (suspect->commit->object.flags & UNINTERESTING)
1718 printf("boundary\n");
1719 if (suspect->previous) {
1720 struct origin *prev = suspect->previous;
1721 printf("previous %s ", sha1_to_hex(prev->commit->object.sha1));
1722 write_name_quoted(prev->path, stdout, '\n');
1725 commit_info_destroy(&ci);
1727 return 1;
1731 * The blame_entry is found to be guilty for the range.
1732 * Show it in incremental output.
1734 static void found_guilty_entry(struct blame_entry *ent)
1736 if (incremental) {
1737 struct origin *suspect = ent->suspect;
1739 printf("%s %d %d %d\n",
1740 sha1_to_hex(suspect->commit->object.sha1),
1741 ent->s_lno + 1, ent->lno + 1, ent->num_lines);
1742 emit_one_suspect_detail(suspect, 0);
1743 write_filename_info(suspect->path);
1744 maybe_flush_or_die(stdout, "stdout");
1749 * The main loop -- while we have blobs with lines whose true origin
1750 * is still unknown, pick one blob, and allow its lines to pass blames
1751 * to its parents. */
1752 static void assign_blame(struct scoreboard *sb, int opt)
1754 struct rev_info *revs = sb->revs;
1755 struct commit *commit = prio_queue_get(&sb->commits);
1757 while (commit) {
1758 struct blame_entry *ent;
1759 struct origin *suspect = commit->util;
1761 /* find one suspect to break down */
1762 while (suspect && !suspect->suspects)
1763 suspect = suspect->next;
1765 if (!suspect) {
1766 commit = prio_queue_get(&sb->commits);
1767 continue;
1770 assert(commit == suspect->commit);
1773 * We will use this suspect later in the loop,
1774 * so hold onto it in the meantime.
1776 origin_incref(suspect);
1777 parse_commit(commit);
1778 if (reverse ||
1779 (!(commit->object.flags & UNINTERESTING) &&
1780 !(revs->max_age != -1 && commit->date < revs->max_age)))
1781 pass_blame(sb, suspect, opt);
1782 else {
1783 commit->object.flags |= UNINTERESTING;
1784 if (commit->object.parsed)
1785 mark_parents_uninteresting(commit);
1787 /* treat root commit as boundary */
1788 if (!commit->parents && !show_root)
1789 commit->object.flags |= UNINTERESTING;
1791 /* Take responsibility for the remaining entries */
1792 ent = suspect->suspects;
1793 if (ent) {
1794 suspect->guilty = 1;
1795 for (;;) {
1796 struct blame_entry *next = ent->next;
1797 found_guilty_entry(ent);
1798 if (next) {
1799 ent = next;
1800 continue;
1802 ent->next = sb->ent;
1803 sb->ent = suspect->suspects;
1804 suspect->suspects = NULL;
1805 break;
1808 origin_decref(suspect);
1810 if (DEBUG) /* sanity */
1811 sanity_check_refcnt(sb);
1815 static const char *format_time(unsigned long time, const char *tz_str,
1816 int show_raw_time)
1818 static struct strbuf time_buf = STRBUF_INIT;
1820 strbuf_reset(&time_buf);
1821 if (show_raw_time) {
1822 strbuf_addf(&time_buf, "%lu %s", time, tz_str);
1824 else {
1825 const char *time_str;
1826 size_t time_width;
1827 int tz;
1828 tz = atoi(tz_str);
1829 time_str = show_date(time, tz, blame_date_mode);
1830 strbuf_addstr(&time_buf, time_str);
1832 * Add space paddings to time_buf to display a fixed width
1833 * string, and use time_width for display width calibration.
