Merge branch 'nk/ref-doc' into maint
[git] / builtin / blame.c
1 /*
2  * Blame
3  *
4  * Copyright (c) 2006, Junio C Hamano
5  */
6
7 #include "cache.h"
8 #include "builtin.h"
9 #include "blob.h"
10 #include "commit.h"
11 #include "tag.h"
12 #include "tree-walk.h"
13 #include "diff.h"
14 #include "diffcore.h"
15 #include "revision.h"
16 #include "quote.h"
17 #include "xdiff-interface.h"
18 #include "cache-tree.h"
19 #include "string-list.h"
20 #include "mailmap.h"
21 #include "parse-options.h"
22 #include "utf8.h"
23 #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 static int abbrev = -1;
45
46 static enum date_mode blame_date_mode = DATE_ISO8601;
47 static size_t blame_date_width;
48
49 static struct string_list mailmap;
50
51 #ifndef DEBUG
52 #define DEBUG 0
53 #endif
54
55 /* stats */
56 static int num_read_blob;
57 static int num_get_patch;
58 static int num_commits;
59
60 #define PICKAXE_BLAME_MOVE              01
61 #define PICKAXE_BLAME_COPY              02
62 #define PICKAXE_BLAME_COPY_HARDER       04
63 #define PICKAXE_BLAME_COPY_HARDEST      010
64
65 /*
66  * blame for a blame_entry with score lower than these thresholds
67  * is not passed to the parent using move/copy logic.
68  */
69 static unsigned blame_move_score;
70 static unsigned blame_copy_score;
71 #define BLAME_DEFAULT_MOVE_SCORE        20
72 #define BLAME_DEFAULT_COPY_SCORE        40
73
74 /* bits #0..7 in revision.h, #8..11 used for merge_bases() in commit.c */
75 #define METAINFO_SHOWN          (1u<<12)
76 #define MORE_THAN_ONE_PATH      (1u<<13)
77
78 /*
79  * One blob in a commit that is being suspected
80  */
81 struct origin {
82         int refcnt;
83         struct origin *previous;
84         struct commit *commit;
85         mmfile_t file;
86         unsigned char blob_sha1[20];
87         unsigned mode;
88         char path[FLEX_ARRAY];
89 };
90
91 /*
92  * Prepare diff_filespec and convert it using diff textconv API
93  * if the textconv driver exists.
94  * Return 1 if the conversion succeeds, 0 otherwise.
95  */
96 int textconv_object(const char *path,
97                     unsigned mode,
98                     const unsigned char *sha1,
99                     char **buf,
100                     unsigned long *buf_size)
101 {
102         struct diff_filespec *df;
103         struct userdiff_driver *textconv;
104
105         df = alloc_filespec(path);
106         fill_filespec(df, sha1, mode);
107         textconv = get_textconv(df);
108         if (!textconv) {
109                 free_filespec(df);
110                 return 0;
111         }
112
113         *buf_size = fill_textconv(textconv, df, buf);
114         free_filespec(df);
115         return 1;
116 }
117
118 /*
119  * Given an origin, prepare mmfile_t structure to be used by the
120  * diff machinery
121  */
122 static void fill_origin_blob(struct diff_options *opt,
123                              struct origin *o, mmfile_t *file)
124 {
125         if (!o->file.ptr) {
126                 enum object_type type;
127                 unsigned long file_size;
128
129                 num_read_blob++;
130                 if (DIFF_OPT_TST(opt, ALLOW_TEXTCONV) &&
131                     textconv_object(o->path, o->mode, o->blob_sha1, &file->ptr, &file_size))
132                         ;
133                 else
134                         file->ptr = read_sha1_file(o->blob_sha1, &type, &file_size);
135                 file->size = file_size;
136
137                 if (!file->ptr)
138                         die("Cannot read blob %s for path %s",
139                             sha1_to_hex(o->blob_sha1),
140                             o->path);
141                 o->file = *file;
142         }
143         else
144                 *file = o->file;
145 }
146
147 /*
148  * Origin is refcounted and usually we keep the blob contents to be
149  * reused.
150  */
151 static inline struct origin *origin_incref(struct origin *o)
152 {
153         if (o)
154                 o->refcnt++;
155         return o;
156 }
157
158 static void origin_decref(struct origin *o)
159 {
160         if (o && --o->refcnt <= 0) {
161                 if (o->previous)
162                         origin_decref(o->previous);
163                 free(o->file.ptr);
164                 free(o);
165         }
166 }
167
168 static void drop_origin_blob(struct origin *o)
169 {
170         if (o->file.ptr) {
171                 free(o->file.ptr);
172                 o->file.ptr = NULL;
173         }
174 }
175
176 /*
177  * Each group of lines is described by a blame_entry; it can be split
178  * as we pass blame to the parents.  They form a linked list in the
179  * scoreboard structure, sorted by the target line number.
180  */
181 struct blame_entry {
182         struct blame_entry *prev;
183         struct blame_entry *next;
184
185         /* the first line of this group in the final image;
186          * internally all line numbers are 0 based.
187          */
188         int lno;
189
190         /* how many lines this group has */
191         int num_lines;
192
193         /* the commit that introduced this group into the final image */
194         struct origin *suspect;
195
196         /* true if the suspect is truly guilty; false while we have not
197          * checked if the group came from one of its parents.
198          */
199         char guilty;
200
201         /* true if the entry has been scanned for copies in the current parent
202          */
203         char scanned;
204
205         /* the line number of the first line of this group in the
206          * suspect's file; internally all line numbers are 0 based.
207          */
208         int s_lno;
209
210         /* how significant this entry is -- cached to avoid
211          * scanning the lines over and over.
212          */
213         unsigned score;
214 };
215
216 /*
217  * The current state of the blame assignment.
218  */
219 struct scoreboard {
220         /* the final commit (i.e. where we started digging from) */
221         struct commit *final;
222         struct rev_info *revs;
223         const char *path;
224
225         /*
226          * The contents in the final image.
227          * Used by many functions to obtain contents of the nth line,
228          * indexed with scoreboard.lineno[blame_entry.lno].
229          */
230         const char *final_buf;
231         unsigned long final_buf_size;
232
233         /* linked list of blames */
234         struct blame_entry *ent;
235
236         /* look-up a line in the final buffer */
237         int num_lines;
238         int *lineno;
239 };
240
241 static inline int same_suspect(struct origin *a, struct origin *b)
242 {
243         if (a == b)
244                 return 1;
245         if (a->commit != b->commit)
246                 return 0;
247         return !strcmp(a->path, b->path);
248 }
249
250 static void sanity_check_refcnt(struct scoreboard *);
251
252 /*
253  * If two blame entries that are next to each other came from
254  * contiguous lines in the same origin (i.e. <commit, path> pair),
255  * merge them together.
256  */
257 static void coalesce(struct scoreboard *sb)
258 {
259         struct blame_entry *ent, *next;
260
261         for (ent = sb->ent; ent && (next = ent->next); ent = next) {
262                 if (same_suspect(ent->suspect, next->suspect) &&
263                     ent->guilty == next->guilty &&
264                     ent->s_lno + ent->num_lines == next->s_lno) {
265                         ent->num_lines += next->num_lines;
266                         ent->next = next->next;
267                         if (ent->next)
268                                 ent->next->prev = ent;
269                         origin_decref(next->suspect);
270                         free(next);
271                         ent->score = 0;
272                         next = ent; /* again */
273                 }
274         }
275
276         if (DEBUG) /* sanity */
277                 sanity_check_refcnt(sb);
278 }
279
280 /*
281  * Given a commit and a path in it, create a new origin structure.
282  * The callers that add blame to the scoreboard should use
283  * get_origin() to obtain shared, refcounted copy instead of calling
284  * this function directly.
285  */
286 static struct origin *make_origin(struct commit *commit, const char *path)
287 {
288         struct origin *o;
289         o = xcalloc(1, sizeof(*o) + strlen(path) + 1);
290         o->commit = commit;
291         o->refcnt = 1;
292         strcpy(o->path, path);
293         return o;
294 }
295
296 /*
297  * Locate an existing origin or create a new one.
298  */
299 static struct origin *get_origin(struct scoreboard *sb,
300                                  struct commit *commit,
301                                  const char *path)
302 {
303         struct blame_entry *e;
304
305         for (e = sb->ent; e; e = e->next) {
306                 if (e->suspect->commit == commit &&
307                     !strcmp(e->suspect->path, path))
308                         return origin_incref(e->suspect);
309         }
310         return make_origin(commit, path);
311 }
312
313 /*
314  * Fill the blob_sha1 field of an origin if it hasn't, so that later
315  * call to fill_origin_blob() can use it to locate the data.  blob_sha1
316  * for an origin is also used to pass the blame for the entire file to
317  * the parent to detect the case where a child's blob is identical to
318  * that of its parent's.
319  *
320  * This also fills origin->mode for corresponding tree path.
321  */
322 static int fill_blob_sha1_and_mode(struct origin *origin)
323 {
324         if (!is_null_sha1(origin->blob_sha1))
325                 return 0;
326         if (get_tree_entry(origin->commit->object.sha1,
327                            origin->path,
328                            origin->blob_sha1, &origin->mode))
329                 goto error_out;
330         if (sha1_object_info(origin->blob_sha1, NULL) != OBJ_BLOB)
331                 goto error_out;
332         return 0;
333  error_out:
334         hashclr(origin->blob_sha1);
335         origin->mode = S_IFINVALID;
336         return -1;
337 }
338
339 /*
340  * We have an origin -- check if the same path exists in the
341  * parent and return an origin structure to represent it.
342  */
343 static struct origin *find_origin(struct scoreboard *sb,
344                                   struct commit *parent,
345                                   struct origin *origin)
346 {
347         struct origin *porigin = NULL;
348         struct diff_options diff_opts;
349         const char *paths[2];
350
351         if (parent->util) {
352                 /*
353                  * Each commit object can cache one origin in that
354                  * commit.  This is a freestanding copy of origin and
355                  * not refcounted.
356                  */
357                 struct origin *cached = parent->util;
358                 if (!strcmp(cached->path, origin->path)) {
359                         /*
360                          * The same path between origin and its parent
361                          * without renaming -- the most common case.
362                          */
363                         porigin = get_origin(sb, parent, cached->path);
364
365                         /*
366                          * If the origin was newly created (i.e. get_origin
367                          * would call make_origin if none is found in the
368                          * scoreboard), it does not know the blob_sha1/mode,
369                          * so copy it.  Otherwise porigin was in the
370                          * scoreboard and already knows blob_sha1/mode.
371                          */
372                         if (porigin->refcnt == 1) {
373                                 hashcpy(porigin->blob_sha1, cached->blob_sha1);
374                                 porigin->mode = cached->mode;
375                         }
376                         return porigin;
377                 }
378                 /* otherwise it was not very useful; free it */
379                 free(parent->util);
380                 parent->util = NULL;
381         }
382
383         /* See if the origin->path is different between parent
384          * and origin first.  Most of the time they are the
385          * same and diff-tree is fairly efficient about this.
