Merge branch 'jc/maint-doc-em-dash'
[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 #include "userdiff.h"
24
25 static char blame_usage[] = "git blame [options] [rev-opts] [rev] [--] file";
26
27 static const char *blame_opt_usage[] = {
28         blame_usage,
29         "",
30         "[rev-opts] are documented in git-rev-list(1)",
31         NULL
32 };
33
34 static int longest_file;
35 static int longest_author;
36 static int max_orig_digits;
37 static int max_digits;
38 static int max_score_digits;
39 static int show_root;
40 static int reverse;
41 static int blank_boundary;
42 static int incremental;
43 static int xdl_opts;
44
45 static enum date_mode blame_date_mode = DATE_ISO8601;
46 static size_t blame_date_width;
47
48 static struct string_list mailmap;
49
50 #ifndef DEBUG
51 #define DEBUG 0
52 #endif
53
54 /* stats */
55 static int num_read_blob;
56 static int num_get_patch;
57 static int num_commits;
58
59 #define PICKAXE_BLAME_MOVE              01
60 #define PICKAXE_BLAME_COPY              02
61 #define PICKAXE_BLAME_COPY_HARDER       04
62 #define PICKAXE_BLAME_COPY_HARDEST      010
63
64 /*
65  * blame for a blame_entry with score lower than these thresholds
66  * is not passed to the parent using move/copy logic.
67  */
68 static unsigned blame_move_score;
69 static unsigned blame_copy_score;
70 #define BLAME_DEFAULT_MOVE_SCORE        20
71 #define BLAME_DEFAULT_COPY_SCORE        40
72
73 /* bits #0..7 in revision.h, #8..11 used for merge_bases() in commit.c */
74 #define METAINFO_SHOWN          (1u<<12)
75 #define MORE_THAN_ONE_PATH      (1u<<13)
76
77 /*
78  * One blob in a commit that is being suspected
79  */
80 struct origin {
81         int refcnt;
82         struct origin *previous;
83         struct commit *commit;
84         mmfile_t file;
85         unsigned char blob_sha1[20];
86         char path[FLEX_ARRAY];
87 };
88
89 /*
90  * Prepare diff_filespec and convert it using diff textconv API
91  * if the textconv driver exists.
92  * Return 1 if the conversion succeeds, 0 otherwise.
93  */
94 int textconv_object(const char *path,
95                     const unsigned char *sha1,
96                     char **buf,
97                     unsigned long *buf_size)
98 {
99         struct diff_filespec *df;
100         struct userdiff_driver *textconv;
101
102         df = alloc_filespec(path);
103         fill_filespec(df, sha1, S_IFREG | 0664);
104         textconv = get_textconv(df);
105         if (!textconv) {
106                 free_filespec(df);
107                 return 0;
108         }
109
110         *buf_size = fill_textconv(textconv, df, buf);
111         free_filespec(df);
112         return 1;
113 }
114
115 /*
116  * Given an origin, prepare mmfile_t structure to be used by the
117  * diff machinery
118  */
119 static void fill_origin_blob(struct diff_options *opt,
120                              struct origin *o, mmfile_t *file)
121 {
122         if (!o->file.ptr) {
123                 enum object_type type;
124                 unsigned long file_size;
125
126                 num_read_blob++;
127                 if (DIFF_OPT_TST(opt, ALLOW_TEXTCONV) &&
128                     textconv_object(o->path, o->blob_sha1, &file->ptr, &file_size))
129                         ;
130                 else
131                         file->ptr = read_sha1_file(o->blob_sha1, &type, &file_size);
132                 file->size = file_size;
133
134                 if (!file->ptr)
135                         die("Cannot read blob %s for path %s",
136                             sha1_to_hex(o->blob_sha1),
137                             o->path);
138                 o->file = *file;
139         }
140         else
141                 *file = o->file;
142 }
143
144 /*
145  * Origin is refcounted and usually we keep the blob contents to be
146  * reused.
147  */
148 static inline struct origin *origin_incref(struct origin *o)
149 {
150         if (o)
151                 o->refcnt++;
152         return o;
153 }
154
155 static void origin_decref(struct origin *o)
156 {
157         if (o && --o->refcnt <= 0) {
158                 if (o->previous)
159                         origin_decref(o->previous);
160                 free(o->file.ptr);
161                 free(o);
162         }
163 }
164
165 static void drop_origin_blob(struct origin *o)
166 {
167         if (o->file.ptr) {
168                 free(o->file.ptr);
169                 o->file.ptr = NULL;
170         }
171 }
172
173 /*
174  * Each group of lines is described by a blame_entry; it can be split
175  * as we pass blame to the parents.  They form a linked list in the
176  * scoreboard structure, sorted by the target line number.
177  */
178 struct blame_entry {
179         struct blame_entry *prev;
180         struct blame_entry *next;
181
182         /* the first line of this group in the final image;
183          * internally all line numbers are 0 based.
184          */
185         int lno;
186
187         /* how many lines this group has */
188         int num_lines;
189
190         /* the commit that introduced this group into the final image */
191         struct origin *suspect;
192
193         /* true if the suspect is truly guilty; false while we have not
194          * checked if the group came from one of its parents.
195          */
196         char guilty;
197
198         /* true if the entry has been scanned for copies in the current parent
199          */
200         char scanned;
201
202         /* the line number of the first line of this group in the
203          * suspect's file; internally all line numbers are 0 based.
204          */
205         int s_lno;
206
207         /* how significant this entry is -- cached to avoid
208          * scanning the lines over and over.
209          */
210         unsigned score;
211 };
212
213 /*
214  * The current state of the blame assignment.
215  */
216 struct scoreboard {
217         /* the final commit (i.e. where we started digging from) */
218         struct commit *final;
219         struct rev_info *revs;
220         const char *path;
221
222         /*
223          * The contents in the final image.
224          * Used by many functions to obtain contents of the nth line,
225          * indexed with scoreboard.lineno[blame_entry.lno].
226          */
227         const char *final_buf;
228         unsigned long final_buf_size;
229
230         /* linked list of blames */
231         struct blame_entry *ent;
232
233         /* look-up a line in the final buffer */
234         int num_lines;
235         int *lineno;
236 };
237
238 static inline int same_suspect(struct origin *a, struct origin *b)
239 {
240         if (a == b)
241                 return 1;
242         if (a->commit != b->commit)
243                 return 0;
244         return !strcmp(a->path, b->path);
245 }
246
247 static void sanity_check_refcnt(struct scoreboard *);
248
249 /*
250  * If two blame entries that are next to each other came from
251  * contiguous lines in the same origin (i.e. <commit, path> pair),
252  * merge them together.
253  */
254 static void coalesce(struct scoreboard *sb)
255 {
256         struct blame_entry *ent, *next;
257
258         for (ent = sb->ent; ent && (next = ent->next); ent = next) {
259                 if (same_suspect(ent->suspect, next->suspect) &&
260                     ent->guilty == next->guilty &&
261                     ent->s_lno + ent->num_lines == next->s_lno) {
262                         ent->num_lines += next->num_lines;
263                         ent->next = next->next;
264                         if (ent->next)
265                                 ent->next->prev = ent;
266                         origin_decref(next->suspect);
267                         free(next);
268                         ent->score = 0;
269                         next = ent; /* again */
270                 }
271         }
272
273         if (DEBUG) /* sanity */
274                 sanity_check_refcnt(sb);
275 }
276
277 /*
278  * Given a commit and a path in it, create a new origin structure.
279  * The callers that add blame to the scoreboard should use
280  * get_origin() to obtain shared, refcounted copy instead of calling
281  * this function directly.
282  */
283 static struct origin *make_origin(struct commit *commit, const char *path)
284 {
285         struct origin *o;
286         o = xcalloc(1, sizeof(*o) + strlen(path) + 1);
287         o->commit = commit;
288         o->refcnt = 1;
289         strcpy(o->path, path);
290         return o;
291 }
292
293 /*
294  * Locate an existing origin or create a new one.
295  */
296 static struct origin *get_origin(struct scoreboard *sb,
297                                  struct commit *commit,
298                                  const char *path)
299 {
300         struct blame_entry *e;
301
302         for (e = sb->ent; e; e = e->next) {
303                 if (e->suspect->commit == commit &&
304                     !strcmp(e->suspect->path, path))
305                         return origin_incref(e->suspect);
306         }
307         return make_origin(commit, path);
308 }
309
310 /*
311  * Fill the blob_sha1 field of an origin if it hasn't, so that later
312  * call to fill_origin_blob() can use it to locate the data.  blob_sha1
313  * for an origin is also used to pass the blame for the entire file to
314  * the parent to detect the case where a child's blob is identical to
315  * that of its parent's.
316  */
317 static int fill_blob_sha1(struct origin *origin)
318 {
319         unsigned mode;
320         if (!is_null_sha1(origin->blob_sha1))
321                 return 0;
322         if (get_tree_entry(origin->commit->object.sha1,
323                            origin->path,
324                            origin->blob_sha1, &mode))
325                 goto error_out;
326         if (sha1_object_info(origin->blob_sha1, NULL) != OBJ_BLOB)
327                 goto error_out;
328         return 0;
329  error_out:
330         hashclr(origin->blob_sha1);
331         return -1;
332 }
333
334 /*
335  * We have an origin -- check if the same path exists in the
336  * parent and return an origin structure to represent it.
337  */
338 static struct origin *find_origin(struct scoreboard *sb,
339                                   struct commit *parent,
340                                   struct origin *origin)
341 {
342         struct origin *porigin = NULL;
343         struct diff_options diff_opts;
344         const char *paths[2];
345
346         if (parent->util) {
347                 /*
348                  * Each commit object can cache one origin in that
349                  * commit.  This is a freestanding copy of origin and
350                  * not refcounted.
351                  */
352                 struct origin *cached = parent->util;
353                 if (!strcmp(cached->path, origin->path)) {
354                         /*
355                          * The same path between origin and its parent
356                          * without renaming -- the most common case.
357                          */
358                         porigin = get_origin(sb, parent, cached->path);
359
360                         /*
361                          * If the origin was newly created (i.e. get_origin
362                          * would call make_origin if none is found in the
363                          * scoreboard), it does not know the blob_sha1,
364                          * so copy it.  Otherwise porigin was in the
365                          * scoreboard and already knows blob_sha1.
366                          */
367                         if (porigin->refcnt == 1)
368                                 hashcpy(porigin->blob_sha1, cached->blob_sha1);
369                         return porigin;
370                 }
371                 /* otherwise it was not very useful; free it */
372                 free(parent->util);
373                 parent->util = NULL;
374         }
375
376         /* See if the origin->path is different between parent
377          * and origin first.  Most of the time they are the
378          * same and diff-tree is fairly efficient about this.
