parseopt: add PARSE_OPT_KEEP_UNKNOWN
[alt-git.git] / builtin-blame.c
blob971126a80d3919c07b068b15540e297f64222248
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"
22 #include "utf8.h"
24 static char blame_usage[] = "git blame [options] [rev-opts] [rev] [--] file";
26 static const char *blame_opt_usage[] = {
27 blame_usage,
28 "",
29 "[rev-opts] are documented in git-rev-list(1)",
30 NULL
33 static int longest_file;
34 static int longest_author;
35 static int max_orig_digits;
36 static int max_digits;
37 static int max_score_digits;
38 static int show_root;
39 static int reverse;
40 static int blank_boundary;
41 static int incremental;
42 static int xdl_opts = XDF_NEED_MINIMAL;
43 static struct string_list mailmap;
45 #ifndef DEBUG
46 #define DEBUG 0
47 #endif
49 /* stats */
50 static int num_read_blob;
51 static int num_get_patch;
52 static int num_commits;
54 #define PICKAXE_BLAME_MOVE 01
55 #define PICKAXE_BLAME_COPY 02
56 #define PICKAXE_BLAME_COPY_HARDER 04
57 #define PICKAXE_BLAME_COPY_HARDEST 010
60 * blame for a blame_entry with score lower than these thresholds
61 * is not passed to the parent using move/copy logic.
63 static unsigned blame_move_score;
64 static unsigned blame_copy_score;
65 #define BLAME_DEFAULT_MOVE_SCORE 20
66 #define BLAME_DEFAULT_COPY_SCORE 40
68 /* bits #0..7 in revision.h, #8..11 used for merge_bases() in commit.c */
69 #define METAINFO_SHOWN (1u<<12)
70 #define MORE_THAN_ONE_PATH (1u<<13)
73 * One blob in a commit that is being suspected
75 struct origin {
76 int refcnt;
77 struct origin *previous;
78 struct commit *commit;
79 mmfile_t file;
80 unsigned char blob_sha1[20];
81 char path[FLEX_ARRAY];
85 * Given an origin, prepare mmfile_t structure to be used by the
86 * diff machinery
88 static void fill_origin_blob(struct origin *o, mmfile_t *file)
90 if (!o->file.ptr) {
91 enum object_type type;
92 num_read_blob++;
93 file->ptr = read_sha1_file(o->blob_sha1, &type,
94 (unsigned long *)(&(file->size)));
95 if (!file->ptr)
96 die("Cannot read blob %s for path %s",
97 sha1_to_hex(o->blob_sha1),
98 o->path);
99 o->file = *file;
101 else
102 *file = o->file;
106 * Origin is refcounted and usually we keep the blob contents to be
107 * reused.
109 static inline struct origin *origin_incref(struct origin *o)
111 if (o)
112 o->refcnt++;
113 return o;
116 static void origin_decref(struct origin *o)
118 if (o && --o->refcnt <= 0) {
119 if (o->previous)
120 origin_decref(o->previous);
121 free(o->file.ptr);
122 free(o);
126 static void drop_origin_blob(struct origin *o)
128 if (o->file.ptr) {
129 free(o->file.ptr);
130 o->file.ptr = NULL;
135 * Each group of lines is described by a blame_entry; it can be split
136 * as we pass blame to the parents. They form a linked list in the
137 * scoreboard structure, sorted by the target line number.
139 struct blame_entry {
140 struct blame_entry *prev;
141 struct blame_entry *next;
143 /* the first line of this group in the final image;
144 * internally all line numbers are 0 based.
146 int lno;
148 /* how many lines this group has */
149 int num_lines;
151 /* the commit that introduced this group into the final image */
152 struct origin *suspect;
154 /* true if the suspect is truly guilty; false while we have not
155 * checked if the group came from one of its parents.
157 char guilty;
159 /* true if the entry has been scanned for copies in the current parent
161 char scanned;
163 /* the line number of the first line of this group in the
164 * suspect's file; internally all line numbers are 0 based.
166 int s_lno;
168 /* how significant this entry is -- cached to avoid
169 * scanning the lines over and over.
171 unsigned score;
175 * The current state of the blame assignment.
177 struct scoreboard {
178 /* the final commit (i.e. where we started digging from) */
179 struct commit *final;
180 struct rev_info *revs;
181 const char *path;
184 * The contents in the final image.
185 * Used by many functions to obtain contents of the nth line,
186 * indexed with scoreboard.lineno[blame_entry.lno].
188 const char *final_buf;
189 unsigned long final_buf_size;
191 /* linked list of blames */
192 struct blame_entry *ent;
194 /* look-up a line in the final buffer */
195 int num_lines;
196 int *lineno;
199 static inline int same_suspect(struct origin *a, struct origin *b)
201 if (a == b)
202 return 1;
203 if (a->commit != b->commit)
204 return 0;
205 return !strcmp(a->path, b->path);
208 static void sanity_check_refcnt(struct scoreboard *);
211 * If two blame entries that are next to each other came from
212 * contiguous lines in the same origin (i.e. <commit, path> pair),
213 * merge them together.
215 static void coalesce(struct scoreboard *sb)
217 struct blame_entry *ent, *next;
219 for (ent = sb->ent; ent && (next = ent->next); ent = next) {
220 if (same_suspect(ent->suspect, next->suspect) &&
221 ent->guilty == next->guilty &&
222 ent->s_lno + ent->num_lines == next->s_lno) {
223 ent->num_lines += next->num_lines;
224 ent->next = next->next;
225 if (ent->next)
226 ent->next->prev = ent;
227 origin_decref(next->suspect);
228 free(next);
229 ent->score = 0;
230 next = ent; /* again */
234 if (DEBUG) /* sanity */
235 sanity_check_refcnt(sb);
239 * Given a commit and a path in it, create a new origin structure.
240 * The callers that add blame to the scoreboard should use
241 * get_origin() to obtain shared, refcounted copy instead of calling
242 * this function directly.
244 static struct origin *make_origin(struct commit *commit, const char *path)
246 struct origin *o;
247 o = xcalloc(1, sizeof(*o) + strlen(path) + 1);
248 o->commit = commit;
249 o->refcnt = 1;
250 strcpy(o->path, path);
251 return o;
255 * Locate an existing origin or create a new one.
257 static struct origin *get_origin(struct scoreboard *sb,
258 struct commit *commit,
259 const char *path)
261 struct blame_entry *e;
263 for (e = sb->ent; e; e = e->next) {
264 if (e->suspect->commit == commit &&
265 !strcmp(e->suspect->path, path))
266 return origin_incref(e->suspect);
268 return make_origin(commit, path);
272 * Fill the blob_sha1 field of an origin if it hasn't, so that later
273 * call to fill_origin_blob() can use it to locate the data. blob_sha1
274 * for an origin is also used to pass the blame for the entire file to
275 * the parent to detect the case where a child's blob is identical to
276 * that of its parent's.
278 static int fill_blob_sha1(struct origin *origin)
280 unsigned mode;
282 if (!is_null_sha1(origin->blob_sha1))
283 return 0;
284 if (get_tree_entry(origin->commit->object.sha1,
285 origin->path,
286 origin->blob_sha1, &mode))
287 goto error_out;
288 if (sha1_object_info(origin->blob_sha1, NULL) != OBJ_BLOB)
289 goto error_out;
290 return 0;
291 error_out:
292 hashclr(origin->blob_sha1);
293 return -1;
297 * We have an origin -- check if the same path exists in the
298 * parent and return an origin structure to represent it.
300 static struct origin *find_origin(struct scoreboard *sb,
301 struct commit *parent,
302 struct origin *origin)
304 struct origin *porigin = NULL;
305 struct diff_options diff_opts;
306 const char *paths[2];
308 if (parent->util) {
310 * Each commit object can cache one origin in that
311 * commit. This is a freestanding copy of origin and
312 * not refcounted.
314 struct origin *cached = parent->util;
315 if (!strcmp(cached->path, origin->path)) {
317 * The same path between origin and its parent
318 * without renaming -- the most common case.
320 porigin = get_origin(sb, parent, cached->path);
323 * If the origin was newly created (i.e. get_origin
324 * would call make_origin if none is found in the
325 * scoreboard), it does not know the blob_sha1,
326 * so copy it. Otherwise porigin was in the
327 * scoreboard and already knows blob_sha1.
329 if (porigin->refcnt == 1)
330 hashcpy(porigin->blob_sha1, cached->blob_sha1);
331 return porigin;
333 /* otherwise it was not very useful; free it */
334 free(parent->util);
335 parent->util = NULL;
338 /* See if the origin->path is different between parent
339 * and origin first. Most of the time they are the
340 * same and diff-tree is fairly efficient about this.
342 diff_setup(&diff_opts);
343 DIFF_OPT_SET(&diff_opts, RECURSIVE);
344 diff_opts.detect_rename = 0;
345 diff_opts.output_format = DIFF_FORMAT_NO_OUTPUT;
346 paths[0] = origin->path;
347 paths[1] = NULL;
349 diff_tree_setup_paths(paths, &diff_opts);
350 if (diff_setup_done(&diff_opts) < 0)
351 die("diff-setup");
353 if (is_null_sha1(origin->commit->object.sha1))
354 do_diff_cache(parent->tree->object.sha1, &diff_opts);
355 else
356 diff_tree_sha1(parent->tree->object.sha1,
357 origin->commit->tree->object.sha1,
358 "", &diff_opts);
359 diffcore_std(&diff_opts);
361 /* It is either one entry that says "modified", or "created",
362 * or nothing.
