git-am: propagate -C<n>, -p<n> options as well
[git/msvc.git] / builtin-blame.c
bloba0d60145f26d32e45f4d04c22164cbe8060739ca
1 /*
2 * Pickaxe
4 * Copyright (c) 2006, Junio C Hamano
5 */
7 #include "cache.h"
8 #include "builtin.h"
9 #include "blob.h"
10 #include "commit.h"
11 #include "tag.h"
12 #include "tree-walk.h"
13 #include "diff.h"
14 #include "diffcore.h"
15 #include "revision.h"
16 #include "quote.h"
17 #include "xdiff-interface.h"
18 #include "cache-tree.h"
19 #include "string-list.h"
20 #include "mailmap.h"
21 #include "parse-options.h"
23 static char blame_usage[] = "git blame [options] [rev-opts] [rev] [--] file";
25 static const char *blame_opt_usage[] = {
26 blame_usage,
27 "",
28 "[rev-opts] are documented in git-rev-list(1)",
29 NULL
32 static int longest_file;
33 static int longest_author;
34 static int max_orig_digits;
35 static int max_digits;
36 static int max_score_digits;
37 static int show_root;
38 static int reverse;
39 static int blank_boundary;
40 static int incremental;
41 static int xdl_opts = XDF_NEED_MINIMAL;
42 static struct string_list mailmap;
44 #ifndef DEBUG
45 #define DEBUG 0
46 #endif
48 /* stats */
49 static int num_read_blob;
50 static int num_get_patch;
51 static int num_commits;
53 #define PICKAXE_BLAME_MOVE 01
54 #define PICKAXE_BLAME_COPY 02
55 #define PICKAXE_BLAME_COPY_HARDER 04
56 #define PICKAXE_BLAME_COPY_HARDEST 010
59 * blame for a blame_entry with score lower than these thresholds
60 * is not passed to the parent using move/copy logic.
62 static unsigned blame_move_score;
63 static unsigned blame_copy_score;
64 #define BLAME_DEFAULT_MOVE_SCORE 20
65 #define BLAME_DEFAULT_COPY_SCORE 40
67 /* bits #0..7 in revision.h, #8..11 used for merge_bases() in commit.c */
68 #define METAINFO_SHOWN (1u<<12)
69 #define MORE_THAN_ONE_PATH (1u<<13)
72 * One blob in a commit that is being suspected
74 struct origin {
75 int refcnt;
76 struct commit *commit;
77 mmfile_t file;
78 unsigned char blob_sha1[20];
79 char path[FLEX_ARRAY];
83 * Given an origin, prepare mmfile_t structure to be used by the
84 * diff machinery
86 static void fill_origin_blob(struct origin *o, mmfile_t *file)
88 if (!o->file.ptr) {
89 enum object_type type;
90 num_read_blob++;
91 file->ptr = read_sha1_file(o->blob_sha1, &type,
92 (unsigned long *)(&(file->size)));
93 if (!file->ptr)
94 die("Cannot read blob %s for path %s",
95 sha1_to_hex(o->blob_sha1),
96 o->path);
97 o->file = *file;
99 else
100 *file = o->file;
104 * Origin is refcounted and usually we keep the blob contents to be
105 * reused.
107 static inline struct origin *origin_incref(struct origin *o)
109 if (o)
110 o->refcnt++;
111 return o;
114 static void origin_decref(struct origin *o)
116 if (o && --o->refcnt <= 0) {
117 free(o->file.ptr);
118 free(o);
122 static void drop_origin_blob(struct origin *o)
124 if (o->file.ptr) {
125 free(o->file.ptr);
126 o->file.ptr = NULL;
131 * Each group of lines is described by a blame_entry; it can be split
132 * as we pass blame to the parents. They form a linked list in the
133 * scoreboard structure, sorted by the target line number.
135 struct blame_entry {
136 struct blame_entry *prev;
137 struct blame_entry *next;
139 /* the first line of this group in the final image;
140 * internally all line numbers are 0 based.
142 int lno;
144 /* how many lines this group has */
145 int num_lines;
147 /* the commit that introduced this group into the final image */
148 struct origin *suspect;
150 /* true if the suspect is truly guilty; false while we have not
151 * checked if the group came from one of its parents.
153 char guilty;
155 /* true if the entry has been scanned for copies in the current parent
157 char scanned;
159 /* the line number of the first line of this group in the
160 * suspect's file; internally all line numbers are 0 based.
162 int s_lno;
164 /* how significant this entry is -- cached to avoid
165 * scanning the lines over and over.
167 unsigned score;
171 * The current state of the blame assignment.
173 struct scoreboard {
174 /* the final commit (i.e. where we started digging from) */
175 struct commit *final;
176 struct rev_info *revs;
177 const char *path;
180 * The contents in the final image.
181 * Used by many functions to obtain contents of the nth line,
182 * indexed with scoreboard.lineno[blame_entry.lno].
184 const char *final_buf;
185 unsigned long final_buf_size;
187 /* linked list of blames */
188 struct blame_entry *ent;
190 /* look-up a line in the final buffer */
191 int num_lines;
192 int *lineno;
195 static inline int same_suspect(struct origin *a, struct origin *b)
197 if (a == b)
198 return 1;
199 if (a->commit != b->commit)
200 return 0;
201 return !strcmp(a->path, b->path);
204 static void sanity_check_refcnt(struct scoreboard *);
207 * If two blame entries that are next to each other came from
208 * contiguous lines in the same origin (i.e. <commit, path> pair),
209 * merge them together.
211 static void coalesce(struct scoreboard *sb)
213 struct blame_entry *ent, *next;
215 for (ent = sb->ent; ent && (next = ent->next); ent = next) {
216 if (same_suspect(ent->suspect, next->suspect) &&
217 ent->guilty == next->guilty &&
218 ent->s_lno + ent->num_lines == next->s_lno) {
219 ent->num_lines += next->num_lines;
220 ent->next = next->next;
221 if (ent->next)
222 ent->next->prev = ent;
223 origin_decref(next->suspect);
224 free(next);
225 ent->score = 0;
226 next = ent; /* again */
230 if (DEBUG) /* sanity */
231 sanity_check_refcnt(sb);
235 * Given a commit and a path in it, create a new origin structure.
236 * The callers that add blame to the scoreboard should use
237 * get_origin() to obtain shared, refcounted copy instead of calling
238 * this function directly.
240 static struct origin *make_origin(struct commit *commit, const char *path)
242 struct origin *o;
243 o = xcalloc(1, sizeof(*o) + strlen(path) + 1);
244 o->commit = commit;
245 o->refcnt = 1;
246 strcpy(o->path, path);
247 return o;
251 * Locate an existing origin or create a new one.
253 static struct origin *get_origin(struct scoreboard *sb,
254 struct commit *commit,
255 const char *path)
257 struct blame_entry *e;
259 for (e = sb->ent; e; e = e->next) {
260 if (e->suspect->commit == commit &&
261 !strcmp(e->suspect->path, path))
262 return origin_incref(e->suspect);
264 return make_origin(commit, path);
268 * Fill the blob_sha1 field of an origin if it hasn't, so that later
269 * call to fill_origin_blob() can use it to locate the data. blob_sha1
270 * for an origin is also used to pass the blame for the entire file to
271 * the parent to detect the case where a child's blob is identical to
272 * that of its parent's.
274 static int fill_blob_sha1(struct origin *origin)
276 unsigned mode;
278 if (!is_null_sha1(origin->blob_sha1))
279 return 0;
280 if (get_tree_entry(origin->commit->object.sha1,
281 origin->path,
282 origin->blob_sha1, &mode))
283 goto error_out;
284 if (sha1_object_info(origin->blob_sha1, NULL) != OBJ_BLOB)
285 goto error_out;
286 return 0;
287 error_out:
288 hashclr(origin->blob_sha1);
289 return -1;
293 * We have an origin -- check if the same path exists in the
294 * parent and return an origin structure to represent it.
296 static struct origin *find_origin(struct scoreboard *sb,
297 struct commit *parent,
298 struct origin *origin)
300 struct origin *porigin = NULL;
301 struct diff_options diff_opts;
302 const char *paths[2];
304 if (parent->util) {
306 * Each commit object can cache one origin in that
307 * commit. This is a freestanding copy of origin and
308 * not refcounted.
310 struct origin *cached = parent->util;
311 if (!strcmp(cached->path, origin->path)) {
313 * The same path between origin and its parent
314 * without renaming -- the most common case.
316 porigin = get_origin(sb, parent, cached->path);
319 * If the origin was newly created (i.e. get_origin
320 * would call make_origin if none is found in the
321 * scoreboard), it does not know the blob_sha1,
322 * so copy it. Otherwise porigin was in the
323 * scoreboard and already knows blob_sha1.
325 if (porigin->refcnt == 1)
326 hashcpy(porigin->blob_sha1, cached->blob_sha1);
327 return porigin;
329 /* otherwise it was not very useful; free it */
330 free(parent->util);
331 parent->util = NULL;
334 /* See if the origin->path is different between parent
335 * and origin first. Most of the time they are the
336 * same and diff-tree is fairly efficient about this.
338 diff_setup(&diff_opts);
339 DIFF_OPT_SET(&diff_opts, RECURSIVE);
340 diff_opts.detect_rename = 0;
341 diff_opts.output_format = DIFF_FORMAT_NO_OUTPUT;
342 paths[0] = origin->path;
343 paths[1] = NULL;
345 diff_tree_setup_paths(paths, &diff_opts);
346 if (diff_setup_done(&diff_opts) < 0)
347 die("diff-setup");
349 if (is_null_sha1(origin->commit->object.sha1))
350 do_diff_cache(parent->tree->object.sha1, &diff_opts);
351 else
352 diff_tree_sha1(parent->tree->object.sha1,
353 origin->commit->tree->object.sha1,
354 "", &diff_opts);
355 diffcore_std(&diff_opts);
357 /* It is either one entry that says "modified", or "created",
358 * or nothing.