379          */
380         diff_setup(&diff_opts);
381         DIFF_OPT_SET(&diff_opts, RECURSIVE);
382         diff_opts.detect_rename = 0;
383         diff_opts.output_format = DIFF_FORMAT_NO_OUTPUT;
384         paths[0] = origin->path;
385         paths[1] = NULL;
386
387         diff_tree_setup_paths(paths, &diff_opts);
388         if (diff_setup_done(&diff_opts) < 0)
389                 die("diff-setup");
390
391         if (is_null_sha1(origin->commit->object.sha1))
392                 do_diff_cache(parent->tree->object.sha1, &diff_opts);
393         else
394                 diff_tree_sha1(parent->tree->object.sha1,
395                                origin->commit->tree->object.sha1,
396                                "", &diff_opts);
397         diffcore_std(&diff_opts);
398
399         if (!diff_queued_diff.nr) {
400                 /* The path is the same as parent */
401                 porigin = get_origin(sb, parent, origin->path);
402                 hashcpy(porigin->blob_sha1, origin->blob_sha1);
403         } else {
404                 /*
405                  * Since origin->path is a pathspec, if the parent
406                  * commit had it as a directory, we will see a whole
407                  * bunch of deletion of files in the directory that we
408                  * do not care about.
409                  */
410                 int i;
411                 struct diff_filepair *p = NULL;
412                 for (i = 0; i < diff_queued_diff.nr; i++) {
413                         const char *name;
414                         p = diff_queued_diff.queue[i];
415                         name = p->one->path ? p->one->path : p->two->path;
416                         if (!strcmp(name, origin->path))
417                                 break;
418                 }
419                 if (!p)
420                         die("internal error in blame::find_origin");
421                 switch (p->status) {
422                 default:
423                         die("internal error in blame::find_origin (%c)",
424                             p->status);
425                 case 'M':
426                         porigin = get_origin(sb, parent, origin->path);
427                         hashcpy(porigin->blob_sha1, p->one->sha1);
428                         break;
429                 case 'A':
430                 case 'T':
431                         /* Did not exist in parent, or type changed */
432                         break;
433                 }
434         }
435         diff_flush(&diff_opts);
436         diff_tree_release_paths(&diff_opts);
437         if (porigin) {
438                 /*
439                  * Create a freestanding copy that is not part of
440                  * the refcounted origin found in the scoreboard, and
441                  * cache it in the commit.
442                  */
443                 struct origin *cached;
444
445                 cached = make_origin(porigin->commit, porigin->path);
446                 hashcpy(cached->blob_sha1, porigin->blob_sha1);
447                 parent->util = cached;
448         }
449         return porigin;
450 }
451
452 /*
453  * We have an origin -- find the path that corresponds to it in its
454  * parent and return an origin structure to represent it.
455  */
456 static struct origin *find_rename(struct scoreboard *sb,
457                                   struct commit *parent,
458                                   struct origin *origin)
459 {
460         struct origin *porigin = NULL;
461         struct diff_options diff_opts;
462         int i;
463         const char *paths[2];
464
465         diff_setup(&diff_opts);
466         DIFF_OPT_SET(&diff_opts, RECURSIVE);
467         diff_opts.detect_rename = DIFF_DETECT_RENAME;
468         diff_opts.output_format = DIFF_FORMAT_NO_OUTPUT;
469         diff_opts.single_follow = origin->path;
470         paths[0] = NULL;
471         diff_tree_setup_paths(paths, &diff_opts);
472         if (diff_setup_done(&diff_opts) < 0)
473                 die("diff-setup");
474
475         if (is_null_sha1(origin->commit->object.sha1))
476                 do_diff_cache(parent->tree->object.sha1, &diff_opts);
477         else
478                 diff_tree_sha1(parent->tree->object.sha1,
479                                origin->commit->tree->object.sha1,
480                                "", &diff_opts);
481         diffcore_std(&diff_opts);
482
483         for (i = 0; i < diff_queued_diff.nr; i++) {
484                 struct diff_filepair *p = diff_queued_diff.queue[i];
485                 if ((p->status == 'R' || p->status == 'C') &&
486                     !strcmp(p->two->path, origin->path)) {
487                         porigin = get_origin(sb, parent, p->one->path);
488                         hashcpy(porigin->blob_sha1, p->one->sha1);
489                         break;
490                 }
491         }
492         diff_flush(&diff_opts);
493         diff_tree_release_paths(&diff_opts);
494         return porigin;
495 }
496
497 /*
498  * Link in a new blame entry to the scoreboard.  Entries that cover the
499  * same line range have been removed from the scoreboard previously.
500  */
501 static void add_blame_entry(struct scoreboard *sb, struct blame_entry *e)
502 {
503         struct blame_entry *ent, *prev = NULL;
504
505         origin_incref(e->suspect);
506
507         for (ent = sb->ent; ent && ent->lno < e->lno; ent = ent->next)
508                 prev = ent;
509
510         /* prev, if not NULL, is the last one that is below e */
511         e->prev = prev;
512         if (prev) {
513                 e->next = prev->next;
514                 prev->next = e;
515         }
516         else {
517                 e->next = sb->ent;
518                 sb->ent = e;
519         }
520         if (e->next)
521                 e->next->prev = e;
522 }
523
524 /*
525  * src typically is on-stack; we want to copy the information in it to
526  * a malloced blame_entry that is already on the linked list of the
527  * scoreboard.  The origin of dst loses a refcnt while the origin of src
528  * gains one.
529  */
530 static void dup_entry(struct blame_entry *dst, struct blame_entry *src)
531 {
532         struct blame_entry *p, *n;
533
534         p = dst->prev;
535         n = dst->next;
536         origin_incref(src->suspect);
537         origin_decref(dst->suspect);
538         memcpy(dst, src, sizeof(*src));
539         dst->prev = p;
540         dst->next = n;
541         dst->score = 0;
542 }
543
544 static const char *nth_line(struct scoreboard *sb, int lno)
545 {
546         return sb->final_buf + sb->lineno[lno];
547 }
548
549 /*
550  * It is known that lines between tlno to same came from parent, and e
551  * has an overlap with that range.  it also is known that parent's
552  * line plno corresponds to e's line tlno.
553  *
554  *                <---- e ----->
555  *                   <------>
556  *                   <------------>
557  *             <------------>
558  *             <------------------>
559  *
560  * Split e into potentially three parts; before this chunk, the chunk
561  * to be blamed for the parent, and after that portion.
562  */
563 static void split_overlap(struct blame_entry *split,
564                           struct blame_entry *e,
565                           int tlno, int plno, int same,
566                           struct origin *parent)
567 {
568         int chunk_end_lno;
569         memset(split, 0, sizeof(struct blame_entry [3]));
570
571         if (e->s_lno < tlno) {
572                 /* there is a pre-chunk part not blamed on parent */
573                 split[0].suspect = origin_incref(e->suspect);
574                 split[0].lno = e->lno;
575                 split[0].s_lno = e->s_lno;
576                 split[0].num_lines = tlno - e->s_lno;
577                 split[1].lno = e->lno + tlno - e->s_lno;
578                 split[1].s_lno = plno;
579         }
580         else {
581                 split[1].lno = e->lno;
582                 split[1].s_lno = plno + (e->s_lno - tlno);
583         }
584
585         if (same < e->s_lno + e->num_lines) {
586                 /* there is a post-chunk part not blamed on parent */
587                 split[2].suspect = origin_incref(e->suspect);
588                 split[2].lno = e->lno + (same - e->s_lno);
589                 split[2].s_lno = e->s_lno + (same - e->s_lno);
590                 split[2].num_lines = e->s_lno + e->num_lines - same;
591                 chunk_end_lno = split[2].lno;
592         }
593         else
594                 chunk_end_lno = e->lno + e->num_lines;
595         split[1].num_lines = chunk_end_lno - split[1].lno;
596
597         /*
598          * if it turns out there is nothing to blame the parent for,
599          * forget about the splitting.  !split[1].suspect signals this.
600          */
601         if (split[1].num_lines < 1)
602                 return;
603         split[1].suspect = origin_incref(parent);
604 }
605
606 /*
607  * split_overlap() divided an existing blame e into up to three parts
608  * in split.  Adjust the linked list of blames in the scoreboard to
609  * reflect the split.
610  */
611 static void split_blame(struct scoreboard *sb,
612                         struct blame_entry *split,
613                         struct blame_entry *e)
614 {
615         struct blame_entry *new_entry;
616
617         if (split[0].suspect && split[2].suspect) {
618                 /* The first part (reuse storage for the existing entry e) */
619                 dup_entry(e, &split[0]);
620
621                 /* The last part -- me */
622                 new_entry = xmalloc(sizeof(*new_entry));
623                 memcpy(new_entry, &(split[2]), sizeof(struct blame_entry));
624                 add_blame_entry(sb, new_entry);
625
626                 /* ... and the middle part -- parent */
627                 new_entry = xmalloc(sizeof(*new_entry));
628                 memcpy(new_entry, &(split[1]), sizeof(struct blame_entry));
629                 add_blame_entry(sb, new_entry);
630         }
631         else if (!split[0].suspect && !split[2].suspect)
632                 /*
633                  * The parent covers the entire area; reuse storage for
634                  * e and replace it with the parent.
635                  */
636                 dup_entry(e, &split[1]);
637         else if (split[0].suspect) {
638                 /* me and then parent */
639                 dup_entry(e, &split[0]);
640
641                 new_entry = xmalloc(sizeof(*new_entry));
642                 memcpy(new_entry, &(split[1]), sizeof(struct blame_entry));
643                 add_blame_entry(sb, new_entry);
644         }
645         else {
646                 /* parent and then me */
647                 dup_entry(e, &split[1]);
648
649                 new_entry = xmalloc(sizeof(*new_entry));
650                 memcpy(new_entry, &(split[2]), sizeof(struct blame_entry));
651                 add_blame_entry(sb, new_entry);
652         }
653
654         if (DEBUG) { /* sanity */
655                 struct blame_entry *ent;
656                 int lno = sb->ent->lno, corrupt = 0;
657
658                 for (ent = sb->ent; ent; ent = ent->next) {
659                         if (lno != ent->lno)
660                                 corrupt = 1;
661                         if (ent->s_lno < 0)
662                                 corrupt = 1;
663                         lno += ent->num_lines;
664                 }
665                 if (corrupt) {
666                         lno = sb->ent->lno;
667                         for (ent = sb->ent; ent; ent = ent->next) {
668                                 printf("L %8d l %8d n %8d\n",
669                                        lno, ent->lno, ent->num_lines);
670                                 lno = ent->lno + ent->num_lines;
671                         }
672                         die("oops");
673                 }
674         }
675 }
676
677 /*
678  * After splitting the blame, the origins used by the
679  * on-stack blame_entry should lose one refcnt each.
680  */
681 static void decref_split(struct blame_entry *split)
682 {
683         int i;
684
685         for (i = 0; i < 3; i++)
686                 origin_decref(split[i].suspect);
687 }
688
689 /*
690  * Helper for blame_chunk().  blame_entry e is known to overlap with
691  * the patch hunk; split it and pass blame to the parent.
692  */
693 static void blame_overlap(struct scoreboard *sb, struct blame_entry *e,
694                           int tlno, int plno, int same,
695                           struct origin *parent)
696 {
697         struct blame_entry split[3];
698
699         split_overlap(split, e, tlno, plno, same, parent);
700         if (split[1].suspect)
701                 split_blame(sb, split, e);
702         decref_split(split);
703 }
704
705 /*
706  * Find the line number of the last line the target is suspected for.
