Merge branch 'db/delta-applier' into db/text-delta
[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         unsigned mode;
87         char path[FLEX_ARRAY];
88 };
89
90 /*
91  * Prepare diff_filespec and convert it using diff textconv API
92  * if the textconv driver exists.
93  * Return 1 if the conversion succeeds, 0 otherwise.
94  */
95 int textconv_object(const char *path,
96                     unsigned mode,
97                     const unsigned char *sha1,
98                     char **buf,
99                     unsigned long *buf_size)
100 {
101         struct diff_filespec *df;
102         struct userdiff_driver *textconv;
103
104         df = alloc_filespec(path);
105         fill_filespec(df, sha1, mode);
106         textconv = get_textconv(df);
107         if (!textconv) {
108                 free_filespec(df);
109                 return 0;
110         }
111
112         *buf_size = fill_textconv(textconv, df, buf);
113         free_filespec(df);
114         return 1;
115 }
116
117 /*
118  * Given an origin, prepare mmfile_t structure to be used by the
119  * diff machinery
120  */
121 static void fill_origin_blob(struct diff_options *opt,
122                              struct origin *o, mmfile_t *file)
123 {
124         if (!o->file.ptr) {
125                 enum object_type type;
126                 unsigned long file_size;
127
128                 num_read_blob++;
129                 if (DIFF_OPT_TST(opt, ALLOW_TEXTCONV) &&
130                     textconv_object(o->path, o->mode, o->blob_sha1, &file->ptr, &file_size))
131                         ;
132                 else
133                         file->ptr = read_sha1_file(o->blob_sha1, &type, &file_size);
134                 file->size = file_size;
135
136                 if (!file->ptr)
137                         die("Cannot read blob %s for path %s",
138                             sha1_to_hex(o->blob_sha1),
139                             o->path);
140                 o->file = *file;
141         }
142         else
143                 *file = o->file;
144 }
145
146 /*
147  * Origin is refcounted and usually we keep the blob contents to be
148  * reused.
149  */
150 static inline struct origin *origin_incref(struct origin *o)
151 {
152         if (o)
153                 o->refcnt++;
154         return o;
155 }
156
157 static void origin_decref(struct origin *o)
158 {
159         if (o && --o->refcnt <= 0) {
160                 if (o->previous)
161                         origin_decref(o->previous);
162                 free(o->file.ptr);
163                 free(o);
164         }
165 }
166
167 static void drop_origin_blob(struct origin *o)
168 {
169         if (o->file.ptr) {
170                 free(o->file.ptr);
171                 o->file.ptr = NULL;
172         }
173 }
174
175 /*
176  * Each group of lines is described by a blame_entry; it can be split
177  * as we pass blame to the parents.  They form a linked list in the
178  * scoreboard structure, sorted by the target line number.
179  */
180 struct blame_entry {
181         struct blame_entry *prev;
182         struct blame_entry *next;
183
184         /* the first line of this group in the final image;
185          * internally all line numbers are 0 based.
186          */
187         int lno;
188
189         /* how many lines this group has */
190         int num_lines;
191
192         /* the commit that introduced this group into the final image */
193         struct origin *suspect;
194
195         /* true if the suspect is truly guilty; false while we have not
196          * checked if the group came from one of its parents.
197          */
198         char guilty;
199
200         /* true if the entry has been scanned for copies in the current parent
201          */
202         char scanned;
203
204         /* the line number of the first line of this group in the
205          * suspect's file; internally all line numbers are 0 based.
206          */
207         int s_lno;
208
209         /* how significant this entry is -- cached to avoid
210          * scanning the lines over and over.
211          */
212         unsigned score;
213 };
214
215 /*
216  * The current state of the blame assignment.
217  */
218 struct scoreboard {
219         /* the final commit (i.e. where we started digging from) */
220         struct commit *final;
221         struct rev_info *revs;
222         const char *path;
223
224         /*
225          * The contents in the final image.
226          * Used by many functions to obtain contents of the nth line,
227          * indexed with scoreboard.lineno[blame_entry.lno].
228          */
229         const char *final_buf;
230         unsigned long final_buf_size;
231
232         /* linked list of blames */
233         struct blame_entry *ent;
234
235         /* look-up a line in the final buffer */
236         int num_lines;
237         int *lineno;
238 };
239
240 static inline int same_suspect(struct origin *a, struct origin *b)
241 {
242         if (a == b)
243                 return 1;
244         if (a->commit != b->commit)
245                 return 0;
246         return !strcmp(a->path, b->path);
247 }
248
249 static void sanity_check_refcnt(struct scoreboard *);
250
251 /*
252  * If two blame entries that are next to each other came from
253  * contiguous lines in the same origin (i.e. <commit, path> pair),
254  * merge them together.
255  */
256 static void coalesce(struct scoreboard *sb)
257 {
258         struct blame_entry *ent, *next;
259
260         for (ent = sb->ent; ent && (next = ent->next); ent = next) {
261                 if (same_suspect(ent->suspect, next->suspect) &&
262                     ent->guilty == next->guilty &&
263                     ent->s_lno + ent->num_lines == next->s_lno) {
264                         ent->num_lines += next->num_lines;
265                         ent->next = next->next;
266                         if (ent->next)
267                                 ent->next->prev = ent;
268                         origin_decref(next->suspect);
269                         free(next);
270                         ent->score = 0;
271                         next = ent; /* again */
272                 }
273         }
274
275         if (DEBUG) /* sanity */
276                 sanity_check_refcnt(sb);
277 }
278
279 /*
280  * Given a commit and a path in it, create a new origin structure.
281  * The callers that add blame to the scoreboard should use
282  * get_origin() to obtain shared, refcounted copy instead of calling
283  * this function directly.
284  */
285 static struct origin *make_origin(struct commit *commit, const char *path)
286 {
287         struct origin *o;
288         o = xcalloc(1, sizeof(*o) + strlen(path) + 1);
289         o->commit = commit;
290         o->refcnt = 1;
291         strcpy(o->path, path);
292         return o;
293 }
294
295 /*
296  * Locate an existing origin or create a new one.
297  */
298 static struct origin *get_origin(struct scoreboard *sb,
299                                  struct commit *commit,
300                                  const char *path)
301 {
302         struct blame_entry *e;
303
304         for (e = sb->ent; e; e = e->next) {
305                 if (e->suspect->commit == commit &&
306                     !strcmp(e->suspect->path, path))
307                         return origin_incref(e->suspect);
308         }
309         return make_origin(commit, path);
310 }
311
312 /*
313  * Fill the blob_sha1 field of an origin if it hasn't, so that later
314  * call to fill_origin_blob() can use it to locate the data.  blob_sha1
315  * for an origin is also used to pass the blame for the entire file to
316  * the parent to detect the case where a child's blob is identical to
317  * that of its parent's.
318  *
319  * This also fills origin->mode for corresponding tree path.
320  */
321 static int fill_blob_sha1_and_mode(struct origin *origin)
322 {
323         if (!is_null_sha1(origin->blob_sha1))
324                 return 0;
325         if (get_tree_entry(origin->commit->object.sha1,
326                            origin->path,
327                            origin->blob_sha1, &origin->mode))
328                 goto error_out;
329         if (sha1_object_info(origin->blob_sha1, NULL) != OBJ_BLOB)
330                 goto error_out;
331         return 0;
332  error_out:
333         hashclr(origin->blob_sha1);
334         origin->mode = S_IFINVALID;
335         return -1;
336 }
337
338 /*
339  * We have an origin -- check if the same path exists in the
340  * parent and return an origin structure to represent it.
341  */
342 static struct origin *find_origin(struct scoreboard *sb,
343                                   struct commit *parent,
344                                   struct origin *origin)
345 {
346         struct origin *porigin = NULL;
347         struct diff_options diff_opts;
348         const char *paths[2];
349
350         if (parent->util) {
351                 /*
352                  * Each commit object can cache one origin in that
353                  * commit.  This is a freestanding copy of origin and
354                  * not refcounted.
355                  */
356                 struct origin *cached = parent->util;
357                 if (!strcmp(cached->path, origin->path)) {
358                         /*
359                          * The same path between origin and its parent
360                          * without renaming -- the most common case.
361                          */
362                         porigin = get_origin(sb, parent, cached->path);
363
364                         /*
365                          * If the origin was newly created (i.e. get_origin
366                          * would call make_origin if none is found in the
367                          * scoreboard), it does not know the blob_sha1/mode,
368                          * so copy it.  Otherwise porigin was in the
369                          * scoreboard and already knows blob_sha1/mode.
370                          */
371                         if (porigin->refcnt == 1) {
372                                 hashcpy(porigin->blob_sha1, cached->blob_sha1);
373                                 porigin->mode = cached->mode;
374                         }
375                         return porigin;
376                 }
377                 /* otherwise it was not very useful; free it */
378                 free(parent->util);
379                 parent->util = NULL;
380         }
381
382         /* See if the origin->path is different between parent
383          * and origin first.  Most of the time they are the
384          * same and diff-tree is fairly efficient about this.
385          */
386         diff_setup(&diff_opts);
387         DIFF_OPT_SET(&diff_opts, RECURSIVE);
388         diff_opts.detect_rename = 0;
389         diff_opts.output_format = DIFF_FORMAT_NO_OUTPUT;
390         paths[0] = origin->path;
391         paths[1] = NULL;
392
393         diff_tree_setup_paths(paths, &diff_opts);
394         if (diff_setup_done(&diff_opts) < 0)
395                 die("diff-setup");
396
397         if (is_null_sha1(origin->commit->object.sha1))
398                 do_diff_cache(parent->tree->object.sha1, &diff_opts);
399         else
400                 diff_tree_sha1(parent->tree->object.sha1,
401                                origin->commit->tree->object.sha1,
402                                "", &diff_opts);
403         diffcore_std(&diff_opts);
404
405         if (!diff_queued_diff.nr) {
406                 /* The path is the same as parent */
407                 porigin = get_origin(sb, parent, origin->path);
408                 hashcpy(porigin->blob_sha1, origin->blob_sha1);
409                 porigin->mode = origin->mode;
410         } else {
411                 /*
412                  * Since origin->path is a pathspec, if the parent
413                  * commit had it as a directory, we will see a whole
414                  * bunch of deletion of files in the directory that we
415                  * do not care about.
416                  */
417                 int i;
418                 struct diff_filepair *p = NULL;
419                 for (i = 0; i < diff_queued_diff.nr; i++) {
420                         const char *name;
421                         p = diff_queued_diff.queue[i];
422                         name = p->one->path ? p->one->path : p->two->path;
423                         if (!strcmp(name, origin->path))
424                                 break;
425                 }
426                 if (!p)
427                         die("internal error in blame::find_origin");
428                 switch (p->status) {
429                 default:
430                         die("internal error in blame::find_origin (%c)",
431                             p->status);
432                 case 'M':
433                         porigin = get_origin(sb, parent, origin->path);
434                         hashcpy(porigin->blob_sha1, p->one->sha1);
435                         porigin->mode = p->one->mode;
436                         break;
437                 case 'A':
438                 case 'T':
439                         /* Did not exist in parent, or type changed */
440                         break;
441                 }
442         }
443         diff_flush(&diff_opts);
444         diff_tree_release_paths(&diff_opts);
445         if (porigin) {
446                 /*
447                  * Create a freestanding copy that is not part of
448                  * the refcounted origin found in the scoreboard, and
449                  * cache it in the commit.