386          */
387         diff_setup(&diff_opts);
388         DIFF_OPT_SET(&diff_opts, RECURSIVE);
389         diff_opts.detect_rename = 0;
390         diff_opts.output_format = DIFF_FORMAT_NO_OUTPUT;
391         paths[0] = origin->path;
392         paths[1] = NULL;
393
394         diff_tree_setup_paths(paths, &diff_opts);
395         if (diff_setup_done(&diff_opts) < 0)
396                 die("diff-setup");
397
398         if (is_null_sha1(origin->commit->object.sha1))
399                 do_diff_cache(parent->tree->object.sha1, &diff_opts);
400         else
401                 diff_tree_sha1(parent->tree->object.sha1,
402                                origin->commit->tree->object.sha1,
403                                "", &diff_opts);
404         diffcore_std(&diff_opts);
405
406         if (!diff_queued_diff.nr) {
407                 /* The path is the same as parent */
408                 porigin = get_origin(sb, parent, origin->path);
409                 hashcpy(porigin->blob_sha1, origin->blob_sha1);
410                 porigin->mode = origin->mode;
411         } else {
412                 /*
413                  * Since origin->path is a pathspec, if the parent
414                  * commit had it as a directory, we will see a whole
415                  * bunch of deletion of files in the directory that we
416                  * do not care about.
417                  */
418                 int i;
419                 struct diff_filepair *p = NULL;
420                 for (i = 0; i < diff_queued_diff.nr; i++) {
421                         const char *name;
422                         p = diff_queued_diff.queue[i];
423                         name = p->one->path ? p->one->path : p->two->path;
424                         if (!strcmp(name, origin->path))
425                                 break;
426                 }
427                 if (!p)
428                         die("internal error in blame::find_origin");
429                 switch (p->status) {
430                 default:
431                         die("internal error in blame::find_origin (%c)",
432                             p->status);
433                 case 'M':
434                         porigin = get_origin(sb, parent, origin->path);
435                         hashcpy(porigin->blob_sha1, p->one->sha1);
436                         porigin->mode = p->one->mode;
437                         break;
438                 case 'A':
439                 case 'T':
440                         /* Did not exist in parent, or type changed */
441                         break;
442                 }
443         }
444         diff_flush(&diff_opts);
445         diff_tree_release_paths(&diff_opts);
446         if (porigin) {
447                 /*
448                  * Create a freestanding copy that is not part of
449                  * the refcounted origin found in the scoreboard, and
450                  * cache it in the commit.
451                  */
452                 struct origin *cached;
453
454                 cached = make_origin(porigin->commit, porigin->path);
455                 hashcpy(cached->blob_sha1, porigin->blob_sha1);
456                 cached->mode = porigin->mode;
457                 parent->util = cached;
458         }
459         return porigin;
460 }
461
462 /*
463  * We have an origin -- find the path that corresponds to it in its
464  * parent and return an origin structure to represent it.
465  */
466 static struct origin *find_rename(struct scoreboard *sb,
467                                   struct commit *parent,
468                                   struct origin *origin)
469 {
470         struct origin *porigin = NULL;
471         struct diff_options diff_opts;
472         int i;
473         const char *paths[2];
474
475         diff_setup(&diff_opts);
476         DIFF_OPT_SET(&diff_opts, RECURSIVE);
477         diff_opts.detect_rename = DIFF_DETECT_RENAME;
478         diff_opts.output_format = DIFF_FORMAT_NO_OUTPUT;
479         diff_opts.single_follow = origin->path;
480         paths[0] = NULL;
481         diff_tree_setup_paths(paths, &diff_opts);
482         if (diff_setup_done(&diff_opts) < 0)
483                 die("diff-setup");
484
485         if (is_null_sha1(origin->commit->object.sha1))
486                 do_diff_cache(parent->tree->object.sha1, &diff_opts);
487         else
488                 diff_tree_sha1(parent->tree->object.sha1,
489                                origin->commit->tree->object.sha1,
490                                "", &diff_opts);
491         diffcore_std(&diff_opts);
492
493         for (i = 0; i < diff_queued_diff.nr; i++) {
494                 struct diff_filepair *p = diff_queued_diff.queue[i];
495                 if ((p->status == 'R' || p->status == 'C') &&
496                     !strcmp(p->two->path, origin->path)) {
497                         porigin = get_origin(sb, parent, p->one->path);
498                         hashcpy(porigin->blob_sha1, p->one->sha1);
499                         porigin->mode = p->one->mode;
500                         break;
501                 }
502         }
503         diff_flush(&diff_opts);
504         diff_tree_release_paths(&diff_opts);
505         return porigin;
506 }
507
508 /*
509  * Link in a new blame entry to the scoreboard.  Entries that cover the
510  * same line range have been removed from the scoreboard previously.
511  */
512 static void add_blame_entry(struct scoreboard *sb, struct blame_entry *e)
513 {
514         struct blame_entry *ent, *prev = NULL;
515
516         origin_incref(e->suspect);
517
518         for (ent = sb->ent; ent && ent->lno < e->lno; ent = ent->next)
519                 prev = ent;
520
521         /* prev, if not NULL, is the last one that is below e */
522         e->prev = prev;
523         if (prev) {
524                 e->next = prev->next;
525                 prev->next = e;
526         }
527         else {
528                 e->next = sb->ent;
529                 sb->ent = e;
530         }
531         if (e->next)
532                 e->next->prev = e;
533 }
534
535 /*
536  * src typically is on-stack; we want to copy the information in it to
537  * a malloced blame_entry that is already on the linked list of the
538  * scoreboard.  The origin of dst loses a refcnt while the origin of src
539  * gains one.
540  */
541 static void dup_entry(struct blame_entry *dst, struct blame_entry *src)
542 {
543         struct blame_entry *p, *n;
544
545         p = dst->prev;
546         n = dst->next;
547         origin_incref(src->suspect);
548         origin_decref(dst->suspect);
549         memcpy(dst, src, sizeof(*src));
550         dst->prev = p;
551         dst->next = n;
552         dst->score = 0;
553 }
554
555 static const char *nth_line(struct scoreboard *sb, int lno)
556 {
557         return sb->final_buf + sb->lineno[lno];
558 }
559
560 /*
561  * It is known that lines between tlno to same came from parent, and e
562  * has an overlap with that range.  it also is known that parent's
563  * line plno corresponds to e's line tlno.
564  *
565  *                <---- e ----->
566  *                   <------>
567  *                   <------------>
568  *             <------------>
569  *             <------------------>
570  *
571  * Split e into potentially three parts; before this chunk, the chunk
572  * to be blamed for the parent, and after that portion.
573  */
574 static void split_overlap(struct blame_entry *split,
575                           struct blame_entry *e,
576                           int tlno, int plno, int same,
577                           struct origin *parent)
578 {
579         int chunk_end_lno;
580         memset(split, 0, sizeof(struct blame_entry [3]));
581
582         if (e->s_lno < tlno) {
583                 /* there is a pre-chunk part not blamed on parent */
584                 split[0].suspect = origin_incref(e->suspect);
585                 split[0].lno = e->lno;
586                 split[0].s_lno = e->s_lno;
587                 split[0].num_lines = tlno - e->s_lno;
588                 split[1].lno = e->lno + tlno - e->s_lno;
589                 split[1].s_lno = plno;
590         }
591         else {
592                 split[1].lno = e->lno;
593                 split[1].s_lno = plno + (e->s_lno - tlno);
594         }
595
596         if (same < e->s_lno + e->num_lines) {
597                 /* there is a post-chunk part not blamed on parent */
598                 split[2].suspect = origin_incref(e->suspect);
599                 split[2].lno = e->lno + (same - e->s_lno);
600                 split[2].s_lno = e->s_lno + (same - e->s_lno);
601                 split[2].num_lines = e->s_lno + e->num_lines - same;
602                 chunk_end_lno = split[2].lno;
603         }
604         else
605                 chunk_end_lno = e->lno + e->num_lines;
606         split[1].num_lines = chunk_end_lno - split[1].lno;
607
608         /*
609          * if it turns out there is nothing to blame the parent for,
610          * forget about the splitting.  !split[1].suspect signals this.
611          */
612         if (split[1].num_lines < 1)
613                 return;
614         split[1].suspect = origin_incref(parent);
615 }
616
617 /*
618  * split_overlap() divided an existing blame e into up to three parts
619  * in split.  Adjust the linked list of blames in the scoreboard to
620  * reflect the split.
621  */
622 static void split_blame(struct scoreboard *sb,
623                         struct blame_entry *split,
624                         struct blame_entry *e)
625 {
626         struct blame_entry *new_entry;
627
628         if (split[0].suspect && split[2].suspect) {
629                 /* The first part (reuse storage for the existing entry e) */
630                 dup_entry(e, &split[0]);
631
632                 /* The last part -- me */
633                 new_entry = xmalloc(sizeof(*new_entry));
634                 memcpy(new_entry, &(split[2]), sizeof(struct blame_entry));
635                 add_blame_entry(sb, new_entry);
636
637                 /* ... and the middle part -- parent */
638                 new_entry = xmalloc(sizeof(*new_entry));
639                 memcpy(new_entry, &(split[1]), sizeof(struct blame_entry));
640                 add_blame_entry(sb, new_entry);
641         }
642         else if (!split[0].suspect && !split[2].suspect)
643                 /*
644                  * The parent covers the entire area; reuse storage for
645                  * e and replace it with the parent.
646                  */
647                 dup_entry(e, &split[1]);
648         else if (split[0].suspect) {
649                 /* me and then parent */
650                 dup_entry(e, &split[0]);
651
652                 new_entry = xmalloc(sizeof(*new_entry));
653                 memcpy(new_entry, &(split[1]), sizeof(struct blame_entry));
654                 add_blame_entry(sb, new_entry);
655         }
656         else {
657                 /* parent and then me */
658                 dup_entry(e, &split[1]);
659
660                 new_entry = xmalloc(sizeof(*new_entry));
661                 memcpy(new_entry, &(split[2]), sizeof(struct blame_entry));
662                 add_blame_entry(sb, new_entry);
663         }
664
665         if (DEBUG) { /* sanity */
666                 struct blame_entry *ent;
667                 int lno = sb->ent->lno, corrupt = 0;
668
669                 for (ent = sb->ent; ent; ent = ent->next) {
670                         if (lno != ent->lno)
671                                 corrupt = 1;
672                         if (ent->s_lno < 0)
673                                 corrupt = 1;
674                         lno += ent->num_lines;
675                 }
676                 if (corrupt) {
677                         lno = sb->ent->lno;
678                         for (ent = sb->ent; ent; ent = ent->next) {
679                                 printf("L %8d l %8d n %8d\n",
680                                        lno, ent->lno, ent->num_lines);
681                                 lno = ent->lno + ent->num_lines;
682                         }
683                         die("oops");
684                 }
685         }
686 }
687
688 /*
689  * After splitting the blame, the origins used by the
690  * on-stack blame_entry should lose one refcnt each.
691  */
692 static void decref_split(struct blame_entry *split)
693 {
694         int i;
695
696         for (i = 0; i < 3; i++)
697                 origin_decref(split[i].suspect);
698 }
699
700 /*
701  * Helper for blame_chunk().  blame_entry e is known to overlap with
702  * the patch hunk; split it and pass blame to the parent.
703  */
704 static void blame_overlap(struct scoreboard *sb, struct blame_entry *e,
705                           int tlno, int plno, int same,
706                           struct origin *parent)
707 {
708         struct blame_entry split[3];
709
710         split_overlap(split, e, tlno, plno, same, parent);
711         if (split[1].suspect)
712                 split_blame(sb, split, e);
713         decref_split(split);
714 }
715
716 /*
717  * Find the line number of the last line the target is suspected for.
718  */
719 static int find_last_in_target(struct scoreboard *sb, struct origin *target)
720 {
721         struct blame_entry *e;
722         int last_in_target = -1;
723
724         for (e = sb->ent; e; e = e->next) {
725                 if (e->guilty || !same_suspect(e->suspect, target))
726                         continue;
727                 if (last_in_target < e->s_lno + e->num_lines)
728                         last_in_target = e->s_lno + e->num_lines;
729         }
730         return last_in_target;
731 }
732
733 /*
734  * Process one hunk from the patch between the current suspect for
735  * blame_entry e and its parent.  Find and split the overlap, and
736  * pass blame to the overlapping part to the parent.