1835 for (time_width = utf8_strwidth(time_str);
1836 time_width < blame_date_width;
1837 time_width++)
1838 strbuf_addch(&time_buf, ' ');
1840 return time_buf.buf;
1843 #define OUTPUT_ANNOTATE_COMPAT 001
1844 #define OUTPUT_LONG_OBJECT_NAME 002
1845 #define OUTPUT_RAW_TIMESTAMP 004
1846 #define OUTPUT_PORCELAIN 010
1847 #define OUTPUT_SHOW_NAME 020
1848 #define OUTPUT_SHOW_NUMBER 040
1849 #define OUTPUT_SHOW_SCORE 0100
1850 #define OUTPUT_NO_AUTHOR 0200
1851 #define OUTPUT_SHOW_EMAIL 0400
1852 #define OUTPUT_LINE_PORCELAIN 01000
1854 static void emit_porcelain_details(struct origin *suspect, int repeat)
1856 if (emit_one_suspect_detail(suspect, repeat) ||
1857 (suspect->commit->object.flags & MORE_THAN_ONE_PATH))
1858 write_filename_info(suspect->path);
1861 static void emit_porcelain(struct scoreboard *sb, struct blame_entry *ent,
1862 int opt)
1864 int repeat = opt & OUTPUT_LINE_PORCELAIN;
1865 int cnt;
1866 const char *cp;
1867 struct origin *suspect = ent->suspect;
1868 char hex[41];
1870 strcpy(hex, sha1_to_hex(suspect->commit->object.sha1));
1871 printf("%s %d %d %d\n",
1872 hex,
1873 ent->s_lno + 1,
1874 ent->lno + 1,
1875 ent->num_lines);
1876 emit_porcelain_details(suspect, repeat);
1878 cp = nth_line(sb, ent->lno);
1879 for (cnt = 0; cnt < ent->num_lines; cnt++) {
1880 char ch;
1881 if (cnt) {
1882 printf("%s %d %d\n", hex,
1883 ent->s_lno + 1 + cnt,
1884 ent->lno + 1 + cnt);
1885 if (repeat)
1886 emit_porcelain_details(suspect, 1);
1888 putchar('\t');
1889 do {
1890 ch = *cp++;
1891 putchar(ch);
1892 } while (ch != '\n' &&
1893 cp < sb->final_buf + sb->final_buf_size);
1896 if (sb->final_buf_size && cp[-1] != '\n')
1897 putchar('\n');
1900 static void emit_other(struct scoreboard *sb, struct blame_entry *ent, int opt)
1902 int cnt;
1903 const char *cp;
1904 struct origin *suspect = ent->suspect;
1905 struct commit_info ci;
1906 char hex[41];
1907 int show_raw_time = !!(opt & OUTPUT_RAW_TIMESTAMP);
1909 get_commit_info(suspect->commit, &ci, 1);
1910 strcpy(hex, sha1_to_hex(suspect->commit->object.sha1));
1912 cp = nth_line(sb, ent->lno);
1913 for (cnt = 0; cnt < ent->num_lines; cnt++) {
1914 char ch;
1915 int length = (opt & OUTPUT_LONG_OBJECT_NAME) ? 40 : abbrev;
1917 if (suspect->commit->object.flags & UNINTERESTING) {
1918 if (blank_boundary)
1919 memset(hex, ' ', length);
1920 else if (!(opt & OUTPUT_ANNOTATE_COMPAT)) {
1921 length--;
1922 putchar('^');
1926 printf("%.*s", length, hex);
1927 if (opt & OUTPUT_ANNOTATE_COMPAT) {
1928 const char *name;
1929 if (opt & OUTPUT_SHOW_EMAIL)
1930 name = ci.author_mail.buf;
1931 else
1932 name = ci.author.buf;
1933 printf("\t(%10s\t%10s\t%d)", name,
1934 format_time(ci.author_time, ci.author_tz.buf,
1935 show_raw_time),
1936 ent->lno + 1 + cnt);
1937 } else {
1938 if (opt & OUTPUT_SHOW_SCORE)
1939 printf(" %*d %02d",
1940 max_score_digits, ent->score,
1941 ent->suspect->refcnt);
1942 if (opt & OUTPUT_SHOW_NAME)
1943 printf(" %-*.*s", longest_file, longest_file,
1944 suspect->path);
1945 if (opt & OUTPUT_SHOW_NUMBER)
1946 printf(" %*d", max_orig_digits,
1947 ent->s_lno + 1 + cnt);
1949 if (!(opt & OUTPUT_NO_AUTHOR)) {
1950 const char *name;
1951 int pad;
1952 if (opt & OUTPUT_SHOW_EMAIL)
1953 name = ci.author_mail.buf;
1954 else
1955 name = ci.author.buf;
1956 pad = longest_author - utf8_strwidth(name);
1957 printf(" (%s%*s %10s",
1958 name, pad, "",
1959 format_time(ci.author_time,
1960 ci.author_tz.buf,
1961 show_raw_time));
1963 printf(" %*d) ",
1964 max_digits, ent->lno + 1 + cnt);
1966 do {
1967 ch = *cp++;
1968 putchar(ch);
1969 } while (ch != '\n' &&
1970 cp < sb->final_buf + sb->final_buf_size);
1973 if (sb->final_buf_size && cp[-1] != '\n')
1974 putchar('\n');
1976 commit_info_destroy(&ci);
1979 static void output(struct scoreboard *sb, int option)
1981 struct blame_entry *ent;
1983 if (option & OUTPUT_PORCELAIN) {
1984 for (ent = sb->ent; ent; ent = ent->next) {
1985 int count = 0;
1986 struct origin *suspect;
1987 struct commit *commit = ent->suspect->commit;
1988 if (commit->object.flags & MORE_THAN_ONE_PATH)
1989 continue;
1990 for (suspect = commit->util; suspect; suspect = suspect->next) {
1991 if (suspect->guilty && count++) {
1992 commit->object.flags |= MORE_THAN_ONE_PATH;
1993 break;
1999 for (ent = sb->ent; ent; ent = ent->next) {
2000 if (option & OUTPUT_PORCELAIN)
2001 emit_porcelain(sb, ent, option);
2002 else {
2003 emit_other(sb, ent, option);
2008 static const char *get_next_line(const char *start, const char *end)
2010 const char *nl = memchr(start, '\n', end - start);
2011 return nl ? nl + 1 : end;
2015 * To allow quick access to the contents of nth line in the
2016 * final image, prepare an index in the scoreboard.