450                  */
451                 struct origin *cached;
452
453                 cached = make_origin(porigin->commit, porigin->path);
454                 hashcpy(cached->blob_sha1, porigin->blob_sha1);
455                 cached->mode = porigin->mode;
456                 parent->util = cached;
457         }
458         return porigin;
459 }
460
461 /*
462  * We have an origin -- find the path that corresponds to it in its
463  * parent and return an origin structure to represent it.
464  */
465 static struct origin *find_rename(struct scoreboard *sb,
466                                   struct commit *parent,
467                                   struct origin *origin)
468 {
469         struct origin *porigin = NULL;
470         struct diff_options diff_opts;
471         int i;
472         const char *paths[2];
473
474         diff_setup(&diff_opts);
475         DIFF_OPT_SET(&diff_opts, RECURSIVE);
476         diff_opts.detect_rename = DIFF_DETECT_RENAME;
477         diff_opts.output_format = DIFF_FORMAT_NO_OUTPUT;
478         diff_opts.single_follow = origin->path;
479         paths[0] = NULL;
480         diff_tree_setup_paths(paths, &diff_opts);
481         if (diff_setup_done(&diff_opts) < 0)
482                 die("diff-setup");
483
484         if (is_null_sha1(origin->commit->object.sha1))
485                 do_diff_cache(parent->tree->object.sha1, &diff_opts);
486         else
487                 diff_tree_sha1(parent->tree->object.sha1,
488                                origin->commit->tree->object.sha1,
489                                "", &diff_opts);
490         diffcore_std(&diff_opts);
491
492         for (i = 0; i < diff_queued_diff.nr; i++) {
493                 struct diff_filepair *p = diff_queued_diff.queue[i];
494                 if ((p->status == 'R' || p->status == 'C') &&
495                     !strcmp(p->two->path, origin->path)) {
496                         porigin = get_origin(sb, parent, p->one->path);
497                         hashcpy(porigin->blob_sha1, p->one->sha1);
498                         porigin->mode = p->one->mode;
499                         break;
500                 }
501         }
502         diff_flush(&diff_opts);
503         diff_tree_release_paths(&diff_opts);
504         return porigin;
505 }
506
507 /*
508  * Link in a new blame entry to the scoreboard.  Entries that cover the
509  * same line range have been removed from the scoreboard previously.
510  */
511 static void add_blame_entry(struct scoreboard *sb, struct blame_entry *e)
512 {
513         struct blame_entry *ent, *prev = NULL;
514
515         origin_incref(e->suspect);
516
517         for (ent = sb->ent; ent && ent->lno < e->lno; ent = ent->next)
518                 prev = ent;
519
520         /* prev, if not NULL, is the last one that is below e */
521         e->prev = prev;
522         if (prev) {
523                 e->next = prev->next;
524                 prev->next = e;
525         }
526         else {
527                 e->next = sb->ent;
528                 sb->ent = e;
529         }
530         if (e->next)
531                 e->next->prev = e;
532 }
533
534 /*
535  * src typically is on-stack; we want to copy the information in it to
536  * a malloced blame_entry that is already on the linked list of the
537  * scoreboard.  The origin of dst loses a refcnt while the origin of src
538  * gains one.
539  */
540 static void dup_entry(struct blame_entry *dst, struct blame_entry *src)
541 {
542         struct blame_entry *p, *n;
543
544         p = dst->prev;
545         n = dst->next;
546         origin_incref(src->suspect);
547         origin_decref(dst->suspect);
548         memcpy(dst, src, sizeof(*src));
549         dst->prev = p;
550         dst->next = n;
551         dst->score = 0;
552 }
553
554 static const char *nth_line(struct scoreboard *sb, int lno)
555 {
556         return sb->final_buf + sb->lineno[lno];
557 }
558
559 /*
560  * It is known that lines between tlno to same came from parent, and e
561  * has an overlap with that range.  it also is known that parent's
562  * line plno corresponds to e's line tlno.
563  *
564  *                <---- e ----->
565  *                   <------>
566  *                   <------------>
567  *             <------------>
568  *             <------------------>
569  *
570  * Split e into potentially three parts; before this chunk, the chunk
571  * to be blamed for the parent, and after that portion.
572  */
573 static void split_overlap(struct blame_entry *split,
574                           struct blame_entry *e,
575                           int tlno, int plno, int same,
576                           struct origin *parent)
577 {
578         int chunk_end_lno;
579         memset(split, 0, sizeof(struct blame_entry [3]));
580
581         if (e->s_lno < tlno) {
582                 /* there is a pre-chunk part not blamed on parent */
583                 split[0].suspect = origin_incref(e->suspect);
584                 split[0].lno = e->lno;
585                 split[0].s_lno = e->s_lno;
586                 split[0].num_lines = tlno - e->s_lno;
587                 split[1].lno = e->lno + tlno - e->s_lno;
588                 split[1].s_lno = plno;
589         }
590         else {
591                 split[1].lno = e->lno;
592                 split[1].s_lno = plno + (e->s_lno - tlno);
593         }
594
595         if (same < e->s_lno + e->num_lines) {
596                 /* there is a post-chunk part not blamed on parent */
597                 split[2].suspect = origin_incref(e->suspect);
598                 split[2].lno = e->lno + (same - e->s_lno);
599                 split[2].s_lno = e->s_lno + (same - e->s_lno);
600                 split[2].num_lines = e->s_lno + e->num_lines - same;
601                 chunk_end_lno = split[2].lno;
602         }
603         else
604                 chunk_end_lno = e->lno + e->num_lines;
605         split[1].num_lines = chunk_end_lno - split[1].lno;
606
607         /*
608          * if it turns out there is nothing to blame the parent for,
609          * forget about the splitting.  !split[1].suspect signals this.
610          */
611         if (split[1].num_lines < 1)
612                 return;
613         split[1].suspect = origin_incref(parent);
614 }
615
616 /*
617  * split_overlap() divided an existing blame e into up to three parts
618  * in split.  Adjust the linked list of blames in the scoreboard to
619  * reflect the split.
620  */
621 static void split_blame(struct scoreboard *sb,
622                         struct blame_entry *split,
623                         struct blame_entry *e)
624 {
625         struct blame_entry *new_entry;
626
627         if (split[0].suspect && split[2].suspect) {
628                 /* The first part (reuse storage for the existing entry e) */
629                 dup_entry(e, &split[0]);
630
631                 /* The last part -- me */
632                 new_entry = xmalloc(sizeof(*new_entry));
633                 memcpy(new_entry, &(split[2]), sizeof(struct blame_entry));
634                 add_blame_entry(sb, new_entry);
635
636                 /* ... and the middle part -- parent */
637                 new_entry = xmalloc(sizeof(*new_entry));
638                 memcpy(new_entry, &(split[1]), sizeof(struct blame_entry));
639                 add_blame_entry(sb, new_entry);
640         }
641         else if (!split[0].suspect && !split[2].suspect)
642                 /*
643                  * The parent covers the entire area; reuse storage for
644                  * e and replace it with the parent.
645                  */
646                 dup_entry(e, &split[1]);
647         else if (split[0].suspect) {
648                 /* me and then parent */
649                 dup_entry(e, &split[0]);
650
651                 new_entry = xmalloc(sizeof(*new_entry));
652                 memcpy(new_entry, &(split[1]), sizeof(struct blame_entry));
653                 add_blame_entry(sb, new_entry);
654         }
655         else {
656                 /* parent and then me */
657                 dup_entry(e, &split[1]);
658
659                 new_entry = xmalloc(sizeof(*new_entry));
660                 memcpy(new_entry, &(split[2]), sizeof(struct blame_entry));
661                 add_blame_entry(sb, new_entry);
662         }
663
664         if (DEBUG) { /* sanity */
665                 struct blame_entry *ent;
666                 int lno = sb->ent->lno, corrupt = 0;
667
668                 for (ent = sb->ent; ent; ent = ent->next) {
669                         if (lno != ent->lno)
670                                 corrupt = 1;
671                         if (ent->s_lno < 0)
672                                 corrupt = 1;
673                         lno += ent->num_lines;
674                 }
675                 if (corrupt) {
676                         lno = sb->ent->lno;
677                         for (ent = sb->ent; ent; ent = ent->next) {
678                                 printf("L %8d l %8d n %8d\n",
679                                        lno, ent->lno, ent->num_lines);
680                                 lno = ent->lno + ent->num_lines;
681                         }
682                         die("oops");
683                 }
684         }
685 }
686
687 /*
688  * After splitting the blame, the origins used by the
689  * on-stack blame_entry should lose one refcnt each.
690  */
691 static void decref_split(struct blame_entry *split)
692 {
693         int i;
694
695         for (i = 0; i < 3; i++)
696                 origin_decref(split[i].suspect);
697 }
698
699 /*
700  * Helper for blame_chunk().  blame_entry e is known to overlap with
701  * the patch hunk; split it and pass blame to the parent.
702  */
703 static void blame_overlap(struct scoreboard *sb, struct blame_entry *e,
704                           int tlno, int plno, int same,
705                           struct origin *parent)
706 {
707         struct blame_entry split[3];
708
709         split_overlap(split, e, tlno, plno, same, parent);
710         if (split[1].suspect)
711                 split_blame(sb, split, e);
712         decref_split(split);
713 }
714
715 /*
716  * Find the line number of the last line the target is suspected for.
717  */
718 static int find_last_in_target(struct scoreboard *sb, struct origin *target)
719 {
720         struct blame_entry *e;
721         int last_in_target = -1;
722
723         for (e = sb->ent; e; e = e->next) {
724                 if (e->guilty || !same_suspect(e->suspect, target))
725                         continue;
726                 if (last_in_target < e->s_lno + e->num_lines)
727                         last_in_target = e->s_lno + e->num_lines;
728         }
729         return last_in_target;
730 }
731
732 /*
733  * Process one hunk from the patch between the current suspect for
734  * blame_entry e and its parent.  Find and split the overlap, and
735  * pass blame to the overlapping part to the parent.
736  */
737 static void blame_chunk(struct scoreboard *sb,
738                         int tlno, int plno, int same,
739                         struct origin *target, struct origin *parent)
740 {
741         struct blame_entry *e;
742
743         for (e = sb->ent; e; e = e->next) {
744                 if (e->guilty || !same_suspect(e->suspect, target))
745                         continue;
746                 if (same <= e->s_lno)
747                         continue;
748                 if (tlno < e->s_lno + e->num_lines)
749                         blame_overlap(sb, e, tlno, plno, same, parent);
750         }
751 }
752
753 struct blame_chunk_cb_data {
754         struct scoreboard *sb;
755         struct origin *target;
756         struct origin *parent;
757         long plno;
758         long tlno;
759 };
760
761 static void blame_chunk_cb(void *data, long same, long p_next, long t_next)
762 {
763         struct blame_chunk_cb_data *d = data;
764         blame_chunk(d->sb, d->tlno, d->plno, same, d->target, d->parent);
765         d->plno = p_next;
766         d->tlno = t_next;
767 }
768
769 /*
770  * We are looking at the origin 'target' and aiming to pass blame
771  * for the lines it is suspected to its parent.  Run diff to find
772  * which lines came from parent and pass blame for them.