360 if (!diff_queued_diff.nr) {
361 /* The path is the same as parent */
362 porigin = get_origin(sb, parent, origin->path);
363 hashcpy(porigin->blob_sha1, origin->blob_sha1);
365 else if (diff_queued_diff.nr != 1)
366 die("internal error in blame::find_origin");
367 else {
368 struct diff_filepair *p = diff_queued_diff.queue[0];
369 switch (p->status) {
370 default:
371 die("internal error in blame::find_origin (%c)",
372 p->status);
373 case 'M':
374 porigin = get_origin(sb, parent, origin->path);
375 hashcpy(porigin->blob_sha1, p->one->sha1);
376 break;
377 case 'A':
378 case 'T':
379 /* Did not exist in parent, or type changed */
380 break;
383 diff_flush(&diff_opts);
384 diff_tree_release_paths(&diff_opts);
385 if (porigin) {
387 * Create a freestanding copy that is not part of
388 * the refcounted origin found in the scoreboard, and
389 * cache it in the commit.
391 struct origin *cached;
393 cached = make_origin(porigin->commit, porigin->path);
394 hashcpy(cached->blob_sha1, porigin->blob_sha1);
395 parent->util = cached;
397 return porigin;
401 * We have an origin -- find the path that corresponds to it in its
402 * parent and return an origin structure to represent it.
404 static struct origin *find_rename(struct scoreboard *sb,
405 struct commit *parent,
406 struct origin *origin)
408 struct origin *porigin = NULL;
409 struct diff_options diff_opts;
410 int i;
411 const char *paths[2];
413 diff_setup(&diff_opts);
414 DIFF_OPT_SET(&diff_opts, RECURSIVE);
415 diff_opts.detect_rename = DIFF_DETECT_RENAME;
416 diff_opts.output_format = DIFF_FORMAT_NO_OUTPUT;
417 diff_opts.single_follow = origin->path;
418 paths[0] = NULL;
419 diff_tree_setup_paths(paths, &diff_opts);
420 if (diff_setup_done(&diff_opts) < 0)
421 die("diff-setup");
423 if (is_null_sha1(origin->commit->object.sha1))
424 do_diff_cache(parent->tree->object.sha1, &diff_opts);
425 else
426 diff_tree_sha1(parent->tree->object.sha1,
427 origin->commit->tree->object.sha1,
428 "", &diff_opts);
429 diffcore_std(&diff_opts);
431 for (i = 0; i < diff_queued_diff.nr; i++) {
432 struct diff_filepair *p = diff_queued_diff.queue[i];
433 if ((p->status == 'R' || p->status == 'C') &&
434 !strcmp(p->two->path, origin->path)) {
435 porigin = get_origin(sb, parent, p->one->path);
436 hashcpy(porigin->blob_sha1, p->one->sha1);
437 break;
440 diff_flush(&diff_opts);
441 diff_tree_release_paths(&diff_opts);
442 return porigin;
446 * Link in a new blame entry to the scoreboard. Entries that cover the
447 * same line range have been removed from the scoreboard previously.
449 static void add_blame_entry(struct scoreboard *sb, struct blame_entry *e)
451 struct blame_entry *ent, *prev = NULL;
453 origin_incref(e->suspect);
455 for (ent = sb->ent; ent && ent->lno < e->lno; ent = ent->next)
456 prev = ent;
458 /* prev, if not NULL, is the last one that is below e */
459 e->prev = prev;
460 if (prev) {
461 e->next = prev->next;
462 prev->next = e;
464 else {
465 e->next = sb->ent;
466 sb->ent = e;
468 if (e->next)
469 e->next->prev = e;
473 * src typically is on-stack; we want to copy the information in it to
474 * a malloced blame_entry that is already on the linked list of the
475 * scoreboard. The origin of dst loses a refcnt while the origin of src
476 * gains one.
478 static void dup_entry(struct blame_entry *dst, struct blame_entry *src)
480 struct blame_entry *p, *n;
482 p = dst->prev;
483 n = dst->next;
484 origin_incref(src->suspect);
485 origin_decref(dst->suspect);
486 memcpy(dst, src, sizeof(*src));
487 dst->prev = p;
488 dst->next = n;
489 dst->score = 0;
492 static const char *nth_line(struct scoreboard *sb, int lno)
494 return sb->final_buf + sb->lineno[lno];
498 * It is known that lines between tlno to same came from parent, and e
499 * has an overlap with that range. it also is known that parent's
500 * line plno corresponds to e's line tlno.
502 * <---- e ----->
503 * <------>
504 * <------------>
505 * <------------>
506 * <------------------>
508 * Split e into potentially three parts; before this chunk, the chunk
509 * to be blamed for the parent, and after that portion.
511 static void split_overlap(struct blame_entry *split,
512 struct blame_entry *e,
513 int tlno, int plno, int same,
514 struct origin *parent)
516 int chunk_end_lno;
517 memset(split, 0, sizeof(struct blame_entry [3]));
519 if (e->s_lno < tlno) {
520 /* there is a pre-chunk part not blamed on parent */
521 split[0].suspect = origin_incref(e->suspect);
522 split[0].lno = e->lno;
523 split[0].s_lno = e->s_lno;
524 split[0].num_lines = tlno - e->s_lno;
525 split[1].lno = e->lno + tlno - e->s_lno;
526 split[1].s_lno = plno;
528 else {
529 split[1].lno = e->lno;
530 split[1].s_lno = plno + (e->s_lno - tlno);
533 if (same < e->s_lno + e->num_lines) {
534 /* there is a post-chunk part not blamed on parent */
535 split[2].suspect = origin_incref(e->suspect);
536 split[2].lno = e->lno + (same - e->s_lno);
537 split[2].s_lno = e->s_lno + (same - e->s_lno);
538 split[2].num_lines = e->s_lno + e->num_lines - same;
539 chunk_end_lno = split[2].lno;
541 else
542 chunk_end_lno = e->lno + e->num_lines;
543 split[1].num_lines = chunk_end_lno - split[1].lno;
546 * if it turns out there is nothing to blame the parent for,
547 * forget about the splitting. !split[1].suspect signals this.
549 if (split[1].num_lines < 1)
550 return;
551 split[1].suspect = origin_incref(parent);
555 * split_overlap() divided an existing blame e into up to three parts
556 * in split. Adjust the linked list of blames in the scoreboard to
557 * reflect the split.
559 static void split_blame(struct scoreboard *sb,
560 struct blame_entry *split,
561 struct blame_entry *e)
563 struct blame_entry *new_entry;
565 if (split[0].suspect && split[2].suspect) {
566 /* The first part (reuse storage for the existing entry e) */
567 dup_entry(e, &split[0]);
569 /* The last part -- me */
570 new_entry = xmalloc(sizeof(*new_entry));
571 memcpy(new_entry, &(split[2]), sizeof(struct blame_entry));
572 add_blame_entry(sb, new_entry);
574 /* ... and the middle part -- parent */
575 new_entry = xmalloc(sizeof(*new_entry));
576 memcpy(new_entry, &(split[1]), sizeof(struct blame_entry));
577 add_blame_entry(sb, new_entry);
579 else if (!split[0].suspect && !split[2].suspect)
581 * The parent covers the entire area; reuse storage for
582 * e and replace it with the parent.
584 dup_entry(e, &split[1]);
585 else if (split[0].suspect) {
586 /* me and then parent */
587 dup_entry(e, &split[0]);
589 new_entry = xmalloc(sizeof(*new_entry));
590 memcpy(new_entry, &(split[1]), sizeof(struct blame_entry));
591 add_blame_entry(sb, new_entry);
593 else {
594 /* parent and then me */
595 dup_entry(e, &split[1]);
597 new_entry = xmalloc(sizeof(*new_entry));
598 memcpy(new_entry, &(split[2]), sizeof(struct blame_entry));
599 add_blame_entry(sb, new_entry);
602 if (DEBUG) { /* sanity */
603 struct blame_entry *ent;
604 int lno = sb->ent->lno, corrupt = 0;
606 for (ent = sb->ent; ent; ent = ent->next) {
607 if (lno != ent->lno)
608 corrupt = 1;
609 if (ent->s_lno < 0)
610 corrupt = 1;
611 lno += ent->num_lines;
613 if (corrupt) {
614 lno = sb->ent->lno;
615 for (ent = sb->ent; ent; ent = ent->next) {
616 printf("L %8d l %8d n %8d\n",
617 lno, ent->lno, ent->num_lines);
618 lno = ent->lno + ent->num_lines;
620 die("oops");
626 * After splitting the blame, the origins used by the
627 * on-stack blame_entry should lose one refcnt each.
629 static void decref_split(struct blame_entry *split)
631 int i;
633 for (i = 0; i < 3; i++)
634 origin_decref(split[i].suspect);
638 * Helper for blame_chunk(). blame_entry e is known to overlap with
639 * the patch hunk; split it and pass blame to the parent.
641 static void blame_overlap(struct scoreboard *sb, struct blame_entry *e,
642 int tlno, int plno, int same,
643 struct origin *parent)
645 struct blame_entry split[3];
647 split_overlap(split, e, tlno, plno, same, parent);
648 if (split[1].suspect)
649 split_blame(sb, split, e);
650 decref_split(split);
654 * Find the line number of the last line the target is suspected for.
656 static int find_last_in_target(struct scoreboard *sb, struct origin *target)
658 struct blame_entry *e;
659 int last_in_target = -1;
661 for (e = sb->ent; e; e = e->next) {
662 if (e->guilty || !same_suspect(e->suspect, target))
663 continue;
664 if (last_in_target < e->s_lno + e->num_lines)
665 last_in_target = e->s_lno + e->num_lines;
667 return last_in_target;
671 * Process one hunk from the patch between the current suspect for
672 * blame_entry e and its parent. Find and split the overlap, and
673 * pass blame to the overlapping part to the parent.
