travis-ci: add static analysis build job to run coccicheck
[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 = STRING_LIST_INIT_NODUP;
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         struct object_id blob_oid;
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,
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.hunk_func = hunk_func;
146         ecb.priv = cb_data;
147         return xdi_diff(file_a, file_b, &xpp, &xecfg, &ecb);
148 }
149
150 /*
151  * Prepare diff_filespec and convert it using diff textconv API
152  * if the textconv driver exists.
153  * Return 1 if the conversion succeeds, 0 otherwise.
154  */
155 int textconv_object(const char *path,
156                     unsigned mode,
157                     const struct object_id *oid,
158                     int oid_valid,
159                     char **buf,
160                     unsigned long *buf_size)
161 {
162         struct diff_filespec *df;
163         struct userdiff_driver *textconv;
164
165         df = alloc_filespec(path);
166         fill_filespec(df, oid->hash, oid_valid, mode);
167         textconv = get_textconv(df);
168         if (!textconv) {
169                 free_filespec(df);
170                 return 0;
171         }
172
173         *buf_size = fill_textconv(textconv, df, buf);
174         free_filespec(df);
175         return 1;
176 }
177
178 /*
179  * Given an origin, prepare mmfile_t structure to be used by the
180  * diff machinery
181  */
182 static void fill_origin_blob(struct diff_options *opt,
183                              struct origin *o, mmfile_t *file)
184 {
185         if (!o->file.ptr) {
186                 enum object_type type;
187                 unsigned long file_size;
188
189                 num_read_blob++;
190                 if (DIFF_OPT_TST(opt, ALLOW_TEXTCONV) &&
191                     textconv_object(o->path, o->mode, &o->blob_oid, 1, &file->ptr, &file_size))
192                         ;
193                 else
194                         file->ptr = read_sha1_file(o->blob_oid.hash, &type,
195                                                    &file_size);
196                 file->size = file_size;
197
198                 if (!file->ptr)
199                         die("Cannot read blob %s for path %s",
200                             oid_to_hex(&o->blob_oid),
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_oid(&origin->blob_oid))
513                 return 0;
514         if (get_tree_entry(origin->commit->object.oid.hash,
515                            origin->path,
516                            origin->blob_oid.hash, &origin->mode))
517                 goto error_out;
518         if (sha1_object_info(origin->blob_oid.hash, NULL) != OBJ_BLOB)
519                 goto error_out;
520         return 0;
521  error_out:
522         oidclr(&origin->blob_oid);
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                 oidcpy(&porigin->blob_oid, &origin->blob_oid);
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                         oidcpy(&porigin->blob_oid, &p->one->oid);
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         clear_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                         oidcpy(&porigin->blob_oid, &p->one->oid);
649                         porigin->mode = p->one->mode;
650                         break;
651                 }
652         }
653         diff_flush(&diff_opts);
654         clear_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,
662                             const struct blame_entry *src)
663 {
664         struct blame_entry *e = xmalloc(sizeof(*e));
665         memcpy(e, src, sizeof(*e));
666         origin_incref(e->suspect);
667
668         e->next = **queue;
669         **queue = e;
670         *queue = &e->next;
671 }
672
673 /*
674  * src typically is on-stack; we want to copy the information in it to
675  * a malloced blame_entry that gets added to the given queue.  The
676  * origin of dst loses a refcnt.
677  */
678 static void dup_entry(struct blame_entry ***queue,
679                       struct blame_entry *dst, struct blame_entry *src)
680 {
681         origin_incref(src->suspect);
682         origin_decref(dst->suspect);
683         memcpy(dst, src, sizeof(*src));
684         dst->next = **queue;
685         **queue = dst;
686         *queue = &dst->next;
687 }
688
689 static const char *nth_line(struct scoreboard *sb, long lno)
690 {
691         return sb->final_buf + sb->lineno[lno];
692 }
693
694 static const char *nth_line_cb(void *data, long lno)
695 {
696         return nth_line((struct scoreboard *)data, lno);
697 }
698
699 /*
700  * It is known that lines between tlno to same came from parent, and e
701  * has an overlap with that range.  it also is known that parent's
702  * line plno corresponds to e's line tlno.
703  *
704  *                <---- e ----->
705  *                   <------>
706  *                   <------------>
707  *             <------------>
708  *             <------------------>
709  *
710  * Split e into potentially three parts; before this chunk, the chunk
711  * to be blamed for the parent, and after that portion.
712  */
713 static void split_overlap(struct blame_entry *split,
714                           struct blame_entry *e,
715                           int tlno, int plno, int same,
716                           struct origin *parent)
717 {
718         int chunk_end_lno;
719         memset(split, 0, sizeof(struct blame_entry [3]));
720
721         if (e->s_lno < tlno) {
722                 /* there is a pre-chunk part not blamed on parent */
723                 split[0].suspect = origin_incref(e->suspect);
724                 split[0].lno = e->lno;
725                 split[0].s_lno = e->s_lno;
726                 split[0].num_lines = tlno - e->s_lno;
727                 split[1].lno = e->lno + tlno - e->s_lno;
728                 split[1].s_lno = plno;
729         }
730         else {
731                 split[1].lno = e->lno;
732                 split[1].s_lno = plno + (e->s_lno - tlno);
733         }
734
735         if (same < e->s_lno + e->num_lines) {
736                 /* there is a post-chunk part not blamed on parent */
737                 split[2].suspect = origin_incref(e->suspect);
738                 split[2].lno = e->lno + (same - e->s_lno);
739                 split[2].s_lno = e->s_lno + (same - e->s_lno);
740                 split[2].num_lines = e->s_lno + e->num_lines - same;
741                 chunk_end_lno = split[2].lno;
742         }
743         else
744                 chunk_end_lno = e->lno + e->num_lines;
745         split[1].num_lines = chunk_end_lno - split[1].lno;
746
747         /*
748          * if it turns out there is nothing to blame the parent for,
749          * forget about the splitting.  !split[1].suspect signals this.
750          */
751         if (split[1].num_lines < 1)
752                 return;
753         split[1].suspect = origin_incref(parent);
754 }
755
756 /*
757  * split_overlap() divided an existing blame e into up to three parts
758  * in split.  Any assigned blame is moved to queue to
759  * reflect the split.
760  */
761 static void split_blame(struct blame_entry ***blamed,
762                         struct blame_entry ***unblamed,
763                         struct blame_entry *split,
764                         struct blame_entry *e)
765 {
766         if (split[0].suspect && split[2].suspect) {
767                 /* The first part (reuse storage for the existing entry e) */
768                 dup_entry(unblamed, e, &split[0]);
769
770                 /* The last part -- me */
771                 add_blame_entry(unblamed, &split[2]);
772
773                 /* ... and the middle part -- parent */
774                 add_blame_entry(blamed, &split[1]);
775         }
776         else if (!split[0].suspect && !split[2].suspect)
777                 /*
778                  * The parent covers the entire area; reuse storage for
779                  * e and replace it with the parent.
780                  */
781                 dup_entry(blamed, e, &split[1]);
782         else if (split[0].suspect) {
783                 /* me and then parent */
784                 dup_entry(unblamed, e, &split[0]);
785                 add_blame_entry(blamed, &split[1]);
786         }
787         else {
788                 /* parent and then me */
789                 dup_entry(blamed, e, &split[1]);
790                 add_blame_entry(unblamed, &split[2]);
791         }
792 }
793
794 /*
795  * After splitting the blame, the origins used by the
796  * on-stack blame_entry should lose one refcnt each.
797  */
798 static void decref_split(struct blame_entry *split)
799 {
800         int i;
801
802         for (i = 0; i < 3; i++)
803                 origin_decref(split[i].suspect);
804 }
805
806 /*
807  * reverse_blame reverses the list given in head, appending tail.
808  * That allows us to build lists in reverse order, then reverse them
809  * afterwards.  This can be faster than building the list in proper
810  * order right away.  The reason is that building in proper order
811  * requires writing a link in the _previous_ element, while building
812  * in reverse order just requires placing the list head into the
813  * _current_ element.
814  */
815
816 static struct blame_entry *reverse_blame(struct blame_entry *head,
817                                          struct blame_entry *tail)
818 {
819         while (head) {
820                 struct blame_entry *next = head->next;
821                 head->next = tail;
822                 tail = head;
823                 head = next;
824         }
825         return tail;
826 }
827
828 /*
829  * Process one hunk from the patch between the current suspect for
830  * blame_entry e and its parent.  This first blames any unfinished
831  * entries before the chunk (which is where target and parent start
832  * differing) on the parent, and then splits blame entries at the
833  * start and at the end of the difference region.  Since use of -M and
834  * -C options may lead to overlapping/duplicate source line number
835  * ranges, all we can rely on from sorting/merging is the order of the
836  * first suspect line number.
837  */
838 static void blame_chunk(struct blame_entry ***dstq, struct blame_entry ***srcq,
839                         int tlno, int offset, int same,
840                         struct origin *parent)
841 {
842         struct blame_entry *e = **srcq;
843         struct blame_entry *samep = NULL, *diffp = NULL;
844
845         while (e && e->s_lno < tlno) {
846                 struct blame_entry *next = e->next;
847                 /*
848                  * current record starts before differing portion.  If
849                  * it reaches into it, we need to split it up and
850                  * examine the second part separately.
851                  */
852                 if (e->s_lno + e->num_lines > tlno) {
853                         /* Move second half to a new record */
854                         int len = tlno - e->s_lno;
855                         struct blame_entry *n = xcalloc(1, sizeof (struct blame_entry));
856                         n->suspect = e->suspect;
857                         n->lno = e->lno + len;
858                         n->s_lno = e->s_lno + len;
859                         n->num_lines = e->num_lines - len;
860                         e->num_lines = len;
861                         e->score = 0;
862                         /* Push new record to diffp */
863                         n->next = diffp;
864                         diffp = n;
865                 } else
866                         origin_decref(e->suspect);
867                 /* Pass blame for everything before the differing
868                  * chunk to the parent */
869                 e->suspect = origin_incref(parent);
870                 e->s_lno += offset;
871                 e->next = samep;
872                 samep = e;
873                 e = next;
874         }
875         /*
876          * As we don't know how much of a common stretch after this
877          * diff will occur, the currently blamed parts are all that we
878          * can assign to the parent for now.
879          */
880
881         if (samep) {
882                 **dstq = reverse_blame(samep, **dstq);
883                 *dstq = &samep->next;
884         }
885         /*
886          * Prepend the split off portions: everything after e starts
887          * after the blameable portion.
888          */
889         e = reverse_blame(diffp, e);
890
891         /*
892          * Now retain records on the target while parts are different
893          * from the parent.
894          */
895         samep = NULL;
896         diffp = NULL;
897         while (e && e->s_lno < same) {
898                 struct blame_entry *next = e->next;
899
900                 /*
901                  * If current record extends into sameness, need to split.