773  */
774 static int pass_blame_to_parent(struct scoreboard *sb,
775                                 struct origin *target,
776                                 struct origin *parent)
777 {
778         int last_in_target;
779         mmfile_t file_p, file_o;
780         struct blame_chunk_cb_data d;
781         xpparam_t xpp;
782         xdemitconf_t xecfg;
783         memset(&d, 0, sizeof(d));
784         d.sb = sb; d.target = target; d.parent = parent;
785         last_in_target = find_last_in_target(sb, target);
786         if (last_in_target < 0)
787                 return 1; /* nothing remains for this target */
788
789         fill_origin_blob(&sb->revs->diffopt, parent, &file_p);
790         fill_origin_blob(&sb->revs->diffopt, target, &file_o);
791         num_get_patch++;
792
793         memset(&xpp, 0, sizeof(xpp));
794         xpp.flags = xdl_opts;
795         memset(&xecfg, 0, sizeof(xecfg));
796         xecfg.ctxlen = 0;
797         xdi_diff_hunks(&file_p, &file_o, blame_chunk_cb, &d, &xpp, &xecfg);
798         /* The rest (i.e. anything after tlno) are the same as the parent */
799         blame_chunk(sb, d.tlno, d.plno, last_in_target, target, parent);
800
801         return 0;
802 }
803
804 /*
805  * The lines in blame_entry after splitting blames many times can become
806  * very small and trivial, and at some point it becomes pointless to
807  * blame the parents.  E.g. "\t\t}\n\t}\n\n" appears everywhere in any
808  * ordinary C program, and it is not worth to say it was copied from
809  * totally unrelated file in the parent.
810  *
811  * Compute how trivial the lines in the blame_entry are.
812  */
813 static unsigned ent_score(struct scoreboard *sb, struct blame_entry *e)
814 {
815         unsigned score;
816         const char *cp, *ep;
817
818         if (e->score)
819                 return e->score;
820
821         score = 1;
822         cp = nth_line(sb, e->lno);
823         ep = nth_line(sb, e->lno + e->num_lines);
824         while (cp < ep) {
825                 unsigned ch = *((unsigned char *)cp);
826                 if (isalnum(ch))
827                         score++;
828                 cp++;
829         }
830         e->score = score;
831         return score;
832 }
833
834 /*
835  * best_so_far[] and this[] are both a split of an existing blame_entry
836  * that passes blame to the parent.  Maintain best_so_far the best split
837  * so far, by comparing this and best_so_far and copying this into
838  * bst_so_far as needed.
839  */
840 static void copy_split_if_better(struct scoreboard *sb,
841                                  struct blame_entry *best_so_far,
842                                  struct blame_entry *this)
843 {
844         int i;
845
846         if (!this[1].suspect)
847                 return;
848         if (best_so_far[1].suspect) {
849                 if (ent_score(sb, &this[1]) < ent_score(sb, &best_so_far[1]))
850                         return;
851         }
852
853         for (i = 0; i < 3; i++)
854                 origin_incref(this[i].suspect);
855         decref_split(best_so_far);
856         memcpy(best_so_far, this, sizeof(struct blame_entry [3]));
857 }
858
859 /*
860  * We are looking at a part of the final image represented by
861  * ent (tlno and same are offset by ent->s_lno).
862  * tlno is where we are looking at in the final image.
863  * up to (but not including) same match preimage.
864  * plno is where we are looking at in the preimage.
865  *
866  * <-------------- final image ---------------------->
867  *       <------ent------>
868  *         ^tlno ^same
869  *    <---------preimage----->
870  *         ^plno
871  *
872  * All line numbers are 0-based.
873  */
874 static void handle_split(struct scoreboard *sb,
875                          struct blame_entry *ent,
876                          int tlno, int plno, int same,
877                          struct origin *parent,
878                          struct blame_entry *split)
879 {
880         if (ent->num_lines <= tlno)
881                 return;
882         if (tlno < same) {
883                 struct blame_entry this[3];
884                 tlno += ent->s_lno;
885                 same += ent->s_lno;
886                 split_overlap(this, ent, tlno, plno, same, parent);
887                 copy_split_if_better(sb, split, this);
888                 decref_split(this);
889         }
890 }
891
892 struct handle_split_cb_data {
893         struct scoreboard *sb;
894         struct blame_entry *ent;
895         struct origin *parent;
896         struct blame_entry *split;
897         long plno;
898         long tlno;
899 };
900
901 static void handle_split_cb(void *data, long same, long p_next, long t_next)
902 {
903         struct handle_split_cb_data *d = data;
904         handle_split(d->sb, d->ent, d->tlno, d->plno, same, d->parent, d->split);
905         d->plno = p_next;
906         d->tlno = t_next;
907 }
908
909 /*
910  * Find the lines from parent that are the same as ent so that
911  * we can pass blames to it.  file_p has the blob contents for
912  * the parent.
913  */
914 static void find_copy_in_blob(struct scoreboard *sb,
915                               struct blame_entry *ent,
916                               struct origin *parent,
917                               struct blame_entry *split,
918                               mmfile_t *file_p)
919 {
920         const char *cp;
921         int cnt;
922         mmfile_t file_o;
923         struct handle_split_cb_data d;
924         xpparam_t xpp;
925         xdemitconf_t xecfg;
926         memset(&d, 0, sizeof(d));
927         d.sb = sb; d.ent = ent; d.parent = parent; d.split = split;
928         /*
929          * Prepare mmfile that contains only the lines in ent.
930          */
931         cp = nth_line(sb, ent->lno);
932         file_o.ptr = (char *) cp;
933         cnt = ent->num_lines;
934
935         while (cnt && cp < sb->final_buf + sb->final_buf_size) {
936                 if (*cp++ == '\n')
937                         cnt--;
938         }
939         file_o.size = cp - file_o.ptr;
940
941         /*
942          * file_o is a part of final image we are annotating.
943          * file_p partially may match that image.
944          */
945         memset(&xpp, 0, sizeof(xpp));
946         xpp.flags = xdl_opts;
947         memset(&xecfg, 0, sizeof(xecfg));
948         xecfg.ctxlen = 1;
949         memset(split, 0, sizeof(struct blame_entry [3]));
950         xdi_diff_hunks(file_p, &file_o, handle_split_cb, &d, &xpp, &xecfg);
951         /* remainder, if any, all match the preimage */
952         handle_split(sb, ent, d.tlno, d.plno, ent->num_lines, parent, split);
953 }
954
955 /*
956  * See if lines currently target is suspected for can be attributed to
957  * parent.
958  */
959 static int find_move_in_parent(struct scoreboard *sb,
960                                struct origin *target,
961                                struct origin *parent)
962 {
963         int last_in_target, made_progress;
964         struct blame_entry *e, split[3];
965         mmfile_t file_p;
966
967         last_in_target = find_last_in_target(sb, target);
968         if (last_in_target < 0)
969                 return 1; /* nothing remains for this target */
970
971         fill_origin_blob(&sb->revs->diffopt, parent, &file_p);
972         if (!file_p.ptr)
973                 return 0;
974
975         made_progress = 1;
976         while (made_progress) {
977                 made_progress = 0;
978                 for (e = sb->ent; e; e = e->next) {
979                         if (e->guilty || !same_suspect(e->suspect, target) ||
980                             ent_score(sb, e) < blame_move_score)
981                                 continue;
982                         find_copy_in_blob(sb, e, parent, split, &file_p);
983                         if (split[1].suspect &&
984                             blame_move_score < ent_score(sb, &split[1])) {
985                                 split_blame(sb, split, e);
986                                 made_progress = 1;
987                         }
988                         decref_split(split);
989                 }
990         }
991         return 0;
992 }
993
994 struct blame_list {
995         struct blame_entry *ent;
996         struct blame_entry split[3];
997 };
998
999 /*
1000  * Count the number of entries the target is suspected for,
1001  * and prepare a list of entry and the best split.
1002  */
1003 static struct blame_list *setup_blame_list(struct scoreboard *sb,
1004                                            struct origin *target,
1005                                            int min_score,
1006                                            int *num_ents_p)
1007 {
1008         struct blame_entry *e;
1009         int num_ents, i;
1010         struct blame_list *blame_list = NULL;
1011
1012         for (e = sb->ent, num_ents = 0; e; e = e->next)
1013                 if (!e->scanned && !e->guilty &&
1014                     same_suspect(e->suspect, target) &&
1015                     min_score < ent_score(sb, e))
1016                         num_ents++;
1017         if (num_ents) {
1018                 blame_list = xcalloc(num_ents, sizeof(struct blame_list));
1019                 for (e = sb->ent, i = 0; e; e = e->next)
1020                         if (!e->scanned && !e->guilty &&
1021                             same_suspect(e->suspect, target) &&
1022                             min_score < ent_score(sb, e))
1023                                 blame_list[i++].ent = e;
1024         }
1025         *num_ents_p = num_ents;
1026         return blame_list;
1027 }
1028
1029 /*
1030  * Reset the scanned status on all entries.
1031  */
1032 static void reset_scanned_flag(struct scoreboard *sb)
1033 {
1034         struct blame_entry *e;
1035         for (e = sb->ent; e; e = e->next)
1036                 e->scanned = 0;
1037 }
1038
1039 /*
1040  * For lines target is suspected for, see if we can find code movement
1041  * across file boundary from the parent commit.  porigin is the path
1042  * in the parent we already tried.
1043  */
1044 static int find_copy_in_parent(struct scoreboard *sb,
1045                                struct origin *target,
1046                                struct commit *parent,
1047                                struct origin *porigin,
1048                                int opt)
1049 {
1050         struct diff_options diff_opts;
1051         const char *paths[1];
1052         int i, j;
1053         int retval;
1054         struct blame_list *blame_list;
1055         int num_ents;
1056
1057         blame_list = setup_blame_list(sb, target, blame_copy_score, &num_ents);
1058         if (!blame_list)
1059                 return 1; /* nothing remains for this target */
1060
1061         diff_setup(&diff_opts);
1062         DIFF_OPT_SET(&diff_opts, RECURSIVE);
1063         diff_opts.output_format = DIFF_FORMAT_NO_OUTPUT;
1064
1065         paths[0] = NULL;
1066         diff_tree_setup_paths(paths, &diff_opts);
1067         if (diff_setup_done(&diff_opts) < 0)
1068                 die("diff-setup");
1069
1070         /* Try "find copies harder" on new path if requested;
1071          * we do not want to use diffcore_rename() actually to
1072          * match things up; find_copies_harder is set only to
1073          * force diff_tree_sha1() to feed all filepairs to diff_queue,
1074          * and this code needs to be after diff_setup_done(), which
1075          * usually makes find-copies-harder imply copy detection.