675 static void blame_chunk(struct scoreboard *sb,
676 int tlno, int plno, int same,
677 struct origin *target, struct origin *parent)
679 struct blame_entry *e;
681 for (e = sb->ent; e; e = e->next) {
682 if (e->guilty || !same_suspect(e->suspect, target))
683 continue;
684 if (same <= e->s_lno)
685 continue;
686 if (tlno < e->s_lno + e->num_lines)
687 blame_overlap(sb, e, tlno, plno, same, parent);
691 struct blame_chunk_cb_data {
692 struct scoreboard *sb;
693 struct origin *target;
694 struct origin *parent;
695 long plno;
696 long tlno;
699 static void blame_chunk_cb(void *data, long same, long p_next, long t_next)
701 struct blame_chunk_cb_data *d = data;
702 blame_chunk(d->sb, d->tlno, d->plno, same, d->target, d->parent);
703 d->plno = p_next;
704 d->tlno = t_next;
708 * We are looking at the origin 'target' and aiming to pass blame
709 * for the lines it is suspected to its parent. Run diff to find
710 * which lines came from parent and pass blame for them.
712 static int pass_blame_to_parent(struct scoreboard *sb,
713 struct origin *target,
714 struct origin *parent)
716 int last_in_target;
717 mmfile_t file_p, file_o;
718 struct blame_chunk_cb_data d = { sb, target, parent, 0, 0 };
719 xpparam_t xpp;
720 xdemitconf_t xecfg;
722 last_in_target = find_last_in_target(sb, target);
723 if (last_in_target < 0)
724 return 1; /* nothing remains for this target */
726 fill_origin_blob(parent, &file_p);
727 fill_origin_blob(target, &file_o);
728 num_get_patch++;
730 memset(&xpp, 0, sizeof(xpp));
731 xpp.flags = xdl_opts;
732 memset(&xecfg, 0, sizeof(xecfg));
733 xecfg.ctxlen = 0;
734 xdi_diff_hunks(&file_p, &file_o, blame_chunk_cb, &d, &xpp, &xecfg);
735 /* The rest (i.e. anything after tlno) are the same as the parent */
736 blame_chunk(sb, d.tlno, d.plno, last_in_target, target, parent);
738 return 0;
742 * The lines in blame_entry after splitting blames many times can become
743 * very small and trivial, and at some point it becomes pointless to
744 * blame the parents. E.g. "\t\t}\n\t}\n\n" appears everywhere in any
745 * ordinary C program, and it is not worth to say it was copied from
746 * totally unrelated file in the parent.
748 * Compute how trivial the lines in the blame_entry are.
750 static unsigned ent_score(struct scoreboard *sb, struct blame_entry *e)
752 unsigned score;
753 const char *cp, *ep;
755 if (e->score)
756 return e->score;
758 score = 1;
759 cp = nth_line(sb, e->lno);
760 ep = nth_line(sb, e->lno + e->num_lines);
761 while (cp < ep) {
762 unsigned ch = *((unsigned char *)cp);
763 if (isalnum(ch))
764 score++;
765 cp++;
767 e->score = score;
768 return score;
772 * best_so_far[] and this[] are both a split of an existing blame_entry
773 * that passes blame to the parent. Maintain best_so_far the best split
774 * so far, by comparing this and best_so_far and copying this into
775 * bst_so_far as needed.
777 static void copy_split_if_better(struct scoreboard *sb,
778 struct blame_entry *best_so_far,
779 struct blame_entry *this)
781 int i;
783 if (!this[1].suspect)
784 return;
785 if (best_so_far[1].suspect) {
786 if (ent_score(sb, &this[1]) < ent_score(sb, &best_so_far[1]))
787 return;
790 for (i = 0; i < 3; i++)
791 origin_incref(this[i].suspect);
792 decref_split(best_so_far);
793 memcpy(best_so_far, this, sizeof(struct blame_entry [3]));
797 * We are looking at a part of the final image represented by
798 * ent (tlno and same are offset by ent->s_lno).
799 * tlno is where we are looking at in the final image.
800 * up to (but not including) same match preimage.
801 * plno is where we are looking at in the preimage.
803 * <-------------- final image ---------------------->
804 * <------ent------>
805 * ^tlno ^same
806 * <---------preimage----->
807 * ^plno
809 * All line numbers are 0-based.
811 static void handle_split(struct scoreboard *sb,
812 struct blame_entry *ent,
813 int tlno, int plno, int same,
814 struct origin *parent,
815 struct blame_entry *split)
817 if (ent->num_lines <= tlno)
818 return;
819 if (tlno < same) {
820 struct blame_entry this[3];
821 tlno += ent->s_lno;
822 same += ent->s_lno;
823 split_overlap(this, ent, tlno, plno, same, parent);
824 copy_split_if_better(sb, split, this);
825 decref_split(this);
829 struct handle_split_cb_data {
830 struct scoreboard *sb;
831 struct blame_entry *ent;
832 struct origin *parent;
833 struct blame_entry *split;
834 long plno;
835 long tlno;
838 static void handle_split_cb(void *data, long same, long p_next, long t_next)
840 struct handle_split_cb_data *d = data;
841 handle_split(d->sb, d->ent, d->tlno, d->plno, same, d->parent, d->split);
842 d->plno = p_next;
843 d->tlno = t_next;
847 * Find the lines from parent that are the same as ent so that
848 * we can pass blames to it. file_p has the blob contents for
849 * the parent.
851 static void find_copy_in_blob(struct scoreboard *sb,
852 struct blame_entry *ent,
853 struct origin *parent,
854 struct blame_entry *split,
855 mmfile_t *file_p)
857 const char *cp;
858 int cnt;
859 mmfile_t file_o;
860 struct handle_split_cb_data d = { sb, ent, parent, split, 0, 0 };
861 xpparam_t xpp;
862 xdemitconf_t xecfg;
865 * Prepare mmfile that contains only the lines in ent.
867 cp = nth_line(sb, ent->lno);
868 file_o.ptr = (char*) cp;
869 cnt = ent->num_lines;
871 while (cnt && cp < sb->final_buf + sb->final_buf_size) {
872 if (*cp++ == '\n')
873 cnt--;
875 file_o.size = cp - file_o.ptr;
878 * file_o is a part of final image we are annotating.
879 * file_p partially may match that image.
881 memset(&xpp, 0, sizeof(xpp));
882 xpp.flags = xdl_opts;
883 memset(&xecfg, 0, sizeof(xecfg));
884 xecfg.ctxlen = 1;
885 memset(split, 0, sizeof(struct blame_entry [3]));
886 xdi_diff_hunks(file_p, &file_o, handle_split_cb, &d, &xpp, &xecfg);
887 /* remainder, if any, all match the preimage */
888 handle_split(sb, ent, d.tlno, d.plno, ent->num_lines, parent, split);
892 * See if lines currently target is suspected for can be attributed to
893 * parent.
895 static int find_move_in_parent(struct scoreboard *sb,
896 struct origin *target,
897 struct origin *parent)
899 int last_in_target, made_progress;
900 struct blame_entry *e, split[3];
901 mmfile_t file_p;
903 last_in_target = find_last_in_target(sb, target);
904 if (last_in_target < 0)
905 return 1; /* nothing remains for this target */
907 fill_origin_blob(parent, &file_p);
908 if (!file_p.ptr)
909 return 0;
911 made_progress = 1;
912 while (made_progress) {
913 made_progress = 0;
914 for (e = sb->ent; e; e = e->next) {
915 if (e->guilty || !same_suspect(e->suspect, target) ||
916 ent_score(sb, e) < blame_move_score)
917 continue;
918 find_copy_in_blob(sb, e, parent, split, &file_p);
919 if (split[1].suspect &&
920 blame_move_score < ent_score(sb, &split[1])) {
921 split_blame(sb, split, e);
922 made_progress = 1;
924 decref_split(split);
927 return 0;
930 struct blame_list {
931 struct blame_entry *ent;
932 struct blame_entry split[3];
936 * Count the number of entries the target is suspected for,
937 * and prepare a list of entry and the best split.
939 static struct blame_list *setup_blame_list(struct scoreboard *sb,
940 struct origin *target,
941 int min_score,
942 int *num_ents_p)
944 struct blame_entry *e;
945 int num_ents, i;
946 struct blame_list *blame_list = NULL;
948 for (e = sb->ent, num_ents = 0; e; e = e->next)
949 if (!e->scanned && !e->guilty &&
950 same_suspect(e->suspect, target) &&
951 min_score < ent_score(sb, e))
952 num_ents++;
953 if (num_ents) {
954 blame_list = xcalloc(num_ents, sizeof(struct blame_list));
955 for (e = sb->ent, i = 0; e; e = e->next)
956 if (!e->scanned && !e->guilty &&
957 same_suspect(e->suspect, target) &&
958 min_score < ent_score(sb, e))
959 blame_list[i++].ent = e;
961 *num_ents_p = num_ents;
962 return blame_list;
966 * Reset the scanned status on all entries.
968 static void reset_scanned_flag(struct scoreboard *sb)
970 struct blame_entry *e;
971 for (e = sb->ent; e; e = e->next)
972 e->scanned = 0;
976 * For lines target is suspected for, see if we can find code movement
977 * across file boundary from the parent commit. porigin is the path
978 * in the parent we already tried.