902                  */
903                 if (e->s_lno + e->num_lines > same) {
904                         /*
905                          * Move second half to a new record to be
906                          * processed by later chunks
907                          */
908                         int len = same - e->s_lno;
909                         struct blame_entry *n = xcalloc(1, sizeof (struct blame_entry));
910                         n->suspect = origin_incref(e->suspect);
911                         n->lno = e->lno + len;
912                         n->s_lno = e->s_lno + len;
913                         n->num_lines = e->num_lines - len;
914                         e->num_lines = len;
915                         e->score = 0;
916                         /* Push new record to samep */
917                         n->next = samep;
918                         samep = n;
919                 }
920                 e->next = diffp;
921                 diffp = e;
922                 e = next;
923         }
924         **srcq = reverse_blame(diffp, reverse_blame(samep, e));
925         /* Move across elements that are in the unblamable portion */
926         if (diffp)
927                 *srcq = &diffp->next;
928 }
929
930 struct blame_chunk_cb_data {
931         struct origin *parent;
932         long offset;
933         struct blame_entry **dstq;
934         struct blame_entry **srcq;
935 };
936
937 /* diff chunks are from parent to target */
938 static int blame_chunk_cb(long start_a, long count_a,
939                           long start_b, long count_b, void *data)
940 {
941         struct blame_chunk_cb_data *d = data;
942         if (start_a - start_b != d->offset)
943                 die("internal error in blame::blame_chunk_cb");
944         blame_chunk(&d->dstq, &d->srcq, start_b, start_a - start_b,
945                     start_b + count_b, d->parent);
946         d->offset = start_a + count_a - (start_b + count_b);
947         return 0;
948 }
949
950 /*
951  * We are looking at the origin 'target' and aiming to pass blame
952  * for the lines it is suspected to its parent.  Run diff to find
953  * which lines came from parent and pass blame for them.
954  */
955 static void pass_blame_to_parent(struct scoreboard *sb,
956                                  struct origin *target,
957                                  struct origin *parent)
958 {
959         mmfile_t file_p, file_o;
960         struct blame_chunk_cb_data d;
961         struct blame_entry *newdest = NULL;
962
963         if (!target->suspects)
964                 return; /* nothing remains for this target */
965
966         d.parent = parent;
967         d.offset = 0;
968         d.dstq = &newdest; d.srcq = &target->suspects;
969
970         fill_origin_blob(&sb->revs->diffopt, parent, &file_p);
971         fill_origin_blob(&sb->revs->diffopt, target, &file_o);
972         num_get_patch++;
973
974         if (diff_hunks(&file_p, &file_o, blame_chunk_cb, &d))
975                 die("unable to generate diff (%s -> %s)",
976                     oid_to_hex(&parent->commit->object.oid),
977                     oid_to_hex(&target->commit->object.oid));
978         /* The rest are the same as the parent */
979         blame_chunk(&d.dstq, &d.srcq, INT_MAX, d.offset, INT_MAX, parent);
980         *d.dstq = NULL;
981         queue_blames(sb, parent, newdest);
982
983         return;
984 }
985
986 /*
987  * The lines in blame_entry after splitting blames many times can become
988  * very small and trivial, and at some point it becomes pointless to
989  * blame the parents.  E.g. "\t\t}\n\t}\n\n" appears everywhere in any
990  * ordinary C program, and it is not worth to say it was copied from
991  * totally unrelated file in the parent.
992  *
993  * Compute how trivial the lines in the blame_entry are.
994  */
995 static unsigned ent_score(struct scoreboard *sb, struct blame_entry *e)
996 {
997         unsigned score;
998         const char *cp, *ep;
999
1000         if (e->score)
1001                 return e->score;
1002
1003         score = 1;
1004         cp = nth_line(sb, e->lno);
1005         ep = nth_line(sb, e->lno + e->num_lines);
1006         while (cp < ep) {
1007                 unsigned ch = *((unsigned char *)cp);
1008                 if (isalnum(ch))
1009                         score++;
1010                 cp++;
1011         }
1012         e->score = score;
1013         return score;
1014 }
1015
1016 /*
1017  * best_so_far[] and this[] are both a split of an existing blame_entry
1018  * that passes blame to the parent.  Maintain best_so_far the best split
1019  * so far, by comparing this and best_so_far and copying this into
1020  * bst_so_far as needed.
1021  */
1022 static void copy_split_if_better(struct scoreboard *sb,
1023                                  struct blame_entry *best_so_far,
1024                                  struct blame_entry *this)
1025 {
1026         int i;
1027
1028         if (!this[1].suspect)
1029                 return;
1030         if (best_so_far[1].suspect) {
1031                 if (ent_score(sb, &this[1]) < ent_score(sb, &best_so_far[1]))
1032                         return;
1033         }
1034
1035         for (i = 0; i < 3; i++)
1036                 origin_incref(this[i].suspect);
1037         decref_split(best_so_far);
1038         memcpy(best_so_far, this, sizeof(struct blame_entry [3]));
1039 }
1040
1041 /*
1042  * We are looking at a part of the final image represented by
1043  * ent (tlno and same are offset by ent->s_lno).
1044  * tlno is where we are looking at in the final image.
1045  * up to (but not including) same match preimage.
1046  * plno is where we are looking at in the preimage.
1047  *
1048  * <-------------- final image ---------------------->
1049  *       <------ent------>
1050  *         ^tlno ^same
1051  *    <---------preimage----->
1052  *         ^plno
1053  *
1054  * All line numbers are 0-based.
1055  */
1056 static void handle_split(struct scoreboard *sb,
1057                          struct blame_entry *ent,
1058                          int tlno, int plno, int same,
1059                          struct origin *parent,
1060                          struct blame_entry *split)
1061 {
1062         if (ent->num_lines <= tlno)
1063                 return;
1064         if (tlno < same) {
1065                 struct blame_entry this[3];
1066                 tlno += ent->s_lno;
1067                 same += ent->s_lno;
1068                 split_overlap(this, ent, tlno, plno, same, parent);
1069                 copy_split_if_better(sb, split, this);
1070                 decref_split(this);
1071         }
1072 }
1073
1074 struct handle_split_cb_data {
1075         struct scoreboard *sb;
1076         struct blame_entry *ent;
1077         struct origin *parent;
1078         struct blame_entry *split;
1079         long plno;
1080         long tlno;
1081 };
1082
1083 static int handle_split_cb(long start_a, long count_a,
1084                            long start_b, long count_b, void *data)
1085 {
1086         struct handle_split_cb_data *d = data;
1087         handle_split(d->sb, d->ent, d->tlno, d->plno, start_b, d->parent,
1088                      d->split);
1089         d->plno = start_a + count_a;
1090         d->tlno = start_b + count_b;
1091         return 0;
1092 }
1093
1094 /*
1095  * Find the lines from parent that are the same as ent so that
1096  * we can pass blames to it.  file_p has the blob contents for
1097  * the parent.
1098  */
1099 static void find_copy_in_blob(struct scoreboard *sb,
1100                               struct blame_entry *ent,
1101                               struct origin *parent,
1102                               struct blame_entry *split,
1103                               mmfile_t *file_p)
1104 {
1105         const char *cp;
1106         mmfile_t file_o;
1107         struct handle_split_cb_data d;
1108
1109         memset(&d, 0, sizeof(d));
1110         d.sb = sb; d.ent = ent; d.parent = parent; d.split = split;
1111         /*
1112          * Prepare mmfile that contains only the lines in ent.
1113          */
1114         cp = nth_line(sb, ent->lno);
1115         file_o.ptr = (char *) cp;
1116         file_o.size = nth_line(sb, ent->lno + ent->num_lines) - cp;
1117
1118         /*
1119          * file_o is a part of final image we are annotating.
1120          * file_p partially may match that image.
1121          */
1122         memset(split, 0, sizeof(struct blame_entry [3]));
1123         if (diff_hunks(file_p, &file_o, handle_split_cb, &d))
1124                 die("unable to generate diff (%s)",
1125                     oid_to_hex(&parent->commit->object.oid));
1126         /* remainder, if any, all match the preimage */
1127         handle_split(sb, ent, d.tlno, d.plno, ent->num_lines, parent, split);
1128 }
1129
1130 /* Move all blame entries from list *source that have a score smaller
1131  * than score_min to the front of list *small.
1132  * Returns a pointer to the link pointing to the old head of the small list.
1133  */
1134
1135 static struct blame_entry **filter_small(struct scoreboard *sb,
1136                                          struct blame_entry **small,
1137                                          struct blame_entry **source,
1138                                          unsigned score_min)
1139 {
1140         struct blame_entry *p = *source;
1141         struct blame_entry *oldsmall = *small;
1142         while (p) {
1143                 if (ent_score(sb, p) <= score_min) {
1144                         *small = p;
1145                         small = &p->next;
1146                         p = *small;
1147                 } else {
1148                         *source = p;
1149                         source = &p->next;
1150                         p = *source;
1151                 }
1152         }
1153         *small = oldsmall;
1154         *source = NULL;
1155         return small;
1156 }
1157
1158 /*
1159  * See if lines currently target is suspected for can be attributed to
1160  * parent.
1161  */
1162 static void find_move_in_parent(struct scoreboard *sb,
1163                                 struct blame_entry ***blamed,
1164                                 struct blame_entry **toosmall,
1165                                 struct origin *target,
1166                                 struct origin *parent)
1167 {
1168         struct blame_entry *e, split[3];
1169         struct blame_entry *unblamed = target->suspects;
1170         struct blame_entry *leftover = NULL;
1171         mmfile_t file_p;
1172
1173         if (!unblamed)
1174                 return; /* nothing remains for this target */
1175
1176         fill_origin_blob(&sb->revs->diffopt, parent, &file_p);
1177         if (!file_p.ptr)
1178                 return;
1179
1180         /* At each iteration, unblamed has a NULL-terminated list of
1181          * entries that have not yet been tested for blame.  leftover
1182          * contains the reversed list of entries that have been tested
1183          * without being assignable to the parent.
1184          */
1185         do {
1186                 struct blame_entry **unblamedtail = &unblamed;
1187                 struct blame_entry *next;
1188                 for (e = unblamed; e; e = next) {
1189                         next = e->next;
1190                         find_copy_in_blob(sb, e, parent, split, &file_p);
1191                         if (split[1].suspect &&
1192                             blame_move_score < ent_score(sb, &split[1])) {
1193                                 split_blame(blamed, &unblamedtail, split, e);
1194                         } else {
1195                                 e->next = leftover;
1196                                 leftover = e;
1197                         }
1198                         decref_split(split);
1199                 }
1200                 *unblamedtail = NULL;
1201                 toosmall = filter_small(sb, toosmall, &unblamed, blame_move_score);
1202         } while (unblamed);
1203         target->suspects = reverse_blame(leftover, NULL);
1204 }
1205
1206 struct blame_list {
1207         struct blame_entry *ent;
1208         struct blame_entry split[3];
1209 };
1210
1211 /*
1212  * Count the number of entries the target is suspected for,
1213  * and prepare a list of entry and the best split.
1214  */
1215 static struct blame_list *setup_blame_list(struct blame_entry *unblamed,
1216                                            int *num_ents_p)
1217 {
1218         struct blame_entry *e;
1219         int num_ents, i;
1220         struct blame_list *blame_list = NULL;
1221
1222         for (e = unblamed, num_ents = 0; e; e = e->next)
1223                 num_ents++;
1224         if (num_ents) {
1225                 blame_list = xcalloc(num_ents, sizeof(struct blame_list));
1226                 for (e = unblamed, i = 0; e; e = e->next)
1227                         blame_list[i++].ent = e;
1228         }
1229         *num_ents_p = num_ents;
1230         return blame_list;
1231 }
1232
1233 /*
1234  * For lines target is suspected for, see if we can find code movement
1235  * across file boundary from the parent commit.  porigin is the path
1236  * in the parent we already tried.