364 if (!diff_queued_diff.nr) {
365 /* The path is the same as parent */
366 porigin = get_origin(sb, parent, origin->path);
367 hashcpy(porigin->blob_sha1, origin->blob_sha1);
369 else if (diff_queued_diff.nr != 1)
370 die("internal error in blame::find_origin");
371 else {
372 struct diff_filepair *p = diff_queued_diff.queue[0];
373 switch (p->status) {
374 default:
375 die("internal error in blame::find_origin (%c)",
376 p->status);
377 case 'M':
378 porigin = get_origin(sb, parent, origin->path);
379 hashcpy(porigin->blob_sha1, p->one->sha1);
380 break;
381 case 'A':
382 case 'T':
383 /* Did not exist in parent, or type changed */
384 break;
387 diff_flush(&diff_opts);
388 diff_tree_release_paths(&diff_opts);
389 if (porigin) {
391 * Create a freestanding copy that is not part of
392 * the refcounted origin found in the scoreboard, and
393 * cache it in the commit.
395 struct origin *cached;
397 cached = make_origin(porigin->commit, porigin->path);
398 hashcpy(cached->blob_sha1, porigin->blob_sha1);
399 parent->util = cached;
401 return porigin;
405 * We have an origin -- find the path that corresponds to it in its
406 * parent and return an origin structure to represent it.
408 static struct origin *find_rename(struct scoreboard *sb,
409 struct commit *parent,
410 struct origin *origin)
412 struct origin *porigin = NULL;
413 struct diff_options diff_opts;
414 int i;
415 const char *paths[2];
417 diff_setup(&diff_opts);
418 DIFF_OPT_SET(&diff_opts, RECURSIVE);
419 diff_opts.detect_rename = DIFF_DETECT_RENAME;
420 diff_opts.output_format = DIFF_FORMAT_NO_OUTPUT;
421 diff_opts.single_follow = origin->path;
422 paths[0] = NULL;
423 diff_tree_setup_paths(paths, &diff_opts);
424 if (diff_setup_done(&diff_opts) < 0)
425 die("diff-setup");
427 if (is_null_sha1(origin->commit->object.sha1))
428 do_diff_cache(parent->tree->object.sha1, &diff_opts);
429 else
430 diff_tree_sha1(parent->tree->object.sha1,
431 origin->commit->tree->object.sha1,
432 "", &diff_opts);
433 diffcore_std(&diff_opts);
435 for (i = 0; i < diff_queued_diff.nr; i++) {
436 struct diff_filepair *p = diff_queued_diff.queue[i];
437 if ((p->status == 'R' || p->status == 'C') &&
438 !strcmp(p->two->path, origin->path)) {
439 porigin = get_origin(sb, parent, p->one->path);
440 hashcpy(porigin->blob_sha1, p->one->sha1);
441 break;
444 diff_flush(&diff_opts);
445 diff_tree_release_paths(&diff_opts);
446 return porigin;
450 * Link in a new blame entry to the scoreboard. Entries that cover the
451 * same line range have been removed from the scoreboard previously.
453 static void add_blame_entry(struct scoreboard *sb, struct blame_entry *e)
455 struct blame_entry *ent, *prev = NULL;
457 origin_incref(e->suspect);
459 for (ent = sb->ent; ent && ent->lno < e->lno; ent = ent->next)
460 prev = ent;
462 /* prev, if not NULL, is the last one that is below e */
463 e->prev = prev;
464 if (prev) {
465 e->next = prev->next;
466 prev->next = e;
468 else {
469 e->next = sb->ent;
470 sb->ent = e;
472 if (e->next)
473 e->next->prev = e;
477 * src typically is on-stack; we want to copy the information in it to
478 * a malloced blame_entry that is already on the linked list of the
479 * scoreboard. The origin of dst loses a refcnt while the origin of src
480 * gains one.
482 static void dup_entry(struct blame_entry *dst, struct blame_entry *src)
484 struct blame_entry *p, *n;
486 p = dst->prev;
487 n = dst->next;
488 origin_incref(src->suspect);
489 origin_decref(dst->suspect);
490 memcpy(dst, src, sizeof(*src));
491 dst->prev = p;
492 dst->next = n;
493 dst->score = 0;
496 static const char *nth_line(struct scoreboard *sb, int lno)
498 return sb->final_buf + sb->lineno[lno];
502 * It is known that lines between tlno to same came from parent, and e
503 * has an overlap with that range. it also is known that parent's
504 * line plno corresponds to e's line tlno.
506 * <---- e ----->
507 * <------>
508 * <------------>
509 * <------------>
510 * <------------------>
512 * Split e into potentially three parts; before this chunk, the chunk
513 * to be blamed for the parent, and after that portion.
515 static void split_overlap(struct blame_entry *split,
516 struct blame_entry *e,
517 int tlno, int plno, int same,
518 struct origin *parent)
520 int chunk_end_lno;
521 memset(split, 0, sizeof(struct blame_entry [3]));
523 if (e->s_lno < tlno) {
524 /* there is a pre-chunk part not blamed on parent */
525 split[0].suspect = origin_incref(e->suspect);
526 split[0].lno = e->lno;
527 split[0].s_lno = e->s_lno;
528 split[0].num_lines = tlno - e->s_lno;
529 split[1].lno = e->lno + tlno - e->s_lno;
530 split[1].s_lno = plno;
532 else {
533 split[1].lno = e->lno;
534 split[1].s_lno = plno + (e->s_lno - tlno);
537 if (same < e->s_lno + e->num_lines) {
538 /* there is a post-chunk part not blamed on parent */
539 split[2].suspect = origin_incref(e->suspect);
540 split[2].lno = e->lno + (same - e->s_lno);
541 split[2].s_lno = e->s_lno + (same - e->s_lno);
542 split[2].num_lines = e->s_lno + e->num_lines - same;
543 chunk_end_lno = split[2].lno;
545 else
546 chunk_end_lno = e->lno + e->num_lines;
547 split[1].num_lines = chunk_end_lno - split[1].lno;
550 * if it turns out there is nothing to blame the parent for,
551 * forget about the splitting. !split[1].suspect signals this.
553 if (split[1].num_lines < 1)
554 return;
555 split[1].suspect = origin_incref(parent);
559 * split_overlap() divided an existing blame e into up to three parts
560 * in split. Adjust the linked list of blames in the scoreboard to
561 * reflect the split.
563 static void split_blame(struct scoreboard *sb,
564 struct blame_entry *split,
565 struct blame_entry *e)
567 struct blame_entry *new_entry;
569 if (split[0].suspect && split[2].suspect) {
570 /* The first part (reuse storage for the existing entry e) */
571 dup_entry(e, &split[0]);
573 /* The last part -- me */
574 new_entry = xmalloc(sizeof(*new_entry));
575 memcpy(new_entry, &(split[2]), sizeof(struct blame_entry));
576 add_blame_entry(sb, new_entry);
578 /* ... and the middle part -- parent */
579 new_entry = xmalloc(sizeof(*new_entry));
580 memcpy(new_entry, &(split[1]), sizeof(struct blame_entry));
581 add_blame_entry(sb, new_entry);
583 else if (!split[0].suspect && !split[2].suspect)
585 * The parent covers the entire area; reuse storage for
586 * e and replace it with the parent.
588 dup_entry(e, &split[1]);
589 else if (split[0].suspect) {
590 /* me and then parent */
591 dup_entry(e, &split[0]);
593 new_entry = xmalloc(sizeof(*new_entry));
594 memcpy(new_entry, &(split[1]), sizeof(struct blame_entry));
595 add_blame_entry(sb, new_entry);
597 else {
598 /* parent and then me */
599 dup_entry(e, &split[1]);
601 new_entry = xmalloc(sizeof(*new_entry));
602 memcpy(new_entry, &(split[2]), sizeof(struct blame_entry));
603 add_blame_entry(sb, new_entry);
606 if (DEBUG) { /* sanity */
607 struct blame_entry *ent;
608 int lno = sb->ent->lno, corrupt = 0;
610 for (ent = sb->ent; ent; ent = ent->next) {
611 if (lno != ent->lno)
612 corrupt = 1;
613 if (ent->s_lno < 0)
614 corrupt = 1;
615 lno += ent->num_lines;
617 if (corrupt) {
618 lno = sb->ent->lno;
619 for (ent = sb->ent; ent; ent = ent->next) {
620 printf("L %8d l %8d n %8d\n",
621 lno, ent->lno, ent->num_lines);
622 lno = ent->lno + ent->num_lines;
624 die("oops");
630 * After splitting the blame, the origins used by the
631 * on-stack blame_entry should lose one refcnt each.
633 static void decref_split(struct blame_entry *split)
635 int i;
637 for (i = 0; i < 3; i++)
638 origin_decref(split[i].suspect);
642 * Helper for blame_chunk(). blame_entry e is known to overlap with
643 * the patch hunk; split it and pass blame to the parent.
645 static void blame_overlap(struct scoreboard *sb, struct blame_entry *e,
646 int tlno, int plno, int same,
647 struct origin *parent)
649 struct blame_entry split[3];
651 split_overlap(split, e, tlno, plno, same, parent);
652 if (split[1].suspect)
653 split_blame(sb, split, e);
654 decref_split(split);
658 * Find the line number of the last line the target is suspected for.
660 static int find_last_in_target(struct scoreboard *sb, struct origin *target)
662 struct blame_entry *e;
663 int last_in_target = -1;
665 for (e = sb->ent; e; e = e->next) {
666 if (e->guilty || !same_suspect(e->suspect, target))
667 continue;
668 if (last_in_target < e->s_lno + e->num_lines)
669 last_in_target = e->s_lno + e->num_lines;
671 return last_in_target;
675 * Process one hunk from the patch between the current suspect for
676 * blame_entry e and its parent. Find and split the overlap, and
677 * pass blame to the overlapping part to the parent.