1076          */
1077         if ((opt & PICKAXE_BLAME_COPY_HARDEST)
1078             || ((opt & PICKAXE_BLAME_COPY_HARDER)
1079                 && (!porigin || strcmp(target->path, porigin->path))))
1080                 DIFF_OPT_SET(&diff_opts, FIND_COPIES_HARDER);
1081
1082         if (is_null_sha1(target->commit->object.sha1))
1083                 do_diff_cache(parent->tree->object.sha1, &diff_opts);
1084         else
1085                 diff_tree_sha1(parent->tree->object.sha1,
1086                                target->commit->tree->object.sha1,
1087                                "", &diff_opts);
1088
1089         if (!DIFF_OPT_TST(&diff_opts, FIND_COPIES_HARDER))
1090                 diffcore_std(&diff_opts);
1091
1092         retval = 0;
1093         while (1) {
1094                 int made_progress = 0;
1095
1096                 for (i = 0; i < diff_queued_diff.nr; i++) {
1097                         struct diff_filepair *p = diff_queued_diff.queue[i];
1098                         struct origin *norigin;
1099                         mmfile_t file_p;
1100                         struct blame_entry this[3];
1101
1102                         if (!DIFF_FILE_VALID(p->one))
1103                                 continue; /* does not exist in parent */
1104                         if (S_ISGITLINK(p->one->mode))
1105                                 continue; /* ignore git links */
1106                         if (porigin && !strcmp(p->one->path, porigin->path))
1107                                 /* find_move already dealt with this path */
1108                                 continue;
1109
1110                         norigin = get_origin(sb, parent, p->one->path);
1111                         hashcpy(norigin->blob_sha1, p->one->sha1);
1112                         norigin->mode = p->one->mode;
1113                         fill_origin_blob(&sb->revs->diffopt, norigin, &file_p);
1114                         if (!file_p.ptr)
1115                                 continue;
1116
1117                         for (j = 0; j < num_ents; j++) {
1118                                 find_copy_in_blob(sb, blame_list[j].ent,
1119                                                   norigin, this, &file_p);
1120                                 copy_split_if_better(sb, blame_list[j].split,
1121                                                      this);
1122                                 decref_split(this);
1123                         }
1124                         origin_decref(norigin);
1125                 }
1126
1127                 for (j = 0; j < num_ents; j++) {
1128                         struct blame_entry *split = blame_list[j].split;
1129                         if (split[1].suspect &&
1130                             blame_copy_score < ent_score(sb, &split[1])) {
1131                                 split_blame(sb, split, blame_list[j].ent);
1132                                 made_progress = 1;
1133                         }
1134                         else
1135                                 blame_list[j].ent->scanned = 1;
1136                         decref_split(split);
1137                 }
1138                 free(blame_list);
1139
1140                 if (!made_progress)
1141                         break;
1142                 blame_list = setup_blame_list(sb, target, blame_copy_score, &num_ents);
1143                 if (!blame_list) {
1144                         retval = 1;
1145                         break;
1146                 }
1147         }
1148         reset_scanned_flag(sb);
1149         diff_flush(&diff_opts);
1150         diff_tree_release_paths(&diff_opts);
1151         return retval;
1152 }
1153
1154 /*
1155  * The blobs of origin and porigin exactly match, so everything
1156  * origin is suspected for can be blamed on the parent.
1157  */
1158 static void pass_whole_blame(struct scoreboard *sb,
1159                              struct origin *origin, struct origin *porigin)
1160 {
1161         struct blame_entry *e;
1162
1163         if (!porigin->file.ptr && origin->file.ptr) {
1164                 /* Steal its file */
1165                 porigin->file = origin->file;
1166                 origin->file.ptr = NULL;
1167         }
1168         for (e = sb->ent; e; e = e->next) {
1169                 if (!same_suspect(e->suspect, origin))
1170                         continue;
1171                 origin_incref(porigin);
1172                 origin_decref(e->suspect);
1173                 e->suspect = porigin;
1174         }
1175 }
1176
1177 /*
1178  * We pass blame from the current commit to its parents.  We keep saying
1179  * "parent" (and "porigin"), but what we mean is to find scapegoat to
1180  * exonerate ourselves.
1181  */
1182 static struct commit_list *first_scapegoat(struct rev_info *revs, struct commit *commit)
1183 {
1184         if (!reverse)
1185                 return commit->parents;
1186         return lookup_decoration(&revs->children, &commit->object);
1187 }
1188
1189 static int num_scapegoats(struct rev_info *revs, struct commit *commit)
1190 {
1191         int cnt;
1192         struct commit_list *l = first_scapegoat(revs, commit);
1193         for (cnt = 0; l; l = l->next)
1194                 cnt++;
1195         return cnt;
1196 }
1197
1198 #define MAXSG 16
1199
1200 static void pass_blame(struct scoreboard *sb, struct origin *origin, int opt)
1201 {
1202         struct rev_info *revs = sb->revs;
1203         int i, pass, num_sg;
1204         struct commit *commit = origin->commit;
1205         struct commit_list *sg;
1206         struct origin *sg_buf[MAXSG];
1207         struct origin *porigin, **sg_origin = sg_buf;
1208
1209         num_sg = num_scapegoats(revs, commit);
1210         if (!num_sg)
1211                 goto finish;
1212         else if (num_sg < ARRAY_SIZE(sg_buf))
1213                 memset(sg_buf, 0, sizeof(sg_buf));
1214         else
1215                 sg_origin = xcalloc(num_sg, sizeof(*sg_origin));
1216
1217         /*
1218          * The first pass looks for unrenamed path to optimize for
1219          * common cases, then we look for renames in the second pass.
1220          */
1221         for (pass = 0; pass < 2; pass++) {
1222                 struct origin *(*find)(struct scoreboard *,
1223                                        struct commit *, struct origin *);
1224                 find = pass ? find_rename : find_origin;
1225
1226                 for (i = 0, sg = first_scapegoat(revs, commit);
1227                      i < num_sg && sg;
1228                      sg = sg->next, i++) {
1229                         struct commit *p = sg->item;
1230                         int j, same;
1231
1232                         if (sg_origin[i])
1233                                 continue;
1234                         if (parse_commit(p))
1235                                 continue;
1236                         porigin = find(sb, p, origin);
1237                         if (!porigin)
1238                                 continue;
1239                         if (!hashcmp(porigin->blob_sha1, origin->blob_sha1)) {
1240                                 pass_whole_blame(sb, origin, porigin);
1241                                 origin_decref(porigin);
1242                                 goto finish;
1243                         }
1244                         for (j = same = 0; j < i; j++)
1245                                 if (sg_origin[j] &&
1246                                     !hashcmp(sg_origin[j]->blob_sha1,
1247                                              porigin->blob_sha1)) {
1248                                         same = 1;
1249                                         break;
1250                                 }
1251                         if (!same)
1252                                 sg_origin[i] = porigin;
1253                         else
1254                                 origin_decref(porigin);
1255                 }
1256         }
1257
1258         num_commits++;
1259         for (i = 0, sg = first_scapegoat(revs, commit);
1260              i < num_sg && sg;
1261              sg = sg->next, i++) {
1262                 struct origin *porigin = sg_origin[i];
1263                 if (!porigin)
1264                         continue;
1265                 if (!origin->previous) {
1266                         origin_incref(porigin);
1267                         origin->previous = porigin;
1268                 }
1269                 if (pass_blame_to_parent(sb, origin, porigin))
1270                         goto finish;
1271         }
1272
1273         /*
1274          * Optionally find moves in parents' files.
1275          */
1276         if (opt & PICKAXE_BLAME_MOVE)
1277                 for (i = 0, sg = first_scapegoat(revs, commit);
1278                      i < num_sg && sg;
1279                      sg = sg->next, i++) {
1280                         struct origin *porigin = sg_origin[i];
1281                         if (!porigin)
1282                                 continue;
1283                         if (find_move_in_parent(sb, origin, porigin))
1284                                 goto finish;
1285                 }
1286
1287         /*
1288          * Optionally find copies from parents' files.
1289          */
1290         if (opt & PICKAXE_BLAME_COPY)
1291                 for (i = 0, sg = first_scapegoat(revs, commit);
1292                      i < num_sg && sg;
1293                      sg = sg->next, i++) {
1294                         struct origin *porigin = sg_origin[i];
1295                         if (find_copy_in_parent(sb, origin, sg->item,
1296                                                 porigin, opt))
1297                                 goto finish;
1298                 }
1299
1300  finish:
1301         for (i = 0; i < num_sg; i++) {
1302                 if (sg_origin[i]) {
1303                         drop_origin_blob(sg_origin[i]);
1304                         origin_decref(sg_origin[i]);
1305                 }
1306         }
1307         drop_origin_blob(origin);
1308         if (sg_buf != sg_origin)
1309                 free(sg_origin);
1310 }
1311
1312 /*
1313  * Information on commits, used for output.
1314  */
1315 struct commit_info {
1316         const char *author;
1317         const char *author_mail;
1318         unsigned long author_time;
1319         const char *author_tz;
1320
1321         /* filled only when asked for details */
1322         const char *committer;
1323         const char *committer_mail;
1324         unsigned long committer_time;
1325         const char *committer_tz;
1326
1327         const char *summary;
1328 };
1329
1330 /*
1331  * Parse author/committer line in the commit object buffer
1332  */
1333 static void get_ac_line(const char *inbuf, const char *what,
1334                         int person_len, char *person,
1335                         int mail_len, char *mail,
1336                         unsigned long *time, const char **tz)
1337 {
1338         int len, tzlen, maillen;
1339         char *tmp, *endp, *timepos, *mailpos;
1340
1341         tmp = strstr(inbuf, what);
1342         if (!tmp)
1343                 goto error_out;
1344         tmp += strlen(what);
1345         endp = strchr(tmp, '\n');
1346         if (!endp)
1347                 len = strlen(tmp);
1348         else
1349                 len = endp - tmp;
1350         if (person_len <= len) {
1351         error_out:
1352                 /* Ugh */
1353                 *tz = "(unknown)";
1354                 strcpy(person, *tz);
1355                 strcpy(mail, *tz);
1356                 *time = 0;
1357                 return;
1358         }
1359         memcpy(person, tmp, len);
1360
1361         tmp = person;
1362         tmp += len;
1363         *tmp = 0;
1364         while (person < tmp && *tmp != ' ')
1365                 tmp--;
1366         if (tmp <= person)
1367                 goto error_out;
1368         *tz = tmp+1;
1369         tzlen = (person+len)-(tmp+1);
1370
1371         *tmp = 0;
1372         while (person < tmp && *tmp != ' ')
1373                 tmp--;
1374         if (tmp <= person)
1375                 goto error_out;
1376         *time = strtoul(tmp, NULL, 10);
1377         timepos = tmp;
1378
1379         *tmp = 0;
1380         while (person < tmp && *tmp != ' ')
1381                 tmp--;
1382         if (tmp <= person)
1383                 return;
1384         mailpos = tmp + 1;
1385         *tmp = 0;
1386         maillen = timepos - tmp;
1387         memcpy(mail, mailpos, maillen);
1388
1389         if (!mailmap.nr)
1390                 return;
1391
1392         /*
1393          * mailmap expansion may make the name longer.
1394          * make room by pushing stuff down.
