Merge branch 'jk/t0008-sigpipe-fix'
[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 #include "line-range.h"
25
26 static char blame_usage[] = N_("git blame [options] [rev-opts] [rev] [--] file");
27
28 static const char *blame_opt_usage[] = {
29         blame_usage,
30         "",
31         N_("[rev-opts] are documented in git-rev-list(1)"),
32         NULL
33 };
34
35 static int longest_file;
36 static int longest_author;
37 static int max_orig_digits;
38 static int max_digits;
39 static int max_score_digits;
40 static int show_root;
41 static int reverse;
42 static int blank_boundary;
43 static int incremental;
44 static int xdl_opts;
45 static int abbrev = -1;
46 static int no_whole_file_rename;
47
48 static enum date_mode blame_date_mode = DATE_ISO8601;
49 static size_t blame_date_width;
50
51 static struct string_list mailmap;
52
53 #ifndef DEBUG
54 #define DEBUG 0
55 #endif
56
57 /* stats */
58 static int num_read_blob;
59 static int num_get_patch;
60 static int num_commits;
61
62 #define PICKAXE_BLAME_MOVE              01
63 #define PICKAXE_BLAME_COPY              02
64 #define PICKAXE_BLAME_COPY_HARDER       04
65 #define PICKAXE_BLAME_COPY_HARDEST      010
66
67 /*
68  * blame for a blame_entry with score lower than these thresholds
69  * is not passed to the parent using move/copy logic.
70  */
71 static unsigned blame_move_score;
72 static unsigned blame_copy_score;
73 #define BLAME_DEFAULT_MOVE_SCORE        20
74 #define BLAME_DEFAULT_COPY_SCORE        40
75
76 /* bits #0..7 in revision.h, #8..11 used for merge_bases() in commit.c */
77 #define METAINFO_SHOWN          (1u<<12)
78 #define MORE_THAN_ONE_PATH      (1u<<13)
79
80 /*
81  * One blob in a commit that is being suspected
82  */
83 struct origin {
84         int refcnt;
85         struct origin *previous;
86         struct commit *commit;
87         mmfile_t file;
88         unsigned char blob_sha1[20];
89         unsigned mode;
90         char path[FLEX_ARRAY];
91 };
92
93 static int diff_hunks(mmfile_t *file_a, mmfile_t *file_b, long ctxlen,
94                       xdl_emit_hunk_consume_func_t hunk_func, void *cb_data)
95 {
96         xpparam_t xpp = {0};
97         xdemitconf_t xecfg = {0};
98         xdemitcb_t ecb = {NULL};
99
100         xpp.flags = xdl_opts;
101         xecfg.ctxlen = ctxlen;
102         xecfg.hunk_func = hunk_func;
103         ecb.priv = cb_data;
104         return xdi_diff(file_a, file_b, &xpp, &xecfg, &ecb);
105 }
106
107 /*
108  * Prepare diff_filespec and convert it using diff textconv API
109  * if the textconv driver exists.
110  * Return 1 if the conversion succeeds, 0 otherwise.
111  */
112 int textconv_object(const char *path,
113                     unsigned mode,
114                     const unsigned char *sha1,
115                     int sha1_valid,
116                     char **buf,
117                     unsigned long *buf_size)
118 {
119         struct diff_filespec *df;
120         struct userdiff_driver *textconv;
121
122         df = alloc_filespec(path);
123         fill_filespec(df, sha1, sha1_valid, mode);
124         textconv = get_textconv(df);
125         if (!textconv) {
126                 free_filespec(df);
127                 return 0;
128         }
129
130         *buf_size = fill_textconv(textconv, df, buf);
131         free_filespec(df);
132         return 1;
133 }
134
135 /*
136  * Given an origin, prepare mmfile_t structure to be used by the
137  * diff machinery
138  */
139 static void fill_origin_blob(struct diff_options *opt,
140                              struct origin *o, mmfile_t *file)
141 {
142         if (!o->file.ptr) {
143                 enum object_type type;
144                 unsigned long file_size;
145
146                 num_read_blob++;
147                 if (DIFF_OPT_TST(opt, ALLOW_TEXTCONV) &&
148                     textconv_object(o->path, o->mode, o->blob_sha1, 1, &file->ptr, &file_size))
149                         ;
150                 else
151                         file->ptr = read_sha1_file(o->blob_sha1, &type, &file_size);
152                 file->size = file_size;
153
154                 if (!file->ptr)
155                         die("Cannot read blob %s for path %s",
156                             sha1_to_hex(o->blob_sha1),
157                             o->path);
158                 o->file = *file;
159         }
160         else
161                 *file = o->file;
162 }
163
164 /*
165  * Origin is refcounted and usually we keep the blob contents to be
166  * reused.
167  */
168 static inline struct origin *origin_incref(struct origin *o)
169 {
170         if (o)
171                 o->refcnt++;
172         return o;
173 }
174
175 static void origin_decref(struct origin *o)
176 {
177         if (o && --o->refcnt <= 0) {
178                 if (o->previous)
179                         origin_decref(o->previous);
180                 free(o->file.ptr);
181                 free(o);
182         }
183 }
184
185 static void drop_origin_blob(struct origin *o)
186 {
187         if (o->file.ptr) {
188                 free(o->file.ptr);
189                 o->file.ptr = NULL;
190         }
191 }
192
193 /*
194  * Each group of lines is described by a blame_entry; it can be split
195  * as we pass blame to the parents.  They form a linked list in the
196  * scoreboard structure, sorted by the target line number.
197  */
198 struct blame_entry {
199         struct blame_entry *prev;
200         struct blame_entry *next;
201
202         /* the first line of this group in the final image;
203          * internally all line numbers are 0 based.
204          */
205         int lno;
206
207         /* how many lines this group has */
208         int num_lines;
209
210         /* the commit that introduced this group into the final image */
211         struct origin *suspect;
212
213         /* true if the suspect is truly guilty; false while we have not
214          * checked if the group came from one of its parents.
215          */
216         char guilty;
217
218         /* true if the entry has been scanned for copies in the current parent
219          */
220         char scanned;
221
222         /* the line number of the first line of this group in the
223          * suspect's file; internally all line numbers are 0 based.
224          */
225         int s_lno;
226
227         /* how significant this entry is -- cached to avoid
228          * scanning the lines over and over.
229          */
230         unsigned score;
231 };
232
233 /*
234  * The current state of the blame assignment.
235  */
236 struct scoreboard {
237         /* the final commit (i.e. where we started digging from) */
238         struct commit *final;
239         struct rev_info *revs;
240         const char *path;
241
242         /*
243          * The contents in the final image.
244          * Used by many functions to obtain contents of the nth line,
245          * indexed with scoreboard.lineno[blame_entry.lno].
246          */
247         const char *final_buf;
248         unsigned long final_buf_size;
249
250         /* linked list of blames */
251         struct blame_entry *ent;
252
253         /* look-up a line in the final buffer */
254         int num_lines;
255         int *lineno;
256 };
257
258 static inline int same_suspect(struct origin *a, struct origin *b)
259 {
260         if (a == b)
261                 return 1;
262         if (a->commit != b->commit)
263                 return 0;
264         return !strcmp(a->path, b->path);
265 }
266
267 static void sanity_check_refcnt(struct scoreboard *);
268
269 /*
270  * If two blame entries that are next to each other came from
271  * contiguous lines in the same origin (i.e. <commit, path> pair),
272  * merge them together.
