4 * Copyright (c) 2006, 2014 by its authors
5 * See COPYING for licensing conditions
13 #include "tree-walk.h"
18 #include "xdiff-interface.h"
19 #include "cache-tree.h"
20 #include "string-list.h"
22 #include "mergesort.h"
23 #include "parse-options.h"
24 #include "prio-queue.h"
27 #include "line-range.h"
31 static char blame_usage[] = N_("git blame [<options>] [<rev-opts>] [<rev>] [--] <file>");
33 static const char *blame_opt_usage[] = {
36 N_("<rev-opts> are documented in git-rev-list(1)"),
40 static int longest_file;
41 static int longest_author;
42 static int max_orig_digits;
43 static int max_digits;
44 static int max_score_digits;
47 static int blank_boundary;
48 static int incremental;
50 static int abbrev = -1;
51 static int no_whole_file_rename;
53 static enum date_mode blame_date_mode = DATE_ISO8601;
54 static size_t blame_date_width;
56 static struct string_list mailmap;
63 static int num_read_blob;
64 static int num_get_patch;
65 static int num_commits;
67 #define PICKAXE_BLAME_MOVE 01
68 #define PICKAXE_BLAME_COPY 02
69 #define PICKAXE_BLAME_COPY_HARDER 04
70 #define PICKAXE_BLAME_COPY_HARDEST 010
73 * blame for a blame_entry with score lower than these thresholds
74 * is not passed to the parent using move/copy logic.
76 static unsigned blame_move_score;
77 static unsigned blame_copy_score;
78 #define BLAME_DEFAULT_MOVE_SCORE 20
79 #define BLAME_DEFAULT_COPY_SCORE 40
81 /* Remember to update object flag allocation in object.h */
82 #define METAINFO_SHOWN (1u<<12)
83 #define MORE_THAN_ONE_PATH (1u<<13)
86 * One blob in a commit that is being suspected
90 /* Record preceding blame record for this blob */
91 struct origin *previous;
92 /* origins are put in a list linked via `next' hanging off the
93 * corresponding commit's util field in order to make finding
94 * them fast. The presence in this chain does not count
95 * towards the origin's reference count. It is tempting to
96 * let it count as long as the commit is pending examination,
97 * but even under circumstances where the commit will be
98 * present multiple times in the priority queue of unexamined
99 * commits, processing the first instance will not leave any
100 * work requiring the origin data for the second instance. An
101 * interspersed commit changing that would have to be
102 * preexisting with a different ancestry and with the same
103 * commit date in order to wedge itself between two instances
104 * of the same commit in the priority queue _and_ produce
105 * blame entries relevant for it. While we don't want to let
106 * us get tripped up by this case, it certainly does not seem
107 * worth optimizing for.
110 struct commit *commit;
111 /* `suspects' contains blame entries that may be attributed to
112 * this origin's commit or to parent commits. When a commit
113 * is being processed, all suspects will be moved, either by
114 * assigning them to an origin in a different commit, or by
115 * shipping them to the scoreboard's ent list because they
116 * cannot be attributed to a different commit.
118 struct blame_entry *suspects;
120 unsigned char blob_sha1[20];
122 /* guilty gets set when shipping any suspects to the final
123 * blame list instead of other commits
126 char path[FLEX_ARRAY];
129 static int diff_hunks(mmfile_t *file_a, mmfile_t *file_b, long ctxlen,
130 xdl_emit_hunk_consume_func_t hunk_func, void *cb_data)
133 xdemitconf_t xecfg = {0};
134 xdemitcb_t ecb = {NULL};
136 xpp.flags = xdl_opts;
137 xecfg.ctxlen = ctxlen;
138 xecfg.hunk_func = hunk_func;
140 return xdi_diff(file_a, file_b, &xpp, &xecfg, &ecb);
144 * Prepare diff_filespec and convert it using diff textconv API
145 * if the textconv driver exists.
146 * Return 1 if the conversion succeeds, 0 otherwise.
148 int textconv_object(const char *path,
150 const unsigned char *sha1,
153 unsigned long *buf_size)
155 struct diff_filespec *df;
156 struct userdiff_driver *textconv;
158 df = alloc_filespec(path);
159 fill_filespec(df, sha1, sha1_valid, mode);
160 textconv = get_textconv(df);
166 *buf_size = fill_textconv(textconv, df, buf);
172 * Given an origin, prepare mmfile_t structure to be used by the
175 static void fill_origin_blob(struct diff_options *opt,
176 struct origin *o, mmfile_t *file)
179 enum object_type type;
180 unsigned long file_size;
183 if (DIFF_OPT_TST(opt, ALLOW_TEXTCONV) &&
184 textconv_object(o->path, o->mode, o->blob_sha1, 1, &file->ptr, &file_size))
187 file->ptr = read_sha1_file(o->blob_sha1, &type, &file_size);
188 file->size = file_size;
191 die("Cannot read blob %s for path %s",
192 sha1_to_hex(o->blob_sha1),
201 * Origin is refcounted and usually we keep the blob contents to be
204 static inline struct origin *origin_incref(struct origin *o)
211 static void origin_decref(struct origin *o)
213 if (o && --o->refcnt <= 0) {
214 struct origin *p, *l = NULL;
216 origin_decref(o->previous);
218 /* Should be present exactly once in commit chain */
219 for (p = o->commit->util; p; l = p, p = p->next) {
224 o->commit->util = p->next;
229 die("internal error in blame::origin_decref");
233 static void drop_origin_blob(struct origin *o)
242 * Each group of lines is described by a blame_entry; it can be split
243 * as we pass blame to the parents. They are arranged in linked lists
244 * kept as `suspects' of some unprocessed origin, or entered (when the
245 * blame origin has been finalized) into the scoreboard structure.
246 * While the scoreboard structure is only sorted at the end of
247 * processing (according to final image line number), the lists
248 * attached to an origin are sorted by the target line number.
251 struct blame_entry *next;
253 /* the first line of this group in the final image;
254 * internally all line numbers are 0 based.
258 /* how many lines this group has */
261 /* the commit that introduced this group into the final image */
262 struct origin *suspect;
264 /* the line number of the first line of this group in the
265 * suspect's file; internally all line numbers are 0 based.
269 /* how significant this entry is -- cached to avoid
270 * scanning the lines over and over.
276 * Any merge of blames happens on lists of blames that arrived via
277 * different parents in a single suspect. In this case, we want to
278 * sort according to the suspect line numbers as opposed to the final
279 * image line numbers. The function body is somewhat longish because
280 * it avoids unnecessary writes.
283 static struct blame_entry *blame_merge(struct blame_entry *list1,
284 struct blame_entry *list2)
286 struct blame_entry *p1 = list1, *p2 = list2,
294 if (p1->s_lno <= p2->s_lno) {
297 if ((p1 = *tail) == NULL) {
301 } while (p1->s_lno <= p2->s_lno);
307 if ((p2 = *tail) == NULL) {
311 } while (p1->s_lno > p2->s_lno);
315 if ((p1 = *tail) == NULL) {
319 } while (p1->s_lno <= p2->s_lno);
323 static void *get_next_blame(const void *p)
325 return ((struct blame_entry *)p)->next;
328 static void set_next_blame(void *p1, void *p2)
330 ((struct blame_entry *)p1)->next = p2;
334 * Final image line numbers are all different, so we don't need a
335 * three-way comparison here.
338 static int compare_blame_final(const void *p1, const void *p2)
340 return ((struct blame_entry *)p1)->lno > ((struct blame_entry *)p2)->lno
344 static int compare_blame_suspect(const void *p1, const void *p2)
346 const struct blame_entry *s1 = p1, *s2 = p2;
348 * to allow for collating suspects, we sort according to the
349 * respective pointer value as the primary sorting criterion.
350 * The actual relation is pretty unimportant as long as it
351 * establishes a total order. Comparing as integers gives us
354 if (s1->suspect != s2->suspect)
355 return (intptr_t)s1->suspect > (intptr_t)s2->suspect ? 1 : -1;
356 if (s1->s_lno == s2->s_lno)
358 return s1->s_lno > s2->s_lno ? 1 : -1;
361 static struct blame_entry *blame_sort(struct blame_entry *head,
362 int (*compare_fn)(const void *, const void *))
364 return llist_mergesort (head, get_next_blame, set_next_blame, compare_fn);
367 static int compare_commits_by_reverse_commit_date(const void *a,
371 return -compare_commits_by_commit_date(a, b, c);
375 * The current state of the blame assignment.
378 /* the final commit (i.e. where we started digging from) */
379 struct commit *final;
380 /* Priority queue for commits with unassigned blame records */
381 struct prio_queue commits;
382 struct rev_info *revs;
386 * The contents in the final image.
387 * Used by many functions to obtain contents of the nth line,
388 * indexed with scoreboard.lineno[blame_entry.lno].
390 const char *final_buf;
391 unsigned long final_buf_size;
393 /* linked list of blames */
394 struct blame_entry *ent;
396 /* look-up a line in the final buffer */
401 static void sanity_check_refcnt(struct scoreboard *);
404 * If two blame entries that are next to each other came from
405 * contiguous lines in the same origin (i.e. <commit, path> pair),
406 * merge them together.
408 static void coalesce(struct scoreboard *sb)
410 struct blame_entry *ent, *next;
412 for (ent = sb->ent; ent && (next = ent->next); ent = next) {
413 if (ent->suspect == next->suspect &&
414 ent->s_lno + ent->num_lines == next->s_lno) {
415 ent->num_lines += next->num_lines;
416 ent->next = next->next;
417 origin_decref(next->suspect);
420 next = ent; /* again */
424 if (DEBUG) /* sanity */
425 sanity_check_refcnt(sb);
429 * Merge the given sorted list of blames into a preexisting origin.
430 * If there were no previous blames to that commit, it is entered into
431 * the commit priority queue of the score board.
434 static void queue_blames(struct scoreboard *sb, struct origin *porigin,
435 struct blame_entry *sorted)
437 if (porigin->suspects)
438 porigin->suspects = blame_merge(porigin->suspects, sorted);
441 for (o = porigin->commit->util; o; o = o->next) {
443 porigin->suspects = sorted;
447 porigin->suspects = sorted;
448 prio_queue_put(&sb->commits, porigin->commit);
453 * Given a commit and a path in it, create a new origin structure.