707  */
708 static int find_last_in_target(struct scoreboard *sb, struct origin *target)
709 {
710         struct blame_entry *e;
711         int last_in_target = -1;
712
713         for (e = sb->ent; e; e = e->next) {
714                 if (e->guilty || !same_suspect(e->suspect, target))
715                         continue;
716                 if (last_in_target < e->s_lno + e->num_lines)
717                         last_in_target = e->s_lno + e->num_lines;
718         }
719         return last_in_target;
720 }
721
722 /*
723  * Process one hunk from the patch between the current suspect for
724  * blame_entry e and its parent.  Find and split the overlap, and
725  * pass blame to the overlapping part to the parent.
726  */
727 static void blame_chunk(struct scoreboard *sb,
728                         int tlno, int plno, int same,
729                         struct origin *target, struct origin *parent)
730 {
731         struct blame_entry *e;
732
733         for (e = sb->ent; e; e = e->next) {
734                 if (e->guilty || !same_suspect(e->suspect, target))
735                         continue;
736                 if (same <= e->s_lno)
737                         continue;
738                 if (tlno < e->s_lno + e->num_lines)
739                         blame_overlap(sb, e, tlno, plno, same, parent);
740         }
741 }
742
743 struct blame_chunk_cb_data {
744         struct scoreboard *sb;
745         struct origin *target;
746         struct origin *parent;
747         long plno;
748         long tlno;
749 };
750
751 static void blame_chunk_cb(void *data, long same, long p_next, long t_next)
752 {
753         struct blame_chunk_cb_data *d = data;
754         blame_chunk(d->sb, d->tlno, d->plno, same, d->target, d->parent);
755         d->plno = p_next;
756         d->tlno = t_next;
757 }
758
759 /*
760  * We are looking at the origin 'target' and aiming to pass blame
761  * for the lines it is suspected to its parent.  Run diff to find
762  * which lines came from parent and pass blame for them.
763  */
764 static int pass_blame_to_parent(struct scoreboard *sb,
765                                 struct origin *target,
766                                 struct origin *parent)
767 {
768         int last_in_target;
769         mmfile_t file_p, file_o;
770         struct blame_chunk_cb_data d;
771         xpparam_t xpp;
772         xdemitconf_t xecfg;
773         memset(&d, 0, sizeof(d));
774         d.sb = sb; d.target = target; d.parent = parent;
775         last_in_target = find_last_in_target(sb, target);
776         if (last_in_target < 0)
777                 return 1; /* nothing remains for this target */
778
779         fill_origin_blob(&sb->revs->diffopt, parent, &file_p);
780         fill_origin_blob(&sb->revs->diffopt, target, &file_o);
781         num_get_patch++;
782
783         memset(&xpp, 0, sizeof(xpp));
784         xpp.flags = xdl_opts;
785         memset(&xecfg, 0, sizeof(xecfg));
786         xecfg.ctxlen = 0;
787         xdi_diff_hunks(&file_p, &file_o, blame_chunk_cb, &d, &xpp, &xecfg);
788         /* The rest (i.e. anything after tlno) are the same as the parent */
789         blame_chunk(sb, d.tlno, d.plno, last_in_target, target, parent);
790
791         return 0;
792 }
793
794 /*
795  * The lines in blame_entry after splitting blames many times can become
796  * very small and trivial, and at some point it becomes pointless to
797  * blame the parents.  E.g. "\t\t}\n\t}\n\n" appears everywhere in any
798  * ordinary C program, and it is not worth to say it was copied from
799  * totally unrelated file in the parent.
800  *
801  * Compute how trivial the lines in the blame_entry are.
802  */
803 static unsigned ent_score(struct scoreboard *sb, struct blame_entry *e)
804 {
805         unsigned score;
806         const char *cp, *ep;
807
808         if (e->score)
809                 return e->score;
810
811         score = 1;
812         cp = nth_line(sb, e->lno);
813         ep = nth_line(sb, e->lno + e->num_lines);
814         while (cp < ep) {
815                 unsigned ch = *((unsigned char *)cp);
816                 if (isalnum(ch))
817                         score++;
818                 cp++;
819         }
820         e->score = score;
821         return score;
822 }
823
824 /*
825  * best_so_far[] and this[] are both a split of an existing blame_entry
826  * that passes blame to the parent.  Maintain best_so_far the best split
827  * so far, by comparing this and best_so_far and copying this into
828  * bst_so_far as needed.
829  */
830 static void copy_split_if_better(struct scoreboard *sb,
831                                  struct blame_entry *best_so_far,
832                                  struct blame_entry *this)
833 {
834         int i;
835
836         if (!this[1].suspect)
837                 return;
838         if (best_so_far[1].suspect) {
839                 if (ent_score(sb, &this[1]) < ent_score(sb, &best_so_far[1]))
840                         return;
841         }
842
843         for (i = 0; i < 3; i++)
844                 origin_incref(this[i].suspect);
845         decref_split(best_so_far);
846         memcpy(best_so_far, this, sizeof(struct blame_entry [3]));
847 }
848
849 /*
850  * We are looking at a part of the final image represented by
851  * ent (tlno and same are offset by ent->s_lno).
852  * tlno is where we are looking at in the final image.
853  * up to (but not including) same match preimage.
854  * plno is where we are looking at in the preimage.
855  *
856  * <-------------- final image ---------------------->
857  *       <------ent------>
858  *         ^tlno ^same
859  *    <---------preimage----->
860  *         ^plno
861  *
862  * All line numbers are 0-based.
863  */
864 static void handle_split(struct scoreboard *sb,
865                          struct blame_entry *ent,
866                          int tlno, int plno, int same,
867                          struct origin *parent,
868                          struct blame_entry *split)
869 {
870         if (ent->num_lines <= tlno)
871                 return;
872         if (tlno < same) {
873                 struct blame_entry this[3];
874                 tlno += ent->s_lno;
875                 same += ent->s_lno;
876                 split_overlap(this, ent, tlno, plno, same, parent);
877                 copy_split_if_better(sb, split, this);
878                 decref_split(this);
879         }
880 }
881
882 struct handle_split_cb_data {
883         struct scoreboard *sb;
884         struct blame_entry *ent;
885         struct origin *parent;
886         struct blame_entry *split;
887         long plno;
888         long tlno;
889 };
890
891 static void handle_split_cb(void *data, long same, long p_next, long t_next)
892 {
893         struct handle_split_cb_data *d = data;
894         handle_split(d->sb, d->ent, d->tlno, d->plno, same, d->parent, d->split);
895         d->plno = p_next;
896         d->tlno = t_next;
897 }
898
899 /*
900  * Find the lines from parent that are the same as ent so that
901  * we can pass blames to it.  file_p has the blob contents for
902  * the parent.
903  */
904 static void find_copy_in_blob(struct scoreboard *sb,
905                               struct blame_entry *ent,
906                               struct origin *parent,
907                               struct blame_entry *split,
908                               mmfile_t *file_p)
909 {
910         const char *cp;
911         int cnt;
912         mmfile_t file_o;
913         struct handle_split_cb_data d;
914         xpparam_t xpp;
915         xdemitconf_t xecfg;
916         memset(&d, 0, sizeof(d));
917         d.sb = sb; d.ent = ent; d.parent = parent; d.split = split;
918         /*
919          * Prepare mmfile that contains only the lines in ent.
920          */
921         cp = nth_line(sb, ent->lno);
922         file_o.ptr = (char *) cp;
923         cnt = ent->num_lines;
924
925         while (cnt && cp < sb->final_buf + sb->final_buf_size) {
926                 if (*cp++ == '\n')
927                         cnt--;
928         }
929         file_o.size = cp - file_o.ptr;
930
931         /*
932          * file_o is a part of final image we are annotating.
933          * file_p partially may match that image.
934          */
935         memset(&xpp, 0, sizeof(xpp));
936         xpp.flags = xdl_opts;
937         memset(&xecfg, 0, sizeof(xecfg));
938         xecfg.ctxlen = 1;
939         memset(split, 0, sizeof(struct blame_entry [3]));
940         xdi_diff_hunks(file_p, &file_o, handle_split_cb, &d, &xpp, &xecfg);
941         /* remainder, if any, all match the preimage */
942         handle_split(sb, ent, d.tlno, d.plno, ent->num_lines, parent, split);
943 }
944
945 /*
946  * See if lines currently target is suspected for can be attributed to
947  * parent.
948  */
949 static int find_move_in_parent(struct scoreboard *sb,
950                                struct origin *target,
951                                struct origin *parent)
952 {
953         int last_in_target, made_progress;
954         struct blame_entry *e, split[3];
955         mmfile_t file_p;
956
957         last_in_target = find_last_in_target(sb, target);
958         if (last_in_target < 0)
959                 return 1; /* nothing remains for this target */
960
961         fill_origin_blob(&sb->revs->diffopt, parent, &file_p);
962         if (!file_p.ptr)
963                 return 0;
964
965         made_progress = 1;
966         while (made_progress) {
967                 made_progress = 0;
968                 for (e = sb->ent; e; e = e->next) {
969                         if (e->guilty || !same_suspect(e->suspect, target) ||
970                             ent_score(sb, e) < blame_move_score)
971                                 continue;
972                         find_copy_in_blob(sb, e, parent, split, &file_p);
973                         if (split[1].suspect &&
974                             blame_move_score < ent_score(sb, &split[1])) {
975                                 split_blame(sb, split, e);
976                                 made_progress = 1;
977                         }
978                         decref_split(split);
979                 }
980         }
981         return 0;
982 }
983
984 struct blame_list {
985         struct blame_entry *ent;
986         struct blame_entry split[3];
987 };
988
989 /*
990  * Count the number of entries the target is suspected for,
991  * and prepare a list of entry and the best split.
992  */
993 static struct blame_list *setup_blame_list(struct scoreboard *sb,
994                                            struct origin *target,
995                                            int min_score,
996                                            int *num_ents_p)
997 {
998         struct blame_entry *e;
999         int num_ents, i;
1000         struct blame_list *blame_list = NULL;
1001
1002         for (e = sb->ent, num_ents = 0; e; e = e->next)
1003                 if (!e->scanned && !e->guilty &&
1004                     same_suspect(e->suspect, target) &&
1005                     min_score < ent_score(sb, e))
1006                         num_ents++;
1007         if (num_ents) {
1008                 blame_list = xcalloc(num_ents, sizeof(struct blame_list));
1009                 for (e = sb->ent, i = 0; e; e = e->next)
1010                         if (!e->scanned && !e->guilty &&
1011                             same_suspect(e->suspect, target) &&
1012                             min_score < ent_score(sb, e))
1013                                 blame_list[i++].ent = e;
1014         }
1015         *num_ents_p = num_ents;
1016         return blame_list;
1017 }
1018
1019 /*
1020  * Reset the scanned status on all entries.
1021  */
1022 static void reset_scanned_flag(struct scoreboard *sb)
1023 {
1024         struct blame_entry *e;
1025         for (e = sb->ent; e; e = e->next)
1026                 e->scanned = 0;
1027 }
1028
1029 /*
1030  * For lines target is suspected for, see if we can find code movement
1031  * across file boundary from the parent commit.  porigin is the path
1032  * in the parent we already tried.