2018 static int prepare_lines(struct scoreboard *sb)
2020 const char *buf = sb->final_buf;
2021 unsigned long len = sb->final_buf_size;
2022 const char *end = buf + len;
2023 const char *p;
2024 int *lineno;
2025 int num = 0;
2027 for (p = buf; p < end; p = get_next_line(p, end))
2028 num++;
2030 sb->lineno = lineno = xmalloc(sizeof(*sb->lineno) * (num + 1));
2032 for (p = buf; p < end; p = get_next_line(p, end))
2033 *lineno++ = p - buf;
2035 *lineno = len;
2037 sb->num_lines = num;
2038 return sb->num_lines;
2042 * Add phony grafts for use with -S; this is primarily to
2043 * support git's cvsserver that wants to give a linear history
2044 * to its clients.
2046 static int read_ancestry(const char *graft_file)
2048 FILE *fp = fopen(graft_file, "r");
2049 struct strbuf buf = STRBUF_INIT;
2050 if (!fp)
2051 return -1;
2052 while (!strbuf_getwholeline(&buf, fp, '\n')) {
2053 /* The format is just "Commit Parent1 Parent2 ...\n" */
2054 struct commit_graft *graft = read_graft_line(buf.buf, buf.len);
2055 if (graft)
2056 register_commit_graft(graft, 0);
2058 fclose(fp);
2059 strbuf_release(&buf);
2060 return 0;
2063 static int update_auto_abbrev(int auto_abbrev, struct origin *suspect)
2065 const char *uniq = find_unique_abbrev(suspect->commit->object.sha1,
2066 auto_abbrev);
2067 int len = strlen(uniq);
2068 if (auto_abbrev < len)
2069 return len;
2070 return auto_abbrev;
2074 * How many columns do we need to show line numbers, authors,
2075 * and filenames?
2077 static void find_alignment(struct scoreboard *sb, int *option)
2079 int longest_src_lines = 0;
2080 int longest_dst_lines = 0;
2081 unsigned largest_score = 0;
2082 struct blame_entry *e;
2083 int compute_auto_abbrev = (abbrev < 0);
2084 int auto_abbrev = default_abbrev;
2086 for (e = sb->ent; e; e = e->next) {
2087 struct origin *suspect = e->suspect;
2088 struct commit_info ci;
2089 int num;
2091 if (compute_auto_abbrev)
2092 auto_abbrev = update_auto_abbrev(auto_abbrev, suspect);
2093 if (strcmp(suspect->path, sb->path))
2094 *option |= OUTPUT_SHOW_NAME;
2095 num = strlen(suspect->path);
2096 if (longest_file < num)
2097 longest_file = num;
2098 if (!(suspect->commit->object.flags & METAINFO_SHOWN)) {
2099 suspect->commit->object.flags |= METAINFO_SHOWN;
2100 get_commit_info(suspect->commit, &ci, 1);
2101 if (*option & OUTPUT_SHOW_EMAIL)
2102 num = utf8_strwidth(ci.author_mail.buf);
2103 else
2104 num = utf8_strwidth(ci.author.buf);
2105 if (longest_author < num)
2106 longest_author = num;
2108 num = e->s_lno + e->num_lines;
2109 if (longest_src_lines < num)
2110 longest_src_lines = num;
2111 num = e->lno + e->num_lines;
2112 if (longest_dst_lines < num)
2113 longest_dst_lines = num;
2114 if (largest_score < ent_score(sb, e))
2115 largest_score = ent_score(sb, e);
2117 commit_info_destroy(&ci);
2119 max_orig_digits = decimal_width(longest_src_lines);
2120 max_digits = decimal_width(longest_dst_lines);
2121 max_score_digits = decimal_width(largest_score);
2123 if (compute_auto_abbrev)
2124 /* one more abbrev length is needed for the boundary commit */
2125 abbrev = auto_abbrev + 1;
2129 * For debugging -- origin is refcounted, and this asserts that
2130 * we do not underflow.
2132 static void sanity_check_refcnt(struct scoreboard *sb)
2134 int baa = 0;
2135 struct blame_entry *ent;
2137 for (ent = sb->ent; ent; ent = ent->next) {
2138 /* Nobody should have zero or negative refcnt */
2139 if (ent->suspect->refcnt <= 0) {
2140 fprintf(stderr, "%s in %s has negative refcnt %d\n",
2141 ent->suspect->path,
2142 sha1_to_hex(ent->suspect->commit->object.sha1),
2143 ent->suspect->refcnt);
2144 baa = 1;
2147 if (baa) {
2148 int opt = 0160;
2149 find_alignment(sb, &opt);
2150 output(sb, opt);
2151 die("Baa %d!", baa);
2156 * Used for the command line parsing; check if the path exists
2157 * in the working tree.