454 * The callers that add blame to the scoreboard should use
455 * get_origin() to obtain shared, refcounted copy instead of calling
456 * this function directly.
458 static struct origin *make_origin(struct commit *commit, const char *path)
461 o = xcalloc(1, sizeof(*o) + strlen(path) + 1);
464 o->next = commit->util;
466 strcpy(o->path, path);
471 * Locate an existing origin or create a new one.
472 * This moves the origin to front position in the commit util list.
474 static struct origin *get_origin(struct scoreboard *sb,
475 struct commit *commit,
478 struct origin *o, *l;
480 for (o = commit->util, l = NULL; o; l = o, o = o->next) {
481 if (!strcmp(o->path, path)) {
485 o->next = commit->util;
488 return origin_incref(o);
491 return make_origin(commit, path);
495 * Fill the blob_sha1 field of an origin if it hasn't, so that later
496 * call to fill_origin_blob() can use it to locate the data. blob_sha1
497 * for an origin is also used to pass the blame for the entire file to
498 * the parent to detect the case where a child's blob is identical to
499 * that of its parent's.
501 * This also fills origin->mode for corresponding tree path.
503 static int fill_blob_sha1_and_mode(struct origin *origin)
505 if (!is_null_sha1(origin->blob_sha1))
507 if (get_tree_entry(origin->commit->object.sha1,
509 origin->blob_sha1, &origin->mode))
511 if (sha1_object_info(origin->blob_sha1, NULL) != OBJ_BLOB)
515 hashclr(origin->blob_sha1);
516 origin->mode = S_IFINVALID;
521 * We have an origin -- check if the same path exists in the
522 * parent and return an origin structure to represent it.
524 static struct origin *find_origin(struct scoreboard *sb,
525 struct commit *parent,
526 struct origin *origin)
528 struct origin *porigin;
529 struct diff_options diff_opts;
530 const char *paths[2];
532 /* First check any existing origins */
533 for (porigin = parent->util; porigin; porigin = porigin->next)
534 if (!strcmp(porigin->path, origin->path)) {
536 * The same path between origin and its parent
537 * without renaming -- the most common case.
539 return origin_incref (porigin);
542 /* See if the origin->path is different between parent
543 * and origin first. Most of the time they are the
544 * same and diff-tree is fairly efficient about this.
546 diff_setup(&diff_opts);
547 DIFF_OPT_SET(&diff_opts, RECURSIVE);
548 diff_opts.detect_rename = 0;
549 diff_opts.output_format = DIFF_FORMAT_NO_OUTPUT;
550 paths[0] = origin->path;
553 parse_pathspec(&diff_opts.pathspec,
554 PATHSPEC_ALL_MAGIC & ~PATHSPEC_LITERAL,
555 PATHSPEC_LITERAL_PATH, "", paths);
556 diff_setup_done(&diff_opts);
558 if (is_null_sha1(origin->commit->object.sha1))
559 do_diff_cache(parent->tree->object.sha1, &diff_opts);
561 diff_tree_sha1(parent->tree->object.sha1,
562 origin->commit->tree->object.sha1,
564 diffcore_std(&diff_opts);
566 if (!diff_queued_diff.nr) {
567 /* The path is the same as parent */
568 porigin = get_origin(sb, parent, origin->path);
569 hashcpy(porigin->blob_sha1, origin->blob_sha1);
570 porigin->mode = origin->mode;
573 * Since origin->path is a pathspec, if the parent
574 * commit had it as a directory, we will see a whole
575 * bunch of deletion of files in the directory that we
579 struct diff_filepair *p = NULL;
580 for (i = 0; i < diff_queued_diff.nr; i++) {
582 p = diff_queued_diff.queue[i];
583 name = p->one->path ? p->one->path : p->two->path;
584 if (!strcmp(name, origin->path))
588 die("internal error in blame::find_origin");
591 die("internal error in blame::find_origin (%c)",
594 porigin = get_origin(sb, parent, origin->path);
595 hashcpy(porigin->blob_sha1, p->one->sha1);
596 porigin->mode = p->one->mode;
600 /* Did not exist in parent, or type changed */
604 diff_flush(&diff_opts);
605 free_pathspec(&diff_opts.pathspec);
610 * We have an origin -- find the path that corresponds to it in its
611 * parent and return an origin structure to represent it.
613 static struct origin *find_rename(struct scoreboard *sb,
614 struct commit *parent,
615 struct origin *origin)
617 struct origin *porigin = NULL;
618 struct diff_options diff_opts;
621 diff_setup(&diff_opts);
622 DIFF_OPT_SET(&diff_opts, RECURSIVE);
623 diff_opts.detect_rename = DIFF_DETECT_RENAME;
624 diff_opts.output_format = DIFF_FORMAT_NO_OUTPUT;
625 diff_opts.single_follow = origin->path;
626 diff_setup_done(&diff_opts);
628 if (is_null_sha1(origin->commit->object.sha1))
629 do_diff_cache(parent->tree->object.sha1, &diff_opts);
631 diff_tree_sha1(parent->tree->object.sha1,
632 origin->commit->tree->object.sha1,
634 diffcore_std(&diff_opts);
636 for (i = 0; i < diff_queued_diff.nr; i++) {
637 struct diff_filepair *p = diff_queued_diff.queue[i];
638 if ((p->status == 'R' || p->status == 'C') &&
639 !strcmp(p->two->path, origin->path)) {
640 porigin = get_origin(sb, parent, p->one->path);
641 hashcpy(porigin->blob_sha1, p->one->sha1);
642 porigin->mode = p->one->mode;
646 diff_flush(&diff_opts);
647 free_pathspec(&diff_opts.pathspec);
652 * Append a new blame entry to a given output queue.
654 static void add_blame_entry(struct blame_entry ***queue, struct blame_entry *e)
656 origin_incref(e->suspect);
664 * src typically is on-stack; we want to copy the information in it to
665 * a malloced blame_entry that gets added to the given queue. The
666 * origin of dst loses a refcnt.
668 static void dup_entry(struct blame_entry ***queue,
669 struct blame_entry *dst, struct blame_entry *src)
671 origin_incref(src->suspect);
672 origin_decref(dst->suspect);
673 memcpy(dst, src, sizeof(*src));
679 static const char *nth_line(struct scoreboard *sb, long lno)
681 return sb->final_buf + sb->lineno[lno];
684 static const char *nth_line_cb(void *data, long lno)
686 return nth_line((struct scoreboard *)data, lno);
690 * It is known that lines between tlno to same came from parent, and e
691 * has an overlap with that range. it also is known that parent's
692 * line plno corresponds to e's line tlno.
698 * <------------------>
700 * Split e into potentially three parts; before this chunk, the chunk
701 * to be blamed for the parent, and after that portion.
703 static void split_overlap(struct blame_entry *split,
704 struct blame_entry *e,
705 int tlno, int plno, int same,
706 struct origin *parent)
709 memset(split, 0, sizeof(struct blame_entry [3]));
711 if (e->s_lno < tlno) {
712 /* there is a pre-chunk part not blamed on parent */
713 split[0].suspect = origin_incref(e->suspect);
714 split[0].lno = e->lno;
715 split[0].s_lno = e->s_lno;
716 split[0].num_lines = tlno - e->s_lno;
717 split[1].lno = e->lno + tlno - e->s_lno;
718 split[1].s_lno = plno;
721 split[1].lno = e->lno;
722 split[1].s_lno = plno + (e->s_lno - tlno);
725 if (same < e->s_lno + e->num_lines) {
726 /* there is a post-chunk part not blamed on parent */
727 split[2].suspect = origin_incref(e->suspect);
728 split[2].lno = e->lno + (same - e->s_lno);
729 split[2].s_lno = e->s_lno + (same - e->s_lno);
730 split[2].num_lines = e->s_lno + e->num_lines - same;
731 chunk_end_lno = split[2].lno;
734 chunk_end_lno = e->lno + e->num_lines;
735 split[1].num_lines = chunk_end_lno - split[1].lno;
738 * if it turns out there is nothing to blame the parent for,
739 * forget about the splitting. !split[1].suspect signals this.
741 if (split[1].num_lines < 1)
743 split[1].suspect = origin_incref(parent);
747 * split_overlap() divided an existing blame e into up to three parts
748 * in split. Any assigned blame is moved to queue to
751 static void split_blame(struct blame_entry ***blamed,
752 struct blame_entry ***unblamed,
753 struct blame_entry *split,
754 struct blame_entry *e)
756 struct blame_entry *new_entry;
758 if (split[0].suspect && split[2].suspect) {
759 /* The first part (reuse storage for the existing entry e) */
760 dup_entry(unblamed, e, &split[0]);
762 /* The last part -- me */
763 new_entry = xmalloc(sizeof(*new_entry));
764 memcpy(new_entry, &(split[2]), sizeof(struct blame_entry));
765 add_blame_entry(unblamed, new_entry);
767 /* ... and the middle part -- parent */
768 new_entry = xmalloc(sizeof(*new_entry));
769 memcpy(new_entry, &(split[1]), sizeof(struct blame_entry));
770 add_blame_entry(blamed, new_entry);
772 else if (!split[0].suspect && !split[2].suspect)
774 * The parent covers the entire area; reuse storage for
775 * e and replace it with the parent.
777 dup_entry(blamed, e, &split[1]);
778 else if (split[0].suspect) {
779 /* me and then parent */
780 dup_entry(unblamed, e, &split[0]);
782 new_entry = xmalloc(sizeof(*new_entry));
783 memcpy(new_entry, &(split[1]), sizeof(struct blame_entry));
784 add_blame_entry(blamed, new_entry);
787 /* parent and then me */
788 dup_entry(blamed, e, &split[1]);
790 new_entry = xmalloc(sizeof(*new_entry));
791 memcpy(new_entry, &(split[2]), sizeof(struct blame_entry));
792 add_blame_entry(unblamed, new_entry);
797 * After splitting the blame, the origins used by the
798 * on-stack blame_entry should lose one refcnt each.