1237  */
1238 static void find_copy_in_parent(struct scoreboard *sb,
1239                                 struct blame_entry ***blamed,
1240                                 struct blame_entry **toosmall,
1241                                 struct origin *target,
1242                                 struct commit *parent,
1243                                 struct origin *porigin,
1244                                 int opt)
1245 {
1246         struct diff_options diff_opts;
1247         int i, j;
1248         struct blame_list *blame_list;
1249         int num_ents;
1250         struct blame_entry *unblamed = target->suspects;
1251         struct blame_entry *leftover = NULL;
1252
1253         if (!unblamed)
1254                 return; /* nothing remains for this target */
1255
1256         diff_setup(&diff_opts);
1257         DIFF_OPT_SET(&diff_opts, RECURSIVE);
1258         diff_opts.output_format = DIFF_FORMAT_NO_OUTPUT;
1259
1260         diff_setup_done(&diff_opts);
1261
1262         /* Try "find copies harder" on new path if requested;
1263          * we do not want to use diffcore_rename() actually to
1264          * match things up; find_copies_harder is set only to
1265          * force diff_tree_sha1() to feed all filepairs to diff_queue,
1266          * and this code needs to be after diff_setup_done(), which
1267          * usually makes find-copies-harder imply copy detection.
1268          */
1269         if ((opt & PICKAXE_BLAME_COPY_HARDEST)
1270             || ((opt & PICKAXE_BLAME_COPY_HARDER)
1271                 && (!porigin || strcmp(target->path, porigin->path))))
1272                 DIFF_OPT_SET(&diff_opts, FIND_COPIES_HARDER);
1273
1274         if (is_null_oid(&target->commit->object.oid))
1275                 do_diff_cache(parent->tree->object.oid.hash, &diff_opts);
1276         else
1277                 diff_tree_sha1(parent->tree->object.oid.hash,
1278                                target->commit->tree->object.oid.hash,
1279                                "", &diff_opts);
1280
1281         if (!DIFF_OPT_TST(&diff_opts, FIND_COPIES_HARDER))
1282                 diffcore_std(&diff_opts);
1283
1284         do {
1285                 struct blame_entry **unblamedtail = &unblamed;
1286                 blame_list = setup_blame_list(unblamed, &num_ents);
1287
1288                 for (i = 0; i < diff_queued_diff.nr; i++) {
1289                         struct diff_filepair *p = diff_queued_diff.queue[i];
1290                         struct origin *norigin;
1291                         mmfile_t file_p;
1292                         struct blame_entry this[3];
1293
1294                         if (!DIFF_FILE_VALID(p->one))
1295                                 continue; /* does not exist in parent */
1296                         if (S_ISGITLINK(p->one->mode))
1297                                 continue; /* ignore git links */
1298                         if (porigin && !strcmp(p->one->path, porigin->path))
1299                                 /* find_move already dealt with this path */
1300                                 continue;
1301
1302                         norigin = get_origin(sb, parent, p->one->path);
1303                         oidcpy(&norigin->blob_oid, &p->one->oid);
1304                         norigin->mode = p->one->mode;
1305                         fill_origin_blob(&sb->revs->diffopt, norigin, &file_p);
1306                         if (!file_p.ptr)
1307                                 continue;
1308
1309                         for (j = 0; j < num_ents; j++) {
1310                                 find_copy_in_blob(sb, blame_list[j].ent,
1311                                                   norigin, this, &file_p);
1312                                 copy_split_if_better(sb, blame_list[j].split,
1313                                                      this);
1314                                 decref_split(this);
1315                         }
1316                         origin_decref(norigin);
1317                 }
1318
1319                 for (j = 0; j < num_ents; j++) {
1320                         struct blame_entry *split = blame_list[j].split;
1321                         if (split[1].suspect &&
1322                             blame_copy_score < ent_score(sb, &split[1])) {
1323                                 split_blame(blamed, &unblamedtail, split,
1324                                             blame_list[j].ent);
1325                         } else {
1326                                 blame_list[j].ent->next = leftover;
1327                                 leftover = blame_list[j].ent;
1328                         }
1329                         decref_split(split);
1330                 }
1331                 free(blame_list);
1332                 *unblamedtail = NULL;
1333                 toosmall = filter_small(sb, toosmall, &unblamed, blame_copy_score);
1334         } while (unblamed);
1335         target->suspects = reverse_blame(leftover, NULL);
1336         diff_flush(&diff_opts);
1337         clear_pathspec(&diff_opts.pathspec);
1338 }
1339
1340 /*
1341  * The blobs of origin and porigin exactly match, so everything
1342  * origin is suspected for can be blamed on the parent.
1343  */
1344 static void pass_whole_blame(struct scoreboard *sb,
1345                              struct origin *origin, struct origin *porigin)
1346 {
1347         struct blame_entry *e, *suspects;
1348
1349         if (!porigin->file.ptr && origin->file.ptr) {
1350                 /* Steal its file */
1351                 porigin->file = origin->file;
1352                 origin->file.ptr = NULL;
1353         }
1354         suspects = origin->suspects;
1355         origin->suspects = NULL;
1356         for (e = suspects; e; e = e->next) {
1357                 origin_incref(porigin);
1358                 origin_decref(e->suspect);
1359                 e->suspect = porigin;
1360         }
1361         queue_blames(sb, porigin, suspects);
1362 }
1363
1364 /*
1365  * We pass blame from the current commit to its parents.  We keep saying
1366  * "parent" (and "porigin"), but what we mean is to find scapegoat to
1367  * exonerate ourselves.
1368  */
1369 static struct commit_list *first_scapegoat(struct rev_info *revs, struct commit *commit)
1370 {
1371         if (!reverse) {
1372                 if (revs->first_parent_only &&
1373                     commit->parents &&
1374                     commit->parents->next) {
1375                         free_commit_list(commit->parents->next);
1376                         commit->parents->next = NULL;
1377                 }
1378                 return commit->parents;
1379         }
1380         return lookup_decoration(&revs->children, &commit->object);
1381 }
1382
1383 static int num_scapegoats(struct rev_info *revs, struct commit *commit)
1384 {
1385         struct commit_list *l = first_scapegoat(revs, commit);
1386         return commit_list_count(l);
1387 }
1388
1389 /* Distribute collected unsorted blames to the respected sorted lists
1390  * in the various origins.
1391  */
1392 static void distribute_blame(struct scoreboard *sb, struct blame_entry *blamed)
1393 {
1394         blamed = blame_sort(blamed, compare_blame_suspect);
1395         while (blamed)
1396         {
1397                 struct origin *porigin = blamed->suspect;
1398                 struct blame_entry *suspects = NULL;
1399                 do {
1400                         struct blame_entry *next = blamed->next;
1401                         blamed->next = suspects;
1402                         suspects = blamed;
1403                         blamed = next;
1404                 } while (blamed && blamed->suspect == porigin);
1405                 suspects = reverse_blame(suspects, NULL);
1406                 queue_blames(sb, porigin, suspects);
1407         }
1408 }
1409
1410 #define MAXSG 16
1411
1412 static void pass_blame(struct scoreboard *sb, struct origin *origin, int opt)
1413 {
1414         struct rev_info *revs = sb->revs;
1415         int i, pass, num_sg;
1416         struct commit *commit = origin->commit;
1417         struct commit_list *sg;
1418         struct origin *sg_buf[MAXSG];
1419         struct origin *porigin, **sg_origin = sg_buf;
1420         struct blame_entry *toosmall = NULL;
1421         struct blame_entry *blames, **blametail = &blames;
1422
1423         num_sg = num_scapegoats(revs, commit);
1424         if (!num_sg)
1425                 goto finish;
1426         else if (num_sg < ARRAY_SIZE(sg_buf))
1427                 memset(sg_buf, 0, sizeof(sg_buf));
1428         else
1429                 sg_origin = xcalloc(num_sg, sizeof(*sg_origin));
1430
1431         /*
1432          * The first pass looks for unrenamed path to optimize for
1433          * common cases, then we look for renames in the second pass.
1434          */
1435         for (pass = 0; pass < 2 - no_whole_file_rename; pass++) {
1436                 struct origin *(*find)(struct scoreboard *,
1437                                        struct commit *, struct origin *);
1438                 find = pass ? find_rename : find_origin;
1439
1440                 for (i = 0, sg = first_scapegoat(revs, commit);
1441                      i < num_sg && sg;
1442                      sg = sg->next, i++) {
1443                         struct commit *p = sg->item;
1444                         int j, same;
1445
1446                         if (sg_origin[i])
1447                                 continue;
1448                         if (parse_commit(p))
1449                                 continue;
1450                         porigin = find(sb, p, origin);
1451                         if (!porigin)
1452                                 continue;
1453                         if (!oidcmp(&porigin->blob_oid, &origin->blob_oid)) {
1454                                 pass_whole_blame(sb, origin, porigin);
1455                                 origin_decref(porigin);
1456                                 goto finish;
1457                         }
1458                         for (j = same = 0; j < i; j++)
1459                                 if (sg_origin[j] &&
1460                                     !oidcmp(&sg_origin[j]->blob_oid, &porigin->blob_oid)) {
1461                                         same = 1;
1462                                         break;
1463                                 }
1464                         if (!same)
1465                                 sg_origin[i] = porigin;
1466                         else
1467                                 origin_decref(porigin);
1468                 }
1469         }
1470
1471         num_commits++;
1472         for (i = 0, sg = first_scapegoat(revs, commit);
1473              i < num_sg && sg;
1474              sg = sg->next, i++) {
1475                 struct origin *porigin = sg_origin[i];
1476                 if (!porigin)
1477                         continue;
1478                 if (!origin->previous) {
1479                         origin_incref(porigin);
1480                         origin->previous = porigin;
1481                 }
1482                 pass_blame_to_parent(sb, origin, porigin);
1483                 if (!origin->suspects)
1484                         goto finish;
1485         }
1486
1487         /*
1488          * Optionally find moves in parents' files.
1489          */
1490         if (opt & PICKAXE_BLAME_MOVE) {
1491                 filter_small(sb, &toosmall, &origin->suspects, blame_move_score);
1492                 if (origin->suspects) {
1493                         for (i = 0, sg = first_scapegoat(revs, commit);
1494                              i < num_sg && sg;
1495                              sg = sg->next, i++) {
1496                                 struct origin *porigin = sg_origin[i];
1497                                 if (!porigin)
1498                                         continue;
1499                                 find_move_in_parent(sb, &blametail, &toosmall, origin, porigin);
1500                                 if (!origin->suspects)
1501                                         break;
1502                         }
1503                 }
1504         }
1505
1506         /*
1507          * Optionally find copies from parents' files.