2159 static int has_string_in_work_tree(const char *path)
2161 struct stat st;
2162 return !lstat(path, &st);
2165 static unsigned parse_score(const char *arg)
2167 char *end;
2168 unsigned long score = strtoul(arg, &end, 10);
2169 if (*end)
2170 return 0;
2171 return score;
2174 static const char *add_prefix(const char *prefix, const char *path)
2176 return prefix_path(prefix, prefix ? strlen(prefix) : 0, path);
2179 static int git_blame_config(const char *var, const char *value, void *cb)
2181 if (!strcmp(var, "blame.showroot")) {
2182 show_root = git_config_bool(var, value);
2183 return 0;
2185 if (!strcmp(var, "blame.blankboundary")) {
2186 blank_boundary = git_config_bool(var, value);
2187 return 0;
2189 if (!strcmp(var, "blame.date")) {
2190 if (!value)
2191 return config_error_nonbool(var);
2192 blame_date_mode = parse_date_format(value);
2193 return 0;
2196 if (userdiff_config(var, value) < 0)
2197 return -1;
2199 return git_default_config(var, value, cb);
2202 static void verify_working_tree_path(struct commit *work_tree, const char *path)
2204 struct commit_list *parents;
2206 for (parents = work_tree->parents; parents; parents = parents->next) {
2207 const unsigned char *commit_sha1 = parents->item->object.sha1;
2208 unsigned char blob_sha1[20];
2209 unsigned mode;
2211 if (!get_tree_entry(commit_sha1, path, blob_sha1, &mode) &&
2212 sha1_object_info(blob_sha1, NULL) == OBJ_BLOB)
2213 return;
2215 die("no such path '%s' in HEAD", path);
2218 static struct commit_list **append_parent(struct commit_list **tail, const unsigned char *sha1)
2220 struct commit *parent;
2222 parent = lookup_commit_reference(sha1);
2223 if (!parent)
2224 die("no such commit %s", sha1_to_hex(sha1));
2225 return &commit_list_insert(parent, tail)->next;
2228 static void append_merge_parents(struct commit_list **tail)
2230 int merge_head;
2231 const char *merge_head_file = git_path("MERGE_HEAD");
2232 struct strbuf line = STRBUF_INIT;
2234 merge_head = open(merge_head_file, O_RDONLY);
2235 if (merge_head < 0) {
2236 if (errno == ENOENT)
2237 return;
2238 die("cannot open '%s' for reading", merge_head_file);
2241 while (!strbuf_getwholeline_fd(&line, merge_head, '\n')) {
2242 unsigned char sha1[20];
2243 if (line.len < 40 || get_sha1_hex(line.buf, sha1))
2244 die("unknown line in '%s': %s", merge_head_file, line.buf);
2245 tail = append_parent(tail, sha1);
2247 close(merge_head);
2248 strbuf_release(&line);
2252 * This isn't as simple as passing sb->buf and sb->len, because we
2253 * want to transfer ownership of the buffer to the commit (so we
2254 * must use detach).
2256 static void set_commit_buffer_from_strbuf(struct commit *c, struct strbuf *sb)
2258 size_t len;
2259 void *buf = strbuf_detach(sb, &len);
2260 set_commit_buffer(c, buf, len);
2264 * Prepare a dummy commit that represents the work tree (or staged) item.
2265 * Note that annotating work tree item never works in the reverse.
2267 static struct commit *fake_working_tree_commit(struct diff_options *opt,
2268 const char *path,
2269 const char *contents_from)
2271 struct commit *commit;
2272 struct origin *origin;
2273 struct commit_list **parent_tail, *parent;
2274 unsigned char head_sha1[20];
2275 struct strbuf buf = STRBUF_INIT;
2276 const char *ident;
2277 time_t now;
2278 int size, len;
2279 struct cache_entry *ce;
2280 unsigned mode;
2281 struct strbuf msg = STRBUF_INIT;
2283 time(&now);
2284 commit = alloc_commit_node();
2285 commit->object.parsed = 1;
2286 commit->date = now;
2287 commit->object.type = OBJ_COMMIT;
2288 parent_tail = &commit->parents;
2290 if (!resolve_ref_unsafe("HEAD", head_sha1, 1, NULL))
2291 die("no such ref: HEAD");
2293 parent_tail = append_parent(parent_tail, head_sha1);
2294 append_merge_parents(parent_tail);
2295 verify_working_tree_path(commit, path);
2297 origin = make_origin(commit, path);
2299 ident = fmt_ident("Not Committed Yet", "not.committed.yet", NULL, 0);
