Merge branch 'pb/maint-gitweb-blob-lineno' into maint
[git] / builtin-blame.c
1 /*
2  * Blame
3  *
4  * Copyright (c) 2006, Junio C Hamano
5  */
6
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"
23
24 static char blame_usage[] = "git blame [options] [rev-opts] [rev] [--] file";
25
26 static const char *blame_opt_usage[] = {
27         blame_usage,
28         "",
29         "[rev-opts] are documented in git-rev-list(1)",
30         NULL
31 };
32
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
44 static enum date_mode blame_date_mode = DATE_ISO8601;
45 static size_t blame_date_width;
46
47 static struct string_list mailmap;
48
49 #ifndef DEBUG
50 #define DEBUG 0
51 #endif
52
53 /* stats */
54 static int num_read_blob;
55 static int num_get_patch;
56 static int num_commits;
57
58 #define PICKAXE_BLAME_MOVE              01
59 #define PICKAXE_BLAME_COPY              02
60 #define PICKAXE_BLAME_COPY_HARDER       04
61 #define PICKAXE_BLAME_COPY_HARDEST      010
62
63 /*
64  * blame for a blame_entry with score lower than these thresholds
65  * is not passed to the parent using move/copy logic.
66  */
67 static unsigned blame_move_score;
68 static unsigned blame_copy_score;
69 #define BLAME_DEFAULT_MOVE_SCORE        20
70 #define BLAME_DEFAULT_COPY_SCORE        40
71
72 /* bits #0..7 in revision.h, #8..11 used for merge_bases() in commit.c */
73 #define METAINFO_SHOWN          (1u<<12)
74 #define MORE_THAN_ONE_PATH      (1u<<13)
75
76 /*
77  * One blob in a commit that is being suspected
78  */
79 struct origin {
80         int refcnt;
81         struct origin *previous;
82         struct commit *commit;
83         mmfile_t file;
84         unsigned char blob_sha1[20];
85         char path[FLEX_ARRAY];
86 };
87
88 /*
89  * Given an origin, prepare mmfile_t structure to be used by the
90  * diff machinery
91  */
92 static void fill_origin_blob(struct origin *o, mmfile_t *file)
93 {
94         if (!o->file.ptr) {
95                 enum object_type type;
96                 num_read_blob++;
97                 file->ptr = read_sha1_file(o->blob_sha1, &type,
98                                            (unsigned long *)(&(file->size)));
99                 if (!file->ptr)
100                         die("Cannot read blob %s for path %s",
101                             sha1_to_hex(o->blob_sha1),
102                             o->path);
103                 o->file = *file;
104         }
105         else
106                 *file = o->file;
107 }
108
109 /*
110  * Origin is refcounted and usually we keep the blob contents to be
111  * reused.
112  */
113 static inline struct origin *origin_incref(struct origin *o)
114 {
115         if (o)
116                 o->refcnt++;
117         return o;
118 }
119
120 static void origin_decref(struct origin *o)
121 {
122         if (o && --o->refcnt <= 0) {
123                 if (o->previous)
124                         origin_decref(o->previous);
125                 free(o->file.ptr);
126                 free(o);
127         }
128 }
129
130 static void drop_origin_blob(struct origin *o)
131 {
132         if (o->file.ptr) {
133                 free(o->file.ptr);
134                 o->file.ptr = NULL;
135         }
136 }
137
138 /*
139  * Each group of lines is described by a blame_entry; it can be split
140  * as we pass blame to the parents.  They form a linked list in the
141  * scoreboard structure, sorted by the target line number.
142  */
143 struct blame_entry {
144         struct blame_entry *prev;
145         struct blame_entry *next;
146
147         /* the first line of this group in the final image;
148          * internally all line numbers are 0 based.
149          */
150         int lno;
151
152         /* how many lines this group has */
153         int num_lines;
154
155         /* the commit that introduced this group into the final image */
156         struct origin *suspect;
157
158         /* true if the suspect is truly guilty; false while we have not
159          * checked if the group came from one of its parents.
160          */
161         char guilty;
162
163         /* true if the entry has been scanned for copies in the current parent
164          */
165         char scanned;
166
167         /* the line number of the first line of this group in the
168          * suspect's file; internally all line numbers are 0 based.
169          */
170         int s_lno;
171
172         /* how significant this entry is -- cached to avoid
173          * scanning the lines over and over.
174          */
175         unsigned score;
176 };
177
178 /*
179  * The current state of the blame assignment.
180  */
181 struct scoreboard {
182         /* the final commit (i.e. where we started digging from) */
183         struct commit *final;
184         struct rev_info *revs;
185         const char *path;
186
187         /*
188          * The contents in the final image.
189          * Used by many functions to obtain contents of the nth line,
190          * indexed with scoreboard.lineno[blame_entry.lno].
191          */
192         const char *final_buf;
193         unsigned long final_buf_size;
194
195         /* linked list of blames */
196         struct blame_entry *ent;
197
198         /* look-up a line in the final buffer */
199         int num_lines;
200         int *lineno;
201 };
202
203 static inline int same_suspect(struct origin *a, struct origin *b)
204 {
205         if (a == b)
206                 return 1;
207         if (a->commit != b->commit)
208                 return 0;
209         return !strcmp(a->path, b->path);
210 }
211
212 static void sanity_check_refcnt(struct scoreboard *);
213
214 /*
215  * If two blame entries that are next to each other came from
216  * contiguous lines in the same origin (i.e. <commit, path> pair),
217  * merge them together.
218  */
219 static void coalesce(struct scoreboard *sb)
220 {
221         struct blame_entry *ent, *next;
222
223         for (ent = sb->ent; ent && (next = ent->next); ent = next) {
224                 if (same_suspect(ent->suspect, next->suspect) &&
225                     ent->guilty == next->guilty &&
226                     ent->s_lno + ent->num_lines == next->s_lno) {
227                         ent->num_lines += next->num_lines;
228                         ent->next = next->next;
229                         if (ent->next)
230                                 ent->next->prev = ent;
231                         origin_decref(next->suspect);
232                         free(next);
233                         ent->score = 0;
234                         next = ent; /* again */
235                 }
236         }
237
238         if (DEBUG) /* sanity */
239                 sanity_check_refcnt(sb);
240 }
241
242 /*
243  * Given a commit and a path in it, create a new origin structure.
244  * The callers that add blame to the scoreboard should use
245  * get_origin() to obtain shared, refcounted copy instead of calling
246  * this function directly.
247  */
248 static struct origin *make_origin(struct commit *commit, const char *path)
249 {
250         struct origin *o;
251         o = xcalloc(1, sizeof(*o) + strlen(path) + 1);
252         o->commit = commit;
253         o->refcnt = 1;
254         strcpy(o->path, path);
255         return o;
256 }
257
258 /*
259  * Locate an existing origin or create a new one.
260  */
261 static struct origin *get_origin(struct scoreboard *sb,
262                                  struct commit *commit,
263                                  const char *path)
264 {
265         struct blame_entry *e;
266
267         for (e = sb->ent; e; e = e->next) {
268                 if (e->suspect->commit == commit &&
269                     !strcmp(e->suspect->path, path))
270                         return origin_incref(e->suspect);
271         }
272         return make_origin(commit, path);
273 }
274
275 /*
276  * Fill the blob_sha1 field of an origin if it hasn't, so that later
277  * call to fill_origin_blob() can use it to locate the data.  blob_sha1
278  * for an origin is also used to pass the blame for the entire file to
279  * the parent to detect the case where a child's blob is identical to
280  * that of its parent's.
281  */
282 static int fill_blob_sha1(struct origin *origin)
283 {
284         unsigned mode;
285
286         if (!is_null_sha1(origin->blob_sha1))
287                 return 0;
288         if (get_tree_entry(origin->commit->object.sha1,
289                            origin->path,
290                            origin->blob_sha1, &mode))
291                 goto error_out;
292         if (sha1_object_info(origin->blob_sha1, NULL) != OBJ_BLOB)
293                 goto error_out;
294         return 0;
295  error_out:
296         hashclr(origin->blob_sha1);
297         return -1;
298 }
299
300 /*
301  * We have an origin -- check if the same path exists in the
302  * parent and return an origin structure to represent it.
303  */
304 static struct origin *find_origin(struct scoreboard *sb,
305                                   struct commit *parent,
306                                   struct origin *origin)
307 {
308         struct origin *porigin = NULL;
309         struct diff_options diff_opts;
310         const char *paths[2];
311
312         if (parent->util) {
313                 /*
314                  * Each commit object can cache one origin in that
315                  * commit.  This is a freestanding copy of origin and
316                  * not refcounted.
317                  */
318                 struct origin *cached = parent->util;
319                 if (!strcmp(cached->path, origin->path)) {
320                         /*
321                          * The same path between origin and its parent
322                          * without renaming -- the most common case.
323                          */
324                         porigin = get_origin(sb, parent, cached->path);
325
326                         /*
327                          * If the origin was newly created (i.e. get_origin
328                          * would call make_origin if none is found in the
329                          * scoreboard), it does not know the blob_sha1,
330                          * so copy it.  Otherwise porigin was in the
331                          * scoreboard and already knows blob_sha1.
332                          */
333                         if (porigin->refcnt == 1)
334                                 hashcpy(porigin->blob_sha1, cached->blob_sha1);
335                         return porigin;
336                 }
337                 /* otherwise it was not very useful; free it */
338                 free(parent->util);
339                 parent->util = NULL;
340         }
341
342         /* See if the origin->path is different between parent
343          * and origin first.  Most of the time they are the
344          * same and diff-tree is fairly efficient about this.
345          */
346         diff_setup(&diff_opts);
347         DIFF_OPT_SET(&diff_opts, RECURSIVE);
348         diff_opts.detect_rename = 0;
349         diff_opts.output_format = DIFF_FORMAT_NO_OUTPUT;
350         paths[0] = origin->path;
351         paths[1] = NULL;
352
353         diff_tree_setup_paths(paths, &diff_opts);
354         if (diff_setup_done(&diff_opts) < 0)
355                 die("diff-setup");
356
357         if (is_null_sha1(origin->commit->object.sha1))
358                 do_diff_cache(parent->tree->object.sha1, &diff_opts);
359         else
360                 diff_tree_sha1(parent->tree->object.sha1,
361                                origin->commit->tree->object.sha1,
362                                "", &diff_opts);
363         diffcore_std(&diff_opts);
364
365         if (!diff_queued_diff.nr) {
366                 /* The path is the same as parent */
367                 porigin = get_origin(sb, parent, origin->path);
368                 hashcpy(porigin->blob_sha1, origin->blob_sha1);
369         } else {
370                 /*
371                  * Since origin->path is a pathspec, if the parent
372                  * commit had it as a directory, we will see a whole
373                  * bunch of deletion of files in the directory that we
374                  * do not care about.