1508          */
1509         if (opt & PICKAXE_BLAME_COPY) {
1510                 if (blame_copy_score > blame_move_score)
1511                         filter_small(sb, &toosmall, &origin->suspects, blame_copy_score);
1512                 else if (blame_copy_score < blame_move_score) {
1513                         origin->suspects = blame_merge(origin->suspects, toosmall);
1514                         toosmall = NULL;
1515                         filter_small(sb, &toosmall, &origin->suspects, blame_copy_score);
1516                 }
1517                 if (!origin->suspects)
1518                         goto finish;
1519
1520                 for (i = 0, sg = first_scapegoat(revs, commit);
1521                      i < num_sg && sg;
1522                      sg = sg->next, i++) {
1523                         struct origin *porigin = sg_origin[i];
1524                         find_copy_in_parent(sb, &blametail, &toosmall,
1525                                             origin, sg->item, porigin, opt);
1526                         if (!origin->suspects)
1527                                 goto finish;
1528                 }
1529         }
1530
1531 finish:
1532         *blametail = NULL;
1533         distribute_blame(sb, blames);
1534         /*
1535          * prepend toosmall to origin->suspects
1536          *
1537          * There is no point in sorting: this ends up on a big
1538          * unsorted list in the caller anyway.
1539          */
1540         if (toosmall) {
1541                 struct blame_entry **tail = &toosmall;
1542                 while (*tail)
1543                         tail = &(*tail)->next;
1544                 *tail = origin->suspects;
1545                 origin->suspects = toosmall;
1546         }
1547         for (i = 0; i < num_sg; i++) {
1548                 if (sg_origin[i]) {
1549                         drop_origin_blob(sg_origin[i]);
1550                         origin_decref(sg_origin[i]);
1551                 }
1552         }
1553         drop_origin_blob(origin);
1554         if (sg_buf != sg_origin)
1555                 free(sg_origin);
1556 }
1557
1558 /*
1559  * Information on commits, used for output.
1560  */
1561 struct commit_info {
1562         struct strbuf author;
1563         struct strbuf author_mail;
1564         unsigned long author_time;
1565         struct strbuf author_tz;
1566
1567         /* filled only when asked for details */
1568         struct strbuf committer;
1569         struct strbuf committer_mail;
1570         unsigned long committer_time;
1571         struct strbuf committer_tz;
1572
1573         struct strbuf summary;
1574 };
1575
1576 /*
1577  * Parse author/committer line in the commit object buffer
1578  */
1579 static void get_ac_line(const char *inbuf, const char *what,
1580         struct strbuf *name, struct strbuf *mail,
1581         unsigned long *time, struct strbuf *tz)
1582 {
1583         struct ident_split ident;
1584         size_t len, maillen, namelen;
1585         char *tmp, *endp;
1586         const char *namebuf, *mailbuf;
1587
1588         tmp = strstr(inbuf, what);
1589         if (!tmp)
1590                 goto error_out;
1591         tmp += strlen(what);
1592         endp = strchr(tmp, '\n');
1593         if (!endp)
1594                 len = strlen(tmp);
1595         else
1596                 len = endp - tmp;
1597
1598         if (split_ident_line(&ident, tmp, len)) {
1599         error_out:
1600                 /* Ugh */
1601                 tmp = "(unknown)";
1602                 strbuf_addstr(name, tmp);
1603                 strbuf_addstr(mail, tmp);
1604                 strbuf_addstr(tz, tmp);
1605                 *time = 0;
1606                 return;
1607         }
1608
1609         namelen = ident.name_end - ident.name_begin;
1610         namebuf = ident.name_begin;
1611
1612         maillen = ident.mail_end - ident.mail_begin;
1613         mailbuf = ident.mail_begin;
1614
1615         if (ident.date_begin && ident.date_end)
1616                 *time = strtoul(ident.date_begin, NULL, 10);
1617         else
1618                 *time = 0;
1619
1620         if (ident.tz_begin && ident.tz_end)
1621                 strbuf_add(tz, ident.tz_begin, ident.tz_end - ident.tz_begin);
1622         else
1623                 strbuf_addstr(tz, "(unknown)");
1624
1625         /*
1626          * Now, convert both name and e-mail using mailmap
1627          */
1628         map_user(&mailmap, &mailbuf, &maillen,
1629                  &namebuf, &namelen);
1630
1631         strbuf_addf(mail, "<%.*s>", (int)maillen, mailbuf);
1632         strbuf_add(name, namebuf, namelen);
1633 }
1634
1635 static void commit_info_init(struct commit_info *ci)
1636 {
1637
1638         strbuf_init(&ci->author, 0);
1639         strbuf_init(&ci->author_mail, 0);
1640         strbuf_init(&ci->author_tz, 0);
1641         strbuf_init(&ci->committer, 0);
1642         strbuf_init(&ci->committer_mail, 0);
1643         strbuf_init(&ci->committer_tz, 0);
1644         strbuf_init(&ci->summary, 0);
1645 }
1646
1647 static void commit_info_destroy(struct commit_info *ci)
1648 {
1649
1650         strbuf_release(&ci->author);
1651         strbuf_release(&ci->author_mail);
1652         strbuf_release(&ci->author_tz);
1653         strbuf_release(&ci->committer);
1654         strbuf_release(&ci->committer_mail);
1655         strbuf_release(&ci->committer_tz);
1656         strbuf_release(&ci->summary);
1657 }
1658
1659 static void get_commit_info(struct commit *commit,
1660                             struct commit_info *ret,
1661                             int detailed)
1662 {
1663         int len;
1664         const char *subject, *encoding;
1665         const char *message;
1666
1667         commit_info_init(ret);
1668
1669         encoding = get_log_output_encoding();
1670         message = logmsg_reencode(commit, NULL, encoding);
1671         get_ac_line(message, "\nauthor ",
1672                     &ret->author, &ret->author_mail,
1673                     &ret->author_time, &ret->author_tz);
1674
1675         if (!detailed) {
1676                 unuse_commit_buffer(commit, message);
1677                 return;
1678         }
1679
1680         get_ac_line(message, "\ncommitter ",
1681                     &ret->committer, &ret->committer_mail,
1682                     &ret->committer_time, &ret->committer_tz);
1683
1684         len = find_commit_subject(message, &subject);
1685         if (len)
1686                 strbuf_add(&ret->summary, subject, len);
1687         else
1688                 strbuf_addf(&ret->summary, "(%s)", oid_to_hex(&commit->object.oid));
1689
1690         unuse_commit_buffer(commit, message);
1691 }
1692
1693 /*
1694  * Write out any suspect information which depends on the path. This must be
1695  * handled separately from emit_one_suspect_detail(), because a given commit
1696  * may have changes in multiple paths. So this needs to appear each time
1697  * we mention a new group.
1698  *
1699  * To allow LF and other nonportable characters in pathnames,
1700  * they are c-style quoted as needed.
1701  */
1702 static void write_filename_info(struct origin *suspect)
1703 {
1704         if (suspect->previous) {
1705                 struct origin *prev = suspect->previous;
1706                 printf("previous %s ", oid_to_hex(&prev->commit->object.oid));
1707                 write_name_quoted(prev->path, stdout, '\n');
1708         }
1709         printf("filename ");
1710         write_name_quoted(suspect->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
1740         commit_info_destroy(&ci);
1741
1742         return 1;
1743 }
1744
1745 /*
1746  * The blame_entry is found to be guilty for the range.
1747  * Show it in incremental output.
1748  */
1749 static void found_guilty_entry(struct blame_entry *ent,
1750                            struct progress_info *pi)
1751 {
1752         if (incremental) {
1753                 struct origin *suspect = ent->suspect;
1754
1755                 printf("%s %d %d %d\n",
1756                        oid_to_hex(&suspect->commit->object.oid),
1757                        ent->s_lno + 1, ent->lno + 1, ent->num_lines);
1758                 emit_one_suspect_detail(suspect, 0);
1759                 write_filename_info(suspect);
1760                 maybe_flush_or_die(stdout, "stdout");
1761         }
1762         pi->blamed_lines += ent->num_lines;
1763         display_progress(pi->progress, pi->blamed_lines);
1764 }
1765
1766 /*
1767  * The main loop -- while we have blobs with lines whose true origin
1768  * is still unknown, pick one blob, and allow its lines to pass blames
1769  * to its parents. */
1770 static void assign_blame(struct scoreboard *sb, int opt)
1771 {
1772         struct rev_info *revs = sb->revs;
1773         struct commit *commit = prio_queue_get(&sb->commits);
1774         struct progress_info pi = { NULL, 0 };
1775
1776         if (show_progress)
1777                 pi.progress = start_progress_delay(_("Blaming lines"),
1778                                                    sb->num_lines, 50, 1);
1779
1780         while (commit) {
1781                 struct blame_entry *ent;
1782                 struct origin *suspect = commit->util;
1783
1784                 /* find one suspect to break down */
1785                 while (suspect && !suspect->suspects)
1786                         suspect = suspect->next;
1787
1788                 if (!suspect) {
1789                         commit = prio_queue_get(&sb->commits);
1790                         continue;
1791                 }
1792
1793                 assert(commit == suspect->commit);
1794
1795                 /*
1796                  * We will use this suspect later in the loop,
1797                  * so hold onto it in the meantime.
1798                  */
1799                 origin_incref(suspect);
1800                 parse_commit(commit);
1801                 if (reverse ||
1802                     (!(commit->object.flags & UNINTERESTING) &&
1803                      !(revs->max_age != -1 && commit->date < revs->max_age)))
1804                         pass_blame(sb, suspect, opt);
1805                 else {
1806                         commit->object.flags |= UNINTERESTING;
1807                         if (commit->object.parsed)
1808                                 mark_parents_uninteresting(commit);
1809                 }
1810                 /* treat root commit as boundary */
1811                 if (!commit->parents && !show_root)
1812                         commit->object.flags |= UNINTERESTING;
1813
1814                 /* Take responsibility for the remaining entries */
1815                 ent = suspect->suspects;
1816                 if (ent) {
1817                         suspect->guilty = 1;
1818                         for (;;) {
1819                                 struct blame_entry *next = ent->next;
1820                                 found_guilty_entry(ent, &pi);
1821                                 if (next) {
1822                                         ent = next;
1823                                         continue;
1824                                 }
1825                                 ent->next = sb->ent;
1826                                 sb->ent = suspect->suspects;
1827                                 suspect->suspects = NULL;
1828                                 break;
1829                         }
1830                 }
1831                 origin_decref(suspect);
1832
1833                 if (DEBUG) /* sanity */
1834                         sanity_check_refcnt(sb);
1835         }
1836
1837         stop_progress(&pi.progress);
1838 }
1839
1840 static const char *format_time(unsigned long time, const char *tz_str,
1841                                int show_raw_time)
1842 {
1843         static struct strbuf time_buf = STRBUF_INIT;
1844
1845         strbuf_reset(&time_buf);
1846         if (show_raw_time) {
1847                 strbuf_addf(&time_buf, "%lu %s", time, tz_str);
1848         }
1849         else {
1850                 const char *time_str;
1851                 size_t time_width;
1852                 int tz;
1853                 tz = atoi(tz_str);
1854                 time_str = show_date(time, tz, &blame_date_mode);
1855                 strbuf_addstr(&time_buf, time_str);
1856                 /*
1857                  * Add space paddings to time_buf to display a fixed width
1858                  * string, and use time_width for display width calibration.