2300 strbuf_addstr(&msg, "tree 0000000000000000000000000000000000000000\n");
2301 for (parent = commit->parents; parent; parent = parent->next)
2302 strbuf_addf(&msg, "parent %s\n",
2303 sha1_to_hex(parent->item->object.sha1));
2304 strbuf_addf(&msg,
2305 "author %s\n"
2306 "committer %s\n\n"
2307 "Version of %s from %s\n",
2308 ident, ident, path,
2309 (!contents_from ? path :
2310 (!strcmp(contents_from, "-") ? "standard input" : contents_from)));
2311 set_commit_buffer_from_strbuf(commit, &msg);
2313 if (!contents_from || strcmp("-", contents_from)) {
2314 struct stat st;
2315 const char *read_from;
2316 char *buf_ptr;
2317 unsigned long buf_len;
2319 if (contents_from) {
2320 if (stat(contents_from, &st) < 0)
2321 die_errno("Cannot stat '%s'", contents_from);
2322 read_from = contents_from;
2324 else {
2325 if (lstat(path, &st) < 0)
2326 die_errno("Cannot lstat '%s'", path);
2327 read_from = path;
2329 mode = canon_mode(st.st_mode);
2331 switch (st.st_mode & S_IFMT) {
2332 case S_IFREG:
2333 if (DIFF_OPT_TST(opt, ALLOW_TEXTCONV) &&
2334 textconv_object(read_from, mode, null_sha1, 0, &buf_ptr, &buf_len))
2335 strbuf_attach(&buf, buf_ptr, buf_len, buf_len + 1);
2336 else if (strbuf_read_file(&buf, read_from, st.st_size) != st.st_size)
2337 die_errno("cannot open or read '%s'", read_from);
2338 break;
2339 case S_IFLNK:
2340 if (strbuf_readlink(&buf, read_from, st.st_size) < 0)
2341 die_errno("cannot readlink '%s'", read_from);
2342 break;
2343 default:
2344 die("unsupported file type %s", read_from);
2347 else {
2348 /* Reading from stdin */
2349 mode = 0;
2350 if (strbuf_read(&buf, 0, 0) < 0)
2351 die_errno("failed to read from stdin");
2353 origin->file.ptr = buf.buf;
2354 origin->file.size = buf.len;
2355 pretend_sha1_file(buf.buf, buf.len, OBJ_BLOB, origin->blob_sha1);
2358 * Read the current index, replace the path entry with
2359 * origin->blob_sha1 without mucking with its mode or type
2360 * bits; we are not going to write this index out -- we just
2361 * want to run "diff-index --cached".
2363 discard_cache();
2364 read_cache();
2366 len = strlen(path);
2367 if (!mode) {
2368 int pos = cache_name_pos(path, len);
2369 if (0 <= pos)
2370 mode = active_cache[pos]->ce_mode;
2371 else
2372 /* Let's not bother reading from HEAD tree */
2373 mode = S_IFREG | 0644;
2375 size = cache_entry_size(len);
2376 ce = xcalloc(1, size);
2377 hashcpy(ce->sha1, origin->blob_sha1);
2378 memcpy(ce->name, path, len);
2379 ce->ce_flags = create_ce_flags(0);
2380 ce->ce_namelen = len;
2381 ce->ce_mode = create_ce_mode(mode);
2382 add_cache_entry(ce, ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE);
2385 * We are not going to write this out, so this does not matter
2386 * right now, but someday we might optimize diff-index --cached
2387 * with cache-tree information.
2389 cache_tree_invalidate_path(active_cache_tree, path);
2391 return commit;
2394 static const char *prepare_final(struct scoreboard *sb)
2396 int i;
2397 const char *final_commit_name = NULL;
2398 struct rev_info *revs = sb->revs;
2401 * There must be one and only one positive commit in the
2402 * revs->pending array.
2404 for (i = 0; i < revs->pending.nr; i++) {
2405 struct object *obj = revs->pending.objects[i].item;
2406 if (obj->flags & UNINTERESTING)
2407 continue;
2408 while (obj->type == OBJ_TAG)
2409 obj = deref_tag(obj, NULL, 0);
2410 if (obj->type != OBJ_COMMIT)
2411 die("Non commit %s?", revs->pending.objects[i].name);
2412 if (sb->final)
2413 die("More than one commit to dig from %s and %s?",
2414 revs->pending.objects[i].name,
2415 final_commit_name);
2416 sb->final = (struct commit *) obj;
2417 final_commit_name = revs->pending.objects[i].name;
2419 return final_commit_name;
2422 static const char *prepare_initial(struct scoreboard *sb)
2424 int i;
2425 const char *final_commit_name = NULL;
2426 struct rev_info *revs = sb->revs;
2429 * There must be one and only one negative commit, and it must be
2430 * the boundary.