1395          */
1396         tmp = person + person_len - (tzlen + 1);
1397         memmove(tmp, *tz, tzlen);
1398         tmp[tzlen] = 0;
1399         *tz = tmp;
1400
1401         /*
1402          * Now, convert both name and e-mail using mailmap
1403          */
1404         if (map_user(&mailmap, mail+1, mail_len-1, person, tmp-person-1)) {
1405                 /* Add a trailing '>' to email, since map_user returns plain emails
1406                    Note: It already has '<', since we replace from mail+1 */
1407                 mailpos = memchr(mail, '\0', mail_len);
1408                 if (mailpos && mailpos-mail < mail_len - 1) {
1409                         *mailpos = '>';
1410                         *(mailpos+1) = '\0';
1411                 }
1412         }
1413 }
1414
1415 static void get_commit_info(struct commit *commit,
1416                             struct commit_info *ret,
1417                             int detailed)
1418 {
1419         int len;
1420         const char *subject;
1421         char *reencoded, *message;
1422         static char author_name[1024];
1423         static char author_mail[1024];
1424         static char committer_name[1024];
1425         static char committer_mail[1024];
1426         static char summary_buf[1024];
1427
1428         /*
1429          * We've operated without save_commit_buffer, so
1430          * we now need to populate them for output.
1431          */
1432         if (!commit->buffer) {
1433                 enum object_type type;
1434                 unsigned long size;
1435                 commit->buffer =
1436                         read_sha1_file(commit->object.sha1, &type, &size);
1437                 if (!commit->buffer)
1438                         die("Cannot read commit %s",
1439                             sha1_to_hex(commit->object.sha1));
1440         }
1441         reencoded = reencode_commit_message(commit, NULL);
1442         message   = reencoded ? reencoded : commit->buffer;
1443         ret->author = author_name;
1444         ret->author_mail = author_mail;
1445         get_ac_line(message, "\nauthor ",
1446                     sizeof(author_name), author_name,
1447                     sizeof(author_mail), author_mail,
1448                     &ret->author_time, &ret->author_tz);
1449
1450         if (!detailed) {
1451                 free(reencoded);
1452                 return;
1453         }
1454
1455         ret->committer = committer_name;
1456         ret->committer_mail = committer_mail;
1457         get_ac_line(message, "\ncommitter ",
1458                     sizeof(committer_name), committer_name,
1459                     sizeof(committer_mail), committer_mail,
1460                     &ret->committer_time, &ret->committer_tz);
1461
1462         ret->summary = summary_buf;
1463         len = find_commit_subject(message, &subject);
1464         if (len && len < sizeof(summary_buf)) {
1465                 memcpy(summary_buf, subject, len);
1466                 summary_buf[len] = 0;
1467         } else {
1468                 sprintf(summary_buf, "(%s)", sha1_to_hex(commit->object.sha1));
1469         }
1470         free(reencoded);
1471 }
1472
1473 /*
1474  * To allow LF and other nonportable characters in pathnames,
1475  * they are c-style quoted as needed.
1476  */
1477 static void write_filename_info(const char *path)
1478 {
1479         printf("filename ");
1480         write_name_quoted(path, stdout, '\n');
1481 }
1482
1483 /*
1484  * Porcelain/Incremental format wants to show a lot of details per
1485  * commit.  Instead of repeating this every line, emit it only once,
1486  * the first time each commit appears in the output.
1487  */
1488 static int emit_one_suspect_detail(struct origin *suspect)
1489 {
1490         struct commit_info ci;
1491
1492         if (suspect->commit->object.flags & METAINFO_SHOWN)
1493                 return 0;
1494
1495         suspect->commit->object.flags |= METAINFO_SHOWN;
1496         get_commit_info(suspect->commit, &ci, 1);
1497         printf("author %s\n", ci.author);
1498         printf("author-mail %s\n", ci.author_mail);
1499         printf("author-time %lu\n", ci.author_time);
1500         printf("author-tz %s\n", ci.author_tz);
1501         printf("committer %s\n", ci.committer);
1502         printf("committer-mail %s\n", ci.committer_mail);
1503         printf("committer-time %lu\n", ci.committer_time);
1504         printf("committer-tz %s\n", ci.committer_tz);
1505         printf("summary %s\n", ci.summary);
1506         if (suspect->commit->object.flags & UNINTERESTING)
1507                 printf("boundary\n");
1508         if (suspect->previous) {
1509                 struct origin *prev = suspect->previous;
1510                 printf("previous %s ", sha1_to_hex(prev->commit->object.sha1));
1511                 write_name_quoted(prev->path, stdout, '\n');
1512         }
1513         return 1;
1514 }
1515
1516 /*
1517  * The blame_entry is found to be guilty for the range.  Mark it
1518  * as such, and show it in incremental output.
1519  */
1520 static void found_guilty_entry(struct blame_entry *ent)
1521 {
1522         if (ent->guilty)
1523                 return;
1524         ent->guilty = 1;
1525         if (incremental) {
1526                 struct origin *suspect = ent->suspect;
1527
1528                 printf("%s %d %d %d\n",
1529                        sha1_to_hex(suspect->commit->object.sha1),
1530                        ent->s_lno + 1, ent->lno + 1, ent->num_lines);
1531                 emit_one_suspect_detail(suspect);
1532                 write_filename_info(suspect->path);
1533                 maybe_flush_or_die(stdout, "stdout");
1534         }
1535 }
1536
1537 /*
1538  * The main loop -- while the scoreboard has lines whose true origin
1539  * is still unknown, pick one blame_entry, and allow its current
1540  * suspect to pass blames to its parents.
1541  */
1542 static void assign_blame(struct scoreboard *sb, int opt)
1543 {
1544         struct rev_info *revs = sb->revs;
1545
1546         while (1) {
1547                 struct blame_entry *ent;
1548                 struct commit *commit;
1549                 struct origin *suspect = NULL;
1550
1551                 /* find one suspect to break down */
1552                 for (ent = sb->ent; !suspect && ent; ent = ent->next)
1553                         if (!ent->guilty)
1554                                 suspect = ent->suspect;
1555                 if (!suspect)
1556                         return; /* all done */
1557
1558                 /*
1559                  * We will use this suspect later in the loop,
1560                  * so hold onto it in the meantime.
1561                  */
1562                 origin_incref(suspect);
1563                 commit = suspect->commit;
1564                 if (!commit->object.parsed)
1565                         parse_commit(commit);
1566                 if (reverse ||
1567                     (!(commit->object.flags & UNINTERESTING) &&
1568                      !(revs->max_age != -1 && commit->date < revs->max_age)))
1569                         pass_blame(sb, suspect, opt);
1570                 else {
1571                         commit->object.flags |= UNINTERESTING;
1572                         if (commit->object.parsed)
1573                                 mark_parents_uninteresting(commit);
1574                 }
1575                 /* treat root commit as boundary */
1576                 if (!commit->parents && !show_root)
1577                         commit->object.flags |= UNINTERESTING;
1578
1579                 /* Take responsibility for the remaining entries */
1580                 for (ent = sb->ent; ent; ent = ent->next)
1581                         if (same_suspect(ent->suspect, suspect))
1582                                 found_guilty_entry(ent);
1583                 origin_decref(suspect);
1584
1585                 if (DEBUG) /* sanity */
1586                         sanity_check_refcnt(sb);
1587         }
1588 }
1589
1590 static const char *format_time(unsigned long time, const char *tz_str,
1591                                int show_raw_time)
1592 {
1593         static char time_buf[128];
1594         const char *time_str;
1595         int time_len;
1596         int tz;
1597
1598         if (show_raw_time) {
1599                 sprintf(time_buf, "%lu %s", time, tz_str);
1600         }
1601         else {
1602                 tz = atoi(tz_str);
1603                 time_str = show_date(time, tz, blame_date_mode);
1604                 time_len = strlen(time_str);
1605                 memcpy(time_buf, time_str, time_len);
1606                 memset(time_buf + time_len, ' ', blame_date_width - time_len);
1607         }
1608         return time_buf;
1609 }
1610
1611 #define OUTPUT_ANNOTATE_COMPAT  001
1612 #define OUTPUT_LONG_OBJECT_NAME 002
1613 #define OUTPUT_RAW_TIMESTAMP    004
1614 #define OUTPUT_PORCELAIN        010
1615 #define OUTPUT_SHOW_NAME        020
1616 #define OUTPUT_SHOW_NUMBER      040
1617 #define OUTPUT_SHOW_SCORE      0100
1618 #define OUTPUT_NO_AUTHOR       0200
1619 #define OUTPUT_SHOW_EMAIL       0400
1620
1621 static void emit_porcelain(struct scoreboard *sb, struct blame_entry *ent)
1622 {
1623         int cnt;
1624         const char *cp;
1625         struct origin *suspect = ent->suspect;
1626         char hex[41];
1627
1628         strcpy(hex, sha1_to_hex(suspect->commit->object.sha1));
1629         printf("%s%c%d %d %d\n",
1630                hex,
1631                ent->guilty ? ' ' : '*', /* purely for debugging */
1632                ent->s_lno + 1,
1633                ent->lno + 1,
1634                ent->num_lines);
1635         if (emit_one_suspect_detail(suspect) ||
1636             (suspect->commit->object.flags & MORE_THAN_ONE_PATH))
1637                 write_filename_info(suspect->path);
1638
1639         cp = nth_line(sb, ent->lno);
1640         for (cnt = 0; cnt < ent->num_lines; cnt++) {
1641                 char ch;
1642                 if (cnt)
1643                         printf("%s %d %d\n", hex,
1644                                ent->s_lno + 1 + cnt,
1645                                ent->lno + 1 + cnt);
1646                 putchar('\t');
1647                 do {
1648                         ch = *cp++;
1649                         putchar(ch);
1650                 } while (ch != '\n' &&
1651                          cp < sb->final_buf + sb->final_buf_size);
1652         }
1653
1654         if (sb->final_buf_size && cp[-1] != '\n')
1655                 putchar('\n');
1656 }
1657
1658 static void emit_other(struct scoreboard *sb, struct blame_entry *ent, int opt)
1659 {
1660         int cnt;
1661         const char *cp;
1662         struct origin *suspect = ent->suspect;
1663         struct commit_info ci;
1664         char hex[41];
1665         int show_raw_time = !!(opt & OUTPUT_RAW_TIMESTAMP);
1666
1667         get_commit_info(suspect->commit, &ci, 1);
1668         strcpy(hex, sha1_to_hex(suspect->commit->object.sha1));
1669
1670         cp = nth_line(sb, ent->lno);
1671         for (cnt = 0; cnt < ent->num_lines; cnt++) {
1672                 char ch;
1673                 int length = (opt & OUTPUT_LONG_OBJECT_NAME) ? 40 : 8;
1674
1675                 if (suspect->commit->object.flags & UNINTERESTING) {
1676                         if (blank_boundary)
1677                                 memset(hex, ' ', length);
1678                         else if (!(opt & OUTPUT_ANNOTATE_COMPAT)) {
1679                                 length--;
1680                                 putchar('^');
1681                         }
1682                 }
1683
1684                 printf("%.*s", length, hex);
1685                 if (opt & OUTPUT_ANNOTATE_COMPAT) {
1686                         const char *name;
1687                         if (opt & OUTPUT_SHOW_EMAIL)
1688                                 name = ci.author_mail;
1689                         else
1690                                 name = ci.author;
1691                         printf("\t(%10s\t%10s\t%d)", name,
1692                                format_time(ci.author_time, ci.author_tz,
1693                                            show_raw_time),
1694                                ent->lno + 1 + cnt);
1695                 } else {
1696                         if (opt & OUTPUT_SHOW_SCORE)
1697                                 printf(" %*d %02d",
1698                                        max_score_digits, ent->score,
1699                                        ent->suspect->refcnt);
1700                         if (opt & OUTPUT_SHOW_NAME)
1701                                 printf(" %-*.*s", longest_file, longest_file,
1702                                        suspect->path);
1703                         if (opt & OUTPUT_SHOW_NUMBER)
1704                                 printf(" %*d", max_orig_digits,
1705                                        ent->s_lno + 1 + cnt);
1706
1707                         if (!(opt & OUTPUT_NO_AUTHOR)) {
1708                                 const char *name;
1709                                 int pad;
1710                                 if (opt & OUTPUT_SHOW_EMAIL)
1711                                         name = ci.author_mail;
1712                                 else
1713                                         name = ci.author;
1714                                 pad = longest_author - utf8_strwidth(name);
1715                                 printf(" (%s%*s %10s",
1716                                        name, pad, "",
1717                                        format_time(ci.author_time,
1718                                                    ci.author_tz,
1719                                                    show_raw_time));
1720                         }
1721                         printf(" %*d) ",
1722                                max_digits, ent->lno + 1 + cnt);
1723                 }
1724                 do {
1725                         ch = *cp++;
1726                         putchar(ch);
1727                 } while (ch != '\n' &&
1728                          cp < sb->final_buf + sb->final_buf_size);
1729         }
1730
1731         if (sb->final_buf_size && cp[-1] != '\n')
1732                 putchar('\n');
1733 }
1734
1735 static void output(struct scoreboard *sb, int option)
1736 {
1737         struct blame_entry *ent;
1738
1739         if (option & OUTPUT_PORCELAIN) {
1740                 for (ent = sb->ent; ent; ent = ent->next) {
1741                         struct blame_entry *oth;
1742                         struct origin *suspect = ent->suspect;
1743                         struct commit *commit = suspect->commit;
1744                         if (commit->object.flags & MORE_THAN_ONE_PATH)
1745                                 continue;
1746                         for (oth = ent->next; oth; oth = oth->next) {
1747                                 if ((oth->suspect->commit != commit) ||
1748                                     !strcmp(oth->suspect->path, suspect->path))
1749                                         continue;
1750                                 commit->object.flags |= MORE_THAN_ONE_PATH;
1751                                 break;
1752                         }
1753                 }
1754         }
1755
1756         for (ent = sb->ent; ent; ent = ent->next) {
1757                 if (option & OUTPUT_PORCELAIN)
1758                         emit_porcelain(sb, ent);
1759                 else {
1760                         emit_other(sb, ent, option);
1761                 }
1762         }
1763 }
1764
1765 /*
1766  * To allow quick access to the contents of nth line in the
1767  * final image, prepare an index in the scoreboard.