980 static int find_copy_in_parent(struct scoreboard *sb,
981 struct origin *target,
982 struct commit *parent,
983 struct origin *porigin,
984 int opt)
986 struct diff_options diff_opts;
987 const char *paths[1];
988 int i, j;
989 int retval;
990 struct blame_list *blame_list;
991 int num_ents;
993 blame_list = setup_blame_list(sb, target, blame_copy_score, &num_ents);
994 if (!blame_list)
995 return 1; /* nothing remains for this target */
997 diff_setup(&diff_opts);
998 DIFF_OPT_SET(&diff_opts, RECURSIVE);
999 diff_opts.output_format = DIFF_FORMAT_NO_OUTPUT;
1001 paths[0] = NULL;
1002 diff_tree_setup_paths(paths, &diff_opts);
1003 if (diff_setup_done(&diff_opts) < 0)
1004 die("diff-setup");
1006 /* Try "find copies harder" on new path if requested;
1007 * we do not want to use diffcore_rename() actually to
1008 * match things up; find_copies_harder is set only to
1009 * force diff_tree_sha1() to feed all filepairs to diff_queue,
1010 * and this code needs to be after diff_setup_done(), which
1011 * usually makes find-copies-harder imply copy detection.
1013 if ((opt & PICKAXE_BLAME_COPY_HARDEST)
1014 || ((opt & PICKAXE_BLAME_COPY_HARDER)
1015 && (!porigin || strcmp(target->path, porigin->path))))
1016 DIFF_OPT_SET(&diff_opts, FIND_COPIES_HARDER);
1018 if (is_null_sha1(target->commit->object.sha1))
1019 do_diff_cache(parent->tree->object.sha1, &diff_opts);
1020 else
1021 diff_tree_sha1(parent->tree->object.sha1,
1022 target->commit->tree->object.sha1,
1023 "", &diff_opts);
1025 if (!DIFF_OPT_TST(&diff_opts, FIND_COPIES_HARDER))
1026 diffcore_std(&diff_opts);
1028 retval = 0;
1029 while (1) {
1030 int made_progress = 0;
1032 for (i = 0; i < diff_queued_diff.nr; i++) {
1033 struct diff_filepair *p = diff_queued_diff.queue[i];
1034 struct origin *norigin;
1035 mmfile_t file_p;
1036 struct blame_entry this[3];
1038 if (!DIFF_FILE_VALID(p->one))
1039 continue; /* does not exist in parent */
1040 if (S_ISGITLINK(p->one->mode))
1041 continue; /* ignore git links */
1042 if (porigin && !strcmp(p->one->path, porigin->path))
1043 /* find_move already dealt with this path */
1044 continue;
1046 norigin = get_origin(sb, parent, p->one->path);
1047 hashcpy(norigin->blob_sha1, p->one->sha1);
1048 fill_origin_blob(norigin, &file_p);
1049 if (!file_p.ptr)
1050 continue;
1052 for (j = 0; j < num_ents; j++) {
1053 find_copy_in_blob(sb, blame_list[j].ent,
1054 norigin, this, &file_p);
1055 copy_split_if_better(sb, blame_list[j].split,
1056 this);
1057 decref_split(this);
1059 origin_decref(norigin);
1062 for (j = 0; j < num_ents; j++) {
1063 struct blame_entry *split = blame_list[j].split;
1064 if (split[1].suspect &&
1065 blame_copy_score < ent_score(sb, &split[1])) {
1066 split_blame(sb, split, blame_list[j].ent);
1067 made_progress = 1;
1069 else
1070 blame_list[j].ent->scanned = 1;
1071 decref_split(split);
1073 free(blame_list);
1075 if (!made_progress)
1076 break;
1077 blame_list = setup_blame_list(sb, target, blame_copy_score, &num_ents);
1078 if (!blame_list) {
1079 retval = 1;
1080 break;
1083 reset_scanned_flag(sb);
1084 diff_flush(&diff_opts);
1085 diff_tree_release_paths(&diff_opts);
1086 return retval;
1090 * The blobs of origin and porigin exactly match, so everything
1091 * origin is suspected for can be blamed on the parent.
1093 static void pass_whole_blame(struct scoreboard *sb,
1094 struct origin *origin, struct origin *porigin)
1096 struct blame_entry *e;
1098 if (!porigin->file.ptr && origin->file.ptr) {
1099 /* Steal its file */
1100 porigin->file = origin->file;
1101 origin->file.ptr = NULL;
1103 for (e = sb->ent; e; e = e->next) {
1104 if (!same_suspect(e->suspect, origin))
1105 continue;
1106 origin_incref(porigin);
1107 origin_decref(e->suspect);
1108 e->suspect = porigin;
1113 * We pass blame from the current commit to its parents. We keep saying
1114 * "parent" (and "porigin"), but what we mean is to find scapegoat to
1115 * exonerate ourselves.
1117 static struct commit_list *first_scapegoat(struct rev_info *revs, struct commit *commit)
1119 if (!reverse)
1120 return commit->parents;
1121 return lookup_decoration(&revs->children, &commit->object);
1124 static int num_scapegoats(struct rev_info *revs, struct commit *commit)
1126 int cnt;
1127 struct commit_list *l = first_scapegoat(revs, commit);
1128 for (cnt = 0; l; l = l->next)
1129 cnt++;
1130 return cnt;
1133 #define MAXSG 16
1135 static void pass_blame(struct scoreboard *sb, struct origin *origin, int opt)
1137 struct rev_info *revs = sb->revs;
1138 int i, pass, num_sg;
1139 struct commit *commit = origin->commit;
1140 struct commit_list *sg;
1141 struct origin *sg_buf[MAXSG];
1142 struct origin *porigin, **sg_origin = sg_buf;
1144 num_sg = num_scapegoats(revs, commit);
1145 if (!num_sg)
1146 goto finish;
1147 else if (num_sg < ARRAY_SIZE(sg_buf))
1148 memset(sg_buf, 0, sizeof(sg_buf));
1149 else
1150 sg_origin = xcalloc(num_sg, sizeof(*sg_origin));
1153 * The first pass looks for unrenamed path to optimize for
1154 * common cases, then we look for renames in the second pass.
1156 for (pass = 0; pass < 2; pass++) {
1157 struct origin *(*find)(struct scoreboard *,
1158 struct commit *, struct origin *);
1159 find = pass ? find_rename : find_origin;
1161 for (i = 0, sg = first_scapegoat(revs, commit);
1162 i < num_sg && sg;
1163 sg = sg->next, i++) {
1164 struct commit *p = sg->item;
1165 int j, same;
1167 if (sg_origin[i])
1168 continue;
1169 if (parse_commit(p))
1170 continue;
1171 porigin = find(sb, p, origin);
1172 if (!porigin)
1173 continue;
1174 if (!hashcmp(porigin->blob_sha1, origin->blob_sha1)) {
1175 pass_whole_blame(sb, origin, porigin);
1176 origin_decref(porigin);
1177 goto finish;
1179 for (j = same = 0; j < i; j++)
1180 if (sg_origin[j] &&
1181 !hashcmp(sg_origin[j]->blob_sha1,
1182 porigin->blob_sha1)) {
1183 same = 1;
1184 break;
1186 if (!same)
1187 sg_origin[i] = porigin;
1188 else
1189 origin_decref(porigin);
1193 num_commits++;
1194 for (i = 0, sg = first_scapegoat(revs, commit);
1195 i < num_sg && sg;
1196 sg = sg->next, i++) {
1197 struct origin *porigin = sg_origin[i];
1198 if (!porigin)
1199 continue;
1200 if (pass_blame_to_parent(sb, origin, porigin))
1201 goto finish;
1205 * Optionally find moves in parents' files.
1207 if (opt & PICKAXE_BLAME_MOVE)
1208 for (i = 0, sg = first_scapegoat(revs, commit);
1209 i < num_sg && sg;
1210 sg = sg->next, i++) {
1211 struct origin *porigin = sg_origin[i];
1212 if (!porigin)
1213 continue;
1214 if (find_move_in_parent(sb, origin, porigin))
1215 goto finish;
1219 * Optionally find copies from parents' files.
1221 if (opt & PICKAXE_BLAME_COPY)
1222 for (i = 0, sg = first_scapegoat(revs, commit);
1223 i < num_sg && sg;
1224 sg = sg->next, i++) {
1225 struct origin *porigin = sg_origin[i];
1226 if (find_copy_in_parent(sb, origin, sg->item,
1227 porigin, opt))
1228 goto finish;
1231 finish:
1232 for (i = 0; i < num_sg; i++) {
1233 if (sg_origin[i]) {
1234 drop_origin_blob(sg_origin[i]);
1235 origin_decref(sg_origin[i]);
1238 drop_origin_blob(origin);
1239 if (sg_buf != sg_origin)
1240 free(sg_origin);
1244 * Information on commits, used for output.
1246 struct commit_info
1248 const char *author;
1249 const char *author_mail;
1250 unsigned long author_time;
1251 const char *author_tz;
1253 /* filled only when asked for details */
1254 const char *committer;
1255 const char *committer_mail;
1256 unsigned long committer_time;
1257 const char *committer_tz;
1259 const char *summary;
1263 * Parse author/committer line in the commit object buffer
1265 static void get_ac_line(const char *inbuf, const char *what,
1266 int bufsz, char *person, const char **mail,
1267 unsigned long *time, const char **tz)
1269 int len, tzlen, maillen;
1270 char *tmp, *endp, *timepos;
1272 tmp = strstr(inbuf, what);
1273 if (!tmp)
1274 goto error_out;
1275 tmp += strlen(what);
1276 endp = strchr(tmp, '\n');
1277 if (!endp)
1278 len = strlen(tmp);
1279 else
1280 len = endp - tmp;
1281 if (bufsz <= len) {
1282 error_out:
1283 /* Ugh */
1284 *mail = *tz = "(unknown)";
1285 *time = 0;
1286 return;
1288 memcpy(person, tmp, len);
1290 tmp = person;
1291 tmp += len;
1292 *tmp = 0;
1293 while (*tmp != ' ')
1294 tmp--;
1295 *tz = tmp+1;
1296 tzlen = (person+len)-(tmp+1);
1298 *tmp = 0;
1299 while (*tmp != ' ')
1300 tmp--;
1301 *time = strtoul(tmp, NULL, 10);
1302 timepos = tmp;
1304 *tmp = 0;
1305 while (*tmp != ' ')
1306 tmp--;
1307 *mail = tmp + 1;
1308 *tmp = 0;
1309 maillen = timepos - tmp;
1311 if (!mailmap.nr)
1312 return;
1315 * mailmap expansion may make the name longer.