273  */
274 static void coalesce(struct scoreboard *sb)
275 {
276         struct blame_entry *ent, *next;
277
278         for (ent = sb->ent; ent && (next = ent->next); ent = next) {
279                 if (same_suspect(ent->suspect, next->suspect) &&
280                     ent->guilty == next->guilty &&
281                     ent->s_lno + ent->num_lines == next->s_lno) {
282                         ent->num_lines += next->num_lines;
283                         ent->next = next->next;
284                         if (ent->next)
285                                 ent->next->prev = ent;
286                         origin_decref(next->suspect);
287                         free(next);
288                         ent->score = 0;
289                         next = ent; /* again */
290                 }
291         }
292
293         if (DEBUG) /* sanity */
294                 sanity_check_refcnt(sb);
295 }
296
297 /*
298  * Given a commit and a path in it, create a new origin structure.
299  * The callers that add blame to the scoreboard should use
300  * get_origin() to obtain shared, refcounted copy instead of calling
301  * this function directly.
302  */
303 static struct origin *make_origin(struct commit *commit, const char *path)
304 {
305         struct origin *o;
306         o = xcalloc(1, sizeof(*o) + strlen(path) + 1);
307         o->commit = commit;
308         o->refcnt = 1;
309         strcpy(o->path, path);
310         return o;
311 }
312
313 /*
314  * Locate an existing origin or create a new one.
315  */
316 static struct origin *get_origin(struct scoreboard *sb,
317                                  struct commit *commit,
318                                  const char *path)
319 {
320         struct blame_entry *e;
321
322         for (e = sb->ent; e; e = e->next) {
323                 if (e->suspect->commit == commit &&
324                     !strcmp(e->suspect->path, path))
325                         return origin_incref(e->suspect);
326         }
327         return make_origin(commit, path);
328 }
329
330 /*
331  * Fill the blob_sha1 field of an origin if it hasn't, so that later
332  * call to fill_origin_blob() can use it to locate the data.  blob_sha1
333  * for an origin is also used to pass the blame for the entire file to
334  * the parent to detect the case where a child's blob is identical to
335  * that of its parent's.
336  *
337  * This also fills origin->mode for corresponding tree path.
338  */
339 static int fill_blob_sha1_and_mode(struct origin *origin)
340 {
341         if (!is_null_sha1(origin->blob_sha1))
342                 return 0;
343         if (get_tree_entry(origin->commit->object.sha1,
344                            origin->path,
345                            origin->blob_sha1, &origin->mode))
346                 goto error_out;
347         if (sha1_object_info(origin->blob_sha1, NULL) != OBJ_BLOB)
348                 goto error_out;
349         return 0;
350  error_out:
351         hashclr(origin->blob_sha1);
352         origin->mode = S_IFINVALID;
353         return -1;
354 }
355
356 /*
357  * We have an origin -- check if the same path exists in the
358  * parent and return an origin structure to represent it.
359  */
360 static struct origin *find_origin(struct scoreboard *sb,
361                                   struct commit *parent,
362                                   struct origin *origin)
363 {
364         struct origin *porigin = NULL;
365         struct diff_options diff_opts;
366         const char *paths[2];
367
368         if (parent->util) {
369                 /*
370                  * Each commit object can cache one origin in that
371                  * commit.  This is a freestanding copy of origin and
372                  * not refcounted.
373                  */
374                 struct origin *cached = parent->util;
375                 if (!strcmp(cached->path, origin->path)) {
376                         /*
377                          * The same path between origin and its parent
378                          * without renaming -- the most common case.
379                          */
380                         porigin = get_origin(sb, parent, cached->path);
381
382                         /*
383                          * If the origin was newly created (i.e. get_origin
384                          * would call make_origin if none is found in the
385                          * scoreboard), it does not know the blob_sha1/mode,
386                          * so copy it.  Otherwise porigin was in the
387                          * scoreboard and already knows blob_sha1/mode.
388                          */
389                         if (porigin->refcnt == 1) {
390                                 hashcpy(porigin->blob_sha1, cached->blob_sha1);
391                                 porigin->mode = cached->mode;
392                         }
393                         return porigin;
394                 }
395                 /* otherwise it was not very useful; free it */
396                 free(parent->util);
397                 parent->util = NULL;
398         }
399
400         /* See if the origin->path is different between parent
401          * and origin first.  Most of the time they are the
402          * same and diff-tree is fairly efficient about this.
403          */
404         diff_setup(&diff_opts);
405         DIFF_OPT_SET(&diff_opts, RECURSIVE);
406         diff_opts.detect_rename = 0;
407         diff_opts.output_format = DIFF_FORMAT_NO_OUTPUT;
408         paths[0] = origin->path;
409         paths[1] = NULL;
410
411         diff_tree_setup_paths(paths, &diff_opts);
412         diff_setup_done(&diff_opts);
413
414         if (is_null_sha1(origin->commit->object.sha1))
415                 do_diff_cache(parent->tree->object.sha1, &diff_opts);
416         else
417                 diff_tree_sha1(parent->tree->object.sha1,
418                                origin->commit->tree->object.sha1,
419                                "", &diff_opts);
420         diffcore_std(&diff_opts);
421
422         if (!diff_queued_diff.nr) {
423                 /* The path is the same as parent */
424                 porigin = get_origin(sb, parent, origin->path);
425                 hashcpy(porigin->blob_sha1, origin->blob_sha1);
426                 porigin->mode = origin->mode;
427         } else {
428                 /*
429                  * Since origin->path is a pathspec, if the parent
430                  * commit had it as a directory, we will see a whole
431                  * bunch of deletion of files in the directory that we
432                  * do not care about.
433                  */
434                 int i;
435                 struct diff_filepair *p = NULL;
436                 for (i = 0; i < diff_queued_diff.nr; i++) {
437                         const char *name;
438                         p = diff_queued_diff.queue[i];
439                         name = p->one->path ? p->one->path : p->two->path;
440                         if (!strcmp(name, origin->path))
441                                 break;
442                 }
443                 if (!p)
444                         die("internal error in blame::find_origin");
445                 switch (p->status) {
446                 default:
447                         die("internal error in blame::find_origin (%c)",
448                             p->status);
449                 case 'M':
450                         porigin = get_origin(sb, parent, origin->path);
451                         hashcpy(porigin->blob_sha1, p->one->sha1);
452                         porigin->mode = p->one->mode;
453                         break;
454                 case 'A':
455                 case 'T':
456                         /* Did not exist in parent, or type changed */
457                         break;
458                 }
459         }
460         diff_flush(&diff_opts);
461         diff_tree_release_paths(&diff_opts);
462         if (porigin) {
463                 /*
464                  * Create a freestanding copy that is not part of
465                  * the refcounted origin found in the scoreboard, and
466                  * cache it in the commit.
467                  */
468                 struct origin *cached;
469
470                 cached = make_origin(porigin->commit, porigin->path);
471                 hashcpy(cached->blob_sha1, porigin->blob_sha1);
472                 cached->mode = porigin->mode;
473                 parent->util = cached;
474         }
475         return porigin;
476 }
477
478 /*
479  * We have an origin -- find the path that corresponds to it in its
480  * parent and return an origin structure to represent it.