1768  */
1769 static int prepare_lines(struct scoreboard *sb)
1770 {
1771         const char *buf = sb->final_buf;
1772         unsigned long len = sb->final_buf_size;
1773         int num = 0, incomplete = 0, bol = 1;
1774
1775         if (len && buf[len-1] != '\n')
1776                 incomplete++; /* incomplete line at the end */
1777         while (len--) {
1778                 if (bol) {
1779                         sb->lineno = xrealloc(sb->lineno,
1780                                               sizeof(int *) * (num + 1));
1781                         sb->lineno[num] = buf - sb->final_buf;
1782                         bol = 0;
1783                 }
1784                 if (*buf++ == '\n') {
1785                         num++;
1786                         bol = 1;
1787                 }
1788         }
1789         sb->lineno = xrealloc(sb->lineno,
1790                               sizeof(int *) * (num + incomplete + 1));
1791         sb->lineno[num + incomplete] = buf - sb->final_buf;
1792         sb->num_lines = num + incomplete;
1793         return sb->num_lines;
1794 }
1795
1796 /*
1797  * Add phony grafts for use with -S; this is primarily to
1798  * support git's cvsserver that wants to give a linear history
1799  * to its clients.
1800  */
1801 static int read_ancestry(const char *graft_file)
1802 {
1803         FILE *fp = fopen(graft_file, "r");
1804         char buf[1024];
1805         if (!fp)
1806                 return -1;
1807         while (fgets(buf, sizeof(buf), fp)) {
1808                 /* The format is just "Commit Parent1 Parent2 ...\n" */
1809                 int len = strlen(buf);
1810                 struct commit_graft *graft = read_graft_line(buf, len);
1811                 if (graft)
1812                         register_commit_graft(graft, 0);
1813         }
1814         fclose(fp);
1815         return 0;
1816 }
1817
1818 /*
1819  * How many columns do we need to show line numbers in decimal?
1820  */
1821 static int lineno_width(int lines)
1822 {
1823         int i, width;
1824
1825         for (width = 1, i = 10; i <= lines; width++)
1826                 i *= 10;
1827         return width;
1828 }
1829
1830 /*
1831  * How many columns do we need to show line numbers, authors,
1832  * and filenames?
1833  */
1834 static void find_alignment(struct scoreboard *sb, int *option)
1835 {
1836         int longest_src_lines = 0;
1837         int longest_dst_lines = 0;
1838         unsigned largest_score = 0;
1839         struct blame_entry *e;
1840
1841         for (e = sb->ent; e; e = e->next) {
1842                 struct origin *suspect = e->suspect;
1843                 struct commit_info ci;
1844                 int num;
1845
1846                 if (strcmp(suspect->path, sb->path))
1847                         *option |= OUTPUT_SHOW_NAME;
1848                 num = strlen(suspect->path);
1849                 if (longest_file < num)
1850                         longest_file = num;
1851                 if (!(suspect->commit->object.flags & METAINFO_SHOWN)) {
1852                         suspect->commit->object.flags |= METAINFO_SHOWN;
1853                         get_commit_info(suspect->commit, &ci, 1);
1854                         if (*option & OUTPUT_SHOW_EMAIL)
1855                                 num = utf8_strwidth(ci.author_mail);
1856                         else
1857                                 num = utf8_strwidth(ci.author);
1858                         if (longest_author < num)
1859                                 longest_author = num;
1860                 }
1861                 num = e->s_lno + e->num_lines;
1862                 if (longest_src_lines < num)
1863                         longest_src_lines = num;
1864                 num = e->lno + e->num_lines;
1865                 if (longest_dst_lines < num)
1866                         longest_dst_lines = num;
1867                 if (largest_score < ent_score(sb, e))
1868                         largest_score = ent_score(sb, e);
1869         }
1870         max_orig_digits = lineno_width(longest_src_lines);
1871         max_digits = lineno_width(longest_dst_lines);
1872         max_score_digits = lineno_width(largest_score);
1873 }
1874
1875 /*
1876  * For debugging -- origin is refcounted, and this asserts that
1877  * we do not underflow.
1878  */
1879 static void sanity_check_refcnt(struct scoreboard *sb)
1880 {
1881         int baa = 0;
1882         struct blame_entry *ent;
1883
1884         for (ent = sb->ent; ent; ent = ent->next) {
1885                 /* Nobody should have zero or negative refcnt */
1886                 if (ent->suspect->refcnt <= 0) {
1887                         fprintf(stderr, "%s in %s has negative refcnt %d\n",
1888                                 ent->suspect->path,
1889                                 sha1_to_hex(ent->suspect->commit->object.sha1),
1890                                 ent->suspect->refcnt);
1891                         baa = 1;
1892                 }
1893         }
1894         if (baa) {
1895                 int opt = 0160;
1896                 find_alignment(sb, &opt);
1897                 output(sb, opt);
1898                 die("Baa %d!", baa);
1899         }
1900 }
1901
1902 /*
1903  * Used for the command line parsing; check if the path exists
1904  * in the working tree.
1905  */
1906 static int has_string_in_work_tree(const char *path)
1907 {
1908         struct stat st;
1909         return !lstat(path, &st);
1910 }
1911
1912 static unsigned parse_score(const char *arg)
1913 {
1914         char *end;
1915         unsigned long score = strtoul(arg, &end, 10);
1916         if (*end)
1917                 return 0;
1918         return score;
1919 }
1920
1921 static const char *add_prefix(const char *prefix, const char *path)
1922 {
1923         return prefix_path(prefix, prefix ? strlen(prefix) : 0, path);
1924 }
1925
1926 /*
1927  * Parsing of (comma separated) one item in the -L option
1928  */
1929 static const char *parse_loc(const char *spec,
1930                              struct scoreboard *sb, long lno,
1931                              long begin, long *ret)
1932 {
1933         char *term;
1934         const char *line;
1935         long num;
1936         int reg_error;
1937         regex_t regexp;
1938         regmatch_t match[1];
1939
1940         /* Allow "-L <something>,+20" to mean starting at <something>
1941          * for 20 lines, or "-L <something>,-5" for 5 lines ending at
1942          * <something>.
1943          */
1944         if (1 < begin && (spec[0] == '+' || spec[0] == '-')) {
1945                 num = strtol(spec + 1, &term, 10);
1946                 if (term != spec + 1) {
1947                         if (spec[0] == '-')
1948                                 num = 0 - num;
1949                         if (0 < num)
1950                                 *ret = begin + num - 2;
1951                         else if (!num)
1952                                 *ret = begin;
1953                         else
1954                                 *ret = begin + num;
1955                         return term;
1956                 }
1957                 return spec;
1958         }
1959         num = strtol(spec, &term, 10);
1960         if (term != spec) {
1961                 *ret = num;
1962                 return term;
1963         }
1964         if (spec[0] != '/')
1965                 return spec;
1966
1967         /* it could be a regexp of form /.../ */
1968         for (term = (char *) spec + 1; *term && *term != '/'; term++) {
1969                 if (*term == '\\')
1970                         term++;
1971         }
1972         if (*term != '/')
1973                 return spec;
1974
1975         /* try [spec+1 .. term-1] as regexp */
1976         *term = 0;
1977         begin--; /* input is in human terms */
1978         line = nth_line(sb, begin);
1979
1980         if (!(reg_error = regcomp(&regexp, spec + 1, REG_NEWLINE)) &&
1981             !(reg_error = regexec(&regexp, line, 1, match, 0))) {
1982                 const char *cp = line + match[0].rm_so;
1983                 const char *nline;
1984
1985                 while (begin++ < lno) {
1986                         nline = nth_line(sb, begin);
1987                         if (line <= cp && cp < nline)
1988                                 break;
1989                         line = nline;
1990                 }
1991                 *ret = begin;
1992                 regfree(&regexp);
1993                 *term++ = '/';
1994                 return term;
1995         }
1996         else {
1997                 char errbuf[1024];
1998                 regerror(reg_error, &regexp, errbuf, 1024);
1999                 die("-L parameter '%s': %s", spec + 1, errbuf);
2000         }
2001 }
2002
2003 /*
2004  * Parsing of -L option
2005  */
2006 static void prepare_blame_range(struct scoreboard *sb,
2007                                 const char *bottomtop,
2008                                 long lno,
2009                                 long *bottom, long *top)
2010 {
2011         const char *term;
2012
2013         term = parse_loc(bottomtop, sb, lno, 1, bottom);
2014         if (*term == ',') {
2015                 term = parse_loc(term + 1, sb, lno, *bottom + 1, top);
2016                 if (*term)
2017                         usage(blame_usage);
2018         }
2019         if (*term)
2020                 usage(blame_usage);
2021 }
2022
2023 static int git_blame_config(const char *var, const char *value, void *cb)
2024 {
2025         if (!strcmp(var, "blame.showroot")) {
2026                 show_root = git_config_bool(var, value);
2027                 return 0;
2028         }
2029         if (!strcmp(var, "blame.blankboundary")) {
2030                 blank_boundary = git_config_bool(var, value);
2031                 return 0;
2032         }
2033         if (!strcmp(var, "blame.date")) {
2034                 if (!value)
2035                         return config_error_nonbool(var);
2036                 blame_date_mode = parse_date_format(value);
2037                 return 0;
2038         }
2039
2040         switch (userdiff_config(var, value)) {
2041         case 0:
2042                 break;
2043         case -1:
2044                 return -1;
2045         default:
2046                 return 0;
2047         }
2048
2049         return git_default_config(var, value, cb);
2050 }
2051
2052 /*
2053  * Prepare a dummy commit that represents the work tree (or staged) item.