1316 * make room by pushing stuff down.
1318 tmp = person + bufsz - (tzlen + 1);
1319 memmove(tmp, *tz, tzlen);
1320 tmp[tzlen] = 0;
1321 *tz = tmp;
1323 tmp = tmp - (maillen + 1);
1324 memmove(tmp, *mail, maillen);
1325 tmp[maillen] = 0;
1326 *mail = tmp;
1329 * Now, convert e-mail using mailmap
1331 map_email(&mailmap, tmp + 1, person, tmp-person-1);
1334 static void get_commit_info(struct commit *commit,
1335 struct commit_info *ret,
1336 int detailed)
1338 int len;
1339 char *tmp, *endp, *reencoded, *message;
1340 static char author_buf[1024];
1341 static char committer_buf[1024];
1342 static char summary_buf[1024];
1345 * We've operated without save_commit_buffer, so
1346 * we now need to populate them for output.
1348 if (!commit->buffer) {
1349 enum object_type type;
1350 unsigned long size;
1351 commit->buffer =
1352 read_sha1_file(commit->object.sha1, &type, &size);
1353 if (!commit->buffer)
1354 die("Cannot read commit %s",
1355 sha1_to_hex(commit->object.sha1));
1357 reencoded = reencode_commit_message(commit, NULL);
1358 message = reencoded ? reencoded : commit->buffer;
1359 ret->author = author_buf;
1360 get_ac_line(message, "\nauthor ",
1361 sizeof(author_buf), author_buf, &ret->author_mail,
1362 &ret->author_time, &ret->author_tz);
1364 if (!detailed) {
1365 free(reencoded);
1366 return;
1369 ret->committer = committer_buf;
1370 get_ac_line(message, "\ncommitter ",
1371 sizeof(committer_buf), committer_buf, &ret->committer_mail,
1372 &ret->committer_time, &ret->committer_tz);
1374 ret->summary = summary_buf;
1375 tmp = strstr(message, "\n\n");
1376 if (!tmp) {
1377 error_out:
1378 sprintf(summary_buf, "(%s)", sha1_to_hex(commit->object.sha1));
1379 free(reencoded);
1380 return;
1382 tmp += 2;
1383 endp = strchr(tmp, '\n');
1384 if (!endp)
1385 endp = tmp + strlen(tmp);
1386 len = endp - tmp;
1387 if (len >= sizeof(summary_buf) || len == 0)
1388 goto error_out;
1389 memcpy(summary_buf, tmp, len);
1390 summary_buf[len] = 0;
1391 free(reencoded);
1395 * To allow LF and other nonportable characters in pathnames,
1396 * they are c-style quoted as needed.
1398 static void write_filename_info(const char *path)
1400 printf("filename ");
1401 write_name_quoted(path, stdout, '\n');
1405 * The blame_entry is found to be guilty for the range. Mark it
1406 * as such, and show it in incremental output.
1408 static void found_guilty_entry(struct blame_entry *ent)
1410 if (ent->guilty)
1411 return;
1412 ent->guilty = 1;
1413 if (incremental) {
1414 struct origin *suspect = ent->suspect;
1416 printf("%s %d %d %d\n",
1417 sha1_to_hex(suspect->commit->object.sha1),
1418 ent->s_lno + 1, ent->lno + 1, ent->num_lines);
1419 if (!(suspect->commit->object.flags & METAINFO_SHOWN)) {
1420 struct commit_info ci;
1421 suspect->commit->object.flags |= METAINFO_SHOWN;
1422 get_commit_info(suspect->commit, &ci, 1);
1423 printf("author %s\n", ci.author);
1424 printf("author-mail %s\n", ci.author_mail);
1425 printf("author-time %lu\n", ci.author_time);
1426 printf("author-tz %s\n", ci.author_tz);
1427 printf("committer %s\n", ci.committer);
1428 printf("committer-mail %s\n", ci.committer_mail);
1429 printf("committer-time %lu\n", ci.committer_time);
1430 printf("committer-tz %s\n", ci.committer_tz);
1431 printf("summary %s\n", ci.summary);
1432 if (suspect->commit->object.flags & UNINTERESTING)
1433 printf("boundary\n");
1435 write_filename_info(suspect->path);
1436 maybe_flush_or_die(stdout, "stdout");
1441 * The main loop -- while the scoreboard has lines whose true origin
1442 * is still unknown, pick one blame_entry, and allow its current
1443 * suspect to pass blames to its parents.
1445 static void assign_blame(struct scoreboard *sb, int opt)
1447 struct rev_info *revs = sb->revs;
1449 while (1) {
1450 struct blame_entry *ent;
1451 struct commit *commit;
1452 struct origin *suspect = NULL;
1454 /* find one suspect to break down */
1455 for (ent = sb->ent; !suspect && ent; ent = ent->next)
1456 if (!ent->guilty)
1457 suspect = ent->suspect;
1458 if (!suspect)
1459 return; /* all done */
1462 * We will use this suspect later in the loop,
1463 * so hold onto it in the meantime.
1465 origin_incref(suspect);
1466 commit = suspect->commit;
1467 if (!commit->object.parsed)
1468 parse_commit(commit);
1469 if (reverse ||
1470 (!(commit->object.flags & UNINTERESTING) &&
1471 !(revs->max_age != -1 && commit->date < revs->max_age)))
1472 pass_blame(sb, suspect, opt);
1473 else {
1474 commit->object.flags |= UNINTERESTING;
1475 if (commit->object.parsed)
1476 mark_parents_uninteresting(commit);
1478 /* treat root commit as boundary */
1479 if (!commit->parents && !show_root)
1480 commit->object.flags |= UNINTERESTING;
1482 /* Take responsibility for the remaining entries */
1483 for (ent = sb->ent; ent; ent = ent->next)
1484 if (same_suspect(ent->suspect, suspect))
1485 found_guilty_entry(ent);
1486 origin_decref(suspect);
1488 if (DEBUG) /* sanity */
1489 sanity_check_refcnt(sb);
1493 static const char *format_time(unsigned long time, const char *tz_str,
1494 int show_raw_time)
1496 static char time_buf[128];
1497 time_t t = time;
1498 int minutes, tz;
1499 struct tm *tm;
1501 if (show_raw_time) {
1502 sprintf(time_buf, "%lu %s", time, tz_str);
1503 return time_buf;
1506 tz = atoi(tz_str);
1507 minutes = tz < 0 ? -tz : tz;
1508 minutes = (minutes / 100)*60 + (minutes % 100);
1509 minutes = tz < 0 ? -minutes : minutes;
1510 t = time + minutes * 60;
1511 tm = gmtime(&t);
1513 strftime(time_buf, sizeof(time_buf), "%Y-%m-%d %H:%M:%S ", tm);
1514 strcat(time_buf, tz_str);
1515 return time_buf;
1518 #define OUTPUT_ANNOTATE_COMPAT 001
1519 #define OUTPUT_LONG_OBJECT_NAME 002
1520 #define OUTPUT_RAW_TIMESTAMP 004
1521 #define OUTPUT_PORCELAIN 010
1522 #define OUTPUT_SHOW_NAME 020
1523 #define OUTPUT_SHOW_NUMBER 040
1524 #define OUTPUT_SHOW_SCORE 0100
1525 #define OUTPUT_NO_AUTHOR 0200
1527 static void emit_porcelain(struct scoreboard *sb, struct blame_entry *ent)
1529 int cnt;
1530 const char *cp;
1531 struct origin *suspect = ent->suspect;
1532 char hex[41];
1534 strcpy(hex, sha1_to_hex(suspect->commit->object.sha1));
1535 printf("%s%c%d %d %d\n",
1536 hex,
1537 ent->guilty ? ' ' : '*', // purely for debugging
1538 ent->s_lno + 1,
1539 ent->lno + 1,
1540 ent->num_lines);
1541 if (!(suspect->commit->object.flags & METAINFO_SHOWN)) {
1542 struct commit_info ci;
1543 suspect->commit->object.flags |= METAINFO_SHOWN;
1544 get_commit_info(suspect->commit, &ci, 1);
1545 printf("author %s\n", ci.author);
1546 printf("author-mail %s\n", ci.author_mail);
1547 printf("author-time %lu\n", ci.author_time);
1548 printf("author-tz %s\n", ci.author_tz);
1549 printf("committer %s\n", ci.committer);
1550 printf("committer-mail %s\n", ci.committer_mail);
1551 printf("committer-time %lu\n", ci.committer_time);
1552 printf("committer-tz %s\n", ci.committer_tz);
1553 write_filename_info(suspect->path);
1554 printf("summary %s\n", ci.summary);
1555 if (suspect->commit->object.flags & UNINTERESTING)
1556 printf("boundary\n");
1558 else if (suspect->commit->object.flags & MORE_THAN_ONE_PATH)
1559 write_filename_info(suspect->path);
1561 cp = nth_line(sb, ent->lno);
1562 for (cnt = 0; cnt < ent->num_lines; cnt++) {
1563 char ch;
1564 if (cnt)
1565 printf("%s %d %d\n", hex,
1566 ent->s_lno + 1 + cnt,
1567 ent->lno + 1 + cnt);
1568 putchar('\t');
1569 do {
1570 ch = *cp++;
1571 putchar(ch);
1572 } while (ch != '\n' &&
1573 cp < sb->final_buf + sb->final_buf_size);
1577 static void emit_other(struct scoreboard *sb, struct blame_entry *ent, int opt)
1579 int cnt;
1580 const char *cp;
