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