800 static void decref_split(struct blame_entry *split)
804 for (i = 0; i < 3; i++)
805 origin_decref(split[i].suspect);
809 * reverse_blame reverses the list given in head, appending tail.
810 * That allows us to build lists in reverse order, then reverse them
811 * afterwards. This can be faster than building the list in proper
812 * order right away. The reason is that building in proper order
813 * requires writing a link in the _previous_ element, while building
814 * in reverse order just requires placing the list head into the
818 static struct blame_entry *reverse_blame(struct blame_entry *head,
819 struct blame_entry *tail)
822 struct blame_entry *next = head->next;
831 * Process one hunk from the patch between the current suspect for
832 * blame_entry e and its parent. This first blames any unfinished
833 * entries before the chunk (which is where target and parent start
834 * differing) on the parent, and then splits blame entries at the
835 * start and at the end of the difference region. Since use of -M and
836 * -C options may lead to overlapping/duplicate source line number
837 * ranges, all we can rely on from sorting/merging is the order of the
838 * first suspect line number.
840 static void blame_chunk(struct blame_entry ***dstq, struct blame_entry ***srcq,
841 int tlno, int offset, int same,
842 struct origin *parent)
844 struct blame_entry *e = **srcq;
845 struct blame_entry *samep = NULL, *diffp = NULL;
847 while (e && e->s_lno < tlno) {
848 struct blame_entry *next = e->next;
850 * current record starts before differing portion. If
851 * it reaches into it, we need to split it up and
852 * examine the second part separately.
854 if (e->s_lno + e->num_lines > tlno) {
855 /* Move second half to a new record */
856 int len = tlno - e->s_lno;
857 struct blame_entry *n = xcalloc(1, sizeof (struct blame_entry));
858 n->suspect = e->suspect;
859 n->lno = e->lno + len;
860 n->s_lno = e->s_lno + len;
861 n->num_lines = e->num_lines - len;
864 /* Push new record to diffp */
868 origin_decref(e->suspect);
869 /* Pass blame for everything before the differing
870 * chunk to the parent */
871 e->suspect = origin_incref(parent);
878 * As we don't know how much of a common stretch after this
879 * diff will occur, the currently blamed parts are all that we
880 * can assign to the parent for now.
884 **dstq = reverse_blame(samep, **dstq);
885 *dstq = &samep->next;
888 * Prepend the split off portions: everything after e starts
889 * after the blameable portion.
891 e = reverse_blame(diffp, e);
894 * Now retain records on the target while parts are different
899 while (e && e->s_lno < same) {
900 struct blame_entry *next = e->next;
903 * If current record extends into sameness, need to split.
905 if (e->s_lno + e->num_lines > same) {
907 * Move second half to a new record to be
908 * processed by later chunks
910 int len = same - e->s_lno;
911 struct blame_entry *n = xcalloc(1, sizeof (struct blame_entry));
912 n->suspect = origin_incref(e->suspect);
913 n->lno = e->lno + len;
914 n->s_lno = e->s_lno + len;
915 n->num_lines = e->num_lines - len;
918 /* Push new record to samep */
926 **srcq = reverse_blame(diffp, reverse_blame(samep, e));
927 /* Move across elements that are in the unblamable portion */
929 *srcq = &diffp->next;
932 struct blame_chunk_cb_data {
933 struct origin *parent;
935 struct blame_entry **dstq;
936 struct blame_entry **srcq;
939 /* diff chunks are from parent to target */
940 static int blame_chunk_cb(long start_a, long count_a,
941 long start_b, long count_b, void *data)
943 struct blame_chunk_cb_data *d = data;
944 if (start_a - start_b != d->offset)
945 die("internal error in blame::blame_chunk_cb");
946 blame_chunk(&d->dstq, &d->srcq, start_b, start_a - start_b,
947 start_b + count_b, d->parent);
948 d->offset = start_a + count_a - (start_b + count_b);
953 * We are looking at the origin 'target' and aiming to pass blame
954 * for the lines it is suspected to its parent. Run diff to find
955 * which lines came from parent and pass blame for them.
957 static void pass_blame_to_parent(struct scoreboard *sb,
958 struct origin *target,
959 struct origin *parent)
961 mmfile_t file_p, file_o;
962 struct blame_chunk_cb_data d;
963 struct blame_entry *newdest = NULL;
965 if (!target->suspects)
966 return; /* nothing remains for this target */
970 d.dstq = &newdest; d.srcq = &target->suspects;
972 fill_origin_blob(&sb->revs->diffopt, parent, &file_p);
973 fill_origin_blob(&sb->revs->diffopt, target, &file_o);
976 diff_hunks(&file_p, &file_o, 0, blame_chunk_cb, &d);
977 /* The rest are the same as the parent */
978 blame_chunk(&d.dstq, &d.srcq, INT_MAX, d.offset, INT_MAX, parent);
980 queue_blames(sb, parent, newdest);
986 * The lines in blame_entry after splitting blames many times can become
987 * very small and trivial, and at some point it becomes pointless to
988 * blame the parents. E.g. "\t\t}\n\t}\n\n" appears everywhere in any
989 * ordinary C program, and it is not worth to say it was copied from
990 * totally unrelated file in the parent.
992 * Compute how trivial the lines in the blame_entry are.
994 static unsigned ent_score(struct scoreboard *sb, struct blame_entry *e)
1003 cp = nth_line(sb, e->lno);
1004 ep = nth_line(sb, e->lno + e->num_lines);
1006 unsigned ch = *((unsigned char *)cp);
1016 * best_so_far[] and this[] are both a split of an existing blame_entry
1017 * that passes blame to the parent. Maintain best_so_far the best split
1018 * so far, by comparing this and best_so_far and copying this into
1019 * bst_so_far as needed.
1021 static void copy_split_if_better(struct scoreboard *sb,
1022 struct blame_entry *best_so_far,
1023 struct blame_entry *this)
1027 if (!this[1].suspect)
1029 if (best_so_far[1].suspect) {
1030 if (ent_score(sb, &this[1]) < ent_score(sb, &best_so_far[1]))
1034 for (i = 0; i < 3; i++)
1035 origin_incref(this[i].suspect);
1036 decref_split(best_so_far);
1037 memcpy(best_so_far, this, sizeof(struct blame_entry [3]));
1041 * We are looking at a part of the final image represented by
1042 * ent (tlno and same are offset by ent->s_lno).
1043 * tlno is where we are looking at in the final image.
1044 * up to (but not including) same match preimage.
1045 * plno is where we are looking at in the preimage.
1047 * <-------------- final image ---------------------->
1050 * <---------preimage----->
1053 * All line numbers are 0-based.
1055 static void handle_split(struct scoreboard *sb,
1056 struct blame_entry *ent,
1057 int tlno, int plno, int same,
1058 struct origin *parent,
1059 struct blame_entry *split)
1061 if (ent->num_lines <= tlno)
1064 struct blame_entry this[3];
1067 split_overlap(this, ent, tlno, plno, same, parent);
1068 copy_split_if_better(sb, split, this);
1073 struct handle_split_cb_data {
1074 struct scoreboard *sb;
1075 struct blame_entry *ent;
1076 struct origin *parent;
1077 struct blame_entry *split;
1082 static int handle_split_cb(long start_a, long count_a,
1083 long start_b, long count_b, void *data)
1085 struct handle_split_cb_data *d = data;
1086 handle_split(d->sb, d->ent, d->tlno, d->plno, start_b, d->parent,
1088 d->plno = start_a + count_a;
1089 d->tlno = start_b + count_b;
1094 * Find the lines from parent that are the same as ent so that
1095 * we can pass blames to it. file_p has the blob contents for
1098 static void find_copy_in_blob(struct scoreboard *sb,
1099 struct blame_entry *ent,
1100 struct origin *parent,
1101 struct blame_entry *split,
1106 struct handle_split_cb_data d;
1108 memset(&d, 0, sizeof(d));
1109 d.sb = sb; d.ent = ent; d.parent = parent; d.split = split;
1111 * Prepare mmfile that contains only the lines in ent.
1113 cp = nth_line(sb, ent->lno);
1114 file_o.ptr = (char *) cp;
1115 file_o.size = nth_line(sb, ent->lno + ent->num_lines) - cp;
1118 * file_o is a part of final image we are annotating.
1119 * file_p partially may match that image.
1121 memset(split, 0, sizeof(struct blame_entry [3]));
1122 diff_hunks(file_p, &file_o, 1, handle_split_cb, &d);
1123 /* remainder, if any, all match the preimage */
1124 handle_split(sb, ent, d.tlno, d.plno, ent->num_lines, parent, split);
1127 /* Move all blame entries from list *source that have a score smaller
1128 * than score_min to the front of list *small.
1129 * Returns a pointer to the link pointing to the old head of the small list.
1132 static struct blame_entry **filter_small(struct scoreboard *sb,
1133 struct blame_entry **small,
1134 struct blame_entry **source,
1137 struct blame_entry *p = *source;
1138 struct blame_entry *oldsmall = *small;
1140 if (ent_score(sb, p) <= score_min) {
1156 * See if lines currently target is suspected for can be attributed to
1159 static void find_move_in_parent(struct scoreboard *sb,
1160 struct blame_entry ***blamed,
1161 struct blame_entry **toosmall,
1162 struct origin *target,
1163 struct origin *parent)
1165 struct blame_entry *e, split[3];
1166 struct blame_entry *unblamed = target->suspects;
1167 struct blame_entry *leftover = NULL;
1171 return; /* nothing remains for this target */
1173 fill_origin_blob(&sb->revs->diffopt, parent, &file_p);
1177 /* At each iteration, unblamed has a NULL-terminated list of
1178 * entries that have not yet been tested for blame. leftover
1179 * contains the reversed list of entries that have been tested
1180 * without being assignable to the parent.