1581 struct origin *suspect = ent->suspect;
1582 struct commit_info ci;
1583 char hex[41];
1584 int show_raw_time = !!(opt & OUTPUT_RAW_TIMESTAMP);
1586 get_commit_info(suspect->commit, &ci, 1);
1587 strcpy(hex, sha1_to_hex(suspect->commit->object.sha1));
1589 cp = nth_line(sb, ent->lno);
1590 for (cnt = 0; cnt < ent->num_lines; cnt++) {
1591 char ch;
1592 int length = (opt & OUTPUT_LONG_OBJECT_NAME) ? 40 : 8;
1594 if (suspect->commit->object.flags & UNINTERESTING) {
1595 if (blank_boundary)
1596 memset(hex, ' ', length);
1597 else if (!(opt & OUTPUT_ANNOTATE_COMPAT)) {
1598 length--;
1599 putchar('^');
1603 printf("%.*s", length, hex);
1604 if (opt & OUTPUT_ANNOTATE_COMPAT)
1605 printf("\t(%10s\t%10s\t%d)", ci.author,
1606 format_time(ci.author_time, ci.author_tz,
1607 show_raw_time),
1608 ent->lno + 1 + cnt);
1609 else {
1610 if (opt & OUTPUT_SHOW_SCORE)
1611 printf(" %*d %02d",
1612 max_score_digits, ent->score,
1613 ent->suspect->refcnt);
1614 if (opt & OUTPUT_SHOW_NAME)
1615 printf(" %-*.*s", longest_file, longest_file,
1616 suspect->path);
1617 if (opt & OUTPUT_SHOW_NUMBER)
1618 printf(" %*d", max_orig_digits,
1619 ent->s_lno + 1 + cnt);
1621 if (!(opt & OUTPUT_NO_AUTHOR))
1622 printf(" (%-*.*s %10s",
1623 longest_author, longest_author,
1624 ci.author,
1625 format_time(ci.author_time,
1626 ci.author_tz,
1627 show_raw_time));
1628 printf(" %*d) ",
1629 max_digits, ent->lno + 1 + cnt);
1631 do {
1632 ch = *cp++;
1633 putchar(ch);
1634 } while (ch != '\n' &&
1635 cp < sb->final_buf + sb->final_buf_size);
1639 static void output(struct scoreboard *sb, int option)
1641 struct blame_entry *ent;
1643 if (option & OUTPUT_PORCELAIN) {
1644 for (ent = sb->ent; ent; ent = ent->next) {
1645 struct blame_entry *oth;
1646 struct origin *suspect = ent->suspect;
1647 struct commit *commit = suspect->commit;
1648 if (commit->object.flags & MORE_THAN_ONE_PATH)
1649 continue;
1650 for (oth = ent->next; oth; oth = oth->next) {
1651 if ((oth->suspect->commit != commit) ||
1652 !strcmp(oth->suspect->path, suspect->path))
1653 continue;
1654 commit->object.flags |= MORE_THAN_ONE_PATH;
1655 break;
1660 for (ent = sb->ent; ent; ent = ent->next) {
1661 if (option & OUTPUT_PORCELAIN)
1662 emit_porcelain(sb, ent);
1663 else {
1664 emit_other(sb, ent, option);
1670 * To allow quick access to the contents of nth line in the
1671 * final image, prepare an index in the scoreboard.
1673 static int prepare_lines(struct scoreboard *sb)
1675 const char *buf = sb->final_buf;
1676 unsigned long len = sb->final_buf_size;
1677 int num = 0, incomplete = 0, bol = 1;
1679 if (len && buf[len-1] != '\n')
1680 incomplete++; /* incomplete line at the end */
1681 while (len--) {
1682 if (bol) {
1683 sb->lineno = xrealloc(sb->lineno,
1684 sizeof(int* ) * (num + 1));
1685 sb->lineno[num] = buf - sb->final_buf;
1686 bol = 0;
1688 if (*buf++ == '\n') {
1689 num++;
1690 bol = 1;
1693 sb->lineno = xrealloc(sb->lineno,
1694 sizeof(int* ) * (num + incomplete + 1));
1695 sb->lineno[num + incomplete] = buf - sb->final_buf;
1696 sb->num_lines = num + incomplete;
1697 return sb->num_lines;
1701 * Add phony grafts for use with -S; this is primarily to
1702 * support git's cvsserver that wants to give a linear history
1703 * to its clients.
1705 static int read_ancestry(const char *graft_file)
1707 FILE *fp = fopen(graft_file, "r");
1708 char buf[1024];
1709 if (!fp)
1710 return -1;
1711 while (fgets(buf, sizeof(buf), fp)) {
1712 /* The format is just "Commit Parent1 Parent2 ...\n" */
1713 int len = strlen(buf);
1714 struct commit_graft *graft = read_graft_line(buf, len);
1715 if (graft)
1716 register_commit_graft(graft, 0);
1718 fclose(fp);
1719 return 0;
1723 * How many columns do we need to show line numbers in decimal?
1725 static int lineno_width(int lines)
1727 int i, width;
1729 for (width = 1, i = 10; i <= lines + 1; width++)
1730 i *= 10;
1731 return width;
1735 * How many columns do we need to show line numbers, authors,
1736 * and filenames?
1738 static void find_alignment(struct scoreboard *sb, int *option)
1740 int longest_src_lines = 0;
1741 int longest_dst_lines = 0;
1742 unsigned largest_score = 0;
1743 struct blame_entry *e;
1745 for (e = sb->ent; e; e = e->next) {
1746 struct origin *suspect = e->suspect;
1747 struct commit_info ci;
1748 int num;
1750 if (strcmp(suspect->path, sb->path))
1751 *option |= OUTPUT_SHOW_NAME;
1752 num = strlen(suspect->path);
1753 if (longest_file < num)
1754 longest_file = num;
1755 if (!(suspect->commit->object.flags & METAINFO_SHOWN)) {
1756 suspect->commit->object.flags |= METAINFO_SHOWN;
1757 get_commit_info(suspect->commit, &ci, 1);
1758 num = strlen(ci.author);
1759 if (longest_author < num)
1760 longest_author = num;
1762 num = e->s_lno + e->num_lines;
1763 if (longest_src_lines < num)
1764 longest_src_lines = num;
1765 num = e->lno + e->num_lines;
1766 if (longest_dst_lines < num)
1767 longest_dst_lines = num;
1768 if (largest_score < ent_score(sb, e))
1769 largest_score = ent_score(sb, e);
1771 max_orig_digits = lineno_width(longest_src_lines);
1772 max_digits = lineno_width(longest_dst_lines);
1773 max_score_digits = lineno_width(largest_score);
1777 * For debugging -- origin is refcounted, and this asserts that
1778 * we do not underflow.
1780 static void sanity_check_refcnt(struct scoreboard *sb)
1782 int baa = 0;
1783 struct blame_entry *ent;
1785 for (ent = sb->ent; ent; ent = ent->next) {
1786 /* Nobody should have zero or negative refcnt */
1787 if (ent->suspect->refcnt <= 0) {
1788 fprintf(stderr, "%s in %s has negative refcnt %d\n",
1789 ent->suspect->path,
1790 sha1_to_hex(ent->suspect->commit->object.sha1),
1791 ent->suspect->refcnt);
1792 baa = 1;
1795 for (ent = sb->ent; ent; ent = ent->next) {
1796 /* Mark the ones that haven't been checked */
1797 if (0 < ent->suspect->refcnt)
1798 ent->suspect->refcnt = -ent->suspect->refcnt;
1800 for (ent = sb->ent; ent; ent = ent->next) {
1802 * ... then pick each and see if they have the the
1803 * correct refcnt.
1805 int found;
1806 struct blame_entry *e;
1807 struct origin *suspect = ent->suspect;
1809 if (0 < suspect->refcnt)
1810 continue;
1811 suspect->refcnt = -suspect->refcnt; /* Unmark */
1812 for (found = 0, e = sb->ent; e; e = e->next) {
1813 if (e->suspect != suspect)
1814 continue;
1815 found++;
1817 if (suspect->refcnt != found) {
1818 fprintf(stderr, "%s in %s has refcnt %d, not %d\n",
1819 ent->suspect->path,
1820 sha1_to_hex(ent->suspect->commit->object.sha1),
1821 ent->suspect->refcnt, found);
1822 baa = 2;
1825 if (baa) {
1826 int opt = 0160;
1827 find_alignment(sb, &opt);
1828 output(sb, opt);
1829 die("Baa %d!", baa);
1834 * Used for the command line parsing; check if the path exists
1835 * in the working tree.
1837 static int has_string_in_work_tree(const char *path)
1839 struct stat st;
1840 return !lstat(path, &st);
1843 static unsigned parse_score(const char *arg)
1845 char *end;
1846 unsigned long score = strtoul(arg, &end, 10);
1847 if (*end)
1848 return 0;
1849 return score;
1852 static const char *add_prefix(const char *prefix, const char *path)
1854 return prefix_path(prefix, prefix ? strlen(prefix) : 0, path);
1858 * Parsing of (comma separated) one item in the -L option
1860 static const char *parse_loc(const char *spec,
1861 struct scoreboard *sb, long lno,
1862 long begin, long *ret)
1864 char *term;
1865 const char *line;
1866 long num;
1867 int reg_error;
1868 regex_t regexp;
1869 regmatch_t match[1];
1871 /* Allow "-L <something>,+20" to mean starting at <something>
1872 * for 20 lines, or "-L <something>,-5" for 5 lines ending at
1873 * <something>.