679 static void blame_chunk(struct scoreboard *sb,
680 int tlno, int plno, int same,
681 struct origin *target, struct origin *parent)
683 struct blame_entry *e;
685 for (e = sb->ent; e; e = e->next) {
686 if (e->guilty || !same_suspect(e->suspect, target))
687 continue;
688 if (same <= e->s_lno)
689 continue;
690 if (tlno < e->s_lno + e->num_lines)
691 blame_overlap(sb, e, tlno, plno, same, parent);
695 struct blame_chunk_cb_data {
696 struct scoreboard *sb;
697 struct origin *target;
698 struct origin *parent;
699 long plno;
700 long tlno;
703 static void blame_chunk_cb(void *data, long same, long p_next, long t_next)
705 struct blame_chunk_cb_data *d = data;
706 blame_chunk(d->sb, d->tlno, d->plno, same, d->target, d->parent);
707 d->plno = p_next;
708 d->tlno = t_next;
712 * We are looking at the origin 'target' and aiming to pass blame
713 * for the lines it is suspected to its parent. Run diff to find
714 * which lines came from parent and pass blame for them.
716 static int pass_blame_to_parent(struct scoreboard *sb,
717 struct origin *target,
718 struct origin *parent)
720 int last_in_target;
721 mmfile_t file_p, file_o;
722 struct blame_chunk_cb_data d = { sb, target, parent, 0, 0 };
723 xpparam_t xpp;
724 xdemitconf_t xecfg;
726 last_in_target = find_last_in_target(sb, target);
727 if (last_in_target < 0)
728 return 1; /* nothing remains for this target */
730 fill_origin_blob(parent, &file_p);
731 fill_origin_blob(target, &file_o);
732 num_get_patch++;
734 memset(&xpp, 0, sizeof(xpp));
735 xpp.flags = xdl_opts;
736 memset(&xecfg, 0, sizeof(xecfg));
737 xecfg.ctxlen = 0;
738 xdi_diff_hunks(&file_p, &file_o, blame_chunk_cb, &d, &xpp, &xecfg);
739 /* The rest (i.e. anything after tlno) are the same as the parent */
740 blame_chunk(sb, d.tlno, d.plno, last_in_target, target, parent);
742 return 0;
746 * The lines in blame_entry after splitting blames many times can become
747 * very small and trivial, and at some point it becomes pointless to
748 * blame the parents. E.g. "\t\t}\n\t}\n\n" appears everywhere in any
749 * ordinary C program, and it is not worth to say it was copied from
750 * totally unrelated file in the parent.
752 * Compute how trivial the lines in the blame_entry are.
754 static unsigned ent_score(struct scoreboard *sb, struct blame_entry *e)
756 unsigned score;
757 const char *cp, *ep;
759 if (e->score)
760 return e->score;
762 score = 1;
763 cp = nth_line(sb, e->lno);
764 ep = nth_line(sb, e->lno + e->num_lines);
765 while (cp < ep) {
766 unsigned ch = *((unsigned char *)cp);
767 if (isalnum(ch))
768 score++;
769 cp++;
771 e->score = score;
772 return score;
776 * best_so_far[] and this[] are both a split of an existing blame_entry
777 * that passes blame to the parent. Maintain best_so_far the best split
778 * so far, by comparing this and best_so_far and copying this into
779 * bst_so_far as needed.
781 static void copy_split_if_better(struct scoreboard *sb,
782 struct blame_entry *best_so_far,
783 struct blame_entry *this)
785 int i;
787 if (!this[1].suspect)
788 return;
789 if (best_so_far[1].suspect) {
790 if (ent_score(sb, &this[1]) < ent_score(sb, &best_so_far[1]))
791 return;
794 for (i = 0; i < 3; i++)
795 origin_incref(this[i].suspect);
796 decref_split(best_so_far);
797 memcpy(best_so_far, this, sizeof(struct blame_entry [3]));
801 * We are looking at a part of the final image represented by
802 * ent (tlno and same are offset by ent->s_lno).
803 * tlno is where we are looking at in the final image.
804 * up to (but not including) same match preimage.
805 * plno is where we are looking at in the preimage.
807 * <-------------- final image ---------------------->
808 * <------ent------>
809 * ^tlno ^same
810 * <---------preimage----->
811 * ^plno
813 * All line numbers are 0-based.
815 static void handle_split(struct scoreboard *sb,
816 struct blame_entry *ent,
817 int tlno, int plno, int same,
818 struct origin *parent,
819 struct blame_entry *split)
821 if (ent->num_lines <= tlno)
822 return;
823 if (tlno < same) {
824 struct blame_entry this[3];
825 tlno += ent->s_lno;
826 same += ent->s_lno;
827 split_overlap(this, ent, tlno, plno, same, parent);
828 copy_split_if_better(sb, split, this);
829 decref_split(this);
833 struct handle_split_cb_data {
834 struct scoreboard *sb;
835 struct blame_entry *ent;
836 struct origin *parent;
837 struct blame_entry *split;
838 long plno;
839 long tlno;
842 static void handle_split_cb(void *data, long same, long p_next, long t_next)
844 struct handle_split_cb_data *d = data;
845 handle_split(d->sb, d->ent, d->tlno, d->plno, same, d->parent, d->split);
846 d->plno = p_next;
847 d->tlno = t_next;
851 * Find the lines from parent that are the same as ent so that
852 * we can pass blames to it. file_p has the blob contents for
853 * the parent.
855 static void find_copy_in_blob(struct scoreboard *sb,
856 struct blame_entry *ent,
857 struct origin *parent,
858 struct blame_entry *split,
859 mmfile_t *file_p)
861 const char *cp;
862 int cnt;
863 mmfile_t file_o;
864 struct handle_split_cb_data d = { sb, ent, parent, split, 0, 0 };
865 xpparam_t xpp;
866 xdemitconf_t xecfg;
869 * Prepare mmfile that contains only the lines in ent.
871 cp = nth_line(sb, ent->lno);
872 file_o.ptr = (char*) cp;
873 cnt = ent->num_lines;
875 while (cnt && cp < sb->final_buf + sb->final_buf_size) {
876 if (*cp++ == '\n')
877 cnt--;
879 file_o.size = cp - file_o.ptr;
882 * file_o is a part of final image we are annotating.
883 * file_p partially may match that image.
885 memset(&xpp, 0, sizeof(xpp));
886 xpp.flags = xdl_opts;
887 memset(&xecfg, 0, sizeof(xecfg));
888 xecfg.ctxlen = 1;
889 memset(split, 0, sizeof(struct blame_entry [3]));
890 xdi_diff_hunks(file_p, &file_o, handle_split_cb, &d, &xpp, &xecfg);
891 /* remainder, if any, all match the preimage */
892 handle_split(sb, ent, d.tlno, d.plno, ent->num_lines, parent, split);
896 * See if lines currently target is suspected for can be attributed to
897 * parent.
899 static int find_move_in_parent(struct scoreboard *sb,
900 struct origin *target,
901 struct origin *parent)
903 int last_in_target, made_progress;
904 struct blame_entry *e, split[3];
905 mmfile_t file_p;
907 last_in_target = find_last_in_target(sb, target);
908 if (last_in_target < 0)
909 return 1; /* nothing remains for this target */
911 fill_origin_blob(parent, &file_p);
912 if (!file_p.ptr)
913 return 0;
915 made_progress = 1;
916 while (made_progress) {
917 made_progress = 0;
918 for (e = sb->ent; e; e = e->next) {
919 if (e->guilty || !same_suspect(e->suspect, target) ||
920 ent_score(sb, e) < blame_move_score)
921 continue;
922 find_copy_in_blob(sb, e, parent, split, &file_p);
923 if (split[1].suspect &&
924 blame_move_score < ent_score(sb, &split[1])) {
925 split_blame(sb, split, e);
926 made_progress = 1;
928 decref_split(split);
931 return 0;
934 struct blame_list {
935 struct blame_entry *ent;
936 struct blame_entry split[3];
940 * Count the number of entries the target is suspected for,
941 * and prepare a list of entry and the best split.
943 static struct blame_list *setup_blame_list(struct scoreboard *sb,
944 struct origin *target,
945 int min_score,
946 int *num_ents_p)
948 struct blame_entry *e;
949 int num_ents, i;
950 struct blame_list *blame_list = NULL;
952 for (e = sb->ent, num_ents = 0; e; e = e->next)
953 if (!e->scanned && !e->guilty &&
954 same_suspect(e->suspect, target) &&
955 min_score < ent_score(sb, e))
956 num_ents++;
957 if (num_ents) {
958 blame_list = xcalloc(num_ents, sizeof(struct blame_list));
959 for (e = sb->ent, i = 0; e; e = e->next)
960 if (!e->scanned && !e->guilty &&
961 same_suspect(e->suspect, target) &&
962 min_score < ent_score(sb, e))
963 blame_list[i++].ent = e;
965 *num_ents_p = num_ents;
966 return blame_list;
970 * Reset the scanned status on all entries.
972 static void reset_scanned_flag(struct scoreboard *sb)
974 struct blame_entry *e;
975 for (e = sb->ent; e; e = e->next)
976 e->scanned = 0;
980 * For lines target is suspected for, see if we can find code movement
981 * across file boundary from the parent commit. porigin is the path
982 * in the parent we already tried.
