Merge branch 'rt/completion-help'
[git] / builtin / blame.c
1 /*
2  * Blame
3  *
4  * Copyright (c) 2006, 2014 by its authors
5  * See COPYING for licensing conditions
6  */
7
8 #include "cache.h"
9 #include "refs.h"
10 #include "builtin.h"
11 #include "blob.h"
12 #include "commit.h"
13 #include "tag.h"
14 #include "tree-walk.h"
15 #include "diff.h"
16 #include "diffcore.h"
17 #include "revision.h"
18 #include "quote.h"
19 #include "xdiff-interface.h"
20 #include "cache-tree.h"
21 #include "string-list.h"
22 #include "mailmap.h"
23 #include "mergesort.h"
24 #include "parse-options.h"
25 #include "prio-queue.h"
26 #include "utf8.h"
27 #include "userdiff.h"
28 #include "line-range.h"
29 #include "line-log.h"
30 #include "dir.h"
31 #include "progress.h"
32
33 static char blame_usage[] = N_("git blame [<options>] [<rev-opts>] [<rev>] [--] <file>");
34
35 static const char *blame_opt_usage[] = {
36         blame_usage,
37         "",
38         N_("<rev-opts> are documented in git-rev-list(1)"),
39         NULL
40 };
41
42 static int longest_file;
43 static int longest_author;
44 static int max_orig_digits;
45 static int max_digits;
46 static int max_score_digits;
47 static int show_root;
48 static int reverse;
49 static int blank_boundary;
50 static int incremental;
51 static int xdl_opts;
52 static int abbrev = -1;
53 static int no_whole_file_rename;
54 static int show_progress;
55
56 static struct date_mode blame_date_mode = { DATE_ISO8601 };
57 static size_t blame_date_width;
58
59 static struct string_list mailmap;
60
61 #ifndef DEBUG
62 #define DEBUG 0
63 #endif
64
65 /* stats */
66 static int num_read_blob;
67 static int num_get_patch;
68 static int num_commits;
69
70 #define PICKAXE_BLAME_MOVE              01
71 #define PICKAXE_BLAME_COPY              02
72 #define PICKAXE_BLAME_COPY_HARDER       04
73 #define PICKAXE_BLAME_COPY_HARDEST      010
74
75 /*
76  * blame for a blame_entry with score lower than these thresholds
77  * is not passed to the parent using move/copy logic.
78  */
79 static unsigned blame_move_score;
80 static unsigned blame_copy_score;
81 #define BLAME_DEFAULT_MOVE_SCORE        20
82 #define BLAME_DEFAULT_COPY_SCORE        40
83
84 /* Remember to update object flag allocation in object.h */
85 #define METAINFO_SHOWN          (1u<<12)
86 #define MORE_THAN_ONE_PATH      (1u<<13)
87
88 /*
89  * One blob in a commit that is being suspected
90  */
91 struct origin {
92         int refcnt;
93         /* Record preceding blame record for this blob */
94         struct origin *previous;
95         /* origins are put in a list linked via `next' hanging off the
96          * corresponding commit's util field in order to make finding
97          * them fast.  The presence in this chain does not count
98          * towards the origin's reference count.  It is tempting to
99          * let it count as long as the commit is pending examination,
100          * but even under circumstances where the commit will be
101          * present multiple times in the priority queue of unexamined
102          * commits, processing the first instance will not leave any
103          * work requiring the origin data for the second instance.  An
104          * interspersed commit changing that would have to be
105          * preexisting with a different ancestry and with the same
106          * commit date in order to wedge itself between two instances
107          * of the same commit in the priority queue _and_ produce
108          * blame entries relevant for it.  While we don't want to let
109          * us get tripped up by this case, it certainly does not seem
110          * worth optimizing for.
111          */
112         struct origin *next;
113         struct commit *commit;
114         /* `suspects' contains blame entries that may be attributed to
115          * this origin's commit or to parent commits.  When a commit
116          * is being processed, all suspects will be moved, either by
117          * assigning them to an origin in a different commit, or by
118          * shipping them to the scoreboard's ent list because they
119          * cannot be attributed to a different commit.
120          */
121         struct blame_entry *suspects;
122         mmfile_t file;
123         unsigned char blob_sha1[20];
124         unsigned mode;
125         /* guilty gets set when shipping any suspects to the final
126          * blame list instead of other commits
127          */
128         char guilty;
129         char path[FLEX_ARRAY];
130 };
131
132 struct progress_info {
133         struct progress *progress;
134         int blamed_lines;
135 };
136
137 static int diff_hunks(mmfile_t *file_a, mmfile_t *file_b, long ctxlen,
138                       xdl_emit_hunk_consume_func_t hunk_func, void *cb_data)
139 {
140         xpparam_t xpp = {0};
141         xdemitconf_t xecfg = {0};
142         xdemitcb_t ecb = {NULL};
143
144         xpp.flags = xdl_opts;
145         xecfg.ctxlen = ctxlen;
146         xecfg.hunk_func = hunk_func;
147         ecb.priv = cb_data;
148         return xdi_diff(file_a, file_b, &xpp, &xecfg, &ecb);
149 }
150
151 /*
152  * Prepare diff_filespec and convert it using diff textconv API
153  * if the textconv driver exists.
154  * Return 1 if the conversion succeeds, 0 otherwise.
155  */
156 int textconv_object(const char *path,
157                     unsigned mode,
158                     const unsigned char *sha1,
159                     int sha1_valid,
160                     char **buf,
161                     unsigned long *buf_size)
162 {
163         struct diff_filespec *df;
164         struct userdiff_driver *textconv;
165
166         df = alloc_filespec(path);
167         fill_filespec(df, sha1, sha1_valid, mode);
168         textconv = get_textconv(df);
169         if (!textconv) {
170                 free_filespec(df);
171                 return 0;
172         }
173
174         *buf_size = fill_textconv(textconv, df, buf);
175         free_filespec(df);
176         return 1;
177 }
178
179 /*
180  * Given an origin, prepare mmfile_t structure to be used by the
181  * diff machinery
182  */
183 static void fill_origin_blob(struct diff_options *opt,
184                              struct origin *o, mmfile_t *file)
185 {
186         if (!o->file.ptr) {
187                 enum object_type type;
188                 unsigned long file_size;
189
190                 num_read_blob++;
191                 if (DIFF_OPT_TST(opt, ALLOW_TEXTCONV) &&
192                     textconv_object(o->path, o->mode, o->blob_sha1, 1, &file->ptr, &file_size))
193                         ;
194                 else
195                         file->ptr = read_sha1_file(o->blob_sha1, &type, &file_size);
196                 file->size = file_size;
197
198                 if (!file->ptr)
199                         die("Cannot read blob %s for path %s",
200                             sha1_to_hex(o->blob_sha1),
201                             o->path);
202                 o->file = *file;
203         }
204         else
205                 *file = o->file;
206 }
207
208 /*
209  * Origin is refcounted and usually we keep the blob contents to be
210  * reused.
211  */
212 static inline struct origin *origin_incref(struct origin *o)
213 {
214         if (o)
215                 o->refcnt++;
216         return o;
217 }
218
219 static void origin_decref(struct origin *o)
220 {
221         if (o && --o->refcnt <= 0) {
222                 struct origin *p, *l = NULL;
223                 if (o->previous)
224                         origin_decref(o->previous);
225                 free(o->file.ptr);
226                 /* Should be present exactly once in commit chain */
227                 for (p = o->commit->util; p; l = p, p = p->next) {
228                         if (p == o) {
229                                 if (l)
230                                         l->next = p->next;
231                                 else
232                                         o->commit->util = p->next;
233                                 free(o);
234                                 return;
235                         }
236                 }
237                 die("internal error in blame::origin_decref");
238         }
239 }
240
241 static void drop_origin_blob(struct origin *o)
242 {
243         if (o->file.ptr) {
244                 free(o->file.ptr);
245                 o->file.ptr = NULL;
246         }
247 }
248
249 /*
250  * Each group of lines is described by a blame_entry; it can be split
251  * as we pass blame to the parents.  They are arranged in linked lists
252  * kept as `suspects' of some unprocessed origin, or entered (when the
253  * blame origin has been finalized) into the scoreboard structure.
254  * While the scoreboard structure is only sorted at the end of
255  * processing (according to final image line number), the lists
256  * attached to an origin are sorted by the target line number.
257  */
258 struct blame_entry {
259         struct blame_entry *next;
260
261         /* the first line of this group in the final image;
262          * internally all line numbers are 0 based.
263          */
264         int lno;
265
266         /* how many lines this group has */
267         int num_lines;
268
269         /* the commit that introduced this group into the final image */
270         struct origin *suspect;
271
272         /* the line number of the first line of this group in the
273          * suspect's file; internally all line numbers are 0 based.
274          */
275         int s_lno;
276
277         /* how significant this entry is -- cached to avoid
278          * scanning the lines over and over.
279          */
280         unsigned score;
281 };
282
283 /*
284  * Any merge of blames happens on lists of blames that arrived via
285  * different parents in a single suspect.  In this case, we want to
286  * sort according to the suspect line numbers as opposed to the final
287  * image line numbers.  The function body is somewhat longish because
288  * it avoids unnecessary writes.
289  */
290
291 static struct blame_entry *blame_merge(struct blame_entry *list1,
292                                        struct blame_entry *list2)
293 {
294         struct blame_entry *p1 = list1, *p2 = list2,
295                 **tail = &list1;
296
297         if (!p1)
298                 return p2;
299         if (!p2)
300                 return p1;
301
302         if (p1->s_lno <= p2->s_lno) {
303                 do {
304                         tail = &p1->next;
305                         if ((p1 = *tail) == NULL) {
306                                 *tail = p2;
307                                 return list1;
308                         }
309                 } while (p1->s_lno <= p2->s_lno);
310         }
311         for (;;) {
312                 *tail = p2;
313                 do {
314                         tail = &p2->next;
315                         if ((p2 = *tail) == NULL)  {
316                                 *tail = p1;
317                                 return list1;
318                         }
319                 } while (p1->s_lno > p2->s_lno);
320                 *tail = p1;
321                 do {
322                         tail = &p1->next;
323                         if ((p1 = *tail) == NULL) {
324                                 *tail = p2;
325                                 return list1;
326                         }
327                 } while (p1->s_lno <= p2->s_lno);
328         }
329 }
330
331 static void *get_next_blame(const void *p)
332 {
333         return ((struct blame_entry *)p)->next;
334 }
335
336 static void set_next_blame(void *p1, void *p2)
337 {
338         ((struct blame_entry *)p1)->next = p2;
339 }
340
341 /*
342  * Final image line numbers are all different, so we don't need a
343  * three-way comparison here.
344  */
345
346 static int compare_blame_final(const void *p1, const void *p2)
347 {
348         return ((struct blame_entry *)p1)->lno > ((struct blame_entry *)p2)->lno
349                 ? 1 : -1;
350 }
351
352 static int compare_blame_suspect(const void *p1, const void *p2)
353 {
354         const struct blame_entry *s1 = p1, *s2 = p2;
355         /*
356          * to allow for collating suspects, we sort according to the
357          * respective pointer value as the primary sorting criterion.
358          * The actual relation is pretty unimportant as long as it
359          * establishes a total order.  Comparing as integers gives us
360          * that.
361          */
362         if (s1->suspect != s2->suspect)
363                 return (intptr_t)s1->suspect > (intptr_t)s2->suspect ? 1 : -1;
364         if (s1->s_lno == s2->s_lno)
365                 return 0;
366         return s1->s_lno > s2->s_lno ? 1 : -1;
367 }
368
369 static struct blame_entry *blame_sort(struct blame_entry *head,
370                                       int (*compare_fn)(const void *, const void *))
371 {
372         return llist_mergesort (head, get_next_blame, set_next_blame, compare_fn);
373 }
374
375 static int compare_commits_by_reverse_commit_date(const void *a,
376                                                   const void *b,
377                                                   void *c)
378 {
379         return -compare_commits_by_commit_date(a, b, c);
380 }
381
382 /*
383  * The current state of the blame assignment.
384  */
385 struct scoreboard {
386         /* the final commit (i.e. where we started digging from) */
387         struct commit *final;
388         /* Priority queue for commits with unassigned blame records */
389         struct prio_queue commits;
390         struct rev_info *revs;
391         const char *path;
392
393         /*
394          * The contents in the final image.
395          * Used by many functions to obtain contents of the nth line,
396          * indexed with scoreboard.lineno[blame_entry.lno].
397          */
398         const char *final_buf;
399         unsigned long final_buf_size;
400
401         /* linked list of blames */
402         struct blame_entry *ent;
403
404         /* look-up a line in the final buffer */
405         int num_lines;
406         int *lineno;
407 };
408
409 static void sanity_check_refcnt(struct scoreboard *);
410
411 /*
412  * If two blame entries that are next to each other came from
413  * contiguous lines in the same origin (i.e. <commit, path> pair),
414  * merge them together.
415  */
416 static void coalesce(struct scoreboard *sb)
417 {
418         struct blame_entry *ent, *next;
419
420         for (ent = sb->ent; ent && (next = ent->next); ent = next) {
421                 if (ent->suspect == next->suspect &&
422                     ent->s_lno + ent->num_lines == next->s_lno) {
423                         ent->num_lines += next->num_lines;
424                         ent->next = next->next;
425                         origin_decref(next->suspect);
426                         free(next);
427                         ent->score = 0;
428                         next = ent; /* again */
429                 }
430         }
431
432         if (DEBUG) /* sanity */
433                 sanity_check_refcnt(sb);
434 }
435
436 /*
437  * Merge the given sorted list of blames into a preexisting origin.