1875 if (1 < begin && (spec[0] == '+' || spec[0] == '-')) {
1876 num = strtol(spec + 1, &term, 10);
1877 if (term != spec + 1) {
1878 if (spec[0] == '-')
1879 num = 0 - num;
1880 if (0 < num)
1881 *ret = begin + num - 2;
1882 else if (!num)
1883 *ret = begin;
1884 else
1885 *ret = begin + num;
1886 return term;
1888 return spec;
1890 num = strtol(spec, &term, 10);
1891 if (term != spec) {
1892 *ret = num;
1893 return term;
1895 if (spec[0] != '/')
1896 return spec;
1898 /* it could be a regexp of form /.../ */
1899 for (term = (char*) spec + 1; *term && *term != '/'; term++) {
1900 if (*term == '\\')
1901 term++;
1903 if (*term != '/')
1904 return spec;
1906 /* try [spec+1 .. term-1] as regexp */
1907 *term = 0;
1908 begin--; /* input is in human terms */
1909 line = nth_line(sb, begin);
1911 if (!(reg_error = regcomp(&regexp, spec + 1, REG_NEWLINE)) &&
1912 !(reg_error = regexec(&regexp, line, 1, match, 0))) {
1913 const char *cp = line + match[0].rm_so;
1914 const char *nline;
1916 while (begin++ < lno) {
1917 nline = nth_line(sb, begin);
1918 if (line <= cp && cp < nline)
1919 break;
1920 line = nline;
1922 *ret = begin;
1923 regfree(&regexp);
1924 *term++ = '/';
1925 return term;
1927 else {
1928 char errbuf[1024];
1929 regerror(reg_error, &regexp, errbuf, 1024);
1930 die("-L parameter '%s': %s", spec + 1, errbuf);
1935 * Parsing of -L option
1937 static void prepare_blame_range(struct scoreboard *sb,
1938 const char *bottomtop,
1939 long lno,
1940 long *bottom, long *top)
1942 const char *term;
1944 term = parse_loc(bottomtop, sb, lno, 1, bottom);
1945 if (*term == ',') {
1946 term = parse_loc(term + 1, sb, lno, *bottom + 1, top);
1947 if (*term)
1948 usage(blame_usage);
1950 if (*term)
1951 usage(blame_usage);
1954 static int git_blame_config(const char *var, const char *value, void *cb)
1956 if (!strcmp(var, "blame.showroot")) {
1957 show_root = git_config_bool(var, value);
1958 return 0;
1960 if (!strcmp(var, "blame.blankboundary")) {
1961 blank_boundary = git_config_bool(var, value);
1962 return 0;
1964 return git_default_config(var, value, cb);
1968 * Prepare a dummy commit that represents the work tree (or staged) item.
1969 * Note that annotating work tree item never works in the reverse.
1971 static struct commit *fake_working_tree_commit(const char *path, const char *contents_from)
1973 struct commit *commit;
1974 struct origin *origin;
1975 unsigned char head_sha1[20];
1976 struct strbuf buf = STRBUF_INIT;
1977 const char *ident;
1978 time_t now;
1979 int size, len;
1980 struct cache_entry *ce;
1981 unsigned mode;
1983 if (get_sha1("HEAD", head_sha1))
1984 die("No such ref: HEAD");
1986 time(&now);
1987 commit = xcalloc(1, sizeof(*commit));
1988 commit->parents = xcalloc(1, sizeof(*commit->parents));
1989 commit->parents->item = lookup_commit_reference(head_sha1);
1990 commit->object.parsed = 1;
1991 commit->date = now;
1992 commit->object.type = OBJ_COMMIT;
1994 origin = make_origin(commit, path);
1996 if (!contents_from || strcmp("-", contents_from)) {
1997 struct stat st;
1998 const char *read_from;
1999 unsigned long fin_size;
2001 if (contents_from) {
2002 if (stat(contents_from, &st) < 0)
2003 die("Cannot stat %s", contents_from);
2004 read_from = contents_from;
2006 else {
2007 if (lstat(path, &st) < 0)
2008 die("Cannot lstat %s", path);
2009 read_from = path;
2011 fin_size = xsize_t(st.st_size);
2012 mode = canon_mode(st.st_mode);
2013 switch (st.st_mode & S_IFMT) {
2014 case S_IFREG:
2015 if (strbuf_read_file(&buf, read_from, st.st_size) != st.st_size)
2016 die("cannot open or read %s", read_from);
2017 break;
2018 case S_IFLNK:
2019 if (readlink(read_from, buf.buf, buf.alloc) != fin_size)
2020 die("cannot readlink %s", read_from);
2021 buf.len = fin_size;
2022 break;
2023 default:
2024 die("unsupported file type %s", read_from);
2027 else {
2028 /* Reading from stdin */
2029 contents_from = "standard input";
2030 mode = 0;
2031 if (strbuf_read(&buf, 0, 0) < 0)
2032 die("read error %s from stdin", strerror(errno));
2034 convert_to_git(path, buf.buf, buf.len, &buf, 0);
2035 origin->file.ptr = buf.buf;
2036 origin->file.size = buf.len;
2037 pretend_sha1_file(buf.buf, buf.len, OBJ_BLOB, origin->blob_sha1);
2038 commit->util = origin;
2041 * Read the current index, replace the path entry with
2042 * origin->blob_sha1 without mucking with its mode or type
2043 * bits; we are not going to write this index out -- we just
2044 * want to run "diff-index --cached".
2046 discard_cache();
2047 read_cache();
2049 len = strlen(path);
2050 if (!mode) {
2051 int pos = cache_name_pos(path, len);
2052 if (0 <= pos)
2053 mode = active_cache[pos]->ce_mode;
2054 else
2055 /* Let's not bother reading from HEAD tree */
2056 mode = S_IFREG | 0644;
2058 size = cache_entry_size(len);
2059 ce = xcalloc(1, size);
2060 hashcpy(ce->sha1, origin->blob_sha1);
2061 memcpy(ce->name, path, len);
2062 ce->ce_flags = create_ce_flags(len, 0);
2063 ce->ce_mode = create_ce_mode(mode);
2064 add_cache_entry(ce, ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE);
2067 * We are not going to write this out, so this does not matter
2068 * right now, but someday we might optimize diff-index --cached
2069 * with cache-tree information.
2071 cache_tree_invalidate_path(active_cache_tree, path);
2073 commit->buffer = xmalloc(400);
2074 ident = fmt_ident("Not Committed Yet", "not.committed.yet", NULL, 0);
2075 snprintf(commit->buffer, 400,
2076 "tree 0000000000000000000000000000000000000000\n"
2077 "parent %s\n"
2078 "author %s\n"
2079 "committer %s\n\n"
2080 "Version of %s from %s\n",
2081 sha1_to_hex(head_sha1),
2082 ident, ident, path, contents_from ? contents_from : path);
2083 return commit;
2086 static const char *prepare_final(struct scoreboard *sb)
2088 int i;
2089 const char *final_commit_name = NULL;
2090 struct rev_info *revs = sb->revs;
2093 * There must be one and only one positive commit in the
2094 * revs->pending array.
2096 for (i = 0; i < revs->pending.nr; i++) {
2097 struct object *obj = revs->pending.objects[i].item;
2098 if (obj->flags & UNINTERESTING)
2099 continue;
2100 while (obj->type == OBJ_TAG)
2101 obj = deref_tag(obj, NULL, 0);
2102 if (obj->type != OBJ_COMMIT)
2103 die("Non commit %s?", revs->pending.objects[i].name);
2104 if (sb->final)
2105 die("More than one commit to dig from %s and %s?",
2106 revs->pending.objects[i].name,
2107 final_commit_name);
2108 sb->final = (struct commit *) obj;
2109 final_commit_name = revs->pending.objects[i].name;
2111 return final_commit_name;
2114 static const char *prepare_initial(struct scoreboard *sb)
2116 int i;
2117 const char *final_commit_name = NULL;
2118 struct rev_info *revs = sb->revs;
2121 * There must be one and only one negative commit, and it must be
2122 * the boundary.