481  */
482 static struct origin *find_rename(struct scoreboard *sb,
483                                   struct commit *parent,
484                                   struct origin *origin)
485 {
486         struct origin *porigin = NULL;
487         struct diff_options diff_opts;
488         int i;
489         const char *paths[2];
490
491         diff_setup(&diff_opts);
492         DIFF_OPT_SET(&diff_opts, RECURSIVE);
493         diff_opts.detect_rename = DIFF_DETECT_RENAME;
494         diff_opts.output_format = DIFF_FORMAT_NO_OUTPUT;
495         diff_opts.single_follow = origin->path;
496         paths[0] = NULL;
497         diff_tree_setup_paths(paths, &diff_opts);
498         diff_setup_done(&diff_opts);
499
500         if (is_null_sha1(origin->commit->object.sha1))
501                 do_diff_cache(parent->tree->object.sha1, &diff_opts);
502         else
503                 diff_tree_sha1(parent->tree->object.sha1,
504                                origin->commit->tree->object.sha1,
505                                "", &diff_opts);
506         diffcore_std(&diff_opts);
507
508         for (i = 0; i < diff_queued_diff.nr; i++) {
509                 struct diff_filepair *p = diff_queued_diff.queue[i];
510                 if ((p->status == 'R' || p->status == 'C') &&
511                     !strcmp(p->two->path, origin->path)) {
512                         porigin = get_origin(sb, parent, p->one->path);
513                         hashcpy(porigin->blob_sha1, p->one->sha1);
514                         porigin->mode = p->one->mode;
515                         break;
516                 }
517         }
518         diff_flush(&diff_opts);
519         diff_tree_release_paths(&diff_opts);
520         return porigin;
521 }
522
523 /*
524  * Link in a new blame entry to the scoreboard.  Entries that cover the
525  * same line range have been removed from the scoreboard previously.
526  */
527 static void add_blame_entry(struct scoreboard *sb, struct blame_entry *e)
528 {
529         struct blame_entry *ent, *prev = NULL;
530
531         origin_incref(e->suspect);
532
533         for (ent = sb->ent; ent && ent->lno < e->lno; ent = ent->next)
534                 prev = ent;
535
536         /* prev, if not NULL, is the last one that is below e */
537         e->prev = prev;
538         if (prev) {
539                 e->next = prev->next;
540                 prev->next = e;
541         }
542         else {
543                 e->next = sb->ent;
544                 sb->ent = e;
545         }
546         if (e->next)
547                 e->next->prev = e;
548 }
549
550 /*
551  * src typically is on-stack; we want to copy the information in it to
552  * a malloced blame_entry that is already on the linked list of the
553  * scoreboard.  The origin of dst loses a refcnt while the origin of src
554  * gains one.
555  */
556 static void dup_entry(struct blame_entry *dst, struct blame_entry *src)
557 {
558         struct blame_entry *p, *n;
559
560         p = dst->prev;
561         n = dst->next;
562         origin_incref(src->suspect);
563         origin_decref(dst->suspect);
564         memcpy(dst, src, sizeof(*src));
565         dst->prev = p;
566         dst->next = n;
567         dst->score = 0;
568 }
569
570 static const char *nth_line(struct scoreboard *sb, long lno)
571 {
572         return sb->final_buf + sb->lineno[lno];
573 }
574
575 static const char *nth_line_cb(void *data, long lno)
576 {
577         return nth_line((struct scoreboard *)data, lno);
578 }
579
580 /*
581  * It is known that lines between tlno to same came from parent, and e
582  * has an overlap with that range.  it also is known that parent's
583  * line plno corresponds to e's line tlno.
584  *
585  *                <---- e ----->
586  *                   <------>
587  *                   <------------>
588  *             <------------>
589  *             <------------------>
590  *
591  * Split e into potentially three parts; before this chunk, the chunk
592  * to be blamed for the parent, and after that portion.
593  */
594 static void split_overlap(struct blame_entry *split,
595                           struct blame_entry *e,
596                           int tlno, int plno, int same,
597                           struct origin *parent)
598 {
599         int chunk_end_lno;
600         memset(split, 0, sizeof(struct blame_entry [3]));
601
602         if (e->s_lno < tlno) {
603                 /* there is a pre-chunk part not blamed on parent */
604                 split[0].suspect = origin_incref(e->suspect);
605                 split[0].lno = e->lno;
606                 split[0].s_lno = e->s_lno;
607                 split[0].num_lines = tlno - e->s_lno;
608                 split[1].lno = e->lno + tlno - e->s_lno;
609                 split[1].s_lno = plno;
610         }
611         else {
612                 split[1].lno = e->lno;
613                 split[1].s_lno = plno + (e->s_lno - tlno);
614         }
615
616         if (same < e->s_lno + e->num_lines) {
617                 /* there is a post-chunk part not blamed on parent */
618                 split[2].suspect = origin_incref(e->suspect);
619                 split[2].lno = e->lno + (same - e->s_lno);
620                 split[2].s_lno = e->s_lno + (same - e->s_lno);
621                 split[2].num_lines = e->s_lno + e->num_lines - same;
622                 chunk_end_lno = split[2].lno;
623         }
624         else
625                 chunk_end_lno = e->lno + e->num_lines;
626         split[1].num_lines = chunk_end_lno - split[1].lno;
627
628         /*
629          * if it turns out there is nothing to blame the parent for,
630          * forget about the splitting.  !split[1].suspect signals this.
631          */
632         if (split[1].num_lines < 1)
633                 return;
634         split[1].suspect = origin_incref(parent);
635 }
636
637 /*
638  * split_overlap() divided an existing blame e into up to three parts
639  * in split.  Adjust the linked list of blames in the scoreboard to
640  * reflect the split.
641  */
642 static void split_blame(struct scoreboard *sb,
643                         struct blame_entry *split,
644                         struct blame_entry *e)
645 {
646         struct blame_entry *new_entry;
647
648         if (split[0].suspect && split[2].suspect) {
649                 /* The first part (reuse storage for the existing entry e) */
650                 dup_entry(e, &split[0]);
651
652                 /* The last part -- me */
653                 new_entry = xmalloc(sizeof(*new_entry));
654                 memcpy(new_entry, &(split[2]), sizeof(struct blame_entry));
655                 add_blame_entry(sb, new_entry);
656
657                 /* ... and the middle part -- parent */
658                 new_entry = xmalloc(sizeof(*new_entry));
659                 memcpy(new_entry, &(split[1]), sizeof(struct blame_entry));
660                 add_blame_entry(sb, new_entry);
661         }
662         else if (!split[0].suspect && !split[2].suspect)
663                 /*
664                  * The parent covers the entire area; reuse storage for
665                  * e and replace it with the parent.