438  * If there were no previous blames to that commit, it is entered into
439  * the commit priority queue of the score board.
440  */
441
442 static void queue_blames(struct scoreboard *sb, struct origin *porigin,
443                          struct blame_entry *sorted)
444 {
445         if (porigin->suspects)
446                 porigin->suspects = blame_merge(porigin->suspects, sorted);
447         else {
448                 struct origin *o;
449                 for (o = porigin->commit->util; o; o = o->next) {
450                         if (o->suspects) {
451                                 porigin->suspects = sorted;
452                                 return;
453                         }
454                 }
455                 porigin->suspects = sorted;
456                 prio_queue_put(&sb->commits, porigin->commit);
457         }
458 }
459
460 /*
461  * Given a commit and a path in it, create a new origin structure.
462  * The callers that add blame to the scoreboard should use
463  * get_origin() to obtain shared, refcounted copy instead of calling
464  * this function directly.
465  */
466 static struct origin *make_origin(struct commit *commit, const char *path)
467 {
468         struct origin *o;
469         FLEX_ALLOC_STR(o, path, path);
470         o->commit = commit;
471         o->refcnt = 1;
472         o->next = commit->util;
473         commit->util = o;
474         return o;
475 }
476
477 /*
478  * Locate an existing origin or create a new one.
479  * This moves the origin to front position in the commit util list.
480  */
481 static struct origin *get_origin(struct scoreboard *sb,
482                                  struct commit *commit,
483                                  const char *path)
484 {
485         struct origin *o, *l;
486
487         for (o = commit->util, l = NULL; o; l = o, o = o->next) {
488                 if (!strcmp(o->path, path)) {
489                         /* bump to front */
490                         if (l) {
491                                 l->next = o->next;
492                                 o->next = commit->util;
493                                 commit->util = o;
494                         }
495                         return origin_incref(o);
496                 }
497         }
498         return make_origin(commit, path);
499 }
500
501 /*
502  * Fill the blob_sha1 field of an origin if it hasn't, so that later
503  * call to fill_origin_blob() can use it to locate the data.  blob_sha1
504  * for an origin is also used to pass the blame for the entire file to
505  * the parent to detect the case where a child's blob is identical to
506  * that of its parent's.
507  *
508  * This also fills origin->mode for corresponding tree path.
509  */
510 static int fill_blob_sha1_and_mode(struct origin *origin)
511 {
512         if (!is_null_sha1(origin->blob_sha1))
513                 return 0;
514         if (get_tree_entry(origin->commit->object.oid.hash,
515                            origin->path,
516                            origin->blob_sha1, &origin->mode))
517                 goto error_out;
518         if (sha1_object_info(origin->blob_sha1, NULL) != OBJ_BLOB)
519                 goto error_out;
520         return 0;
521  error_out:
522         hashclr(origin->blob_sha1);
523         origin->mode = S_IFINVALID;
524         return -1;
525 }
526
527 /*
528  * We have an origin -- check if the same path exists in the
529  * parent and return an origin structure to represent it.
530  */
531 static struct origin *find_origin(struct scoreboard *sb,
532                                   struct commit *parent,
533                                   struct origin *origin)
534 {
535         struct origin *porigin;
536         struct diff_options diff_opts;
537         const char *paths[2];
538
539         /* First check any existing origins */
540         for (porigin = parent->util; porigin; porigin = porigin->next)
541                 if (!strcmp(porigin->path, origin->path)) {
542                         /*
543                          * The same path between origin and its parent
544                          * without renaming -- the most common case.
545                          */
546                         return origin_incref (porigin);
547                 }
548
549         /* See if the origin->path is different between parent
550          * and origin first.  Most of the time they are the
551          * same and diff-tree is fairly efficient about this.
552          */
553         diff_setup(&diff_opts);
554         DIFF_OPT_SET(&diff_opts, RECURSIVE);
555         diff_opts.detect_rename = 0;
556         diff_opts.output_format = DIFF_FORMAT_NO_OUTPUT;
557         paths[0] = origin->path;
558         paths[1] = NULL;
559
560         parse_pathspec(&diff_opts.pathspec,
561                        PATHSPEC_ALL_MAGIC & ~PATHSPEC_LITERAL,
562                        PATHSPEC_LITERAL_PATH, "", paths);
563         diff_setup_done(&diff_opts);
564
565         if (is_null_oid(&origin->commit->object.oid))
566                 do_diff_cache(parent->tree->object.oid.hash, &diff_opts);
567         else
568                 diff_tree_sha1(parent->tree->object.oid.hash,
569                                origin->commit->tree->object.oid.hash,
570                                "", &diff_opts);
571         diffcore_std(&diff_opts);
572
573         if (!diff_queued_diff.nr) {
574                 /* The path is the same as parent */
575                 porigin = get_origin(sb, parent, origin->path);
576                 hashcpy(porigin->blob_sha1, origin->blob_sha1);
577                 porigin->mode = origin->mode;
578         } else {
579                 /*
580                  * Since origin->path is a pathspec, if the parent
581                  * commit had it as a directory, we will see a whole
582                  * bunch of deletion of files in the directory that we
583                  * do not care about.
584                  */
585                 int i;
586                 struct diff_filepair *p = NULL;
587                 for (i = 0; i < diff_queued_diff.nr; i++) {
588                         const char *name;
589                         p = diff_queued_diff.queue[i];
590                         name = p->one->path ? p->one->path : p->two->path;
591                         if (!strcmp(name, origin->path))
592                                 break;
593                 }
594                 if (!p)
595                         die("internal error in blame::find_origin");
596                 switch (p->status) {
597                 default:
598                         die("internal error in blame::find_origin (%c)",
599                             p->status);
600                 case 'M':
601                         porigin = get_origin(sb, parent, origin->path);
602                         hashcpy(porigin->blob_sha1, p->one->sha1);
603                         porigin->mode = p->one->mode;
604                         break;
605                 case 'A':
606                 case 'T':
607                         /* Did not exist in parent, or type changed */
608                         break;
609                 }
610         }
611         diff_flush(&diff_opts);
612         free_pathspec(&diff_opts.pathspec);
613         return porigin;
614 }
615
616 /*
617  * We have an origin -- find the path that corresponds to it in its
618  * parent and return an origin structure to represent it.
619  */
620 static struct origin *find_rename(struct scoreboard *sb,
621                                   struct commit *parent,
622                                   struct origin *origin)
623 {
624         struct origin *porigin = NULL;
625         struct diff_options diff_opts;
626         int i;
627
628         diff_setup(&diff_opts);
629         DIFF_OPT_SET(&diff_opts, RECURSIVE);
630         diff_opts.detect_rename = DIFF_DETECT_RENAME;
631         diff_opts.output_format = DIFF_FORMAT_NO_OUTPUT;
632         diff_opts.single_follow = origin->path;
633         diff_setup_done(&diff_opts);
634
635         if (is_null_oid(&origin->commit->object.oid))
636                 do_diff_cache(parent->tree->object.oid.hash, &diff_opts);
637         else
638                 diff_tree_sha1(parent->tree->object.oid.hash,
639                                origin->commit->tree->object.oid.hash,
640                                "", &diff_opts);
641         diffcore_std(&diff_opts);
642
643         for (i = 0; i < diff_queued_diff.nr; i++) {
644                 struct diff_filepair *p = diff_queued_diff.queue[i];
645                 if ((p->status == 'R' || p->status == 'C') &&
646                     !strcmp(p->two->path, origin->path)) {
647                         porigin = get_origin(sb, parent, p->one->path);
648                         hashcpy(porigin->blob_sha1, p->one->sha1);
649                         porigin->mode = p->one->mode;
650                         break;
651                 }
652         }
653         diff_flush(&diff_opts);
654         free_pathspec(&diff_opts.pathspec);
655         return porigin;
656 }
657
658 /*
659  * Append a new blame entry to a given output queue.
660  */
661 static void add_blame_entry(struct blame_entry ***queue, struct blame_entry *e)
662 {
663         origin_incref(e->suspect);
664
665         e->next = **queue;
666         **queue = e;
667         *queue = &e->next;
668 }
669
670 /*
671  * src typically is on-stack; we want to copy the information in it to
672  * a malloced blame_entry that gets added to the given queue.  The
673  * origin of dst loses a refcnt.
674  */
675 static void dup_entry(struct blame_entry ***queue,
676                       struct blame_entry *dst, struct blame_entry *src)
677 {
678         origin_incref(src->suspect);
679         origin_decref(dst->suspect);
680         memcpy(dst, src, sizeof(*src));
681         dst->next = **queue;
682         **queue = dst;
683         *queue = &dst->next;
684 }
685
686 static const char *nth_line(struct scoreboard *sb, long lno)
687 {
688         return sb->final_buf + sb->lineno[lno];
689 }
690
691 static const char *nth_line_cb(void *data, long lno)
692 {
693         return nth_line((struct scoreboard *)data, lno);
694 }
695
696 /*
697  * It is known that lines between tlno to same came from parent, and e
698  * has an overlap with that range.  it also is known that parent's
699  * line plno corresponds to e's line tlno.
700  *
701  *                <---- e ----->
702  *                   <------>
703  *                   <------------>
704  *             <------------>
705  *             <------------------>
706  *
707  * Split e into potentially three parts; before this chunk, the chunk
708  * to be blamed for the parent, and after that portion.
709  */
710 static void split_overlap(struct blame_entry *split,
711                           struct blame_entry *e,
712                           int tlno, int plno, int same,
713                           struct origin *parent)
714 {
715         int chunk_end_lno;
716         memset(split, 0, sizeof(struct blame_entry [3]));
717
718         if (e->s_lno < tlno) {
719                 /* there is a pre-chunk part not blamed on parent */
720                 split[0].suspect = origin_incref(e->suspect);
721                 split[0].lno = e->lno;
722                 split[0].s_lno = e->s_lno;
723                 split[0].num_lines = tlno - e->s_lno;
724                 split[1].lno = e->lno + tlno - e->s_lno;
725                 split[1].s_lno = plno;
726         }
727         else {
728                 split[1].lno = e->lno;
729                 split[1].s_lno = plno + (e->s_lno - tlno);
730         }
731
732         if (same < e->s_lno + e->num_lines) {
733                 /* there is a post-chunk part not blamed on parent */
734                 split[2].suspect = origin_incref(e->suspect);
735                 split[2].lno = e->lno + (same - e->s_lno);
736                 split[2].s_lno = e->s_lno + (same - e->s_lno);
737                 split[2].num_lines = e->s_lno + e->num_lines - same;
738                 chunk_end_lno = split[2].lno;
739         }
740         else
741                 chunk_end_lno = e->lno + e->num_lines;
742         split[1].num_lines = chunk_end_lno - split[1].lno;
743
744         /*
745          * if it turns out there is nothing to blame the parent for,
746          * forget about the splitting.  !split[1].suspect signals this.
747          */
748         if (split[1].num_lines < 1)
749                 return;
750         split[1].suspect = origin_incref(parent);
751 }
752
753 /*
754  * split_overlap() divided an existing blame e into up to three parts
755  * in split.  Any assigned blame is moved to queue to
756  * reflect the split.
757  */
758 static void split_blame(struct blame_entry ***blamed,
759                         struct blame_entry ***unblamed,
760                         struct blame_entry *split,
761                         struct blame_entry *e)
762 {
763         struct blame_entry *new_entry;
764
765         if (split[0].suspect && split[2].suspect) {
766                 /* The first part (reuse storage for the existing entry e) */
767                 dup_entry(unblamed, e, &split[0]);
768
769                 /* The last part -- me */
770                 new_entry = xmalloc(sizeof(*new_entry));
771                 memcpy(new_entry, &(split[2]), sizeof(struct blame_entry));
772                 add_blame_entry(unblamed, new_entry);
773
774                 /* ... and the middle part -- parent */
775                 new_entry = xmalloc(sizeof(*new_entry));
776                 memcpy(new_entry, &(split[1]), sizeof(struct blame_entry));
777                 add_blame_entry(blamed, new_entry);
778         }
779         else if (!split[0].suspect && !split[2].suspect)
780                 /*
781                  * The parent covers the entire area; reuse storage for
782                  * e and replace it with the parent.
783                  */
784                 dup_entry(blamed, e, &split[1]);
785         else if (split[0].suspect) {
786                 /* me and then parent */
787                 dup_entry(unblamed, e, &split[0]);
788
789                 new_entry = xmalloc(sizeof(*new_entry));
790                 memcpy(new_entry, &(split[1]), sizeof(struct blame_entry));
791                 add_blame_entry(blamed, new_entry);
792         }
793         else {
794                 /* parent and then me */
795                 dup_entry(blamed, e, &split[1]);
796
797                 new_entry = xmalloc(sizeof(*new_entry));
798                 memcpy(new_entry, &(split[2]), sizeof(struct blame_entry));
799                 add_blame_entry(unblamed, new_entry);
800         }
801 }
802
803 /*
804  * After splitting the blame, the origins used by the
805  * on-stack blame_entry should lose one refcnt each.