984 static int find_copy_in_parent(struct scoreboard *sb,
985 struct origin *target,
986 struct commit *parent,
987 struct origin *porigin,
988 int opt)
990 struct diff_options diff_opts;
991 const char *paths[1];
992 int i, j;
993 int retval;
994 struct blame_list *blame_list;
995 int num_ents;
997 blame_list = setup_blame_list(sb, target, blame_copy_score, &num_ents);
998 if (!blame_list)
999 return 1; /* nothing remains for this target */
1001 diff_setup(&diff_opts);
1002 DIFF_OPT_SET(&diff_opts, RECURSIVE);
1003 diff_opts.output_format = DIFF_FORMAT_NO_OUTPUT;
1005 paths[0] = NULL;
1006 diff_tree_setup_paths(paths, &diff_opts);
1007 if (diff_setup_done(&diff_opts) < 0)
1008 die("diff-setup");
1010 /* Try "find copies harder" on new path if requested;
1011 * we do not want to use diffcore_rename() actually to
1012 * match things up; find_copies_harder is set only to
1013 * force diff_tree_sha1() to feed all filepairs to diff_queue,
1014 * and this code needs to be after diff_setup_done(), which
1015 * usually makes find-copies-harder imply copy detection.
1017 if ((opt & PICKAXE_BLAME_COPY_HARDEST)
1018 || ((opt & PICKAXE_BLAME_COPY_HARDER)
1019 && (!porigin || strcmp(target->path, porigin->path))))
1020 DIFF_OPT_SET(&diff_opts, FIND_COPIES_HARDER);
1022 if (is_null_sha1(target->commit->object.sha1))
1023 do_diff_cache(parent->tree->object.sha1, &diff_opts);
1024 else
1025 diff_tree_sha1(parent->tree->object.sha1,
1026 target->commit->tree->object.sha1,
1027 "", &diff_opts);
1029 if (!DIFF_OPT_TST(&diff_opts, FIND_COPIES_HARDER))
1030 diffcore_std(&diff_opts);
1032 retval = 0;
1033 while (1) {
1034 int made_progress = 0;
1036 for (i = 0; i < diff_queued_diff.nr; i++) {
1037 struct diff_filepair *p = diff_queued_diff.queue[i];
1038 struct origin *norigin;
1039 mmfile_t file_p;
1040 struct blame_entry this[3];
1042 if (!DIFF_FILE_VALID(p->one))
1043 continue; /* does not exist in parent */
1044 if (S_ISGITLINK(p->one->mode))
1045 continue; /* ignore git links */
1046 if (porigin && !strcmp(p->one->path, porigin->path))
1047 /* find_move already dealt with this path */
1048 continue;
1050 norigin = get_origin(sb, parent, p->one->path);
1051 hashcpy(norigin->blob_sha1, p->one->sha1);
1052 fill_origin_blob(norigin, &file_p);
1053 if (!file_p.ptr)
1054 continue;
1056 for (j = 0; j < num_ents; j++) {
1057 find_copy_in_blob(sb, blame_list[j].ent,
1058 norigin, this, &file_p);
1059 copy_split_if_better(sb, blame_list[j].split,
1060 this);
1061 decref_split(this);
1063 origin_decref(norigin);
1066 for (j = 0; j < num_ents; j++) {
1067 struct blame_entry *split = blame_list[j].split;
1068 if (split[1].suspect &&
1069 blame_copy_score < ent_score(sb, &split[1])) {
1070 split_blame(sb, split, blame_list[j].ent);
1071 made_progress = 1;
1073 else
1074 blame_list[j].ent->scanned = 1;
1075 decref_split(split);
1077 free(blame_list);
1079 if (!made_progress)
1080 break;
1081 blame_list = setup_blame_list(sb, target, blame_copy_score, &num_ents);
1082 if (!blame_list) {
1083 retval = 1;
1084 break;
1087 reset_scanned_flag(sb);
1088 diff_flush(&diff_opts);
1089 diff_tree_release_paths(&diff_opts);
1090 return retval;
1094 * The blobs of origin and porigin exactly match, so everything
1095 * origin is suspected for can be blamed on the parent.
1097 static void pass_whole_blame(struct scoreboard *sb,
1098 struct origin *origin, struct origin *porigin)
1100 struct blame_entry *e;
1102 if (!porigin->file.ptr && origin->file.ptr) {
1103 /* Steal its file */
1104 porigin->file = origin->file;
1105 origin->file.ptr = NULL;
1107 for (e = sb->ent; e; e = e->next) {
1108 if (!same_suspect(e->suspect, origin))
1109 continue;
1110 origin_incref(porigin);
1111 origin_decref(e->suspect);
1112 e->suspect = porigin;
1117 * We pass blame from the current commit to its parents. We keep saying
1118 * "parent" (and "porigin"), but what we mean is to find scapegoat to
1119 * exonerate ourselves.
1121 static struct commit_list *first_scapegoat(struct rev_info *revs, struct commit *commit)
1123 if (!reverse)
1124 return commit->parents;
1125 return lookup_decoration(&revs->children, &commit->object);
1128 static int num_scapegoats(struct rev_info *revs, struct commit *commit)
1130 int cnt;
1131 struct commit_list *l = first_scapegoat(revs, commit);
1132 for (cnt = 0; l; l = l->next)
1133 cnt++;
1134 return cnt;
1137 #define MAXSG 16
1139 static void pass_blame(struct scoreboard *sb, struct origin *origin, int opt)
1141 struct rev_info *revs = sb->revs;
1142 int i, pass, num_sg;
1143 struct commit *commit = origin->commit;
1144 struct commit_list *sg;
1145 struct origin *sg_buf[MAXSG];
1146 struct origin *porigin, **sg_origin = sg_buf;
1148 num_sg = num_scapegoats(revs, commit);
1149 if (!num_sg)
1150 goto finish;
1151 else if (num_sg < ARRAY_SIZE(sg_buf))
1152 memset(sg_buf, 0, sizeof(sg_buf));
1153 else
1154 sg_origin = xcalloc(num_sg, sizeof(*sg_origin));
1157 * The first pass looks for unrenamed path to optimize for
1158 * common cases, then we look for renames in the second pass.
1160 for (pass = 0; pass < 2; pass++) {
1161 struct origin *(*find)(struct scoreboard *,
1162 struct commit *, struct origin *);
1163 find = pass ? find_rename : find_origin;
1165 for (i = 0, sg = first_scapegoat(revs, commit);
1166 i < num_sg && sg;
1167 sg = sg->next, i++) {
1168 struct commit *p = sg->item;
1169 int j, same;
1171 if (sg_origin[i])
1172 continue;
1173 if (parse_commit(p))
1174 continue;
1175 porigin = find(sb, p, origin);
1176 if (!porigin)
1177 continue;
1178 if (!hashcmp(porigin->blob_sha1, origin->blob_sha1)) {
1179 pass_whole_blame(sb, origin, porigin);
1180 origin_decref(porigin);
1181 goto finish;
1183 for (j = same = 0; j < i; j++)
1184 if (sg_origin[j] &&
1185 !hashcmp(sg_origin[j]->blob_sha1,
1186 porigin->blob_sha1)) {
1187 same = 1;
1188 break;
1190 if (!same)
1191 sg_origin[i] = porigin;
1192 else
1193 origin_decref(porigin);
1197 num_commits++;
1198 for (i = 0, sg = first_scapegoat(revs, commit);
1199 i < num_sg && sg;
1200 sg = sg->next, i++) {
1201 struct origin *porigin = sg_origin[i];
1202 if (!porigin)
1203 continue;
1204 if (!origin->previous) {
1205 origin_incref(porigin);
1206 origin->previous = porigin;
1208 if (pass_blame_to_parent(sb, origin, porigin))
1209 goto finish;
1213 * Optionally find moves in parents' files.
1215 if (opt & PICKAXE_BLAME_MOVE)
1216 for (i = 0, sg = first_scapegoat(revs, commit);
1217 i < num_sg && sg;
1218 sg = sg->next, i++) {
1219 struct origin *porigin = sg_origin[i];
1220 if (!porigin)
1221 continue;
1222 if (find_move_in_parent(sb, origin, porigin))
1223 goto finish;
1227 * Optionally find copies from parents' files.
1229 if (opt & PICKAXE_BLAME_COPY)
1230 for (i = 0, sg = first_scapegoat(revs, commit);
1231 i < num_sg && sg;
1232 sg = sg->next, i++) {
1233 struct origin *porigin = sg_origin[i];
1234 if (find_copy_in_parent(sb, origin, sg->item,
1235 porigin, opt))
1236 goto finish;
1239 finish:
1240 for (i = 0; i < num_sg; i++) {
1241 if (sg_origin[i]) {
1242 drop_origin_blob(sg_origin[i]);
1243 origin_decref(sg_origin[i]);
1246 drop_origin_blob(origin);
1247 if (sg_buf != sg_origin)
1248 free(sg_origin);
1252 * Information on commits, used for output.
1254 struct commit_info
1256 const char *author;
1257 const char *author_mail;
1258 unsigned long author_time;
1259 const char *author_tz;
1261 /* filled only when asked for details */
1262 const char *committer;
1263 const char *committer_mail;
1264 unsigned long committer_time;
1265 const char *committer_tz;
1267 const char *summary;
1271 * Parse author/committer line in the commit object buffer
1273 static void get_ac_line(const char *inbuf, const char *what,
1274 int person_len, char *person,
1275 int mail_len, char *mail,
1276 unsigned long *time, const char **tz)
1278 int len, tzlen, maillen;
1279 char *tmp, *endp, *timepos, *mailpos;
1281 tmp = strstr(inbuf, what);
1282 if (!tmp)
1283 goto error_out;
1284 tmp += strlen(what);
1285 endp = strchr(tmp, '\n');
1286 if (!endp)
1287 len = strlen(tmp);
1288 else
1289 len = endp - tmp;
1290 if (person_len <= len) {
1291 error_out:
1292 /* Ugh */
1293 *tz = "(unknown)";
1294 strcpy(mail, *tz);
1295 *time = 0;
1296 return;
1298 memcpy(person, tmp, len);
1300 tmp = person;
1301 tmp += len;
1302 *tmp = 0;
1303 while (*tmp != ' ')
1304 tmp--;
1305 *tz = tmp+1;
1306 tzlen = (person+len)-(tmp+1);
1308 *tmp = 0;
1309 while (*tmp != ' ')
1310 tmp--;
1311 *time = strtoul(tmp, NULL, 10);
1312 timepos = tmp;
1314 *tmp = 0;
1315 while (*tmp != ' ')
1316 tmp--;
1317 mailpos = tmp + 1;
1318 *tmp = 0;
1319 maillen = timepos - tmp;
1320 memcpy(mail, mailpos, maillen);
1322 if (!mailmap.nr)
1323 return;
1326 * mailmap expansion may make the name longer.