1859                  */
1860                 for (time_width = utf8_strwidth(time_str);
1861                      time_width < blame_date_width;
1862                      time_width++)
1863                         strbuf_addch(&time_buf, ' ');
1864         }
1865         return time_buf.buf;
1866 }
1867
1868 #define OUTPUT_ANNOTATE_COMPAT  001
1869 #define OUTPUT_LONG_OBJECT_NAME 002
1870 #define OUTPUT_RAW_TIMESTAMP    004
1871 #define OUTPUT_PORCELAIN        010
1872 #define OUTPUT_SHOW_NAME        020
1873 #define OUTPUT_SHOW_NUMBER      040
1874 #define OUTPUT_SHOW_SCORE      0100
1875 #define OUTPUT_NO_AUTHOR       0200
1876 #define OUTPUT_SHOW_EMAIL       0400
1877 #define OUTPUT_LINE_PORCELAIN 01000
1878
1879 static void emit_porcelain_details(struct origin *suspect, int repeat)
1880 {
1881         if (emit_one_suspect_detail(suspect, repeat) ||
1882             (suspect->commit->object.flags & MORE_THAN_ONE_PATH))
1883                 write_filename_info(suspect);
1884 }
1885
1886 static void emit_porcelain(struct scoreboard *sb, struct blame_entry *ent,
1887                            int opt)
1888 {
1889         int repeat = opt & OUTPUT_LINE_PORCELAIN;
1890         int cnt;
1891         const char *cp;
1892         struct origin *suspect = ent->suspect;
1893         char hex[GIT_SHA1_HEXSZ + 1];
1894
1895         oid_to_hex_r(hex, &suspect->commit->object.oid);
1896         printf("%s %d %d %d\n",
1897                hex,
1898                ent->s_lno + 1,
1899                ent->lno + 1,
1900                ent->num_lines);
1901         emit_porcelain_details(suspect, repeat);
1902
1903         cp = nth_line(sb, ent->lno);
1904         for (cnt = 0; cnt < ent->num_lines; cnt++) {
1905                 char ch;
1906                 if (cnt) {
1907                         printf("%s %d %d\n", hex,
1908                                ent->s_lno + 1 + cnt,
1909                                ent->lno + 1 + cnt);
1910                         if (repeat)
1911                                 emit_porcelain_details(suspect, 1);
1912                 }
1913                 putchar('\t');
1914                 do {
1915                         ch = *cp++;
1916                         putchar(ch);
1917                 } while (ch != '\n' &&
1918                          cp < sb->final_buf + sb->final_buf_size);
1919         }
1920
1921         if (sb->final_buf_size && cp[-1] != '\n')
1922                 putchar('\n');
1923 }
1924
1925 static void emit_other(struct scoreboard *sb, struct blame_entry *ent, int opt)
1926 {
1927         int cnt;
1928         const char *cp;
1929         struct origin *suspect = ent->suspect;
1930         struct commit_info ci;
1931         char hex[GIT_SHA1_HEXSZ + 1];
1932         int show_raw_time = !!(opt & OUTPUT_RAW_TIMESTAMP);
1933
1934         get_commit_info(suspect->commit, &ci, 1);
1935         oid_to_hex_r(hex, &suspect->commit->object.oid);
1936
1937         cp = nth_line(sb, ent->lno);
1938         for (cnt = 0; cnt < ent->num_lines; cnt++) {
1939                 char ch;
1940                 int length = (opt & OUTPUT_LONG_OBJECT_NAME) ? GIT_SHA1_HEXSZ : abbrev;
1941
1942                 if (suspect->commit->object.flags & UNINTERESTING) {
1943                         if (blank_boundary)
1944                                 memset(hex, ' ', length);
1945                         else if (!(opt & OUTPUT_ANNOTATE_COMPAT)) {
1946                                 length--;
1947                                 putchar('^');
1948                         }
1949                 }
1950
1951                 printf("%.*s", length, hex);
1952                 if (opt & OUTPUT_ANNOTATE_COMPAT) {
1953                         const char *name;
1954                         if (opt & OUTPUT_SHOW_EMAIL)
1955                                 name = ci.author_mail.buf;
1956                         else
1957                                 name = ci.author.buf;
1958                         printf("\t(%10s\t%10s\t%d)", name,
1959                                format_time(ci.author_time, ci.author_tz.buf,
1960                                            show_raw_time),
1961                                ent->lno + 1 + cnt);
1962                 } else {
1963                         if (opt & OUTPUT_SHOW_SCORE)
1964                                 printf(" %*d %02d",
1965                                        max_score_digits, ent->score,
1966                                        ent->suspect->refcnt);
1967                         if (opt & OUTPUT_SHOW_NAME)
1968                                 printf(" %-*.*s", longest_file, longest_file,
1969                                        suspect->path);
1970                         if (opt & OUTPUT_SHOW_NUMBER)
1971                                 printf(" %*d", max_orig_digits,
1972                                        ent->s_lno + 1 + cnt);
1973
1974                         if (!(opt & OUTPUT_NO_AUTHOR)) {
1975                                 const char *name;
1976                                 int pad;
1977                                 if (opt & OUTPUT_SHOW_EMAIL)
1978                                         name = ci.author_mail.buf;
1979                                 else
1980                                         name = ci.author.buf;
1981                                 pad = longest_author - utf8_strwidth(name);
1982                                 printf(" (%s%*s %10s",
1983                                        name, pad, "",
1984                                        format_time(ci.author_time,
1985                                                    ci.author_tz.buf,
1986                                                    show_raw_time));
1987                         }
1988                         printf(" %*d) ",
1989                                max_digits, ent->lno + 1 + cnt);
1990                 }
1991                 do {
1992                         ch = *cp++;
1993                         putchar(ch);
1994                 } while (ch != '\n' &&
1995                          cp < sb->final_buf + sb->final_buf_size);
1996         }
1997
1998         if (sb->final_buf_size && cp[-1] != '\n')
1999                 putchar('\n');
2000
2001         commit_info_destroy(&ci);
2002 }
2003
2004 static void output(struct scoreboard *sb, int option)
2005 {
2006         struct blame_entry *ent;
2007
2008         if (option & OUTPUT_PORCELAIN) {
2009                 for (ent = sb->ent; ent; ent = ent->next) {
2010                         int count = 0;
2011                         struct origin *suspect;
2012                         struct commit *commit = ent->suspect->commit;
2013                         if (commit->object.flags & MORE_THAN_ONE_PATH)
2014                                 continue;
2015                         for (suspect = commit->util; suspect; suspect = suspect->next) {
2016                                 if (suspect->guilty && count++) {
2017                                         commit->object.flags |= MORE_THAN_ONE_PATH;
2018                                         break;
2019                                 }
2020                         }
2021                 }
2022         }
2023
2024         for (ent = sb->ent; ent; ent = ent->next) {
2025                 if (option & OUTPUT_PORCELAIN)
2026                         emit_porcelain(sb, ent, option);
2027                 else {
2028                         emit_other(sb, ent, option);
2029                 }
2030         }
2031 }
2032
2033 static const char *get_next_line(const char *start, const char *end)
2034 {
2035         const char *nl = memchr(start, '\n', end - start);
2036         return nl ? nl + 1 : end;
2037 }
2038
2039 /*
2040  * To allow quick access to the contents of nth line in the
2041  * final image, prepare an index in the scoreboard.
2042  */
2043 static int prepare_lines(struct scoreboard *sb)
2044 {
2045         const char *buf = sb->final_buf;
2046         unsigned long len = sb->final_buf_size;
2047         const char *end = buf + len;
2048         const char *p;
2049         int *lineno;
2050         int num = 0;
2051
2052         for (p = buf; p < end; p = get_next_line(p, end))
2053                 num++;
2054
2055         ALLOC_ARRAY(sb->lineno, num + 1);
2056         lineno = sb->lineno;
2057
2058         for (p = buf; p < end; p = get_next_line(p, end))
2059                 *lineno++ = p - buf;
2060
2061         *lineno = len;
2062
2063         sb->num_lines = num;
2064         return sb->num_lines;
2065 }
2066
2067 /*
2068  * Add phony grafts for use with -S; this is primarily to
2069  * support git's cvsserver that wants to give a linear history
2070  * to its clients.
2071  */
2072 static int read_ancestry(const char *graft_file)
2073 {
2074         FILE *fp = fopen(graft_file, "r");
2075         struct strbuf buf = STRBUF_INIT;
2076         if (!fp)
2077                 return -1;
2078         while (!strbuf_getwholeline(&buf, fp, '\n')) {
2079                 /* The format is just "Commit Parent1 Parent2 ...\n" */
2080                 struct commit_graft *graft = read_graft_line(buf.buf, buf.len);
2081                 if (graft)
2082                         register_commit_graft(graft, 0);
2083         }
2084         fclose(fp);
2085         strbuf_release(&buf);
2086         return 0;
2087 }
2088
2089 static int update_auto_abbrev(int auto_abbrev, struct origin *suspect)
2090 {
2091         const char *uniq = find_unique_abbrev(suspect->commit->object.oid.hash,
2092                                               auto_abbrev);
2093         int len = strlen(uniq);
2094         if (auto_abbrev < len)
2095                 return len;
2096         return auto_abbrev;
2097 }
2098
2099 /*
2100  * How many columns do we need to show line numbers, authors,
2101  * and filenames?
2102  */
2103 static void find_alignment(struct scoreboard *sb, int *option)
2104 {
2105         int longest_src_lines = 0;
2106         int longest_dst_lines = 0;
2107         unsigned largest_score = 0;
2108         struct blame_entry *e;
2109         int compute_auto_abbrev = (abbrev < 0);
2110         int auto_abbrev = DEFAULT_ABBREV;
2111
2112         for (e = sb->ent; e; e = e->next) {
2113                 struct origin *suspect = e->suspect;
2114                 int num;
2115
2116                 if (compute_auto_abbrev)
2117                         auto_abbrev = update_auto_abbrev(auto_abbrev, suspect);
2118                 if (strcmp(suspect->path, sb->path))
2119                         *option |= OUTPUT_SHOW_NAME;
2120                 num = strlen(suspect->path);
2121                 if (longest_file < num)
2122                         longest_file = num;
2123                 if (!(suspect->commit->object.flags & METAINFO_SHOWN)) {
2124                         struct commit_info ci;
2125                         suspect->commit->object.flags |= METAINFO_SHOWN;
2126                         get_commit_info(suspect->commit, &ci, 1);
2127                         if (*option & OUTPUT_SHOW_EMAIL)
2128                                 num = utf8_strwidth(ci.author_mail.buf);
2129                         else
2130                                 num = utf8_strwidth(ci.author.buf);
2131                         if (longest_author < num)
2132                                 longest_author = num;
2133                         commit_info_destroy(&ci);
2134                 }
2135                 num = e->s_lno + e->num_lines;
2136                 if (longest_src_lines < num)
2137                         longest_src_lines = num;
2138                 num = e->lno + e->num_lines;
2139                 if (longest_dst_lines < num)
2140                         longest_dst_lines = num;
2141                 if (largest_score < ent_score(sb, e))
2142                         largest_score = ent_score(sb, e);
2143         }
2144         max_orig_digits = decimal_width(longest_src_lines);
2145         max_digits = decimal_width(longest_dst_lines);
2146         max_score_digits = decimal_width(largest_score);
2147
2148         if (compute_auto_abbrev)
2149                 /* one more abbrev length is needed for the boundary commit */
2150                 abbrev = auto_abbrev + 1;
2151 }
2152
2153 /*
2154  * For debugging -- origin is refcounted, and this asserts that
2155  * we do not underflow.