1033  */
1034 static int find_copy_in_parent(struct scoreboard *sb,
1035                                struct origin *target,
1036                                struct commit *parent,
1037                                struct origin *porigin,
1038                                int opt)
1039 {
1040         struct diff_options diff_opts;
1041         const char *paths[1];
1042         int i, j;
1043         int retval;
1044         struct blame_list *blame_list;
1045         int num_ents;
1046
1047         blame_list = setup_blame_list(sb, target, blame_copy_score, &num_ents);
1048         if (!blame_list)
1049                 return 1; /* nothing remains for this target */
1050
1051         diff_setup(&diff_opts);
1052         DIFF_OPT_SET(&diff_opts, RECURSIVE);
1053         diff_opts.output_format = DIFF_FORMAT_NO_OUTPUT;
1054
1055         paths[0] = NULL;
1056         diff_tree_setup_paths(paths, &diff_opts);
1057         if (diff_setup_done(&diff_opts) < 0)
1058                 die("diff-setup");
1059
1060         /* Try "find copies harder" on new path if requested;
1061          * we do not want to use diffcore_rename() actually to
1062          * match things up; find_copies_harder is set only to
1063          * force diff_tree_sha1() to feed all filepairs to diff_queue,
1064          * and this code needs to be after diff_setup_done(), which
1065          * usually makes find-copies-harder imply copy detection.
1066          */
1067         if ((opt & PICKAXE_BLAME_COPY_HARDEST)
1068             || ((opt & PICKAXE_BLAME_COPY_HARDER)
1069                 && (!porigin || strcmp(target->path, porigin->path))))
1070                 DIFF_OPT_SET(&diff_opts, FIND_COPIES_HARDER);
1071
1072         if (is_null_sha1(target->commit->object.sha1))
1073                 do_diff_cache(parent->tree->object.sha1, &diff_opts);
1074         else
1075                 diff_tree_sha1(parent->tree->object.sha1,
1076                                target->commit->tree->object.sha1,
1077                                "", &diff_opts);
1078
1079         if (!DIFF_OPT_TST(&diff_opts, FIND_COPIES_HARDER))
1080                 diffcore_std(&diff_opts);
1081
1082         retval = 0;
1083         while (1) {
1084                 int made_progress = 0;
1085
1086                 for (i = 0; i < diff_queued_diff.nr; i++) {
1087                         struct diff_filepair *p = diff_queued_diff.queue[i];
1088                         struct origin *norigin;
1089                         mmfile_t file_p;
1090                         struct blame_entry this[3];
1091
1092                         if (!DIFF_FILE_VALID(p->one))
1093                                 continue; /* does not exist in parent */
1094                         if (S_ISGITLINK(p->one->mode))
1095                                 continue; /* ignore git links */
1096                         if (porigin && !strcmp(p->one->path, porigin->path))
1097                                 /* find_move already dealt with this path */
1098                                 continue;
1099
1100                         norigin = get_origin(sb, parent, p->one->path);
1101                         hashcpy(norigin->blob_sha1, p->one->sha1);
1102                         fill_origin_blob(&sb->revs->diffopt, norigin, &file_p);
1103                         if (!file_p.ptr)
1104                                 continue;
1105
1106                         for (j = 0; j < num_ents; j++) {
1107                                 find_copy_in_blob(sb, blame_list[j].ent,
1108                                                   norigin, this, &file_p);
1109                                 copy_split_if_better(sb, blame_list[j].split,
1110                                                      this);
1111                                 decref_split(this);
1112                         }
1113                         origin_decref(norigin);
1114                 }
1115
1116                 for (j = 0; j < num_ents; j++) {
1117                         struct blame_entry *split = blame_list[j].split;
1118                         if (split[1].suspect &&
1119                             blame_copy_score < ent_score(sb, &split[1])) {
1120                                 split_blame(sb, split, blame_list[j].ent);
1121                                 made_progress = 1;
1122                         }
1123                         else
1124                                 blame_list[j].ent->scanned = 1;
1125                         decref_split(split);
1126                 }
1127                 free(blame_list);
1128
1129                 if (!made_progress)
1130                         break;
1131                 blame_list = setup_blame_list(sb, target, blame_copy_score, &num_ents);
1132                 if (!blame_list) {
1133                         retval = 1;
1134                         break;
1135                 }
1136         }
1137         reset_scanned_flag(sb);
1138         diff_flush(&diff_opts);
1139         diff_tree_release_paths(&diff_opts);
1140         return retval;
1141 }
1142
1143 /*
1144  * The blobs of origin and porigin exactly match, so everything
1145  * origin is suspected for can be blamed on the parent.
1146  */
1147 static void pass_whole_blame(struct scoreboard *sb,
1148                              struct origin *origin, struct origin *porigin)
1149 {
1150         struct blame_entry *e;
1151
1152         if (!porigin->file.ptr && origin->file.ptr) {
1153                 /* Steal its file */
1154                 porigin->file = origin->file;
1155                 origin->file.ptr = NULL;
1156         }
1157         for (e = sb->ent; e; e = e->next) {
1158                 if (!same_suspect(e->suspect, origin))
1159                         continue;
1160                 origin_incref(porigin);
1161                 origin_decref(e->suspect);
1162                 e->suspect = porigin;
1163         }
1164 }
1165
1166 /*
1167  * We pass blame from the current commit to its parents.  We keep saying
1168  * "parent" (and "porigin"), but what we mean is to find scapegoat to
1169  * exonerate ourselves.
1170  */
1171 static struct commit_list *first_scapegoat(struct rev_info *revs, struct commit *commit)
1172 {
1173         if (!reverse)
1174                 return commit->parents;
1175         return lookup_decoration(&revs->children, &commit->object);
1176 }
1177
1178 static int num_scapegoats(struct rev_info *revs, struct commit *commit)
1179 {
1180         int cnt;
1181         struct commit_list *l = first_scapegoat(revs, commit);
1182         for (cnt = 0; l; l = l->next)
1183                 cnt++;
1184         return cnt;
1185 }
1186
1187 #define MAXSG 16
1188
1189 static void pass_blame(struct scoreboard *sb, struct origin *origin, int opt)
1190 {
1191         struct rev_info *revs = sb->revs;
1192         int i, pass, num_sg;
1193         struct commit *commit = origin->commit;
1194         struct commit_list *sg;
1195         struct origin *sg_buf[MAXSG];
1196         struct origin *porigin, **sg_origin = sg_buf;
1197
1198         num_sg = num_scapegoats(revs, commit);
1199         if (!num_sg)
1200                 goto finish;
1201         else if (num_sg < ARRAY_SIZE(sg_buf))
1202                 memset(sg_buf, 0, sizeof(sg_buf));
1203         else
1204                 sg_origin = xcalloc(num_sg, sizeof(*sg_origin));
1205
1206         /*
1207          * The first pass looks for unrenamed path to optimize for
1208          * common cases, then we look for renames in the second pass.
1209          */
1210         for (pass = 0; pass < 2; pass++) {
1211                 struct origin *(*find)(struct scoreboard *,
1212                                        struct commit *, struct origin *);
1213                 find = pass ? find_rename : find_origin;
1214
1215                 for (i = 0, sg = first_scapegoat(revs, commit);
1216                      i < num_sg && sg;
1217                      sg = sg->next, i++) {
1218                         struct commit *p = sg->item;
1219                         int j, same;
1220
1221                         if (sg_origin[i])
1222                                 continue;
1223                         if (parse_commit(p))
1224                                 continue;
1225                         porigin = find(sb, p, origin);
1226                         if (!porigin)
1227                                 continue;
1228                         if (!hashcmp(porigin->blob_sha1, origin->blob_sha1)) {
1229                                 pass_whole_blame(sb, origin, porigin);
1230                                 origin_decref(porigin);
1231                                 goto finish;
1232                         }
1233                         for (j = same = 0; j < i; j++)
1234                                 if (sg_origin[j] &&
1235                                     !hashcmp(sg_origin[j]->blob_sha1,
1236                                              porigin->blob_sha1)) {
1237                                         same = 1;
1238                                         break;
1239                                 }
1240                         if (!same)
1241                                 sg_origin[i] = porigin;
1242                         else
1243                                 origin_decref(porigin);
1244                 }
1245         }
1246
1247         num_commits++;
1248         for (i = 0, sg = first_scapegoat(revs, commit);
1249              i < num_sg && sg;
1250              sg = sg->next, i++) {
1251                 struct origin *porigin = sg_origin[i];
1252                 if (!porigin)
1253                         continue;
1254                 if (!origin->previous) {
1255                         origin_incref(porigin);
1256                         origin->previous = porigin;
1257                 }
1258                 if (pass_blame_to_parent(sb, origin, porigin))
1259                         goto finish;
1260         }
1261
1262         /*
1263          * Optionally find moves in parents' files.
1264          */
1265         if (opt & PICKAXE_BLAME_MOVE)
1266                 for (i = 0, sg = first_scapegoat(revs, commit);
1267                      i < num_sg && sg;
1268                      sg = sg->next, i++) {
1269                         struct origin *porigin = sg_origin[i];
1270                         if (!porigin)
1271                                 continue;
1272                         if (find_move_in_parent(sb, origin, porigin))
1273                                 goto finish;
1274                 }
1275
1276         /*
1277          * Optionally find copies from parents' files.
1278          */
1279         if (opt & PICKAXE_BLAME_COPY)
1280                 for (i = 0, sg = first_scapegoat(revs, commit);
1281                      i < num_sg && sg;
1282                      sg = sg->next, i++) {
1283                         struct origin *porigin = sg_origin[i];
1284                         if (find_copy_in_parent(sb, origin, sg->item,
1285                                                 porigin, opt))
1286                                 goto finish;
1287                 }
1288
1289  finish:
1290         for (i = 0; i < num_sg; i++) {
1291                 if (sg_origin[i]) {
1292                         drop_origin_blob(sg_origin[i]);
1293                         origin_decref(sg_origin[i]);
1294                 }
1295         }
1296         drop_origin_blob(origin);
1297         if (sg_buf != sg_origin)
1298                 free(sg_origin);
1299 }
1300
1301 /*
1302  * Information on commits, used for output.
1303  */
1304 struct commit_info
1305 {
1306         const char *author;
1307         const char *author_mail;
1308         unsigned long author_time;
1309         const char *author_tz;
1310
1311         /* filled only when asked for details */
1312         const char *committer;
1313         const char *committer_mail;
1314         unsigned long committer_time;
1315         const char *committer_tz;
1316
1317         const char *summary;
1318 };
1319
1320 /*
1321  * Parse author/committer line in the commit object buffer
1322  */
1323 static void get_ac_line(const char *inbuf, const char *what,
1324                         int person_len, char *person,
1325                         int mail_len, char *mail,
1326                         unsigned long *time, const char **tz)
1327 {
1328         int len, tzlen, maillen;
1329         char *tmp, *endp, *timepos, *mailpos;
1330
1331         tmp = strstr(inbuf, what);
1332         if (!tmp)
1333                 goto error_out;
1334         tmp += strlen(what);
1335         endp = strchr(tmp, '\n');
1336         if (!endp)
1337                 len = strlen(tmp);
1338         else
1339                 len = endp - tmp;
1340         if (person_len <= len) {
1341         error_out:
1342                 /* Ugh */
1343                 *tz = "(unknown)";
1344                 strcpy(person, *tz);
1345                 strcpy(mail, *tz);
1346                 *time = 0;
1347                 return;
1348         }
1349         memcpy(person, tmp, len);
1350
1351         tmp = person;
1352         tmp += len;
1353         *tmp = 0;
1354         while (person < tmp && *tmp != ' ')
1355                 tmp--;
1356         if (tmp <= person)
1357                 goto error_out;
1358         *tz = tmp+1;
1359         tzlen = (person+len)-(tmp+1);
1360
1361         *tmp = 0;
1362         while (person < tmp && *tmp != ' ')
1363                 tmp--;
1364         if (tmp <= person)
1365                 goto error_out;
1366         *time = strtoul(tmp, NULL, 10);
1367         timepos = tmp;
1368
1369         *tmp = 0;
1370         while (person < tmp && *tmp != ' ')
1371                 tmp--;
1372         if (tmp <= person)
1373                 return;
1374         mailpos = tmp + 1;
1375         *tmp = 0;
1376         maillen = timepos - tmp;
1377         memcpy(mail, mailpos, maillen);
1378
1379         if (!mailmap.nr)
1380                 return;
1381
1382         /*
1383          * mailmap expansion may make the name longer.