1327 * make room by pushing stuff down.
1329 tmp = person + person_len - (tzlen + 1);
1330 memmove(tmp, *tz, tzlen);
1331 tmp[tzlen] = 0;
1332 *tz = tmp;
1335 * Now, convert both name and e-mail using mailmap
1337 if(map_user(&mailmap, mail+1, mail_len-1, person, tmp-person-1)) {
1338 /* Add a trailing '>' to email, since map_user returns plain emails
1339 Note: It already has '<', since we replace from mail+1 */
1340 mailpos = memchr(mail, '\0', mail_len);
1341 if (mailpos && mailpos-mail < mail_len - 1) {
1342 *mailpos = '>';
1343 *(mailpos+1) = '\0';
1348 static void get_commit_info(struct commit *commit,
1349 struct commit_info *ret,
1350 int detailed)
1352 int len;
1353 char *tmp, *endp, *reencoded, *message;
1354 static char author_name[1024];
1355 static char author_mail[1024];
1356 static char committer_name[1024];
1357 static char committer_mail[1024];
1358 static char summary_buf[1024];
1361 * We've operated without save_commit_buffer, so
1362 * we now need to populate them for output.
1364 if (!commit->buffer) {
1365 enum object_type type;
1366 unsigned long size;
1367 commit->buffer =
1368 read_sha1_file(commit->object.sha1, &type, &size);
1369 if (!commit->buffer)
1370 die("Cannot read commit %s",
1371 sha1_to_hex(commit->object.sha1));
1373 reencoded = reencode_commit_message(commit, NULL);
1374 message = reencoded ? reencoded : commit->buffer;
1375 ret->author = author_name;
1376 ret->author_mail = author_mail;
1377 get_ac_line(message, "\nauthor ",
1378 sizeof(author_name), author_name,
1379 sizeof(author_mail), author_mail,
1380 &ret->author_time, &ret->author_tz);
1382 if (!detailed) {
1383 free(reencoded);
1384 return;
1387 ret->committer = committer_name;
1388 ret->committer_mail = committer_mail;
1389 get_ac_line(message, "\ncommitter ",
1390 sizeof(committer_name), committer_name,
1391 sizeof(committer_mail), committer_mail,
1392 &ret->committer_time, &ret->committer_tz);
1394 ret->summary = summary_buf;
1395 tmp = strstr(message, "\n\n");
1396 if (!tmp) {
1397 error_out:
1398 sprintf(summary_buf, "(%s)", sha1_to_hex(commit->object.sha1));
1399 free(reencoded);
1400 return;
1402 tmp += 2;
1403 endp = strchr(tmp, '\n');
1404 if (!endp)
1405 endp = tmp + strlen(tmp);
1406 len = endp - tmp;
1407 if (len >= sizeof(summary_buf) || len == 0)
1408 goto error_out;
1409 memcpy(summary_buf, tmp, len);
1410 summary_buf[len] = 0;
1411 free(reencoded);
1415 * To allow LF and other nonportable characters in pathnames,
1416 * they are c-style quoted as needed.
1418 static void write_filename_info(const char *path)
1420 printf("filename ");
1421 write_name_quoted(path, stdout, '\n');
1425 * Porcelain/Incremental format wants to show a lot of details per
1426 * commit. Instead of repeating this every line, emit it only once,
1427 * the first time each commit appears in the output.
1429 static int emit_one_suspect_detail(struct origin *suspect)
1431 struct commit_info ci;
1433 if (suspect->commit->object.flags & METAINFO_SHOWN)
1434 return 0;
1436 suspect->commit->object.flags |= METAINFO_SHOWN;
1437 get_commit_info(suspect->commit, &ci, 1);
1438 printf("author %s\n", ci.author);
1439 printf("author-mail %s\n", ci.author_mail);
1440 printf("author-time %lu\n", ci.author_time);
1441 printf("author-tz %s\n", ci.author_tz);
1442 printf("committer %s\n", ci.committer);
1443 printf("committer-mail %s\n", ci.committer_mail);
1444 printf("committer-time %lu\n", ci.committer_time);
1445 printf("committer-tz %s\n", ci.committer_tz);
1446 printf("summary %s\n", ci.summary);
1447 if (suspect->commit->object.flags & UNINTERESTING)
1448 printf("boundary\n");
1449 if (suspect->previous) {
1450 struct origin *prev = suspect->previous;
1451 printf("previous %s ", sha1_to_hex(prev->commit->object.sha1));
1452 write_name_quoted(prev->path, stdout, '\n');
1454 return 1;
1458 * The blame_entry is found to be guilty for the range. Mark it
1459 * as such, and show it in incremental output.
1461 static void found_guilty_entry(struct blame_entry *ent)
1463 if (ent->guilty)
1464 return;
1465 ent->guilty = 1;
1466 if (incremental) {
1467 struct origin *suspect = ent->suspect;
1469 printf("%s %d %d %d\n",
1470 sha1_to_hex(suspect->commit->object.sha1),
1471 ent->s_lno + 1, ent->lno + 1, ent->num_lines);
1472 emit_one_suspect_detail(suspect);
1473 write_filename_info(suspect->path);
1474 maybe_flush_or_die(stdout, "stdout");
1479 * The main loop -- while the scoreboard has lines whose true origin
1480 * is still unknown, pick one blame_entry, and allow its current
1481 * suspect to pass blames to its parents.
1483 static void assign_blame(struct scoreboard *sb, int opt)
1485 struct rev_info *revs = sb->revs;
1487 while (1) {
1488 struct blame_entry *ent;
1489 struct commit *commit;
1490 struct origin *suspect = NULL;
1492 /* find one suspect to break down */
1493 for (ent = sb->ent; !suspect && ent; ent = ent->next)
1494 if (!ent->guilty)
1495 suspect = ent->suspect;
1496 if (!suspect)
1497 return; /* all done */
1500 * We will use this suspect later in the loop,
1501 * so hold onto it in the meantime.
1503 origin_incref(suspect);
1504 commit = suspect->commit;
1505 if (!commit->object.parsed)
1506 parse_commit(commit);
1507 if (reverse ||
1508 (!(commit->object.flags & UNINTERESTING) &&
1509 !(revs->max_age != -1 && commit->date < revs->max_age)))
1510 pass_blame(sb, suspect, opt);
1511 else {
1512 commit->object.flags |= UNINTERESTING;
1513 if (commit->object.parsed)
1514 mark_parents_uninteresting(commit);
1516 /* treat root commit as boundary */
1517 if (!commit->parents && !show_root)
1518 commit->object.flags |= UNINTERESTING;
1520 /* Take responsibility for the remaining entries */
1521 for (ent = sb->ent; ent; ent = ent->next)
1522 if (same_suspect(ent->suspect, suspect))
1523 found_guilty_entry(ent);
1524 origin_decref(suspect);
1526 if (DEBUG) /* sanity */
1527 sanity_check_refcnt(sb);
1531 static const char *format_time(unsigned long time, const char *tz_str,
1532 int show_raw_time)
1534 static char time_buf[128];
1535 time_t t = time;
1536 int minutes, tz;
1537 struct tm *tm;
1539 if (show_raw_time) {
1540 sprintf(time_buf, "%lu %s", time, tz_str);
1541 return time_buf;
1544 tz = atoi(tz_str);
1545 minutes = tz < 0 ? -tz : tz;
1546 minutes = (minutes / 100)*60 + (minutes % 100);
1547 minutes = tz < 0 ? -minutes : minutes;
1548 t = time + minutes * 60;
1549 tm = gmtime(&t);
1551 strftime(time_buf, sizeof(time_buf), "%Y-%m-%d %H:%M:%S ", tm);
1552 strcat(time_buf, tz_str);
1553 return time_buf;
1556 #define OUTPUT_ANNOTATE_COMPAT 001
1557 #define OUTPUT_LONG_OBJECT_NAME 002
1558 #define OUTPUT_RAW_TIMESTAMP 004
1559 #define OUTPUT_PORCELAIN 010
1560 #define OUTPUT_SHOW_NAME 020
1561 #define OUTPUT_SHOW_NUMBER 040
1562 #define OUTPUT_SHOW_SCORE 0100
1563 #define OUTPUT_NO_AUTHOR 0200
1565 static void emit_porcelain(struct scoreboard *sb, struct blame_entry *ent)
1567 int cnt;
1568 const char *cp;
1569 struct origin *suspect = ent->suspect;
1570 char hex[41];
1572 strcpy(hex, sha1_to_hex(suspect->commit->object.sha1));
1573 printf("%s%c%d %d %d\n",
1574 hex,
1575 ent->guilty ? ' ' : '*', // purely for debugging
1576 ent->s_lno + 1,
1577 ent->lno + 1,
1578 ent->num_lines);
1579 if (emit_one_suspect_detail(suspect) ||
1580 (suspect->commit->object.flags & MORE_THAN_ONE_PATH))