737  */
738 static void blame_chunk(struct scoreboard *sb,
739                         int tlno, int plno, int same,
740                         struct origin *target, struct origin *parent)
741 {
742         struct blame_entry *e;
743
744         for (e = sb->ent; e; e = e->next) {
745                 if (e->guilty || !same_suspect(e->suspect, target))
746                         continue;
747                 if (same <= e->s_lno)
748                         continue;
749                 if (tlno < e->s_lno + e->num_lines)
750                         blame_overlap(sb, e, tlno, plno, same, parent);
751         }
752 }
753
754 struct blame_chunk_cb_data {
755         struct scoreboard *sb;
756         struct origin *target;
757         struct origin *parent;
758         long plno;
759         long tlno;
760 };
761
762 static void blame_chunk_cb(void *data, long same, long p_next, long t_next)
763 {
764         struct blame_chunk_cb_data *d = data;
765         blame_chunk(d->sb, d->tlno, d->plno, same, d->target, d->parent);
766         d->plno = p_next;
767         d->tlno = t_next;
768 }
769
770 /*
771  * We are looking at the origin 'target' and aiming to pass blame
772  * for the lines it is suspected to its parent.  Run diff to find
773  * which lines came from parent and pass blame for them.
774  */
775 static int pass_blame_to_parent(struct scoreboard *sb,
776                                 struct origin *target,
777                                 struct origin *parent)
778 {
779         int last_in_target;
780         mmfile_t file_p, file_o;
781         struct blame_chunk_cb_data d;
782         xpparam_t xpp;
783         xdemitconf_t xecfg;
784         memset(&d, 0, sizeof(d));
785         d.sb = sb; d.target = target; d.parent = parent;
786         last_in_target = find_last_in_target(sb, target);
787         if (last_in_target < 0)
788                 return 1; /* nothing remains for this target */
789
790         fill_origin_blob(&sb->revs->diffopt, parent, &file_p);
791         fill_origin_blob(&sb->revs->diffopt, target, &file_o);
792         num_get_patch++;
793
794         memset(&xpp, 0, sizeof(xpp));
795         xpp.flags = xdl_opts;
796         memset(&xecfg, 0, sizeof(xecfg));
797         xecfg.ctxlen = 0;
798         xdi_diff_hunks(&file_p, &file_o, blame_chunk_cb, &d, &xpp, &xecfg);
799         /* The rest (i.e. anything after tlno) are the same as the parent */
800         blame_chunk(sb, d.tlno, d.plno, last_in_target, target, parent);
801
802         return 0;
803 }
804
805 /*
806  * The lines in blame_entry after splitting blames many times can become
807  * very small and trivial, and at some point it becomes pointless to
808  * blame the parents.  E.g. "\t\t}\n\t}\n\n" appears everywhere in any
809  * ordinary C program, and it is not worth to say it was copied from
810  * totally unrelated file in the parent.
811  *
812  * Compute how trivial the lines in the blame_entry are.
813  */
814 static unsigned ent_score(struct scoreboard *sb, struct blame_entry *e)
815 {
816         unsigned score;
817         const char *cp, *ep;
818
819         if (e->score)
820                 return e->score;
821
822         score = 1;
823         cp = nth_line(sb, e->lno);
824         ep = nth_line(sb, e->lno + e->num_lines);
825         while (cp < ep) {
826                 unsigned ch = *((unsigned char *)cp);
827                 if (isalnum(ch))
828                         score++;
829                 cp++;
830         }
831         e->score = score;
832         return score;
833 }
834
835 /*
836  * best_so_far[] and this[] are both a split of an existing blame_entry
837  * that passes blame to the parent.  Maintain best_so_far the best split
838  * so far, by comparing this and best_so_far and copying this into
839  * bst_so_far as needed.
840  */
841 static void copy_split_if_better(struct scoreboard *sb,
842                                  struct blame_entry *best_so_far,
843                                  struct blame_entry *this)
844 {
845         int i;
846
847         if (!this[1].suspect)
848                 return;
849         if (best_so_far[1].suspect) {
850                 if (ent_score(sb, &this[1]) < ent_score(sb, &best_so_far[1]))
851                         return;
852         }
853
854         for (i = 0; i < 3; i++)
855                 origin_incref(this[i].suspect);
856         decref_split(best_so_far);
857         memcpy(best_so_far, this, sizeof(struct blame_entry [3]));
858 }
859
860 /*
861  * We are looking at a part of the final image represented by
862  * ent (tlno and same are offset by ent->s_lno).
863  * tlno is where we are looking at in the final image.
864  * up to (but not including) same match preimage.
865  * plno is where we are looking at in the preimage.
866  *
867  * <-------------- final image ---------------------->
868  *       <------ent------>
869  *         ^tlno ^same
870  *    <---------preimage----->
871  *         ^plno
872  *
873  * All line numbers are 0-based.
874  */
875 static void handle_split(struct scoreboard *sb,
876                          struct blame_entry *ent,
877                          int tlno, int plno, int same,
878                          struct origin *parent,
879                          struct blame_entry *split)
880 {
881         if (ent->num_lines <= tlno)
882                 return;
883         if (tlno < same) {
884                 struct blame_entry this[3];
885                 tlno += ent->s_lno;
886                 same += ent->s_lno;
887                 split_overlap(this, ent, tlno, plno, same, parent);
888                 copy_split_if_better(sb, split, this);
889                 decref_split(this);
890         }
891 }
892
893 struct handle_split_cb_data {
894         struct scoreboard *sb;
895         struct blame_entry *ent;
896         struct origin *parent;
897         struct blame_entry *split;
898         long plno;
899         long tlno;
900 };
901
902 static void handle_split_cb(void *data, long same, long p_next, long t_next)
903 {
904         struct handle_split_cb_data *d = data;
905         handle_split(d->sb, d->ent, d->tlno, d->plno, same, d->parent, d->split);
906         d->plno = p_next;
907         d->tlno = t_next;
908 }
909
910 /*
911  * Find the lines from parent that are the same as ent so that
912  * we can pass blames to it.  file_p has the blob contents for
913  * the parent.
914  */
915 static void find_copy_in_blob(struct scoreboard *sb,
916                               struct blame_entry *ent,
917                               struct origin *parent,
918                               struct blame_entry *split,
919                               mmfile_t *file_p)
920 {
921         const char *cp;
922         int cnt;
923         mmfile_t file_o;
924         struct handle_split_cb_data d;
925         xpparam_t xpp;
926         xdemitconf_t xecfg;
927         memset(&d, 0, sizeof(d));
928         d.sb = sb; d.ent = ent; d.parent = parent; d.split = split;
929         /*
930          * Prepare mmfile that contains only the lines in ent.
931          */
932         cp = nth_line(sb, ent->lno);
933         file_o.ptr = (char *) cp;
934         cnt = ent->num_lines;
935
936         while (cnt && cp < sb->final_buf + sb->final_buf_size) {
937                 if (*cp++ == '\n')
938                         cnt--;
939         }
940         file_o.size = cp - file_o.ptr;
941
942         /*
943          * file_o is a part of final image we are annotating.
944          * file_p partially may match that image.
945          */
946         memset(&xpp, 0, sizeof(xpp));
947         xpp.flags = xdl_opts;
948         memset(&xecfg, 0, sizeof(xecfg));
949         xecfg.ctxlen = 1;
950         memset(split, 0, sizeof(struct blame_entry [3]));
951         xdi_diff_hunks(file_p, &file_o, handle_split_cb, &d, &xpp, &xecfg);
952         /* remainder, if any, all match the preimage */
953         handle_split(sb, ent, d.tlno, d.plno, ent->num_lines, parent, split);
954 }
955
956 /*
957  * See if lines currently target is suspected for can be attributed to
958  * parent.
959  */
960 static int find_move_in_parent(struct scoreboard *sb,
961                                struct origin *target,
962                                struct origin *parent)
963 {
964         int last_in_target, made_progress;
965         struct blame_entry *e, split[3];
966         mmfile_t file_p;
967
968         last_in_target = find_last_in_target(sb, target);
969         if (last_in_target < 0)
970                 return 1; /* nothing remains for this target */
971
972         fill_origin_blob(&sb->revs->diffopt, parent, &file_p);
973         if (!file_p.ptr)
974                 return 0;
975
976         made_progress = 1;
977         while (made_progress) {
978                 made_progress = 0;
979                 for (e = sb->ent; e; e = e->next) {
980                         if (e->guilty || !same_suspect(e->suspect, target) ||
981                             ent_score(sb, e) < blame_move_score)
982                                 continue;
983                         find_copy_in_blob(sb, e, parent, split, &file_p);
984                         if (split[1].suspect &&
985                             blame_move_score < ent_score(sb, &split[1])) {
986                                 split_blame(sb, split, e);
987                                 made_progress = 1;
988                         }
989                         decref_split(split);
990                 }
991         }
992         return 0;
993 }
994
995 struct blame_list {
996         struct blame_entry *ent;
997         struct blame_entry split[3];
998 };
999
1000 /*
1001  * Count the number of entries the target is suspected for,
1002  * and prepare a list of entry and the best split.
1003  */
1004 static struct blame_list *setup_blame_list(struct scoreboard *sb,
1005                                            struct origin *target,
1006                                            int min_score,
1007                                            int *num_ents_p)
1008 {
1009         struct blame_entry *e;
1010         int num_ents, i;
1011         struct blame_list *blame_list = NULL;
1012
1013         for (e = sb->ent, num_ents = 0; e; e = e->next)
1014                 if (!e->scanned && !e->guilty &&
1015                     same_suspect(e->suspect, target) &&
1016                     min_score < ent_score(sb, e))
1017                         num_ents++;
1018         if (num_ents) {
1019                 blame_list = xcalloc(num_ents, sizeof(struct blame_list));
1020                 for (e = sb->ent, i = 0; e; e = e->next)
1021                         if (!e->scanned && !e->guilty &&
1022                             same_suspect(e->suspect, target) &&
1023                             min_score < ent_score(sb, e))
1024                                 blame_list[i++].ent = e;
1025         }
1026         *num_ents_p = num_ents;
1027         return blame_list;
1028 }
1029
1030 /*
1031  * Reset the scanned status on all entries.
1032  */
1033 static void reset_scanned_flag(struct scoreboard *sb)
1034 {
1035         struct blame_entry *e;
1036         for (e = sb->ent; e; e = e->next)
1037                 e->scanned = 0;
1038 }
1039
1040 /*
1041  * For lines target is suspected for, see if we can find code movement
1042  * across file boundary from the parent commit.  porigin is the path
1043  * in the parent we already tried.
1044  */
1045 static int find_copy_in_parent(struct scoreboard *sb,
1046                                struct origin *target,
1047                                struct commit *parent,
1048                                struct origin *porigin,
1049                                int opt)
1050 {
1051         struct diff_options diff_opts;
1052         const char *paths[1];
1053         int i, j;
1054         int retval;
1055         struct blame_list *blame_list;
1056         int num_ents;
1057
1058         blame_list = setup_blame_list(sb, target, blame_copy_score, &num_ents);
1059         if (!blame_list)
1060                 return 1; /* nothing remains for this target */
1061
1062         diff_setup(&diff_opts);
1063         DIFF_OPT_SET(&diff_opts, RECURSIVE);
1064         diff_opts.output_format = DIFF_FORMAT_NO_OUTPUT;
1065
1066         paths[0] = NULL;
1067         diff_tree_setup_paths(paths, &diff_opts);
1068         if (diff_setup_done(&diff_opts) < 0)
1069                 die("diff-setup");
1070
1071         /* Try "find copies harder" on new path if requested;
1072          * we do not want to use diffcore_rename() actually to
1073          * match things up; find_copies_harder is set only to
1074          * force diff_tree_sha1() to feed all filepairs to diff_queue,
1075          * and this code needs to be after diff_setup_done(), which
1076          * usually makes find-copies-harder imply copy detection.