2156  */
2157 static void sanity_check_refcnt(struct scoreboard *sb)
2158 {
2159         int baa = 0;
2160         struct blame_entry *ent;
2161
2162         for (ent = sb->ent; ent; ent = ent->next) {
2163                 /* Nobody should have zero or negative refcnt */
2164                 if (ent->suspect->refcnt <= 0) {
2165                         fprintf(stderr, "%s in %s has negative refcnt %d\n",
2166                                 ent->suspect->path,
2167                                 oid_to_hex(&ent->suspect->commit->object.oid),
2168                                 ent->suspect->refcnt);
2169                         baa = 1;
2170                 }
2171         }
2172         if (baa) {
2173                 int opt = 0160;
2174                 find_alignment(sb, &opt);
2175                 output(sb, opt);
2176                 die("Baa %d!", baa);
2177         }
2178 }
2179
2180 static unsigned parse_score(const char *arg)
2181 {
2182         char *end;
2183         unsigned long score = strtoul(arg, &end, 10);
2184         if (*end)
2185                 return 0;
2186         return score;
2187 }
2188
2189 static const char *add_prefix(const char *prefix, const char *path)
2190 {
2191         return prefix_path(prefix, prefix ? strlen(prefix) : 0, path);
2192 }
2193
2194 static int git_blame_config(const char *var, const char *value, void *cb)
2195 {
2196         if (!strcmp(var, "blame.showroot")) {
2197                 show_root = git_config_bool(var, value);
2198                 return 0;
2199         }
2200         if (!strcmp(var, "blame.blankboundary")) {
2201                 blank_boundary = git_config_bool(var, value);
2202                 return 0;
2203         }
2204         if (!strcmp(var, "blame.showemail")) {
2205                 int *output_option = cb;
2206                 if (git_config_bool(var, value))
2207                         *output_option |= OUTPUT_SHOW_EMAIL;
2208                 else
2209                         *output_option &= ~OUTPUT_SHOW_EMAIL;
2210                 return 0;
2211         }
2212         if (!strcmp(var, "blame.date")) {
2213                 if (!value)
2214                         return config_error_nonbool(var);
2215                 parse_date_format(value, &blame_date_mode);
2216                 return 0;
2217         }
2218
2219         if (git_diff_heuristic_config(var, value, cb) < 0)
2220                 return -1;
2221         if (userdiff_config(var, value) < 0)
2222                 return -1;
2223
2224         return git_default_config(var, value, cb);
2225 }
2226
2227 static void verify_working_tree_path(struct commit *work_tree, const char *path)
2228 {
2229         struct commit_list *parents;
2230         int pos;
2231
2232         for (parents = work_tree->parents; parents; parents = parents->next) {
2233                 const struct object_id *commit_oid = &parents->item->object.oid;
2234                 struct object_id blob_oid;
2235                 unsigned mode;
2236
2237                 if (!get_tree_entry(commit_oid->hash, path, blob_oid.hash, &mode) &&
2238                     sha1_object_info(blob_oid.hash, NULL) == OBJ_BLOB)
2239                         return;
2240         }
2241
2242         pos = cache_name_pos(path, strlen(path));
2243         if (pos >= 0)
2244                 ; /* path is in the index */
2245         else if (-1 - pos < active_nr &&
2246                  !strcmp(active_cache[-1 - pos]->name, path))
2247                 ; /* path is in the index, unmerged */
2248         else
2249                 die("no such path '%s' in HEAD", path);
2250 }
2251
2252 static struct commit_list **append_parent(struct commit_list **tail, const struct object_id *oid)
2253 {
2254         struct commit *parent;
2255
2256         parent = lookup_commit_reference(oid->hash);
2257         if (!parent)
2258                 die("no such commit %s", oid_to_hex(oid));
2259         return &commit_list_insert(parent, tail)->next;
2260 }
2261
2262 static void append_merge_parents(struct commit_list **tail)
2263 {
2264         int merge_head;
2265         struct strbuf line = STRBUF_INIT;
2266
2267         merge_head = open(git_path_merge_head(), O_RDONLY);
2268         if (merge_head < 0) {
2269                 if (errno == ENOENT)
2270                         return;
2271                 die("cannot open '%s' for reading", git_path_merge_head());
2272         }
2273
2274         while (!strbuf_getwholeline_fd(&line, merge_head, '\n')) {
2275                 struct object_id oid;
2276                 if (line.len < GIT_SHA1_HEXSZ || get_oid_hex(line.buf, &oid))
2277                         die("unknown line in '%s': %s", git_path_merge_head(), line.buf);
2278                 tail = append_parent(tail, &oid);
2279         }
2280         close(merge_head);
2281         strbuf_release(&line);
2282 }
2283
2284 /*
2285  * This isn't as simple as passing sb->buf and sb->len, because we
2286  * want to transfer ownership of the buffer to the commit (so we
2287  * must use detach).
2288  */
2289 static void set_commit_buffer_from_strbuf(struct commit *c, struct strbuf *sb)
2290 {
2291         size_t len;
2292         void *buf = strbuf_detach(sb, &len);
2293         set_commit_buffer(c, buf, len);
2294 }
2295
2296 /*
2297  * Prepare a dummy commit that represents the work tree (or staged) item.
2298  * Note that annotating work tree item never works in the reverse.
2299  */
2300 static struct commit *fake_working_tree_commit(struct diff_options *opt,
2301                                                const char *path,
2302                                                const char *contents_from)
2303 {
2304         struct commit *commit;
2305         struct origin *origin;
2306         struct commit_list **parent_tail, *parent;
2307         struct object_id head_oid;
2308         struct strbuf buf = STRBUF_INIT;
2309         const char *ident;
2310         time_t now;
2311         int size, len;
2312         struct cache_entry *ce;
2313         unsigned mode;
2314         struct strbuf msg = STRBUF_INIT;
2315
2316         read_cache();
2317         time(&now);
2318         commit = alloc_commit_node();
2319         commit->object.parsed = 1;
2320         commit->date = now;
2321         parent_tail = &commit->parents;
2322
2323         if (!resolve_ref_unsafe("HEAD", RESOLVE_REF_READING, head_oid.hash, NULL))
2324                 die("no such ref: HEAD");
2325
2326         parent_tail = append_parent(parent_tail, &head_oid);
2327         append_merge_parents(parent_tail);
2328         verify_working_tree_path(commit, path);
2329
2330         origin = make_origin(commit, path);
2331
2332         ident = fmt_ident("Not Committed Yet", "not.committed.yet", NULL, 0);
2333         strbuf_addstr(&msg, "tree 0000000000000000000000000000000000000000\n");
2334         for (parent = commit->parents; parent; parent = parent->next)
2335                 strbuf_addf(&msg, "parent %s\n",
2336                             oid_to_hex(&parent->item->object.oid));
2337         strbuf_addf(&msg,
2338                     "author %s\n"
2339                     "committer %s\n\n"
2340                     "Version of %s from %s\n",
2341                     ident, ident, path,
2342                     (!contents_from ? path :
2343                      (!strcmp(contents_from, "-") ? "standard input" : contents_from)));
2344         set_commit_buffer_from_strbuf(commit, &msg);
2345
2346         if (!contents_from || strcmp("-", contents_from)) {
2347                 struct stat st;
2348                 const char *read_from;
2349                 char *buf_ptr;
2350                 unsigned long buf_len;
2351
2352                 if (contents_from) {
2353                         if (stat(contents_from, &st) < 0)
2354                                 die_errno("Cannot stat '%s'", contents_from);
2355                         read_from = contents_from;
2356                 }
2357                 else {
2358                         if (lstat(path, &st) < 0)
2359                                 die_errno("Cannot lstat '%s'", path);
2360                         read_from = path;
2361                 }
2362                 mode = canon_mode(st.st_mode);
2363
2364                 switch (st.st_mode & S_IFMT) {
2365                 case S_IFREG:
2366                         if (DIFF_OPT_TST(opt, ALLOW_TEXTCONV) &&
2367                             textconv_object(read_from, mode, &null_oid, 0, &buf_ptr, &buf_len))
2368                                 strbuf_attach(&buf, buf_ptr, buf_len, buf_len + 1);
2369                         else if (strbuf_read_file(&buf, read_from, st.st_size) != st.st_size)
2370                                 die_errno("cannot open or read '%s'", read_from);
2371                         break;
2372                 case S_IFLNK:
2373                         if (strbuf_readlink(&buf, read_from, st.st_size) < 0)
2374                                 die_errno("cannot readlink '%s'", read_from);
2375                         break;
2376                 default:
2377                         die("unsupported file type %s", read_from);
2378                 }
2379         }
2380         else {
2381                 /* Reading from stdin */
2382                 mode = 0;
2383                 if (strbuf_read(&buf, 0, 0) < 0)
2384                         die_errno("failed to read from stdin");
2385         }
2386         convert_to_git(path, buf.buf, buf.len, &buf, 0);
2387         origin->file.ptr = buf.buf;
2388         origin->file.size = buf.len;
2389         pretend_sha1_file(buf.buf, buf.len, OBJ_BLOB, origin->blob_oid.hash);
2390
2391         /*
2392          * Read the current index, replace the path entry with
2393          * origin->blob_sha1 without mucking with its mode or type
2394          * bits; we are not going to write this index out -- we just
2395          * want to run "diff-index --cached".
2396          */
2397         discard_cache();
2398         read_cache();
2399
2400         len = strlen(path);
2401         if (!mode) {
2402                 int pos = cache_name_pos(path, len);
2403                 if (0 <= pos)
2404                         mode = active_cache[pos]->ce_mode;
2405                 else
2406                         /* Let's not bother reading from HEAD tree */
2407                         mode = S_IFREG | 0644;
2408         }
2409         size = cache_entry_size(len);
2410         ce = xcalloc(1, size);
2411         oidcpy(&ce->oid, &origin->blob_oid);
2412         memcpy(ce->name, path, len);
2413         ce->ce_flags = create_ce_flags(0);
2414         ce->ce_namelen = len;
2415         ce->ce_mode = create_ce_mode(mode);
2416         add_cache_entry(ce, ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE);
2417
2418         cache_tree_invalidate_path(&the_index, path);
2419
2420         return commit;
2421 }
2422
2423 static struct commit *find_single_final(struct rev_info *revs,
2424                                         const char **name_p)
2425 {
2426         int i;
2427         struct commit *found = NULL;
2428         const char *name = NULL;
2429
2430         for (i = 0; i < revs->pending.nr; i++) {
2431                 struct object *obj = revs->pending.objects[i].item;
2432                 if (obj->flags & UNINTERESTING)
2433                         continue;
2434                 obj = deref_tag(obj, NULL, 0);
2435                 if (obj->type != OBJ_COMMIT)
2436                         die("Non commit %s?", revs->pending.objects[i].name);
2437                 if (found)
2438                         die("More than one commit to dig from %s and %s?",
2439                             revs->pending.objects[i].name, name);
2440                 found = (struct commit *)obj;
2441                 name = revs->pending.objects[i].name;
2442         }
2443         if (name_p)
2444                 *name_p = name;
2445         return found;
2446 }
2447
2448 static char *prepare_final(struct scoreboard *sb)
2449 {
2450         const char *name;
2451         sb->final = find_single_final(sb->revs, &name);
2452         return xstrdup_or_null(name);
2453 }
2454
2455 static const char *dwim_reverse_initial(struct scoreboard *sb)
2456 {
2457         /*
2458          * DWIM "git blame --reverse ONE -- PATH" as
2459          * "git blame --reverse ONE..HEAD -- PATH" but only do so
2460          * when it makes sense.