1581 write_filename_info(suspect->path);
1583 cp = nth_line(sb, ent->lno);
1584 for (cnt = 0; cnt < ent->num_lines; cnt++) {
1585 char ch;
1586 if (cnt)
1587 printf("%s %d %d\n", hex,
1588 ent->s_lno + 1 + cnt,
1589 ent->lno + 1 + cnt);
1590 putchar('\t');
1591 do {
1592 ch = *cp++;
1593 putchar(ch);
1594 } while (ch != '\n' &&
1595 cp < sb->final_buf + sb->final_buf_size);
1599 static void emit_other(struct scoreboard *sb, struct blame_entry *ent, int opt)
1601 int cnt;
1602 const char *cp;
1603 struct origin *suspect = ent->suspect;
1604 struct commit_info ci;
1605 char hex[41];
1606 int show_raw_time = !!(opt & OUTPUT_RAW_TIMESTAMP);
1608 get_commit_info(suspect->commit, &ci, 1);
1609 strcpy(hex, sha1_to_hex(suspect->commit->object.sha1));
1611 cp = nth_line(sb, ent->lno);
1612 for (cnt = 0; cnt < ent->num_lines; cnt++) {
1613 char ch;
1614 int length = (opt & OUTPUT_LONG_OBJECT_NAME) ? 40 : 8;
1616 if (suspect->commit->object.flags & UNINTERESTING) {
1617 if (blank_boundary)
1618 memset(hex, ' ', length);
1619 else if (!(opt & OUTPUT_ANNOTATE_COMPAT)) {
1620 length--;
1621 putchar('^');
1625 printf("%.*s", length, hex);
1626 if (opt & OUTPUT_ANNOTATE_COMPAT)
1627 printf("\t(%10s\t%10s\t%d)", ci.author,
1628 format_time(ci.author_time, ci.author_tz,
1629 show_raw_time),
1630 ent->lno + 1 + cnt);
1631 else {
1632 if (opt & OUTPUT_SHOW_SCORE)
1633 printf(" %*d %02d",
1634 max_score_digits, ent->score,
1635 ent->suspect->refcnt);
1636 if (opt & OUTPUT_SHOW_NAME)
1637 printf(" %-*.*s", longest_file, longest_file,
1638 suspect->path);
1639 if (opt & OUTPUT_SHOW_NUMBER)
1640 printf(" %*d", max_orig_digits,
1641 ent->s_lno + 1 + cnt);
1643 if (!(opt & OUTPUT_NO_AUTHOR)) {
1644 int pad = longest_author - utf8_strwidth(ci.author);
1645 printf(" (%s%*s %10s",
1646 ci.author, pad, "",
1647 format_time(ci.author_time,
1648 ci.author_tz,
1649 show_raw_time));
1651 printf(" %*d) ",
1652 max_digits, ent->lno + 1 + cnt);
1654 do {
1655 ch = *cp++;
1656 putchar(ch);
1657 } while (ch != '\n' &&
1658 cp < sb->final_buf + sb->final_buf_size);
1662 static void output(struct scoreboard *sb, int option)
1664 struct blame_entry *ent;
1666 if (option & OUTPUT_PORCELAIN) {
1667 for (ent = sb->ent; ent; ent = ent->next) {
1668 struct blame_entry *oth;
1669 struct origin *suspect = ent->suspect;
1670 struct commit *commit = suspect->commit;
1671 if (commit->object.flags & MORE_THAN_ONE_PATH)
1672 continue;
1673 for (oth = ent->next; oth; oth = oth->next) {
1674 if ((oth->suspect->commit != commit) ||
1675 !strcmp(oth->suspect->path, suspect->path))
1676 continue;
1677 commit->object.flags |= MORE_THAN_ONE_PATH;
1678 break;
1683 for (ent = sb->ent; ent; ent = ent->next) {
1684 if (option & OUTPUT_PORCELAIN)
1685 emit_porcelain(sb, ent);
1686 else {
1687 emit_other(sb, ent, option);
1693 * To allow quick access to the contents of nth line in the
1694 * final image, prepare an index in the scoreboard.
1696 static int prepare_lines(struct scoreboard *sb)
1698 const char *buf = sb->final_buf;
1699 unsigned long len = sb->final_buf_size;
1700 int num = 0, incomplete = 0, bol = 1;
1702 if (len && buf[len-1] != '\n')
1703 incomplete++; /* incomplete line at the end */
1704 while (len--) {
1705 if (bol) {
1706 sb->lineno = xrealloc(sb->lineno,
1707 sizeof(int* ) * (num + 1));
1708 sb->lineno[num] = buf - sb->final_buf;
1709 bol = 0;
1711 if (*buf++ == '\n') {
1712 num++;
1713 bol = 1;
1716 sb->lineno = xrealloc(sb->lineno,
1717 sizeof(int* ) * (num + incomplete + 1));
1718 sb->lineno[num + incomplete] = buf - sb->final_buf;
1719 sb->num_lines = num + incomplete;
1720 return sb->num_lines;
1724 * Add phony grafts for use with -S; this is primarily to
1725 * support git's cvsserver that wants to give a linear history
1726 * to its clients.
1728 static int read_ancestry(const char *graft_file)
1730 FILE *fp = fopen(graft_file, "r");
1731 char buf[1024];
1732 if (!fp)
1733 return -1;
1734 while (fgets(buf, sizeof(buf), fp)) {
1735 /* The format is just "Commit Parent1 Parent2 ...\n" */
1736 int len = strlen(buf);
1737 struct commit_graft *graft = read_graft_line(buf, len);
1738 if (graft)
1739 register_commit_graft(graft, 0);
1741 fclose(fp);
1742 return 0;
1746 * How many columns do we need to show line numbers in decimal?
1748 static int lineno_width(int lines)
1750 int i, width;
1752 for (width = 1, i = 10; i <= lines + 1; width++)
1753 i *= 10;
1754 return width;
1758 * How many columns do we need to show line numbers, authors,
1759 * and filenames?
1761 static void find_alignment(struct scoreboard *sb, int *option)
1763 int longest_src_lines = 0;
1764 int longest_dst_lines = 0;
1765 unsigned largest_score = 0;
1766 struct blame_entry *e;
1768 for (e = sb->ent; e; e = e->next) {
1769 struct origin *suspect = e->suspect;
1770 struct commit_info ci;
1771 int num;
1773 if (strcmp(suspect->path, sb->path))
1774 *option |= OUTPUT_SHOW_NAME;
1775 num = strlen(suspect->path);
1776 if (longest_file < num)
1777 longest_file = num;
1778 if (!(suspect->commit->object.flags & METAINFO_SHOWN)) {
1779 suspect->commit->object.flags |= METAINFO_SHOWN;
1780 get_commit_info(suspect->commit, &ci, 1);
1781 num = utf8_strwidth(ci.author);
1782 if (longest_author < num)
1783 longest_author = num;
1785 num = e->s_lno + e->num_lines;
1786 if (longest_src_lines < num)
1787 longest_src_lines = num;
1788 num = e->lno + e->num_lines;
1789 if (longest_dst_lines < num)
1790 longest_dst_lines = num;
1791 if (largest_score < ent_score(sb, e))
1792 largest_score = ent_score(sb, e);
1794 max_orig_digits = lineno_width(longest_src_lines);
1795 max_digits = lineno_width(longest_dst_lines);
1796 max_score_digits = lineno_width(largest_score);
1800 * For debugging -- origin is refcounted, and this asserts that
1801 * we do not underflow.
1803 static void sanity_check_refcnt(struct scoreboard *sb)
1805 int baa = 0;
1806 struct blame_entry *ent;
1808 for (ent = sb->ent; ent; ent = ent->next) {
1809 /* Nobody should have zero or negative refcnt */
1810 if (ent->suspect->refcnt <= 0) {
1811 fprintf(stderr, "%s in %s has negative refcnt %d\n",
1812 ent->suspect->path,
1813 sha1_to_hex(ent->suspect->commit->object.sha1),
1814 ent->suspect->refcnt);
1815 baa = 1;
1818 if (baa) {
1819 int opt = 0160;
1820 find_alignment(sb, &opt);
1821 output(sb, opt);
1822 die("Baa %d!", baa);
1827 * Used for the command line parsing; check if the path exists
1828 * in the working tree.
1830 static int has_string_in_work_tree(const char *path)
1832 struct stat st;
1833 return !lstat(path, &st);
1836 static unsigned parse_score(const char *arg)
1838 char *end;
1839 unsigned long score = strtoul(arg, &end, 10);
1840 if (*end)
1841 return 0;
1842 return score;
1845 static const char *add_prefix(const char *prefix, const char *path)
1847 return prefix_path(prefix, prefix ? strlen(prefix) : 0, path);
1851 * Parsing of (comma separated) one item in the -L option
1853 static const char *parse_loc(const char *spec,
1854 struct scoreboard *sb, long lno,
1855 long begin, long *ret)
1857 char *term;
1858 const char *line;
1859 long num;
1860 int reg_error;
1861 regex_t regexp;
1862 regmatch_t match[1];
1864 /* Allow "-L <something>,+20" to mean starting at <something>
1865 * for 20 lines, or "-L <something>,-5" for 5 lines ending at
1866 * <something>.