1077          */
1078         if ((opt & PICKAXE_BLAME_COPY_HARDEST)
1079             || ((opt & PICKAXE_BLAME_COPY_HARDER)
1080                 && (!porigin || strcmp(target->path, porigin->path))))
1081                 DIFF_OPT_SET(&diff_opts, FIND_COPIES_HARDER);
1082
1083         if (is_null_sha1(target->commit->object.sha1))
1084                 do_diff_cache(parent->tree->object.sha1, &diff_opts);
1085         else
1086                 diff_tree_sha1(parent->tree->object.sha1,
1087                                target->commit->tree->object.sha1,
1088                                "", &diff_opts);
1089
1090         if (!DIFF_OPT_TST(&diff_opts, FIND_COPIES_HARDER))
1091                 diffcore_std(&diff_opts);
1092
1093         retval = 0;
1094         while (1) {
1095                 int made_progress = 0;
1096
1097                 for (i = 0; i < diff_queued_diff.nr; i++) {
1098                         struct diff_filepair *p = diff_queued_diff.queue[i];
1099                         struct origin *norigin;
1100                         mmfile_t file_p;
1101                         struct blame_entry this[3];
1102
1103                         if (!DIFF_FILE_VALID(p->one))
1104                                 continue; /* does not exist in parent */
1105                         if (S_ISGITLINK(p->one->mode))
1106                                 continue; /* ignore git links */
1107                         if (porigin && !strcmp(p->one->path, porigin->path))
1108                                 /* find_move already dealt with this path */
1109                                 continue;
1110
1111                         norigin = get_origin(sb, parent, p->one->path);
1112                         hashcpy(norigin->blob_sha1, p->one->sha1);
1113                         norigin->mode = p->one->mode;
1114                         fill_origin_blob(&sb->revs->diffopt, norigin, &file_p);
1115                         if (!file_p.ptr)
1116                                 continue;
1117
1118                         for (j = 0; j < num_ents; j++) {
1119                                 find_copy_in_blob(sb, blame_list[j].ent,
1120                                                   norigin, this, &file_p);
1121                                 copy_split_if_better(sb, blame_list[j].split,
1122                                                      this);
1123                                 decref_split(this);
1124                         }
1125                         origin_decref(norigin);
1126                 }
1127
1128                 for (j = 0; j < num_ents; j++) {
1129                         struct blame_entry *split = blame_list[j].split;
1130                         if (split[1].suspect &&
1131                             blame_copy_score < ent_score(sb, &split[1])) {
1132                                 split_blame(sb, split, blame_list[j].ent);
1133                                 made_progress = 1;
1134                         }
1135                         else
1136                                 blame_list[j].ent->scanned = 1;
1137                         decref_split(split);
1138                 }
1139                 free(blame_list);
1140
1141                 if (!made_progress)
1142                         break;
1143                 blame_list = setup_blame_list(sb, target, blame_copy_score, &num_ents);
1144                 if (!blame_list) {
1145                         retval = 1;
1146                         break;
1147                 }
1148         }
1149         reset_scanned_flag(sb);
1150         diff_flush(&diff_opts);
1151         diff_tree_release_paths(&diff_opts);
1152         return retval;
1153 }
1154
1155 /*
1156  * The blobs of origin and porigin exactly match, so everything
1157  * origin is suspected for can be blamed on the parent.
1158  */
1159 static void pass_whole_blame(struct scoreboard *sb,
1160                              struct origin *origin, struct origin *porigin)
1161 {
1162         struct blame_entry *e;
1163
1164         if (!porigin->file.ptr && origin->file.ptr) {
1165                 /* Steal its file */
1166                 porigin->file = origin->file;
1167                 origin->file.ptr = NULL;
1168         }
1169         for (e = sb->ent; e; e = e->next) {
1170                 if (!same_suspect(e->suspect, origin))
1171                         continue;
1172                 origin_incref(porigin);
1173                 origin_decref(e->suspect);
1174                 e->suspect = porigin;
1175         }
1176 }
1177
1178 /*
1179  * We pass blame from the current commit to its parents.  We keep saying
1180  * "parent" (and "porigin"), but what we mean is to find scapegoat to
1181  * exonerate ourselves.
1182  */
1183 static struct commit_list *first_scapegoat(struct rev_info *revs, struct commit *commit)
1184 {
1185         if (!reverse)
1186                 return commit->parents;
1187         return lookup_decoration(&revs->children, &commit->object);
1188 }
1189
1190 static int num_scapegoats(struct rev_info *revs, struct commit *commit)
1191 {
1192         int cnt;
1193         struct commit_list *l = first_scapegoat(revs, commit);
1194         for (cnt = 0; l; l = l->next)
1195                 cnt++;
1196         return cnt;
1197 }
1198
1199 #define MAXSG 16
1200
1201 static void pass_blame(struct scoreboard *sb, struct origin *origin, int opt)
1202 {
1203         struct rev_info *revs = sb->revs;
1204         int i, pass, num_sg;
1205         struct commit *commit = origin->commit;
1206         struct commit_list *sg;
1207         struct origin *sg_buf[MAXSG];
1208         struct origin *porigin, **sg_origin = sg_buf;
1209
1210         num_sg = num_scapegoats(revs, commit);
1211         if (!num_sg)
1212                 goto finish;
1213         else if (num_sg < ARRAY_SIZE(sg_buf))
1214                 memset(sg_buf, 0, sizeof(sg_buf));
1215         else
1216                 sg_origin = xcalloc(num_sg, sizeof(*sg_origin));
1217
1218         /*
1219          * The first pass looks for unrenamed path to optimize for
1220          * common cases, then we look for renames in the second pass.
1221          */
1222         for (pass = 0; pass < 2; pass++) {
1223                 struct origin *(*find)(struct scoreboard *,
1224                                        struct commit *, struct origin *);
1225                 find = pass ? find_rename : find_origin;
1226
1227                 for (i = 0, sg = first_scapegoat(revs, commit);
1228                      i < num_sg && sg;
1229                      sg = sg->next, i++) {
1230                         struct commit *p = sg->item;
1231                         int j, same;
1232
1233                         if (sg_origin[i])
1234                                 continue;
1235                         if (parse_commit(p))
1236                                 continue;
1237                         porigin = find(sb, p, origin);
1238                         if (!porigin)
1239                                 continue;
1240                         if (!hashcmp(porigin->blob_sha1, origin->blob_sha1)) {
1241                                 pass_whole_blame(sb, origin, porigin);
1242                                 origin_decref(porigin);
1243                                 goto finish;
1244                         }
1245                         for (j = same = 0; j < i; j++)
1246                                 if (sg_origin[j] &&
1247                                     !hashcmp(sg_origin[j]->blob_sha1,
1248                                              porigin->blob_sha1)) {
1249                                         same = 1;
1250                                         break;
1251                                 }
1252                         if (!same)
1253                                 sg_origin[i] = porigin;
1254                         else
1255                                 origin_decref(porigin);
1256                 }
1257         }
1258
1259         num_commits++;
1260         for (i = 0, sg = first_scapegoat(revs, commit);
1261              i < num_sg && sg;
1262              sg = sg->next, i++) {
1263                 struct origin *porigin = sg_origin[i];
1264                 if (!porigin)
1265                         continue;
1266                 if (!origin->previous) {
1267                         origin_incref(porigin);
1268                         origin->previous = porigin;
1269                 }
1270                 if (pass_blame_to_parent(sb, origin, porigin))
1271                         goto finish;
1272         }
1273
1274         /*
1275          * Optionally find moves in parents' files.
1276          */
1277         if (opt & PICKAXE_BLAME_MOVE)
1278                 for (i = 0, sg = first_scapegoat(revs, commit);
1279                      i < num_sg && sg;
1280                      sg = sg->next, i++) {
1281                         struct origin *porigin = sg_origin[i];
1282                         if (!porigin)
1283                                 continue;
1284                         if (find_move_in_parent(sb, origin, porigin))
1285                                 goto finish;
1286                 }
1287
1288         /*
1289          * Optionally find copies from parents' files.
1290          */
1291         if (opt & PICKAXE_BLAME_COPY)
1292                 for (i = 0, sg = first_scapegoat(revs, commit);
1293                      i < num_sg && sg;
1294                      sg = sg->next, i++) {
1295                         struct origin *porigin = sg_origin[i];
1296                         if (find_copy_in_parent(sb, origin, sg->item,
1297                                                 porigin, opt))
1298                                 goto finish;
1299                 }
1300
1301  finish:
1302         for (i = 0; i < num_sg; i++) {
1303                 if (sg_origin[i]) {
1304                         drop_origin_blob(sg_origin[i]);
1305                         origin_decref(sg_origin[i]);
1306                 }
1307         }
1308         drop_origin_blob(origin);
1309         if (sg_buf != sg_origin)
1310                 free(sg_origin);
1311 }
1312
1313 /*
1314  * Information on commits, used for output.
1315  */
1316 struct commit_info {
1317         const char *author;
1318         const char *author_mail;
1319         unsigned long author_time;
1320         const char *author_tz;
1321
1322         /* filled only when asked for details */
1323         const char *committer;
1324         const char *committer_mail;
1325         unsigned long committer_time;
1326         const char *committer_tz;
1327
1328         const char *summary;
1329 };
1330
1331 /*
1332  * Parse author/committer line in the commit object buffer
1333  */
1334 static void get_ac_line(const char *inbuf, const char *what,
1335                         int person_len, char *person,
1336                         int mail_len, char *mail,
1337                         unsigned long *time, const char **tz)
1338 {
1339         int len, tzlen, maillen;
1340         char *tmp, *endp, *timepos, *mailpos;
1341
1342         tmp = strstr(inbuf, what);
1343         if (!tmp)
1344                 goto error_out;
1345         tmp += strlen(what);
1346         endp = strchr(tmp, '\n');
1347         if (!endp)
1348                 len = strlen(tmp);
1349         else
1350                 len = endp - tmp;
1351         if (person_len <= len) {
1352         error_out:
1353                 /* Ugh */
1354                 *tz = "(unknown)";
1355                 strcpy(person, *tz);
1356                 strcpy(mail, *tz);
1357                 *time = 0;
1358                 return;
1359         }
1360         memcpy(person, tmp, len);
1361
1362         tmp = person;
1363         tmp += len;
1364         *tmp = 0;
1365         while (person < tmp && *tmp != ' ')
1366                 tmp--;
1367         if (tmp <= person)
1368                 goto error_out;
1369         *tz = tmp+1;
1370         tzlen = (person+len)-(tmp+1);
1371
1372         *tmp = 0;
1373         while (person < tmp && *tmp != ' ')
1374                 tmp--;
1375         if (tmp <= person)
1376                 goto error_out;
1377         *time = strtoul(tmp, NULL, 10);
1378         timepos = tmp;
1379
1380         *tmp = 0;
1381         while (person < tmp && !(*tmp == ' ' && tmp[1] == '<'))
1382                 tmp--;
1383         if (tmp <= person)
1384                 return;
1385         mailpos = tmp + 1;
1386         *tmp = 0;
1387         maillen = timepos - tmp;
1388         memcpy(mail, mailpos, maillen);
1389
1390         if (!mailmap.nr)
1391                 return;
1392
1393         /*
1394          * mailmap expansion may make the name longer.