375                  */
376                 int i;
377                 struct diff_filepair *p = NULL;
378                 for (i = 0; i < diff_queued_diff.nr; i++) {
379                         const char *name;
380                         p = diff_queued_diff.queue[i];
381                         name = p->one->path ? p->one->path : p->two->path;
382                         if (!strcmp(name, origin->path))
383                                 break;
384                 }
385                 if (!p)
386                         die("internal error in blame::find_origin");
387                 switch (p->status) {
388                 default:
389                         die("internal error in blame::find_origin (%c)",
390                             p->status);
391                 case 'M':
392                         porigin = get_origin(sb, parent, origin->path);
393                         hashcpy(porigin->blob_sha1, p->one->sha1);
394                         break;
395                 case 'A':
396                 case 'T':
397                         /* Did not exist in parent, or type changed */
398                         break;
399                 }
400         }
401         diff_flush(&diff_opts);
402         diff_tree_release_paths(&diff_opts);
403         if (porigin) {
404                 /*
405                  * Create a freestanding copy that is not part of
406                  * the refcounted origin found in the scoreboard, and
407                  * cache it in the commit.
408                  */
409                 struct origin *cached;
410
411                 cached = make_origin(porigin->commit, porigin->path);
412                 hashcpy(cached->blob_sha1, porigin->blob_sha1);
413                 parent->util = cached;
414         }
415         return porigin;
416 }
417
418 /*
419  * We have an origin -- find the path that corresponds to it in its
420  * parent and return an origin structure to represent it.
421  */
422 static struct origin *find_rename(struct scoreboard *sb,
423                                   struct commit *parent,
424                                   struct origin *origin)
425 {
426         struct origin *porigin = NULL;
427         struct diff_options diff_opts;
428         int i;
429         const char *paths[2];
430
431         diff_setup(&diff_opts);
432         DIFF_OPT_SET(&diff_opts, RECURSIVE);
433         diff_opts.detect_rename = DIFF_DETECT_RENAME;
434         diff_opts.output_format = DIFF_FORMAT_NO_OUTPUT;
435         diff_opts.single_follow = origin->path;
436         paths[0] = NULL;
437         diff_tree_setup_paths(paths, &diff_opts);
438         if (diff_setup_done(&diff_opts) < 0)
439                 die("diff-setup");
440
441         if (is_null_sha1(origin->commit->object.sha1))
442                 do_diff_cache(parent->tree->object.sha1, &diff_opts);
443         else
444                 diff_tree_sha1(parent->tree->object.sha1,
445                                origin->commit->tree->object.sha1,
446                                "", &diff_opts);
447         diffcore_std(&diff_opts);
448
449         for (i = 0; i < diff_queued_diff.nr; i++) {
450                 struct diff_filepair *p = diff_queued_diff.queue[i];
451                 if ((p->status == 'R' || p->status == 'C') &&
452                     !strcmp(p->two->path, origin->path)) {
453                         porigin = get_origin(sb, parent, p->one->path);
454                         hashcpy(porigin->blob_sha1, p->one->sha1);
455                         break;
456                 }
457         }
458         diff_flush(&diff_opts);
459         diff_tree_release_paths(&diff_opts);
460         return porigin;
461 }
462
463 /*
464  * Link in a new blame entry to the scoreboard.  Entries that cover the
465  * same line range have been removed from the scoreboard previously.
466  */
467 static void add_blame_entry(struct scoreboard *sb, struct blame_entry *e)
468 {
469         struct blame_entry *ent, *prev = NULL;
470
471         origin_incref(e->suspect);
472
473         for (ent = sb->ent; ent && ent->lno < e->lno; ent = ent->next)
474                 prev = ent;
475
476         /* prev, if not NULL, is the last one that is below e */
477         e->prev = prev;
478         if (prev) {
479                 e->next = prev->next;
480                 prev->next = e;
481         }
482         else {
483                 e->next = sb->ent;
484                 sb->ent = e;
485         }
486         if (e->next)
487                 e->next->prev = e;
488 }
489
490 /*
491  * src typically is on-stack; we want to copy the information in it to
492  * a malloced blame_entry that is already on the linked list of the
493  * scoreboard.  The origin of dst loses a refcnt while the origin of src
494  * gains one.
495  */
496 static void dup_entry(struct blame_entry *dst, struct blame_entry *src)
497 {
498         struct blame_entry *p, *n;
499
500         p = dst->prev;
501         n = dst->next;
502         origin_incref(src->suspect);
503         origin_decref(dst->suspect);
504         memcpy(dst, src, sizeof(*src));
505         dst->prev = p;
506         dst->next = n;
507         dst->score = 0;
508 }
509
510 static const char *nth_line(struct scoreboard *sb, int lno)
511 {
512         return sb->final_buf + sb->lineno[lno];
513 }
514
515 /*
516  * It is known that lines between tlno to same came from parent, and e
517  * has an overlap with that range.  it also is known that parent's
518  * line plno corresponds to e's line tlno.
519  *
520  *                <---- e ----->
521  *                   <------>
522  *                   <------------>
523  *             <------------>
524  *             <------------------>
525  *
526  * Split e into potentially three parts; before this chunk, the chunk
527  * to be blamed for the parent, and after that portion.
528  */
529 static void split_overlap(struct blame_entry *split,
530                           struct blame_entry *e,
531                           int tlno, int plno, int same,
532                           struct origin *parent)
533 {
534         int chunk_end_lno;
535         memset(split, 0, sizeof(struct blame_entry [3]));
536
537         if (e->s_lno < tlno) {
538                 /* there is a pre-chunk part not blamed on parent */
539                 split[0].suspect = origin_incref(e->suspect);
540                 split[0].lno = e->lno;
541                 split[0].s_lno = e->s_lno;
542                 split[0].num_lines = tlno - e->s_lno;
543                 split[1].lno = e->lno + tlno - e->s_lno;
544                 split[1].s_lno = plno;
545         }
546         else {
547                 split[1].lno = e->lno;
548                 split[1].s_lno = plno + (e->s_lno - tlno);
549         }
550
551         if (same < e->s_lno + e->num_lines) {
552                 /* there is a post-chunk part not blamed on parent */
553                 split[2].suspect = origin_incref(e->suspect);
554                 split[2].lno = e->lno + (same - e->s_lno);
555                 split[2].s_lno = e->s_lno + (same - e->s_lno);
556                 split[2].num_lines = e->s_lno + e->num_lines - same;
557                 chunk_end_lno = split[2].lno;
558         }
559         else
560                 chunk_end_lno = e->lno + e->num_lines;
561         split[1].num_lines = chunk_end_lno - split[1].lno;
562
563         /*
564          * if it turns out there is nothing to blame the parent for,
565          * forget about the splitting.  !split[1].suspect signals this.
566          */
567         if (split[1].num_lines < 1)
568                 return;
569         split[1].suspect = origin_incref(parent);
570 }
571
572 /*
573  * split_overlap() divided an existing blame e into up to three parts
574  * in split.  Adjust the linked list of blames in the scoreboard to
575  * reflect the split.
576  */
577 static void split_blame(struct scoreboard *sb,
578                         struct blame_entry *split,
579                         struct blame_entry *e)
580 {
581         struct blame_entry *new_entry;
582
583         if (split[0].suspect && split[2].suspect) {
584                 /* The first part (reuse storage for the existing entry e) */
585                 dup_entry(e, &split[0]);
586
587                 /* The last part -- me */
588                 new_entry = xmalloc(sizeof(*new_entry));
589                 memcpy(new_entry, &(split[2]), sizeof(struct blame_entry));
590                 add_blame_entry(sb, new_entry);
591
592                 /* ... and the middle part -- parent */
593                 new_entry = xmalloc(sizeof(*new_entry));
594                 memcpy(new_entry, &(split[1]), sizeof(struct blame_entry));
595                 add_blame_entry(sb, new_entry);
596         }
597         else if (!split[0].suspect && !split[2].suspect)
598                 /*
599                  * The parent covers the entire area; reuse storage for
600                  * e and replace it with the parent.
601                  */
602                 dup_entry(e, &split[1]);
603         else if (split[0].suspect) {
604                 /* me and then parent */
605                 dup_entry(e, &split[0]);
606
607                 new_entry = xmalloc(sizeof(*new_entry));
608                 memcpy(new_entry, &(split[1]), sizeof(struct blame_entry));
609                 add_blame_entry(sb, new_entry);
610         }
611         else {
612                 /* parent and then me */
613                 dup_entry(e, &split[1]);
614
615                 new_entry = xmalloc(sizeof(*new_entry));
616                 memcpy(new_entry, &(split[2]), sizeof(struct blame_entry));
617                 add_blame_entry(sb, new_entry);
618         }
619
620         if (DEBUG) { /* sanity */
621                 struct blame_entry *ent;
622                 int lno = sb->ent->lno, corrupt = 0;
623
624                 for (ent = sb->ent; ent; ent = ent->next) {
625                         if (lno != ent->lno)
626                                 corrupt = 1;
627                         if (ent->s_lno < 0)
628                                 corrupt = 1;
629                         lno += ent->num_lines;
630                 }
631                 if (corrupt) {
632                         lno = sb->ent->lno;
633                         for (ent = sb->ent; ent; ent = ent->next) {
634                                 printf("L %8d l %8d n %8d\n",
635                                        lno, ent->lno, ent->num_lines);
636                                 lno = ent->lno + ent->num_lines;
637                         }
638                         die("oops");
639                 }
640         }
641 }
642
643 /*
644  * After splitting the blame, the origins used by the
645  * on-stack blame_entry should lose one refcnt each.
646  */
647 static void decref_split(struct blame_entry *split)
648 {
649         int i;
650
651         for (i = 0; i < 3; i++)
652                 origin_decref(split[i].suspect);
653 }
654
655 /*
656  * Helper for blame_chunk().  blame_entry e is known to overlap with
657  * the patch hunk; split it and pass blame to the parent.
658  */
659 static void blame_overlap(struct scoreboard *sb, struct blame_entry *e,
660                           int tlno, int plno, int same,
661                           struct origin *parent)
662 {
663         struct blame_entry split[3];
664
665         split_overlap(split, e, tlno, plno, same, parent);
666         if (split[1].suspect)
667                 split_blame(sb, split, e);
668         decref_split(split);
669 }
670
671 /*
672  * Find the line number of the last line the target is suspected for.
673  */
674 static int find_last_in_target(struct scoreboard *sb, struct origin *target)
675 {
676         struct blame_entry *e;
677         int last_in_target = -1;
678
679         for (e = sb->ent; e; e = e->next) {
680                 if (e->guilty || !same_suspect(e->suspect, target))
681                         continue;
682                 if (last_in_target < e->s_lno + e->num_lines)
683                         last_in_target = e->s_lno + e->num_lines;
684         }
685         return last_in_target;
686 }
687
688 /*
689  * Process one hunk from the patch between the current suspect for
690  * blame_entry e and its parent.  Find and split the overlap, and
691  * pass blame to the overlapping part to the parent.