806  */
807 static void decref_split(struct blame_entry *split)
808 {
809         int i;
810
811         for (i = 0; i < 3; i++)
812                 origin_decref(split[i].suspect);
813 }
814
815 /*
816  * reverse_blame reverses the list given in head, appending tail.
817  * That allows us to build lists in reverse order, then reverse them
818  * afterwards.  This can be faster than building the list in proper
819  * order right away.  The reason is that building in proper order
820  * requires writing a link in the _previous_ element, while building
821  * in reverse order just requires placing the list head into the
822  * _current_ element.
823  */
824
825 static struct blame_entry *reverse_blame(struct blame_entry *head,
826                                          struct blame_entry *tail)
827 {
828         while (head) {
829                 struct blame_entry *next = head->next;
830                 head->next = tail;
831                 tail = head;
832                 head = next;
833         }
834         return tail;
835 }
836
837 /*
838  * Process one hunk from the patch between the current suspect for
839  * blame_entry e and its parent.  This first blames any unfinished
840  * entries before the chunk (which is where target and parent start
841  * differing) on the parent, and then splits blame entries at the
842  * start and at the end of the difference region.  Since use of -M and
843  * -C options may lead to overlapping/duplicate source line number
844  * ranges, all we can rely on from sorting/merging is the order of the
845  * first suspect line number.
846  */
847 static void blame_chunk(struct blame_entry ***dstq, struct blame_entry ***srcq,
848                         int tlno, int offset, int same,
849                         struct origin *parent)
850 {
851         struct blame_entry *e = **srcq;
852         struct blame_entry *samep = NULL, *diffp = NULL;
853
854         while (e && e->s_lno < tlno) {
855                 struct blame_entry *next = e->next;
856                 /*
857                  * current record starts before differing portion.  If
858                  * it reaches into it, we need to split it up and
859                  * examine the second part separately.
860                  */
861                 if (e->s_lno + e->num_lines > tlno) {
862                         /* Move second half to a new record */
863                         int len = tlno - e->s_lno;
864                         struct blame_entry *n = xcalloc(1, sizeof (struct blame_entry));
865                         n->suspect = e->suspect;
866                         n->lno = e->lno + len;
867                         n->s_lno = e->s_lno + len;
868                         n->num_lines = e->num_lines - len;
869                         e->num_lines = len;
870                         e->score = 0;
871                         /* Push new record to diffp */
872                         n->next = diffp;
873                         diffp = n;
874                 } else
875                         origin_decref(e->suspect);
876                 /* Pass blame for everything before the differing
877                  * chunk to the parent */
878                 e->suspect = origin_incref(parent);
879                 e->s_lno += offset;
880                 e->next = samep;
881                 samep = e;
882                 e = next;
883         }
884         /*
885          * As we don't know how much of a common stretch after this
886          * diff will occur, the currently blamed parts are all that we
887          * can assign to the parent for now.
888          */
889
890         if (samep) {
891                 **dstq = reverse_blame(samep, **dstq);
892                 *dstq = &samep->next;
893         }
894         /*
895          * Prepend the split off portions: everything after e starts
896          * after the blameable portion.
897          */
898         e = reverse_blame(diffp, e);
899
900         /*
901          * Now retain records on the target while parts are different
902          * from the parent.
903          */
904         samep = NULL;
905         diffp = NULL;
906         while (e && e->s_lno < same) {
907                 struct blame_entry *next = e->next;
908
909                 /*
910                  * If current record extends into sameness, need to split.
911                  */
912                 if (e->s_lno + e->num_lines > same) {
913                         /*
914                          * Move second half to a new record to be
915                          * processed by later chunks
916                          */
917                         int len = same - e->s_lno;
918                         struct blame_entry *n = xcalloc(1, sizeof (struct blame_entry));
919                         n->suspect = origin_incref(e->suspect);
920                         n->lno = e->lno + len;
921                         n->s_lno = e->s_lno + len;
922                         n->num_lines = e->num_lines - len;
923                         e->num_lines = len;
924                         e->score = 0;
925                         /* Push new record to samep */
926                         n->next = samep;
927                         samep = n;
928                 }
929                 e->next = diffp;
930                 diffp = e;
931                 e = next;
932         }
933         **srcq = reverse_blame(diffp, reverse_blame(samep, e));
934         /* Move across elements that are in the unblamable portion */
935         if (diffp)
936                 *srcq = &diffp->next;
937 }
938
939 struct blame_chunk_cb_data {
940         struct origin *parent;
941         long offset;
942         struct blame_entry **dstq;
943         struct blame_entry **srcq;
944 };
945
946 /* diff chunks are from parent to target */
947 static int blame_chunk_cb(long start_a, long count_a,
948                           long start_b, long count_b, void *data)
949 {
950         struct blame_chunk_cb_data *d = data;
951         if (start_a - start_b != d->offset)
952                 die("internal error in blame::blame_chunk_cb");
953         blame_chunk(&d->dstq, &d->srcq, start_b, start_a - start_b,
954                     start_b + count_b, d->parent);
955         d->offset = start_a + count_a - (start_b + count_b);
956         return 0;
957 }
958
959 /*
960  * We are looking at the origin 'target' and aiming to pass blame
961  * for the lines it is suspected to its parent.  Run diff to find
962  * which lines came from parent and pass blame for them.
963  */
964 static void pass_blame_to_parent(struct scoreboard *sb,
965                                  struct origin *target,
966                                  struct origin *parent)
967 {
968         mmfile_t file_p, file_o;
969         struct blame_chunk_cb_data d;
970         struct blame_entry *newdest = NULL;
971
972         if (!target->suspects)
973                 return; /* nothing remains for this target */
974
975         d.parent = parent;
976         d.offset = 0;
977         d.dstq = &newdest; d.srcq = &target->suspects;
978
979         fill_origin_blob(&sb->revs->diffopt, parent, &file_p);
980         fill_origin_blob(&sb->revs->diffopt, target, &file_o);
981         num_get_patch++;
982
983         if (diff_hunks(&file_p, &file_o, 0, blame_chunk_cb, &d))
984                 die("unable to generate diff (%s -> %s)",
985                     oid_to_hex(&parent->commit->object.oid),
986                     oid_to_hex(&target->commit->object.oid));
987         /* The rest are the same as the parent */
988         blame_chunk(&d.dstq, &d.srcq, INT_MAX, d.offset, INT_MAX, parent);
989         *d.dstq = NULL;
990         queue_blames(sb, parent, newdest);
991
992         return;
993 }
994
995 /*
996  * The lines in blame_entry after splitting blames many times can become
997  * very small and trivial, and at some point it becomes pointless to
998  * blame the parents.  E.g. "\t\t}\n\t}\n\n" appears everywhere in any
999  * ordinary C program, and it is not worth to say it was copied from
1000  * totally unrelated file in the parent.
1001  *
1002  * Compute how trivial the lines in the blame_entry are.
1003  */
1004 static unsigned ent_score(struct scoreboard *sb, struct blame_entry *e)
1005 {
1006         unsigned score;
1007         const char *cp, *ep;
1008
1009         if (e->score)
1010                 return e->score;
1011
1012         score = 1;
1013         cp = nth_line(sb, e->lno);
1014         ep = nth_line(sb, e->lno + e->num_lines);
1015         while (cp < ep) {
1016                 unsigned ch = *((unsigned char *)cp);
1017                 if (isalnum(ch))
1018                         score++;
1019                 cp++;
1020         }
1021         e->score = score;
1022         return score;
1023 }
1024
1025 /*
1026  * best_so_far[] and this[] are both a split of an existing blame_entry
1027  * that passes blame to the parent.  Maintain best_so_far the best split
1028  * so far, by comparing this and best_so_far and copying this into
1029  * bst_so_far as needed.
1030  */
1031 static void copy_split_if_better(struct scoreboard *sb,
1032                                  struct blame_entry *best_so_far,
1033                                  struct blame_entry *this)
1034 {
1035         int i;
1036
1037         if (!this[1].suspect)
1038                 return;
1039         if (best_so_far[1].suspect) {
1040                 if (ent_score(sb, &this[1]) < ent_score(sb, &best_so_far[1]))
1041                         return;
1042         }
1043
1044         for (i = 0; i < 3; i++)
1045                 origin_incref(this[i].suspect);
1046         decref_split(best_so_far);
1047         memcpy(best_so_far, this, sizeof(struct blame_entry [3]));
1048 }
1049
1050 /*
1051  * We are looking at a part of the final image represented by
1052  * ent (tlno and same are offset by ent->s_lno).
1053  * tlno is where we are looking at in the final image.
1054  * up to (but not including) same match preimage.
1055  * plno is where we are looking at in the preimage.
1056  *
1057  * <-------------- final image ---------------------->
1058  *       <------ent------>
1059  *         ^tlno ^same
1060  *    <---------preimage----->
1061  *         ^plno
1062  *
1063  * All line numbers are 0-based.
1064  */
1065 static void handle_split(struct scoreboard *sb,
1066                          struct blame_entry *ent,
1067                          int tlno, int plno, int same,
1068                          struct origin *parent,
1069                          struct blame_entry *split)
1070 {
1071         if (ent->num_lines <= tlno)
1072                 return;
1073         if (tlno < same) {
1074                 struct blame_entry this[3];
1075                 tlno += ent->s_lno;
1076                 same += ent->s_lno;
1077                 split_overlap(this, ent, tlno, plno, same, parent);
1078                 copy_split_if_better(sb, split, this);
1079                 decref_split(this);
1080         }
1081 }
1082
1083 struct handle_split_cb_data {
1084         struct scoreboard *sb;
1085         struct blame_entry *ent;
1086         struct origin *parent;
1087         struct blame_entry *split;
1088         long plno;
1089         long tlno;
1090 };
1091
1092 static int handle_split_cb(long start_a, long count_a,
1093                            long start_b, long count_b, void *data)
1094 {
1095         struct handle_split_cb_data *d = data;
1096         handle_split(d->sb, d->ent, d->tlno, d->plno, start_b, d->parent,
1097                      d->split);
1098         d->plno = start_a + count_a;
1099         d->tlno = start_b + count_b;
1100         return 0;
1101 }
1102
1103 /*
1104  * Find the lines from parent that are the same as ent so that
1105  * we can pass blames to it.  file_p has the blob contents for
1106  * the parent.
1107  */
1108 static void find_copy_in_blob(struct scoreboard *sb,
1109                               struct blame_entry *ent,
1110                               struct origin *parent,
1111                               struct blame_entry *split,
1112                               mmfile_t *file_p)
1113 {
1114         const char *cp;
1115         mmfile_t file_o;
1116         struct handle_split_cb_data d;
1117
1118         memset(&d, 0, sizeof(d));
1119         d.sb = sb; d.ent = ent; d.parent = parent; d.split = split;
1120         /*
1121          * Prepare mmfile that contains only the lines in ent.
1122          */
1123         cp = nth_line(sb, ent->lno);
1124         file_o.ptr = (char *) cp;
1125         file_o.size = nth_line(sb, ent->lno + ent->num_lines) - cp;
1126
1127         /*
1128          * file_o is a part of final image we are annotating.
1129          * file_p partially may match that image.
1130          */
1131         memset(split, 0, sizeof(struct blame_entry [3]));
1132         if (diff_hunks(file_p, &file_o, 1, handle_split_cb, &d))
1133                 die("unable to generate diff (%s)",
1134                     oid_to_hex(&parent->commit->object.oid));
1135         /* remainder, if any, all match the preimage */
1136         handle_split(sb, ent, d.tlno, d.plno, ent->num_lines, parent, split);
1137 }
1138
1139 /* Move all blame entries from list *source that have a score smaller
1140  * than score_min to the front of list *small.
1141  * Returns a pointer to the link pointing to the old head of the small list.
1142  */
1143
1144 static struct blame_entry **filter_small(struct scoreboard *sb,
1145                                          struct blame_entry **small,
1146                                          struct blame_entry **source,
1147                                          unsigned score_min)
1148 {
1149         struct blame_entry *p = *source;
1150         struct blame_entry *oldsmall = *small;
1151         while (p) {
1152                 if (ent_score(sb, p) <= score_min) {
1153                         *small = p;
1154                         small = &p->next;
1155                         p = *small;
1156                 } else {
1157                         *source = p;
1158                         source = &p->next;
1159                         p = *source;
1160                 }
1161         }
1162         *small = oldsmall;
1163         *source = NULL;
1164         return small;
1165 }
1166
1167 /*
1168  * See if lines currently target is suspected for can be attributed to
1169  * parent.
1170  */
1171 static void find_move_in_parent(struct scoreboard *sb,
1172                                 struct blame_entry ***blamed,
1173                                 struct blame_entry **toosmall,
1174                                 struct origin *target,
1175                                 struct origin *parent)
1176 {
1177         struct blame_entry *e, split[3];
1178         struct blame_entry *unblamed = target->suspects;
1179         struct blame_entry *leftover = NULL;
1180         mmfile_t file_p;
1181
1182         if (!unblamed)
1183                 return; /* nothing remains for this target */
1184
1185         fill_origin_blob(&sb->revs->diffopt, parent, &file_p);
1186         if (!file_p.ptr)
1187                 return;
1188
1189         /* At each iteration, unblamed has a NULL-terminated list of
1190          * entries that have not yet been tested for blame.  leftover
1191          * contains the reversed list of entries that have been tested
1192          * without being assignable to the parent.