1395          * make room by pushing stuff down.
1396          */
1397         tmp = person + person_len - (tzlen + 1);
1398         memmove(tmp, *tz, tzlen);
1399         tmp[tzlen] = 0;
1400         *tz = tmp;
1401
1402         /*
1403          * Now, convert both name and e-mail using mailmap
1404          */
1405         if (map_user(&mailmap, mail+1, mail_len-1, person, tmp-person-1)) {
1406                 /* Add a trailing '>' to email, since map_user returns plain emails
1407                    Note: It already has '<', since we replace from mail+1 */
1408                 mailpos = memchr(mail, '\0', mail_len);
1409                 if (mailpos && mailpos-mail < mail_len - 1) {
1410                         *mailpos = '>';
1411                         *(mailpos+1) = '\0';
1412                 }
1413         }
1414 }
1415
1416 static void get_commit_info(struct commit *commit,
1417                             struct commit_info *ret,
1418                             int detailed)
1419 {
1420         int len;
1421         const char *subject;
1422         char *reencoded, *message;
1423         static char author_name[1024];
1424         static char author_mail[1024];
1425         static char committer_name[1024];
1426         static char committer_mail[1024];
1427         static char summary_buf[1024];
1428
1429         /*
1430          * We've operated without save_commit_buffer, so
1431          * we now need to populate them for output.
1432          */
1433         if (!commit->buffer) {
1434                 enum object_type type;
1435                 unsigned long size;
1436                 commit->buffer =
1437                         read_sha1_file(commit->object.sha1, &type, &size);
1438                 if (!commit->buffer)
1439                         die("Cannot read commit %s",
1440                             sha1_to_hex(commit->object.sha1));
1441         }
1442         reencoded = reencode_commit_message(commit, NULL);
1443         message   = reencoded ? reencoded : commit->buffer;
1444         ret->author = author_name;
1445         ret->author_mail = author_mail;
1446         get_ac_line(message, "\nauthor ",
1447                     sizeof(author_name), author_name,
1448                     sizeof(author_mail), author_mail,
1449                     &ret->author_time, &ret->author_tz);
1450
1451         if (!detailed) {
1452                 free(reencoded);
1453                 return;
1454         }
1455
1456         ret->committer = committer_name;
1457         ret->committer_mail = committer_mail;
1458         get_ac_line(message, "\ncommitter ",
1459                     sizeof(committer_name), committer_name,
1460                     sizeof(committer_mail), committer_mail,
1461                     &ret->committer_time, &ret->committer_tz);
1462
1463         ret->summary = summary_buf;
1464         len = find_commit_subject(message, &subject);
1465         if (len && len < sizeof(summary_buf)) {
1466                 memcpy(summary_buf, subject, len);
1467                 summary_buf[len] = 0;
1468         } else {
1469                 sprintf(summary_buf, "(%s)", sha1_to_hex(commit->object.sha1));
1470         }
1471         free(reencoded);
1472 }
1473
1474 /*
1475  * To allow LF and other nonportable characters in pathnames,
1476  * they are c-style quoted as needed.
1477  */
1478 static void write_filename_info(const char *path)
1479 {
1480         printf("filename ");
1481         write_name_quoted(path, stdout, '\n');
1482 }
1483
1484 /*
1485  * Porcelain/Incremental format wants to show a lot of details per
1486  * commit.  Instead of repeating this every line, emit it only once,
1487  * the first time each commit appears in the output (unless the
1488  * user has specifically asked for us to repeat).
1489  */
1490 static int emit_one_suspect_detail(struct origin *suspect, int repeat)
1491 {
1492         struct commit_info ci;
1493
1494         if (!repeat && (suspect->commit->object.flags & METAINFO_SHOWN))
1495                 return 0;
1496
1497         suspect->commit->object.flags |= METAINFO_SHOWN;
1498         get_commit_info(suspect->commit, &ci, 1);
1499         printf("author %s\n", ci.author);
1500         printf("author-mail %s\n", ci.author_mail);
1501         printf("author-time %lu\n", ci.author_time);
1502         printf("author-tz %s\n", ci.author_tz);
1503         printf("committer %s\n", ci.committer);
1504         printf("committer-mail %s\n", ci.committer_mail);
1505         printf("committer-time %lu\n", ci.committer_time);
1506         printf("committer-tz %s\n", ci.committer_tz);
1507         printf("summary %s\n", ci.summary);
1508         if (suspect->commit->object.flags & UNINTERESTING)
1509                 printf("boundary\n");
1510         if (suspect->previous) {
1511                 struct origin *prev = suspect->previous;
1512                 printf("previous %s ", sha1_to_hex(prev->commit->object.sha1));
1513                 write_name_quoted(prev->path, stdout, '\n');
1514         }
1515         return 1;
1516 }
1517
1518 /*
1519  * The blame_entry is found to be guilty for the range.  Mark it
1520  * as such, and show it in incremental output.
1521  */
1522 static void found_guilty_entry(struct blame_entry *ent)
1523 {
1524         if (ent->guilty)
1525                 return;
1526         ent->guilty = 1;
1527         if (incremental) {
1528                 struct origin *suspect = ent->suspect;
1529
1530                 printf("%s %d %d %d\n",
1531                        sha1_to_hex(suspect->commit->object.sha1),
1532                        ent->s_lno + 1, ent->lno + 1, ent->num_lines);
1533                 emit_one_suspect_detail(suspect, 0);
1534                 write_filename_info(suspect->path);
1535                 maybe_flush_or_die(stdout, "stdout");
1536         }
1537 }
1538
1539 /*
1540  * The main loop -- while the scoreboard has lines whose true origin
1541  * is still unknown, pick one blame_entry, and allow its current
1542  * suspect to pass blames to its parents.
1543  */
1544 static void assign_blame(struct scoreboard *sb, int opt)
1545 {
1546         struct rev_info *revs = sb->revs;
1547
1548         while (1) {
1549                 struct blame_entry *ent;
1550                 struct commit *commit;
1551                 struct origin *suspect = NULL;
1552
1553                 /* find one suspect to break down */
1554                 for (ent = sb->ent; !suspect && ent; ent = ent->next)
1555                         if (!ent->guilty)
1556                                 suspect = ent->suspect;
1557                 if (!suspect)
1558                         return; /* all done */
1559
1560                 /*
1561                  * We will use this suspect later in the loop,
1562                  * so hold onto it in the meantime.
1563                  */
1564                 origin_incref(suspect);
1565                 commit = suspect->commit;
1566                 if (!commit->object.parsed)
1567                         parse_commit(commit);
1568                 if (reverse ||
1569                     (!(commit->object.flags & UNINTERESTING) &&
1570                      !(revs->max_age != -1 && commit->date < revs->max_age)))
1571                         pass_blame(sb, suspect, opt);
1572                 else {
1573                         commit->object.flags |= UNINTERESTING;
1574                         if (commit->object.parsed)
1575                                 mark_parents_uninteresting(commit);
1576                 }
1577                 /* treat root commit as boundary */
1578                 if (!commit->parents && !show_root)
1579                         commit->object.flags |= UNINTERESTING;
1580
1581                 /* Take responsibility for the remaining entries */
1582                 for (ent = sb->ent; ent; ent = ent->next)
1583                         if (same_suspect(ent->suspect, suspect))
1584                                 found_guilty_entry(ent);
1585                 origin_decref(suspect);
1586
1587                 if (DEBUG) /* sanity */
1588                         sanity_check_refcnt(sb);
1589         }
1590 }
1591
1592 static const char *format_time(unsigned long time, const char *tz_str,
1593                                int show_raw_time)
1594 {
1595         static char time_buf[128];
1596         const char *time_str;
1597         int time_len;
1598         int tz;
1599
1600         if (show_raw_time) {
1601                 sprintf(time_buf, "%lu %s", time, tz_str);
1602         }
1603         else {
1604                 tz = atoi(tz_str);
1605                 time_str = show_date(time, tz, blame_date_mode);
1606                 time_len = strlen(time_str);
1607                 memcpy(time_buf, time_str, time_len);
1608                 memset(time_buf + time_len, ' ', blame_date_width - time_len);
1609         }
1610         return time_buf;
1611 }
1612
1613 #define OUTPUT_ANNOTATE_COMPAT  001
1614 #define OUTPUT_LONG_OBJECT_NAME 002
1615 #define OUTPUT_RAW_TIMESTAMP    004
1616 #define OUTPUT_PORCELAIN        010
1617 #define OUTPUT_SHOW_NAME        020
1618 #define OUTPUT_SHOW_NUMBER      040
1619 #define OUTPUT_SHOW_SCORE      0100
1620 #define OUTPUT_NO_AUTHOR       0200
1621 #define OUTPUT_SHOW_EMAIL       0400
1622 #define OUTPUT_LINE_PORCELAIN 01000
1623
1624 static void emit_porcelain_details(struct origin *suspect, int repeat)
1625 {
1626         if (emit_one_suspect_detail(suspect, repeat) ||
1627             (suspect->commit->object.flags & MORE_THAN_ONE_PATH))
1628                 write_filename_info(suspect->path);
1629 }
1630
1631 static void emit_porcelain(struct scoreboard *sb, struct blame_entry *ent,
1632                            int opt)
1633 {
1634         int repeat = opt & OUTPUT_LINE_PORCELAIN;
1635         int cnt;
1636         const char *cp;
1637         struct origin *suspect = ent->suspect;
1638         char hex[41];
1639
1640         strcpy(hex, sha1_to_hex(suspect->commit->object.sha1));
1641         printf("%s%c%d %d %d\n",
1642                hex,
1643                ent->guilty ? ' ' : '*', /* purely for debugging */
1644                ent->s_lno + 1,
1645                ent->lno + 1,
1646                ent->num_lines);
1647         emit_porcelain_details(suspect, repeat);
1648
1649         cp = nth_line(sb, ent->lno);
1650         for (cnt = 0; cnt < ent->num_lines; cnt++) {
1651                 char ch;
1652                 if (cnt) {
1653                         printf("%s %d %d\n", hex,
1654                                ent->s_lno + 1 + cnt,
1655                                ent->lno + 1 + cnt);
1656                         if (repeat)
1657                                 emit_porcelain_details(suspect, 1);
1658                 }
1659                 putchar('\t');
1660                 do {
1661                         ch = *cp++;
1662                         putchar(ch);
1663                 } while (ch != '\n' &&
1664                          cp < sb->final_buf + sb->final_buf_size);
1665         }
1666
1667         if (sb->final_buf_size && cp[-1] != '\n')
1668                 putchar('\n');
1669 }
1670
1671 static void emit_other(struct scoreboard *sb, struct blame_entry *ent, int opt)
1672 {
1673         int cnt;
1674         const char *cp;
1675         struct origin *suspect = ent->suspect;
1676         struct commit_info ci;
1677         char hex[41];
1678         int show_raw_time = !!(opt & OUTPUT_RAW_TIMESTAMP);
1679
1680         get_commit_info(suspect->commit, &ci, 1);
1681         strcpy(hex, sha1_to_hex(suspect->commit->object.sha1));
1682
1683         cp = nth_line(sb, ent->lno);
1684         for (cnt = 0; cnt < ent->num_lines; cnt++) {
1685                 char ch;
1686                 int length = (opt & OUTPUT_LONG_OBJECT_NAME) ? 40 : abbrev;
1687
1688                 if (suspect->commit->object.flags & UNINTERESTING) {
1689                         if (blank_boundary)
1690                                 memset(hex, ' ', length);
1691                         else if (!(opt & OUTPUT_ANNOTATE_COMPAT)) {
1692                                 length--;
1693                                 putchar('^');
1694                         }
1695                 }
1696
1697                 printf("%.*s", length, hex);
1698                 if (opt & OUTPUT_ANNOTATE_COMPAT) {
1699                         const char *name;
1700                         if (opt & OUTPUT_SHOW_EMAIL)
1701                                 name = ci.author_mail;
1702                         else
1703                                 name = ci.author;
1704                         printf("\t(%10s\t%10s\t%d)", name,
1705                                format_time(ci.author_time, ci.author_tz,
1706                                            show_raw_time),
1707                                ent->lno + 1 + cnt);
1708                 } else {
1709                         if (opt & OUTPUT_SHOW_SCORE)
1710                                 printf(" %*d %02d",
1711                                        max_score_digits, ent->score,
1712                                        ent->suspect->refcnt);
1713                         if (opt & OUTPUT_SHOW_NAME)
1714                                 printf(" %-*.*s", longest_file, longest_file,
1715                                        suspect->path);
1716                         if (opt & OUTPUT_SHOW_NUMBER)