2054  * Note that annotating work tree item never works in the reverse.
2055  */
2056 static struct commit *fake_working_tree_commit(struct diff_options *opt,
2057                                                const char *path,
2058                                                const char *contents_from)
2059 {
2060         struct commit *commit;
2061         struct origin *origin;
2062         unsigned char head_sha1[20];
2063         struct strbuf buf = STRBUF_INIT;
2064         const char *ident;
2065         time_t now;
2066         int size, len;
2067         struct cache_entry *ce;
2068         unsigned mode;
2069
2070         if (get_sha1("HEAD", head_sha1))
2071                 die("No such ref: HEAD");
2072
2073         time(&now);
2074         commit = xcalloc(1, sizeof(*commit));
2075         commit->parents = xcalloc(1, sizeof(*commit->parents));
2076         commit->parents->item = lookup_commit_reference(head_sha1);
2077         commit->object.parsed = 1;
2078         commit->date = now;
2079         commit->object.type = OBJ_COMMIT;
2080
2081         origin = make_origin(commit, path);
2082
2083         if (!contents_from || strcmp("-", contents_from)) {
2084                 struct stat st;
2085                 const char *read_from;
2086                 unsigned long buf_len;
2087
2088                 if (contents_from) {
2089                         if (stat(contents_from, &st) < 0)
2090                                 die_errno("Cannot stat '%s'", contents_from);
2091                         read_from = contents_from;
2092                 }
2093                 else {
2094                         if (lstat(path, &st) < 0)
2095                                 die_errno("Cannot lstat '%s'", path);
2096                         read_from = path;
2097                 }
2098                 mode = canon_mode(st.st_mode);
2099
2100                 switch (st.st_mode & S_IFMT) {
2101                 case S_IFREG:
2102                         if (DIFF_OPT_TST(opt, ALLOW_TEXTCONV) &&
2103                             textconv_object(read_from, mode, null_sha1, &buf.buf, &buf_len))
2104                                 buf.len = buf_len;
2105                         else if (strbuf_read_file(&buf, read_from, st.st_size) != st.st_size)
2106                                 die_errno("cannot open or read '%s'", read_from);
2107                         break;
2108                 case S_IFLNK:
2109                         if (strbuf_readlink(&buf, read_from, st.st_size) < 0)
2110                                 die_errno("cannot readlink '%s'", read_from);
2111                         break;
2112                 default:
2113                         die("unsupported file type %s", read_from);
2114                 }
2115         }
2116         else {
2117                 /* Reading from stdin */
2118                 contents_from = "standard input";
2119                 mode = 0;
2120                 if (strbuf_read(&buf, 0, 0) < 0)
2121                         die_errno("failed to read from stdin");
2122         }
2123         convert_to_git(path, buf.buf, buf.len, &buf, 0);
2124         origin->file.ptr = buf.buf;
2125         origin->file.size = buf.len;
2126         pretend_sha1_file(buf.buf, buf.len, OBJ_BLOB, origin->blob_sha1);
2127         commit->util = origin;
2128
2129         /*
2130          * Read the current index, replace the path entry with
2131          * origin->blob_sha1 without mucking with its mode or type
2132          * bits; we are not going to write this index out -- we just
2133          * want to run "diff-index --cached".
2134          */
2135         discard_cache();
2136         read_cache();
2137
2138         len = strlen(path);
2139         if (!mode) {
2140                 int pos = cache_name_pos(path, len);
2141                 if (0 <= pos)
2142                         mode = active_cache[pos]->ce_mode;
2143                 else
2144                         /* Let's not bother reading from HEAD tree */
2145                         mode = S_IFREG | 0644;
2146         }
2147         size = cache_entry_size(len);
2148         ce = xcalloc(1, size);
2149         hashcpy(ce->sha1, origin->blob_sha1);
2150         memcpy(ce->name, path, len);
2151         ce->ce_flags = create_ce_flags(len, 0);
2152         ce->ce_mode = create_ce_mode(mode);
2153         add_cache_entry(ce, ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE);
2154
2155         /*
2156          * We are not going to write this out, so this does not matter
2157          * right now, but someday we might optimize diff-index --cached
2158          * with cache-tree information.
2159          */
2160         cache_tree_invalidate_path(active_cache_tree, path);
2161
2162         commit->buffer = xmalloc(400);
2163         ident = fmt_ident("Not Committed Yet", "not.committed.yet", NULL, 0);
2164         snprintf(commit->buffer, 400,
2165                 "tree 0000000000000000000000000000000000000000\n"
2166                 "parent %s\n"
2167                 "author %s\n"
2168                 "committer %s\n\n"
2169                 "Version of %s from %s\n",
2170                 sha1_to_hex(head_sha1),
2171                 ident, ident, path, contents_from ? contents_from : path);
2172         return commit;
2173 }
2174
2175 static const char *prepare_final(struct scoreboard *sb)
2176 {
2177         int i;
2178         const char *final_commit_name = NULL;
2179         struct rev_info *revs = sb->revs;
2180
2181         /*
2182          * There must be one and only one positive commit in the
2183          * revs->pending array.
2184          */
2185         for (i = 0; i < revs->pending.nr; i++) {
2186                 struct object *obj = revs->pending.objects[i].item;
2187                 if (obj->flags & UNINTERESTING)
2188                         continue;
2189                 while (obj->type == OBJ_TAG)
2190                         obj = deref_tag(obj, NULL, 0);
2191                 if (obj->type != OBJ_COMMIT)
2192                         die("Non commit %s?", revs->pending.objects[i].name);
2193                 if (sb->final)
2194                         die("More than one commit to dig from %s and %s?",
2195                             revs->pending.objects[i].name,
2196                             final_commit_name);
2197                 sb->final = (struct commit *) obj;
2198                 final_commit_name = revs->pending.objects[i].name;
2199         }
2200         return final_commit_name;
2201 }
2202
2203 static const char *prepare_initial(struct scoreboard *sb)
2204 {
2205         int i;
2206         const char *final_commit_name = NULL;
2207         struct rev_info *revs = sb->revs;
2208
2209         /*
2210          * There must be one and only one negative commit, and it must be
2211          * the boundary.
2212          */
2213         for (i = 0; i < revs->pending.nr; i++) {
2214                 struct object *obj = revs->pending.objects[i].item;
2215                 if (!(obj->flags & UNINTERESTING))
2216                         continue;
2217                 while (obj->type == OBJ_TAG)
2218                         obj = deref_tag(obj, NULL, 0);
2219                 if (obj->type != OBJ_COMMIT)
2220                         die("Non commit %s?", revs->pending.objects[i].name);
2221                 if (sb->final)
2222                         die("More than one commit to dig down to %s and %s?",
2223                             revs->pending.objects[i].name,
2224                             final_commit_name);
2225                 sb->final = (struct commit *) obj;
2226                 final_commit_name = revs->pending.objects[i].name;
2227         }
2228         if (!final_commit_name)
2229                 die("No commit to dig down to?");
2230         return final_commit_name;
2231 }
2232
2233 static int blame_copy_callback(const struct option *option, const char *arg, int unset)
2234 {
2235         int *opt = option->value;
2236
2237         /*
2238          * -C enables copy from removed files;
2239          * -C -C enables copy from existing files, but only
2240          *       when blaming a new file;
2241          * -C -C -C enables copy from existing files for
2242          *          everybody
2243          */
2244         if (*opt & PICKAXE_BLAME_COPY_HARDER)
2245                 *opt |= PICKAXE_BLAME_COPY_HARDEST;
2246         if (*opt & PICKAXE_BLAME_COPY)
2247                 *opt |= PICKAXE_BLAME_COPY_HARDER;
2248         *opt |= PICKAXE_BLAME_COPY | PICKAXE_BLAME_MOVE;
2249
2250         if (arg)
2251                 blame_copy_score = parse_score(arg);
2252         return 0;
2253 }
2254
2255 static int blame_move_callback(const struct option *option, const char *arg, int unset)
2256 {
2257         int *opt = option->value;
2258
2259         *opt |= PICKAXE_BLAME_MOVE;
2260
2261         if (arg)
2262                 blame_move_score = parse_score(arg);
2263         return 0;
2264 }
2265
2266 static int blame_bottomtop_callback(const struct option *option, const char *arg, int unset)
2267 {
2268         const char **bottomtop = option->value;
2269         if (!arg)
2270                 return -1;
2271         if (*bottomtop)
2272                 die("More than one '-L n,m' option given");
2273         *bottomtop = arg;
2274         return 0;
2275 }
2276
2277 int cmd_blame(int argc, const char **argv, const char *prefix)
2278 {
2279         struct rev_info revs;
2280         const char *path;
2281         struct scoreboard sb;
2282         struct origin *o;
2283         struct blame_entry *ent;
2284         long dashdash_pos, bottom, top, lno;
2285         const char *final_commit_name = NULL;
2286         enum object_type type;
2287
2288         static const char *bottomtop = NULL;
2289         static int output_option = 0, opt = 0;
2290         static int show_stats = 0;
2291         static const char *revs_file = NULL;
2292         static const char *contents_from = NULL;
2293         static const struct option options[] = {
2294                 OPT_BOOLEAN(0, "incremental", &incremental, "Show blame entries as we find them, incrementally"),
2295                 OPT_BOOLEAN('b', NULL, &blank_boundary, "Show blank SHA-1 for boundary commits (Default: off)"),
2296                 OPT_BOOLEAN(0, "root", &show_root, "Do not treat root commits as boundaries (Default: off)"),
2297                 OPT_BOOLEAN(0, "show-stats", &show_stats, "Show work cost statistics"),
2298                 OPT_BIT(0, "score-debug", &output_option, "Show output score for blame entries", OUTPUT_SHOW_SCORE),
2299                 OPT_BIT('f', "show-name", &output_option, "Show original filename (Default: auto)", OUTPUT_SHOW_NAME),
2300                 OPT_BIT('n', "show-number", &output_option, "Show original linenumber (Default: off)", OUTPUT_SHOW_NUMBER),
2301                 OPT_BIT('p', "porcelain", &output_option, "Show in a format designed for machine consumption", OUTPUT_PORCELAIN),
2302                 OPT_BIT('c', NULL, &output_option, "Use the same output mode as git-annotate (Default: off)", OUTPUT_ANNOTATE_COMPAT),