2461          */
2462         struct object *obj;
2463         struct commit *head_commit;
2464         unsigned char head_sha1[20];
2465
2466         if (sb->revs->pending.nr != 1)
2467                 return NULL;
2468
2469         /* Is that sole rev a committish? */
2470         obj = sb->revs->pending.objects[0].item;
2471         obj = deref_tag(obj, NULL, 0);
2472         if (obj->type != OBJ_COMMIT)
2473                 return NULL;
2474
2475         /* Do we have HEAD? */
2476         if (!resolve_ref_unsafe("HEAD", RESOLVE_REF_READING, head_sha1, NULL))
2477                 return NULL;
2478         head_commit = lookup_commit_reference_gently(head_sha1, 1);
2479         if (!head_commit)
2480                 return NULL;
2481
2482         /* Turn "ONE" into "ONE..HEAD" then */
2483         obj->flags |= UNINTERESTING;
2484         add_pending_object(sb->revs, &head_commit->object, "HEAD");
2485
2486         sb->final = (struct commit *)obj;
2487         return sb->revs->pending.objects[0].name;
2488 }
2489
2490 static char *prepare_initial(struct scoreboard *sb)
2491 {
2492         int i;
2493         const char *final_commit_name = NULL;
2494         struct rev_info *revs = sb->revs;
2495
2496         /*
2497          * There must be one and only one negative commit, and it must be
2498          * the boundary.
2499          */
2500         for (i = 0; i < revs->pending.nr; i++) {
2501                 struct object *obj = revs->pending.objects[i].item;
2502                 if (!(obj->flags & UNINTERESTING))
2503                         continue;
2504                 obj = deref_tag(obj, NULL, 0);
2505                 if (obj->type != OBJ_COMMIT)
2506                         die("Non commit %s?", revs->pending.objects[i].name);
2507                 if (sb->final)
2508                         die("More than one commit to dig up from, %s and %s?",
2509                             revs->pending.objects[i].name,
2510                             final_commit_name);
2511                 sb->final = (struct commit *) obj;
2512                 final_commit_name = revs->pending.objects[i].name;
2513         }
2514
2515         if (!final_commit_name)
2516                 final_commit_name = dwim_reverse_initial(sb);
2517         if (!final_commit_name)
2518                 die("No commit to dig up from?");
2519         return xstrdup(final_commit_name);
2520 }
2521
2522 static int blame_copy_callback(const struct option *option, const char *arg, int unset)
2523 {
2524         int *opt = option->value;
2525
2526         /*
2527          * -C enables copy from removed files;
2528          * -C -C enables copy from existing files, but only
2529          *       when blaming a new file;
2530          * -C -C -C enables copy from existing files for
2531          *          everybody
2532          */
2533         if (*opt & PICKAXE_BLAME_COPY_HARDER)
2534                 *opt |= PICKAXE_BLAME_COPY_HARDEST;
2535         if (*opt & PICKAXE_BLAME_COPY)
2536                 *opt |= PICKAXE_BLAME_COPY_HARDER;
2537         *opt |= PICKAXE_BLAME_COPY | PICKAXE_BLAME_MOVE;
2538
2539         if (arg)
2540                 blame_copy_score = parse_score(arg);
2541         return 0;
2542 }
2543
2544 static int blame_move_callback(const struct option *option, const char *arg, int unset)
2545 {
2546         int *opt = option->value;
2547
2548         *opt |= PICKAXE_BLAME_MOVE;
2549
2550         if (arg)
2551                 blame_move_score = parse_score(arg);
2552         return 0;
2553 }
2554
2555 int cmd_blame(int argc, const char **argv, const char *prefix)
2556 {
2557         struct rev_info revs;
2558         const char *path;
2559         struct scoreboard sb;
2560         struct origin *o;
2561         struct blame_entry *ent = NULL;
2562         long dashdash_pos, lno;
2563         char *final_commit_name = NULL;
2564         enum object_type type;
2565         struct commit *final_commit = NULL;
2566
2567         struct string_list range_list = STRING_LIST_INIT_NODUP;
2568         int output_option = 0, opt = 0;
2569         int show_stats = 0;
2570         const char *revs_file = NULL;
2571         const char *contents_from = NULL;
2572         const struct option options[] = {
2573                 OPT_BOOL(0, "incremental", &incremental, N_("Show blame entries as we find them, incrementally")),
2574                 OPT_BOOL('b', NULL, &blank_boundary, N_("Show blank SHA-1 for boundary commits (Default: off)")),
2575                 OPT_BOOL(0, "root", &show_root, N_("Do not treat root commits as boundaries (Default: off)")),
2576                 OPT_BOOL(0, "show-stats", &show_stats, N_("Show work cost statistics")),
2577                 OPT_BOOL(0, "progress", &show_progress, N_("Force progress reporting")),
2578                 OPT_BIT(0, "score-debug", &output_option, N_("Show output score for blame entries"), OUTPUT_SHOW_SCORE),
2579                 OPT_BIT('f', "show-name", &output_option, N_("Show original filename (Default: auto)"), OUTPUT_SHOW_NAME),
2580                 OPT_BIT('n', "show-number", &output_option, N_("Show original linenumber (Default: off)"), OUTPUT_SHOW_NUMBER),
2581                 OPT_BIT('p', "porcelain", &output_option, N_("Show in a format designed for machine consumption"), OUTPUT_PORCELAIN),
2582                 OPT_BIT(0, "line-porcelain", &output_option, N_("Show porcelain format with per-line commit information"), OUTPUT_PORCELAIN|OUTPUT_LINE_PORCELAIN),
2583                 OPT_BIT('c', NULL, &output_option, N_("Use the same output mode as git-annotate (Default: off)"), OUTPUT_ANNOTATE_COMPAT),
2584                 OPT_BIT('t', NULL, &output_option, N_("Show raw timestamp (Default: off)"), OUTPUT_RAW_TIMESTAMP),
2585                 OPT_BIT('l', NULL, &output_option, N_("Show long commit SHA1 (Default: off)"), OUTPUT_LONG_OBJECT_NAME),
2586                 OPT_BIT('s', NULL, &output_option, N_("Suppress author name and timestamp (Default: off)"), OUTPUT_NO_AUTHOR),
2587                 OPT_BIT('e', "show-email", &output_option, N_("Show author email instead of name (Default: off)"), OUTPUT_SHOW_EMAIL),
2588                 OPT_BIT('w', NULL, &xdl_opts, N_("Ignore whitespace differences"), XDF_IGNORE_WHITESPACE),
2589
2590                 /*
2591                  * The following two options are parsed by parse_revision_opt()
2592                  * and are only included here to get included in the "-h"
2593                  * output:
2594                  */
2595                 { OPTION_LOWLEVEL_CALLBACK, 0, "indent-heuristic", NULL, NULL, N_("Use an experimental heuristic to improve diffs"), PARSE_OPT_NOARG, parse_opt_unknown_cb },
2596
2597                 OPT_BIT(0, "minimal", &xdl_opts, N_("Spend extra cycles to find better match"), XDF_NEED_MINIMAL),
2598                 OPT_STRING('S', NULL, &revs_file, N_("file"), N_("Use revisions from <file> instead of calling git-rev-list")),
2599                 OPT_STRING(0, "contents", &contents_from, N_("file"), N_("Use <file>'s contents as the final image")),
2600                 { OPTION_CALLBACK, 'C', NULL, &opt, N_("score"), N_("Find line copies within and across files"), PARSE_OPT_OPTARG, blame_copy_callback },
2601                 { OPTION_CALLBACK, 'M', NULL, &opt, N_("score"), N_("Find line movements within and across files"), PARSE_OPT_OPTARG, blame_move_callback },
2602                 OPT_STRING_LIST('L', NULL, &range_list, N_("n,m"), N_("Process only line range n,m, counting from 1")),
2603                 OPT__ABBREV(&abbrev),
2604                 OPT_END()
2605         };
2606
2607         struct parse_opt_ctx_t ctx;
2608         int cmd_is_annotate = !strcmp(argv[0], "annotate");
2609         struct range_set ranges;
2610         unsigned int range_i;
2611         long anchor;
2612
2613         git_config(git_blame_config, &output_option);
2614         init_revisions(&revs, NULL);
2615         revs.date_mode = blame_date_mode;
2616         DIFF_OPT_SET(&revs.diffopt, ALLOW_TEXTCONV);
2617         DIFF_OPT_SET(&revs.diffopt, FOLLOW_RENAMES);
2618
2619         save_commit_buffer = 0;
2620         dashdash_pos = 0;
2621         show_progress = -1;
2622
2623         parse_options_start(&ctx, argc, argv, prefix, options,
2624                             PARSE_OPT_KEEP_DASHDASH | PARSE_OPT_KEEP_ARGV0);
2625         for (;;) {
2626                 switch (parse_options_step(&ctx, options, blame_opt_usage)) {
2627                 case PARSE_OPT_HELP:
2628                         exit(129);
2629                 case PARSE_OPT_DONE:
2630                         if (ctx.argv[0])
2631                                 dashdash_pos = ctx.cpidx;
2632                         goto parse_done;
2633                 }
2634
2635                 if (!strcmp(ctx.argv[0], "--reverse")) {
2636                         ctx.argv[0] = "--children";
2637                         reverse = 1;
2638                 }
2639                 parse_revision_opt(&revs, &ctx, options, blame_opt_usage);
2640         }
2641 parse_done:
2642         no_whole_file_rename = !DIFF_OPT_TST(&revs.diffopt, FOLLOW_RENAMES);
2643         xdl_opts |= revs.diffopt.xdl_opts & XDF_INDENT_HEURISTIC;
2644         DIFF_OPT_CLR(&revs.diffopt, FOLLOW_RENAMES);
2645         argc = parse_options_end(&ctx);
2646
2647         if (incremental || (output_option & OUTPUT_PORCELAIN)) {
2648                 if (show_progress > 0)
2649                         die(_("--progress can't be used with --incremental or porcelain formats"));
2650                 show_progress = 0;
2651         } else if (show_progress < 0)
2652                 show_progress = isatty(2);
2653
2654         if (0 < abbrev && abbrev < GIT_SHA1_HEXSZ)
2655                 /* one more abbrev length is needed for the boundary commit */
2656                 abbrev++;
2657         else if (!abbrev)
2658                 abbrev = GIT_SHA1_HEXSZ;
2659
2660         if (revs_file && read_ancestry(revs_file))
2661                 die_errno("reading graft file '%s' failed", revs_file);
2662
2663         if (cmd_is_annotate) {
2664                 output_option |= OUTPUT_ANNOTATE_COMPAT;
2665                 blame_date_mode.type = DATE_ISO8601;
2666         } else {
2667                 blame_date_mode = revs.date_mode;
2668         }
2669
2670         /* The maximum width used to show the dates */
2671         switch (blame_date_mode.type) {
2672         case DATE_RFC2822:
2673                 blame_date_width = sizeof("Thu, 19 Oct 2006 16:00:04 -0700");
2674                 break;
2675         case DATE_ISO8601_STRICT:
2676                 blame_date_width = sizeof("2006-10-19T16:00:04-07:00");
2677                 break;
2678         case DATE_ISO8601:
2679                 blame_date_width = sizeof("2006-10-19 16:00:04 -0700");
2680                 break;
2681         case DATE_RAW:
2682                 blame_date_width = sizeof("1161298804 -0700");
2683                 break;
2684         case DATE_UNIX:
2685                 blame_date_width = sizeof("1161298804");
2686                 break;
2687         case DATE_SHORT:
2688                 blame_date_width = sizeof("2006-10-19");
2689                 break;
2690         case DATE_RELATIVE:
2691                 /* TRANSLATORS: This string is used to tell us the maximum
2692                    display width for a relative timestamp in "git blame"
2693                    output.  For C locale, "4 years, 11 months ago", which
2694                    takes 22 places, is the longest among various forms of
2695                    relative timestamps, but your language may need more or
2696                    fewer display columns. */
2697                 blame_date_width = utf8_strwidth(_("4 years, 11 months ago")) + 1; /* add the null */
2698                 break;
2699         case DATE_NORMAL:
2700                 blame_date_width = sizeof("Thu Oct 19 16:00:04 2006 -0700");
2701                 break;
2702         case DATE_STRFTIME:
2703                 blame_date_width = strlen(show_date(0, 0, &blame_date_mode)) + 1; /* add the null */
2704                 break;
2705         }
2706         blame_date_width -= 1; /* strip the null */
2707
2708         if (DIFF_OPT_TST(&revs.diffopt, FIND_COPIES_HARDER))
2709                 opt |= (PICKAXE_BLAME_COPY | PICKAXE_BLAME_MOVE |
2710                         PICKAXE_BLAME_COPY_HARDER);
2711
2712         if (!blame_move_score)
2713                 blame_move_score = BLAME_DEFAULT_MOVE_SCORE;
2714         if (!blame_copy_score)
2715                 blame_copy_score = BLAME_DEFAULT_COPY_SCORE;
2716
2717         /*
2718          * We have collected options unknown to us in argv[1..unk]
2719          * which are to be passed to revision machinery if we are
2720          * going to do the "bottom" processing.