1183 struct blame_entry **unblamedtail = &unblamed;
1184 struct blame_entry *next;
1185 for (e = unblamed; e; e = next) {
1187 find_copy_in_blob(sb, e, parent, split, &file_p);
1188 if (split[1].suspect &&
1189 blame_move_score < ent_score(sb, &split[1])) {
1190 split_blame(blamed, &unblamedtail, split, e);
1195 decref_split(split);
1197 *unblamedtail = NULL;
1198 toosmall = filter_small(sb, toosmall, &unblamed, blame_move_score);
1200 target->suspects = reverse_blame(leftover, NULL);
1204 struct blame_entry *ent;
1205 struct blame_entry split[3];
1209 * Count the number of entries the target is suspected for,
1210 * and prepare a list of entry and the best split.
1212 static struct blame_list *setup_blame_list(struct blame_entry *unblamed,
1215 struct blame_entry *e;
1217 struct blame_list *blame_list = NULL;
1219 for (e = unblamed, num_ents = 0; e; e = e->next)
1222 blame_list = xcalloc(num_ents, sizeof(struct blame_list));
1223 for (e = unblamed, i = 0; e; e = e->next)
1224 blame_list[i++].ent = e;
1226 *num_ents_p = num_ents;
1231 * For lines target is suspected for, see if we can find code movement
1232 * across file boundary from the parent commit. porigin is the path
1233 * in the parent we already tried.
1235 static void find_copy_in_parent(struct scoreboard *sb,
1236 struct blame_entry ***blamed,
1237 struct blame_entry **toosmall,
1238 struct origin *target,
1239 struct commit *parent,
1240 struct origin *porigin,
1243 struct diff_options diff_opts;
1245 struct blame_list *blame_list;
1247 struct blame_entry *unblamed = target->suspects;
1248 struct blame_entry *leftover = NULL;
1251 return; /* nothing remains for this target */
1253 diff_setup(&diff_opts);
1254 DIFF_OPT_SET(&diff_opts, RECURSIVE);
1255 diff_opts.output_format = DIFF_FORMAT_NO_OUTPUT;
1257 diff_setup_done(&diff_opts);
1259 /* Try "find copies harder" on new path if requested;
1260 * we do not want to use diffcore_rename() actually to
1261 * match things up; find_copies_harder is set only to
1262 * force diff_tree_sha1() to feed all filepairs to diff_queue,
1263 * and this code needs to be after diff_setup_done(), which
1264 * usually makes find-copies-harder imply copy detection.
1266 if ((opt & PICKAXE_BLAME_COPY_HARDEST)
1267 || ((opt & PICKAXE_BLAME_COPY_HARDER)
1268 && (!porigin || strcmp(target->path, porigin->path))))
1269 DIFF_OPT_SET(&diff_opts, FIND_COPIES_HARDER);
1271 if (is_null_sha1(target->commit->object.sha1))
1272 do_diff_cache(parent->tree->object.sha1, &diff_opts);
1274 diff_tree_sha1(parent->tree->object.sha1,
1275 target->commit->tree->object.sha1,
1278 if (!DIFF_OPT_TST(&diff_opts, FIND_COPIES_HARDER))
1279 diffcore_std(&diff_opts);
1282 struct blame_entry **unblamedtail = &unblamed;
1283 blame_list = setup_blame_list(unblamed, &num_ents);
1285 for (i = 0; i < diff_queued_diff.nr; i++) {
1286 struct diff_filepair *p = diff_queued_diff.queue[i];
1287 struct origin *norigin;
1289 struct blame_entry this[3];
1291 if (!DIFF_FILE_VALID(p->one))
1292 continue; /* does not exist in parent */
1293 if (S_ISGITLINK(p->one->mode))
1294 continue; /* ignore git links */
1295 if (porigin && !strcmp(p->one->path, porigin->path))
1296 /* find_move already dealt with this path */
1299 norigin = get_origin(sb, parent, p->one->path);
1300 hashcpy(norigin->blob_sha1, p->one->sha1);
1301 norigin->mode = p->one->mode;
1302 fill_origin_blob(&sb->revs->diffopt, norigin, &file_p);
1306 for (j = 0; j < num_ents; j++) {
1307 find_copy_in_blob(sb, blame_list[j].ent,
1308 norigin, this, &file_p);
1309 copy_split_if_better(sb, blame_list[j].split,
1313 origin_decref(norigin);
1316 for (j = 0; j < num_ents; j++) {
1317 struct blame_entry *split = blame_list[j].split;
1318 if (split[1].suspect &&
1319 blame_copy_score < ent_score(sb, &split[1])) {
1320 split_blame(blamed, &unblamedtail, split,
1323 blame_list[j].ent->next = leftover;
1324 leftover = blame_list[j].ent;
1326 decref_split(split);
1329 *unblamedtail = NULL;
1330 toosmall = filter_small(sb, toosmall, &unblamed, blame_copy_score);
1332 target->suspects = reverse_blame(leftover, NULL);
1333 diff_flush(&diff_opts);
1334 free_pathspec(&diff_opts.pathspec);
1338 * The blobs of origin and porigin exactly match, so everything
1339 * origin is suspected for can be blamed on the parent.
1341 static void pass_whole_blame(struct scoreboard *sb,
1342 struct origin *origin, struct origin *porigin)
1344 struct blame_entry *e, *suspects;
1346 if (!porigin->file.ptr && origin->file.ptr) {
1347 /* Steal its file */
1348 porigin->file = origin->file;
1349 origin->file.ptr = NULL;
1351 suspects = origin->suspects;
1352 origin->suspects = NULL;
1353 for (e = suspects; e; e = e->next) {
1354 origin_incref(porigin);
1355 origin_decref(e->suspect);
1356 e->suspect = porigin;
1358 queue_blames(sb, porigin, suspects);
1362 * We pass blame from the current commit to its parents. We keep saying
1363 * "parent" (and "porigin"), but what we mean is to find scapegoat to
1364 * exonerate ourselves.
1366 static struct commit_list *first_scapegoat(struct rev_info *revs, struct commit *commit)
1369 if (revs->first_parent_only &&
1371 commit->parents->next) {
1372 free_commit_list(commit->parents->next);
1373 commit->parents->next = NULL;
1375 return commit->parents;
1377 return lookup_decoration(&revs->children, &commit->object);
1380 static int num_scapegoats(struct rev_info *revs, struct commit *commit)
1382 struct commit_list *l = first_scapegoat(revs, commit);
1383 return commit_list_count(l);
1386 /* Distribute collected unsorted blames to the respected sorted lists
1387 * in the various origins.
1389 static void distribute_blame(struct scoreboard *sb, struct blame_entry *blamed)
1391 blamed = blame_sort(blamed, compare_blame_suspect);
1394 struct origin *porigin = blamed->suspect;
1395 struct blame_entry *suspects = NULL;
1397 struct blame_entry *next = blamed->next;
1398 blamed->next = suspects;
1401 } while (blamed && blamed->suspect == porigin);
1402 suspects = reverse_blame(suspects, NULL);
1403 queue_blames(sb, porigin, suspects);
1409 static void pass_blame(struct scoreboard *sb, struct origin *origin, int opt)
1411 struct rev_info *revs = sb->revs;
1412 int i, pass, num_sg;
1413 struct commit *commit = origin->commit;
1414 struct commit_list *sg;
1415 struct origin *sg_buf[MAXSG];
1416 struct origin *porigin, **sg_origin = sg_buf;
1417 struct blame_entry *toosmall = NULL;
1418 struct blame_entry *blames, **blametail = &blames;
1420 num_sg = num_scapegoats(revs, commit);
1423 else if (num_sg < ARRAY_SIZE(sg_buf))
1424 memset(sg_buf, 0, sizeof(sg_buf));
1426 sg_origin = xcalloc(num_sg, sizeof(*sg_origin));
1429 * The first pass looks for unrenamed path to optimize for
1430 * common cases, then we look for renames in the second pass.
1432 for (pass = 0; pass < 2 - no_whole_file_rename; pass++) {
1433 struct origin *(*find)(struct scoreboard *,
1434 struct commit *, struct origin *);
1435 find = pass ? find_rename : find_origin;
1437 for (i = 0, sg = first_scapegoat(revs, commit);
1439 sg = sg->next, i++) {
1440 struct commit *p = sg->item;
1445 if (parse_commit(p))
1447 porigin = find(sb, p, origin);
1450 if (!hashcmp(porigin->blob_sha1, origin->blob_sha1)) {
1451 pass_whole_blame(sb, origin, porigin);
1452 origin_decref(porigin);
1455 for (j = same = 0; j < i; j++)
1457 !hashcmp(sg_origin[j]->blob_sha1,
1458 porigin->blob_sha1)) {
1463 sg_origin[i] = porigin;
1465 origin_decref(porigin);
1470 for (i = 0, sg = first_scapegoat(revs, commit);
1472 sg = sg->next, i++) {
1473 struct origin *porigin = sg_origin[i];
1476 if (!origin->previous) {
1477 origin_incref(porigin);
1478 origin->previous = porigin;
1480 pass_blame_to_parent(sb, origin, porigin);
1481 if (!origin->suspects)
1486 * Optionally find moves in parents' files.
1488 if (opt & PICKAXE_BLAME_MOVE) {
1489 filter_small(sb, &toosmall, &origin->suspects, blame_move_score);
1490 if (origin->suspects) {
1491 for (i = 0, sg = first_scapegoat(revs, commit);
1493 sg = sg->next, i++) {
1494 struct origin *porigin = sg_origin[i];
1497 find_move_in_parent(sb, &blametail, &toosmall, origin, porigin);
1498 if (!origin->suspects)
1505 * Optionally find copies from parents' files.
1507 if (opt & PICKAXE_BLAME_COPY) {
1508 if (blame_copy_score > blame_move_score)
1509 filter_small(sb, &toosmall, &origin->suspects, blame_copy_score);
1510 else if (blame_copy_score < blame_move_score) {
1511 origin->suspects = blame_merge(origin->suspects, toosmall);
1513 filter_small(sb, &toosmall, &origin->suspects, blame_copy_score);
1515 if (!origin->suspects)
1518 for (i = 0, sg = first_scapegoat(revs, commit);
1520 sg = sg->next, i++) {
1521 struct origin *porigin = sg_origin[i];
1522 find_copy_in_parent(sb, &blametail, &toosmall,
1523 origin, sg->item, porigin, opt);
1524 if (!origin->suspects)
1531 distribute_blame(sb, blames);
1533 * prepend toosmall to origin->suspects
1535 * There is no point in sorting: this ends up on a big
1536 * unsorted list in the caller anyway.