1384          * make room by pushing stuff down.
1385          */
1386         tmp = person + person_len - (tzlen + 1);
1387         memmove(tmp, *tz, tzlen);
1388         tmp[tzlen] = 0;
1389         *tz = tmp;
1390
1391         /*
1392          * Now, convert both name and e-mail using mailmap
1393          */
1394         if (map_user(&mailmap, mail+1, mail_len-1, person, tmp-person-1)) {
1395                 /* Add a trailing '>' to email, since map_user returns plain emails
1396                    Note: It already has '<', since we replace from mail+1 */
1397                 mailpos = memchr(mail, '\0', mail_len);
1398                 if (mailpos && mailpos-mail < mail_len - 1) {
1399                         *mailpos = '>';
1400                         *(mailpos+1) = '\0';
1401                 }
1402         }
1403 }
1404
1405 static void get_commit_info(struct commit *commit,
1406                             struct commit_info *ret,
1407                             int detailed)
1408 {
1409         int len;
1410         const char *subject;
1411         char *reencoded, *message;
1412         static char author_name[1024];
1413         static char author_mail[1024];
1414         static char committer_name[1024];
1415         static char committer_mail[1024];
1416         static char summary_buf[1024];
1417
1418         /*
1419          * We've operated without save_commit_buffer, so
1420          * we now need to populate them for output.
1421          */
1422         if (!commit->buffer) {
1423                 enum object_type type;
1424                 unsigned long size;
1425                 commit->buffer =
1426                         read_sha1_file(commit->object.sha1, &type, &size);
1427                 if (!commit->buffer)
1428                         die("Cannot read commit %s",
1429                             sha1_to_hex(commit->object.sha1));
1430         }
1431         reencoded = reencode_commit_message(commit, NULL);
1432         message   = reencoded ? reencoded : commit->buffer;
1433         ret->author = author_name;
1434         ret->author_mail = author_mail;
1435         get_ac_line(message, "\nauthor ",
1436                     sizeof(author_name), author_name,
1437                     sizeof(author_mail), author_mail,
1438                     &ret->author_time, &ret->author_tz);
1439
1440         if (!detailed) {
1441                 free(reencoded);
1442                 return;
1443         }
1444
1445         ret->committer = committer_name;
1446         ret->committer_mail = committer_mail;
1447         get_ac_line(message, "\ncommitter ",
1448                     sizeof(committer_name), committer_name,
1449                     sizeof(committer_mail), committer_mail,
1450                     &ret->committer_time, &ret->committer_tz);
1451
1452         ret->summary = summary_buf;
1453         len = find_commit_subject(message, &subject);
1454         if (len && len < sizeof(summary_buf)) {
1455                 memcpy(summary_buf, subject, len);
1456                 summary_buf[len] = 0;
1457         } else {
1458                 sprintf(summary_buf, "(%s)", sha1_to_hex(commit->object.sha1));
1459         }
1460         free(reencoded);
1461 }
1462
1463 /*
1464  * To allow LF and other nonportable characters in pathnames,
1465  * they are c-style quoted as needed.
1466  */
1467 static void write_filename_info(const char *path)
1468 {
1469         printf("filename ");
1470         write_name_quoted(path, stdout, '\n');
1471 }
1472
1473 /*
1474  * Porcelain/Incremental format wants to show a lot of details per
1475  * commit.  Instead of repeating this every line, emit it only once,
1476  * the first time each commit appears in the output.
1477  */
1478 static int emit_one_suspect_detail(struct origin *suspect)
1479 {
1480         struct commit_info ci;
1481
1482         if (suspect->commit->object.flags & METAINFO_SHOWN)
1483                 return 0;
1484
1485         suspect->commit->object.flags |= METAINFO_SHOWN;
1486         get_commit_info(suspect->commit, &ci, 1);
1487         printf("author %s\n", ci.author);
1488         printf("author-mail %s\n", ci.author_mail);
1489         printf("author-time %lu\n", ci.author_time);
1490         printf("author-tz %s\n", ci.author_tz);
1491         printf("committer %s\n", ci.committer);
1492         printf("committer-mail %s\n", ci.committer_mail);
1493         printf("committer-time %lu\n", ci.committer_time);
1494         printf("committer-tz %s\n", ci.committer_tz);
1495         printf("summary %s\n", ci.summary);
1496         if (suspect->commit->object.flags & UNINTERESTING)
1497                 printf("boundary\n");
1498         if (suspect->previous) {
1499                 struct origin *prev = suspect->previous;
1500                 printf("previous %s ", sha1_to_hex(prev->commit->object.sha1));
1501                 write_name_quoted(prev->path, stdout, '\n');
1502         }
1503         return 1;
1504 }
1505
1506 /*
1507  * The blame_entry is found to be guilty for the range.  Mark it
1508  * as such, and show it in incremental output.
1509  */
1510 static void found_guilty_entry(struct blame_entry *ent)
1511 {
1512         if (ent->guilty)
1513                 return;
1514         ent->guilty = 1;
1515         if (incremental) {
1516                 struct origin *suspect = ent->suspect;
1517
1518                 printf("%s %d %d %d\n",
1519                        sha1_to_hex(suspect->commit->object.sha1),
1520                        ent->s_lno + 1, ent->lno + 1, ent->num_lines);
1521                 emit_one_suspect_detail(suspect);
1522                 write_filename_info(suspect->path);
1523                 maybe_flush_or_die(stdout, "stdout");
1524         }
1525 }
1526
1527 /*
1528  * The main loop -- while the scoreboard has lines whose true origin
1529  * is still unknown, pick one blame_entry, and allow its current
1530  * suspect to pass blames to its parents.
1531  */
1532 static void assign_blame(struct scoreboard *sb, int opt)
1533 {
1534         struct rev_info *revs = sb->revs;
1535
1536         while (1) {
1537                 struct blame_entry *ent;
1538                 struct commit *commit;
1539                 struct origin *suspect = NULL;
1540
1541                 /* find one suspect to break down */
1542                 for (ent = sb->ent; !suspect && ent; ent = ent->next)
1543                         if (!ent->guilty)
1544                                 suspect = ent->suspect;
1545                 if (!suspect)
1546                         return; /* all done */
1547
1548                 /*
1549                  * We will use this suspect later in the loop,
1550                  * so hold onto it in the meantime.
1551                  */
1552                 origin_incref(suspect);
1553                 commit = suspect->commit;
1554                 if (!commit->object.parsed)
1555                         parse_commit(commit);
1556                 if (reverse ||
1557                     (!(commit->object.flags & UNINTERESTING) &&
1558                      !(revs->max_age != -1 && commit->date < revs->max_age)))
1559                         pass_blame(sb, suspect, opt);
1560                 else {
1561                         commit->object.flags |= UNINTERESTING;
1562                         if (commit->object.parsed)
1563                                 mark_parents_uninteresting(commit);
1564                 }
1565                 /* treat root commit as boundary */
1566                 if (!commit->parents && !show_root)
1567                         commit->object.flags |= UNINTERESTING;
1568
1569                 /* Take responsibility for the remaining entries */
1570                 for (ent = sb->ent; ent; ent = ent->next)
1571                         if (same_suspect(ent->suspect, suspect))
1572                                 found_guilty_entry(ent);
1573                 origin_decref(suspect);
1574
1575                 if (DEBUG) /* sanity */
1576                         sanity_check_refcnt(sb);
1577         }
1578 }
1579
1580 static const char *format_time(unsigned long time, const char *tz_str,
1581                                int show_raw_time)
1582 {
1583         static char time_buf[128];
1584         const char *time_str;
1585         int time_len;
1586         int tz;
1587
1588         if (show_raw_time) {
1589                 sprintf(time_buf, "%lu %s", time, tz_str);
1590         }
1591         else {
1592                 tz = atoi(tz_str);
1593                 time_str = show_date(time, tz, blame_date_mode);
1594                 time_len = strlen(time_str);
1595                 memcpy(time_buf, time_str, time_len);
1596                 memset(time_buf + time_len, ' ', blame_date_width - time_len);
1597         }
1598         return time_buf;
1599 }
1600
1601 #define OUTPUT_ANNOTATE_COMPAT  001
1602 #define OUTPUT_LONG_OBJECT_NAME 002
1603 #define OUTPUT_RAW_TIMESTAMP    004
1604 #define OUTPUT_PORCELAIN        010
1605 #define OUTPUT_SHOW_NAME        020
1606 #define OUTPUT_SHOW_NUMBER      040
1607 #define OUTPUT_SHOW_SCORE      0100
1608 #define OUTPUT_NO_AUTHOR       0200
1609
1610 static void emit_porcelain(struct scoreboard *sb, struct blame_entry *ent)
1611 {
1612         int cnt;
1613         const char *cp;
1614         struct origin *suspect = ent->suspect;
1615         char hex[41];
1616
1617         strcpy(hex, sha1_to_hex(suspect->commit->object.sha1));
1618         printf("%s%c%d %d %d\n",
1619                hex,
1620                ent->guilty ? ' ' : '*', /* purely for debugging */
1621                ent->s_lno + 1,
1622                ent->lno + 1,
1623                ent->num_lines);
1624         if (emit_one_suspect_detail(suspect) ||
1625             (suspect->commit->object.flags & MORE_THAN_ONE_PATH))
1626                 write_filename_info(suspect->path);
1627
1628         cp = nth_line(sb, ent->lno);
1629         for (cnt = 0; cnt < ent->num_lines; cnt++) {
1630                 char ch;
1631                 if (cnt)
1632                         printf("%s %d %d\n", hex,
1633                                ent->s_lno + 1 + cnt,
1634                                ent->lno + 1 + cnt);
1635                 putchar('\t');
1636                 do {
1637                         ch = *cp++;
1638                         putchar(ch);
1639                 } while (ch != '\n' &&
1640                          cp < sb->final_buf + sb->final_buf_size);
1641         }
1642
1643         if (sb->final_buf_size && cp[-1] != '\n')
1644                 putchar('\n');
1645 }
1646
1647 static void emit_other(struct scoreboard *sb, struct blame_entry *ent, int opt)
1648 {
1649         int cnt;
1650         const char *cp;
1651         struct origin *suspect = ent->suspect;
1652         struct commit_info ci;
1653         char hex[41];
1654         int show_raw_time = !!(opt & OUTPUT_RAW_TIMESTAMP);
1655
1656         get_commit_info(suspect->commit, &ci, 1);
1657         strcpy(hex, sha1_to_hex(suspect->commit->object.sha1));
1658
1659         cp = nth_line(sb, ent->lno);
1660         for (cnt = 0; cnt < ent->num_lines; cnt++) {
1661                 char ch;
1662                 int length = (opt & OUTPUT_LONG_OBJECT_NAME) ? 40 : 8;
1663
1664                 if (suspect->commit->object.flags & UNINTERESTING) {
1665                         if (blank_boundary)
1666                                 memset(hex, ' ', length);
1667                         else if (!(opt & OUTPUT_ANNOTATE_COMPAT)) {
1668                                 length--;
1669                                 putchar('^');
1670                         }
1671                 }
1672
1673                 printf("%.*s", length, hex);
1674                 if (opt & OUTPUT_ANNOTATE_COMPAT)
1675                         printf("\t(%10s\t%10s\t%d)", ci.author,
1676                                format_time(ci.author_time, ci.author_tz,
1677                                            show_raw_time),
1678                                ent->lno + 1 + cnt);
1679                 else {
1680                         if (opt & OUTPUT_SHOW_SCORE)
1681                                 printf(" %*d %02d",
1682                                        max_score_digits, ent->score,
1683                                        ent->suspect->refcnt);
1684                         if (opt & OUTPUT_SHOW_NAME)
1685                                 printf(" %-*.*s", longest_file, longest_file,
1686                                        suspect->path);
1687                         if (opt & OUTPUT_SHOW_NUMBER)
1688                                 printf(" %*d", max_orig_digits,
1689                                        ent->s_lno + 1 + cnt);
1690