666                  */
667                 dup_entry(e, &split[1]);
668         else if (split[0].suspect) {
669                 /* me and then parent */
670                 dup_entry(e, &split[0]);
671
672                 new_entry = xmalloc(sizeof(*new_entry));
673                 memcpy(new_entry, &(split[1]), sizeof(struct blame_entry));
674                 add_blame_entry(sb, new_entry);
675         }
676         else {
677                 /* parent and then me */
678                 dup_entry(e, &split[1]);
679
680                 new_entry = xmalloc(sizeof(*new_entry));
681                 memcpy(new_entry, &(split[2]), sizeof(struct blame_entry));
682                 add_blame_entry(sb, new_entry);
683         }
684
685         if (DEBUG) { /* sanity */
686                 struct blame_entry *ent;
687                 int lno = sb->ent->lno, corrupt = 0;
688
689                 for (ent = sb->ent; ent; ent = ent->next) {
690                         if (lno != ent->lno)
691                                 corrupt = 1;
692                         if (ent->s_lno < 0)
693                                 corrupt = 1;
694                         lno += ent->num_lines;
695                 }
696                 if (corrupt) {
697                         lno = sb->ent->lno;
698                         for (ent = sb->ent; ent; ent = ent->next) {
699                                 printf("L %8d l %8d n %8d\n",
700                                        lno, ent->lno, ent->num_lines);
701                                 lno = ent->lno + ent->num_lines;
702                         }
703                         die("oops");
704                 }
705         }
706 }
707
708 /*
709  * After splitting the blame, the origins used by the
710  * on-stack blame_entry should lose one refcnt each.
711  */
712 static void decref_split(struct blame_entry *split)
713 {
714         int i;
715
716         for (i = 0; i < 3; i++)
717                 origin_decref(split[i].suspect);
718 }
719
720 /*
721  * Helper for blame_chunk().  blame_entry e is known to overlap with
722  * the patch hunk; split it and pass blame to the parent.
723  */
724 static void blame_overlap(struct scoreboard *sb, struct blame_entry *e,
725                           int tlno, int plno, int same,
726                           struct origin *parent)
727 {
728         struct blame_entry split[3];
729
730         split_overlap(split, e, tlno, plno, same, parent);
731         if (split[1].suspect)
732                 split_blame(sb, split, e);
733         decref_split(split);
734 }
735
736 /*
737  * Find the line number of the last line the target is suspected for.
738  */
739 static int find_last_in_target(struct scoreboard *sb, struct origin *target)
740 {
741         struct blame_entry *e;
742         int last_in_target = -1;
743
744         for (e = sb->ent; e; e = e->next) {
745                 if (e->guilty || !same_suspect(e->suspect, target))
746                         continue;
747                 if (last_in_target < e->s_lno + e->num_lines)
748                         last_in_target = e->s_lno + e->num_lines;
749         }
750         return last_in_target;
751 }
752
753 /*
754  * Process one hunk from the patch between the current suspect for
755  * blame_entry e and its parent.  Find and split the overlap, and
756  * pass blame to the overlapping part to the parent.
757  */
758 static void blame_chunk(struct scoreboard *sb,
759                         int tlno, int plno, int same,
760                         struct origin *target, struct origin *parent)
761 {
762         struct blame_entry *e;
763
764         for (e = sb->ent; e; e = e->next) {
765                 if (e->guilty || !same_suspect(e->suspect, target))
766                         continue;
767                 if (same <= e->s_lno)
768                         continue;
769                 if (tlno < e->s_lno + e->num_lines)
770                         blame_overlap(sb, e, tlno, plno, same, parent);
771         }
772 }
773
774 struct blame_chunk_cb_data {
775         struct scoreboard *sb;
776         struct origin *target;
777         struct origin *parent;
778         long plno;
779         long tlno;
780 };
781
782 static int blame_chunk_cb(long start_a, long count_a,
783                           long start_b, long count_b, void *data)
784 {
785         struct blame_chunk_cb_data *d = data;
786         blame_chunk(d->sb, d->tlno, d->plno, start_b, d->target, d->parent);
787         d->plno = start_a + count_a;
788         d->tlno = start_b + count_b;
789         return 0;
790 }
791
792 /*
793  * We are looking at the origin 'target' and aiming to pass blame
794  * for the lines it is suspected to its parent.  Run diff to find
795  * which lines came from parent and pass blame for them.
796  */
797 static int pass_blame_to_parent(struct scoreboard *sb,
798                                 struct origin *target,
799                                 struct origin *parent)
800 {
801         int last_in_target;
802         mmfile_t file_p, file_o;
803         struct blame_chunk_cb_data d;
804
805         memset(&d, 0, sizeof(d));
806         d.sb = sb; d.target = target; d.parent = parent;
807         last_in_target = find_last_in_target(sb, target);
808         if (last_in_target < 0)
809                 return 1; /* nothing remains for this target */
810
811         fill_origin_blob(&sb->revs->diffopt, parent, &file_p);
812         fill_origin_blob(&sb->revs->diffopt, target, &file_o);
813         num_get_patch++;
814
815         diff_hunks(&file_p, &file_o, 0, blame_chunk_cb, &d);
816         /* The rest (i.e. anything after tlno) are the same as the parent */
817         blame_chunk(sb, d.tlno, d.plno, last_in_target, target, parent);
818
819         return 0;
820 }
821
822 /*
823  * The lines in blame_entry after splitting blames many times can become
824  * very small and trivial, and at some point it becomes pointless to
825  * blame the parents.  E.g. "\t\t}\n\t}\n\n" appears everywhere in any
826  * ordinary C program, and it is not worth to say it was copied from
827  * totally unrelated file in the parent.
828  *
829  * Compute how trivial the lines in the blame_entry are.
830  */
831 static unsigned ent_score(struct scoreboard *sb, struct blame_entry *e)
832 {
833         unsigned score;
834         const char *cp, *ep;
835
836         if (e->score)
837                 return e->score;
838
839         score = 1;
840         cp = nth_line(sb, e->lno);
841         ep = nth_line(sb, e->lno + e->num_lines);
842         while (cp < ep) {
843                 unsigned ch = *((unsigned char *)cp);
844                 if (isalnum(ch))
845                         score++;
846                 cp++;
847         }
848         e->score = score;
849         return score;
850 }
851
852 /*
853  * best_so_far[] and this[] are both a split of an existing blame_entry
854  * that passes blame to the parent.  Maintain best_so_far the best split
855  * so far, by comparing this and best_so_far and copying this into
856  * bst_so_far as needed.
857  */
858 static void copy_split_if_better(struct scoreboard *sb,
859                                  struct blame_entry *best_so_far,
860                                  struct blame_entry *this)
861 {
862         int i;
863
864         if (!this[1].suspect)
865                 return;
866         if (best_so_far[1].suspect) {
867                 if (ent_score(sb, &this[1]) < ent_score(sb, &best_so_far[1]))
868                         return;
869         }
870
871         for (i = 0; i < 3; i++)
872                 origin_incref(this[i].suspect);
873         decref_split(best_so_far);
874         memcpy(best_so_far, this, sizeof(struct blame_entry [3]));
875 }
876
877 /*
878  * We are looking at a part of the final image represented by
879  * ent (tlno and same are offset by ent->s_lno).
880  * tlno is where we are looking at in the final image.
881  * up to (but not including) same match preimage.
882  * plno is where we are looking at in the preimage.
883  *
884  * <-------------- final image ---------------------->
885  *       <------ent------>
886  *         ^tlno ^same
887  *    <---------preimage----->
888  *         ^plno
889  *
890  * All line numbers are 0-based.