1539 struct blame_entry **tail = &toosmall;
1541 tail = &(*tail)->next;
1542 *tail = origin->suspects;
1543 origin->suspects = toosmall;
1545 for (i = 0; i < num_sg; i++) {
1547 drop_origin_blob(sg_origin[i]);
1548 origin_decref(sg_origin[i]);
1551 drop_origin_blob(origin);
1552 if (sg_buf != sg_origin)
1557 * Information on commits, used for output.
1559 struct commit_info {
1560 struct strbuf author;
1561 struct strbuf author_mail;
1562 unsigned long author_time;
1563 struct strbuf author_tz;
1565 /* filled only when asked for details */
1566 struct strbuf committer;
1567 struct strbuf committer_mail;
1568 unsigned long committer_time;
1569 struct strbuf committer_tz;
1571 struct strbuf summary;
1575 * Parse author/committer line in the commit object buffer
1577 static void get_ac_line(const char *inbuf, const char *what,
1578 struct strbuf *name, struct strbuf *mail,
1579 unsigned long *time, struct strbuf *tz)
1581 struct ident_split ident;
1582 size_t len, maillen, namelen;
1584 const char *namebuf, *mailbuf;
1586 tmp = strstr(inbuf, what);
1589 tmp += strlen(what);
1590 endp = strchr(tmp, '\n');
1596 if (split_ident_line(&ident, tmp, len)) {
1600 strbuf_addstr(name, tmp);
1601 strbuf_addstr(mail, tmp);
1602 strbuf_addstr(tz, tmp);
1607 namelen = ident.name_end - ident.name_begin;
1608 namebuf = ident.name_begin;
1610 maillen = ident.mail_end - ident.mail_begin;
1611 mailbuf = ident.mail_begin;
1613 if (ident.date_begin && ident.date_end)
1614 *time = strtoul(ident.date_begin, NULL, 10);
1618 if (ident.tz_begin && ident.tz_end)
1619 strbuf_add(tz, ident.tz_begin, ident.tz_end - ident.tz_begin);
1621 strbuf_addstr(tz, "(unknown)");
1624 * Now, convert both name and e-mail using mailmap
1626 map_user(&mailmap, &mailbuf, &maillen,
1627 &namebuf, &namelen);
1629 strbuf_addf(mail, "<%.*s>", (int)maillen, mailbuf);
1630 strbuf_add(name, namebuf, namelen);
1633 static void commit_info_init(struct commit_info *ci)
1636 strbuf_init(&ci->author, 0);
1637 strbuf_init(&ci->author_mail, 0);
1638 strbuf_init(&ci->author_tz, 0);
1639 strbuf_init(&ci->committer, 0);
1640 strbuf_init(&ci->committer_mail, 0);
1641 strbuf_init(&ci->committer_tz, 0);
1642 strbuf_init(&ci->summary, 0);
1645 static void commit_info_destroy(struct commit_info *ci)
1648 strbuf_release(&ci->author);
1649 strbuf_release(&ci->author_mail);
1650 strbuf_release(&ci->author_tz);
1651 strbuf_release(&ci->committer);
1652 strbuf_release(&ci->committer_mail);
1653 strbuf_release(&ci->committer_tz);
1654 strbuf_release(&ci->summary);
1657 static void get_commit_info(struct commit *commit,
1658 struct commit_info *ret,
1662 const char *subject, *encoding;
1663 const char *message;
1665 commit_info_init(ret);
1667 encoding = get_log_output_encoding();
1668 message = logmsg_reencode(commit, NULL, encoding);
1669 get_ac_line(message, "\nauthor ",
1670 &ret->author, &ret->author_mail,
1671 &ret->author_time, &ret->author_tz);
1674 unuse_commit_buffer(commit, message);
1678 get_ac_line(message, "\ncommitter ",
1679 &ret->committer, &ret->committer_mail,
1680 &ret->committer_time, &ret->committer_tz);
1682 len = find_commit_subject(message, &subject);
1684 strbuf_add(&ret->summary, subject, len);
1686 strbuf_addf(&ret->summary, "(%s)", sha1_to_hex(commit->object.sha1));
1688 unuse_commit_buffer(commit, message);
1692 * To allow LF and other nonportable characters in pathnames,
1693 * they are c-style quoted as needed.
1695 static void write_filename_info(const char *path)
1697 printf("filename ");
1698 write_name_quoted(path, stdout, '\n');
1702 * Porcelain/Incremental format wants to show a lot of details per
1703 * commit. Instead of repeating this every line, emit it only once,
1704 * the first time each commit appears in the output (unless the
1705 * user has specifically asked for us to repeat).
1707 static int emit_one_suspect_detail(struct origin *suspect, int repeat)
1709 struct commit_info ci;
1711 if (!repeat && (suspect->commit->object.flags & METAINFO_SHOWN))
1714 suspect->commit->object.flags |= METAINFO_SHOWN;
1715 get_commit_info(suspect->commit, &ci, 1);
1716 printf("author %s\n", ci.author.buf);
1717 printf("author-mail %s\n", ci.author_mail.buf);
1718 printf("author-time %lu\n", ci.author_time);
1719 printf("author-tz %s\n", ci.author_tz.buf);
1720 printf("committer %s\n", ci.committer.buf);
1721 printf("committer-mail %s\n", ci.committer_mail.buf);
1722 printf("committer-time %lu\n", ci.committer_time);
1723 printf("committer-tz %s\n", ci.committer_tz.buf);
1724 printf("summary %s\n", ci.summary.buf);
1725 if (suspect->commit->object.flags & UNINTERESTING)
1726 printf("boundary\n");
1727 if (suspect->previous) {
1728 struct origin *prev = suspect->previous;
1729 printf("previous %s ", sha1_to_hex(prev->commit->object.sha1));
1730 write_name_quoted(prev->path, stdout, '\n');
1733 commit_info_destroy(&ci);
1739 * The blame_entry is found to be guilty for the range.
1740 * Show it in incremental output.
1742 static void found_guilty_entry(struct blame_entry *ent)
1745 struct origin *suspect = ent->suspect;
1747 printf("%s %d %d %d\n",
1748 sha1_to_hex(suspect->commit->object.sha1),
1749 ent->s_lno + 1, ent->lno + 1, ent->num_lines);
1750 emit_one_suspect_detail(suspect, 0);
1751 write_filename_info(suspect->path);
1752 maybe_flush_or_die(stdout, "stdout");
1757 * The main loop -- while we have blobs with lines whose true origin
1758 * is still unknown, pick one blob, and allow its lines to pass blames
1759 * to its parents. */
1760 static void assign_blame(struct scoreboard *sb, int opt)
1762 struct rev_info *revs = sb->revs;
1763 struct commit *commit = prio_queue_get(&sb->commits);
1766 struct blame_entry *ent;
1767 struct origin *suspect = commit->util;
1769 /* find one suspect to break down */
1770 while (suspect && !suspect->suspects)
1771 suspect = suspect->next;
1774 commit = prio_queue_get(&sb->commits);
1778 assert(commit == suspect->commit);
1781 * We will use this suspect later in the loop,
1782 * so hold onto it in the meantime.
1784 origin_incref(suspect);
1785 parse_commit(commit);
1787 (!(commit->object.flags & UNINTERESTING) &&
1788 !(revs->max_age != -1 && commit->date < revs->max_age)))
1789 pass_blame(sb, suspect, opt);
1791 commit->object.flags |= UNINTERESTING;
1792 if (commit->object.parsed)
1793 mark_parents_uninteresting(commit);
1795 /* treat root commit as boundary */
1796 if (!commit->parents && !show_root)
1797 commit->object.flags |= UNINTERESTING;
1799 /* Take responsibility for the remaining entries */
1800 ent = suspect->suspects;
1802 suspect->guilty = 1;
1804 struct blame_entry *next = ent->next;
1805 found_guilty_entry(ent);
1810 ent->next = sb->ent;
1811 sb->ent = suspect->suspects;
1812 suspect->suspects = NULL;
1816 origin_decref(suspect);
1818 if (DEBUG) /* sanity */
1819 sanity_check_refcnt(sb);
1823 static const char *format_time(unsigned long time, const char *tz_str,
1826 static struct strbuf time_buf = STRBUF_INIT;
1828 strbuf_reset(&time_buf);
1829 if (show_raw_time) {
1830 strbuf_addf(&time_buf, "%lu %s", time, tz_str);
1833 const char *time_str;
1837 time_str = show_date(time, tz, blame_date_mode);
1838 strbuf_addstr(&time_buf, time_str);
1840 * Add space paddings to time_buf to display a fixed width
1841 * string, and use time_width for display width calibration.