1691                         if (!(opt & OUTPUT_NO_AUTHOR)) {
1692                                 int pad = longest_author - utf8_strwidth(ci.author);
1693                                 printf(" (%s%*s %10s",
1694                                        ci.author, pad, "",
1695                                        format_time(ci.author_time,
1696                                                    ci.author_tz,
1697                                                    show_raw_time));
1698                         }
1699                         printf(" %*d) ",
1700                                max_digits, ent->lno + 1 + cnt);
1701                 }
1702                 do {
1703                         ch = *cp++;
1704                         putchar(ch);
1705                 } while (ch != '\n' &&
1706                          cp < sb->final_buf + sb->final_buf_size);
1707         }
1708
1709         if (sb->final_buf_size && cp[-1] != '\n')
1710                 putchar('\n');
1711 }
1712
1713 static void output(struct scoreboard *sb, int option)
1714 {
1715         struct blame_entry *ent;
1716
1717         if (option & OUTPUT_PORCELAIN) {
1718                 for (ent = sb->ent; ent; ent = ent->next) {
1719                         struct blame_entry *oth;
1720                         struct origin *suspect = ent->suspect;
1721                         struct commit *commit = suspect->commit;
1722                         if (commit->object.flags & MORE_THAN_ONE_PATH)
1723                                 continue;
1724                         for (oth = ent->next; oth; oth = oth->next) {
1725                                 if ((oth->suspect->commit != commit) ||
1726                                     !strcmp(oth->suspect->path, suspect->path))
1727                                         continue;
1728                                 commit->object.flags |= MORE_THAN_ONE_PATH;
1729                                 break;
1730                         }
1731                 }
1732         }
1733
1734         for (ent = sb->ent; ent; ent = ent->next) {
1735                 if (option & OUTPUT_PORCELAIN)
1736                         emit_porcelain(sb, ent);
1737                 else {
1738                         emit_other(sb, ent, option);
1739                 }
1740         }
1741 }
1742
1743 /*
1744  * To allow quick access to the contents of nth line in the
1745  * final image, prepare an index in the scoreboard.
1746  */
1747 static int prepare_lines(struct scoreboard *sb)
1748 {
1749         const char *buf = sb->final_buf;
1750         unsigned long len = sb->final_buf_size;
1751         int num = 0, incomplete = 0, bol = 1;
1752
1753         if (len && buf[len-1] != '\n')
1754                 incomplete++; /* incomplete line at the end */
1755         while (len--) {
1756                 if (bol) {
1757                         sb->lineno = xrealloc(sb->lineno,
1758                                               sizeof(int *) * (num + 1));
1759                         sb->lineno[num] = buf - sb->final_buf;
1760                         bol = 0;
1761                 }
1762                 if (*buf++ == '\n') {
1763                         num++;
1764                         bol = 1;
1765                 }
1766         }
1767         sb->lineno = xrealloc(sb->lineno,
1768                               sizeof(int *) * (num + incomplete + 1));
1769         sb->lineno[num + incomplete] = buf - sb->final_buf;
1770         sb->num_lines = num + incomplete;
1771         return sb->num_lines;
1772 }
1773
1774 /*
1775  * Add phony grafts for use with -S; this is primarily to
1776  * support git's cvsserver that wants to give a linear history
1777  * to its clients.
1778  */
1779 static int read_ancestry(const char *graft_file)
1780 {
1781         FILE *fp = fopen(graft_file, "r");
1782         char buf[1024];
1783         if (!fp)
1784                 return -1;
1785         while (fgets(buf, sizeof(buf), fp)) {
1786                 /* The format is just "Commit Parent1 Parent2 ...\n" */
1787                 int len = strlen(buf);
1788                 struct commit_graft *graft = read_graft_line(buf, len);
1789                 if (graft)
1790                         register_commit_graft(graft, 0);
1791         }
1792         fclose(fp);
1793         return 0;
1794 }
1795
1796 /*
1797  * How many columns do we need to show line numbers in decimal?
1798  */
1799 static int lineno_width(int lines)
1800 {
1801         int i, width;
1802
1803         for (width = 1, i = 10; i <= lines; width++)
1804                 i *= 10;
1805         return width;
1806 }
1807
1808 /*
1809  * How many columns do we need to show line numbers, authors,
1810  * and filenames?
1811  */
1812 static void find_alignment(struct scoreboard *sb, int *option)
1813 {
1814         int longest_src_lines = 0;
1815         int longest_dst_lines = 0;
1816         unsigned largest_score = 0;
1817         struct blame_entry *e;
1818
1819         for (e = sb->ent; e; e = e->next) {
1820                 struct origin *suspect = e->suspect;
1821                 struct commit_info ci;
1822                 int num;
1823
1824                 if (strcmp(suspect->path, sb->path))
1825                         *option |= OUTPUT_SHOW_NAME;
1826                 num = strlen(suspect->path);
1827                 if (longest_file < num)
1828                         longest_file = num;
1829                 if (!(suspect->commit->object.flags & METAINFO_SHOWN)) {
1830                         suspect->commit->object.flags |= METAINFO_SHOWN;
1831                         get_commit_info(suspect->commit, &ci, 1);
1832                         num = utf8_strwidth(ci.author);
1833                         if (longest_author < num)
1834                                 longest_author = num;
1835                 }
1836                 num = e->s_lno + e->num_lines;
1837                 if (longest_src_lines < num)
1838                         longest_src_lines = num;
1839                 num = e->lno + e->num_lines;
1840                 if (longest_dst_lines < num)
1841                         longest_dst_lines = num;
1842                 if (largest_score < ent_score(sb, e))
1843                         largest_score = ent_score(sb, e);
1844         }
1845         max_orig_digits = lineno_width(longest_src_lines);
1846         max_digits = lineno_width(longest_dst_lines);
1847         max_score_digits = lineno_width(largest_score);
1848 }
1849
1850 /*
1851  * For debugging -- origin is refcounted, and this asserts that
1852  * we do not underflow.
1853  */
1854 static void sanity_check_refcnt(struct scoreboard *sb)
1855 {
1856         int baa = 0;
1857         struct blame_entry *ent;
1858
1859         for (ent = sb->ent; ent; ent = ent->next) {
1860                 /* Nobody should have zero or negative refcnt */
1861                 if (ent->suspect->refcnt <= 0) {
1862                         fprintf(stderr, "%s in %s has negative refcnt %d\n",
1863                                 ent->suspect->path,
1864                                 sha1_to_hex(ent->suspect->commit->object.sha1),
1865                                 ent->suspect->refcnt);
1866                         baa = 1;
1867                 }
1868         }
1869         if (baa) {
1870                 int opt = 0160;
1871                 find_alignment(sb, &opt);
1872                 output(sb, opt);
1873                 die("Baa %d!", baa);
1874         }
1875 }
1876
1877 /*
1878  * Used for the command line parsing; check if the path exists
1879  * in the working tree.
1880  */
1881 static int has_string_in_work_tree(const char *path)
1882 {
1883         struct stat st;
1884         return !lstat(path, &st);
1885 }
1886
1887 static unsigned parse_score(const char *arg)
1888 {
1889         char *end;
1890         unsigned long score = strtoul(arg, &end, 10);
1891         if (*end)
1892                 return 0;
1893         return score;
1894 }
1895
1896 static const char *add_prefix(const char *prefix, const char *path)
1897 {
1898         return prefix_path(prefix, prefix ? strlen(prefix) : 0, path);
1899 }
1900
1901 /*
1902  * Parsing of (comma separated) one item in the -L option
1903  */
1904 static const char *parse_loc(const char *spec,
1905                              struct scoreboard *sb, long lno,
1906                              long begin, long *ret)
1907 {
1908         char *term;
1909         const char *line;
1910         long num;
1911         int reg_error;
1912         regex_t regexp;
1913         regmatch_t match[1];
1914
1915         /* Allow "-L <something>,+20" to mean starting at <something>
1916          * for 20 lines, or "-L <something>,-5" for 5 lines ending at
1917          * <something>.
1918          */
1919         if (1 < begin && (spec[0] == '+' || spec[0] == '-')) {
1920                 num = strtol(spec + 1, &term, 10);
1921                 if (term != spec + 1) {
1922                         if (spec[0] == '-')
1923                                 num = 0 - num;
1924                         if (0 < num)
1925                                 *ret = begin + num - 2;
1926                         else if (!num)
1927                                 *ret = begin;
1928                         else
1929                                 *ret = begin + num;
1930                         return term;
1931                 }
1932                 return spec;
1933         }
1934         num = strtol(spec, &term, 10);
1935         if (term != spec) {
1936                 *ret = num;
1937                 return term;
1938         }
1939         if (spec[0] != '/')
1940                 return spec;
1941
1942         /* it could be a regexp of form /.../ */
1943         for (term = (char *) spec + 1; *term && *term != '/'; term++) {
1944                 if (*term == '\\')
1945                         term++;
1946         }
1947         if (*term != '/')
1948                 return spec;
1949
1950         /* try [spec+1 .. term-1] as regexp */
1951         *term = 0;
1952         begin--; /* input is in human terms */
1953         line = nth_line(sb, begin);
1954
1955         if (!(reg_error = regcomp(&regexp, spec + 1, REG_NEWLINE)) &&
1956             !(reg_error = regexec(&regexp, line, 1, match, 0))) {
1957                 const char *cp = line + match[0].rm_so;
1958                 const char *nline;
1959
1960                 while (begin++ < lno) {
1961                         nline = nth_line(sb, begin);
1962                         if (line <= cp && cp < nline)
1963                                 break;
1964                         line = nline;
1965                 }
1966                 *ret = begin;
1967                 regfree(&regexp);
1968                 *term++ = '/';
1969                 return term;
1970         }
1971         else {
1972                 char errbuf[1024];
1973                 regerror(reg_error, &regexp, errbuf, 1024);
1974                 die("-L parameter '%s': %s", spec + 1, errbuf);
1975         }
1976 }
1977
1978 /*
1979  * Parsing of -L option
1980  */
1981 static void prepare_blame_range(struct scoreboard *sb,
1982                                 const char *bottomtop,
1983                                 long lno,
1984                                 long *bottom, long *top)
1985 {
1986         const char *term;
1987
1988         term = parse_loc(bottomtop, sb, lno, 1, bottom);
1989         if (*term == ',') {
1990                 term = parse_loc(term + 1, sb, lno, *bottom + 1, top);
1991                 if (*term)
1992                         usage(blame_usage);
1993         }
1994         if (*term)
1995                 usage(blame_usage);
1996 }
1997
1998 static int git_blame_config(const char *var, const char *value, void *cb)
1999 {
2000         if (!strcmp(var, "blame.showroot")) {
2001                 show_root = git_config_bool(var, value);
2002                 return 0;
2003         }
2004         if (!strcmp(var, "blame.blankboundary")) {
2005                 blank_boundary = git_config_bool(var, value);
2006                 return 0;
2007         }
2008         if (!strcmp(var, "blame.date")) {
2009                 if (!value)
2010                         return config_error_nonbool(var);
2011                 blame_date_mode = parse_date_format(value);
2012                 return 0;
2013         }
2014
2015         switch (userdiff_config(var, value)) {
2016         case 0:
2017                 break;
2018         case -1:
2019                 return -1;
2020         default:
2021                 return 0;
2022         }
2023
2024         return git_default_config(var, value, cb);
2025 }
2026
2027 /*
2028  * Prepare a dummy commit that represents the work tree (or staged) item.