1193          */
1194         do {
1195                 struct blame_entry **unblamedtail = &unblamed;
1196                 struct blame_entry *next;
1197                 for (e = unblamed; e; e = next) {
1198                         next = e->next;
1199                         find_copy_in_blob(sb, e, parent, split, &file_p);
1200                         if (split[1].suspect &&
1201                             blame_move_score < ent_score(sb, &split[1])) {
1202                                 split_blame(blamed, &unblamedtail, split, e);
1203                         } else {
1204                                 e->next = leftover;
1205                                 leftover = e;
1206                         }
1207                         decref_split(split);
1208                 }
1209                 *unblamedtail = NULL;
1210                 toosmall = filter_small(sb, toosmall, &unblamed, blame_move_score);
1211         } while (unblamed);
1212         target->suspects = reverse_blame(leftover, NULL);
1213 }
1214
1215 struct blame_list {
1216         struct blame_entry *ent;
1217         struct blame_entry split[3];
1218 };
1219
1220 /*
1221  * Count the number of entries the target is suspected for,
1222  * and prepare a list of entry and the best split.
1223  */
1224 static struct blame_list *setup_blame_list(struct blame_entry *unblamed,
1225                                            int *num_ents_p)
1226 {
1227         struct blame_entry *e;
1228         int num_ents, i;
1229         struct blame_list *blame_list = NULL;
1230
1231         for (e = unblamed, num_ents = 0; e; e = e->next)
1232                 num_ents++;
1233         if (num_ents) {
1234                 blame_list = xcalloc(num_ents, sizeof(struct blame_list));
1235                 for (e = unblamed, i = 0; e; e = e->next)
1236                         blame_list[i++].ent = e;
1237         }
1238         *num_ents_p = num_ents;
1239         return blame_list;
1240 }
1241
1242 /*
1243  * For lines target is suspected for, see if we can find code movement
1244  * across file boundary from the parent commit.  porigin is the path
1245  * in the parent we already tried.
1246  */
1247 static void find_copy_in_parent(struct scoreboard *sb,
1248                                 struct blame_entry ***blamed,
1249                                 struct blame_entry **toosmall,
1250                                 struct origin *target,
1251                                 struct commit *parent,
1252                                 struct origin *porigin,
1253                                 int opt)
1254 {
1255         struct diff_options diff_opts;
1256         int i, j;
1257         struct blame_list *blame_list;
1258         int num_ents;
1259         struct blame_entry *unblamed = target->suspects;
1260         struct blame_entry *leftover = NULL;
1261
1262         if (!unblamed)
1263                 return; /* nothing remains for this target */
1264
1265         diff_setup(&diff_opts);
1266         DIFF_OPT_SET(&diff_opts, RECURSIVE);
1267         diff_opts.output_format = DIFF_FORMAT_NO_OUTPUT;
1268
1269         diff_setup_done(&diff_opts);
1270
1271         /* Try "find copies harder" on new path if requested;
1272          * we do not want to use diffcore_rename() actually to
1273          * match things up; find_copies_harder is set only to
1274          * force diff_tree_sha1() to feed all filepairs to diff_queue,
1275          * and this code needs to be after diff_setup_done(), which
1276          * usually makes find-copies-harder imply copy detection.
1277          */
1278         if ((opt & PICKAXE_BLAME_COPY_HARDEST)
1279             || ((opt & PICKAXE_BLAME_COPY_HARDER)
1280                 && (!porigin || strcmp(target->path, porigin->path))))
1281                 DIFF_OPT_SET(&diff_opts, FIND_COPIES_HARDER);
1282
1283         if (is_null_oid(&target->commit->object.oid))
1284                 do_diff_cache(parent->tree->object.oid.hash, &diff_opts);
1285         else
1286                 diff_tree_sha1(parent->tree->object.oid.hash,
1287                                target->commit->tree->object.oid.hash,
1288                                "", &diff_opts);
1289
1290         if (!DIFF_OPT_TST(&diff_opts, FIND_COPIES_HARDER))
1291                 diffcore_std(&diff_opts);
1292
1293         do {
1294                 struct blame_entry **unblamedtail = &unblamed;
1295                 blame_list = setup_blame_list(unblamed, &num_ents);
1296
1297                 for (i = 0; i < diff_queued_diff.nr; i++) {
1298                         struct diff_filepair *p = diff_queued_diff.queue[i];
1299                         struct origin *norigin;
1300                         mmfile_t file_p;
1301                         struct blame_entry this[3];
1302
1303                         if (!DIFF_FILE_VALID(p->one))
1304                                 continue; /* does not exist in parent */
1305                         if (S_ISGITLINK(p->one->mode))
1306                                 continue; /* ignore git links */
1307                         if (porigin && !strcmp(p->one->path, porigin->path))
1308                                 /* find_move already dealt with this path */
1309                                 continue;
1310
1311                         norigin = get_origin(sb, parent, p->one->path);
1312                         hashcpy(norigin->blob_sha1, p->one->sha1);
1313                         norigin->mode = p->one->mode;
1314                         fill_origin_blob(&sb->revs->diffopt, norigin, &file_p);
1315                         if (!file_p.ptr)
1316                                 continue;
1317
1318                         for (j = 0; j < num_ents; j++) {
1319                                 find_copy_in_blob(sb, blame_list[j].ent,
1320                                                   norigin, this, &file_p);
1321                                 copy_split_if_better(sb, blame_list[j].split,
1322                                                      this);
1323                                 decref_split(this);
1324                         }
1325                         origin_decref(norigin);
1326                 }
1327
1328                 for (j = 0; j < num_ents; j++) {
1329                         struct blame_entry *split = blame_list[j].split;
1330                         if (split[1].suspect &&
1331                             blame_copy_score < ent_score(sb, &split[1])) {
1332                                 split_blame(blamed, &unblamedtail, split,
1333                                             blame_list[j].ent);
1334                         } else {
1335                                 blame_list[j].ent->next = leftover;
1336                                 leftover = blame_list[j].ent;
1337                         }
1338                         decref_split(split);
1339                 }
1340                 free(blame_list);
1341                 *unblamedtail = NULL;
1342                 toosmall = filter_small(sb, toosmall, &unblamed, blame_copy_score);
1343         } while (unblamed);
1344         target->suspects = reverse_blame(leftover, NULL);
1345         diff_flush(&diff_opts);
1346         free_pathspec(&diff_opts.pathspec);
1347 }
1348
1349 /*
1350  * The blobs of origin and porigin exactly match, so everything
1351  * origin is suspected for can be blamed on the parent.
1352  */
1353 static void pass_whole_blame(struct scoreboard *sb,
1354                              struct origin *origin, struct origin *porigin)
1355 {
1356         struct blame_entry *e, *suspects;
1357
1358         if (!porigin->file.ptr && origin->file.ptr) {
1359                 /* Steal its file */
1360                 porigin->file = origin->file;
1361                 origin->file.ptr = NULL;
1362         }
1363         suspects = origin->suspects;
1364         origin->suspects = NULL;
1365         for (e = suspects; e; e = e->next) {
1366                 origin_incref(porigin);
1367                 origin_decref(e->suspect);
1368                 e->suspect = porigin;
1369         }
1370         queue_blames(sb, porigin, suspects);
1371 }
1372
1373 /*
1374  * We pass blame from the current commit to its parents.  We keep saying
1375  * "parent" (and "porigin"), but what we mean is to find scapegoat to
1376  * exonerate ourselves.
1377  */
1378 static struct commit_list *first_scapegoat(struct rev_info *revs, struct commit *commit)
1379 {
1380         if (!reverse) {
1381                 if (revs->first_parent_only &&
1382                     commit->parents &&
1383                     commit->parents->next) {
1384                         free_commit_list(commit->parents->next);
1385                         commit->parents->next = NULL;
1386                 }
1387                 return commit->parents;
1388         }
1389         return lookup_decoration(&revs->children, &commit->object);
1390 }
1391
1392 static int num_scapegoats(struct rev_info *revs, struct commit *commit)
1393 {
1394         struct commit_list *l = first_scapegoat(revs, commit);
1395         return commit_list_count(l);
1396 }
1397
1398 /* Distribute collected unsorted blames to the respected sorted lists
1399  * in the various origins.
1400  */
1401 static void distribute_blame(struct scoreboard *sb, struct blame_entry *blamed)
1402 {
1403         blamed = blame_sort(blamed, compare_blame_suspect);
1404         while (blamed)
1405         {
1406                 struct origin *porigin = blamed->suspect;
1407                 struct blame_entry *suspects = NULL;
1408                 do {
1409                         struct blame_entry *next = blamed->next;
1410                         blamed->next = suspects;
1411                         suspects = blamed;
1412                         blamed = next;
1413                 } while (blamed && blamed->suspect == porigin);
1414                 suspects = reverse_blame(suspects, NULL);
1415                 queue_blames(sb, porigin, suspects);
1416         }
1417 }
1418
1419 #define MAXSG 16
1420
1421 static void pass_blame(struct scoreboard *sb, struct origin *origin, int opt)
1422 {
1423         struct rev_info *revs = sb->revs;
1424         int i, pass, num_sg;
1425         struct commit *commit = origin->commit;
1426         struct commit_list *sg;
1427         struct origin *sg_buf[MAXSG];
1428         struct origin *porigin, **sg_origin = sg_buf;
1429         struct blame_entry *toosmall = NULL;
1430         struct blame_entry *blames, **blametail = &blames;
1431
1432         num_sg = num_scapegoats(revs, commit);
1433         if (!num_sg)
1434                 goto finish;
1435         else if (num_sg < ARRAY_SIZE(sg_buf))
1436                 memset(sg_buf, 0, sizeof(sg_buf));
1437         else
1438                 sg_origin = xcalloc(num_sg, sizeof(*sg_origin));
1439
1440         /*
1441          * The first pass looks for unrenamed path to optimize for
1442          * common cases, then we look for renames in the second pass.
1443          */
1444         for (pass = 0; pass < 2 - no_whole_file_rename; pass++) {
1445                 struct origin *(*find)(struct scoreboard *,
1446                                        struct commit *, struct origin *);
1447                 find = pass ? find_rename : find_origin;
1448
1449                 for (i = 0, sg = first_scapegoat(revs, commit);
1450                      i < num_sg && sg;
1451                      sg = sg->next, i++) {
1452                         struct commit *p = sg->item;
1453                         int j, same;
1454
1455                         if (sg_origin[i])
1456                                 continue;
1457                         if (parse_commit(p))
1458                                 continue;
1459                         porigin = find(sb, p, origin);
1460                         if (!porigin)
1461                                 continue;
1462                         if (!hashcmp(porigin->blob_sha1, origin->blob_sha1)) {
1463                                 pass_whole_blame(sb, origin, porigin);
1464                                 origin_decref(porigin);
1465                                 goto finish;
1466                         }
1467                         for (j = same = 0; j < i; j++)
1468                                 if (sg_origin[j] &&
1469                                     !hashcmp(sg_origin[j]->blob_sha1,
1470                                              porigin->blob_sha1)) {
1471                                         same = 1;
1472                                         break;
1473                                 }
1474                         if (!same)
1475                                 sg_origin[i] = porigin;
1476                         else
1477                                 origin_decref(porigin);
1478                 }
1479         }
1480
1481         num_commits++;
1482         for (i = 0, sg = first_scapegoat(revs, commit);
1483              i < num_sg && sg;
1484              sg = sg->next, i++) {
1485                 struct origin *porigin = sg_origin[i];
1486                 if (!porigin)
1487                         continue;
1488                 if (!origin->previous) {
1489                         origin_incref(porigin);
1490                         origin->previous = porigin;
1491                 }
1492                 pass_blame_to_parent(sb, origin, porigin);
1493                 if (!origin->suspects)
1494                         goto finish;
1495         }
1496
1497         /*
1498          * Optionally find moves in parents' files.
1499          */
1500         if (opt & PICKAXE_BLAME_MOVE) {
1501                 filter_small(sb, &toosmall, &origin->suspects, blame_move_score);
1502                 if (origin->suspects) {
1503                         for (i = 0, sg = first_scapegoat(revs, commit);
1504                              i < num_sg && sg;
1505                              sg = sg->next, i++) {
1506                                 struct origin *porigin = sg_origin[i];
1507                                 if (!porigin)
1508                                         continue;
1509                                 find_move_in_parent(sb, &blametail, &toosmall, origin, porigin);
1510                                 if (!origin->suspects)
1511                                         break;
1512                         }
1513                 }
1514         }
1515
1516         /*
1517          * Optionally find copies from parents' files.
1518          */
1519         if (opt & PICKAXE_BLAME_COPY) {
1520                 if (blame_copy_score > blame_move_score)
1521                         filter_small(sb, &toosmall, &origin->suspects, blame_copy_score);
1522                 else if (blame_copy_score < blame_move_score) {
1523                         origin->suspects = blame_merge(origin->suspects, toosmall);
1524                         toosmall = NULL;
1525                         filter_small(sb, &toosmall, &origin->suspects, blame_copy_score);
1526                 }
1527                 if (!origin->suspects)
1528                         goto finish;
1529
1530                 for (i = 0, sg = first_scapegoat(revs, commit);
1531                      i < num_sg && sg;
1532                      sg = sg->next, i++) {
1533                         struct origin *porigin = sg_origin[i];
1534                         find_copy_in_parent(sb, &blametail, &toosmall,
1535                                             origin, sg->item, porigin, opt);
1536                         if (!origin->suspects)
1537                                 goto finish;
1538                 }
1539         }
1540
1541 finish:
1542         *blametail = NULL;
1543         distribute_blame(sb, blames);
1544         /*
1545          * prepend toosmall to origin->suspects
1546          *
1547          * There is no point in sorting: this ends up on a big
1548          * unsorted list in the caller anyway.