1843 for (time_width = utf8_strwidth(time_str);
1844 time_width < blame_date_width;
1846 strbuf_addch(&time_buf, ' ');
1848 return time_buf.buf;
1851 #define OUTPUT_ANNOTATE_COMPAT 001
1852 #define OUTPUT_LONG_OBJECT_NAME 002
1853 #define OUTPUT_RAW_TIMESTAMP 004
1854 #define OUTPUT_PORCELAIN 010
1855 #define OUTPUT_SHOW_NAME 020
1856 #define OUTPUT_SHOW_NUMBER 040
1857 #define OUTPUT_SHOW_SCORE 0100
1858 #define OUTPUT_NO_AUTHOR 0200
1859 #define OUTPUT_SHOW_EMAIL 0400
1860 #define OUTPUT_LINE_PORCELAIN 01000
1862 static void emit_porcelain_details(struct origin *suspect, int repeat)
1864 if (emit_one_suspect_detail(suspect, repeat) ||
1865 (suspect->commit->object.flags & MORE_THAN_ONE_PATH))
1866 write_filename_info(suspect->path);
1869 static void emit_porcelain(struct scoreboard *sb, struct blame_entry *ent,
1872 int repeat = opt & OUTPUT_LINE_PORCELAIN;
1875 struct origin *suspect = ent->suspect;
1878 strcpy(hex, sha1_to_hex(suspect->commit->object.sha1));
1879 printf("%s %d %d %d\n",
1884 emit_porcelain_details(suspect, repeat);
1886 cp = nth_line(sb, ent->lno);
1887 for (cnt = 0; cnt < ent->num_lines; cnt++) {
1890 printf("%s %d %d\n", hex,
1891 ent->s_lno + 1 + cnt,
1892 ent->lno + 1 + cnt);
1894 emit_porcelain_details(suspect, 1);
1900 } while (ch != '\n' &&
1901 cp < sb->final_buf + sb->final_buf_size);
1904 if (sb->final_buf_size && cp[-1] != '\n')
1908 static void emit_other(struct scoreboard *sb, struct blame_entry *ent, int opt)
1912 struct origin *suspect = ent->suspect;
1913 struct commit_info ci;
1915 int show_raw_time = !!(opt & OUTPUT_RAW_TIMESTAMP);
1917 get_commit_info(suspect->commit, &ci, 1);
1918 strcpy(hex, sha1_to_hex(suspect->commit->object.sha1));
1920 cp = nth_line(sb, ent->lno);
1921 for (cnt = 0; cnt < ent->num_lines; cnt++) {
1923 int length = (opt & OUTPUT_LONG_OBJECT_NAME) ? 40 : abbrev;
1925 if (suspect->commit->object.flags & UNINTERESTING) {
1927 memset(hex, ' ', length);
1928 else if (!(opt & OUTPUT_ANNOTATE_COMPAT)) {
1934 printf("%.*s", length, hex);
1935 if (opt & OUTPUT_ANNOTATE_COMPAT) {
1937 if (opt & OUTPUT_SHOW_EMAIL)
1938 name = ci.author_mail.buf;
1940 name = ci.author.buf;
1941 printf("\t(%10s\t%10s\t%d)", name,
1942 format_time(ci.author_time, ci.author_tz.buf,
1944 ent->lno + 1 + cnt);
1946 if (opt & OUTPUT_SHOW_SCORE)
1948 max_score_digits, ent->score,
1949 ent->suspect->refcnt);
1950 if (opt & OUTPUT_SHOW_NAME)
1951 printf(" %-*.*s", longest_file, longest_file,
1953 if (opt & OUTPUT_SHOW_NUMBER)
1954 printf(" %*d", max_orig_digits,
1955 ent->s_lno + 1 + cnt);
1957 if (!(opt & OUTPUT_NO_AUTHOR)) {
1960 if (opt & OUTPUT_SHOW_EMAIL)
1961 name = ci.author_mail.buf;
1963 name = ci.author.buf;
1964 pad = longest_author - utf8_strwidth(name);
1965 printf(" (%s%*s %10s",
1967 format_time(ci.author_time,
1972 max_digits, ent->lno + 1 + cnt);
1977 } while (ch != '\n' &&
1978 cp < sb->final_buf + sb->final_buf_size);
1981 if (sb->final_buf_size && cp[-1] != '\n')
1984 commit_info_destroy(&ci);
1987 static void output(struct scoreboard *sb, int option)
1989 struct blame_entry *ent;
1991 if (option & OUTPUT_PORCELAIN) {
1992 for (ent = sb->ent; ent; ent = ent->next) {
1994 struct origin *suspect;
1995 struct commit *commit = ent->suspect->commit;
1996 if (commit->object.flags & MORE_THAN_ONE_PATH)
1998 for (suspect = commit->util; suspect; suspect = suspect->next) {
1999 if (suspect->guilty && count++) {
2000 commit->object.flags |= MORE_THAN_ONE_PATH;
2007 for (ent = sb->ent; ent; ent = ent->next) {
2008 if (option & OUTPUT_PORCELAIN)
2009 emit_porcelain(sb, ent, option);
2011 emit_other(sb, ent, option);
2016 static const char *get_next_line(const char *start, const char *end)
2018 const char *nl = memchr(start, '\n', end - start);
2019 return nl ? nl + 1 : end;
2023 * To allow quick access to the contents of nth line in the
2024 * final image, prepare an index in the scoreboard.
2026 static int prepare_lines(struct scoreboard *sb)
2028 const char *buf = sb->final_buf;
2029 unsigned long len = sb->final_buf_size;
2030 const char *end = buf + len;
2035 for (p = buf; p < end; p = get_next_line(p, end))
2038 sb->lineno = lineno = xmalloc(sizeof(*sb->lineno) * (num + 1));
2040 for (p = buf; p < end; p = get_next_line(p, end))
2041 *lineno++ = p - buf;
2045 sb->num_lines = num;
2046 return sb->num_lines;
2050 * Add phony grafts for use with -S; this is primarily to
2051 * support git's cvsserver that wants to give a linear history
2054 static int read_ancestry(const char *graft_file)
2056 FILE *fp = fopen(graft_file, "r");
2057 struct strbuf buf = STRBUF_INIT;
2060 while (!strbuf_getwholeline(&buf, fp, '\n')) {
2061 /* The format is just "Commit Parent1 Parent2 ...\n" */
2062 struct commit_graft *graft = read_graft_line(buf.buf, buf.len);
2064 register_commit_graft(graft, 0);
2067 strbuf_release(&buf);
2071 static int update_auto_abbrev(int auto_abbrev, struct origin *suspect)
2073 const char *uniq = find_unique_abbrev(suspect->commit->object.sha1,
2075 int len = strlen(uniq);
2076 if (auto_abbrev < len)
2082 * How many columns do we need to show line numbers, authors,
2085 static void find_alignment(struct scoreboard *sb, int *option)
2087 int longest_src_lines = 0;
2088 int longest_dst_lines = 0;
2089 unsigned largest_score = 0;
2090 struct blame_entry *e;
2091 int compute_auto_abbrev = (abbrev < 0);
2092 int auto_abbrev = default_abbrev;
2094 for (e = sb->ent; e; e = e->next) {
2095 struct origin *suspect = e->suspect;
2098 if (compute_auto_abbrev)
2099 auto_abbrev = update_auto_abbrev(auto_abbrev, suspect);
2100 if (strcmp(suspect->path, sb->path))
2101 *option |= OUTPUT_SHOW_NAME;
2102 num = strlen(suspect->path);
2103 if (longest_file < num)
2105 if (!(suspect->commit->object.flags & METAINFO_SHOWN)) {
2106 struct commit_info ci;
2107 suspect->commit->object.flags |= METAINFO_SHOWN;
2108 get_commit_info(suspect->commit, &ci, 1);
2109 if (*option & OUTPUT_SHOW_EMAIL)
2110 num = utf8_strwidth(ci.author_mail.buf);
2112 num = utf8_strwidth(ci.author.buf);
2113 if (longest_author < num)
2114 longest_author = num;
2115 commit_info_destroy(&ci);
2117 num = e->s_lno + e->num_lines;
2118 if (longest_src_lines < num)
2119 longest_src_lines = num;
2120 num = e->lno + e->num_lines;
2121 if (longest_dst_lines < num)
2122 longest_dst_lines = num;
2123 if (largest_score < ent_score(sb, e))
2124 largest_score = ent_score(sb, e);
2126 max_orig_digits = decimal_width(longest_src_lines);
2127 max_digits = decimal_width(longest_dst_lines);
2128 max_score_digits = decimal_width(largest_score);
2130 if (compute_auto_abbrev)
2131 /* one more abbrev length is needed for the boundary commit */
2132 abbrev = auto_abbrev + 1;
2136 * For debugging -- origin is refcounted, and this asserts that
2137 * we do not underflow.
2139 static void sanity_check_refcnt(struct scoreboard *sb)
2142 struct blame_entry *ent;
2144 for (ent = sb->ent; ent; ent = ent->next) {
2145 /* Nobody should have zero or negative refcnt */
2146 if (ent->suspect->refcnt <= 0) {
2147 fprintf(stderr, "%s in %s has negative refcnt %d\n",
2149 sha1_to_hex(ent->suspect->commit->object.sha1),
2150 ent->suspect->refcnt);
2156 find_alignment(sb, &opt);
2158 die("Baa %d!", baa);
2162 static unsigned parse_score(const char *arg)
2165 unsigned long score = strtoul(arg, &end, 10);
2171 static const char *add_prefix(const char *prefix, const char *path)
2173 return prefix_path(prefix, prefix ? strlen(prefix) : 0, path);
2176 static int git_blame_config(const char *var, const char *value, void *cb)
2178 if (!strcmp(var, "blame.showroot")) {
2179 show_root = git_config_bool(var, value);
2182 if (!strcmp(var, "blame.blankboundary")) {
2183 blank_boundary = git_config_bool(var, value);
2186 if (!strcmp(var, "blame.showemail")) {
2187 int *output_option = cb;
2188 if (git_config_bool(var, value))
2189 *output_option |= OUTPUT_SHOW_EMAIL;
2191 *output_option &= ~OUTPUT_SHOW_EMAIL;
2194 if (!strcmp(var, "blame.date")) {
2196 return config_error_nonbool(var);
2197 blame_date_mode = parse_date_format(value);
2201 if (userdiff_config(var, value) < 0)
2204 return git_default_config(var, value, cb);
2207 static void verify_working_tree_path(struct commit *work_tree, const char *path)
2209 struct commit_list *parents;
2211 for (parents = work_tree->parents; parents; parents = parents->next) {
2212 const unsigned char *commit_sha1 = parents->item->object.sha1;
2213 unsigned char blob_sha1[20];
2216 if (!get_tree_entry(commit_sha1, path, blob_sha1, &mode) &&
2217 sha1_object_info(blob_sha1, NULL) == OBJ_BLOB)
2220 die("no such path '%s' in HEAD", path);
2223 static struct commit_list **append_parent(struct commit_list **tail, const unsigned char *sha1)
2225 struct commit *parent;
2227 parent = lookup_commit_reference(sha1);
2229 die("no such commit %s", sha1_to_hex(sha1));
2230 return &commit_list_insert(parent, tail)->next;
2233 static void append_merge_parents(struct commit_list **tail)
2236 const char *merge_head_file = git_path("MERGE_HEAD");
2237 struct strbuf line = STRBUF_INIT;
2239 merge_head = open(merge_head_file, O_RDONLY);
2240 if (merge_head < 0) {
2241 if (errno == ENOENT)
2243 die("cannot open '%s' for reading", merge_head_file);
2246 while (!strbuf_getwholeline_fd(&line, merge_head, '\n')) {
2247 unsigned char sha1[20];
2248 if (line.len < 40 || get_sha1_hex(line.buf, sha1))
2249 die("unknown line in '%s': %s", merge_head_file, line.buf);
2250 tail = append_parent(tail, sha1);
2253 strbuf_release(&line);
2257 * This isn't as simple as passing sb->buf and sb->len, because we
2258 * want to transfer ownership of the buffer to the commit (so we
2261 static void set_commit_buffer_from_strbuf(struct commit *c, struct strbuf *sb)
2264 void *buf = strbuf_detach(sb, &len);
2265 set_commit_buffer(c, buf, len);
2269 * Prepare a dummy commit that represents the work tree (or staged) item.