2432 for (i = 0; i < revs->pending.nr; i++) {
2433 struct object *obj = revs->pending.objects[i].item;
2434 if (!(obj->flags & UNINTERESTING))
2435 continue;
2436 while (obj->type == OBJ_TAG)
2437 obj = deref_tag(obj, NULL, 0);
2438 if (obj->type != OBJ_COMMIT)
2439 die("Non commit %s?", revs->pending.objects[i].name);
2440 if (sb->final)
2441 die("More than one commit to dig down to %s and %s?",
2442 revs->pending.objects[i].name,
2443 final_commit_name);
2444 sb->final = (struct commit *) obj;
2445 final_commit_name = revs->pending.objects[i].name;
2447 if (!final_commit_name)
2448 die("No commit to dig down to?");
2449 return final_commit_name;
2452 static int blame_copy_callback(const struct option *option, const char *arg, int unset)
2454 int *opt = option->value;
2457 * -C enables copy from removed files;
2458 * -C -C enables copy from existing files, but only
2459 * when blaming a new file;
2460 * -C -C -C enables copy from existing files for
2461 * everybody
2463 if (*opt & PICKAXE_BLAME_COPY_HARDER)
2464 *opt |= PICKAXE_BLAME_COPY_HARDEST;
2465 if (*opt & PICKAXE_BLAME_COPY)
2466 *opt |= PICKAXE_BLAME_COPY_HARDER;
2467 *opt |= PICKAXE_BLAME_COPY | PICKAXE_BLAME_MOVE;
2469 if (arg)
2470 blame_copy_score = parse_score(arg);
2471 return 0;
2474 static int blame_move_callback(const struct option *option, const char *arg, int unset)
2476 int *opt = option->value;
2478 *opt |= PICKAXE_BLAME_MOVE;
2480 if (arg)
2481 blame_move_score = parse_score(arg);
2482 return 0;
2485 int cmd_blame(int argc, const char **argv, const char *prefix)
2487 struct rev_info revs;
2488 const char *path;
2489 struct scoreboard sb;
2490 struct origin *o;
2491 struct blame_entry *ent = NULL;
2492 long dashdash_pos, lno;
2493 const char *final_commit_name = NULL;
2494 enum object_type type;
2496 static struct string_list range_list;
2497 static int output_option = 0, opt = 0;
2498 static int show_stats = 0;
2499 static const char *revs_file = NULL;
2500 static const char *contents_from = NULL;
2501 static const struct option options[] = {
2502 OPT_BOOL(0, "incremental", &incremental, N_("Show blame entries as we find them, incrementally")),
2503 OPT_BOOL('b', NULL, &blank_boundary, N_("Show blank SHA-1 for boundary commits (Default: off)")),
2504 OPT_BOOL(0, "root", &show_root, N_("Do not treat root commits as boundaries (Default: off)")),
2505 OPT_BOOL(0, "show-stats", &show_stats, N_("Show work cost statistics")),
2506 OPT_BIT(0, "score-debug", &output_option, N_("Show output score for blame entries"), OUTPUT_SHOW_SCORE),
2507 OPT_BIT('f', "show-name", &output_option, N_("Show original filename (Default: auto)"), OUTPUT_SHOW_NAME),
2508 OPT_BIT('n', "show-number", &output_option, N_("Show original linenumber (Default: off)"), OUTPUT_SHOW_NUMBER),
2509 OPT_BIT('p', "porcelain", &output_option, N_("Show in a format designed for machine consumption"), OUTPUT_PORCELAIN),
2510 OPT_BIT(0, "line-porcelain", &output_option, N_("Show porcelain format with per-line commit information"), OUTPUT_PORCELAIN|OUTPUT_LINE_PORCELAIN),
2511 OPT_BIT('c', NULL, &output_option, N_("Use the same output mode as git-annotate (Default: off)"), OUTPUT_ANNOTATE_COMPAT),
2512 OPT_BIT('t', NULL, &output_option, N_("Show raw timestamp (Default: off)"), OUTPUT_RAW_TIMESTAMP),
2513 OPT_BIT('l', NULL, &output_option, N_("Show long commit SHA1 (Default: off)"), OUTPUT_LONG_OBJECT_NAME),
2514 OPT_BIT('s', NULL, &output_option, N_("Suppress author name and timestamp (Default: off)"), OUTPUT_NO_AUTHOR),
2515 OPT_BIT('e', "show-email", &output_option, N_("Show author email instead of name (Default: off)"), OUTPUT_SHOW_EMAIL),
2516 OPT_BIT('w', NULL, &xdl_opts, N_("Ignore whitespace differences"), XDF_IGNORE_WHITESPACE),
2517 OPT_BIT(0, "minimal", &xdl_opts, N_("Spend extra cycles to find better match"), XDF_NEED_MINIMAL),
2518 OPT_STRING('S', NULL, &revs_file, N_("file"), N_("Use revisions from <file> instead of calling git-rev-list")),
2519 OPT_STRING(0, "contents", &contents_from, N_("file"), N_("Use <file>'s contents as the final image")),
2520 { OPTION_CALLBACK, 'C', NULL, &opt, N_("score"), N_("Find line copies within and across files"), PARSE_OPT_OPTARG, blame_copy_callback },
2521 { OPTION_CALLBACK, 'M', NULL, &opt, N_("score"), N_("Find line movements within and across files"), PARSE_OPT_OPTARG, blame_move_callback },
2522 OPT_STRING_LIST('L', NULL, &range_list, N_("n,m"), N_("Process only line range n,m, counting from 1")),
2523 OPT__ABBREV(&abbrev),
2524 OPT_END()
2527 struct parse_opt_ctx_t ctx;
2528 int cmd_is_annotate = !strcmp(argv[0], "annotate");
2529 struct range_set ranges;
2530 unsigned int range_i;
2531 long anchor;
2533 git_config(git_blame_config, NULL);