1549          */
1550         if (toosmall) {
1551                 struct blame_entry **tail = &toosmall;
1552                 while (*tail)
1553                         tail = &(*tail)->next;
1554                 *tail = origin->suspects;
1555                 origin->suspects = toosmall;
1556         }
1557         for (i = 0; i < num_sg; i++) {
1558                 if (sg_origin[i]) {
1559                         drop_origin_blob(sg_origin[i]);
1560                         origin_decref(sg_origin[i]);
1561                 }
1562         }
1563         drop_origin_blob(origin);
1564         if (sg_buf != sg_origin)
1565                 free(sg_origin);
1566 }
1567
1568 /*
1569  * Information on commits, used for output.
1570  */
1571 struct commit_info {
1572         struct strbuf author;
1573         struct strbuf author_mail;
1574         unsigned long author_time;
1575         struct strbuf author_tz;
1576
1577         /* filled only when asked for details */
1578         struct strbuf committer;
1579         struct strbuf committer_mail;
1580         unsigned long committer_time;
1581         struct strbuf committer_tz;
1582
1583         struct strbuf summary;
1584 };
1585
1586 /*
1587  * Parse author/committer line in the commit object buffer
1588  */
1589 static void get_ac_line(const char *inbuf, const char *what,
1590         struct strbuf *name, struct strbuf *mail,
1591         unsigned long *time, struct strbuf *tz)
1592 {
1593         struct ident_split ident;
1594         size_t len, maillen, namelen;
1595         char *tmp, *endp;
1596         const char *namebuf, *mailbuf;
1597
1598         tmp = strstr(inbuf, what);
1599         if (!tmp)
1600                 goto error_out;
1601         tmp += strlen(what);
1602         endp = strchr(tmp, '\n');
1603         if (!endp)
1604                 len = strlen(tmp);
1605         else
1606                 len = endp - tmp;
1607
1608         if (split_ident_line(&ident, tmp, len)) {
1609         error_out:
1610                 /* Ugh */
1611                 tmp = "(unknown)";
1612                 strbuf_addstr(name, tmp);
1613                 strbuf_addstr(mail, tmp);
1614                 strbuf_addstr(tz, tmp);
1615                 *time = 0;
1616                 return;
1617         }
1618
1619         namelen = ident.name_end - ident.name_begin;
1620         namebuf = ident.name_begin;
1621
1622         maillen = ident.mail_end - ident.mail_begin;
1623         mailbuf = ident.mail_begin;
1624
1625         if (ident.date_begin && ident.date_end)
1626                 *time = strtoul(ident.date_begin, NULL, 10);
1627         else
1628                 *time = 0;
1629
1630         if (ident.tz_begin && ident.tz_end)
1631                 strbuf_add(tz, ident.tz_begin, ident.tz_end - ident.tz_begin);
1632         else
1633                 strbuf_addstr(tz, "(unknown)");
1634
1635         /*
1636          * Now, convert both name and e-mail using mailmap
1637          */
1638         map_user(&mailmap, &mailbuf, &maillen,
1639                  &namebuf, &namelen);
1640
1641         strbuf_addf(mail, "<%.*s>", (int)maillen, mailbuf);
1642         strbuf_add(name, namebuf, namelen);
1643 }
1644
1645 static void commit_info_init(struct commit_info *ci)
1646 {
1647
1648         strbuf_init(&ci->author, 0);
1649         strbuf_init(&ci->author_mail, 0);
1650         strbuf_init(&ci->author_tz, 0);
1651         strbuf_init(&ci->committer, 0);
1652         strbuf_init(&ci->committer_mail, 0);
1653         strbuf_init(&ci->committer_tz, 0);
1654         strbuf_init(&ci->summary, 0);
1655 }
1656
1657 static void commit_info_destroy(struct commit_info *ci)
1658 {
1659
1660         strbuf_release(&ci->author);
1661         strbuf_release(&ci->author_mail);
1662         strbuf_release(&ci->author_tz);
1663         strbuf_release(&ci->committer);
1664         strbuf_release(&ci->committer_mail);
1665         strbuf_release(&ci->committer_tz);
1666         strbuf_release(&ci->summary);
1667 }
1668
1669 static void get_commit_info(struct commit *commit,
1670                             struct commit_info *ret,
1671                             int detailed)
1672 {
1673         int len;
1674         const char *subject, *encoding;
1675         const char *message;
1676
1677         commit_info_init(ret);
1678
1679         encoding = get_log_output_encoding();
1680         message = logmsg_reencode(commit, NULL, encoding);
1681         get_ac_line(message, "\nauthor ",
1682                     &ret->author, &ret->author_mail,
1683                     &ret->author_time, &ret->author_tz);
1684
1685         if (!detailed) {
1686                 unuse_commit_buffer(commit, message);
1687                 return;
1688         }
1689
1690         get_ac_line(message, "\ncommitter ",
1691                     &ret->committer, &ret->committer_mail,
1692                     &ret->committer_time, &ret->committer_tz);
1693
1694         len = find_commit_subject(message, &subject);
1695         if (len)
1696                 strbuf_add(&ret->summary, subject, len);
1697         else
1698                 strbuf_addf(&ret->summary, "(%s)", oid_to_hex(&commit->object.oid));
1699
1700         unuse_commit_buffer(commit, message);
1701 }
1702
1703 /*
1704  * To allow LF and other nonportable characters in pathnames,
1705  * they are c-style quoted as needed.
1706  */
1707 static void write_filename_info(const char *path)
1708 {
1709         printf("filename ");
1710         write_name_quoted(path, stdout, '\n');
1711 }
1712
1713 /*
1714  * Porcelain/Incremental format wants to show a lot of details per
1715  * commit.  Instead of repeating this every line, emit it only once,
1716  * the first time each commit appears in the output (unless the
1717  * user has specifically asked for us to repeat).
1718  */
1719 static int emit_one_suspect_detail(struct origin *suspect, int repeat)
1720 {
1721         struct commit_info ci;
1722
1723         if (!repeat && (suspect->commit->object.flags & METAINFO_SHOWN))
1724                 return 0;
1725
1726         suspect->commit->object.flags |= METAINFO_SHOWN;
1727         get_commit_info(suspect->commit, &ci, 1);
1728         printf("author %s\n", ci.author.buf);
1729         printf("author-mail %s\n", ci.author_mail.buf);
1730         printf("author-time %lu\n", ci.author_time);
1731         printf("author-tz %s\n", ci.author_tz.buf);
1732         printf("committer %s\n", ci.committer.buf);
1733         printf("committer-mail %s\n", ci.committer_mail.buf);
1734         printf("committer-time %lu\n", ci.committer_time);
1735         printf("committer-tz %s\n", ci.committer_tz.buf);
1736         printf("summary %s\n", ci.summary.buf);
1737         if (suspect->commit->object.flags & UNINTERESTING)
1738                 printf("boundary\n");
1739         if (suspect->previous) {
1740                 struct origin *prev = suspect->previous;
1741                 printf("previous %s ", oid_to_hex(&prev->commit->object.oid));
1742                 write_name_quoted(prev->path, stdout, '\n');
1743         }
1744
1745         commit_info_destroy(&ci);
1746
1747         return 1;
1748 }
1749
1750 /*
1751  * The blame_entry is found to be guilty for the range.
1752  * Show it in incremental output.
1753  */
1754 static void found_guilty_entry(struct blame_entry *ent,
1755                            struct progress_info *pi)
1756 {
1757         if (incremental) {
1758                 struct origin *suspect = ent->suspect;
1759
1760                 printf("%s %d %d %d\n",
1761                        oid_to_hex(&suspect->commit->object.oid),
1762                        ent->s_lno + 1, ent->lno + 1, ent->num_lines);
1763                 emit_one_suspect_detail(suspect, 0);
1764                 write_filename_info(suspect->path);
1765                 maybe_flush_or_die(stdout, "stdout");
1766         }
1767         pi->blamed_lines += ent->num_lines;
1768         display_progress(pi->progress, pi->blamed_lines);
1769 }
1770
1771 /*
1772  * The main loop -- while we have blobs with lines whose true origin
1773  * is still unknown, pick one blob, and allow its lines to pass blames
1774  * to its parents. */
1775 static void assign_blame(struct scoreboard *sb, int opt)
1776 {
1777         struct rev_info *revs = sb->revs;
1778         struct commit *commit = prio_queue_get(&sb->commits);
1779         struct progress_info pi = { NULL, 0 };
1780
1781         if (show_progress)
1782                 pi.progress = start_progress_delay(_("Blaming lines"),
1783                                                    sb->num_lines, 50, 1);
1784
1785         while (commit) {
1786                 struct blame_entry *ent;
1787                 struct origin *suspect = commit->util;
1788
1789                 /* find one suspect to break down */
1790                 while (suspect && !suspect->suspects)
1791                         suspect = suspect->next;
1792
1793                 if (!suspect) {
1794                         commit = prio_queue_get(&sb->commits);
1795                         continue;
1796                 }
1797
1798                 assert(commit == suspect->commit);
1799
1800                 /*
1801                  * We will use this suspect later in the loop,
1802                  * so hold onto it in the meantime.
1803                  */
1804                 origin_incref(suspect);
1805                 parse_commit(commit);
1806                 if (reverse ||
1807                     (!(commit->object.flags & UNINTERESTING) &&
1808                      !(revs->max_age != -1 && commit->date < revs->max_age)))
1809                         pass_blame(sb, suspect, opt);
1810                 else {
1811                         commit->object.flags |= UNINTERESTING;
1812                         if (commit->object.parsed)
1813                                 mark_parents_uninteresting(commit);
1814                 }
1815                 /* treat root commit as boundary */
1816                 if (!commit->parents && !show_root)
1817                         commit->object.flags |= UNINTERESTING;
1818
1819                 /* Take responsibility for the remaining entries */
1820                 ent = suspect->suspects;
1821                 if (ent) {
1822                         suspect->guilty = 1;
1823                         for (;;) {
1824                                 struct blame_entry *next = ent->next;
1825                                 found_guilty_entry(ent, &pi);
1826                                 if (next) {
1827                                         ent = next;
1828                                         continue;
1829                                 }
1830                                 ent->next = sb->ent;
1831                                 sb->ent = suspect->suspects;
1832                                 suspect->suspects = NULL;
1833                                 break;
1834                         }
1835                 }
1836                 origin_decref(suspect);
1837
1838                 if (DEBUG) /* sanity */
1839                         sanity_check_refcnt(sb);
1840         }
1841
1842         stop_progress(&pi.progress);
1843 }
1844
1845 static const char *format_time(unsigned long time, const char *tz_str,
1846                                int show_raw_time)
1847 {
1848         static struct strbuf time_buf = STRBUF_INIT;
1849
1850         strbuf_reset(&time_buf);
1851         if (show_raw_time) {
1852                 strbuf_addf(&time_buf, "%lu %s", time, tz_str);
1853         }
1854         else {
1855                 const char *time_str;
1856                 size_t time_width;
1857                 int tz;
1858                 tz = atoi(tz_str);
1859                 time_str = show_date(time, tz, &blame_date_mode);
1860                 strbuf_addstr(&time_buf, time_str);
1861                 /*
1862                  * Add space paddings to time_buf to display a fixed width
1863                  * string, and use time_width for display width calibration.