692  */
693 static void blame_chunk(struct scoreboard *sb,
694                         int tlno, int plno, int same,
695                         struct origin *target, struct origin *parent)
696 {
697         struct blame_entry *e;
698
699         for (e = sb->ent; e; e = e->next) {
700                 if (e->guilty || !same_suspect(e->suspect, target))
701                         continue;
702                 if (same <= e->s_lno)
703                         continue;
704                 if (tlno < e->s_lno + e->num_lines)
705                         blame_overlap(sb, e, tlno, plno, same, parent);
706         }
707 }
708
709 struct blame_chunk_cb_data {
710         struct scoreboard *sb;
711         struct origin *target;
712         struct origin *parent;
713         long plno;
714         long tlno;
715 };
716
717 static void blame_chunk_cb(void *data, long same, long p_next, long t_next)
718 {
719         struct blame_chunk_cb_data *d = data;
720         blame_chunk(d->sb, d->tlno, d->plno, same, d->target, d->parent);
721         d->plno = p_next;
722         d->tlno = t_next;
723 }
724
725 /*
726  * We are looking at the origin 'target' and aiming to pass blame
727  * for the lines it is suspected to its parent.  Run diff to find
728  * which lines came from parent and pass blame for them.
729  */
730 static int pass_blame_to_parent(struct scoreboard *sb,
731                                 struct origin *target,
732                                 struct origin *parent)
733 {
734         int last_in_target;
735         mmfile_t file_p, file_o;
736         struct blame_chunk_cb_data d = { sb, target, parent, 0, 0 };
737         xpparam_t xpp;
738         xdemitconf_t xecfg;
739
740         last_in_target = find_last_in_target(sb, target);
741         if (last_in_target < 0)
742                 return 1; /* nothing remains for this target */
743
744         fill_origin_blob(parent, &file_p);
745         fill_origin_blob(target, &file_o);
746         num_get_patch++;
747
748         memset(&xpp, 0, sizeof(xpp));
749         xpp.flags = xdl_opts;
750         memset(&xecfg, 0, sizeof(xecfg));
751         xecfg.ctxlen = 0;
752         xdi_diff_hunks(&file_p, &file_o, blame_chunk_cb, &d, &xpp, &xecfg);
753         /* The rest (i.e. anything after tlno) are the same as the parent */
754         blame_chunk(sb, d.tlno, d.plno, last_in_target, target, parent);
755
756         return 0;
757 }
758
759 /*
760  * The lines in blame_entry after splitting blames many times can become
761  * very small and trivial, and at some point it becomes pointless to
762  * blame the parents.  E.g. "\t\t}\n\t}\n\n" appears everywhere in any
763  * ordinary C program, and it is not worth to say it was copied from
764  * totally unrelated file in the parent.
765  *
766  * Compute how trivial the lines in the blame_entry are.
767  */
768 static unsigned ent_score(struct scoreboard *sb, struct blame_entry *e)
769 {
770         unsigned score;
771         const char *cp, *ep;
772
773         if (e->score)
774                 return e->score;
775
776         score = 1;
777         cp = nth_line(sb, e->lno);
778         ep = nth_line(sb, e->lno + e->num_lines);
779         while (cp < ep) {
780                 unsigned ch = *((unsigned char *)cp);
781                 if (isalnum(ch))
782                         score++;
783                 cp++;
784         }
785         e->score = score;
786         return score;
787 }
788
789 /*
790  * best_so_far[] and this[] are both a split of an existing blame_entry
791  * that passes blame to the parent.  Maintain best_so_far the best split
792  * so far, by comparing this and best_so_far and copying this into
793  * bst_so_far as needed.
794  */
795 static void copy_split_if_better(struct scoreboard *sb,
796                                  struct blame_entry *best_so_far,
797                                  struct blame_entry *this)
798 {
799         int i;
800
801         if (!this[1].suspect)
802                 return;
803         if (best_so_far[1].suspect) {
804                 if (ent_score(sb, &this[1]) < ent_score(sb, &best_so_far[1]))
805                         return;
806         }
807
808         for (i = 0; i < 3; i++)
809                 origin_incref(this[i].suspect);
810         decref_split(best_so_far);
811         memcpy(best_so_far, this, sizeof(struct blame_entry [3]));
812 }
813
814 /*
815  * We are looking at a part of the final image represented by
816  * ent (tlno and same are offset by ent->s_lno).
817  * tlno is where we are looking at in the final image.
818  * up to (but not including) same match preimage.
819  * plno is where we are looking at in the preimage.
820  *
821  * <-------------- final image ---------------------->
822  *       <------ent------>
823  *         ^tlno ^same
824  *    <---------preimage----->
825  *         ^plno
826  *
827  * All line numbers are 0-based.
828  */
829 static void handle_split(struct scoreboard *sb,
830                          struct blame_entry *ent,
831                          int tlno, int plno, int same,
832                          struct origin *parent,
833                          struct blame_entry *split)
834 {
835         if (ent->num_lines <= tlno)
836                 return;
837         if (tlno < same) {
838                 struct blame_entry this[3];
839                 tlno += ent->s_lno;
840                 same += ent->s_lno;
841                 split_overlap(this, ent, tlno, plno, same, parent);
842                 copy_split_if_better(sb, split, this);
843                 decref_split(this);
844         }
845 }
846
847 struct handle_split_cb_data {
848         struct scoreboard *sb;
849         struct blame_entry *ent;
850         struct origin *parent;
851         struct blame_entry *split;
852         long plno;
853         long tlno;
854 };
855
856 static void handle_split_cb(void *data, long same, long p_next, long t_next)
857 {
858         struct handle_split_cb_data *d = data;
859         handle_split(d->sb, d->ent, d->tlno, d->plno, same, d->parent, d->split);
860         d->plno = p_next;
861         d->tlno = t_next;
862 }
863
864 /*
865  * Find the lines from parent that are the same as ent so that
866  * we can pass blames to it.  file_p has the blob contents for
867  * the parent.
868  */
869 static void find_copy_in_blob(struct scoreboard *sb,
870                               struct blame_entry *ent,
871                               struct origin *parent,
872                               struct blame_entry *split,
873                               mmfile_t *file_p)
874 {
875         const char *cp;
876         int cnt;
877         mmfile_t file_o;
878         struct handle_split_cb_data d = { sb, ent, parent, split, 0, 0 };
879         xpparam_t xpp;
880         xdemitconf_t xecfg;
881
882         /*
883          * Prepare mmfile that contains only the lines in ent.
884          */
885         cp = nth_line(sb, ent->lno);
886         file_o.ptr = (char *) cp;
887         cnt = ent->num_lines;
888
889         while (cnt && cp < sb->final_buf + sb->final_buf_size) {
890                 if (*cp++ == '\n')
891                         cnt--;
892         }
893         file_o.size = cp - file_o.ptr;
894
895         /*
896          * file_o is a part of final image we are annotating.
897          * file_p partially may match that image.
898          */
899         memset(&xpp, 0, sizeof(xpp));
900         xpp.flags = xdl_opts;
901         memset(&xecfg, 0, sizeof(xecfg));
902         xecfg.ctxlen = 1;
903         memset(split, 0, sizeof(struct blame_entry [3]));
904         xdi_diff_hunks(file_p, &file_o, handle_split_cb, &d, &xpp, &xecfg);
905         /* remainder, if any, all match the preimage */
906         handle_split(sb, ent, d.tlno, d.plno, ent->num_lines, parent, split);
907 }
908
909 /*
910  * See if lines currently target is suspected for can be attributed to
911  * parent.
912  */
913 static int find_move_in_parent(struct scoreboard *sb,
914                                struct origin *target,
915                                struct origin *parent)
916 {
917         int last_in_target, made_progress;
918         struct blame_entry *e, split[3];
919         mmfile_t file_p;
920
921         last_in_target = find_last_in_target(sb, target);
922         if (last_in_target < 0)
923                 return 1; /* nothing remains for this target */
924
925         fill_origin_blob(parent, &file_p);
926         if (!file_p.ptr)
927                 return 0;
928
929         made_progress = 1;
930         while (made_progress) {
931                 made_progress = 0;
932                 for (e = sb->ent; e; e = e->next) {
933                         if (e->guilty || !same_suspect(e->suspect, target) ||
934                             ent_score(sb, e) < blame_move_score)
935                                 continue;
936                         find_copy_in_blob(sb, e, parent, split, &file_p);
937                         if (split[1].suspect &&
938                             blame_move_score < ent_score(sb, &split[1])) {
939                                 split_blame(sb, split, e);
940                                 made_progress = 1;
941                         }
942                         decref_split(split);
943                 }
944         }
945         return 0;
946 }
947
948 struct blame_list {
949         struct blame_entry *ent;
950         struct blame_entry split[3];
951 };
952
953 /*
954  * Count the number of entries the target is suspected for,
955  * and prepare a list of entry and the best split.
956  */
957 static struct blame_list *setup_blame_list(struct scoreboard *sb,
958                                            struct origin *target,
959                                            int min_score,
960                                            int *num_ents_p)
961 {
962         struct blame_entry *e;
963         int num_ents, i;
964         struct blame_list *blame_list = NULL;
965
966         for (e = sb->ent, num_ents = 0; e; e = e->next)
967                 if (!e->scanned && !e->guilty &&
968                     same_suspect(e->suspect, target) &&
969                     min_score < ent_score(sb, e))
970                         num_ents++;
971         if (num_ents) {
972                 blame_list = xcalloc(num_ents, sizeof(struct blame_list));
973                 for (e = sb->ent, i = 0; e; e = e->next)
974                         if (!e->scanned && !e->guilty &&
975                             same_suspect(e->suspect, target) &&
976                             min_score < ent_score(sb, e))
977                                 blame_list[i++].ent = e;
978         }
979         *num_ents_p = num_ents;
980         return blame_list;
981 }
982
983 /*
984  * Reset the scanned status on all entries.
985  */
986 static void reset_scanned_flag(struct scoreboard *sb)
987 {
988         struct blame_entry *e;
989         for (e = sb->ent; e; e = e->next)
990                 e->scanned = 0;
991 }
992
993 /*
994  * For lines target is suspected for, see if we can find code movement
995  * across file boundary from the parent commit.  porigin is the path
996  * in the parent we already tried.
997  */
998 static int find_copy_in_parent(struct scoreboard *sb,
999                                struct origin *target,
1000                                struct commit *parent,
1001                                struct origin *porigin,
1002                                int opt)
1003 {
1004         struct diff_options diff_opts;
1005         const char *paths[1];
1006         int i, j;
1007         int retval;
1008         struct blame_list *blame_list;
1009         int num_ents;
1010
1011         blame_list = setup_blame_list(sb, target, blame_copy_score, &num_ents);
1012         if (!blame_list)
1013                 return 1; /* nothing remains for this target */
1014
1015         diff_setup(&diff_opts);
1016         DIFF_OPT_SET(&diff_opts, RECURSIVE);
1017         diff_opts.output_format = DIFF_FORMAT_NO_OUTPUT;
1018
1019         paths[0] = NULL;
1020         diff_tree_setup_paths(paths, &diff_opts);
1021         if (diff_setup_done(&diff_opts) < 0)
1022                 die("diff-setup");
1023
1024         /* Try "find copies harder" on new path if requested;
1025          * we do not want to use diffcore_rename() actually to
1026          * match things up; find_copies_harder is set only to
1027          * force diff_tree_sha1() to feed all filepairs to diff_queue,
1028          * and this code needs to be after diff_setup_done(), which
1029          * usually makes find-copies-harder imply copy detection.