2303                 OPT_BIT('t', NULL, &output_option, "Show raw timestamp (Default: off)", OUTPUT_RAW_TIMESTAMP),
2304                 OPT_BIT('l', NULL, &output_option, "Show long commit SHA1 (Default: off)", OUTPUT_LONG_OBJECT_NAME),
2305                 OPT_BIT('s', NULL, &output_option, "Suppress author name and timestamp (Default: off)", OUTPUT_NO_AUTHOR),
2306                 OPT_BIT('e', "show-email", &output_option, "Show author email instead of name (Default: off)", OUTPUT_SHOW_EMAIL),
2307                 OPT_BIT('w', NULL, &xdl_opts, "Ignore whitespace differences", XDF_IGNORE_WHITESPACE),
2308                 OPT_STRING('S', NULL, &revs_file, "file", "Use revisions from <file> instead of calling git-rev-list"),
2309                 OPT_STRING(0, "contents", &contents_from, "file", "Use <file>'s contents as the final image"),
2310                 { OPTION_CALLBACK, 'C', NULL, &opt, "score", "Find line copies within and across files", PARSE_OPT_OPTARG, blame_copy_callback },
2311                 { OPTION_CALLBACK, 'M', NULL, &opt, "score", "Find line movements within and across files", PARSE_OPT_OPTARG, blame_move_callback },
2312                 OPT_CALLBACK('L', NULL, &bottomtop, "n,m", "Process only line range n,m, counting from 1", blame_bottomtop_callback),
2313                 OPT_END()
2314         };
2315
2316         struct parse_opt_ctx_t ctx;
2317         int cmd_is_annotate = !strcmp(argv[0], "annotate");
2318
2319         git_config(git_blame_config, NULL);
2320         init_revisions(&revs, NULL);
2321         revs.date_mode = blame_date_mode;
2322         DIFF_OPT_SET(&revs.diffopt, ALLOW_TEXTCONV);
2323
2324         save_commit_buffer = 0;
2325         dashdash_pos = 0;
2326
2327         parse_options_start(&ctx, argc, argv, prefix, options,
2328                             PARSE_OPT_KEEP_DASHDASH | PARSE_OPT_KEEP_ARGV0);
2329         for (;;) {
2330                 switch (parse_options_step(&ctx, options, blame_opt_usage)) {
2331                 case PARSE_OPT_HELP:
2332                         exit(129);
2333                 case PARSE_OPT_DONE:
2334                         if (ctx.argv[0])
2335                                 dashdash_pos = ctx.cpidx;
2336                         goto parse_done;
2337                 }
2338
2339                 if (!strcmp(ctx.argv[0], "--reverse")) {
2340                         ctx.argv[0] = "--children";
2341                         reverse = 1;
2342                 }
2343                 parse_revision_opt(&revs, &ctx, options, blame_opt_usage);
2344         }
2345 parse_done:
2346         argc = parse_options_end(&ctx);
2347
2348         if (revs_file && read_ancestry(revs_file))
2349                 die_errno("reading graft file '%s' failed", revs_file);
2350
2351         if (cmd_is_annotate) {
2352                 output_option |= OUTPUT_ANNOTATE_COMPAT;
2353                 blame_date_mode = DATE_ISO8601;
2354         } else {
2355                 blame_date_mode = revs.date_mode;
2356         }
2357
2358         /* The maximum width used to show the dates */
2359         switch (blame_date_mode) {
2360         case DATE_RFC2822:
2361                 blame_date_width = sizeof("Thu, 19 Oct 2006 16:00:04 -0700");
2362                 break;
2363         case DATE_ISO8601:
2364                 blame_date_width = sizeof("2006-10-19 16:00:04 -0700");
2365                 break;
2366         case DATE_RAW:
2367                 blame_date_width = sizeof("1161298804 -0700");
2368                 break;
2369         case DATE_SHORT:
2370                 blame_date_width = sizeof("2006-10-19");
2371                 break;
2372         case DATE_RELATIVE:
2373                 /* "normal" is used as the fallback for "relative" */
2374         case DATE_LOCAL:
2375         case DATE_NORMAL:
2376                 blame_date_width = sizeof("Thu Oct 19 16:00:04 2006 -0700");
2377                 break;
2378         }
2379         blame_date_width -= 1; /* strip the null */
2380
2381         if (DIFF_OPT_TST(&revs.diffopt, FIND_COPIES_HARDER))
2382                 opt |= (PICKAXE_BLAME_COPY | PICKAXE_BLAME_MOVE |
2383                         PICKAXE_BLAME_COPY_HARDER);
2384
2385         if (!blame_move_score)
2386                 blame_move_score = BLAME_DEFAULT_MOVE_SCORE;
2387         if (!blame_copy_score)
2388                 blame_copy_score = BLAME_DEFAULT_COPY_SCORE;
2389
2390         /*
2391          * We have collected options unknown to us in argv[1..unk]
2392          * which are to be passed to revision machinery if we are
2393          * going to do the "bottom" processing.
2394          *
2395          * The remaining are:
2396          *
2397          * (1) if dashdash_pos != 0, it is either
2398          *     "blame [revisions] -- <path>" or
2399          *     "blame -- <path> <rev>"
2400          *
2401          * (2) otherwise, it is one of the two:
2402          *     "blame [revisions] <path>"
2403          *     "blame <path> <rev>"
2404          *
2405          * Note that we must strip out <path> from the arguments: we do not
2406          * want the path pruning but we may want "bottom" processing.
2407          */
2408         if (dashdash_pos) {
2409                 switch (argc - dashdash_pos - 1) {
2410                 case 2: /* (1b) */
2411                         if (argc != 4)
2412                                 usage_with_options(blame_opt_usage, options);
2413                         /* reorder for the new way: <rev> -- <path> */
2414                         argv[1] = argv[3];
2415                         argv[3] = argv[2];
2416                         argv[2] = "--";
2417                         /* FALLTHROUGH */
2418                 case 1: /* (1a) */
2419                         path = add_prefix(prefix, argv[--argc]);
2420                         argv[argc] = NULL;
2421                         break;
2422                 default:
2423                         usage_with_options(blame_opt_usage, options);
2424                 }
2425         } else {
2426                 if (argc < 2)
2427                         usage_with_options(blame_opt_usage, options);
2428                 path = add_prefix(prefix, argv[argc - 1]);
2429                 if (argc == 3 && !has_string_in_work_tree(path)) { /* (2b) */
2430                         path = add_prefix(prefix, argv[1]);
2431                         argv[1] = argv[2];
2432                 }
2433                 argv[argc - 1] = "--";
2434
2435                 setup_work_tree();
2436                 if (!has_string_in_work_tree(path))
2437                         die_errno("cannot stat path '%s'", path);
2438         }
2439
2440         revs.disable_stdin = 1;
2441         setup_revisions(argc, argv, &revs, NULL);
2442         memset(&sb, 0, sizeof(sb));
2443
2444         sb.revs = &revs;
2445         if (!reverse)
2446                 final_commit_name = prepare_final(&sb);
2447         else if (contents_from)
2448                 die("--contents and --children do not blend well.");
2449         else
2450                 final_commit_name = prepare_initial(&sb);
2451
2452         if (!sb.final) {
2453                 /*
2454                  * "--not A B -- path" without anything positive;
2455                  * do not default to HEAD, but use the working tree
2456                  * or "--contents".
2457                  */
2458                 setup_work_tree();
2459                 sb.final = fake_working_tree_commit(&sb.revs->diffopt,
2460                                                     path, contents_from);
2461                 add_pending_object(&revs, &(sb.final->object), ":");
2462         }
2463         else if (contents_from)
2464                 die("Cannot use --contents with final commit object name");
2465
2466         /*
2467          * If we have bottom, this will mark the ancestors of the
2468          * bottom commits we would reach while traversing as
2469          * uninteresting.
2470          */
2471         if (prepare_revision_walk(&revs))
2472                 die("revision walk setup failed");
2473
2474         if (is_null_sha1(sb.final->object.sha1)) {
2475                 char *buf;
2476                 o = sb.final->util;
2477                 buf = xmalloc(o->file.size + 1);
2478                 memcpy(buf, o->file.ptr, o->file.size + 1);
2479                 sb.final_buf = buf;
2480                 sb.final_buf_size = o->file.size;
2481         }
2482         else {
2483                 o = get_origin(&sb, sb.final, path);
2484                 if (fill_blob_sha1_and_mode(o))
2485                         die("no such path %s in %s", path, final_commit_name);
2486
2487                 if (DIFF_OPT_TST(&sb.revs->diffopt, ALLOW_TEXTCONV) &&
2488                     textconv_object(path, o->mode, o->blob_sha1, (char **) &sb.final_buf,
2489                                     &sb.final_buf_size))
2490                         ;
2491                 else
2492                         sb.final_buf = read_sha1_file(o->blob_sha1, &type,
2493                                                       &sb.final_buf_size);
2494
2495                 if (!sb.final_buf)
2496                         die("Cannot read blob %s for path %s",
2497                             sha1_to_hex(o->blob_sha1),
2498                             path);
2499         }
2500         num_read_blob++;
2501         lno = prepare_lines(&sb);
2502
2503         bottom = top = 0;
2504         if (bottomtop)
2505                 prepare_blame_range(&sb, bottomtop, lno, &bottom, &top);
2506         if (bottom && top && top < bottom) {
2507                 long tmp;
2508                 tmp = top; top = bottom; bottom = tmp;
2509         }
2510         if (bottom < 1)
2511                 bottom = 1;
2512         if (top < 1)
2513                 top = lno;
2514         bottom--;
2515         if (lno < top || lno < bottom)
2516                 die("file %s has only %lu lines", path, lno);
2517
2518         ent = xcalloc(1, sizeof(*ent));
2519         ent->lno = bottom;
2520         ent->num_lines = top - bottom;
2521         ent->suspect = o;
2522         ent->s_lno = bottom;
2523
2524         sb.ent = ent;
2525         sb.path = path;
2526
2527         read_mailmap(&mailmap, NULL);
2528
2529         if (!incremental)
2530                 setup_pager();
2531
2532         assign_blame(&sb, opt);
2533
2534         if (incremental)
2535                 return 0;
2536
2537         coalesce(&sb);
2538
2539         if (!(output_option & OUTPUT_PORCELAIN))
2540                 find_alignment(&sb, &output_option);
2541
2542         output(&sb, output_option);
2543         free((void *)sb.final_buf);
2544         for (ent = sb.ent; ent; ) {
2545                 struct blame_entry *e = ent->next;
2546                 free(ent);
2547                 ent = e;
2548         }
2549
2550         if (show_stats) {
2551                 printf("num read blob: %d\n", num_read_blob);
2552                 printf("num get patch: %d\n", num_get_patch);
2553                 printf("num commits: %d\n", num_commits);
2554         }
2555         return 0;
2556 }