1864                  */
1865                 for (time_width = utf8_strwidth(time_str);
1866                      time_width < blame_date_width;
1867                      time_width++)
1868                         strbuf_addch(&time_buf, ' ');
1869         }
1870         return time_buf.buf;
1871 }
1872
1873 #define OUTPUT_ANNOTATE_COMPAT  001
1874 #define OUTPUT_LONG_OBJECT_NAME 002
1875 #define OUTPUT_RAW_TIMESTAMP    004
1876 #define OUTPUT_PORCELAIN        010
1877 #define OUTPUT_SHOW_NAME        020
1878 #define OUTPUT_SHOW_NUMBER      040
1879 #define OUTPUT_SHOW_SCORE      0100
1880 #define OUTPUT_NO_AUTHOR       0200
1881 #define OUTPUT_SHOW_EMAIL       0400
1882 #define OUTPUT_LINE_PORCELAIN 01000
1883
1884 static void emit_porcelain_details(struct origin *suspect, int repeat)
1885 {
1886         if (emit_one_suspect_detail(suspect, repeat) ||
1887             (suspect->commit->object.flags & MORE_THAN_ONE_PATH))
1888                 write_filename_info(suspect->path);
1889 }
1890
1891 static void emit_porcelain(struct scoreboard *sb, struct blame_entry *ent,
1892                            int opt)
1893 {
1894         int repeat = opt & OUTPUT_LINE_PORCELAIN;
1895         int cnt;
1896         const char *cp;
1897         struct origin *suspect = ent->suspect;
1898         char hex[GIT_SHA1_HEXSZ + 1];
1899
1900         sha1_to_hex_r(hex, suspect->commit->object.oid.hash);
1901         printf("%s %d %d %d\n",
1902                hex,
1903                ent->s_lno + 1,
1904                ent->lno + 1,
1905                ent->num_lines);
1906         emit_porcelain_details(suspect, repeat);
1907
1908         cp = nth_line(sb, ent->lno);
1909         for (cnt = 0; cnt < ent->num_lines; cnt++) {
1910                 char ch;
1911                 if (cnt) {
1912                         printf("%s %d %d\n", hex,
1913                                ent->s_lno + 1 + cnt,
1914                                ent->lno + 1 + cnt);
1915                         if (repeat)
1916                                 emit_porcelain_details(suspect, 1);
1917                 }
1918                 putchar('\t');
1919                 do {
1920                         ch = *cp++;
1921                         putchar(ch);
1922                 } while (ch != '\n' &&
1923                          cp < sb->final_buf + sb->final_buf_size);
1924         }
1925
1926         if (sb->final_buf_size && cp[-1] != '\n')
1927                 putchar('\n');
1928 }
1929
1930 static void emit_other(struct scoreboard *sb, struct blame_entry *ent, int opt)
1931 {
1932         int cnt;
1933         const char *cp;
1934         struct origin *suspect = ent->suspect;
1935         struct commit_info ci;
1936         char hex[GIT_SHA1_HEXSZ + 1];
1937         int show_raw_time = !!(opt & OUTPUT_RAW_TIMESTAMP);
1938
1939         get_commit_info(suspect->commit, &ci, 1);
1940         sha1_to_hex_r(hex, suspect->commit->object.oid.hash);
1941
1942         cp = nth_line(sb, ent->lno);
1943         for (cnt = 0; cnt < ent->num_lines; cnt++) {
1944                 char ch;
1945                 int length = (opt & OUTPUT_LONG_OBJECT_NAME) ? 40 : abbrev;
1946
1947                 if (suspect->commit->object.flags & UNINTERESTING) {
1948                         if (blank_boundary)
1949                                 memset(hex, ' ', length);
1950                         else if (!(opt & OUTPUT_ANNOTATE_COMPAT)) {
1951                                 length--;
1952                                 putchar('^');
1953                         }
1954                 }
1955
1956                 printf("%.*s", length, hex);
1957                 if (opt & OUTPUT_ANNOTATE_COMPAT) {
1958                         const char *name;
1959                         if (opt & OUTPUT_SHOW_EMAIL)
1960                                 name = ci.author_mail.buf;
1961                         else
1962                                 name = ci.author.buf;
1963                         printf("\t(%10s\t%10s\t%d)", name,
1964                                format_time(ci.author_time, ci.author_tz.buf,
1965                                            show_raw_time),
1966                                ent->lno + 1 + cnt);
1967                 } else {
1968                         if (opt & OUTPUT_SHOW_SCORE)
1969                                 printf(" %*d %02d",
1970                                        max_score_digits, ent->score,
1971                                        ent->suspect->refcnt);
1972                         if (opt & OUTPUT_SHOW_NAME)
1973                                 printf(" %-*.*s", longest_file, longest_file,
1974                                        suspect->path);
1975                         if (opt & OUTPUT_SHOW_NUMBER)
1976                                 printf(" %*d", max_orig_digits,
1977                                        ent->s_lno + 1 + cnt);
1978
1979                         if (!(opt & OUTPUT_NO_AUTHOR)) {
1980                                 const char *name;
1981                                 int pad;
1982                                 if (opt & OUTPUT_SHOW_EMAIL)
1983                                         name = ci.author_mail.buf;
1984                                 else
1985                                         name = ci.author.buf;
1986                                 pad = longest_author - utf8_strwidth(name);
1987                                 printf(" (%s%*s %10s",
1988                                        name, pad, "",
1989                                        format_time(ci.author_time,
1990                                                    ci.author_tz.buf,
1991                                                    show_raw_time));
1992                         }
1993                         printf(" %*d) ",
1994                                max_digits, ent->lno + 1 + cnt);
1995                 }
1996                 do {
1997                         ch = *cp++;
1998                         putchar(ch);
1999                 } while (ch != '\n' &&
2000                          cp < sb->final_buf + sb->final_buf_size);
2001         }
2002
2003         if (sb->final_buf_size && cp[-1] != '\n')
2004                 putchar('\n');
2005
2006         commit_info_destroy(&ci);
2007 }
2008
2009 static void output(struct scoreboard *sb, int option)
2010 {
2011         struct blame_entry *ent;
2012
2013         if (option & OUTPUT_PORCELAIN) {
2014                 for (ent = sb->ent; ent; ent = ent->next) {
2015                         int count = 0;
2016                         struct origin *suspect;
2017                         struct commit *commit = ent->suspect->commit;
2018                         if (commit->object.flags & MORE_THAN_ONE_PATH)
2019                                 continue;
2020                         for (suspect = commit->util; suspect; suspect = suspect->next) {
2021                                 if (suspect->guilty && count++) {
2022                                         commit->object.flags |= MORE_THAN_ONE_PATH;
2023                                         break;
2024                                 }
2025                         }
2026                 }
2027         }
2028
2029         for (ent = sb->ent; ent; ent = ent->next) {
2030                 if (option & OUTPUT_PORCELAIN)
2031                         emit_porcelain(sb, ent, option);
2032                 else {
2033                         emit_other(sb, ent, option);
2034                 }
2035         }
2036 }
2037
2038 static const char *get_next_line(const char *start, const char *end)
2039 {
2040         const char *nl = memchr(start, '\n', end - start);
2041         return nl ? nl + 1 : end;
2042 }
2043
2044 /*
2045  * To allow quick access to the contents of nth line in the
2046  * final image, prepare an index in the scoreboard.
2047  */
2048 static int prepare_lines(struct scoreboard *sb)
2049 {
2050         const char *buf = sb->final_buf;
2051         unsigned long len = sb->final_buf_size;
2052         const char *end = buf + len;
2053         const char *p;
2054         int *lineno;
2055         int num = 0;
2056
2057         for (p = buf; p < end; p = get_next_line(p, end))
2058                 num++;
2059
2060         ALLOC_ARRAY(sb->lineno, num + 1);
2061         lineno = sb->lineno;
2062
2063         for (p = buf; p < end; p = get_next_line(p, end))
2064                 *lineno++ = p - buf;
2065
2066         *lineno = len;
2067
2068         sb->num_lines = num;
2069         return sb->num_lines;
2070 }
2071
2072 /*
2073  * Add phony grafts for use with -S; this is primarily to
2074  * support git's cvsserver that wants to give a linear history
2075  * to its clients.
2076  */
2077 static int read_ancestry(const char *graft_file)
2078 {
2079         FILE *fp = fopen(graft_file, "r");
2080         struct strbuf buf = STRBUF_INIT;
2081         if (!fp)
2082                 return -1;
2083         while (!strbuf_getwholeline(&buf, fp, '\n')) {
2084                 /* The format is just "Commit Parent1 Parent2 ...\n" */
2085                 struct commit_graft *graft = read_graft_line(buf.buf, buf.len);
2086                 if (graft)
2087                         register_commit_graft(graft, 0);
2088         }
2089         fclose(fp);
2090         strbuf_release(&buf);
2091         return 0;
2092 }
2093
2094 static int update_auto_abbrev(int auto_abbrev, struct origin *suspect)
2095 {
2096         const char *uniq = find_unique_abbrev(suspect->commit->object.oid.hash,
2097                                               auto_abbrev);
2098         int len = strlen(uniq);
2099         if (auto_abbrev < len)
2100                 return len;
2101         return auto_abbrev;
2102 }
2103
2104 /*
2105  * How many columns do we need to show line numbers, authors,
2106  * and filenames?
2107  */
2108 static void find_alignment(struct scoreboard *sb, int *option)
2109 {
2110         int longest_src_lines = 0;
2111         int longest_dst_lines = 0;
2112         unsigned largest_score = 0;
2113         struct blame_entry *e;
2114         int compute_auto_abbrev = (abbrev < 0);
2115         int auto_abbrev = default_abbrev;
2116
2117         for (e = sb->ent; e; e = e->next) {
2118                 struct origin *suspect = e->suspect;
2119                 int num;
2120
2121                 if (compute_auto_abbrev)
2122                         auto_abbrev = update_auto_abbrev(auto_abbrev, suspect);
2123                 if (strcmp(suspect->path, sb->path))
2124                         *option |= OUTPUT_SHOW_NAME;
2125                 num = strlen(suspect->path);
2126                 if (longest_file < num)
2127                         longest_file = num;
2128                 if (!(suspect->commit->object.flags & METAINFO_SHOWN)) {
2129                         struct commit_info ci;
2130                         suspect->commit->object.flags |= METAINFO_SHOWN;
2131                         get_commit_info(suspect->commit, &ci, 1);
2132                         if (*option & OUTPUT_SHOW_EMAIL)
2133                                 num = utf8_strwidth(ci.author_mail.buf);
2134                         else
2135                                 num = utf8_strwidth(ci.author.buf);
2136                         if (longest_author < num)
2137                                 longest_author = num;
2138                         commit_info_destroy(&ci);
2139                 }
2140                 num = e->s_lno + e->num_lines;
2141                 if (longest_src_lines < num)
2142                         longest_src_lines = num;
2143                 num = e->lno + e->num_lines;
2144                 if (longest_dst_lines < num)
2145                         longest_dst_lines = num;
2146                 if (largest_score < ent_score(sb, e))
2147                         largest_score = ent_score(sb, e);
2148         }
2149         max_orig_digits = decimal_width(longest_src_lines);
2150         max_digits = decimal_width(longest_dst_lines);
2151         max_score_digits = decimal_width(largest_score);
2152
2153         if (compute_auto_abbrev)
2154                 /* one more abbrev length is needed for the boundary commit */
2155                 abbrev = auto_abbrev + 1;
2156 }
2157
2158 /*
2159  * For debugging -- origin is refcounted, and this asserts that
2160  * we do not underflow.
2161  */
2162 static void sanity_check_refcnt(struct scoreboard *sb)
2163 {
2164         int baa = 0;
2165         struct blame_entry *ent;
2166
2167         for (ent = sb->ent; ent; ent = ent->next) {
2168                 /* Nobody should have zero or negative refcnt */
2169                 if (ent->suspect->refcnt <= 0) {
2170                         fprintf(stderr, "%s in %s has negative refcnt %d\n",
2171                                 ent->suspect->path,
2172                                 oid_to_hex(&ent->suspect->commit->object.oid),
2173                                 ent->suspect->refcnt);
2174                         baa = 1;
2175                 }
2176         }
2177         if (baa) {
2178                 int opt = 0160;
2179                 find_alignment(sb, &opt);
2180                 output(sb, opt);
2181                 die("Baa %d!", baa);
2182         }
2183 }
2184
2185 static unsigned parse_score(const char *arg)
2186 {
2187         char *end;
2188         unsigned long score = strtoul(arg, &end, 10);
2189         if (*end)
2190                 return 0;
2191         return score;
2192 }
2193
2194 static const char *add_prefix(const char *prefix, const char *path)
2195 {
2196         return prefix_path(prefix, prefix ? strlen(prefix) : 0, path);
2197 }
2198
2199 static int git_blame_config(const char *var, const char *value, void *cb)
2200 {
2201         if (!strcmp(var, "blame.showroot")) {
2202                 show_root = git_config_bool(var, value);
2203                 return 0;
2204         }
2205         if (!strcmp(var, "blame.blankboundary")) {
2206                 blank_boundary = git_config_bool(var, value);
2207                 return 0;
2208         }
2209         if (!strcmp(var, "blame.showemail")) {
2210                 int *output_option = cb;
2211                 if (git_config_bool(var, value))
2212                         *output_option |= OUTPUT_SHOW_EMAIL;
2213                 else
2214                         *output_option &= ~OUTPUT_SHOW_EMAIL;
2215                 return 0;
2216         }
2217         if (!strcmp(var, "blame.date")) {
2218                 if (!value)
2219                         return config_error_nonbool(var);
2220                 parse_date_format(value, &blame_date_mode);
2221                 return 0;
2222         }
2223
2224         if (userdiff_config(var, value) < 0)
2225                 return -1;
2226
2227         return git_default_config(var, value, cb);
2228 }
2229
2230 static void verify_working_tree_path(struct commit *work_tree, const char *path)
2231 {
2232         struct commit_list *parents;
2233
2234         for (parents = work_tree->parents; parents; parents = parents->next) {
2235                 const unsigned char *commit_sha1 = parents->item->object.oid.hash;
2236                 unsigned char blob_sha1[20];
2237                 unsigned mode;
2238
2239                 if (!get_tree_entry(commit_sha1, path, blob_sha1, &mode) &&
2240                     sha1_object_info(blob_sha1, NULL) == OBJ_BLOB)
2241                         return;
2242         }
2243         die("no such path '%s' in HEAD", path);
2244 }
2245
2246 static struct commit_list **append_parent(struct commit_list **tail, const unsigned char *sha1)
2247 {
2248         struct commit *parent;
2249
2250         parent = lookup_commit_reference(sha1);
2251         if (!parent)
2252                 die("no such commit %s", sha1_to_hex(sha1));
2253         return &commit_list_insert(parent, tail)->next;
2254 }
2255
2256 static void append_merge_parents(struct commit_list **tail)
2257 {
2258         int merge_head;
2259         struct strbuf line = STRBUF_INIT;
2260
2261         merge_head = open(git_path_merge_head(), O_RDONLY);
2262         if (merge_head < 0) {
2263                 if (errno == ENOENT)
2264                         return;
2265                 die("cannot open '%s' for reading", git_path_merge_head());
2266         }
2267
2268         while (!strbuf_getwholeline_fd(&line, merge_head, '\n')) {
2269                 unsigned char sha1[20];
2270                 if (line.len < 40 || get_sha1_hex(line.buf, sha1))
2271                         die("unknown line in '%s': %s", git_path_merge_head(), line.buf);
2272                 tail = append_parent(tail, sha1);
2273         }
2274         close(merge_head);
2275         strbuf_release(&line);
2276 }
2277
2278 /*
2279  * This isn't as simple as passing sb->buf and sb->len, because we
2280  * want to transfer ownership of the buffer to the commit (so we
2281  * must use detach).