1030          */
1031         if ((opt & PICKAXE_BLAME_COPY_HARDEST)
1032             || ((opt & PICKAXE_BLAME_COPY_HARDER)
1033                 && (!porigin || strcmp(target->path, porigin->path))))
1034                 DIFF_OPT_SET(&diff_opts, FIND_COPIES_HARDER);
1035
1036         if (is_null_sha1(target->commit->object.sha1))
1037                 do_diff_cache(parent->tree->object.sha1, &diff_opts);
1038         else
1039                 diff_tree_sha1(parent->tree->object.sha1,
1040                                target->commit->tree->object.sha1,
1041                                "", &diff_opts);
1042
1043         if (!DIFF_OPT_TST(&diff_opts, FIND_COPIES_HARDER))
1044                 diffcore_std(&diff_opts);
1045
1046         retval = 0;
1047         while (1) {
1048                 int made_progress = 0;
1049
1050                 for (i = 0; i < diff_queued_diff.nr; i++) {
1051                         struct diff_filepair *p = diff_queued_diff.queue[i];
1052                         struct origin *norigin;
1053                         mmfile_t file_p;
1054                         struct blame_entry this[3];
1055
1056                         if (!DIFF_FILE_VALID(p->one))
1057                                 continue; /* does not exist in parent */
1058                         if (S_ISGITLINK(p->one->mode))
1059                                 continue; /* ignore git links */
1060                         if (porigin && !strcmp(p->one->path, porigin->path))
1061                                 /* find_move already dealt with this path */
1062                                 continue;
1063
1064                         norigin = get_origin(sb, parent, p->one->path);
1065                         hashcpy(norigin->blob_sha1, p->one->sha1);
1066                         fill_origin_blob(norigin, &file_p);
1067                         if (!file_p.ptr)
1068                                 continue;
1069
1070                         for (j = 0; j < num_ents; j++) {
1071                                 find_copy_in_blob(sb, blame_list[j].ent,
1072                                                   norigin, this, &file_p);
1073                                 copy_split_if_better(sb, blame_list[j].split,
1074                                                      this);
1075                                 decref_split(this);
1076                         }
1077                         origin_decref(norigin);
1078                 }
1079
1080                 for (j = 0; j < num_ents; j++) {
1081                         struct blame_entry *split = blame_list[j].split;
1082                         if (split[1].suspect &&
1083                             blame_copy_score < ent_score(sb, &split[1])) {
1084                                 split_blame(sb, split, blame_list[j].ent);
1085                                 made_progress = 1;
1086                         }
1087                         else
1088                                 blame_list[j].ent->scanned = 1;
1089                         decref_split(split);
1090                 }
1091                 free(blame_list);
1092
1093                 if (!made_progress)
1094                         break;
1095                 blame_list = setup_blame_list(sb, target, blame_copy_score, &num_ents);
1096                 if (!blame_list) {
1097                         retval = 1;
1098                         break;
1099                 }
1100         }
1101         reset_scanned_flag(sb);
1102         diff_flush(&diff_opts);
1103         diff_tree_release_paths(&diff_opts);
1104         return retval;
1105 }
1106
1107 /*
1108  * The blobs of origin and porigin exactly match, so everything
1109  * origin is suspected for can be blamed on the parent.
1110  */
1111 static void pass_whole_blame(struct scoreboard *sb,
1112                              struct origin *origin, struct origin *porigin)
1113 {
1114         struct blame_entry *e;
1115
1116         if (!porigin->file.ptr && origin->file.ptr) {
1117                 /* Steal its file */
1118                 porigin->file = origin->file;
1119                 origin->file.ptr = NULL;
1120         }
1121         for (e = sb->ent; e; e = e->next) {
1122                 if (!same_suspect(e->suspect, origin))
1123                         continue;
1124                 origin_incref(porigin);
1125                 origin_decref(e->suspect);
1126                 e->suspect = porigin;
1127         }
1128 }
1129
1130 /*
1131  * We pass blame from the current commit to its parents.  We keep saying
1132  * "parent" (and "porigin"), but what we mean is to find scapegoat to
1133  * exonerate ourselves.
1134  */
1135 static struct commit_list *first_scapegoat(struct rev_info *revs, struct commit *commit)
1136 {
1137         if (!reverse)
1138                 return commit->parents;
1139         return lookup_decoration(&revs->children, &commit->object);
1140 }
1141
1142 static int num_scapegoats(struct rev_info *revs, struct commit *commit)
1143 {
1144         int cnt;
1145         struct commit_list *l = first_scapegoat(revs, commit);
1146         for (cnt = 0; l; l = l->next)
1147                 cnt++;
1148         return cnt;
1149 }
1150
1151 #define MAXSG 16
1152
1153 static void pass_blame(struct scoreboard *sb, struct origin *origin, int opt)
1154 {
1155         struct rev_info *revs = sb->revs;
1156         int i, pass, num_sg;
1157         struct commit *commit = origin->commit;
1158         struct commit_list *sg;
1159         struct origin *sg_buf[MAXSG];
1160         struct origin *porigin, **sg_origin = sg_buf;
1161
1162         num_sg = num_scapegoats(revs, commit);
1163         if (!num_sg)
1164                 goto finish;
1165         else if (num_sg < ARRAY_SIZE(sg_buf))
1166                 memset(sg_buf, 0, sizeof(sg_buf));
1167         else
1168                 sg_origin = xcalloc(num_sg, sizeof(*sg_origin));
1169
1170         /*
1171          * The first pass looks for unrenamed path to optimize for
1172          * common cases, then we look for renames in the second pass.
1173          */
1174         for (pass = 0; pass < 2; pass++) {
1175                 struct origin *(*find)(struct scoreboard *,
1176                                        struct commit *, struct origin *);
1177                 find = pass ? find_rename : find_origin;
1178
1179                 for (i = 0, sg = first_scapegoat(revs, commit);
1180                      i < num_sg && sg;
1181                      sg = sg->next, i++) {
1182                         struct commit *p = sg->item;
1183                         int j, same;
1184
1185                         if (sg_origin[i])
1186                                 continue;
1187                         if (parse_commit(p))
1188                                 continue;
1189                         porigin = find(sb, p, origin);
1190                         if (!porigin)
1191                                 continue;
1192                         if (!hashcmp(porigin->blob_sha1, origin->blob_sha1)) {
1193                                 pass_whole_blame(sb, origin, porigin);
1194                                 origin_decref(porigin);
1195                                 goto finish;
1196                         }
1197                         for (j = same = 0; j < i; j++)
1198                                 if (sg_origin[j] &&
1199                                     !hashcmp(sg_origin[j]->blob_sha1,
1200                                              porigin->blob_sha1)) {
1201                                         same = 1;
1202                                         break;
1203                                 }
1204                         if (!same)
1205                                 sg_origin[i] = porigin;
1206                         else
1207                                 origin_decref(porigin);
1208                 }
1209         }
1210
1211         num_commits++;
1212         for (i = 0, sg = first_scapegoat(revs, commit);
1213              i < num_sg && sg;
1214              sg = sg->next, i++) {
1215                 struct origin *porigin = sg_origin[i];
1216                 if (!porigin)
1217                         continue;
1218                 if (!origin->previous) {
1219                         origin_incref(porigin);
1220                         origin->previous = porigin;
1221                 }
1222                 if (pass_blame_to_parent(sb, origin, porigin))
1223                         goto finish;
1224         }
1225
1226         /*
1227          * Optionally find moves in parents' files.
1228          */
1229         if (opt & PICKAXE_BLAME_MOVE)
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 (!porigin)
1235                                 continue;
1236                         if (find_move_in_parent(sb, origin, porigin))
1237                                 goto finish;
1238                 }
1239
1240         /*
1241          * Optionally find copies from parents' files.
1242          */
1243         if (opt & PICKAXE_BLAME_COPY)
1244                 for (i = 0, sg = first_scapegoat(revs, commit);
1245                      i < num_sg && sg;
1246                      sg = sg->next, i++) {
1247                         struct origin *porigin = sg_origin[i];
1248                         if (find_copy_in_parent(sb, origin, sg->item,
1249                                                 porigin, opt))
1250                                 goto finish;
1251                 }
1252
1253  finish:
1254         for (i = 0; i < num_sg; i++) {
1255                 if (sg_origin[i]) {
1256                         drop_origin_blob(sg_origin[i]);
1257                         origin_decref(sg_origin[i]);
1258                 }
1259         }
1260         drop_origin_blob(origin);
1261         if (sg_buf != sg_origin)
1262                 free(sg_origin);
1263 }
1264
1265 /*
1266  * Information on commits, used for output.
1267  */
1268 struct commit_info
1269 {
1270         const char *author;
1271         const char *author_mail;
1272         unsigned long author_time;
1273         const char *author_tz;
1274
1275         /* filled only when asked for details */
1276         const char *committer;
1277         const char *committer_mail;
1278         unsigned long committer_time;
1279         const char *committer_tz;
1280
1281         const char *summary;
1282 };
1283
1284 /*
1285  * Parse author/committer line in the commit object buffer
1286  */
1287 static void get_ac_line(const char *inbuf, const char *what,
1288                         int person_len, char *person,
1289                         int mail_len, char *mail,
1290                         unsigned long *time, const char **tz)
1291 {
1292         int len, tzlen, maillen;
1293         char *tmp, *endp, *timepos, *mailpos;
1294
1295         tmp = strstr(inbuf, what);
1296         if (!tmp)
1297                 goto error_out;
1298         tmp += strlen(what);
1299         endp = strchr(tmp, '\n');
1300         if (!endp)
1301                 len = strlen(tmp);
1302         else
1303                 len = endp - tmp;
1304         if (person_len <= len) {
1305         error_out:
1306                 /* Ugh */
1307                 *tz = "(unknown)";
1308                 strcpy(mail, *tz);
1309                 *time = 0;
1310                 return;
1311         }
1312         memcpy(person, tmp, len);
1313
1314         tmp = person;
1315         tmp += len;
1316         *tmp = 0;
1317         while (*tmp != ' ')
1318                 tmp--;
1319         *tz = tmp+1;
1320         tzlen = (person+len)-(tmp+1);
1321
1322         *tmp = 0;
1323         while (*tmp != ' ')
1324                 tmp--;
1325         *time = strtoul(tmp, NULL, 10);
1326         timepos = tmp;
1327
1328         *tmp = 0;
1329         while (*tmp != ' ')
1330                 tmp--;
1331         mailpos = tmp + 1;
1332         *tmp = 0;
1333         maillen = timepos - tmp;
1334         memcpy(mail, mailpos, maillen);
1335
1336         if (!mailmap.nr)
1337                 return;
1338
1339         /*
1340          * mailmap expansion may make the name longer.