891  */
892 static void handle_split(struct scoreboard *sb,
893                          struct blame_entry *ent,
894                          int tlno, int plno, int same,
895                          struct origin *parent,
896                          struct blame_entry *split)
897 {
898         if (ent->num_lines <= tlno)
899                 return;
900         if (tlno < same) {
901                 struct blame_entry this[3];
902                 tlno += ent->s_lno;
903                 same += ent->s_lno;
904                 split_overlap(this, ent, tlno, plno, same, parent);
905                 copy_split_if_better(sb, split, this);
906                 decref_split(this);
907         }
908 }
909
910 struct handle_split_cb_data {
911         struct scoreboard *sb;
912         struct blame_entry *ent;
913         struct origin *parent;
914         struct blame_entry *split;
915         long plno;
916         long tlno;
917 };
918
919 static int handle_split_cb(long start_a, long count_a,
920                            long start_b, long count_b, void *data)
921 {
922         struct handle_split_cb_data *d = data;
923         handle_split(d->sb, d->ent, d->tlno, d->plno, start_b, d->parent,
924                      d->split);
925         d->plno = start_a + count_a;
926         d->tlno = start_b + count_b;
927         return 0;
928 }
929
930 /*
931  * Find the lines from parent that are the same as ent so that
932  * we can pass blames to it.  file_p has the blob contents for
933  * the parent.
934  */
935 static void find_copy_in_blob(struct scoreboard *sb,
936                               struct blame_entry *ent,
937                               struct origin *parent,
938                               struct blame_entry *split,
939                               mmfile_t *file_p)
940 {
941         const char *cp;
942         int cnt;
943         mmfile_t file_o;
944         struct handle_split_cb_data d;
945
946         memset(&d, 0, sizeof(d));
947         d.sb = sb; d.ent = ent; d.parent = parent; d.split = split;
948         /*
949          * Prepare mmfile that contains only the lines in ent.
950          */
951         cp = nth_line(sb, ent->lno);
952         file_o.ptr = (char *) cp;
953         cnt = ent->num_lines;
954
955         while (cnt && cp < sb->final_buf + sb->final_buf_size) {
956                 if (*cp++ == '\n')
957                         cnt--;
958         }
959         file_o.size = cp - file_o.ptr;
960
961         /*
962          * file_o is a part of final image we are annotating.
963          * file_p partially may match that image.
964          */
965         memset(split, 0, sizeof(struct blame_entry [3]));
966         diff_hunks(file_p, &file_o, 1, handle_split_cb, &d);
967         /* remainder, if any, all match the preimage */
968         handle_split(sb, ent, d.tlno, d.plno, ent->num_lines, parent, split);
969 }
970
971 /*
972  * See if lines currently target is suspected for can be attributed to
973  * parent.
974  */
975 static int find_move_in_parent(struct scoreboard *sb,
976                                struct origin *target,
977                                struct origin *parent)
978 {
979         int last_in_target, made_progress;
980         struct blame_entry *e, split[3];
981         mmfile_t file_p;
982
983         last_in_target = find_last_in_target(sb, target);
984         if (last_in_target < 0)
985                 return 1; /* nothing remains for this target */
986
987         fill_origin_blob(&sb->revs->diffopt, parent, &file_p);
988         if (!file_p.ptr)
989                 return 0;
990
991         made_progress = 1;
992         while (made_progress) {
993                 made_progress = 0;
994                 for (e = sb->ent; e; e = e->next) {
995                         if (e->guilty || !same_suspect(e->suspect, target) ||
996                             ent_score(sb, e) < blame_move_score)
997                                 continue;
998                         find_copy_in_blob(sb, e, parent, split, &file_p);
999                         if (split[1].suspect &&
1000                             blame_move_score < ent_score(sb, &split[1])) {
1001                                 split_blame(sb, split, e);
1002                                 made_progress = 1;
1003                         }
1004                         decref_split(split);
1005                 }
1006         }
1007         return 0;
1008 }
1009
1010 struct blame_list {
1011         struct blame_entry *ent;
1012         struct blame_entry split[3];
1013 };
1014
1015 /*
1016  * Count the number of entries the target is suspected for,
1017  * and prepare a list of entry and the best split.
1018  */
1019 static struct blame_list *setup_blame_list(struct scoreboard *sb,
1020                                            struct origin *target,
1021                                            int min_score,
1022                                            int *num_ents_p)
1023 {
1024         struct blame_entry *e;
1025         int num_ents, i;
1026         struct blame_list *blame_list = NULL;
1027
1028         for (e = sb->ent, num_ents = 0; e; e = e->next)
1029                 if (!e->scanned && !e->guilty &&
1030                     same_suspect(e->suspect, target) &&
1031                     min_score < ent_score(sb, e))
1032                         num_ents++;
1033         if (num_ents) {
1034                 blame_list = xcalloc(num_ents, sizeof(struct blame_list));
1035                 for (e = sb->ent, i = 0; e; e = e->next)
1036                         if (!e->scanned && !e->guilty &&
1037                             same_suspect(e->suspect, target) &&
1038                             min_score < ent_score(sb, e))
1039                                 blame_list[i++].ent = e;
1040         }
1041         *num_ents_p = num_ents;
1042         return blame_list;
1043 }
1044
1045 /*
1046  * Reset the scanned status on all entries.
1047  */
1048 static void reset_scanned_flag(struct scoreboard *sb)
1049 {
1050         struct blame_entry *e;
1051         for (e = sb->ent; e; e = e->next)
1052                 e->scanned = 0;
1053 }
1054
1055 /*
1056  * For lines target is suspected for, see if we can find code movement
1057  * across file boundary from the parent commit.  porigin is the path
1058  * in the parent we already tried.
1059  */
1060 static int find_copy_in_parent(struct scoreboard *sb,
1061                                struct origin *target,
1062                                struct commit *parent,
1063                                struct origin *porigin,
1064                                int opt)
1065 {
1066         struct diff_options diff_opts;
1067         const char *paths[1];
1068         int i, j;
1069         int retval;
1070         struct blame_list *blame_list;
1071         int num_ents;
1072
1073         blame_list = setup_blame_list(sb, target, blame_copy_score, &num_ents);
1074         if (!blame_list)
1075                 return 1; /* nothing remains for this target */
1076
1077         diff_setup(&diff_opts);
1078         DIFF_OPT_SET(&diff_opts, RECURSIVE);
1079         diff_opts.output_format = DIFF_FORMAT_NO_OUTPUT;
1080
1081         paths[0] = NULL;
1082         diff_tree_setup_paths(paths, &diff_opts);
1083         diff_setup_done(&diff_opts);
1084
1085         /* Try "find copies harder" on new path if requested;
1086          * we do not want to use diffcore_rename() actually to
1087          * match things up; find_copies_harder is set only to
1088          * force diff_tree_sha1() to feed all filepairs to diff_queue,
1089          * and this code needs to be after diff_setup_done(), which
1090          * usually makes find-copies-harder imply copy detection.