1717                                 printf(" %*d", max_orig_digits,
1718                                        ent->s_lno + 1 + cnt);
1719
1720                         if (!(opt & OUTPUT_NO_AUTHOR)) {
1721                                 const char *name;
1722                                 int pad;
1723                                 if (opt & OUTPUT_SHOW_EMAIL)
1724                                         name = ci.author_mail;
1725                                 else
1726                                         name = ci.author;
1727                                 pad = longest_author - utf8_strwidth(name);
1728                                 printf(" (%s%*s %10s",
1729                                        name, pad, "",
1730                                        format_time(ci.author_time,
1731                                                    ci.author_tz,
1732                                                    show_raw_time));
1733                         }
1734                         printf(" %*d) ",
1735                                max_digits, ent->lno + 1 + cnt);
1736                 }
1737                 do {
1738                         ch = *cp++;
1739                         putchar(ch);
1740                 } while (ch != '\n' &&
1741                          cp < sb->final_buf + sb->final_buf_size);
1742         }
1743
1744         if (sb->final_buf_size && cp[-1] != '\n')
1745                 putchar('\n');
1746 }
1747
1748 static void output(struct scoreboard *sb, int option)
1749 {
1750         struct blame_entry *ent;
1751
1752         if (option & OUTPUT_PORCELAIN) {
1753                 for (ent = sb->ent; ent; ent = ent->next) {
1754                         struct blame_entry *oth;
1755                         struct origin *suspect = ent->suspect;
1756                         struct commit *commit = suspect->commit;
1757                         if (commit->object.flags & MORE_THAN_ONE_PATH)
1758                                 continue;
1759                         for (oth = ent->next; oth; oth = oth->next) {
1760                                 if ((oth->suspect->commit != commit) ||
1761                                     !strcmp(oth->suspect->path, suspect->path))
1762                                         continue;
1763                                 commit->object.flags |= MORE_THAN_ONE_PATH;
1764                                 break;
1765                         }
1766                 }
1767         }
1768
1769         for (ent = sb->ent; ent; ent = ent->next) {
1770                 if (option & OUTPUT_PORCELAIN)
1771                         emit_porcelain(sb, ent, option);
1772                 else {
1773                         emit_other(sb, ent, option);
1774                 }
1775         }
1776 }
1777
1778 /*
1779  * To allow quick access to the contents of nth line in the
1780  * final image, prepare an index in the scoreboard.
1781  */
1782 static int prepare_lines(struct scoreboard *sb)
1783 {
1784         const char *buf = sb->final_buf;
1785         unsigned long len = sb->final_buf_size;
1786         int num = 0, incomplete = 0, bol = 1;
1787
1788         if (len && buf[len-1] != '\n')
1789                 incomplete++; /* incomplete line at the end */
1790         while (len--) {
1791                 if (bol) {
1792                         sb->lineno = xrealloc(sb->lineno,
1793                                               sizeof(int *) * (num + 1));
1794                         sb->lineno[num] = buf - sb->final_buf;
1795                         bol = 0;
1796                 }
1797                 if (*buf++ == '\n') {
1798                         num++;
1799                         bol = 1;
1800                 }
1801         }
1802         sb->lineno = xrealloc(sb->lineno,
1803                               sizeof(int *) * (num + incomplete + 1));
1804         sb->lineno[num + incomplete] = buf - sb->final_buf;
1805         sb->num_lines = num + incomplete;
1806         return sb->num_lines;
1807 }
1808
1809 /*
1810  * Add phony grafts for use with -S; this is primarily to
1811  * support git's cvsserver that wants to give a linear history
1812  * to its clients.
1813  */
1814 static int read_ancestry(const char *graft_file)
1815 {
1816         FILE *fp = fopen(graft_file, "r");
1817         char buf[1024];
1818         if (!fp)
1819                 return -1;
1820         while (fgets(buf, sizeof(buf), fp)) {
1821                 /* The format is just "Commit Parent1 Parent2 ...\n" */
1822                 int len = strlen(buf);
1823                 struct commit_graft *graft = read_graft_line(buf, len);
1824                 if (graft)
1825                         register_commit_graft(graft, 0);
1826         }
1827         fclose(fp);
1828         return 0;
1829 }
1830
1831 /*
1832  * How many columns do we need to show line numbers in decimal?
1833  */
1834 static int lineno_width(int lines)
1835 {
1836         int i, width;
1837
1838         for (width = 1, i = 10; i <= lines; width++)
1839                 i *= 10;
1840         return width;
1841 }
1842
1843 /*
1844  * How many columns do we need to show line numbers, authors,
1845  * and filenames?
1846  */
1847 static void find_alignment(struct scoreboard *sb, int *option)
1848 {
1849         int longest_src_lines = 0;
1850         int longest_dst_lines = 0;
1851         unsigned largest_score = 0;
1852         struct blame_entry *e;
1853
1854         for (e = sb->ent; e; e = e->next) {
1855                 struct origin *suspect = e->suspect;
1856                 struct commit_info ci;
1857                 int num;
1858
1859                 if (strcmp(suspect->path, sb->path))
1860                         *option |= OUTPUT_SHOW_NAME;
1861                 num = strlen(suspect->path);
1862                 if (longest_file < num)
1863                         longest_file = num;
1864                 if (!(suspect->commit->object.flags & METAINFO_SHOWN)) {
1865                         suspect->commit->object.flags |= METAINFO_SHOWN;
1866                         get_commit_info(suspect->commit, &ci, 1);
1867                         if (*option & OUTPUT_SHOW_EMAIL)
1868                                 num = utf8_strwidth(ci.author_mail);
1869                         else
1870                                 num = utf8_strwidth(ci.author);
1871                         if (longest_author < num)
1872                                 longest_author = num;
1873                 }
1874                 num = e->s_lno + e->num_lines;
1875                 if (longest_src_lines < num)
1876                         longest_src_lines = num;
1877                 num = e->lno + e->num_lines;
1878                 if (longest_dst_lines < num)
1879                         longest_dst_lines = num;
1880                 if (largest_score < ent_score(sb, e))
1881                         largest_score = ent_score(sb, e);
1882         }
1883         max_orig_digits = lineno_width(longest_src_lines);
1884         max_digits = lineno_width(longest_dst_lines);
1885         max_score_digits = lineno_width(largest_score);
1886 }
1887
1888 /*
1889  * For debugging -- origin is refcounted, and this asserts that
1890  * we do not underflow.
1891  */
1892 static void sanity_check_refcnt(struct scoreboard *sb)
1893 {
1894         int baa = 0;
1895         struct blame_entry *ent;
1896
1897         for (ent = sb->ent; ent; ent = ent->next) {
1898                 /* Nobody should have zero or negative refcnt */
1899                 if (ent->suspect->refcnt <= 0) {
1900                         fprintf(stderr, "%s in %s has negative refcnt %d\n",
1901                                 ent->suspect->path,
1902                                 sha1_to_hex(ent->suspect->commit->object.sha1),
1903                                 ent->suspect->refcnt);
1904                         baa = 1;
1905                 }
1906         }
1907         if (baa) {
1908                 int opt = 0160;
1909                 find_alignment(sb, &opt);
1910                 output(sb, opt);
1911                 die("Baa %d!", baa);
1912         }
1913 }
1914
1915 /*
1916  * Used for the command line parsing; check if the path exists
1917  * in the working tree.
1918  */
1919 static int has_string_in_work_tree(const char *path)
1920 {
1921         struct stat st;
1922         return !lstat(path, &st);
1923 }
1924
1925 static unsigned parse_score(const char *arg)
1926 {
1927         char *end;
1928         unsigned long score = strtoul(arg, &end, 10);
1929         if (*end)
1930                 return 0;
1931         return score;
1932 }
1933
1934 static const char *add_prefix(const char *prefix, const char *path)
1935 {
1936         return prefix_path(prefix, prefix ? strlen(prefix) : 0, path);
1937 }
1938
1939 /*
1940  * Parsing of (comma separated) one item in the -L option
1941  */
1942 static const char *parse_loc(const char *spec,
1943                              struct scoreboard *sb, long lno,
1944                              long begin, long *ret)
1945 {
1946         char *term;
1947         const char *line;
1948         long num;
1949         int reg_error;
1950         regex_t regexp;
1951         regmatch_t match[1];
1952
1953         /* Allow "-L <something>,+20" to mean starting at <something>
1954          * for 20 lines, or "-L <something>,-5" for 5 lines ending at
1955          * <something>.
1956          */
1957         if (1 < begin && (spec[0] == '+' || spec[0] == '-')) {
1958                 num = strtol(spec + 1, &term, 10);
1959                 if (term != spec + 1) {
1960                         if (spec[0] == '-')
1961                                 num = 0 - num;
1962                         if (0 < num)
1963                                 *ret = begin + num - 2;
1964                         else if (!num)
1965                                 *ret = begin;
1966                         else
1967                                 *ret = begin + num;
1968                         return term;
1969                 }
1970                 return spec;
1971         }
1972         num = strtol(spec, &term, 10);
1973         if (term != spec) {
1974                 *ret = num;
1975                 return term;
1976         }
1977         if (spec[0] != '/')
1978                 return spec;
1979
1980         /* it could be a regexp of form /.../ */
1981         for (term = (char *) spec + 1; *term && *term != '/'; term++) {
1982                 if (*term == '\\')
1983                         term++;
1984         }
1985         if (*term != '/')
1986                 return spec;
1987
1988         /* try [spec+1 .. term-1] as regexp */
1989         *term = 0;
1990         begin--; /* input is in human terms */
1991         line = nth_line(sb, begin);
1992
1993         if (!(reg_error = regcomp(&regexp, spec + 1, REG_NEWLINE)) &&
1994             !(reg_error = regexec(&regexp, line, 1, match, 0))) {
1995                 const char *cp = line + match[0].rm_so;
1996                 const char *nline;
1997
1998                 while (begin++ < lno) {
1999                         nline = nth_line(sb, begin);
2000                         if (line <= cp && cp < nline)
2001                                 break;
2002                         line = nline;
2003                 }
2004                 *ret = begin;
2005                 regfree(&regexp);
2006                 *term++ = '/';
2007                 return term;
2008         }
2009         else {
2010                 char errbuf[1024];
2011                 regerror(reg_error, &regexp, errbuf, 1024);
2012                 die("-L parameter '%s': %s", spec + 1, errbuf);
2013         }
2014 }
2015
2016 /*
2017  * Parsing of -L option
2018  */
2019 static void prepare_blame_range(struct scoreboard *sb,
2020                                 const char *bottomtop,
2021                                 long lno,
2022                                 long *bottom, long *top)
2023 {
2024         const char *term;
2025
2026         term = parse_loc(bottomtop, sb, lno, 1, bottom);
2027         if (*term == ',') {
2028                 term = parse_loc(term + 1, sb, lno, *bottom + 1, top);
2029                 if (*term)
2030                         usage(blame_usage);
2031         }
2032         if (*term)
2033                 usage(blame_usage);
2034 }
2035
2036 static int git_blame_config(const char *var, const char *value, void *cb)
2037 {
2038         if (!strcmp(var, "blame.showroot")) {
2039                 show_root = git_config_bool(var, value);
2040                 return 0;
2041         }
2042         if (!strcmp(var, "blame.blankboundary")) {
2043                 blank_boundary = git_config_bool(var, value);
2044                 return 0;
2045         }
2046         if (!strcmp(var, "blame.date")) {
2047                 if (!value)
2048                         return config_error_nonbool(var);
2049                 blame_date_mode = parse_date_format(value);
2050                 return 0;
2051         }
2052
2053         switch (userdiff_config(var, value)) {
2054         case 0:
2055                 break;
2056         case -1:
2057                 return -1;
2058         default:
2059                 return 0;
2060         }
2061
2062         return git_default_config(var, value, cb);
2063 }
2064
2065 /*
2066  * Prepare a dummy commit that represents the work tree (or staged) item.