1341          * make room by pushing stuff down.
1342          */
1343         tmp = person + person_len - (tzlen + 1);
1344         memmove(tmp, *tz, tzlen);
1345         tmp[tzlen] = 0;
1346         *tz = tmp;
1347
1348         /*
1349          * Now, convert both name and e-mail using mailmap
1350          */
1351         if (map_user(&mailmap, mail+1, mail_len-1, person, tmp-person-1)) {
1352                 /* Add a trailing '>' to email, since map_user returns plain emails
1353                    Note: It already has '<', since we replace from mail+1 */
1354                 mailpos = memchr(mail, '\0', mail_len);
1355                 if (mailpos && mailpos-mail < mail_len - 1) {
1356                         *mailpos = '>';
1357                         *(mailpos+1) = '\0';
1358                 }
1359         }
1360 }
1361
1362 static void get_commit_info(struct commit *commit,
1363                             struct commit_info *ret,
1364                             int detailed)
1365 {
1366         int len;
1367         char *tmp, *endp, *reencoded, *message;
1368         static char author_name[1024];
1369         static char author_mail[1024];
1370         static char committer_name[1024];
1371         static char committer_mail[1024];
1372         static char summary_buf[1024];
1373
1374         /*
1375          * We've operated without save_commit_buffer, so
1376          * we now need to populate them for output.
1377          */
1378         if (!commit->buffer) {
1379                 enum object_type type;
1380                 unsigned long size;
1381                 commit->buffer =
1382                         read_sha1_file(commit->object.sha1, &type, &size);
1383                 if (!commit->buffer)
1384                         die("Cannot read commit %s",
1385                             sha1_to_hex(commit->object.sha1));
1386         }
1387         reencoded = reencode_commit_message(commit, NULL);
1388         message   = reencoded ? reencoded : commit->buffer;
1389         ret->author = author_name;
1390         ret->author_mail = author_mail;
1391         get_ac_line(message, "\nauthor ",
1392                     sizeof(author_name), author_name,
1393                     sizeof(author_mail), author_mail,
1394                     &ret->author_time, &ret->author_tz);
1395
1396         if (!detailed) {
1397                 free(reencoded);
1398                 return;
1399         }
1400
1401         ret->committer = committer_name;
1402         ret->committer_mail = committer_mail;
1403         get_ac_line(message, "\ncommitter ",
1404                     sizeof(committer_name), committer_name,
1405                     sizeof(committer_mail), committer_mail,
1406                     &ret->committer_time, &ret->committer_tz);
1407
1408         ret->summary = summary_buf;
1409         tmp = strstr(message, "\n\n");
1410         if (!tmp) {
1411         error_out:
1412                 sprintf(summary_buf, "(%s)", sha1_to_hex(commit->object.sha1));
1413                 free(reencoded);
1414                 return;
1415         }
1416         tmp += 2;
1417         endp = strchr(tmp, '\n');
1418         if (!endp)
1419                 endp = tmp + strlen(tmp);
1420         len = endp - tmp;
1421         if (len >= sizeof(summary_buf) || len == 0)
1422                 goto error_out;
1423         memcpy(summary_buf, tmp, len);
1424         summary_buf[len] = 0;
1425         free(reencoded);
1426 }
1427
1428 /*
1429  * To allow LF and other nonportable characters in pathnames,
1430  * they are c-style quoted as needed.
1431  */
1432 static void write_filename_info(const char *path)
1433 {
1434         printf("filename ");
1435         write_name_quoted(path, stdout, '\n');
1436 }
1437
1438 /*
1439  * Porcelain/Incremental format wants to show a lot of details per
1440  * commit.  Instead of repeating this every line, emit it only once,
1441  * the first time each commit appears in the output.
1442  */
1443 static int emit_one_suspect_detail(struct origin *suspect)
1444 {
1445         struct commit_info ci;
1446
1447         if (suspect->commit->object.flags & METAINFO_SHOWN)
1448                 return 0;
1449
1450         suspect->commit->object.flags |= METAINFO_SHOWN;
1451         get_commit_info(suspect->commit, &ci, 1);
1452         printf("author %s\n", ci.author);
1453         printf("author-mail %s\n", ci.author_mail);
1454         printf("author-time %lu\n", ci.author_time);
1455         printf("author-tz %s\n", ci.author_tz);
1456         printf("committer %s\n", ci.committer);
1457         printf("committer-mail %s\n", ci.committer_mail);
1458         printf("committer-time %lu\n", ci.committer_time);
1459         printf("committer-tz %s\n", ci.committer_tz);
1460         printf("summary %s\n", ci.summary);
1461         if (suspect->commit->object.flags & UNINTERESTING)
1462                 printf("boundary\n");
1463         if (suspect->previous) {
1464                 struct origin *prev = suspect->previous;
1465                 printf("previous %s ", sha1_to_hex(prev->commit->object.sha1));
1466                 write_name_quoted(prev->path, stdout, '\n');
1467         }
1468         return 1;
1469 }
1470
1471 /*
1472  * The blame_entry is found to be guilty for the range.  Mark it
1473  * as such, and show it in incremental output.
1474  */
1475 static void found_guilty_entry(struct blame_entry *ent)
1476 {
1477         if (ent->guilty)
1478                 return;
1479         ent->guilty = 1;
1480         if (incremental) {
1481                 struct origin *suspect = ent->suspect;
1482
1483                 printf("%s %d %d %d\n",
1484                        sha1_to_hex(suspect->commit->object.sha1),
1485                        ent->s_lno + 1, ent->lno + 1, ent->num_lines);
1486                 emit_one_suspect_detail(suspect);
1487                 write_filename_info(suspect->path);
1488                 maybe_flush_or_die(stdout, "stdout");
1489         }
1490 }
1491
1492 /*
1493  * The main loop -- while the scoreboard has lines whose true origin
1494  * is still unknown, pick one blame_entry, and allow its current
1495  * suspect to pass blames to its parents.
1496  */
1497 static void assign_blame(struct scoreboard *sb, int opt)
1498 {
1499         struct rev_info *revs = sb->revs;
1500
1501         while (1) {
1502                 struct blame_entry *ent;
1503                 struct commit *commit;
1504                 struct origin *suspect = NULL;
1505
1506                 /* find one suspect to break down */
1507                 for (ent = sb->ent; !suspect && ent; ent = ent->next)
1508                         if (!ent->guilty)
1509                                 suspect = ent->suspect;
1510                 if (!suspect)
1511                         return; /* all done */
1512
1513                 /*
1514                  * We will use this suspect later in the loop,
1515                  * so hold onto it in the meantime.
1516                  */
1517                 origin_incref(suspect);
1518                 commit = suspect->commit;
1519                 if (!commit->object.parsed)
1520                         parse_commit(commit);
1521                 if (reverse ||
1522                     (!(commit->object.flags & UNINTERESTING) &&
1523                      !(revs->max_age != -1 && commit->date < revs->max_age)))
1524                         pass_blame(sb, suspect, opt);
1525                 else {
1526                         commit->object.flags |= UNINTERESTING;
1527                         if (commit->object.parsed)
1528                                 mark_parents_uninteresting(commit);
1529                 }
1530                 /* treat root commit as boundary */
1531                 if (!commit->parents && !show_root)
1532                         commit->object.flags |= UNINTERESTING;
1533
1534                 /* Take responsibility for the remaining entries */
1535                 for (ent = sb->ent; ent; ent = ent->next)
1536                         if (same_suspect(ent->suspect, suspect))
1537                                 found_guilty_entry(ent);
1538                 origin_decref(suspect);
1539
1540                 if (DEBUG) /* sanity */
1541                         sanity_check_refcnt(sb);
1542         }
1543 }
1544
1545 static const char *format_time(unsigned long time, const char *tz_str,
1546                                int show_raw_time)
1547 {
1548         static char time_buf[128];
1549         const char *time_str;
1550         int time_len;
1551         int tz;
1552
1553         if (show_raw_time) {
1554                 sprintf(time_buf, "%lu %s", time, tz_str);
1555         }
1556         else {
1557                 tz = atoi(tz_str);
1558                 time_str = show_date(time, tz, blame_date_mode);
1559                 time_len = strlen(time_str);
1560                 memcpy(time_buf, time_str, time_len);
1561                 memset(time_buf + time_len, ' ', blame_date_width - time_len);
1562         }
1563         return time_buf;
1564 }
1565
1566 #define OUTPUT_ANNOTATE_COMPAT  001
1567 #define OUTPUT_LONG_OBJECT_NAME 002
1568 #define OUTPUT_RAW_TIMESTAMP    004
1569 #define OUTPUT_PORCELAIN        010
1570 #define OUTPUT_SHOW_NAME        020
1571 #define OUTPUT_SHOW_NUMBER      040
1572 #define OUTPUT_SHOW_SCORE      0100
1573 #define OUTPUT_NO_AUTHOR       0200
1574
1575 static void emit_porcelain(struct scoreboard *sb, struct blame_entry *ent)
1576 {
1577         int cnt;
1578         const char *cp;
1579         struct origin *suspect = ent->suspect;
1580         char hex[41];
1581
1582         strcpy(hex, sha1_to_hex(suspect->commit->object.sha1));
1583         printf("%s%c%d %d %d\n",
1584                hex,
1585                ent->guilty ? ' ' : '*', // purely for debugging
1586                ent->s_lno + 1,
1587                ent->lno + 1,
1588                ent->num_lines);
1589         if (emit_one_suspect_detail(suspect) ||
1590             (suspect->commit->object.flags & MORE_THAN_ONE_PATH))
1591                 write_filename_info(suspect->path);
1592
1593         cp = nth_line(sb, ent->lno);
1594         for (cnt = 0; cnt < ent->num_lines; cnt++) {
1595                 char ch;
1596                 if (cnt)
1597                         printf("%s %d %d\n", hex,
1598                                ent->s_lno + 1 + cnt,
1599                                ent->lno + 1 + cnt);
1600                 putchar('\t');
1601                 do {
1602                         ch = *cp++;
1603                         putchar(ch);
1604                 } while (ch != '\n' &&
1605                          cp < sb->final_buf + sb->final_buf_size);
1606         }
1607
1608         if (sb->final_buf_size && cp[-1] != '\n')
1609                 putchar('\n');
1610 }
1611