1091          */
1092         if ((opt & PICKAXE_BLAME_COPY_HARDEST)
1093             || ((opt & PICKAXE_BLAME_COPY_HARDER)
1094                 && (!porigin || strcmp(target->path, porigin->path))))
1095                 DIFF_OPT_SET(&diff_opts, FIND_COPIES_HARDER);
1096
1097         if (is_null_sha1(target->commit->object.sha1))
1098                 do_diff_cache(parent->tree->object.sha1, &diff_opts);
1099         else
1100                 diff_tree_sha1(parent->tree->object.sha1,
1101                                target->commit->tree->object.sha1,
1102                                "", &diff_opts);
1103
1104         if (!DIFF_OPT_TST(&diff_opts, FIND_COPIES_HARDER))
1105                 diffcore_std(&diff_opts);
1106
1107         retval = 0;
1108         while (1) {
1109                 int made_progress = 0;
1110
1111                 for (i = 0; i < diff_queued_diff.nr; i++) {
1112                         struct diff_filepair *p = diff_queued_diff.queue[i];
1113                         struct origin *norigin;
1114                         mmfile_t file_p;
1115                         struct blame_entry this[3];
1116
1117                         if (!DIFF_FILE_VALID(p->one))
1118                                 continue; /* does not exist in parent */
1119                         if (S_ISGITLINK(p->one->mode))
1120                                 continue; /* ignore git links */
1121                         if (porigin && !strcmp(p->one->path, porigin->path))
1122                                 /* find_move already dealt with this path */
1123                                 continue;
1124
1125                         norigin = get_origin(sb, parent, p->one->path);
1126                         hashcpy(norigin->blob_sha1, p->one->sha1);
1127                         norigin->mode = p->one->mode;
1128                         fill_origin_blob(&sb->revs->diffopt, norigin, &file_p);
1129                         if (!file_p.ptr)
1130                                 continue;
1131
1132                         for (j = 0; j < num_ents; j++) {
1133                                 find_copy_in_blob(sb, blame_list[j].ent,
1134                                                   norigin, this, &file_p);
1135                                 copy_split_if_better(sb, blame_list[j].split,
1136                                                      this);
1137                                 decref_split(this);
1138                         }
1139                         origin_decref(norigin);
1140                 }
1141
1142                 for (j = 0; j < num_ents; j++) {
1143                         struct blame_entry *split = blame_list[j].split;
1144                         if (split[1].suspect &&
1145                             blame_copy_score < ent_score(sb, &split[1])) {
1146                                 split_blame(sb, split, blame_list[j].ent);
1147                                 made_progress = 1;
1148                         }
1149                         else
1150                                 blame_list[j].ent->scanned = 1;
1151                         decref_split(split);
1152                 }
1153                 free(blame_list);
1154
1155                 if (!made_progress)
1156                         break;
1157                 blame_list = setup_blame_list(sb, target, blame_copy_score, &num_ents);
1158                 if (!blame_list) {
1159                         retval = 1;
1160                         break;
1161                 }
1162         }
1163         reset_scanned_flag(sb);
1164         diff_flush(&diff_opts);
1165         diff_tree_release_paths(&diff_opts);
1166         return retval;
1167 }
1168
1169 /*
1170  * The blobs of origin and porigin exactly match, so everything
1171  * origin is suspected for can be blamed on the parent.
1172  */
1173 static void pass_whole_blame(struct scoreboard *sb,
1174                              struct origin *origin, struct origin *porigin)
1175 {
1176         struct blame_entry *e;
1177
1178         if (!porigin->file.ptr && origin->file.ptr) {
1179                 /* Steal its file */
1180                 porigin->file = origin->file;
1181                 origin->file.ptr = NULL;
1182         }
1183         for (e = sb->ent; e; e = e->next) {
1184                 if (!same_suspect(e->suspect, origin))
1185                         continue;
1186                 origin_incref(porigin);
1187                 origin_decref(e->suspect);
1188                 e->suspect = porigin;
1189         }
1190 }
1191
1192 /*
1193  * We pass blame from the current commit to its parents.  We keep saying
1194  * "parent" (and "porigin"), but what we mean is to find scapegoat to
1195  * exonerate ourselves.
1196  */
1197 static struct commit_list *first_scapegoat(struct rev_info *revs, struct commit *commit)
1198 {
1199         if (!reverse)
1200                 return commit->parents;
1201         return lookup_decoration(&revs->children, &commit->object);
1202 }
1203
1204 static int num_scapegoats(struct rev_info *revs, struct commit *commit)
1205 {
1206         int cnt;
1207         struct commit_list *l = first_scapegoat(revs, commit);
1208         for (cnt = 0; l; l = l->next)
1209                 cnt++;
1210         return cnt;
1211 }
1212
1213 #define MAXSG 16
1214
1215 static void pass_blame(struct scoreboard *sb, struct origin *origin, int opt)
1216 {
1217         struct rev_info *revs = sb->revs;
1218         int i, pass, num_sg;
1219         struct commit *commit = origin->commit;
1220         struct commit_list *sg;
1221         struct origin *sg_buf[MAXSG];
1222         struct origin *porigin, **sg_origin = sg_buf;
1223
1224         num_sg = num_scapegoats(revs, commit);
1225         if (!num_sg)
1226                 goto finish;
1227         else if (num_sg < ARRAY_SIZE(sg_buf))
1228                 memset(sg_buf, 0, sizeof(sg_buf));
1229         else
1230                 sg_origin = xcalloc(num_sg, sizeof(*sg_origin));
1231
1232         /*
1233          * The first pass looks for unrenamed path to optimize for
1234          * common cases, then we look for renames in the second pass.
1235          */
1236         for (pass = 0; pass < 2 - no_whole_file_rename; pass++) {
1237                 struct origin *(*find)(struct scoreboard *,
1238                                        struct commit *, struct origin *);
1239                 find = pass ? find_rename : find_origin;
1240
1241                 for (i = 0, sg = first_scapegoat(revs, commit);
1242                      i < num_sg && sg;
1243                      sg = sg->next, i++) {
1244                         struct commit *p = sg->item;
1245                         int j, same;
1246
1247                         if (sg_origin[i])
1248                                 continue;
1249                         if (parse_commit(p))
1250                                 continue;
1251                         porigin = find(sb, p, origin);
1252                         if (!porigin)
1253                                 continue;
1254                         if (!hashcmp(porigin->blob_sha1, origin->blob_sha1)) {
1255                                 pass_whole_blame(sb, origin, porigin);
1256                                 origin_decref(porigin);
1257                                 goto finish;
1258                         }
1259                         for (j = same = 0; j < i; j++)
1260                                 if (sg_origin[j] &&
1261                                     !hashcmp(sg_origin[j]->blob_sha1,
1262                                              porigin->blob_sha1)) {
1263                                         same = 1;
1264                                         break;
1265                                 }
1266                         if (!same)
1267                                 sg_origin[i] = porigin;
1268                         else
1269                                 origin_decref(porigin);
1270                 }
1271         }
1272
1273         num_commits++;
1274         for (i = 0, sg = first_scapegoat(revs, commit);
1275              i < num_sg && sg;
1276              sg = sg->next, i++) {
1277                 struct origin *porigin = sg_origin[i];
1278                 if (!porigin)
1279                         continue;
1280                 if (!origin->previous) {
1281                         origin_incref(porigin);
1282                         origin->previous = porigin;
1283                 }
1284                 if (pass_blame_to_parent(sb, origin, porigin))
1285                         goto finish;
1286         }
1287
1288         /*
1289          * Optionally find moves in parents' files.
1290          */
1291         if (opt & PICKAXE_BLAME_MOVE)
1292                 for (i = 0, sg = first_scapegoat(revs, commit);
1293                      i < num_sg && sg;
1294                      sg = sg->next, i++) {
1295                         struct origin *porigin = sg_origin[i];
1296                         if (!porigin)
1297                                 continue;
1298                         if (find_move_in_parent(sb, origin, porigin))
1299                                 goto finish;
1300                 }
1301
1302         /*
1303          * Optionally find copies from parents' files.