2067  * Note that annotating work tree item never works in the reverse.
2068  */
2069 static struct commit *fake_working_tree_commit(struct diff_options *opt,
2070                                                const char *path,
2071                                                const char *contents_from)
2072 {
2073         struct commit *commit;
2074         struct origin *origin;
2075         unsigned char head_sha1[20];
2076         struct strbuf buf = STRBUF_INIT;
2077         const char *ident;
2078         time_t now;
2079         int size, len;
2080         struct cache_entry *ce;
2081         unsigned mode;
2082
2083         if (get_sha1("HEAD", head_sha1))
2084                 die("No such ref: HEAD");
2085
2086         time(&now);
2087         commit = xcalloc(1, sizeof(*commit));
2088         commit->parents = xcalloc(1, sizeof(*commit->parents));
2089         commit->parents->item = lookup_commit_reference(head_sha1);
2090         commit->object.parsed = 1;
2091         commit->date = now;
2092         commit->object.type = OBJ_COMMIT;
2093
2094         origin = make_origin(commit, path);
2095
2096         if (!contents_from || strcmp("-", contents_from)) {
2097                 struct stat st;
2098                 const char *read_from;
2099                 unsigned long buf_len;
2100
2101                 if (contents_from) {
2102                         if (stat(contents_from, &st) < 0)
2103                                 die_errno("Cannot stat '%s'", contents_from);
2104                         read_from = contents_from;
2105                 }
2106                 else {
2107                         if (lstat(path, &st) < 0)
2108                                 die_errno("Cannot lstat '%s'", path);
2109                         read_from = path;
2110                 }
2111                 mode = canon_mode(st.st_mode);
2112
2113                 switch (st.st_mode & S_IFMT) {
2114                 case S_IFREG:
2115                         if (DIFF_OPT_TST(opt, ALLOW_TEXTCONV) &&
2116                             textconv_object(read_from, mode, null_sha1, &buf.buf, &buf_len))
2117                                 buf.len = buf_len;
2118                         else if (strbuf_read_file(&buf, read_from, st.st_size) != st.st_size)
2119                                 die_errno("cannot open or read '%s'", read_from);
2120                         break;
2121                 case S_IFLNK:
2122                         if (strbuf_readlink(&buf, read_from, st.st_size) < 0)
2123                                 die_errno("cannot readlink '%s'", read_from);
2124                         break;
2125                 default:
2126                         die("unsupported file type %s", read_from);
2127                 }
2128         }
2129         else {
2130                 /* Reading from stdin */
2131                 contents_from = "standard input";
2132                 mode = 0;
2133                 if (strbuf_read(&buf, 0, 0) < 0)
2134                         die_errno("failed to read from stdin");
2135         }
2136         convert_to_git(path, buf.buf, buf.len, &buf, 0);
2137         origin->file.ptr = buf.buf;
2138         origin->file.size = buf.len;
2139         pretend_sha1_file(buf.buf, buf.len, OBJ_BLOB, origin->blob_sha1);
2140         commit->util = origin;
2141
2142         /*
2143          * Read the current index, replace the path entry with
2144          * origin->blob_sha1 without mucking with its mode or type
2145          * bits; we are not going to write this index out -- we just
2146          * want to run "diff-index --cached".
2147          */
2148         discard_cache();
2149         read_cache();
2150
2151         len = strlen(path);
2152         if (!mode) {
2153                 int pos = cache_name_pos(path, len);
2154                 if (0 <= pos)
2155                         mode = active_cache[pos]->ce_mode;
2156                 else
2157                         /* Let's not bother reading from HEAD tree */
2158                         mode = S_IFREG | 0644;
2159         }
2160         size = cache_entry_size(len);
2161         ce = xcalloc(1, size);
2162         hashcpy(ce->sha1, origin->blob_sha1);
2163         memcpy(ce->name, path, len);
2164         ce->ce_flags = create_ce_flags(len, 0);
2165         ce->ce_mode = create_ce_mode(mode);
2166         add_cache_entry(ce, ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE);
2167
2168         /*
2169          * We are not going to write this out, so this does not matter
2170          * right now, but someday we might optimize diff-index --cached
2171          * with cache-tree information.
2172          */
2173         cache_tree_invalidate_path(active_cache_tree, path);
2174
2175         commit->buffer = xmalloc(400);
2176         ident = fmt_ident("Not Committed Yet", "not.committed.yet", NULL, 0);
2177         snprintf(commit->buffer, 400,
2178                 "tree 0000000000000000000000000000000000000000\n"
2179                 "parent %s\n"
2180                 "author %s\n"
2181                 "committer %s\n\n"
2182                 "Version of %s from %s\n",
2183                 sha1_to_hex(head_sha1),
2184                 ident, ident, path, contents_from ? contents_from : path);
2185         return commit;
2186 }
2187
2188 static const char *prepare_final(struct scoreboard *sb)
2189 {
2190         int i;
2191         const char *final_commit_name = NULL;
2192         struct rev_info *revs = sb->revs;
2193
2194         /*
2195          * There must be one and only one positive commit in the
2196          * revs->pending array.
2197          */
2198         for (i = 0; i < revs->pending.nr; i++) {
2199                 struct object *obj = revs->pending.objects[i].item;
2200                 if (obj->flags & UNINTERESTING)
2201                         continue;
2202                 while (obj->type == OBJ_TAG)
2203                         obj = deref_tag(obj, NULL, 0);
2204                 if (obj->type != OBJ_COMMIT)
2205                         die("Non commit %s?", revs->pending.objects[i].name);
2206                 if (sb->final)
2207                         die("More than one commit to dig from %s and %s?",
2208                             revs->pending.objects[i].name,
2209                             final_commit_name);
2210                 sb->final = (struct commit *) obj;
2211                 final_commit_name = revs->pending.objects[i].name;
2212         }
2213         return final_commit_name;
2214 }
2215
2216 static const char *prepare_initial(struct scoreboard *sb)
2217 {
2218         int i;
2219         const char *final_commit_name = NULL;
2220         struct rev_info *revs = sb->revs;
2221
2222         /*
2223          * There must be one and only one negative commit, and it must be
2224          * the boundary.
2225          */
2226         for (i = 0; i < revs->pending.nr; i++) {
2227                 struct object *obj = revs->pending.objects[i].item;
2228                 if (!(obj->flags & UNINTERESTING))
2229                         continue;
2230                 while (obj->type == OBJ_TAG)
2231                         obj = deref_tag(obj, NULL, 0);
2232                 if (obj->type != OBJ_COMMIT)
2233                         die("Non commit %s?", revs->pending.objects[i].name);
2234                 if (sb->final)
2235                         die("More than one commit to dig down to %s and %s?",
2236                             revs->pending.objects[i].name,
2237                             final_commit_name);
2238                 sb->final = (struct commit *) obj;
2239                 final_commit_name = revs->pending.objects[i].name;
2240         }
2241         if (!final_commit_name)
2242                 die("No commit to dig down to?");
2243         return final_commit_name;
2244 }
2245
2246 static int blame_copy_callback(const struct option *option, const char *arg, int unset)
2247 {
2248         int *opt = option->value;
2249
2250         /*
2251          * -C enables copy from removed files;
2252          * -C -C enables copy from existing files, but only
2253          *       when blaming a new file;
2254          * -C -C -C enables copy from existing files for
2255          *          everybody
2256          */
2257         if (*opt & PICKAXE_BLAME_COPY_HARDER)
2258                 *opt |= PICKAXE_BLAME_COPY_HARDEST;
2259         if (*opt & PICKAXE_BLAME_COPY)
2260                 *opt |= PICKAXE_BLAME_COPY_HARDER;
2261         *opt |= PICKAXE_BLAME_COPY | PICKAXE_BLAME_MOVE;
2262
2263         if (arg)
2264                 blame_copy_score = parse_score(arg);
2265         return 0;
2266 }
2267
2268 static int blame_move_callback(const struct option *option, const char *arg, int unset)
2269 {
2270         int *opt = option->value;
2271
2272         *opt |= PICKAXE_BLAME_MOVE;
2273
2274         if (arg)
2275                 blame_move_score = parse_score(arg);
2276         return 0;
2277 }
2278
2279 static int blame_bottomtop_callback(const struct option *option, const char *arg, int unset)
2280 {
2281         const char **bottomtop = option->value;
2282         if (!arg)
2283                 return -1;
2284         if (*bottomtop)
2285                 die("More than one '-L n,m' option given");
2286         *bottomtop = arg;
2287         return 0;
2288 }
2289
2290 int cmd_blame(int argc, const char **argv, const char *prefix)
2291 {
2292         struct rev_info revs;
2293         const char *path;
2294         struct scoreboard sb;
2295         struct origin *o;
2296         struct blame_entry *ent;
2297         long dashdash_pos, bottom, top, lno;
2298         const char *final_commit_name = NULL;
2299         enum object_type type;
2300
2301         static const char *bottomtop = NULL;
2302         static int output_option = 0, opt = 0;
2303         static int show_stats = 0;
2304         static const char *revs_file = NULL;
2305         static const char *contents_from = NULL;
2306         static const struct option options[] = {
2307                 OPT_BOOLEAN(0, "incremental", &incremental, "Show blame entries as we find them, incrementally"),
2308                 OPT_BOOLEAN('b', NULL, &blank_boundary, "Show blank SHA-1 for boundary commits (Default: off)"),
2309                 OPT_BOOLEAN(0, "root", &show_root, "Do not treat root commits as boundaries (Default: off)"),
2310                 OPT_BOOLEAN(0, "show-stats", &show_stats, "Show work cost statistics"),
2311                 OPT_BIT(0, "score-debug", &output_option, "Show output score for blame entries", OUTPUT_SHOW_SCORE),
2312                 OPT_BIT('f', "show-name", &output_option, "Show original filename (Default: auto)", OUTPUT_SHOW_NAME),
2313                 OPT_BIT('n', "show-number", &output_option, "Show original linenumber (Default: off)", OUTPUT_SHOW_NUMBER),
2314                 OPT_BIT('p', "porcelain", &output_option, "Show in a format designed for machine consumption", OUTPUT_PORCELAIN),
2315                 OPT_BIT(0, "line-porcelain", &output_option, "Show porcelain format with per-line commit information", OUTPUT_PORCELAIN|OUTPUT_LINE_PORCELAIN),
2316                 OPT_BIT('c', NULL, &output_option, "Use the same output mode as git-annotate (Default: off)", OUTPUT_ANNOTATE_COMPAT),