2721          *
2722          * The remaining are:
2723          *
2724          * (1) if dashdash_pos != 0, it is either
2725          *     "blame [revisions] -- <path>" or
2726          *     "blame -- <path> <rev>"
2727          *
2728          * (2) otherwise, it is one of the two:
2729          *     "blame [revisions] <path>"
2730          *     "blame <path> <rev>"
2731          *
2732          * Note that we must strip out <path> from the arguments: we do not
2733          * want the path pruning but we may want "bottom" processing.
2734          */
2735         if (dashdash_pos) {
2736                 switch (argc - dashdash_pos - 1) {
2737                 case 2: /* (1b) */
2738                         if (argc != 4)
2739                                 usage_with_options(blame_opt_usage, options);
2740                         /* reorder for the new way: <rev> -- <path> */
2741                         argv[1] = argv[3];
2742                         argv[3] = argv[2];
2743                         argv[2] = "--";
2744                         /* FALLTHROUGH */
2745                 case 1: /* (1a) */
2746                         path = add_prefix(prefix, argv[--argc]);
2747                         argv[argc] = NULL;
2748                         break;
2749                 default:
2750                         usage_with_options(blame_opt_usage, options);
2751                 }
2752         } else {
2753                 if (argc < 2)
2754                         usage_with_options(blame_opt_usage, options);
2755                 path = add_prefix(prefix, argv[argc - 1]);
2756                 if (argc == 3 && !file_exists(path)) { /* (2b) */
2757                         path = add_prefix(prefix, argv[1]);
2758                         argv[1] = argv[2];
2759                 }
2760                 argv[argc - 1] = "--";
2761
2762                 setup_work_tree();
2763                 if (!file_exists(path))
2764                         die_errno("cannot stat path '%s'", path);
2765         }
2766
2767         revs.disable_stdin = 1;
2768         setup_revisions(argc, argv, &revs, NULL);
2769         memset(&sb, 0, sizeof(sb));
2770
2771         sb.revs = &revs;
2772         if (!reverse) {
2773                 final_commit_name = prepare_final(&sb);
2774                 sb.commits.compare = compare_commits_by_commit_date;
2775         }
2776         else if (contents_from)
2777                 die(_("--contents and --reverse do not blend well."));
2778         else {
2779                 final_commit_name = prepare_initial(&sb);
2780                 sb.commits.compare = compare_commits_by_reverse_commit_date;
2781                 if (revs.first_parent_only)
2782                         revs.children.name = NULL;
2783         }
2784
2785         if (!sb.final) {
2786                 /*
2787                  * "--not A B -- path" without anything positive;
2788                  * do not default to HEAD, but use the working tree
2789                  * or "--contents".
2790                  */
2791                 setup_work_tree();
2792                 sb.final = fake_working_tree_commit(&sb.revs->diffopt,
2793                                                     path, contents_from);
2794                 add_pending_object(&revs, &(sb.final->object), ":");
2795         }
2796         else if (contents_from)
2797                 die(_("cannot use --contents with final commit object name"));
2798
2799         if (reverse && revs.first_parent_only) {
2800                 final_commit = find_single_final(sb.revs, NULL);
2801                 if (!final_commit)
2802                         die(_("--reverse and --first-parent together require specified latest commit"));
2803         }
2804
2805         /*
2806          * If we have bottom, this will mark the ancestors of the
2807          * bottom commits we would reach while traversing as
2808          * uninteresting.
2809          */
2810         if (prepare_revision_walk(&revs))
2811                 die(_("revision walk setup failed"));
2812
2813         if (reverse && revs.first_parent_only) {
2814                 struct commit *c = final_commit;
2815
2816                 sb.revs->children.name = "children";
2817                 while (c->parents &&
2818                        oidcmp(&c->object.oid, &sb.final->object.oid)) {
2819                         struct commit_list *l = xcalloc(1, sizeof(*l));
2820
2821                         l->item = c;
2822                         if (add_decoration(&sb.revs->children,
2823                                            &c->parents->item->object, l))
2824                                 die("BUG: not unique item in first-parent chain");
2825                         c = c->parents->item;
2826                 }
2827
2828                 if (oidcmp(&c->object.oid, &sb.final->object.oid))
2829                         die(_("--reverse --first-parent together require range along first-parent chain"));
2830         }
2831
2832         if (is_null_oid(&sb.final->object.oid)) {
2833                 o = sb.final->util;
2834                 sb.final_buf = xmemdupz(o->file.ptr, o->file.size);
2835                 sb.final_buf_size = o->file.size;
2836         }
2837         else {
2838                 o = get_origin(&sb, sb.final, path);
2839                 if (fill_blob_sha1_and_mode(o))
2840                         die(_("no such path %s in %s"), path, final_commit_name);
2841
2842                 if (DIFF_OPT_TST(&sb.revs->diffopt, ALLOW_TEXTCONV) &&
2843                     textconv_object(path, o->mode, &o->blob_oid, 1, (char **) &sb.final_buf,
2844                                     &sb.final_buf_size))
2845                         ;
2846                 else
2847                         sb.final_buf = read_sha1_file(o->blob_oid.hash, &type,
2848                                                       &sb.final_buf_size);
2849
2850                 if (!sb.final_buf)
2851                         die(_("cannot read blob %s for path %s"),
2852                             oid_to_hex(&o->blob_oid),
2853                             path);
2854         }
2855         num_read_blob++;
2856         lno = prepare_lines(&sb);
2857
2858         if (lno && !range_list.nr)
2859                 string_list_append(&range_list, "1");
2860
2861         anchor = 1;
2862         range_set_init(&ranges, range_list.nr);
2863         for (range_i = 0; range_i < range_list.nr; ++range_i) {
2864                 long bottom, top;
2865                 if (parse_range_arg(range_list.items[range_i].string,
2866                                     nth_line_cb, &sb, lno, anchor,
2867                                     &bottom, &top, sb.path))
2868                         usage(blame_usage);
2869                 if (lno < top || ((lno || bottom) && lno < bottom))
2870                         die(Q_("file %s has only %lu line",
2871                                "file %s has only %lu lines",
2872                                lno), path, lno);
2873                 if (bottom < 1)
2874                         bottom = 1;
2875                 if (top < 1)
2876                         top = lno;
2877                 bottom--;
2878                 range_set_append_unsafe(&ranges, bottom, top);
2879                 anchor = top + 1;
2880         }
2881         sort_and_merge_range_set(&ranges);
2882
2883         for (range_i = ranges.nr; range_i > 0; --range_i) {
2884                 const struct range *r = &ranges.ranges[range_i - 1];
2885                 long bottom = r->start;
2886                 long top = r->end;
2887                 struct blame_entry *next = ent;
2888                 ent = xcalloc(1, sizeof(*ent));
2889                 ent->lno = bottom;
2890                 ent->num_lines = top - bottom;
2891                 ent->suspect = o;
2892                 ent->s_lno = bottom;
2893                 ent->next = next;
2894                 origin_incref(o);
2895         }
2896
2897         o->suspects = ent;
2898         prio_queue_put(&sb.commits, o->commit);
2899
2900         origin_decref(o);
2901
2902         range_set_release(&ranges);
2903         string_list_clear(&range_list, 0);
2904
2905         sb.ent = NULL;
2906         sb.path = path;
2907
2908         read_mailmap(&mailmap, NULL);
2909
2910         assign_blame(&sb, opt);
2911
2912         if (!incremental)
2913                 setup_pager();
2914
2915         free(final_commit_name);
2916
2917         if (incremental)
2918                 return 0;
2919
2920         sb.ent = blame_sort(sb.ent, compare_blame_final);
2921
2922         coalesce(&sb);
2923
2924         if (!(output_option & OUTPUT_PORCELAIN))
2925                 find_alignment(&sb, &output_option);
2926
2927         output(&sb, output_option);
2928         free((void *)sb.final_buf);
2929         for (ent = sb.ent; ent; ) {
2930                 struct blame_entry *e = ent->next;
2931                 free(ent);
2932                 ent = e;
2933         }
2934
2935         if (show_stats) {
2936                 printf("num read blob: %d\n", num_read_blob);
2937                 printf("num get patch: %d\n", num_get_patch);
2938                 printf("num commits: %d\n", num_commits);
2939         }
2940         return 0;
2941 }