2534 init_revisions(&revs, NULL);
2535 revs.date_mode = blame_date_mode;
2536 DIFF_OPT_SET(&revs.diffopt, ALLOW_TEXTCONV);
2537 DIFF_OPT_SET(&revs.diffopt, FOLLOW_RENAMES);
2539 save_commit_buffer = 0;
2540 dashdash_pos = 0;
2542 parse_options_start(&ctx, argc, argv, prefix, options,
2543 PARSE_OPT_KEEP_DASHDASH | PARSE_OPT_KEEP_ARGV0);
2544 for (;;) {
2545 switch (parse_options_step(&ctx, options, blame_opt_usage)) {
2546 case PARSE_OPT_HELP:
2547 exit(129);
2548 case PARSE_OPT_DONE:
2549 if (ctx.argv[0])
2550 dashdash_pos = ctx.cpidx;
2551 goto parse_done;
2554 if (!strcmp(ctx.argv[0], "--reverse")) {
2555 ctx.argv[0] = "--children";
2556 reverse = 1;
2558 parse_revision_opt(&revs, &ctx, options, blame_opt_usage);
2560 parse_done:
2561 no_whole_file_rename = !DIFF_OPT_TST(&revs.diffopt, FOLLOW_RENAMES);
2562 DIFF_OPT_CLR(&revs.diffopt, FOLLOW_RENAMES);
2563 argc = parse_options_end(&ctx);
2565 if (0 < abbrev)
2566 /* one more abbrev length is needed for the boundary commit */
2567 abbrev++;
2569 if (revs_file && read_ancestry(revs_file))
2570 die_errno("reading graft file '%s' failed", revs_file);
2572 if (cmd_is_annotate) {
2573 output_option |= OUTPUT_ANNOTATE_COMPAT;
2574 blame_date_mode = DATE_ISO8601;
2575 } else {
2576 blame_date_mode = revs.date_mode;
2579 /* The maximum width used to show the dates */
2580 switch (blame_date_mode) {
2581 case DATE_RFC2822:
2582 blame_date_width = sizeof("Thu, 19 Oct 2006 16:00:04 -0700");
2583 break;
2584 case DATE_ISO8601:
2585 blame_date_width = sizeof("2006-10-19 16:00:04 -0700");
2586 break;
2587 case DATE_RAW:
2588 blame_date_width = sizeof("1161298804 -0700");
2589 break;
2590 case DATE_SHORT:
2591 blame_date_width = sizeof("2006-10-19");
2592 break;
2593 case DATE_RELATIVE:
2594 /* TRANSLATORS: This string is used to tell us the maximum
2595 display width for a relative timestamp in "git blame"
2596 output. For C locale, "4 years, 11 months ago", which
2597 takes 22 places, is the longest among various forms of
2598 relative timestamps, but your language may need more or
2599 fewer display columns. */
2600 blame_date_width = utf8_strwidth(_("4 years, 11 months ago")) + 1; /* add the null */
2601 break;
2602 case DATE_LOCAL:
2603 case DATE_NORMAL:
2604 blame_date_width = sizeof("Thu Oct 19 16:00:04 2006 -0700");
2605 break;
2607 blame_date_width -= 1; /* strip the null */
2609 if (DIFF_OPT_TST(&revs.diffopt, FIND_COPIES_HARDER))
2610 opt |= (PICKAXE_BLAME_COPY | PICKAXE_BLAME_MOVE |
2611 PICKAXE_BLAME_COPY_HARDER);
2613 if (!blame_move_score)
2614 blame_move_score = BLAME_DEFAULT_MOVE_SCORE;
2615 if (!blame_copy_score)
2616 blame_copy_score = BLAME_DEFAULT_COPY_SCORE;
2619 * We have collected options unknown to us in argv[1..unk]
2620 * which are to be passed to revision machinery if we are
2621 * going to do the "bottom" processing.
2623 * The remaining are:
2625 * (1) if dashdash_pos != 0, it is either
2626 * "blame [revisions] -- <path>" or
2627 * "blame -- <path> <rev>"
2629 * (2) otherwise, it is one of the two:
2630 * "blame [revisions] <path>"
2631 * "blame <path> <rev>"
2633 * Note that we must strip out <path> from the arguments: we do not
2634 * want the path pruning but we may want "bottom" processing.
2636 if (dashdash_pos) {
2637 switch (argc - dashdash_pos - 1) {
2638 case 2: /* (1b) */
2639 if (argc != 4)
2640 usage_with_options(blame_opt_usage, options);
2641 /* reorder for the new way: <rev> -- <path> */
2642 argv[1] = argv[3];
2643 argv[3] = argv[2];
2644 argv[2] = "--";
2645 /* FALLTHROUGH */
2646 case 1: /* (1a) */
2647 path = add_prefix(prefix, argv[--argc]);
2648 argv[argc] = NULL;
2649 break;
2650 default:
2651 usage_with_options(blame_opt_usage, options);
2653 } else {
2654 if (argc < 2)
2655 usage_with_options(blame_opt_usage, options);
2656 path = add_prefix(prefix, argv[argc - 1]);
2657 if (argc == 3 && !has_string_in_work_tree(path)) { /* (2b) */
2658 path = add_prefix(prefix, argv[1]);
2659 argv[1] = argv[2];
2661 argv[argc - 1] = "--";
2663 setup_work_tree();
2664 if (!has_string_in_work_tree(path))
2665 die_errno("cannot stat path '%s'", path);
2668 revs.disable_stdin = 1;
2669 setup_revisions(argc, argv, &revs, NULL);
2670 memset(&sb, 0, sizeof(sb));
2672 sb.revs = &revs;
2673 if (!reverse) {
2674 final_commit_name = prepare_final(&sb);
2675 sb.commits.compare = compare_commits_by_commit_date;
2677 else if (contents_from)
2678 die("--contents and --children do not blend well.");
2679 else {
2680 final_commit_name = prepare_initial(&sb);
2681 sb.commits.compare = compare_commits_by_reverse_commit_date;
2684 if (!sb.final) {
2686 * "--not A B -- path" without anything positive;
2687 * do not default to HEAD, but use the working tree
2688 * or "--contents".