2270 * Note that annotating work tree item never works in the reverse.
2272 static struct commit *fake_working_tree_commit(struct diff_options *opt,
2274 const char *contents_from)
2276 struct commit *commit;
2277 struct origin *origin;
2278 struct commit_list **parent_tail, *parent;
2279 unsigned char head_sha1[20];
2280 struct strbuf buf = STRBUF_INIT;
2284 struct cache_entry *ce;
2286 struct strbuf msg = STRBUF_INIT;
2289 commit = alloc_commit_node();
2290 commit->object.parsed = 1;
2292 parent_tail = &commit->parents;
2294 if (!resolve_ref_unsafe("HEAD", RESOLVE_REF_READING, head_sha1, NULL))
2295 die("no such ref: HEAD");
2297 parent_tail = append_parent(parent_tail, head_sha1);
2298 append_merge_parents(parent_tail);
2299 verify_working_tree_path(commit, path);
2301 origin = make_origin(commit, path);
2303 ident = fmt_ident("Not Committed Yet", "not.committed.yet", NULL, 0);
2304 strbuf_addstr(&msg, "tree 0000000000000000000000000000000000000000\n");
2305 for (parent = commit->parents; parent; parent = parent->next)
2306 strbuf_addf(&msg, "parent %s\n",
2307 sha1_to_hex(parent->item->object.sha1));
2311 "Version of %s from %s\n",
2313 (!contents_from ? path :
2314 (!strcmp(contents_from, "-") ? "standard input" : contents_from)));
2315 set_commit_buffer_from_strbuf(commit, &msg);
2317 if (!contents_from || strcmp("-", contents_from)) {
2319 const char *read_from;
2321 unsigned long buf_len;
2323 if (contents_from) {
2324 if (stat(contents_from, &st) < 0)
2325 die_errno("Cannot stat '%s'", contents_from);
2326 read_from = contents_from;
2329 if (lstat(path, &st) < 0)
2330 die_errno("Cannot lstat '%s'", path);
2333 mode = canon_mode(st.st_mode);
2335 switch (st.st_mode & S_IFMT) {
2337 if (DIFF_OPT_TST(opt, ALLOW_TEXTCONV) &&
2338 textconv_object(read_from, mode, null_sha1, 0, &buf_ptr, &buf_len))
2339 strbuf_attach(&buf, buf_ptr, buf_len, buf_len + 1);
2340 else if (strbuf_read_file(&buf, read_from, st.st_size) != st.st_size)
2341 die_errno("cannot open or read '%s'", read_from);
2344 if (strbuf_readlink(&buf, read_from, st.st_size) < 0)
2345 die_errno("cannot readlink '%s'", read_from);
2348 die("unsupported file type %s", read_from);
2352 /* Reading from stdin */
2354 if (strbuf_read(&buf, 0, 0) < 0)
2355 die_errno("failed to read from stdin");
2357 convert_to_git(path, buf.buf, buf.len, &buf, 0);
2358 origin->file.ptr = buf.buf;
2359 origin->file.size = buf.len;
2360 pretend_sha1_file(buf.buf, buf.len, OBJ_BLOB, origin->blob_sha1);
2363 * Read the current index, replace the path entry with
2364 * origin->blob_sha1 without mucking with its mode or type
2365 * bits; we are not going to write this index out -- we just
2366 * want to run "diff-index --cached".
2373 int pos = cache_name_pos(path, len);
2375 mode = active_cache[pos]->ce_mode;
2377 /* Let's not bother reading from HEAD tree */
2378 mode = S_IFREG | 0644;
2380 size = cache_entry_size(len);
2381 ce = xcalloc(1, size);
2382 hashcpy(ce->sha1, origin->blob_sha1);
2383 memcpy(ce->name, path, len);
2384 ce->ce_flags = create_ce_flags(0);
2385 ce->ce_namelen = len;
2386 ce->ce_mode = create_ce_mode(mode);
2387 add_cache_entry(ce, ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE);
2390 * We are not going to write this out, so this does not matter
2391 * right now, but someday we might optimize diff-index --cached
2392 * with cache-tree information.
2394 cache_tree_invalidate_path(&the_index, path);
2399 static char *prepare_final(struct scoreboard *sb)
2402 const char *final_commit_name = NULL;
2403 struct rev_info *revs = sb->revs;
2406 * There must be one and only one positive commit in the
2407 * revs->pending array.
2409 for (i = 0; i < revs->pending.nr; i++) {
2410 struct object *obj = revs->pending.objects[i].item;
2411 if (obj->flags & UNINTERESTING)
2413 while (obj->type == OBJ_TAG)
2414 obj = deref_tag(obj, NULL, 0);
2415 if (obj->type != OBJ_COMMIT)
2416 die("Non commit %s?", revs->pending.objects[i].name);
2418 die("More than one commit to dig from %s and %s?",
2419 revs->pending.objects[i].name,
2421 sb->final = (struct commit *) obj;
2422 final_commit_name = revs->pending.objects[i].name;
2424 return xstrdup_or_null(final_commit_name);
2427 static char *prepare_initial(struct scoreboard *sb)
2430 const char *final_commit_name = NULL;
2431 struct rev_info *revs = sb->revs;
2434 * There must be one and only one negative commit, and it must be
2437 for (i = 0; i < revs->pending.nr; i++) {
2438 struct object *obj = revs->pending.objects[i].item;
2439 if (!(obj->flags & UNINTERESTING))
2441 while (obj->type == OBJ_TAG)
2442 obj = deref_tag(obj, NULL, 0);
2443 if (obj->type != OBJ_COMMIT)
2444 die("Non commit %s?", revs->pending.objects[i].name);
2446 die("More than one commit to dig down to %s and %s?",
2447 revs->pending.objects[i].name,
2449 sb->final = (struct commit *) obj;
2450 final_commit_name = revs->pending.objects[i].name;
2452 if (!final_commit_name)
2453 die("No commit to dig down to?");
2454 return xstrdup(final_commit_name);
2457 static int blame_copy_callback(const struct option *option, const char *arg, int unset)
2459 int *opt = option->value;
2462 * -C enables copy from removed files;
2463 * -C -C enables copy from existing files, but only
2464 * when blaming a new file;
2465 * -C -C -C enables copy from existing files for
2468 if (*opt & PICKAXE_BLAME_COPY_HARDER)
2469 *opt |= PICKAXE_BLAME_COPY_HARDEST;
2470 if (*opt & PICKAXE_BLAME_COPY)
2471 *opt |= PICKAXE_BLAME_COPY_HARDER;
2472 *opt |= PICKAXE_BLAME_COPY | PICKAXE_BLAME_MOVE;
2475 blame_copy_score = parse_score(arg);
2479 static int blame_move_callback(const struct option *option, const char *arg, int unset)
2481 int *opt = option->value;
2483 *opt |= PICKAXE_BLAME_MOVE;
2486 blame_move_score = parse_score(arg);
2490 int cmd_blame(int argc, const char **argv, const char *prefix)
2492 struct rev_info revs;
2494 struct scoreboard sb;
2496 struct blame_entry *ent = NULL;
2497 long dashdash_pos, lno;
2498 char *final_commit_name = NULL;
2499 enum object_type type;
2501 static struct string_list range_list;
2502 static int output_option = 0, opt = 0;
2503 static int show_stats = 0;
2504 static const char *revs_file = NULL;
2505 static const char *contents_from = NULL;
2506 static const struct option options[] = {
2507 OPT_BOOL(0, "incremental", &incremental, N_("Show blame entries as we find them, incrementally")),
2508 OPT_BOOL('b', NULL, &blank_boundary, N_("Show blank SHA-1 for boundary commits (Default: off)")),
2509 OPT_BOOL(0, "root", &show_root, N_("Do not treat root commits as boundaries (Default: off)")),
2510 OPT_BOOL(0, "show-stats", &show_stats, N_("Show work cost statistics")),
2511 OPT_BIT(0, "score-debug", &output_option, N_("Show output score for blame entries"), OUTPUT_SHOW_SCORE),
2512 OPT_BIT('f', "show-name", &output_option, N_("Show original filename (Default: auto)"), OUTPUT_SHOW_NAME),
2513 OPT_BIT('n', "show-number", &output_option, N_("Show original linenumber (Default: off)"), OUTPUT_SHOW_NUMBER),
2514 OPT_BIT('p', "porcelain", &output_option, N_("Show in a format designed for machine consumption"), OUTPUT_PORCELAIN),
2515 OPT_BIT(0, "line-porcelain", &output_option, N_("Show porcelain format with per-line commit information"), OUTPUT_PORCELAIN|OUTPUT_LINE_PORCELAIN),
2516 OPT_BIT('c', NULL, &output_option, N_("Use the same output mode as git-annotate (Default: off)"), OUTPUT_ANNOTATE_COMPAT),
2517 OPT_BIT('t', NULL, &output_option, N_("Show raw timestamp (Default: off)"), OUTPUT_RAW_TIMESTAMP),
2518 OPT_BIT('l', NULL, &output_option, N_("Show long commit SHA1 (Default: off)"), OUTPUT_LONG_OBJECT_NAME),
2519 OPT_BIT('s', NULL, &output_option, N_("Suppress author name and timestamp (Default: off)"), OUTPUT_NO_AUTHOR),