2029  * Note that annotating work tree item never works in the reverse.
2030  */
2031 static struct commit *fake_working_tree_commit(struct diff_options *opt,
2032                                                const char *path,
2033                                                const char *contents_from)
2034 {
2035         struct commit *commit;
2036         struct origin *origin;
2037         unsigned char head_sha1[20];
2038         struct strbuf buf = STRBUF_INIT;
2039         const char *ident;
2040         time_t now;
2041         int size, len;
2042         struct cache_entry *ce;
2043         unsigned mode;
2044
2045         if (get_sha1("HEAD", head_sha1))
2046                 die("No such ref: HEAD");
2047
2048         time(&now);
2049         commit = xcalloc(1, sizeof(*commit));
2050         commit->parents = xcalloc(1, sizeof(*commit->parents));
2051         commit->parents->item = lookup_commit_reference(head_sha1);
2052         commit->object.parsed = 1;
2053         commit->date = now;
2054         commit->object.type = OBJ_COMMIT;
2055
2056         origin = make_origin(commit, path);
2057
2058         if (!contents_from || strcmp("-", contents_from)) {
2059                 struct stat st;
2060                 const char *read_from;
2061                 unsigned long buf_len;
2062
2063                 if (contents_from) {
2064                         if (stat(contents_from, &st) < 0)
2065                                 die_errno("Cannot stat '%s'", contents_from);
2066                         read_from = contents_from;
2067                 }
2068                 else {
2069                         if (lstat(path, &st) < 0)
2070                                 die_errno("Cannot lstat '%s'", path);
2071                         read_from = path;
2072                 }
2073                 mode = canon_mode(st.st_mode);
2074
2075                 switch (st.st_mode & S_IFMT) {
2076                 case S_IFREG:
2077                         if (DIFF_OPT_TST(opt, ALLOW_TEXTCONV) &&
2078                             textconv_object(read_from, null_sha1, &buf.buf, &buf_len))
2079                                 buf.len = buf_len;
2080                         else if (strbuf_read_file(&buf, read_from, st.st_size) != st.st_size)
2081                                 die_errno("cannot open or read '%s'", read_from);
2082                         break;
2083                 case S_IFLNK:
2084                         if (strbuf_readlink(&buf, read_from, st.st_size) < 0)
2085                                 die_errno("cannot readlink '%s'", read_from);
2086                         break;
2087                 default:
2088                         die("unsupported file type %s", read_from);
2089                 }
2090         }
2091         else {
2092                 /* Reading from stdin */
2093                 contents_from = "standard input";
2094                 mode = 0;
2095                 if (strbuf_read(&buf, 0, 0) < 0)
2096                         die_errno("failed to read from stdin");
2097         }
2098         convert_to_git(path, buf.buf, buf.len, &buf, 0);
2099         origin->file.ptr = buf.buf;
2100         origin->file.size = buf.len;
2101         pretend_sha1_file(buf.buf, buf.len, OBJ_BLOB, origin->blob_sha1);
2102         commit->util = origin;
2103
2104         /*
2105          * Read the current index, replace the path entry with
2106          * origin->blob_sha1 without mucking with its mode or type
2107          * bits; we are not going to write this index out -- we just
2108          * want to run "diff-index --cached".
2109          */
2110         discard_cache();
2111         read_cache();
2112
2113         len = strlen(path);
2114         if (!mode) {
2115                 int pos = cache_name_pos(path, len);
2116                 if (0 <= pos)
2117                         mode = active_cache[pos]->ce_mode;
2118                 else
2119                         /* Let's not bother reading from HEAD tree */
2120                         mode = S_IFREG | 0644;
2121         }
2122         size = cache_entry_size(len);
2123         ce = xcalloc(1, size);
2124         hashcpy(ce->sha1, origin->blob_sha1);
2125         memcpy(ce->name, path, len);
2126         ce->ce_flags = create_ce_flags(len, 0);
2127         ce->ce_mode = create_ce_mode(mode);
2128         add_cache_entry(ce, ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE);
2129
2130         /*
2131          * We are not going to write this out, so this does not matter
2132          * right now, but someday we might optimize diff-index --cached
2133          * with cache-tree information.
2134          */
2135         cache_tree_invalidate_path(active_cache_tree, path);
2136
2137         commit->buffer = xmalloc(400);
2138         ident = fmt_ident("Not Committed Yet", "not.committed.yet", NULL, 0);
2139         snprintf(commit->buffer, 400,
2140                 "tree 0000000000000000000000000000000000000000\n"
2141                 "parent %s\n"
2142                 "author %s\n"
2143                 "committer %s\n\n"
2144                 "Version of %s from %s\n",
2145                 sha1_to_hex(head_sha1),
2146                 ident, ident, path, contents_from ? contents_from : path);
2147         return commit;
2148 }
2149
2150 static const char *prepare_final(struct scoreboard *sb)
2151 {
2152         int i;
2153         const char *final_commit_name = NULL;
2154         struct rev_info *revs = sb->revs;
2155
2156         /*
2157          * There must be one and only one positive commit in the
2158          * revs->pending array.
2159          */
2160         for (i = 0; i < revs->pending.nr; i++) {
2161                 struct object *obj = revs->pending.objects[i].item;
2162                 if (obj->flags & UNINTERESTING)
2163                         continue;
2164                 while (obj->type == OBJ_TAG)
2165                         obj = deref_tag(obj, NULL, 0);
2166                 if (obj->type != OBJ_COMMIT)
2167                         die("Non commit %s?", revs->pending.objects[i].name);
2168                 if (sb->final)
2169                         die("More than one commit to dig from %s and %s?",
2170                             revs->pending.objects[i].name,
2171                             final_commit_name);
2172                 sb->final = (struct commit *) obj;
2173                 final_commit_name = revs->pending.objects[i].name;
2174         }
2175         return final_commit_name;
2176 }
2177
2178 static const char *prepare_initial(struct scoreboard *sb)
2179 {
2180         int i;
2181         const char *final_commit_name = NULL;
2182         struct rev_info *revs = sb->revs;
2183
2184         /*
2185          * There must be one and only one negative commit, and it must be
2186          * the boundary.
2187          */
2188         for (i = 0; i < revs->pending.nr; i++) {
2189                 struct object *obj = revs->pending.objects[i].item;
2190                 if (!(obj->flags & UNINTERESTING))
2191                         continue;
2192                 while (obj->type == OBJ_TAG)
2193                         obj = deref_tag(obj, NULL, 0);
2194                 if (obj->type != OBJ_COMMIT)
2195                         die("Non commit %s?", revs->pending.objects[i].name);
2196                 if (sb->final)
2197                         die("More than one commit to dig down to %s and %s?",
2198                             revs->pending.objects[i].name,
2199                             final_commit_name);
2200                 sb->final = (struct commit *) obj;
2201                 final_commit_name = revs->pending.objects[i].name;
2202         }
2203         if (!final_commit_name)
2204                 die("No commit to dig down to?");
2205         return final_commit_name;
2206 }
2207
2208 static int blame_copy_callback(const struct option *option, const char *arg, int unset)
2209 {
2210         int *opt = option->value;
2211
2212         /*
2213          * -C enables copy from removed files;
2214          * -C -C enables copy from existing files, but only
2215          *       when blaming a new file;
2216          * -C -C -C enables copy from existing files for
2217          *          everybody
2218          */
2219         if (*opt & PICKAXE_BLAME_COPY_HARDER)
2220                 *opt |= PICKAXE_BLAME_COPY_HARDEST;
2221         if (*opt & PICKAXE_BLAME_COPY)
2222                 *opt |= PICKAXE_BLAME_COPY_HARDER;
2223         *opt |= PICKAXE_BLAME_COPY | PICKAXE_BLAME_MOVE;
2224
2225         if (arg)
2226                 blame_copy_score = parse_score(arg);
2227         return 0;
2228 }
2229
2230 static int blame_move_callback(const struct option *option, const char *arg, int unset)
2231 {
2232         int *opt = option->value;
2233
2234         *opt |= PICKAXE_BLAME_MOVE;
2235
2236         if (arg)
2237                 blame_move_score = parse_score(arg);
2238         return 0;
2239 }
2240
2241 static int blame_bottomtop_callback(const struct option *option, const char *arg, int unset)
2242 {
2243         const char **bottomtop = option->value;
2244         if (!arg)
2245                 return -1;
2246         if (*bottomtop)
2247                 die("More than one '-L n,m' option given");
2248         *bottomtop = arg;
2249         return 0;
2250 }
2251
2252 int cmd_blame(int argc, const char **argv, const char *prefix)
2253 {
2254         struct rev_info revs;
2255         const char *path;
2256         struct scoreboard sb;
2257         struct origin *o;
2258         struct blame_entry *ent;
2259         long dashdash_pos, bottom, top, lno;
2260         const char *final_commit_name = NULL;
2261         enum object_type type;
2262
2263         static const char *bottomtop = NULL;
2264         static int output_option = 0, opt = 0;
2265         static int show_stats = 0;
2266         static const char *revs_file = NULL;
2267         static const char *contents_from = NULL;
2268         static const struct option options[] = {
2269                 OPT_BOOLEAN(0, "incremental", &incremental, "Show blame entries as we find them, incrementally"),
2270                 OPT_BOOLEAN('b', NULL, &blank_boundary, "Show blank SHA-1 for boundary commits (Default: off)"),
2271                 OPT_BOOLEAN(0, "root", &show_root, "Do not treat root commits as boundaries (Default: off)"),
2272                 OPT_BOOLEAN(0, "show-stats", &show_stats, "Show work cost statistics"),
2273                 OPT_BIT(0, "score-debug", &output_option, "Show output score for blame entries", OUTPUT_SHOW_SCORE),
2274                 OPT_BIT('f', "show-name", &output_option, "Show original filename (Default: auto)", OUTPUT_SHOW_NAME),
2275                 OPT_BIT('n', "show-number", &output_option, "Show original linenumber (Default: off)", OUTPUT_SHOW_NUMBER),
2276                 OPT_BIT('p', "porcelain", &output_option, "Show in a format designed for machine consumption", OUTPUT_PORCELAIN),
2277                 OPT_BIT('c', NULL, &output_option, "Use the same output mode as git-annotate (Default: off)", OUTPUT_ANNOTATE_COMPAT),