1304          */
1305         if (opt & PICKAXE_BLAME_COPY)
1306                 for (i = 0, sg = first_scapegoat(revs, commit);
1307                      i < num_sg && sg;
1308                      sg = sg->next, i++) {
1309                         struct origin *porigin = sg_origin[i];
1310                         if (find_copy_in_parent(sb, origin, sg->item,
1311                                                 porigin, opt))
1312                                 goto finish;
1313                 }
1314
1315  finish:
1316         for (i = 0; i < num_sg; i++) {
1317                 if (sg_origin[i]) {
1318                         drop_origin_blob(sg_origin[i]);
1319                         origin_decref(sg_origin[i]);
1320                 }
1321         }
1322         drop_origin_blob(origin);
1323         if (sg_buf != sg_origin)
1324                 free(sg_origin);
1325 }
1326
1327 /*
1328  * Information on commits, used for output.
1329  */
1330 struct commit_info {
1331         struct strbuf author;
1332         struct strbuf author_mail;
1333         unsigned long author_time;
1334         struct strbuf author_tz;
1335
1336         /* filled only when asked for details */
1337         struct strbuf committer;
1338         struct strbuf committer_mail;
1339         unsigned long committer_time;
1340         struct strbuf committer_tz;
1341
1342         struct strbuf summary;
1343 };
1344
1345 /*
1346  * Parse author/committer line in the commit object buffer
1347  */
1348 static void get_ac_line(const char *inbuf, const char *what,
1349         struct strbuf *name, struct strbuf *mail,
1350         unsigned long *time, struct strbuf *tz)
1351 {
1352         struct ident_split ident;
1353         size_t len, maillen, namelen;
1354         char *tmp, *endp;
1355         const char *namebuf, *mailbuf;
1356
1357         tmp = strstr(inbuf, what);
1358         if (!tmp)
1359                 goto error_out;
1360         tmp += strlen(what);
1361         endp = strchr(tmp, '\n');
1362         if (!endp)
1363                 len = strlen(tmp);
1364         else
1365                 len = endp - tmp;
1366
1367         if (split_ident_line(&ident, tmp, len)) {
1368         error_out:
1369                 /* Ugh */
1370                 tmp = "(unknown)";
1371                 strbuf_addstr(name, tmp);
1372                 strbuf_addstr(mail, tmp);
1373                 strbuf_addstr(tz, tmp);
1374                 *time = 0;
1375                 return;
1376         }
1377
1378         namelen = ident.name_end - ident.name_begin;
1379         namebuf = ident.name_begin;
1380
1381         maillen = ident.mail_end - ident.mail_begin;
1382         mailbuf = ident.mail_begin;
1383
1384         if (ident.date_begin && ident.date_end)
1385                 *time = strtoul(ident.date_begin, NULL, 10);
1386         else
1387                 *time = 0;
1388
1389         if (ident.tz_begin && ident.tz_end)
1390                 strbuf_add(tz, ident.tz_begin, ident.tz_end - ident.tz_begin);
1391         else
1392                 strbuf_addstr(tz, "(unknown)");
1393
1394         /*
1395          * Now, convert both name and e-mail using mailmap
1396          */
1397         map_user(&mailmap, &mailbuf, &maillen,
1398                  &namebuf, &namelen);
1399
1400         strbuf_addf(mail, "<%.*s>", (int)maillen, mailbuf);
1401         strbuf_add(name, namebuf, namelen);
1402 }
1403
1404 static void commit_info_init(struct commit_info *ci)
1405 {
1406
1407         strbuf_init(&ci->author, 0);
1408         strbuf_init(&ci->author_mail, 0);
1409         strbuf_init(&ci->author_tz, 0);
1410         strbuf_init(&ci->committer, 0);
1411         strbuf_init(&ci->committer_mail, 0);
1412         strbuf_init(&ci->committer_tz, 0);
1413         strbuf_init(&ci->summary, 0);
1414 }
1415
1416 static void commit_info_destroy(struct commit_info *ci)
1417 {
1418
1419         strbuf_release(&ci->author);
1420         strbuf_release(&ci->author_mail);
1421         strbuf_release(&ci->author_tz);
1422         strbuf_release(&ci->committer);
1423         strbuf_release(&ci->committer_mail);
1424         strbuf_release(&ci->committer_tz);
1425         strbuf_release(&ci->summary);
1426 }
1427
1428 static void get_commit_info(struct commit *commit,
1429                             struct commit_info *ret,
1430                             int detailed)
1431 {
1432         int len;
1433         const char *subject, *encoding;
1434         char *message;
1435
1436         commit_info_init(ret);
1437
1438         encoding = get_log_output_encoding();
1439         message = logmsg_reencode(commit, NULL, encoding);
1440         get_ac_line(message, "\nauthor ",
1441                     &ret->author, &ret->author_mail,
1442                     &ret->author_time, &ret->author_tz);
1443
1444         if (!detailed) {
1445                 logmsg_free(message, commit);
1446                 return;
1447         }
1448
1449         get_ac_line(message, "\ncommitter ",
1450                     &ret->committer, &ret->committer_mail,
1451                     &ret->committer_time, &ret->committer_tz);
1452
1453         len = find_commit_subject(message, &subject);
1454         if (len)
1455                 strbuf_add(&ret->summary, subject, len);
1456         else
1457                 strbuf_addf(&ret->summary, "(%s)", sha1_to_hex(commit->object.sha1));
1458
1459         logmsg_free(message, commit);
1460 }
1461
1462 /*
1463  * To allow LF and other nonportable characters in pathnames,
1464  * they are c-style quoted as needed.
1465  */
1466 static void write_filename_info(const char *path)
1467 {
1468         printf("filename ");
1469         write_name_quoted(path, stdout, '\n');
1470 }
1471
1472 /*
1473  * Porcelain/Incremental format wants to show a lot of details per
1474  * commit.  Instead of repeating this every line, emit it only once,
1475  * the first time each commit appears in the output (unless the
1476  * user has specifically asked for us to repeat).
1477  */
1478 static int emit_one_suspect_detail(struct origin *suspect, int repeat)
1479 {
1480         struct commit_info ci;
1481
1482         if (!repeat && (suspect->commit->object.flags & METAINFO_SHOWN))
1483                 return 0;
1484
1485         suspect->commit->object.flags |= METAINFO_SHOWN;
1486         get_commit_info(suspect->commit, &ci, 1);
1487         printf("author %s\n", ci.author.buf);
1488         printf("author-mail %s\n", ci.author_mail.buf);
1489         printf("author-time %lu\n", ci.author_time);
1490         printf("author-tz %s\n", ci.author_tz.buf);
1491         printf("committer %s\n", ci.committer.buf);
1492         printf("committer-mail %s\n", ci.committer_mail.buf);
1493         printf("committer-time %lu\n", ci.committer_time);
1494         printf("committer-tz %s\n", ci.committer_tz.buf);
1495         printf("summary %s\n", ci.summary.buf);
1496         if (suspect->commit->object.flags & UNINTERESTING)
1497                 printf("boundary\n");
1498         if (suspect->previous) {
1499                 struct origin *prev = suspect->previous;
1500                 printf("previous %s ", sha1_to_hex(prev->commit->object.sha1));
1501                 write_name_quoted(prev->path, stdout, '\n');
1502         }
1503
1504         commit_info_destroy(&ci);
1505
1506         return 1;
1507 }
1508
1509 /*
1510  * The blame_entry is found to be guilty for the range.  Mark it
1511  * as such, and show it in incremental output.