2282  */
2283 static void set_commit_buffer_from_strbuf(struct commit *c, struct strbuf *sb)
2284 {
2285         size_t len;
2286         void *buf = strbuf_detach(sb, &len);
2287         set_commit_buffer(c, buf, len);
2288 }
2289
2290 /*
2291  * Prepare a dummy commit that represents the work tree (or staged) item.
2292  * Note that annotating work tree item never works in the reverse.
2293  */
2294 static struct commit *fake_working_tree_commit(struct diff_options *opt,
2295                                                const char *path,
2296                                                const char *contents_from)
2297 {
2298         struct commit *commit;
2299         struct origin *origin;
2300         struct commit_list **parent_tail, *parent;
2301         unsigned char head_sha1[20];
2302         struct strbuf buf = STRBUF_INIT;
2303         const char *ident;
2304         time_t now;
2305         int size, len;
2306         struct cache_entry *ce;
2307         unsigned mode;
2308         struct strbuf msg = STRBUF_INIT;
2309
2310         time(&now);
2311         commit = alloc_commit_node();
2312         commit->object.parsed = 1;
2313         commit->date = now;
2314         parent_tail = &commit->parents;
2315
2316         if (!resolve_ref_unsafe("HEAD", RESOLVE_REF_READING, head_sha1, NULL))
2317                 die("no such ref: HEAD");
2318
2319         parent_tail = append_parent(parent_tail, head_sha1);
2320         append_merge_parents(parent_tail);
2321         verify_working_tree_path(commit, path);
2322
2323         origin = make_origin(commit, path);
2324
2325         ident = fmt_ident("Not Committed Yet", "not.committed.yet", NULL, 0);
2326         strbuf_addstr(&msg, "tree 0000000000000000000000000000000000000000\n");
2327         for (parent = commit->parents; parent; parent = parent->next)
2328                 strbuf_addf(&msg, "parent %s\n",
2329                             oid_to_hex(&parent->item->object.oid));
2330         strbuf_addf(&msg,
2331                     "author %s\n"
2332                     "committer %s\n\n"
2333                     "Version of %s from %s\n",
2334                     ident, ident, path,
2335                     (!contents_from ? path :
2336                      (!strcmp(contents_from, "-") ? "standard input" : contents_from)));
2337         set_commit_buffer_from_strbuf(commit, &msg);
2338
2339         if (!contents_from || strcmp("-", contents_from)) {
2340                 struct stat st;
2341                 const char *read_from;
2342                 char *buf_ptr;
2343                 unsigned long buf_len;
2344
2345                 if (contents_from) {
2346                         if (stat(contents_from, &st) < 0)
2347                                 die_errno("Cannot stat '%s'", contents_from);
2348                         read_from = contents_from;
2349                 }
2350                 else {
2351                         if (lstat(path, &st) < 0)
2352                                 die_errno("Cannot lstat '%s'", path);
2353                         read_from = path;
2354                 }
2355                 mode = canon_mode(st.st_mode);
2356
2357                 switch (st.st_mode & S_IFMT) {
2358                 case S_IFREG:
2359                         if (DIFF_OPT_TST(opt, ALLOW_TEXTCONV) &&
2360                             textconv_object(read_from, mode, null_sha1, 0, &buf_ptr, &buf_len))
2361                                 strbuf_attach(&buf, buf_ptr, buf_len, buf_len + 1);
2362                         else if (strbuf_read_file(&buf, read_from, st.st_size) != st.st_size)
2363                                 die_errno("cannot open or read '%s'", read_from);
2364                         break;
2365                 case S_IFLNK:
2366                         if (strbuf_readlink(&buf, read_from, st.st_size) < 0)
2367                                 die_errno("cannot readlink '%s'", read_from);
2368                         break;
2369                 default:
2370                         die("unsupported file type %s", read_from);
2371                 }
2372         }
2373         else {
2374                 /* Reading from stdin */
2375                 mode = 0;
2376                 if (strbuf_read(&buf, 0, 0) < 0)
2377                         die_errno("failed to read from stdin");
2378         }
2379         convert_to_git(path, buf.buf, buf.len, &buf, 0);
2380         origin->file.ptr = buf.buf;
2381         origin->file.size = buf.len;
2382         pretend_sha1_file(buf.buf, buf.len, OBJ_BLOB, origin->blob_sha1);
2383
2384         /*
2385          * Read the current index, replace the path entry with
2386          * origin->blob_sha1 without mucking with its mode or type
2387          * bits; we are not going to write this index out -- we just
2388          * want to run "diff-index --cached".
2389          */
2390         discard_cache();
2391         read_cache();
2392
2393         len = strlen(path);
2394         if (!mode) {
2395                 int pos = cache_name_pos(path, len);
2396                 if (0 <= pos)
2397                         mode = active_cache[pos]->ce_mode;
2398                 else
2399                         /* Let's not bother reading from HEAD tree */
2400                         mode = S_IFREG | 0644;
2401         }
2402         size = cache_entry_size(len);
2403         ce = xcalloc(1, size);
2404         hashcpy(ce->sha1, origin->blob_sha1);
2405         memcpy(ce->name, path, len);
2406         ce->ce_flags = create_ce_flags(0);
2407         ce->ce_namelen = len;
2408         ce->ce_mode = create_ce_mode(mode);
2409         add_cache_entry(ce, ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE);
2410
2411         cache_tree_invalidate_path(&the_index, path);
2412
2413         return commit;
2414 }
2415
2416 static struct commit *find_single_final(struct rev_info *revs,
2417                                         const char **name_p)
2418 {
2419         int i;
2420         struct commit *found = NULL;
2421         const char *name = NULL;
2422
2423         for (i = 0; i < revs->pending.nr; i++) {
2424                 struct object *obj = revs->pending.objects[i].item;
2425                 if (obj->flags & UNINTERESTING)
2426                         continue;
2427                 while (obj->type == OBJ_TAG)
2428                         obj = deref_tag(obj, NULL, 0);
2429                 if (obj->type != OBJ_COMMIT)
2430                         die("Non commit %s?", revs->pending.objects[i].name);
2431                 if (found)
2432                         die("More than one commit to dig from %s and %s?",
2433                             revs->pending.objects[i].name, name);
2434                 found = (struct commit *)obj;
2435                 name = revs->pending.objects[i].name;
2436         }
2437         if (name_p)
2438                 *name_p = name;
2439         return found;
2440 }
2441
2442 static char *prepare_final(struct scoreboard *sb)
2443 {
2444         const char *name;
2445         sb->final = find_single_final(sb->revs, &name);
2446         return xstrdup_or_null(name);
2447 }
2448
2449 static char *prepare_initial(struct scoreboard *sb)
2450 {
2451         int i;
2452         const char *final_commit_name = NULL;
2453         struct rev_info *revs = sb->revs;
2454
2455         /*
2456          * There must be one and only one negative commit, and it must be
2457          * the boundary.
2458          */
2459         for (i = 0; i < revs->pending.nr; i++) {
2460                 struct object *obj = revs->pending.objects[i].item;
2461                 if (!(obj->flags & UNINTERESTING))
2462                         continue;
2463                 while (obj->type == OBJ_TAG)
2464                         obj = deref_tag(obj, NULL, 0);
2465                 if (obj->type != OBJ_COMMIT)
2466                         die("Non commit %s?", revs->pending.objects[i].name);
2467                 if (sb->final)
2468                         die("More than one commit to dig down to %s and %s?",
2469                             revs->pending.objects[i].name,
2470                             final_commit_name);
2471                 sb->final = (struct commit *) obj;
2472                 final_commit_name = revs->pending.objects[i].name;
2473         }
2474         if (!final_commit_name)
2475                 die("No commit to dig down to?");
2476         return xstrdup(final_commit_name);
2477 }
2478
2479 static int blame_copy_callback(const struct option *option, const char *arg, int unset)
2480 {
2481         int *opt = option->value;
2482
2483         /*
2484          * -C enables copy from removed files;
2485          * -C -C enables copy from existing files, but only
2486          *       when blaming a new file;
2487          * -C -C -C enables copy from existing files for
2488          *          everybody
2489          */
2490         if (*opt & PICKAXE_BLAME_COPY_HARDER)
2491                 *opt |= PICKAXE_BLAME_COPY_HARDEST;
2492         if (*opt & PICKAXE_BLAME_COPY)
2493                 *opt |= PICKAXE_BLAME_COPY_HARDER;
2494         *opt |= PICKAXE_BLAME_COPY | PICKAXE_BLAME_MOVE;
2495
2496         if (arg)
2497                 blame_copy_score = parse_score(arg);
2498         return 0;
2499 }
2500
2501 static int blame_move_callback(const struct option *option, const char *arg, int unset)
2502 {
2503         int *opt = option->value;
2504
2505         *opt |= PICKAXE_BLAME_MOVE;
2506
2507         if (arg)
2508                 blame_move_score = parse_score(arg);
2509         return 0;
2510 }
2511
2512 int cmd_blame(int argc, const char **argv, const char *prefix)
2513 {
2514         struct rev_info revs;
2515         const char *path;
2516         struct scoreboard sb;
2517         struct origin *o;
2518         struct blame_entry *ent = NULL;
2519         long dashdash_pos, lno;
2520         char *final_commit_name = NULL;
2521         enum object_type type;
2522         struct commit *final_commit = NULL;
2523
2524         static struct string_list range_list;
2525         static int output_option = 0, opt = 0;
2526         static int show_stats = 0;
2527         static const char *revs_file = NULL;
2528         static const char *contents_from = NULL;
2529         static const struct option options[] = {
2530                 OPT_BOOL(0, "incremental", &incremental, N_("Show blame entries as we find them, incrementally")),
2531                 OPT_BOOL('b', NULL, &blank_boundary, N_("Show blank SHA-1 for boundary commits (Default: off)")),
2532                 OPT_BOOL(0, "root", &show_root, N_("Do not treat root commits as boundaries (Default: off)")),
2533                 OPT_BOOL(0, "show-stats", &show_stats, N_("Show work cost statistics")),
2534                 OPT_BOOL(0, "progress", &show_progress, N_("Force progress reporting")),
2535                 OPT_BIT(0, "score-debug", &output_option, N_("Show output score for blame entries"), OUTPUT_SHOW_SCORE),
2536                 OPT_BIT('f', "show-name", &output_option, N_("Show original filename (Default: auto)"), OUTPUT_SHOW_NAME),
2537                 OPT_BIT('n', "show-number", &output_option, N_("Show original linenumber (Default: off)"), OUTPUT_SHOW_NUMBER),
2538                 OPT_BIT('p', "porcelain", &output_option, N_("Show in a format designed for machine consumption"), OUTPUT_PORCELAIN),
2539                 OPT_BIT(0, "line-porcelain", &output_option, N_("Show porcelain format with per-line commit information"), OUTPUT_PORCELAIN|OUTPUT_LINE_PORCELAIN),
2540                 OPT_BIT('c', NULL, &output_option, N_("Use the same output mode as git-annotate (Default: off)"), OUTPUT_ANNOTATE_COMPAT),
2541                 OPT_BIT('t', NULL, &output_option, N_("Show raw timestamp (Default: off)"), OUTPUT_RAW_TIMESTAMP),
2542                 OPT_BIT('l', NULL, &output_option, N_("Show long commit SHA1 (Default: off)"), OUTPUT_LONG_OBJECT_NAME),
2543                 OPT_BIT('s', NULL, &output_option, N_("Suppress author name and timestamp (Default: off)"), OUTPUT_NO_AUTHOR),
2544                 OPT_BIT('e', "show-email", &output_option, N_("Show author email instead of name (Default: off)"), OUTPUT_SHOW_EMAIL),
2545                 OPT_BIT('w', NULL, &xdl_opts, N_("Ignore whitespace differences"), XDF_IGNORE_WHITESPACE),
2546                 OPT_BIT(0, "minimal", &xdl_opts, N_("Spend extra cycles to find better match"), XDF_NEED_MINIMAL),
2547                 OPT_STRING('S', NULL, &revs_file, N_("file"), N_("Use revisions from <file> instead of calling git-rev-list")),
2548                 OPT_STRING(0, "contents", &contents_from, N_("file"), N_("Use <file>'s contents as the final image")),
2549                 { OPTION_CALLBACK, 'C', NULL, &opt, N_("score"), N_("Find line copies within and across files"), PARSE_OPT_OPTARG, blame_copy_callback },
2550                 { OPTION_CALLBACK, 'M', NULL, &opt, N_("score"), N_("Find line movements within and across files"), PARSE_OPT_OPTARG, blame_move_callback },
2551                 OPT_STRING_LIST('L', NULL, &range_list, N_("n,m"), N_("Process only line range n,m, counting from 1")),
2552                 OPT__ABBREV(&abbrev),
2553                 OPT_END()
2554         };
2555
2556         struct parse_opt_ctx_t ctx;
2557         int cmd_is_annotate = !strcmp(argv[0], "annotate");
2558         struct range_set ranges;
2559         unsigned int range_i;
2560         long anchor;
2561
2562         git_config(git_blame_config, &output_option);
2563         init_revisions(&revs, NULL);
2564         revs.date_mode = blame_date_mode;
2565         DIFF_OPT_SET(&revs.diffopt, ALLOW_TEXTCONV);