2124 for (i = 0; i < revs->pending.nr; i++) {
2125 struct object *obj = revs->pending.objects[i].item;
2126 if (!(obj->flags & UNINTERESTING))
2127 continue;
2128 while (obj->type == OBJ_TAG)
2129 obj = deref_tag(obj, NULL, 0);
2130 if (obj->type != OBJ_COMMIT)
2131 die("Non commit %s?", revs->pending.objects[i].name);
2132 if (sb->final)
2133 die("More than one commit to dig down to %s and %s?",
2134 revs->pending.objects[i].name,
2135 final_commit_name);
2136 sb->final = (struct commit *) obj;
2137 final_commit_name = revs->pending.objects[i].name;
2139 if (!final_commit_name)
2140 die("No commit to dig down to?");
2141 return final_commit_name;
2144 static int blame_copy_callback(const struct option *option, const char *arg, int unset)
2146 int *opt = option->value;
2149 * -C enables copy from removed files;
2150 * -C -C enables copy from existing files, but only
2151 * when blaming a new file;
2152 * -C -C -C enables copy from existing files for
2153 * everybody
2155 if (*opt & PICKAXE_BLAME_COPY_HARDER)
2156 *opt |= PICKAXE_BLAME_COPY_HARDEST;
2157 if (*opt & PICKAXE_BLAME_COPY)
2158 *opt |= PICKAXE_BLAME_COPY_HARDER;
2159 *opt |= PICKAXE_BLAME_COPY | PICKAXE_BLAME_MOVE;
2161 if (arg)
2162 blame_copy_score = parse_score(arg);
2163 return 0;
2166 static int blame_move_callback(const struct option *option, const char *arg, int unset)
2168 int *opt = option->value;
2170 *opt |= PICKAXE_BLAME_MOVE;
2172 if (arg)
2173 blame_move_score = parse_score(arg);
2174 return 0;
2177 static int blame_bottomtop_callback(const struct option *option, const char *arg, int unset)
2179 const char **bottomtop = option->value;
2180 if (!arg)
2181 return -1;
2182 if (*bottomtop)
2183 die("More than one '-L n,m' option given");
2184 *bottomtop = arg;
2185 return 0;
2188 int cmd_blame(int argc, const char **argv, const char *prefix)
2190 struct rev_info revs;
2191 const char *path;
2192 struct scoreboard sb;
2193 struct origin *o;
2194 struct blame_entry *ent;
2195 long dashdash_pos, bottom, top, lno;
2196 const char *final_commit_name = NULL;
2197 enum object_type type;
2199 static const char *bottomtop = NULL;
2200 static int output_option = 0, opt = 0;
2201 static int show_stats = 0;
2202 static const char *revs_file = NULL;
2203 static const char *contents_from = NULL;
2204 static const struct option options[] = {
2205 OPT_BOOLEAN(0, "incremental", &incremental, "Show blame entries as we find them, incrementally"),
2206 OPT_BOOLEAN('b', NULL, &blank_boundary, "Show blank SHA-1 for boundary commits (Default: off)"),
2207 OPT_BOOLEAN(0, "root", &show_root, "Do not treat root commits as boundaries (Default: off)"),
2208 OPT_BOOLEAN(0, "show-stats", &show_stats, "Show work cost statistics"),
2209 OPT_BIT(0, "score-debug", &output_option, "Show output score for blame entries", OUTPUT_SHOW_SCORE),
2210 OPT_BIT('f', "show-name", &output_option, "Show original filename (Default: auto)", OUTPUT_SHOW_NAME),
2211 OPT_BIT('n', "show-number", &output_option, "Show original linenumber (Default: off)", OUTPUT_SHOW_NUMBER),
2212 OPT_BIT('p', "porcelain", &output_option, "Show in a format designed for machine consumption", OUTPUT_PORCELAIN),
2213 OPT_BIT('c', NULL, &output_option, "Use the same output mode as git-annotate (Default: off)", OUTPUT_ANNOTATE_COMPAT),
2214 OPT_BIT('t', NULL, &output_option, "Show raw timestamp (Default: off)", OUTPUT_RAW_TIMESTAMP),
2215 OPT_BIT('l', NULL, &output_option, "Show long commit SHA1 (Default: off)", OUTPUT_LONG_OBJECT_NAME),
2216 OPT_BIT('s', NULL, &output_option, "Suppress author name and timestamp (Default: off)", OUTPUT_NO_AUTHOR),
2217 OPT_BIT('w', NULL, &xdl_opts, "Ignore whitespace differences", XDF_IGNORE_WHITESPACE),
2218 OPT_STRING('S', NULL, &revs_file, "file", "Use revisions from <file> instead of calling git-rev-list"),
2219 OPT_STRING(0, "contents", &contents_from, "file", "Use <file>'s contents as the final image"),
2220 { OPTION_CALLBACK, 'C', NULL, &opt, "score", "Find line copies within and across files", PARSE_OPT_OPTARG, blame_copy_callback },
2221 { OPTION_CALLBACK, 'M', NULL, &opt, "score", "Find line movements within and across files", PARSE_OPT_OPTARG, blame_move_callback },
2222 OPT_CALLBACK('L', NULL, &bottomtop, "n,m", "Process only line range n,m, counting from 1", blame_bottomtop_callback),
2223 OPT_END()
2226 struct parse_opt_ctx_t ctx;
2227 int cmd_is_annotate = !strcmp(argv[0], "annotate");
2229 git_config(git_blame_config, NULL);
2230 init_revisions(&revs, NULL);
2231 save_commit_buffer = 0;
2232 dashdash_pos = 0;
2234 parse_options_start(&ctx, argc, argv, PARSE_OPT_KEEP_DASHDASH |
2235 PARSE_OPT_KEEP_ARGV0);
2236 for (;;) {
2237 switch (parse_options_step(&ctx, options, blame_opt_usage)) {
2238 case PARSE_OPT_HELP:
2239 exit(129);
2240 case PARSE_OPT_DONE:
2241 if (ctx.argv[0])
2242 dashdash_pos = ctx.cpidx;
2243 goto parse_done;
2246 if (!strcmp(ctx.argv[0], "--reverse")) {
2247 ctx.argv[0] = "--children";
2248 reverse = 1;
2250 parse_revision_opt(&revs, &ctx, options, blame_opt_usage);
2252 parse_done:
2253 argc = parse_options_end(&ctx);
2255 if (cmd_is_annotate)
2256 output_option |= OUTPUT_ANNOTATE_COMPAT;
2258 if (DIFF_OPT_TST(&revs.diffopt, FIND_COPIES_HARDER))
2259 opt |= (PICKAXE_BLAME_COPY | PICKAXE_BLAME_MOVE |
2260 PICKAXE_BLAME_COPY_HARDER);
2262 if (!blame_move_score)
2263 blame_move_score = BLAME_DEFAULT_MOVE_SCORE;
2264 if (!blame_copy_score)
2265 blame_copy_score = BLAME_DEFAULT_COPY_SCORE;
2268 * We have collected options unknown to us in argv[1..unk]
2269 * which are to be passed to revision machinery if we are
2270 * going to do the "bottom" processing.
2272 * The remaining are:
2274 * (1) if dashdash_pos != 0, its either
2275 * "blame [revisions] -- <path>" or
2276 * "blame -- <path> <rev>"
2278 * (2) otherwise, its one of the two:
2279 * "blame [revisions] <path>"
2280 * "blame <path> <rev>"
2282 * Note that we must strip out <path> from the arguments: we do not
2283 * want the path pruning but we may want "bottom" processing.
2285 if (dashdash_pos) {
2286 switch (argc - dashdash_pos - 1) {
2287 case 2: /* (1b) */
2288 if (argc != 4)
2289 usage_with_options(blame_opt_usage, options);
2290 /* reorder for the new way: <rev> -- <path> */
2291 argv[1] = argv[3];
2292 argv[3] = argv[2];
2293 argv[2] = "--";
2294 /* FALLTHROUGH */
2295 case 1: /* (1a) */
2296 path = add_prefix(prefix, argv[--argc]);
2297 argv[argc] = NULL;
2298 break;
2299 default:
2300 usage_with_options(blame_opt_usage, options);
2302 } else {
2303 if (argc < 2)
2304 usage_with_options(blame_opt_usage, options);
2305 path = add_prefix(prefix, argv[argc - 1]);
2306 if (argc == 3 && !has_string_in_work_tree(path)) { /* (2b) */
2307 path = add_prefix(prefix, argv[1]);
2308 argv[1] = argv[2];
2310 argv[argc - 1] = "--";
2312 setup_work_tree();
2313 if (!has_string_in_work_tree(path))
2314 die("cannot stat path %s: %s", path, strerror(errno));
2317 setup_revisions(argc, argv, &revs, NULL);
2318 memset(&sb, 0, sizeof(sb));
2320 sb.revs = &revs;
2321 if (!reverse)
2322 final_commit_name = prepare_final(&sb);
2323 else if (contents_from)
2324 die("--contents and --children do not blend well.");
2325 else
2326 final_commit_name = prepare_initial(&sb);
2328 if (!sb.final) {
2330 * "--not A B -- path" without anything positive;
2331 * do not default to HEAD, but use the working tree
2332 * or "--contents".
2334 setup_work_tree();
2335 sb.final = fake_working_tree_commit(path, contents_from);
2336 add_pending_object(&revs, &(sb.final->object), ":");
2338 else if (contents_from)
2339 die("Cannot use --contents with final commit object name");
2342 * If we have bottom, this will mark the ancestors of the
2343 * bottom commits we would reach while traversing as
2344 * uninteresting.
2346 if (prepare_revision_walk(&revs))
2347 die("revision walk setup failed");
2349 if (is_null_sha1(sb.final->object.sha1)) {
2350 char *buf;
2351 o = sb.final->util;
2352 buf = xmalloc(o->file.size + 1);
2353 memcpy(buf, o->file.ptr, o->file.size + 1);
2354 sb.final_buf = buf;
2355 sb.final_buf_size = o->file.size;
2357 else {
2358 o = get_origin(&sb, sb.final, path);
2359 if (fill_blob_sha1(o))
2360 die("no such path %s in %s", path, final_commit_name);
2362 sb.final_buf = read_sha1_file(o->blob_sha1, &type,
2363 &sb.final_buf_size);
2364 if (!sb.final_buf)
2365 die("Cannot read blob %s for path %s",
2366 sha1_to_hex(o->blob_sha1),
2367 path);
2369 num_read_blob++;
2370 lno = prepare_lines(&sb);
2372 bottom = top = 0;
2373 if (bottomtop)
2374 prepare_blame_range(&sb, bottomtop, lno, &bottom, &top);
2375 if (bottom && top && top < bottom) {
2376 long tmp;
2377 tmp = top; top = bottom; bottom = tmp;
2379 if (bottom < 1)
2380 bottom = 1;
2381 if (top < 1)
2382 top = lno;
2383 bottom--;
2384 if (lno < top)
2385 die("file %s has only %lu lines", path, lno);
2387 ent = xcalloc(1, sizeof(*ent));
2388 ent->lno = bottom;
2389 ent->num_lines = top - bottom;
2390 ent->suspect = o;
2391 ent->s_lno = bottom;
2393 sb.ent = ent;
2394 sb.path = path;
2396 if (revs_file && read_ancestry(revs_file))
2397 die("reading graft file %s failed: %s",
2398 revs_file, strerror(errno));
2400 read_mailmap(&mailmap, ".mailmap", NULL);
2402 if (!incremental)
2403 setup_pager();
2405 assign_blame(&sb, opt);
2407 if (incremental)
2408 return 0;
2410 coalesce(&sb);
2412 if (!(output_option & OUTPUT_PORCELAIN))
2413 find_alignment(&sb, &output_option);
2415 output(&sb, output_option);
2416 free((void *)sb.final_buf);
2417 for (ent = sb.ent; ent; ) {
2418 struct blame_entry *e = ent->next;
2419 free(ent);
2420 ent = e;
2423 if (show_stats) {
2424 printf("num read blob: %d\n", num_read_blob);
2425 printf("num get patch: %d\n", num_get_patch);
2426 printf("num commits: %d\n", num_commits);
2428 return 0;