1612 static void emit_other(struct scoreboard *sb, struct blame_entry *ent, int opt)
1613 {
1614         int cnt;
1615         const char *cp;
1616         struct origin *suspect = ent->suspect;
1617         struct commit_info ci;
1618         char hex[41];
1619         int show_raw_time = !!(opt & OUTPUT_RAW_TIMESTAMP);
1620
1621         get_commit_info(suspect->commit, &ci, 1);
1622         strcpy(hex, sha1_to_hex(suspect->commit->object.sha1));
1623
1624         cp = nth_line(sb, ent->lno);
1625         for (cnt = 0; cnt < ent->num_lines; cnt++) {
1626                 char ch;
1627                 int length = (opt & OUTPUT_LONG_OBJECT_NAME) ? 40 : 8;
1628
1629                 if (suspect->commit->object.flags & UNINTERESTING) {
1630                         if (blank_boundary)
1631                                 memset(hex, ' ', length);
1632                         else if (!(opt & OUTPUT_ANNOTATE_COMPAT)) {
1633                                 length--;
1634                                 putchar('^');
1635                         }
1636                 }
1637
1638                 printf("%.*s", length, hex);
1639                 if (opt & OUTPUT_ANNOTATE_COMPAT)
1640                         printf("\t(%10s\t%10s\t%d)", ci.author,
1641                                format_time(ci.author_time, ci.author_tz,
1642                                            show_raw_time),
1643                                ent->lno + 1 + cnt);
1644                 else {
1645                         if (opt & OUTPUT_SHOW_SCORE)
1646                                 printf(" %*d %02d",
1647                                        max_score_digits, ent->score,
1648                                        ent->suspect->refcnt);
1649                         if (opt & OUTPUT_SHOW_NAME)
1650                                 printf(" %-*.*s", longest_file, longest_file,
1651                                        suspect->path);
1652                         if (opt & OUTPUT_SHOW_NUMBER)
1653                                 printf(" %*d", max_orig_digits,
1654                                        ent->s_lno + 1 + cnt);
1655
1656                         if (!(opt & OUTPUT_NO_AUTHOR)) {
1657                                 int pad = longest_author - utf8_strwidth(ci.author);
1658                                 printf(" (%s%*s %10s",
1659                                        ci.author, pad, "",
1660                                        format_time(ci.author_time,
1661                                                    ci.author_tz,
1662                                                    show_raw_time));
1663                         }
1664                         printf(" %*d) ",
1665                                max_digits, ent->lno + 1 + cnt);
1666                 }
1667                 do {
1668                         ch = *cp++;
1669                         putchar(ch);
1670                 } while (ch != '\n' &&
1671                          cp < sb->final_buf + sb->final_buf_size);
1672         }
1673
1674         if (sb->final_buf_size && cp[-1] != '\n')
1675                 putchar('\n');
1676 }
1677
1678 static void output(struct scoreboard *sb, int option)
1679 {
1680         struct blame_entry *ent;
1681
1682         if (option & OUTPUT_PORCELAIN) {
1683                 for (ent = sb->ent; ent; ent = ent->next) {
1684                         struct blame_entry *oth;
1685                         struct origin *suspect = ent->suspect;
1686                         struct commit *commit = suspect->commit;
1687                         if (commit->object.flags & MORE_THAN_ONE_PATH)
1688                                 continue;
1689                         for (oth = ent->next; oth; oth = oth->next) {
1690                                 if ((oth->suspect->commit != commit) ||
1691                                     !strcmp(oth->suspect->path, suspect->path))
1692                                         continue;
1693                                 commit->object.flags |= MORE_THAN_ONE_PATH;
1694                                 break;
1695                         }
1696                 }
1697         }
1698
1699         for (ent = sb->ent; ent; ent = ent->next) {
1700                 if (option & OUTPUT_PORCELAIN)
1701                         emit_porcelain(sb, ent);
1702                 else {
1703                         emit_other(sb, ent, option);
1704                 }
1705         }
1706 }
1707
1708 /*
1709  * To allow quick access to the contents of nth line in the
1710  * final image, prepare an index in the scoreboard.
1711  */
1712 static int prepare_lines(struct scoreboard *sb)
1713 {
1714         const char *buf = sb->final_buf;
1715         unsigned long len = sb->final_buf_size;
1716         int num = 0, incomplete = 0, bol = 1;
1717
1718         if (len && buf[len-1] != '\n')
1719                 incomplete++; /* incomplete line at the end */
1720         while (len--) {
1721                 if (bol) {
1722                         sb->lineno = xrealloc(sb->lineno,
1723                                               sizeof(int *) * (num + 1));
1724                         sb->lineno[num] = buf - sb->final_buf;
1725                         bol = 0;
1726                 }
1727                 if (*buf++ == '\n') {
1728                         num++;
1729                         bol = 1;
1730                 }
1731         }
1732         sb->lineno = xrealloc(sb->lineno,
1733                               sizeof(int *) * (num + incomplete + 1));
1734         sb->lineno[num + incomplete] = buf - sb->final_buf;
1735         sb->num_lines = num + incomplete;
1736         return sb->num_lines;
1737 }
1738
1739 /*
1740  * Add phony grafts for use with -S; this is primarily to
1741  * support git's cvsserver that wants to give a linear history
1742  * to its clients.
1743  */
1744 static int read_ancestry(const char *graft_file)
1745 {
1746         FILE *fp = fopen(graft_file, "r");
1747         char buf[1024];
1748         if (!fp)
1749                 return -1;
1750         while (fgets(buf, sizeof(buf), fp)) {
1751                 /* The format is just "Commit Parent1 Parent2 ...\n" */
1752                 int len = strlen(buf);
1753                 struct commit_graft *graft = read_graft_line(buf, len);
1754                 if (graft)
1755                         register_commit_graft(graft, 0);
1756         }
1757         fclose(fp);
1758         return 0;
1759 }
1760
1761 /*
1762  * How many columns do we need to show line numbers in decimal?
1763  */
1764 static int lineno_width(int lines)
1765 {
1766         int i, width;
1767
1768         for (width = 1, i = 10; i <= lines + 1; width++)
1769                 i *= 10;
1770         return width;
1771 }
1772
1773 /*
1774  * How many columns do we need to show line numbers, authors,
1775  * and filenames?
1776  */
1777 static void find_alignment(struct scoreboard *sb, int *option)
1778 {
1779         int longest_src_lines = 0;
1780         int longest_dst_lines = 0;
1781         unsigned largest_score = 0;
1782         struct blame_entry *e;
1783
1784         for (e = sb->ent; e; e = e->next) {
1785                 struct origin *suspect = e->suspect;
1786                 struct commit_info ci;
1787                 int num;
1788
1789                 if (strcmp(suspect->path, sb->path))
1790                         *option |= OUTPUT_SHOW_NAME;
1791                 num = strlen(suspect->path);
1792                 if (longest_file < num)
1793                         longest_file = num;
1794                 if (!(suspect->commit->object.flags & METAINFO_SHOWN)) {
1795                         suspect->commit->object.flags |= METAINFO_SHOWN;
1796                         get_commit_info(suspect->commit, &ci, 1);
1797                         num = utf8_strwidth(ci.author);
1798                         if (longest_author < num)
1799                                 longest_author = num;
1800                 }
1801                 num = e->s_lno + e->num_lines;
1802                 if (longest_src_lines < num)
1803                         longest_src_lines = num;
1804                 num = e->lno + e->num_lines;
1805                 if (longest_dst_lines < num)
1806                         longest_dst_lines = num;
1807                 if (largest_score < ent_score(sb, e))
1808                         largest_score = ent_score(sb, e);
1809         }
1810         max_orig_digits = lineno_width(longest_src_lines);
1811         max_digits = lineno_width(longest_dst_lines);
1812         max_score_digits = lineno_width(largest_score);
1813 }
1814
1815 /*
1816  * For debugging -- origin is refcounted, and this asserts that
1817  * we do not underflow.
1818  */
1819 static void sanity_check_refcnt(struct scoreboard *sb)
1820 {
1821         int baa = 0;
1822         struct blame_entry *ent;
1823
1824         for (ent = sb->ent; ent; ent = ent->next) {
1825                 /* Nobody should have zero or negative refcnt */
1826                 if (ent->suspect->refcnt <= 0) {
1827                         fprintf(stderr, "%s in %s has negative refcnt %d\n",
1828                                 ent->suspect->path,
1829                                 sha1_to_hex(ent->suspect->commit->object.sha1),
1830                                 ent->suspect->refcnt);
1831                         baa = 1;
1832                 }
1833         }
1834         if (baa) {
1835                 int opt = 0160;
1836                 find_alignment(sb, &opt);
1837                 output(sb, opt);
1838                 die("Baa %d!", baa);
1839         }
1840 }
1841
1842 /*
1843  * Used for the command line parsing; check if the path exists
1844  * in the working tree.
1845  */
1846 static int has_string_in_work_tree(const char *path)
1847 {
1848         struct stat st;
1849         return !lstat(path, &st);
1850 }
1851
1852 static unsigned parse_score(const char *arg)
1853 {
1854         char *end;
1855         unsigned long score = strtoul(arg, &end, 10);
1856         if (*end)
1857                 return 0;
1858         return score;
1859 }
1860
1861 static const char *add_prefix(const char *prefix, const char *path)
1862 {
1863         return prefix_path(prefix, prefix ? strlen(prefix) : 0, path);
1864 }
1865
1866 /*
1867  * Parsing of (comma separated) one item in the -L option
1868  */
1869 static const char *parse_loc(const char *spec,
1870                              struct scoreboard *sb, long lno,
1871                              long begin, long *ret)
1872 {
1873         char *term;
1874         const char *line;
1875         long num;
1876         int reg_error;
1877         regex_t regexp;
1878         regmatch_t match[1];
1879
1880         /* Allow "-L <something>,+20" to mean starting at <something>
1881          * for 20 lines, or "-L <something>,-5" for 5 lines ending at
1882          * <something>.