2690 setup_work_tree();
2691 sb.final = fake_working_tree_commit(&sb.revs->diffopt,
2692 path, contents_from);
2693 add_pending_object(&revs, &(sb.final->object), ":");
2695 else if (contents_from)
2696 die("Cannot use --contents with final commit object name");
2699 * If we have bottom, this will mark the ancestors of the
2700 * bottom commits we would reach while traversing as
2701 * uninteresting.
2703 if (prepare_revision_walk(&revs))
2704 die("revision walk setup failed");
2706 if (is_null_sha1(sb.final->object.sha1)) {
2707 char *buf;
2708 o = sb.final->util;
2709 buf = xmalloc(o->file.size + 1);
2710 memcpy(buf, o->file.ptr, o->file.size + 1);
2711 sb.final_buf = buf;
2712 sb.final_buf_size = o->file.size;
2714 else {
2715 o = get_origin(&sb, sb.final, path);
2716 if (fill_blob_sha1_and_mode(o))
2717 die("no such path %s in %s", path, final_commit_name);
2719 if (DIFF_OPT_TST(&sb.revs->diffopt, ALLOW_TEXTCONV) &&
2720 textconv_object(path, o->mode, o->blob_sha1, 1, (char **) &sb.final_buf,
2721 &sb.final_buf_size))
2723 else
2724 sb.final_buf = read_sha1_file(o->blob_sha1, &type,
2725 &sb.final_buf_size);
2727 if (!sb.final_buf)
2728 die("Cannot read blob %s for path %s",
2729 sha1_to_hex(o->blob_sha1),
2730 path);
2732 num_read_blob++;
2733 lno = prepare_lines(&sb);
2735 if (lno && !range_list.nr)
2736 string_list_append(&range_list, xstrdup("1"));
2738 anchor = 1;
2739 range_set_init(&ranges, range_list.nr);
2740 for (range_i = 0; range_i < range_list.nr; ++range_i) {
2741 long bottom, top;
2742 if (parse_range_arg(range_list.items[range_i].string,
2743 nth_line_cb, &sb, lno, anchor,
2744 &bottom, &top, sb.path))
2745 usage(blame_usage);
2746 if (lno < top || ((lno || bottom) && lno < bottom))
2747 die("file %s has only %lu lines", path, lno);
2748 if (bottom < 1)
2749 bottom = 1;
2750 if (top < 1)
2751 top = lno;
2752 bottom--;
2753 range_set_append_unsafe(&ranges, bottom, top);
2754 anchor = top + 1;
2756 sort_and_merge_range_set(&ranges);
2758 for (range_i = ranges.nr; range_i > 0; --range_i) {
2759 const struct range *r = &ranges.ranges[range_i - 1];
2760 long bottom = r->start;
2761 long top = r->end;
2762 struct blame_entry *next = ent;
2763 ent = xcalloc(1, sizeof(*ent));
2764 ent->lno = bottom;
2765 ent->num_lines = top - bottom;
2766 ent->suspect = o;
2767 ent->s_lno = bottom;
2768 ent->next = next;
2769 origin_incref(o);
2772 o->suspects = ent;
2773 prio_queue_put(&sb.commits, o->commit);
2775 origin_decref(o);
2777 range_set_release(&ranges);
2778 string_list_clear(&range_list, 0);
2780 sb.ent = NULL;
2781 sb.path = path;
2783 read_mailmap(&mailmap, NULL);
2785 if (!incremental)
2786 setup_pager();
2788 assign_blame(&sb, opt);
2790 if (incremental)
2791 return 0;
2793 sb.ent = blame_sort(sb.ent, compare_blame_final);
2795 coalesce(&sb);
2797 if (!(output_option & OUTPUT_PORCELAIN))
2798 find_alignment(&sb, &output_option);
2800 output(&sb, output_option);
2801 free((void *)sb.final_buf);
2802 for (ent = sb.ent; ent; ) {
2803 struct blame_entry *e = ent->next;
2804 free(ent);
2805 ent = e;
2808 if (show_stats) {
2809 printf("num read blob: %d\n", num_read_blob);
2810 printf("num get patch: %d\n", num_get_patch);
2811 printf("num commits: %d\n", num_commits);
2813 return 0;