1868 if (1 < begin && (spec[0] == '+' || spec[0] == '-')) {
1869 num = strtol(spec + 1, &term, 10);
1870 if (term != spec + 1) {
1871 if (spec[0] == '-')
1872 num = 0 - num;
1873 if (0 < num)
1874 *ret = begin + num - 2;
1875 else if (!num)
1876 *ret = begin;
1877 else
1878 *ret = begin + num;
1879 return term;
1881 return spec;
1883 num = strtol(spec, &term, 10);
1884 if (term != spec) {
1885 *ret = num;
1886 return term;
1888 if (spec[0] != '/')
1889 return spec;
1891 /* it could be a regexp of form /.../ */
1892 for (term = (char*) spec + 1; *term && *term != '/'; term++) {
1893 if (*term == '\\')
1894 term++;
1896 if (*term != '/')
1897 return spec;
1899 /* try [spec+1 .. term-1] as regexp */
1900 *term = 0;
1901 begin--; /* input is in human terms */
1902 line = nth_line(sb, begin);
1904 if (!(reg_error = regcomp(&regexp, spec + 1, REG_NEWLINE)) &&
1905 !(reg_error = regexec(&regexp, line, 1, match, 0))) {
1906 const char *cp = line + match[0].rm_so;
1907 const char *nline;
1909 while (begin++ < lno) {
1910 nline = nth_line(sb, begin);
1911 if (line <= cp && cp < nline)
1912 break;
1913 line = nline;
1915 *ret = begin;
1916 regfree(&regexp);
1917 *term++ = '/';
1918 return term;
1920 else {
1921 char errbuf[1024];
1922 regerror(reg_error, &regexp, errbuf, 1024);
1923 die("-L parameter '%s': %s", spec + 1, errbuf);
1928 * Parsing of -L option
1930 static void prepare_blame_range(struct scoreboard *sb,
1931 const char *bottomtop,
1932 long lno,
1933 long *bottom, long *top)
1935 const char *term;
1937 term = parse_loc(bottomtop, sb, lno, 1, bottom);
1938 if (*term == ',') {
1939 term = parse_loc(term + 1, sb, lno, *bottom + 1, top);
1940 if (*term)
1941 usage(blame_usage);
1943 if (*term)
1944 usage(blame_usage);
1947 static int git_blame_config(const char *var, const char *value, void *cb)
1949 if (!strcmp(var, "blame.showroot")) {
1950 show_root = git_config_bool(var, value);
1951 return 0;
1953 if (!strcmp(var, "blame.blankboundary")) {
1954 blank_boundary = git_config_bool(var, value);
1955 return 0;
1957 return git_default_config(var, value, cb);
1961 * Prepare a dummy commit that represents the work tree (or staged) item.
1962 * Note that annotating work tree item never works in the reverse.
1964 static struct commit *fake_working_tree_commit(const char *path, const char *contents_from)
1966 struct commit *commit;
1967 struct origin *origin;
1968 unsigned char head_sha1[20];
1969 struct strbuf buf = STRBUF_INIT;
1970 const char *ident;
1971 time_t now;
1972 int size, len;
1973 struct cache_entry *ce;
1974 unsigned mode;
1976 if (get_sha1("HEAD", head_sha1))
1977 die("No such ref: HEAD");
1979 time(&now);
1980 commit = xcalloc(1, sizeof(*commit));
1981 commit->parents = xcalloc(1, sizeof(*commit->parents));
1982 commit->parents->item = lookup_commit_reference(head_sha1);
1983 commit->object.parsed = 1;
1984 commit->date = now;
1985 commit->object.type = OBJ_COMMIT;
1987 origin = make_origin(commit, path);
1989 if (!contents_from || strcmp("-", contents_from)) {
1990 struct stat st;
1991 const char *read_from;
1993 if (contents_from) {
1994 if (stat(contents_from, &st) < 0)
1995 die("Cannot stat %s", contents_from);
1996 read_from = contents_from;
1998 else {
1999 if (lstat(path, &st) < 0)
2000 die("Cannot lstat %s", path);
2001 read_from = path;
2003 mode = canon_mode(st.st_mode);
2004 switch (st.st_mode & S_IFMT) {
2005 case S_IFREG:
2006 if (strbuf_read_file(&buf, read_from, st.st_size) != st.st_size)
2007 die("cannot open or read %s", read_from);
2008 break;
2009 case S_IFLNK:
2010 if (strbuf_readlink(&buf, read_from, st.st_size) < 0)
2011 die("cannot readlink %s", read_from);
2012 break;
2013 default:
2014 die("unsupported file type %s", read_from);
2017 else {
2018 /* Reading from stdin */
2019 contents_from = "standard input";
2020 mode = 0;
2021 if (strbuf_read(&buf, 0, 0) < 0)
2022 die("read error %s from stdin", strerror(errno));
2024 convert_to_git(path, buf.buf, buf.len, &buf, 0);
2025 origin->file.ptr = buf.buf;
2026 origin->file.size = buf.len;
2027 pretend_sha1_file(buf.buf, buf.len, OBJ_BLOB, origin->blob_sha1);
2028 commit->util = origin;
2031 * Read the current index, replace the path entry with
2032 * origin->blob_sha1 without mucking with its mode or type
2033 * bits; we are not going to write this index out -- we just
2034 * want to run "diff-index --cached".
2036 discard_cache();
2037 read_cache();
2039 len = strlen(path);
2040 if (!mode) {
2041 int pos = cache_name_pos(path, len);
2042 if (0 <= pos)
2043 mode = active_cache[pos]->ce_mode;
2044 else
2045 /* Let's not bother reading from HEAD tree */
2046 mode = S_IFREG | 0644;
2048 size = cache_entry_size(len);
2049 ce = xcalloc(1, size);
2050 hashcpy(ce->sha1, origin->blob_sha1);
2051 memcpy(ce->name, path, len);
2052 ce->ce_flags = create_ce_flags(len, 0);
2053 ce->ce_mode = create_ce_mode(mode);
2054 add_cache_entry(ce, ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE);
2057 * We are not going to write this out, so this does not matter
2058 * right now, but someday we might optimize diff-index --cached
2059 * with cache-tree information.
2061 cache_tree_invalidate_path(active_cache_tree, path);
2063 commit->buffer = xmalloc(400);
2064 ident = fmt_ident("Not Committed Yet", "not.committed.yet", NULL, 0);
2065 snprintf(commit->buffer, 400,
2066 "tree 0000000000000000000000000000000000000000\n"
2067 "parent %s\n"
2068 "author %s\n"
2069 "committer %s\n\n"
2070 "Version of %s from %s\n",
2071 sha1_to_hex(head_sha1),
2072 ident, ident, path, contents_from ? contents_from : path);
2073 return commit;
2076 static const char *prepare_final(struct scoreboard *sb)
2078 int i;
2079 const char *final_commit_name = NULL;
2080 struct rev_info *revs = sb->revs;
2083 * There must be one and only one positive commit in the
2084 * revs->pending array.
2086 for (i = 0; i < revs->pending.nr; i++) {
2087 struct object *obj = revs->pending.objects[i].item;
2088 if (obj->flags & UNINTERESTING)
2089 continue;
2090 while (obj->type == OBJ_TAG)
2091 obj = deref_tag(obj, NULL, 0);
2092 if (obj->type != OBJ_COMMIT)
2093 die("Non commit %s?", revs->pending.objects[i].name);
2094 if (sb->final)
2095 die("More than one commit to dig from %s and %s?",
2096 revs->pending.objects[i].name,
2097 final_commit_name);
2098 sb->final = (struct commit *) obj;
2099 final_commit_name = revs->pending.objects[i].name;
2101 return final_commit_name;
2104 static const char *prepare_initial(struct scoreboard *sb)
2106 int i;
2107 const char *final_commit_name = NULL;
2108 struct rev_info *revs = sb->revs;
2111 * There must be one and only one negative commit, and it must be
2112 * the boundary.