2278                 OPT_BIT('t', NULL, &output_option, "Show raw timestamp (Default: off)", OUTPUT_RAW_TIMESTAMP),
2279                 OPT_BIT('l', NULL, &output_option, "Show long commit SHA1 (Default: off)", OUTPUT_LONG_OBJECT_NAME),
2280                 OPT_BIT('s', NULL, &output_option, "Suppress author name and timestamp (Default: off)", OUTPUT_NO_AUTHOR),
2281                 OPT_BIT('w', NULL, &xdl_opts, "Ignore whitespace differences", XDF_IGNORE_WHITESPACE),
2282                 OPT_STRING('S', NULL, &revs_file, "file", "Use revisions from <file> instead of calling git-rev-list"),
2283                 OPT_STRING(0, "contents", &contents_from, "file", "Use <file>'s contents as the final image"),
2284                 { OPTION_CALLBACK, 'C', NULL, &opt, "score", "Find line copies within and across files", PARSE_OPT_OPTARG, blame_copy_callback },
2285                 { OPTION_CALLBACK, 'M', NULL, &opt, "score", "Find line movements within and across files", PARSE_OPT_OPTARG, blame_move_callback },
2286                 OPT_CALLBACK('L', NULL, &bottomtop, "n,m", "Process only line range n,m, counting from 1", blame_bottomtop_callback),
2287                 OPT_END()
2288         };
2289
2290         struct parse_opt_ctx_t ctx;
2291         int cmd_is_annotate = !strcmp(argv[0], "annotate");
2292
2293         git_config(git_blame_config, NULL);
2294         init_revisions(&revs, NULL);
2295         revs.date_mode = blame_date_mode;
2296         DIFF_OPT_SET(&revs.diffopt, ALLOW_TEXTCONV);
2297
2298         save_commit_buffer = 0;
2299         dashdash_pos = 0;
2300
2301         parse_options_start(&ctx, argc, argv, prefix, PARSE_OPT_KEEP_DASHDASH |
2302                             PARSE_OPT_KEEP_ARGV0);
2303         for (;;) {
2304                 switch (parse_options_step(&ctx, options, blame_opt_usage)) {
2305                 case PARSE_OPT_HELP:
2306                         exit(129);
2307                 case PARSE_OPT_DONE:
2308                         if (ctx.argv[0])
2309                                 dashdash_pos = ctx.cpidx;
2310                         goto parse_done;
2311                 }
2312
2313                 if (!strcmp(ctx.argv[0], "--reverse")) {
2314                         ctx.argv[0] = "--children";
2315                         reverse = 1;
2316                 }
2317                 parse_revision_opt(&revs, &ctx, options, blame_opt_usage);
2318         }
2319 parse_done:
2320         argc = parse_options_end(&ctx);
2321
2322         if (revs_file && read_ancestry(revs_file))
2323                 die_errno("reading graft file '%s' failed", revs_file);
2324
2325         if (cmd_is_annotate) {
2326                 output_option |= OUTPUT_ANNOTATE_COMPAT;
2327                 blame_date_mode = DATE_ISO8601;
2328         } else {
2329                 blame_date_mode = revs.date_mode;
2330         }
2331
2332         /* The maximum width used to show the dates */
2333         switch (blame_date_mode) {
2334         case DATE_RFC2822:
2335                 blame_date_width = sizeof("Thu, 19 Oct 2006 16:00:04 -0700");
2336                 break;
2337         case DATE_ISO8601:
2338                 blame_date_width = sizeof("2006-10-19 16:00:04 -0700");
2339                 break;
2340         case DATE_RAW:
2341                 blame_date_width = sizeof("1161298804 -0700");
2342                 break;
2343         case DATE_SHORT:
2344                 blame_date_width = sizeof("2006-10-19");
2345                 break;
2346         case DATE_RELATIVE:
2347                 /* "normal" is used as the fallback for "relative" */
2348         case DATE_LOCAL:
2349         case DATE_NORMAL:
2350                 blame_date_width = sizeof("Thu Oct 19 16:00:04 2006 -0700");
2351                 break;
2352         }
2353         blame_date_width -= 1; /* strip the null */
2354
2355         if (DIFF_OPT_TST(&revs.diffopt, FIND_COPIES_HARDER))
2356                 opt |= (PICKAXE_BLAME_COPY | PICKAXE_BLAME_MOVE |
2357                         PICKAXE_BLAME_COPY_HARDER);
2358
2359         if (!blame_move_score)
2360                 blame_move_score = BLAME_DEFAULT_MOVE_SCORE;
2361         if (!blame_copy_score)
2362                 blame_copy_score = BLAME_DEFAULT_COPY_SCORE;
2363
2364         /*
2365          * We have collected options unknown to us in argv[1..unk]
2366          * which are to be passed to revision machinery if we are
2367          * going to do the "bottom" processing.
2368          *
2369          * The remaining are:
2370          *
2371          * (1) if dashdash_pos != 0, it is either
2372          *     "blame [revisions] -- <path>" or
2373          *     "blame -- <path> <rev>"
2374          *
2375          * (2) otherwise, it is one of the two:
2376          *     "blame [revisions] <path>"
2377          *     "blame <path> <rev>"
2378          *
2379          * Note that we must strip out <path> from the arguments: we do not
2380          * want the path pruning but we may want "bottom" processing.
2381          */
2382         if (dashdash_pos) {
2383                 switch (argc - dashdash_pos - 1) {
2384                 case 2: /* (1b) */
2385                         if (argc != 4)
2386                                 usage_with_options(blame_opt_usage, options);
2387                         /* reorder for the new way: <rev> -- <path> */
2388                         argv[1] = argv[3];
2389                         argv[3] = argv[2];
2390                         argv[2] = "--";
2391                         /* FALLTHROUGH */
2392                 case 1: /* (1a) */
2393                         path = add_prefix(prefix, argv[--argc]);
2394                         argv[argc] = NULL;
2395                         break;
2396                 default:
2397                         usage_with_options(blame_opt_usage, options);
2398                 }
2399         } else {
2400                 if (argc < 2)
2401                         usage_with_options(blame_opt_usage, options);
2402                 path = add_prefix(prefix, argv[argc - 1]);
2403                 if (argc == 3 && !has_string_in_work_tree(path)) { /* (2b) */
2404                         path = add_prefix(prefix, argv[1]);
2405                         argv[1] = argv[2];
2406                 }
2407                 argv[argc - 1] = "--";
2408
2409                 setup_work_tree();
2410                 if (!has_string_in_work_tree(path))
2411                         die_errno("cannot stat path '%s'", path);
2412         }
2413
2414         revs.disable_stdin = 1;
2415         setup_revisions(argc, argv, &revs, NULL);
2416         memset(&sb, 0, sizeof(sb));
2417
2418         sb.revs = &revs;
2419         if (!reverse)
2420                 final_commit_name = prepare_final(&sb);
2421         else if (contents_from)
2422                 die("--contents and --children do not blend well.");
2423         else
2424                 final_commit_name = prepare_initial(&sb);
2425
2426         if (!sb.final) {
2427                 /*
2428                  * "--not A B -- path" without anything positive;
2429                  * do not default to HEAD, but use the working tree
2430                  * or "--contents".
2431                  */
2432                 setup_work_tree();
2433                 sb.final = fake_working_tree_commit(&sb.revs->diffopt,
2434                                                     path, contents_from);
2435                 add_pending_object(&revs, &(sb.final->object), ":");
2436         }
2437         else if (contents_from)
2438                 die("Cannot use --contents with final commit object name");
2439
2440         /*
2441          * If we have bottom, this will mark the ancestors of the
2442          * bottom commits we would reach while traversing as
2443          * uninteresting.
2444          */
2445         if (prepare_revision_walk(&revs))
2446                 die("revision walk setup failed");
2447
2448         if (is_null_sha1(sb.final->object.sha1)) {
2449                 char *buf;
2450                 o = sb.final->util;
2451                 buf = xmalloc(o->file.size + 1);
2452                 memcpy(buf, o->file.ptr, o->file.size + 1);
2453                 sb.final_buf = buf;
2454                 sb.final_buf_size = o->file.size;
2455         }
2456         else {
2457                 o = get_origin(&sb, sb.final, path);
2458                 if (fill_blob_sha1(o))
2459                         die("no such path %s in %s", path, final_commit_name);
2460
2461                 if (DIFF_OPT_TST(&sb.revs->diffopt, ALLOW_TEXTCONV) &&
2462                     textconv_object(path, o->blob_sha1, (char **) &sb.final_buf,
2463                                     &sb.final_buf_size))
2464                         ;
2465                 else
2466                         sb.final_buf = read_sha1_file(o->blob_sha1, &type,
2467                                                       &sb.final_buf_size);
2468
2469                 if (!sb.final_buf)
2470                         die("Cannot read blob %s for path %s",
2471                             sha1_to_hex(o->blob_sha1),
2472                             path);
2473         }
2474         num_read_blob++;
2475         lno = prepare_lines(&sb);
2476
2477         bottom = top = 0;
2478         if (bottomtop)
2479                 prepare_blame_range(&sb, bottomtop, lno, &bottom, &top);
2480         if (bottom && top && top < bottom) {
2481                 long tmp;
2482                 tmp = top; top = bottom; bottom = tmp;
2483         }
2484         if (bottom < 1)
2485                 bottom = 1;
2486         if (top < 1)
2487                 top = lno;
2488         bottom--;
2489         if (lno < top || lno < bottom)
2490                 die("file %s has only %lu lines", path, lno);
2491
2492         ent = xcalloc(1, sizeof(*ent));
2493         ent->lno = bottom;
2494         ent->num_lines = top - bottom;
2495         ent->suspect = o;
2496         ent->s_lno = bottom;
2497
2498         sb.ent = ent;
2499         sb.path = path;
2500
2501         read_mailmap(&mailmap, NULL);
2502
2503         if (!incremental)
2504                 setup_pager();
2505
2506         assign_blame(&sb, opt);
2507
2508         if (incremental)
2509                 return 0;
2510
2511         coalesce(&sb);
2512
2513         if (!(output_option & OUTPUT_PORCELAIN))
2514                 find_alignment(&sb, &output_option);
2515
2516         output(&sb, output_option);
2517         free((void *)sb.final_buf);
2518         for (ent = sb.ent; ent; ) {
2519                 struct blame_entry *e = ent->next;
2520                 free(ent);
2521                 ent = e;
2522         }
2523
2524         if (show_stats) {
2525                 printf("num read blob: %d\n", num_read_blob);
2526                 printf("num get patch: %d\n", num_get_patch);
2527                 printf("num commits: %d\n", num_commits);
2528         }
2529         return 0;
2530 }