2566         DIFF_OPT_SET(&revs.diffopt, FOLLOW_RENAMES);
2567
2568         save_commit_buffer = 0;
2569         dashdash_pos = 0;
2570         show_progress = -1;
2571
2572         parse_options_start(&ctx, argc, argv, prefix, options,
2573                             PARSE_OPT_KEEP_DASHDASH | PARSE_OPT_KEEP_ARGV0);
2574         for (;;) {
2575                 switch (parse_options_step(&ctx, options, blame_opt_usage)) {
2576                 case PARSE_OPT_HELP:
2577                         exit(129);
2578                 case PARSE_OPT_DONE:
2579                         if (ctx.argv[0])
2580                                 dashdash_pos = ctx.cpidx;
2581                         goto parse_done;
2582                 }
2583
2584                 if (!strcmp(ctx.argv[0], "--reverse")) {
2585                         ctx.argv[0] = "--children";
2586                         reverse = 1;
2587                 }
2588                 parse_revision_opt(&revs, &ctx, options, blame_opt_usage);
2589         }
2590 parse_done:
2591         no_whole_file_rename = !DIFF_OPT_TST(&revs.diffopt, FOLLOW_RENAMES);
2592         DIFF_OPT_CLR(&revs.diffopt, FOLLOW_RENAMES);
2593         argc = parse_options_end(&ctx);
2594
2595         if (incremental || (output_option & OUTPUT_PORCELAIN)) {
2596                 if (show_progress > 0)
2597                         die("--progress can't be used with --incremental or porcelain formats");
2598                 show_progress = 0;
2599         } else if (show_progress < 0)
2600                 show_progress = isatty(2);
2601
2602         if (0 < abbrev)
2603                 /* one more abbrev length is needed for the boundary commit */
2604                 abbrev++;
2605
2606         if (revs_file && read_ancestry(revs_file))
2607                 die_errno("reading graft file '%s' failed", revs_file);
2608
2609         if (cmd_is_annotate) {
2610                 output_option |= OUTPUT_ANNOTATE_COMPAT;
2611                 blame_date_mode.type = DATE_ISO8601;
2612         } else {
2613                 blame_date_mode = revs.date_mode;
2614         }
2615
2616         /* The maximum width used to show the dates */
2617         switch (blame_date_mode.type) {
2618         case DATE_RFC2822:
2619                 blame_date_width = sizeof("Thu, 19 Oct 2006 16:00:04 -0700");
2620                 break;
2621         case DATE_ISO8601_STRICT:
2622                 blame_date_width = sizeof("2006-10-19T16:00:04-07:00");
2623                 break;
2624         case DATE_ISO8601:
2625                 blame_date_width = sizeof("2006-10-19 16:00:04 -0700");
2626                 break;
2627         case DATE_RAW:
2628                 blame_date_width = sizeof("1161298804 -0700");
2629                 break;
2630         case DATE_SHORT:
2631                 blame_date_width = sizeof("2006-10-19");
2632                 break;
2633         case DATE_RELATIVE:
2634                 /* TRANSLATORS: This string is used to tell us the maximum
2635                    display width for a relative timestamp in "git blame"
2636                    output.  For C locale, "4 years, 11 months ago", which
2637                    takes 22 places, is the longest among various forms of
2638                    relative timestamps, but your language may need more or
2639                    fewer display columns. */
2640                 blame_date_width = utf8_strwidth(_("4 years, 11 months ago")) + 1; /* add the null */
2641                 break;
2642         case DATE_NORMAL:
2643                 blame_date_width = sizeof("Thu Oct 19 16:00:04 2006 -0700");
2644                 break;
2645         case DATE_STRFTIME:
2646                 blame_date_width = strlen(show_date(0, 0, &blame_date_mode)) + 1; /* add the null */
2647                 break;
2648         }
2649         blame_date_width -= 1; /* strip the null */
2650
2651         if (DIFF_OPT_TST(&revs.diffopt, FIND_COPIES_HARDER))
2652                 opt |= (PICKAXE_BLAME_COPY | PICKAXE_BLAME_MOVE |
2653                         PICKAXE_BLAME_COPY_HARDER);
2654
2655         if (!blame_move_score)
2656                 blame_move_score = BLAME_DEFAULT_MOVE_SCORE;
2657         if (!blame_copy_score)
2658                 blame_copy_score = BLAME_DEFAULT_COPY_SCORE;
2659
2660         /*
2661          * We have collected options unknown to us in argv[1..unk]
2662          * which are to be passed to revision machinery if we are
2663          * going to do the "bottom" processing.
2664          *
2665          * The remaining are:
2666          *
2667          * (1) if dashdash_pos != 0, it is either
2668          *     "blame [revisions] -- <path>" or
2669          *     "blame -- <path> <rev>"
2670          *
2671          * (2) otherwise, it is one of the two:
2672          *     "blame [revisions] <path>"
2673          *     "blame <path> <rev>"
2674          *
2675          * Note that we must strip out <path> from the arguments: we do not
2676          * want the path pruning but we may want "bottom" processing.
2677          */
2678         if (dashdash_pos) {
2679                 switch (argc - dashdash_pos - 1) {
2680                 case 2: /* (1b) */
2681                         if (argc != 4)
2682                                 usage_with_options(blame_opt_usage, options);
2683                         /* reorder for the new way: <rev> -- <path> */
2684                         argv[1] = argv[3];
2685                         argv[3] = argv[2];
2686                         argv[2] = "--";
2687                         /* FALLTHROUGH */
2688                 case 1: /* (1a) */
2689                         path = add_prefix(prefix, argv[--argc]);
2690                         argv[argc] = NULL;
2691                         break;
2692                 default:
2693                         usage_with_options(blame_opt_usage, options);
2694                 }
2695         } else {
2696                 if (argc < 2)
2697                         usage_with_options(blame_opt_usage, options);
2698                 path = add_prefix(prefix, argv[argc - 1]);
2699                 if (argc == 3 && !file_exists(path)) { /* (2b) */
2700                         path = add_prefix(prefix, argv[1]);
2701                         argv[1] = argv[2];
2702                 }
2703                 argv[argc - 1] = "--";
2704
2705                 setup_work_tree();
2706                 if (!file_exists(path))
2707                         die_errno("cannot stat path '%s'", path);
2708         }
2709
2710         revs.disable_stdin = 1;
2711         setup_revisions(argc, argv, &revs, NULL);
2712         memset(&sb, 0, sizeof(sb));
2713
2714         sb.revs = &revs;
2715         if (!reverse) {
2716                 final_commit_name = prepare_final(&sb);
2717                 sb.commits.compare = compare_commits_by_commit_date;
2718         }
2719         else if (contents_from)
2720                 die("--contents and --reverse do not blend well.");
2721         else {
2722                 final_commit_name = prepare_initial(&sb);
2723                 sb.commits.compare = compare_commits_by_reverse_commit_date;
2724                 if (revs.first_parent_only)
2725                         revs.children.name = NULL;
2726         }
2727
2728         if (!sb.final) {
2729                 /*
2730                  * "--not A B -- path" without anything positive;
2731                  * do not default to HEAD, but use the working tree
2732                  * or "--contents".
2733                  */
2734                 setup_work_tree();
2735                 sb.final = fake_working_tree_commit(&sb.revs->diffopt,
2736                                                     path, contents_from);
2737                 add_pending_object(&revs, &(sb.final->object), ":");
2738         }
2739         else if (contents_from)
2740                 die("Cannot use --contents with final commit object name");
2741
2742         if (reverse && revs.first_parent_only) {
2743                 final_commit = find_single_final(sb.revs, NULL);
2744                 if (!final_commit)
2745                         die("--reverse and --first-parent together require specified latest commit");
2746         }
2747
2748         /*
2749          * If we have bottom, this will mark the ancestors of the
2750          * bottom commits we would reach while traversing as
2751          * uninteresting.
2752          */
2753         if (prepare_revision_walk(&revs))
2754                 die(_("revision walk setup failed"));
2755
2756         if (reverse && revs.first_parent_only) {
2757                 struct commit *c = final_commit;
2758
2759                 sb.revs->children.name = "children";
2760                 while (c->parents &&
2761                        oidcmp(&c->object.oid, &sb.final->object.oid)) {
2762                         struct commit_list *l = xcalloc(1, sizeof(*l));
2763
2764                         l->item = c;
2765                         if (add_decoration(&sb.revs->children,
2766                                            &c->parents->item->object, l))
2767                                 die("BUG: not unique item in first-parent chain");
2768                         c = c->parents->item;
2769                 }
2770
2771                 if (oidcmp(&c->object.oid, &sb.final->object.oid))
2772                         die("--reverse --first-parent together require range along first-parent chain");
2773         }
2774
2775         if (is_null_oid(&sb.final->object.oid)) {
2776                 o = sb.final->util;
2777                 sb.final_buf = xmemdupz(o->file.ptr, o->file.size);
2778                 sb.final_buf_size = o->file.size;
2779         }
2780         else {
2781                 o = get_origin(&sb, sb.final, path);
2782                 if (fill_blob_sha1_and_mode(o))
2783                         die("no such path %s in %s", path, final_commit_name);
2784
2785                 if (DIFF_OPT_TST(&sb.revs->diffopt, ALLOW_TEXTCONV) &&
2786                     textconv_object(path, o->mode, o->blob_sha1, 1, (char **) &sb.final_buf,
2787                                     &sb.final_buf_size))
2788                         ;
2789                 else
2790                         sb.final_buf = read_sha1_file(o->blob_sha1, &type,
2791                                                       &sb.final_buf_size);
2792
2793                 if (!sb.final_buf)
2794                         die("Cannot read blob %s for path %s",
2795                             sha1_to_hex(o->blob_sha1),
2796                             path);
2797         }
2798         num_read_blob++;
2799         lno = prepare_lines(&sb);
2800
2801         if (lno && !range_list.nr)
2802                 string_list_append(&range_list, xstrdup("1"));
2803
2804         anchor = 1;
2805         range_set_init(&ranges, range_list.nr);
2806         for (range_i = 0; range_i < range_list.nr; ++range_i) {
2807                 long bottom, top;
2808                 if (parse_range_arg(range_list.items[range_i].string,
2809                                     nth_line_cb, &sb, lno, anchor,
2810                                     &bottom, &top, sb.path))
2811                         usage(blame_usage);
2812                 if (lno < top || ((lno || bottom) && lno < bottom))
2813                         die("file %s has only %lu lines", path, lno);
2814                 if (bottom < 1)
2815                         bottom = 1;
2816                 if (top < 1)
2817                         top = lno;
2818                 bottom--;
2819                 range_set_append_unsafe(&ranges, bottom, top);
2820                 anchor = top + 1;
2821         }
2822         sort_and_merge_range_set(&ranges);
2823
2824         for (range_i = ranges.nr; range_i > 0; --range_i) {
2825                 const struct range *r = &ranges.ranges[range_i - 1];
2826                 long bottom = r->start;
2827                 long top = r->end;
2828                 struct blame_entry *next = ent;
2829                 ent = xcalloc(1, sizeof(*ent));
2830                 ent->lno = bottom;
2831                 ent->num_lines = top - bottom;
2832                 ent->suspect = o;
2833                 ent->s_lno = bottom;
2834                 ent->next = next;
2835                 origin_incref(o);
2836         }
2837
2838         o->suspects = ent;
2839         prio_queue_put(&sb.commits, o->commit);
2840
2841         origin_decref(o);
2842
2843         range_set_release(&ranges);
2844         string_list_clear(&range_list, 0);
2845
2846         sb.ent = NULL;
2847         sb.path = path;
2848
2849         read_mailmap(&mailmap, NULL);
2850
2851         assign_blame(&sb, opt);
2852
2853         if (!incremental)
2854                 setup_pager();
2855
2856         free(final_commit_name);
2857
2858         if (incremental)
2859                 return 0;
2860
2861         sb.ent = blame_sort(sb.ent, compare_blame_final);
2862
2863         coalesce(&sb);
2864
2865         if (!(output_option & OUTPUT_PORCELAIN))
2866                 find_alignment(&sb, &output_option);
2867
2868         output(&sb, output_option);
2869         free((void *)sb.final_buf);
2870         for (ent = sb.ent; ent; ) {
2871                 struct blame_entry *e = ent->next;
2872                 free(ent);
2873                 ent = e;
2874         }
2875
2876         if (show_stats) {
2877                 printf("num read blob: %d\n", num_read_blob);
2878                 printf("num get patch: %d\n", num_get_patch);
2879                 printf("num commits: %d\n", num_commits);
2880         }
2881         return 0;
2882 }