2317                 OPT_BIT('t', NULL, &output_option, "Show raw timestamp (Default: off)", OUTPUT_RAW_TIMESTAMP),
2318                 OPT_BIT('l', NULL, &output_option, "Show long commit SHA1 (Default: off)", OUTPUT_LONG_OBJECT_NAME),
2319                 OPT_BIT('s', NULL, &output_option, "Suppress author name and timestamp (Default: off)", OUTPUT_NO_AUTHOR),
2320                 OPT_BIT('e', "show-email", &output_option, "Show author email instead of name (Default: off)", OUTPUT_SHOW_EMAIL),
2321                 OPT_BIT('w', NULL, &xdl_opts, "Ignore whitespace differences", XDF_IGNORE_WHITESPACE),
2322                 OPT_STRING('S', NULL, &revs_file, "file", "Use revisions from <file> instead of calling git-rev-list"),
2323                 OPT_STRING(0, "contents", &contents_from, "file", "Use <file>'s contents as the final image"),
2324                 { OPTION_CALLBACK, 'C', NULL, &opt, "score", "Find line copies within and across files", PARSE_OPT_OPTARG, blame_copy_callback },
2325                 { OPTION_CALLBACK, 'M', NULL, &opt, "score", "Find line movements within and across files", PARSE_OPT_OPTARG, blame_move_callback },
2326                 OPT_CALLBACK('L', NULL, &bottomtop, "n,m", "Process only line range n,m, counting from 1", blame_bottomtop_callback),
2327                 OPT__ABBREV(&abbrev),
2328                 OPT_END()
2329         };
2330
2331         struct parse_opt_ctx_t ctx;
2332         int cmd_is_annotate = !strcmp(argv[0], "annotate");
2333
2334         git_config(git_blame_config, NULL);
2335         init_revisions(&revs, NULL);
2336         revs.date_mode = blame_date_mode;
2337         DIFF_OPT_SET(&revs.diffopt, ALLOW_TEXTCONV);
2338
2339         save_commit_buffer = 0;
2340         dashdash_pos = 0;
2341
2342         parse_options_start(&ctx, argc, argv, prefix, options,
2343                             PARSE_OPT_KEEP_DASHDASH | PARSE_OPT_KEEP_ARGV0);
2344         for (;;) {
2345                 switch (parse_options_step(&ctx, options, blame_opt_usage)) {
2346                 case PARSE_OPT_HELP:
2347                         exit(129);
2348                 case PARSE_OPT_DONE:
2349                         if (ctx.argv[0])
2350                                 dashdash_pos = ctx.cpidx;
2351                         goto parse_done;
2352                 }
2353
2354                 if (!strcmp(ctx.argv[0], "--reverse")) {
2355                         ctx.argv[0] = "--children";
2356                         reverse = 1;
2357                 }
2358                 parse_revision_opt(&revs, &ctx, options, blame_opt_usage);
2359         }
2360 parse_done:
2361         argc = parse_options_end(&ctx);
2362
2363         if (abbrev == -1)
2364                 abbrev = default_abbrev;
2365         /* one more abbrev length is needed for the boundary commit */
2366         abbrev++;
2367
2368         if (revs_file && read_ancestry(revs_file))
2369                 die_errno("reading graft file '%s' failed", revs_file);
2370
2371         if (cmd_is_annotate) {
2372                 output_option |= OUTPUT_ANNOTATE_COMPAT;
2373                 blame_date_mode = DATE_ISO8601;
2374         } else {
2375                 blame_date_mode = revs.date_mode;
2376         }
2377
2378         /* The maximum width used to show the dates */
2379         switch (blame_date_mode) {
2380         case DATE_RFC2822:
2381                 blame_date_width = sizeof("Thu, 19 Oct 2006 16:00:04 -0700");
2382                 break;
2383         case DATE_ISO8601:
2384                 blame_date_width = sizeof("2006-10-19 16:00:04 -0700");
2385                 break;
2386         case DATE_RAW:
2387                 blame_date_width = sizeof("1161298804 -0700");
2388                 break;
2389         case DATE_SHORT:
2390                 blame_date_width = sizeof("2006-10-19");
2391                 break;
2392         case DATE_RELATIVE:
2393                 /* "normal" is used as the fallback for "relative" */
2394         case DATE_LOCAL:
2395         case DATE_NORMAL:
2396                 blame_date_width = sizeof("Thu Oct 19 16:00:04 2006 -0700");
2397                 break;
2398         }
2399         blame_date_width -= 1; /* strip the null */
2400
2401         if (DIFF_OPT_TST(&revs.diffopt, FIND_COPIES_HARDER))
2402                 opt |= (PICKAXE_BLAME_COPY | PICKAXE_BLAME_MOVE |
2403                         PICKAXE_BLAME_COPY_HARDER);
2404
2405         if (!blame_move_score)
2406                 blame_move_score = BLAME_DEFAULT_MOVE_SCORE;
2407         if (!blame_copy_score)
2408                 blame_copy_score = BLAME_DEFAULT_COPY_SCORE;
2409
2410         /*
2411          * We have collected options unknown to us in argv[1..unk]
2412          * which are to be passed to revision machinery if we are
2413          * going to do the "bottom" processing.
2414          *
2415          * The remaining are:
2416          *
2417          * (1) if dashdash_pos != 0, it is either
2418          *     "blame [revisions] -- <path>" or
2419          *     "blame -- <path> <rev>"
2420          *
2421          * (2) otherwise, it is one of the two:
2422          *     "blame [revisions] <path>"
2423          *     "blame <path> <rev>"
2424          *
2425          * Note that we must strip out <path> from the arguments: we do not
2426          * want the path pruning but we may want "bottom" processing.
2427          */
2428         if (dashdash_pos) {
2429                 switch (argc - dashdash_pos - 1) {
2430                 case 2: /* (1b) */
2431                         if (argc != 4)
2432                                 usage_with_options(blame_opt_usage, options);
2433                         /* reorder for the new way: <rev> -- <path> */
2434                         argv[1] = argv[3];
2435                         argv[3] = argv[2];
2436                         argv[2] = "--";
2437                         /* FALLTHROUGH */
2438                 case 1: /* (1a) */
2439                         path = add_prefix(prefix, argv[--argc]);
2440                         argv[argc] = NULL;
2441                         break;
2442                 default:
2443                         usage_with_options(blame_opt_usage, options);
2444                 }
2445         } else {
2446                 if (argc < 2)
2447                         usage_with_options(blame_opt_usage, options);
2448                 path = add_prefix(prefix, argv[argc - 1]);
2449                 if (argc == 3 && !has_string_in_work_tree(path)) { /* (2b) */
2450                         path = add_prefix(prefix, argv[1]);
2451                         argv[1] = argv[2];
2452                 }
2453                 argv[argc - 1] = "--";
2454
2455                 setup_work_tree();
2456                 if (!has_string_in_work_tree(path))
2457                         die_errno("cannot stat path '%s'", path);
2458         }
2459
2460         revs.disable_stdin = 1;
2461         setup_revisions(argc, argv, &revs, NULL);
2462         memset(&sb, 0, sizeof(sb));
2463
2464         sb.revs = &revs;
2465         if (!reverse)
2466                 final_commit_name = prepare_final(&sb);
2467         else if (contents_from)
2468                 die("--contents and --children do not blend well.");
2469         else
2470                 final_commit_name = prepare_initial(&sb);
2471
2472         if (!sb.final) {
2473                 /*
2474                  * "--not A B -- path" without anything positive;
2475                  * do not default to HEAD, but use the working tree
2476                  * or "--contents".
2477                  */
2478                 setup_work_tree();
2479                 sb.final = fake_working_tree_commit(&sb.revs->diffopt,
2480                                                     path, contents_from);
2481                 add_pending_object(&revs, &(sb.final->object), ":");
2482         }
2483         else if (contents_from)
2484                 die("Cannot use --contents with final commit object name");
2485
2486         /*
2487          * If we have bottom, this will mark the ancestors of the
2488          * bottom commits we would reach while traversing as
2489          * uninteresting.
2490          */
2491         if (prepare_revision_walk(&revs))
2492                 die("revision walk setup failed");
2493
2494         if (is_null_sha1(sb.final->object.sha1)) {
2495                 char *buf;
2496                 o = sb.final->util;
2497                 buf = xmalloc(o->file.size + 1);
2498                 memcpy(buf, o->file.ptr, o->file.size + 1);
2499                 sb.final_buf = buf;
2500                 sb.final_buf_size = o->file.size;
2501         }
2502         else {
2503                 o = get_origin(&sb, sb.final, path);
2504                 if (fill_blob_sha1_and_mode(o))
2505                         die("no such path %s in %s", path, final_commit_name);
2506
2507                 if (DIFF_OPT_TST(&sb.revs->diffopt, ALLOW_TEXTCONV) &&
2508                     textconv_object(path, o->mode, o->blob_sha1, (char **) &sb.final_buf,
2509                                     &sb.final_buf_size))
2510                         ;
2511                 else
2512                         sb.final_buf = read_sha1_file(o->blob_sha1, &type,
2513                                                       &sb.final_buf_size);
2514
2515                 if (!sb.final_buf)
2516                         die("Cannot read blob %s for path %s",
2517                             sha1_to_hex(o->blob_sha1),
2518                             path);
2519         }
2520         num_read_blob++;
2521         lno = prepare_lines(&sb);
2522
2523         bottom = top = 0;
2524         if (bottomtop)
2525                 prepare_blame_range(&sb, bottomtop, lno, &bottom, &top);
2526         if (bottom && top && top < bottom) {
2527                 long tmp;
2528                 tmp = top; top = bottom; bottom = tmp;
2529         }
2530         if (bottom < 1)
2531                 bottom = 1;
2532         if (top < 1)
2533                 top = lno;
2534         bottom--;
2535         if (lno < top || lno < bottom)
2536                 die("file %s has only %lu lines", path, lno);
2537
2538         ent = xcalloc(1, sizeof(*ent));
2539         ent->lno = bottom;
2540         ent->num_lines = top - bottom;
2541         ent->suspect = o;
2542         ent->s_lno = bottom;
2543
2544         sb.ent = ent;
2545         sb.path = path;
2546
2547         read_mailmap(&mailmap, NULL);
2548
2549         if (!incremental)
2550                 setup_pager();
2551
2552         assign_blame(&sb, opt);
2553
2554         if (incremental)
2555                 return 0;
2556
2557         coalesce(&sb);
2558
2559         if (!(output_option & OUTPUT_PORCELAIN))
2560                 find_alignment(&sb, &output_option);
2561
2562         output(&sb, output_option);
2563         free((void *)sb.final_buf);
2564         for (ent = sb.ent; ent; ) {
2565                 struct blame_entry *e = ent->next;
2566                 free(ent);
2567                 ent = e;
2568         }
2569
2570         if (show_stats) {
2571                 printf("num read blob: %d\n", num_read_blob);
2572                 printf("num get patch: %d\n", num_get_patch);
2573                 printf("num commits: %d\n", num_commits);
2574         }
2575         return 0;
2576 }