1883          */
1884         if (1 < begin && (spec[0] == '+' || spec[0] == '-')) {
1885                 num = strtol(spec + 1, &term, 10);
1886                 if (term != spec + 1) {
1887                         if (spec[0] == '-')
1888                                 num = 0 - num;
1889                         if (0 < num)
1890                                 *ret = begin + num - 2;
1891                         else if (!num)
1892                                 *ret = begin;
1893                         else
1894                                 *ret = begin + num;
1895                         return term;
1896                 }
1897                 return spec;
1898         }
1899         num = strtol(spec, &term, 10);
1900         if (term != spec) {
1901                 *ret = num;
1902                 return term;
1903         }
1904         if (spec[0] != '/')
1905                 return spec;
1906
1907         /* it could be a regexp of form /.../ */
1908         for (term = (char *) spec + 1; *term && *term != '/'; term++) {
1909                 if (*term == '\\')
1910                         term++;
1911         }
1912         if (*term != '/')
1913                 return spec;
1914
1915         /* try [spec+1 .. term-1] as regexp */
1916         *term = 0;
1917         begin--; /* input is in human terms */
1918         line = nth_line(sb, begin);
1919
1920         if (!(reg_error = regcomp(&regexp, spec + 1, REG_NEWLINE)) &&
1921             !(reg_error = regexec(&regexp, line, 1, match, 0))) {
1922                 const char *cp = line + match[0].rm_so;
1923                 const char *nline;
1924
1925                 while (begin++ < lno) {
1926                         nline = nth_line(sb, begin);
1927                         if (line <= cp && cp < nline)
1928                                 break;
1929                         line = nline;
1930                 }
1931                 *ret = begin;
1932                 regfree(&regexp);
1933                 *term++ = '/';
1934                 return term;
1935         }
1936         else {
1937                 char errbuf[1024];
1938                 regerror(reg_error, &regexp, errbuf, 1024);
1939                 die("-L parameter '%s': %s", spec + 1, errbuf);
1940         }
1941 }
1942
1943 /*
1944  * Parsing of -L option
1945  */
1946 static void prepare_blame_range(struct scoreboard *sb,
1947                                 const char *bottomtop,
1948                                 long lno,
1949                                 long *bottom, long *top)
1950 {
1951         const char *term;
1952
1953         term = parse_loc(bottomtop, sb, lno, 1, bottom);
1954         if (*term == ',') {
1955                 term = parse_loc(term + 1, sb, lno, *bottom + 1, top);
1956                 if (*term)
1957                         usage(blame_usage);
1958         }
1959         if (*term)
1960                 usage(blame_usage);
1961 }
1962
1963 static int git_blame_config(const char *var, const char *value, void *cb)
1964 {
1965         if (!strcmp(var, "blame.showroot")) {
1966                 show_root = git_config_bool(var, value);
1967                 return 0;
1968         }
1969         if (!strcmp(var, "blame.blankboundary")) {
1970                 blank_boundary = git_config_bool(var, value);
1971                 return 0;
1972         }
1973         if (!strcmp(var, "blame.date")) {
1974                 if (!value)
1975                         return config_error_nonbool(var);
1976                 blame_date_mode = parse_date_format(value);
1977                 return 0;
1978         }
1979         return git_default_config(var, value, cb);
1980 }
1981
1982 /*
1983  * Prepare a dummy commit that represents the work tree (or staged) item.
1984  * Note that annotating work tree item never works in the reverse.
1985  */
1986 static struct commit *fake_working_tree_commit(const char *path, const char *contents_from)
1987 {
1988         struct commit *commit;
1989         struct origin *origin;
1990         unsigned char head_sha1[20];
1991         struct strbuf buf = STRBUF_INIT;
1992         const char *ident;
1993         time_t now;
1994         int size, len;
1995         struct cache_entry *ce;
1996         unsigned mode;
1997
1998         if (get_sha1("HEAD", head_sha1))
1999                 die("No such ref: HEAD");
2000
2001         time(&now);
2002         commit = xcalloc(1, sizeof(*commit));
2003         commit->parents = xcalloc(1, sizeof(*commit->parents));
2004         commit->parents->item = lookup_commit_reference(head_sha1);
2005         commit->object.parsed = 1;
2006         commit->date = now;
2007         commit->object.type = OBJ_COMMIT;
2008
2009         origin = make_origin(commit, path);
2010
2011         if (!contents_from || strcmp("-", contents_from)) {
2012                 struct stat st;
2013                 const char *read_from;
2014
2015                 if (contents_from) {
2016                         if (stat(contents_from, &st) < 0)
2017                                 die_errno("Cannot stat '%s'", contents_from);
2018                         read_from = contents_from;
2019                 }
2020                 else {
2021                         if (lstat(path, &st) < 0)
2022                                 die_errno("Cannot lstat '%s'", path);
2023                         read_from = path;
2024                 }
2025                 mode = canon_mode(st.st_mode);
2026                 switch (st.st_mode & S_IFMT) {
2027                 case S_IFREG:
2028                         if (strbuf_read_file(&buf, read_from, st.st_size) != st.st_size)
2029                                 die_errno("cannot open or read '%s'", read_from);
2030                         break;
2031                 case S_IFLNK:
2032                         if (strbuf_readlink(&buf, read_from, st.st_size) < 0)
2033                                 die_errno("cannot readlink '%s'", read_from);
2034                         break;
2035                 default:
2036                         die("unsupported file type %s", read_from);
2037                 }
2038         }
2039         else {
2040                 /* Reading from stdin */
2041                 contents_from = "standard input";
2042                 mode = 0;
2043                 if (strbuf_read(&buf, 0, 0) < 0)
2044                         die_errno("failed to read from stdin");
2045         }
2046         convert_to_git(path, buf.buf, buf.len, &buf, 0);
2047         origin->file.ptr = buf.buf;
2048         origin->file.size = buf.len;
2049         pretend_sha1_file(buf.buf, buf.len, OBJ_BLOB, origin->blob_sha1);
2050         commit->util = origin;
2051
2052         /*
2053          * Read the current index, replace the path entry with
2054          * origin->blob_sha1 without mucking with its mode or type
2055          * bits; we are not going to write this index out -- we just
2056          * want to run "diff-index --cached".
2057          */
2058         discard_cache();
2059         read_cache();
2060
2061         len = strlen(path);
2062         if (!mode) {
2063                 int pos = cache_name_pos(path, len);
2064                 if (0 <= pos)
2065                         mode = active_cache[pos]->ce_mode;
2066                 else
2067                         /* Let's not bother reading from HEAD tree */
2068                         mode = S_IFREG | 0644;
2069         }
2070         size = cache_entry_size(len);
2071         ce = xcalloc(1, size);
2072         hashcpy(ce->sha1, origin->blob_sha1);
2073         memcpy(ce->name, path, len);
2074         ce->ce_flags = create_ce_flags(len, 0);
2075         ce->ce_mode = create_ce_mode(mode);
2076         add_cache_entry(ce, ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE);
2077
2078         /*
2079          * We are not going to write this out, so this does not matter
2080          * right now, but someday we might optimize diff-index --cached
2081          * with cache-tree information.
2082          */
2083         cache_tree_invalidate_path(active_cache_tree, path);
2084
2085         commit->buffer = xmalloc(400);
2086         ident = fmt_ident("Not Committed Yet", "not.committed.yet", NULL, 0);
2087         snprintf(commit->buffer, 400,
2088                 "tree 0000000000000000000000000000000000000000\n"
2089                 "parent %s\n"
2090                 "author %s\n"
2091                 "committer %s\n\n"
2092                 "Version of %s from %s\n",
2093                 sha1_to_hex(head_sha1),
2094                 ident, ident, path, contents_from ? contents_from : path);
2095         return commit;
2096 }
2097
2098 static const char *prepare_final(struct scoreboard *sb)
2099 {
2100         int i;
2101         const char *final_commit_name = NULL;
2102         struct rev_info *revs = sb->revs;
2103
2104         /*
2105          * There must be one and only one positive commit in the
2106          * revs->pending array.
2107          */
2108         for (i = 0; i < revs->pending.nr; i++) {
2109                 struct object *obj = revs->pending.objects[i].item;
2110                 if (obj->flags & UNINTERESTING)
2111                         continue;
2112                 while (obj->type == OBJ_TAG)
2113                         obj = deref_tag(obj, NULL, 0);
2114                 if (obj->type != OBJ_COMMIT)
2115                         die("Non commit %s?", revs->pending.objects[i].name);
2116                 if (sb->final)
2117                         die("More than one commit to dig from %s and %s?",
2118                             revs->pending.objects[i].name,
2119                             final_commit_name);
2120                 sb->final = (struct commit *) obj;
2121                 final_commit_name = revs->pending.objects[i].name;
2122         }
2123         return final_commit_name;
2124 }
2125
2126 static const char *prepare_initial(struct scoreboard *sb)
2127 {
2128         int i;
2129         const char *final_commit_name = NULL;
2130         struct rev_info *revs = sb->revs;
2131
2132         /*
2133          * There must be one and only one negative commit, and it must be
2134          * the boundary.
2135          */
2136         for (i = 0; i < revs->pending.nr; i++) {
2137                 struct object *obj = revs->pending.objects[i].item;
2138                 if (!(obj->flags & UNINTERESTING))
2139                         continue;
2140                 while (obj->type == OBJ_TAG)
2141                         obj = deref_tag(obj, NULL, 0);
2142                 if (obj->type != OBJ_COMMIT)
2143                         die("Non commit %s?", revs->pending.objects[i].name);
2144                 if (sb->final)
2145                         die("More than one commit to dig down to %s and %s?",
2146                             revs->pending.objects[i].name,
2147                             final_commit_name);
2148                 sb->final = (struct commit *) obj;
2149                 final_commit_name = revs->pending.objects[i].name;
2150         }
2151         if (!final_commit_name)
2152                 die("No commit to dig down to?");
2153         return final_commit_name;
2154 }
2155
2156 static int blame_copy_callback(const struct option *option, const char *arg, int unset)
2157 {
2158         int *opt = option->value;
2159
2160         /*
2161          * -C enables copy from removed files;
2162          * -C -C enables copy from existing files, but only
2163          *       when blaming a new file;
2164          * -C -C -C enables copy from existing files for
2165          *          everybody
2166          */
2167         if (*opt & PICKAXE_BLAME_COPY_HARDER)
2168                 *opt |= PICKAXE_BLAME_COPY_HARDEST;
2169         if (*opt & PICKAXE_BLAME_COPY)
2170                 *opt |= PICKAXE_BLAME_COPY_HARDER;
2171         *opt |= PICKAXE_BLAME_COPY | PICKAXE_BLAME_MOVE;
2172
2173         if (arg)
2174                 blame_copy_score = parse_score(arg);
2175         return 0;
2176 }
2177
2178 static int blame_move_callback(const struct option *option, const char *arg, int unset)
2179 {
2180         int *opt = option->value;
2181
2182         *opt |= PICKAXE_BLAME_MOVE;
2183
2184         if (arg)
2185                 blame_move_score = parse_score(arg);
2186         return 0;
2187 }
2188
2189 static int blame_bottomtop_callback(const struct option *option, const char *arg, int unset)
2190 {
2191         const char **bottomtop = option->value;