2520 OPT_BIT('e', "show-email", &output_option, N_("Show author email instead of name (Default: off)"), OUTPUT_SHOW_EMAIL),
2521 OPT_BIT('w', NULL, &xdl_opts, N_("Ignore whitespace differences"), XDF_IGNORE_WHITESPACE),
2522 OPT_BIT(0, "minimal", &xdl_opts, N_("Spend extra cycles to find better match"), XDF_NEED_MINIMAL),
2523 OPT_STRING('S', NULL, &revs_file, N_("file"), N_("Use revisions from <file> instead of calling git-rev-list")),
2524 OPT_STRING(0, "contents", &contents_from, N_("file"), N_("Use <file>'s contents as the final image")),
2525 { OPTION_CALLBACK, 'C', NULL, &opt, N_("score"), N_("Find line copies within and across files"), PARSE_OPT_OPTARG, blame_copy_callback },
2526 { OPTION_CALLBACK, 'M', NULL, &opt, N_("score"), N_("Find line movements within and across files"), PARSE_OPT_OPTARG, blame_move_callback },
2527 OPT_STRING_LIST('L', NULL, &range_list, N_("n,m"), N_("Process only line range n,m, counting from 1")),
2528 OPT__ABBREV(&abbrev),
2532 struct parse_opt_ctx_t ctx;
2533 int cmd_is_annotate = !strcmp(argv[0], "annotate");
2534 struct range_set ranges;
2535 unsigned int range_i;
2538 git_config(git_blame_config, &output_option);
2539 init_revisions(&revs, NULL);
2540 revs.date_mode = blame_date_mode;
2541 DIFF_OPT_SET(&revs.diffopt, ALLOW_TEXTCONV);
2542 DIFF_OPT_SET(&revs.diffopt, FOLLOW_RENAMES);
2544 save_commit_buffer = 0;
2547 parse_options_start(&ctx, argc, argv, prefix, options,
2548 PARSE_OPT_KEEP_DASHDASH | PARSE_OPT_KEEP_ARGV0);
2550 switch (parse_options_step(&ctx, options, blame_opt_usage)) {
2551 case PARSE_OPT_HELP:
2553 case PARSE_OPT_DONE:
2555 dashdash_pos = ctx.cpidx;
2559 if (!strcmp(ctx.argv[0], "--reverse")) {
2560 ctx.argv[0] = "--children";
2563 parse_revision_opt(&revs, &ctx, options, blame_opt_usage);
2566 no_whole_file_rename = !DIFF_OPT_TST(&revs.diffopt, FOLLOW_RENAMES);
2567 DIFF_OPT_CLR(&revs.diffopt, FOLLOW_RENAMES);
2568 argc = parse_options_end(&ctx);
2571 /* one more abbrev length is needed for the boundary commit */
2574 if (revs_file && read_ancestry(revs_file))
2575 die_errno("reading graft file '%s' failed", revs_file);
2577 if (cmd_is_annotate) {
2578 output_option |= OUTPUT_ANNOTATE_COMPAT;
2579 blame_date_mode = DATE_ISO8601;
2581 blame_date_mode = revs.date_mode;
2584 /* The maximum width used to show the dates */
2585 switch (blame_date_mode) {
2587 blame_date_width = sizeof("Thu, 19 Oct 2006 16:00:04 -0700");
2589 case DATE_ISO8601_STRICT:
2590 blame_date_width = sizeof("2006-10-19T16:00:04-07:00");
2593 blame_date_width = sizeof("2006-10-19 16:00:04 -0700");
2596 blame_date_width = sizeof("1161298804 -0700");
2599 blame_date_width = sizeof("2006-10-19");
2602 /* TRANSLATORS: This string is used to tell us the maximum
2603 display width for a relative timestamp in "git blame"
2604 output. For C locale, "4 years, 11 months ago", which
2605 takes 22 places, is the longest among various forms of
2606 relative timestamps, but your language may need more or
2607 fewer display columns. */
2608 blame_date_width = utf8_strwidth(_("4 years, 11 months ago")) + 1; /* add the null */
2612 blame_date_width = sizeof("Thu Oct 19 16:00:04 2006 -0700");
2615 blame_date_width -= 1; /* strip the null */
2617 if (DIFF_OPT_TST(&revs.diffopt, FIND_COPIES_HARDER))
2618 opt |= (PICKAXE_BLAME_COPY | PICKAXE_BLAME_MOVE |
2619 PICKAXE_BLAME_COPY_HARDER);
2621 if (!blame_move_score)
2622 blame_move_score = BLAME_DEFAULT_MOVE_SCORE;
2623 if (!blame_copy_score)
2624 blame_copy_score = BLAME_DEFAULT_COPY_SCORE;
2627 * We have collected options unknown to us in argv[1..unk]
2628 * which are to be passed to revision machinery if we are
2629 * going to do the "bottom" processing.
2631 * The remaining are:
2633 * (1) if dashdash_pos != 0, it is either
2634 * "blame [revisions] -- <path>" or
2635 * "blame -- <path> <rev>"
2637 * (2) otherwise, it is one of the two:
2638 * "blame [revisions] <path>"
2639 * "blame <path> <rev>"
2641 * Note that we must strip out <path> from the arguments: we do not
2642 * want the path pruning but we may want "bottom" processing.
2645 switch (argc - dashdash_pos - 1) {
2648 usage_with_options(blame_opt_usage, options);
2649 /* reorder for the new way: <rev> -- <path> */
2655 path = add_prefix(prefix, argv[--argc]);
2659 usage_with_options(blame_opt_usage, options);
2663 usage_with_options(blame_opt_usage, options);
2664 path = add_prefix(prefix, argv[argc - 1]);
2665 if (argc == 3 && !file_exists(path)) { /* (2b) */
2666 path = add_prefix(prefix, argv[1]);
2669 argv[argc - 1] = "--";
2672 if (!file_exists(path))
2673 die_errno("cannot stat path '%s'", path);
2676 revs.disable_stdin = 1;
2677 setup_revisions(argc, argv, &revs, NULL);
2678 memset(&sb, 0, sizeof(sb));
2682 final_commit_name = prepare_final(&sb);
2683 sb.commits.compare = compare_commits_by_commit_date;
2685 else if (contents_from)
2686 die("--contents and --children do not blend well.");
2687 else if (revs.first_parent_only)
2688 die("combining --first-parent and --reverse is not supported");
2690 final_commit_name = prepare_initial(&sb);
2691 sb.commits.compare = compare_commits_by_reverse_commit_date;
2696 * "--not A B -- path" without anything positive;
2697 * do not default to HEAD, but use the working tree
2701 sb.final = fake_working_tree_commit(&sb.revs->diffopt,
2702 path, contents_from);
2703 add_pending_object(&revs, &(sb.final->object), ":");
2705 else if (contents_from)
2706 die("Cannot use --contents with final commit object name");
2709 * If we have bottom, this will mark the ancestors of the
2710 * bottom commits we would reach while traversing as
2713 if (prepare_revision_walk(&revs))
2714 die(_("revision walk setup failed"));
2716 if (is_null_sha1(sb.final->object.sha1)) {
2718 sb.final_buf = xmemdupz(o->file.ptr, o->file.size);
2719 sb.final_buf_size = o->file.size;
2722 o = get_origin(&sb, sb.final, path);
2723 if (fill_blob_sha1_and_mode(o))
2724 die("no such path %s in %s", path, final_commit_name);
2726 if (DIFF_OPT_TST(&sb.revs->diffopt, ALLOW_TEXTCONV) &&
2727 textconv_object(path, o->mode, o->blob_sha1, 1, (char **) &sb.final_buf,
2728 &sb.final_buf_size))
2731 sb.final_buf = read_sha1_file(o->blob_sha1, &type,
2732 &sb.final_buf_size);
2735 die("Cannot read blob %s for path %s",
2736 sha1_to_hex(o->blob_sha1),
2740 lno = prepare_lines(&sb);
2742 if (lno && !range_list.nr)
2743 string_list_append(&range_list, xstrdup("1"));
2746 range_set_init(&ranges, range_list.nr);
2747 for (range_i = 0; range_i < range_list.nr; ++range_i) {
2749 if (parse_range_arg(range_list.items[range_i].string,
2750 nth_line_cb, &sb, lno, anchor,
2751 &bottom, &top, sb.path))
2753 if (lno < top || ((lno || bottom) && lno < bottom))
2754 die("file %s has only %lu lines", path, lno);
2760 range_set_append_unsafe(&ranges, bottom, top);
2763 sort_and_merge_range_set(&ranges);
2765 for (range_i = ranges.nr; range_i > 0; --range_i) {
2766 const struct range *r = &ranges.ranges[range_i - 1];
2767 long bottom = r->start;
2769 struct blame_entry *next = ent;
2770 ent = xcalloc(1, sizeof(*ent));
2772 ent->num_lines = top - bottom;
2774 ent->s_lno = bottom;
2780 prio_queue_put(&sb.commits, o->commit);
2784 range_set_release(&ranges);
2785 string_list_clear(&range_list, 0);
2790 read_mailmap(&mailmap, NULL);
2795 assign_blame(&sb, opt);
2797 free(final_commit_name);
2802 sb.ent = blame_sort(sb.ent, compare_blame_final);
2806 if (!(output_option & OUTPUT_PORCELAIN))
2807 find_alignment(&sb, &output_option);
2809 output(&sb, output_option);
2810 free((void *)sb.final_buf);
2811 for (ent = sb.ent; ent; ) {
2812 struct blame_entry *e = ent->next;
2818 printf("num read blob: %d\n", num_read_blob);
2819 printf("num get patch: %d\n", num_get_patch);
2820 printf("num commits: %d\n", num_commits);