2114 for (i = 0; i < revs->pending.nr; i++) {
2115 struct object *obj = revs->pending.objects[i].item;
2116 if (!(obj->flags & UNINTERESTING))
2117 continue;
2118 while (obj->type == OBJ_TAG)
2119 obj = deref_tag(obj, NULL, 0);
2120 if (obj->type != OBJ_COMMIT)
2121 die("Non commit %s?", revs->pending.objects[i].name);
2122 if (sb->final)
2123 die("More than one commit to dig down to %s and %s?",
2124 revs->pending.objects[i].name,
2125 final_commit_name);
2126 sb->final = (struct commit *) obj;
2127 final_commit_name = revs->pending.objects[i].name;
2129 if (!final_commit_name)
2130 die("No commit to dig down to?");
2131 return final_commit_name;
2134 static int blame_copy_callback(const struct option *option, const char *arg, int unset)
2136 int *opt = option->value;
2139 * -C enables copy from removed files;
2140 * -C -C enables copy from existing files, but only
2141 * when blaming a new file;
2142 * -C -C -C enables copy from existing files for
2143 * everybody
2145 if (*opt & PICKAXE_BLAME_COPY_HARDER)
2146 *opt |= PICKAXE_BLAME_COPY_HARDEST;
2147 if (*opt & PICKAXE_BLAME_COPY)
2148 *opt |= PICKAXE_BLAME_COPY_HARDER;
2149 *opt |= PICKAXE_BLAME_COPY | PICKAXE_BLAME_MOVE;
2151 if (arg)
2152 blame_copy_score = parse_score(arg);
2153 return 0;
2156 static int blame_move_callback(const struct option *option, const char *arg, int unset)
2158 int *opt = option->value;
2160 *opt |= PICKAXE_BLAME_MOVE;
2162 if (arg)
2163 blame_move_score = parse_score(arg);
2164 return 0;
2167 static int blame_bottomtop_callback(const struct option *option, const char *arg, int unset)
2169 const char **bottomtop = option->value;
2170 if (!arg)
2171 return -1;
2172 if (*bottomtop)
2173 die("More than one '-L n,m' option given");
2174 *bottomtop = arg;
2175 return 0;
2178 int cmd_blame(int argc, const char **argv, const char *prefix)
2180 struct rev_info revs;
2181 const char *path;
2182 struct scoreboard sb;
2183 struct origin *o;
2184 struct blame_entry *ent;
2185 long dashdash_pos, bottom, top, lno;
2186 const char *final_commit_name = NULL;
2187 enum object_type type;
2189 static const char *bottomtop = NULL;
2190 static int output_option = 0, opt = 0;
2191 static int show_stats = 0;
2192 static const char *revs_file = NULL;
2193 static const char *contents_from = NULL;
2194 static const struct option options[] = {
2195 OPT_BOOLEAN(0, "incremental", &incremental, "Show blame entries as we find them, incrementally"),
2196 OPT_BOOLEAN('b', NULL, &blank_boundary, "Show blank SHA-1 for boundary commits (Default: off)"),
2197 OPT_BOOLEAN(0, "root", &show_root, "Do not treat root commits as boundaries (Default: off)"),
2198 OPT_BOOLEAN(0, "show-stats", &show_stats, "Show work cost statistics"),
2199 OPT_BIT(0, "score-debug", &output_option, "Show output score for blame entries", OUTPUT_SHOW_SCORE),
2200 OPT_BIT('f', "show-name", &output_option, "Show original filename (Default: auto)", OUTPUT_SHOW_NAME),
2201 OPT_BIT('n', "show-number", &output_option, "Show original linenumber (Default: off)", OUTPUT_SHOW_NUMBER),
2202 OPT_BIT('p', "porcelain", &output_option, "Show in a format designed for machine consumption", OUTPUT_PORCELAIN),
2203 OPT_BIT('c', NULL, &output_option, "Use the same output mode as git-annotate (Default: off)", OUTPUT_ANNOTATE_COMPAT),
2204 OPT_BIT('t', NULL, &output_option, "Show raw timestamp (Default: off)", OUTPUT_RAW_TIMESTAMP),
2205 OPT_BIT('l', NULL, &output_option, "Show long commit SHA1 (Default: off)", OUTPUT_LONG_OBJECT_NAME),
2206 OPT_BIT('s', NULL, &output_option, "Suppress author name and timestamp (Default: off)", OUTPUT_NO_AUTHOR),
2207 OPT_BIT('w', NULL, &xdl_opts, "Ignore whitespace differences", XDF_IGNORE_WHITESPACE),
2208 OPT_STRING('S', NULL, &revs_file, "file", "Use revisions from <file> instead of calling git-rev-list"),
2209 OPT_STRING(0, "contents", &contents_from, "file", "Use <file>'s contents as the final image"),
2210 { OPTION_CALLBACK, 'C', NULL, &opt, "score", "Find line copies within and across files", PARSE_OPT_OPTARG, blame_copy_callback },
2211 { OPTION_CALLBACK, 'M', NULL, &opt, "score", "Find line movements within and across files", PARSE_OPT_OPTARG, blame_move_callback },
2212 OPT_CALLBACK('L', NULL, &bottomtop, "n,m", "Process only line range n,m, counting from 1", blame_bottomtop_callback),
2213 OPT_END()
2216 struct parse_opt_ctx_t ctx;
2217 int cmd_is_annotate = !strcmp(argv[0], "annotate");
2219 git_config(git_blame_config, NULL);
2220 init_revisions(&revs, NULL);
2221 save_commit_buffer = 0;
2222 dashdash_pos = 0;
2224 parse_options_start(&ctx, argc, argv, PARSE_OPT_KEEP_DASHDASH |
2225 PARSE_OPT_KEEP_ARGV0);
2226 for (;;) {
2227 switch (parse_options_step(&ctx, options, blame_opt_usage)) {
2228 case PARSE_OPT_HELP:
2229 exit(129);
2230 case PARSE_OPT_DONE:
2231 if (ctx.argv[0])
2232 dashdash_pos = ctx.cpidx;
2233 goto parse_done;
2236 if (!strcmp(ctx.argv[0], "--reverse")) {
2237 ctx.argv[0] = "--children";
2238 reverse = 1;
2240 parse_revision_opt(&revs, &ctx, options, blame_opt_usage);
2242 parse_done:
2243 argc = parse_options_end(&ctx);
2245 if (cmd_is_annotate)
2246 output_option |= OUTPUT_ANNOTATE_COMPAT;
2248 if (DIFF_OPT_TST(&revs.diffopt, FIND_COPIES_HARDER))
2249 opt |= (PICKAXE_BLAME_COPY | PICKAXE_BLAME_MOVE |
2250 PICKAXE_BLAME_COPY_HARDER);
2252 if (!blame_move_score)
2253 blame_move_score = BLAME_DEFAULT_MOVE_SCORE;
2254 if (!blame_copy_score)
2255 blame_copy_score = BLAME_DEFAULT_COPY_SCORE;
2258 * We have collected options unknown to us in argv[1..unk]
2259 * which are to be passed to revision machinery if we are
2260 * going to do the "bottom" processing.
2262 * The remaining are:
2264 * (1) if dashdash_pos != 0, its either
2265 * "blame [revisions] -- <path>" or
2266 * "blame -- <path> <rev>"
2268 * (2) otherwise, its one of the two:
2269 * "blame [revisions] <path>"
2270 * "blame <path> <rev>"
2272 * Note that we must strip out <path> from the arguments: we do not
2273 * want the path pruning but we may want "bottom" processing.
2275 if (dashdash_pos) {
2276 switch (argc - dashdash_pos - 1) {
2277 case 2: /* (1b) */
2278 if (argc != 4)
2279 usage_with_options(blame_opt_usage, options);
2280 /* reorder for the new way: <rev> -- <path> */
2281 argv[1] = argv[3];
2282 argv[3] = argv[2];
2283 argv[2] = "--";
2284 /* FALLTHROUGH */
2285 case 1: /* (1a) */
2286 path = add_prefix(prefix, argv[--argc]);
2287 argv[argc] = NULL;
2288 break;
2289 default:
2290 usage_with_options(blame_opt_usage, options);
2292 } else {
2293 if (argc < 2)
2294 usage_with_options(blame_opt_usage, options);
2295 path = add_prefix(prefix, argv[argc - 1]);
2296 if (argc == 3 && !has_string_in_work_tree(path)) { /* (2b) */
2297 path = add_prefix(prefix, argv[1]);
2298 argv[1] = argv[2];
2300 argv[argc - 1] = "--";
2302 setup_work_tree();
2303 if (!has_string_in_work_tree(path))
2304 die("cannot stat path %s: %s", path, strerror(errno));
2307 setup_revisions(argc, argv, &revs, NULL);
2308 memset(&sb, 0, sizeof(sb));
2310 sb.revs = &revs;
2311 if (!reverse)
2312 final_commit_name = prepare_final(&sb);
2313 else if (contents_from)
2314 die("--contents and --children do not blend well.");
2315 else
2316 final_commit_name = prepare_initial(&sb);
2318 if (!sb.final) {
2320 * "--not A B -- path" without anything positive;
2321 * do not default to HEAD, but use the working tree
2322 * or "--contents".
2324 setup_work_tree();
2325 sb.final = fake_working_tree_commit(path, contents_from);
2326 add_pending_object(&revs, &(sb.final->object), ":");
2328 else if (contents_from)
2329 die("Cannot use --contents with final commit object name");
2332 * If we have bottom, this will mark the ancestors of the
2333 * bottom commits we would reach while traversing as
2334 * uninteresting.
2336 if (prepare_revision_walk(&revs))
2337 die("revision walk setup failed");
2339 if (is_null_sha1(sb.final->object.sha1)) {
2340 char *buf;
2341 o = sb.final->util;
2342 buf = xmalloc(o->file.size + 1);
2343 memcpy(buf, o->file.ptr, o->file.size + 1);
2344 sb.final_buf = buf;
2345 sb.final_buf_size = o->file.size;
2347 else {
2348 o = get_origin(&sb, sb.final, path);
2349 if (fill_blob_sha1(o))
2350 die("no such path %s in %s", path, final_commit_name);
2352 sb.final_buf = read_sha1_file(o->blob_sha1, &type,
2353 &sb.final_buf_size);
2354 if (!sb.final_buf)
2355 die("Cannot read blob %s for path %s",
2356 sha1_to_hex(o->blob_sha1),
2357 path);
2359 num_read_blob++;
2360 lno = prepare_lines(&sb);
2362 bottom = top = 0;
2363 if (bottomtop)
2364 prepare_blame_range(&sb, bottomtop, lno, &bottom, &top);
2365 if (bottom && top && top < bottom) {
2366 long tmp;
2367 tmp = top; top = bottom; bottom = tmp;
2369 if (bottom < 1)
2370 bottom = 1;
2371 if (top < 1)
2372 top = lno;
2373 bottom--;
2374 if (lno < top)
2375 die("file %s has only %lu lines", path, lno);
2377 ent = xcalloc(1, sizeof(*ent));
2378 ent->lno = bottom;
2379 ent->num_lines = top - bottom;
2380 ent->suspect = o;
2381 ent->s_lno = bottom;
2383 sb.ent = ent;
2384 sb.path = path;
2386 if (revs_file && read_ancestry(revs_file))
2387 die("reading graft file %s failed: %s",
2388 revs_file, strerror(errno));
2390 read_mailmap(&mailmap, NULL);
2392 if (!incremental)
2393 setup_pager();
2395 assign_blame(&sb, opt);
2397 if (incremental)
2398 return 0;
2400 coalesce(&sb);
2402 if (!(output_option & OUTPUT_PORCELAIN))
2403 find_alignment(&sb, &output_option);
2405 output(&sb, output_option);
2406 free((void *)sb.final_buf);
2407 for (ent = sb.ent; ent; ) {
2408 struct blame_entry *e = ent->next;
2409 free(ent);
2410 ent = e;
2413 if (show_stats) {
2414 printf("num read blob: %d\n", num_read_blob);
2415 printf("num get patch: %d\n", num_get_patch);
2416 printf("num commits: %d\n", num_commits);
2418 return 0;