2192         if (!arg)
2193                 return -1;
2194         if (*bottomtop)
2195                 die("More than one '-L n,m' option given");
2196         *bottomtop = arg;
2197         return 0;
2198 }
2199
2200 int cmd_blame(int argc, const char **argv, const char *prefix)
2201 {
2202         struct rev_info revs;
2203         const char *path;
2204         struct scoreboard sb;
2205         struct origin *o;
2206         struct blame_entry *ent;
2207         long dashdash_pos, bottom, top, lno;
2208         const char *final_commit_name = NULL;
2209         enum object_type type;
2210
2211         static const char *bottomtop = NULL;
2212         static int output_option = 0, opt = 0;
2213         static int show_stats = 0;
2214         static const char *revs_file = NULL;
2215         static const char *contents_from = NULL;
2216         static const struct option options[] = {
2217                 OPT_BOOLEAN(0, "incremental", &incremental, "Show blame entries as we find them, incrementally"),
2218                 OPT_BOOLEAN('b', NULL, &blank_boundary, "Show blank SHA-1 for boundary commits (Default: off)"),
2219                 OPT_BOOLEAN(0, "root", &show_root, "Do not treat root commits as boundaries (Default: off)"),
2220                 OPT_BOOLEAN(0, "show-stats", &show_stats, "Show work cost statistics"),
2221                 OPT_BIT(0, "score-debug", &output_option, "Show output score for blame entries", OUTPUT_SHOW_SCORE),
2222                 OPT_BIT('f', "show-name", &output_option, "Show original filename (Default: auto)", OUTPUT_SHOW_NAME),
2223                 OPT_BIT('n', "show-number", &output_option, "Show original linenumber (Default: off)", OUTPUT_SHOW_NUMBER),
2224                 OPT_BIT('p', "porcelain", &output_option, "Show in a format designed for machine consumption", OUTPUT_PORCELAIN),
2225                 OPT_BIT('c', NULL, &output_option, "Use the same output mode as git-annotate (Default: off)", OUTPUT_ANNOTATE_COMPAT),
2226                 OPT_BIT('t', NULL, &output_option, "Show raw timestamp (Default: off)", OUTPUT_RAW_TIMESTAMP),
2227                 OPT_BIT('l', NULL, &output_option, "Show long commit SHA1 (Default: off)", OUTPUT_LONG_OBJECT_NAME),
2228                 OPT_BIT('s', NULL, &output_option, "Suppress author name and timestamp (Default: off)", OUTPUT_NO_AUTHOR),
2229                 OPT_BIT('w', NULL, &xdl_opts, "Ignore whitespace differences", XDF_IGNORE_WHITESPACE),
2230                 OPT_STRING('S', NULL, &revs_file, "file", "Use revisions from <file> instead of calling git-rev-list"),
2231                 OPT_STRING(0, "contents", &contents_from, "file", "Use <file>'s contents as the final image"),
2232                 { OPTION_CALLBACK, 'C', NULL, &opt, "score", "Find line copies within and across files", PARSE_OPT_OPTARG, blame_copy_callback },
2233                 { OPTION_CALLBACK, 'M', NULL, &opt, "score", "Find line movements within and across files", PARSE_OPT_OPTARG, blame_move_callback },
2234                 OPT_CALLBACK('L', NULL, &bottomtop, "n,m", "Process only line range n,m, counting from 1", blame_bottomtop_callback),
2235                 OPT_END()
2236         };
2237
2238         struct parse_opt_ctx_t ctx;
2239         int cmd_is_annotate = !strcmp(argv[0], "annotate");
2240
2241         git_config(git_blame_config, NULL);
2242         init_revisions(&revs, NULL);
2243         revs.date_mode = blame_date_mode;
2244
2245         save_commit_buffer = 0;
2246         dashdash_pos = 0;
2247
2248         parse_options_start(&ctx, argc, argv, prefix, PARSE_OPT_KEEP_DASHDASH |
2249                             PARSE_OPT_KEEP_ARGV0);
2250         for (;;) {
2251                 switch (parse_options_step(&ctx, options, blame_opt_usage)) {
2252                 case PARSE_OPT_HELP:
2253                         exit(129);
2254                 case PARSE_OPT_DONE:
2255                         if (ctx.argv[0])
2256                                 dashdash_pos = ctx.cpidx;
2257                         goto parse_done;
2258                 }
2259
2260                 if (!strcmp(ctx.argv[0], "--reverse")) {
2261                         ctx.argv[0] = "--children";
2262                         reverse = 1;
2263                 }
2264                 parse_revision_opt(&revs, &ctx, options, blame_opt_usage);
2265         }
2266 parse_done:
2267         argc = parse_options_end(&ctx);
2268
2269         if (revs_file && read_ancestry(revs_file))
2270                 die_errno("reading graft file '%s' failed", revs_file);
2271
2272         if (cmd_is_annotate) {
2273                 output_option |= OUTPUT_ANNOTATE_COMPAT;
2274                 blame_date_mode = DATE_ISO8601;
2275         } else {
2276                 blame_date_mode = revs.date_mode;
2277         }
2278
2279         /* The maximum width used to show the dates */
2280         switch (blame_date_mode) {
2281         case DATE_RFC2822:
2282                 blame_date_width = sizeof("Thu, 19 Oct 2006 16:00:04 -0700");
2283                 break;
2284         case DATE_ISO8601:
2285                 blame_date_width = sizeof("2006-10-19 16:00:04 -0700");
2286                 break;
2287         case DATE_RAW:
2288                 blame_date_width = sizeof("1161298804 -0700");
2289                 break;
2290         case DATE_SHORT:
2291                 blame_date_width = sizeof("2006-10-19");
2292                 break;
2293         case DATE_RELATIVE:
2294                 /* "normal" is used as the fallback for "relative" */
2295         case DATE_LOCAL:
2296         case DATE_NORMAL:
2297                 blame_date_width = sizeof("Thu Oct 19 16:00:04 2006 -0700");
2298                 break;
2299         }
2300         blame_date_width -= 1; /* strip the null */
2301
2302         if (DIFF_OPT_TST(&revs.diffopt, FIND_COPIES_HARDER))
2303                 opt |= (PICKAXE_BLAME_COPY | PICKAXE_BLAME_MOVE |
2304                         PICKAXE_BLAME_COPY_HARDER);
2305
2306         if (!blame_move_score)
2307                 blame_move_score = BLAME_DEFAULT_MOVE_SCORE;
2308         if (!blame_copy_score)
2309                 blame_copy_score = BLAME_DEFAULT_COPY_SCORE;
2310
2311         /*
2312          * We have collected options unknown to us in argv[1..unk]
2313          * which are to be passed to revision machinery if we are
2314          * going to do the "bottom" processing.
2315          *
2316          * The remaining are:
2317          *
2318          * (1) if dashdash_pos != 0, its either
2319          *     "blame [revisions] -- <path>" or
2320          *     "blame -- <path> <rev>"
2321          *
2322          * (2) otherwise, its one of the two:
2323          *     "blame [revisions] <path>"
2324          *     "blame <path> <rev>"
2325          *
2326          * Note that we must strip out <path> from the arguments: we do not
2327          * want the path pruning but we may want "bottom" processing.
2328          */
2329         if (dashdash_pos) {
2330                 switch (argc - dashdash_pos - 1) {
2331                 case 2: /* (1b) */
2332                         if (argc != 4)
2333                                 usage_with_options(blame_opt_usage, options);
2334                         /* reorder for the new way: <rev> -- <path> */
2335                         argv[1] = argv[3];
2336                         argv[3] = argv[2];
2337                         argv[2] = "--";
2338                         /* FALLTHROUGH */
2339                 case 1: /* (1a) */
2340                         path = add_prefix(prefix, argv[--argc]);
2341                         argv[argc] = NULL;
2342                         break;
2343                 default:
2344                         usage_with_options(blame_opt_usage, options);
2345                 }
2346         } else {
2347                 if (argc < 2)
2348                         usage_with_options(blame_opt_usage, options);
2349                 path = add_prefix(prefix, argv[argc - 1]);
2350                 if (argc == 3 && !has_string_in_work_tree(path)) { /* (2b) */
2351                         path = add_prefix(prefix, argv[1]);
2352                         argv[1] = argv[2];
2353                 }
2354                 argv[argc - 1] = "--";
2355
2356                 setup_work_tree();
2357                 if (!has_string_in_work_tree(path))
2358                         die_errno("cannot stat path '%s'", path);
2359         }
2360
2361         setup_revisions(argc, argv, &revs, NULL);
2362         memset(&sb, 0, sizeof(sb));
2363
2364         sb.revs = &revs;
2365         if (!reverse)
2366                 final_commit_name = prepare_final(&sb);
2367         else if (contents_from)
2368                 die("--contents and --children do not blend well.");
2369         else
2370                 final_commit_name = prepare_initial(&sb);
2371
2372         if (!sb.final) {
2373                 /*
2374                  * "--not A B -- path" without anything positive;
2375                  * do not default to HEAD, but use the working tree
2376                  * or "--contents".
2377                  */
2378                 setup_work_tree();
2379                 sb.final = fake_working_tree_commit(path, contents_from);
2380                 add_pending_object(&revs, &(sb.final->object), ":");
2381         }
2382         else if (contents_from)
2383                 die("Cannot use --contents with final commit object name");
2384
2385         /*
2386          * If we have bottom, this will mark the ancestors of the
2387          * bottom commits we would reach while traversing as
2388          * uninteresting.
2389          */
2390         if (prepare_revision_walk(&revs))
2391                 die("revision walk setup failed");
2392
2393         if (is_null_sha1(sb.final->object.sha1)) {
2394                 char *buf;
2395                 o = sb.final->util;
2396                 buf = xmalloc(o->file.size + 1);
2397                 memcpy(buf, o->file.ptr, o->file.size + 1);
2398                 sb.final_buf = buf;
2399                 sb.final_buf_size = o->file.size;
2400         }
2401         else {
2402                 o = get_origin(&sb, sb.final, path);
2403                 if (fill_blob_sha1(o))
2404                         die("no such path %s in %s", path, final_commit_name);
2405
2406                 sb.final_buf = read_sha1_file(o->blob_sha1, &type,
2407                                               &sb.final_buf_size);
2408                 if (!sb.final_buf)
2409                         die("Cannot read blob %s for path %s",
2410                             sha1_to_hex(o->blob_sha1),
2411                             path);
2412         }
2413         num_read_blob++;
2414         lno = prepare_lines(&sb);
2415
2416         bottom = top = 0;
2417         if (bottomtop)
2418                 prepare_blame_range(&sb, bottomtop, lno, &bottom, &top);
2419         if (bottom && top && top < bottom) {
2420                 long tmp;
2421                 tmp = top; top = bottom; bottom = tmp;
2422         }
2423         if (bottom < 1)
2424                 bottom = 1;
2425         if (top < 1)
2426                 top = lno;
2427         bottom--;
2428         if (lno < top)
2429                 die("file %s has only %lu lines", path, lno);
2430
2431         ent = xcalloc(1, sizeof(*ent));
2432         ent->lno = bottom;
2433         ent->num_lines = top - bottom;
2434         ent->suspect = o;
2435         ent->s_lno = bottom;
2436
2437         sb.ent = ent;
2438         sb.path = path;
2439
2440         read_mailmap(&mailmap, NULL);
2441
2442         if (!incremental)
2443                 setup_pager();
2444
2445         assign_blame(&sb, opt);
2446
2447         if (incremental)
2448                 return 0;
2449
2450         coalesce(&sb);
2451
2452         if (!(output_option & OUTPUT_PORCELAIN))
2453                 find_alignment(&sb, &output_option);
2454
2455         output(&sb, output_option);
2456         free((void *)sb.final_buf);
2457         for (ent = sb.ent; ent; ) {
2458                 struct blame_entry *e = ent->next;
2459                 free(ent);
2460                 ent = e;
2461         }
2462
2463         if (show_stats) {
2464                 printf("num read blob: %d\n", num_read_blob);
2465                 printf("num get patch: %d\n", num_get_patch);
2466                 printf("num commits: %d\n", num_commits);
2467         }
2468         return 0;
2469 }