1512  */
1513 static void found_guilty_entry(struct blame_entry *ent)
1514 {
1515         if (ent->guilty)
1516                 return;
1517         ent->guilty = 1;
1518         if (incremental) {
1519                 struct origin *suspect = ent->suspect;
1520
1521                 printf("%s %d %d %d\n",
1522                        sha1_to_hex(suspect->commit->object.sha1),
1523                        ent->s_lno + 1, ent->lno + 1, ent->num_lines);
1524                 emit_one_suspect_detail(suspect, 0);
1525                 write_filename_info(suspect->path);
1526                 maybe_flush_or_die(stdout, "stdout");
1527         }
1528 }
1529
1530 /*
1531  * The main loop -- while the scoreboard has lines whose true origin
1532  * is still unknown, pick one blame_entry, and allow its current
1533  * suspect to pass blames to its parents.
1534  */
1535 static void assign_blame(struct scoreboard *sb, int opt)
1536 {
1537         struct rev_info *revs = sb->revs;
1538
1539         while (1) {
1540                 struct blame_entry *ent;
1541                 struct commit *commit;
1542                 struct origin *suspect = NULL;
1543
1544                 /* find one suspect to break down */
1545                 for (ent = sb->ent; !suspect && ent; ent = ent->next)
1546                         if (!ent->guilty)
1547                                 suspect = ent->suspect;
1548                 if (!suspect)
1549                         return; /* all done */
1550
1551                 /*
1552                  * We will use this suspect later in the loop,
1553                  * so hold onto it in the meantime.
1554                  */
1555                 origin_incref(suspect);
1556                 commit = suspect->commit;
1557                 if (!commit->object.parsed)
1558                         parse_commit(commit);
1559                 if (reverse ||
1560                     (!(commit->object.flags & UNINTERESTING) &&
1561                      !(revs->max_age != -1 && commit->date < revs->max_age)))
1562                         pass_blame(sb, suspect, opt);
1563                 else {
1564                         commit->object.flags |= UNINTERESTING;
1565                         if (commit->object.parsed)
1566                                 mark_parents_uninteresting(commit);
1567                 }
1568                 /* treat root commit as boundary */
1569                 if (!commit->parents && !show_root)
1570                         commit->object.flags |= UNINTERESTING;
1571
1572                 /* Take responsibility for the remaining entries */
1573                 for (ent = sb->ent; ent; ent = ent->next)
1574                         if (same_suspect(ent->suspect, suspect))
1575                                 found_guilty_entry(ent);
1576                 origin_decref(suspect);
1577
1578                 if (DEBUG) /* sanity */
1579                         sanity_check_refcnt(sb);
1580         }
1581 }
1582
1583 static const char *format_time(unsigned long time, const char *tz_str,
1584                                int show_raw_time)
1585 {
1586         static char time_buf[128];
1587         const char *time_str;
1588         int time_len;
1589         int tz;
1590
1591         if (show_raw_time) {
1592                 snprintf(time_buf, sizeof(time_buf), "%lu %s", time, tz_str);
1593         }
1594         else {
1595                 tz = atoi(tz_str);
1596                 time_str = show_date(time, tz, blame_date_mode);
1597                 time_len = strlen(time_str);
1598                 memcpy(time_buf, time_str, time_len);
1599                 memset(time_buf + time_len, ' ', blame_date_width - time_len);
1600         }
1601         return time_buf;
1602 }
1603
1604 #define OUTPUT_ANNOTATE_COMPAT  001
1605 #define OUTPUT_LONG_OBJECT_NAME 002
1606 #define OUTPUT_RAW_TIMESTAMP    004
1607 #define OUTPUT_PORCELAIN        010
1608 #define OUTPUT_SHOW_NAME        020
1609 #define OUTPUT_SHOW_NUMBER      040
1610 #define OUTPUT_SHOW_SCORE      0100
1611 #define OUTPUT_NO_AUTHOR       0200
1612 #define OUTPUT_SHOW_EMAIL       0400
1613 #define OUTPUT_LINE_PORCELAIN 01000
1614
1615 static void emit_porcelain_details(struct origin *suspect, int repeat)
1616 {
1617         if (emit_one_suspect_detail(suspect, repeat) ||
1618             (suspect->commit->object.flags & MORE_THAN_ONE_PATH))
1619                 write_filename_info(suspect->path);
1620 }
1621
1622 static void emit_porcelain(struct scoreboard *sb, struct blame_entry *ent,
1623                            int opt)
1624 {
1625         int repeat = opt & OUTPUT_LINE_PORCELAIN;
1626         int cnt;
1627         const char *cp;
1628         struct origin *suspect = ent->suspect;
1629         char hex[41];
1630
1631         strcpy(hex, sha1_to_hex(suspect->commit->object.sha1));
1632         printf("%s%c%d %d %d\n",
1633                hex,
1634                ent->guilty ? ' ' : '*', /* purely for debugging */
1635                ent->s_lno + 1,
1636                ent->lno + 1,
1637                ent->num_lines);
1638         emit_porcelain_details(suspect, repeat);
1639
1640         cp = nth_line(sb, ent->lno);
1641         for (cnt = 0; cnt < ent->num_lines; cnt++) {
1642                 char ch;
1643                 if (cnt) {
1644                         printf("%s %d %d\n", hex,
1645                                ent->s_lno + 1 + cnt,
1646                                ent->lno + 1 + cnt);
1647                         if (repeat)
1648                                 emit_porcelain_details(suspect, 1);
1649                 }
1650                 putchar('\t');
1651                 do {
1652                         ch = *cp++;
1653                         putchar(ch);
1654                 } while (ch != '\n' &&
1655                          cp < sb->final_buf + sb->final_buf_size);
1656         }
1657
1658         if (sb->final_buf_size && cp[-1] != '\n')
1659                 putchar('\n');
1660 }
1661
1662 static void emit_other(struct scoreboard *sb, struct blame_entry *ent, int opt)
1663 {
1664         int cnt;
1665         const char *cp;
1666         struct origin *suspect = ent->suspect;
1667         struct commit_info ci;
1668         char hex[41];
1669         int show_raw_time = !!(opt & OUTPUT_RAW_TIMESTAMP);
1670
1671         get_commit_info(suspect->commit, &ci, 1);
1672         strcpy(hex, sha1_to_hex(suspect->commit->object.sha1));
1673
1674         cp = nth_line(sb, ent->lno);
1675         for (cnt = 0; cnt < ent->num_lines; cnt++) {
1676                 char ch;
1677                 int length = (opt & OUTPUT_LONG_OBJECT_NAME) ? 40 : abbrev;
1678
1679                 if (suspect->commit->object.flags & UNINTERESTING) {
1680                         if (blank_boundary)
1681                                 memset(hex, ' ', length);
1682                         else if (!(opt & OUTPUT_ANNOTATE_COMPAT)) {
1683                                 length--;
1684                                 putchar('^');
1685                         }
1686                 }
1687
1688                 printf("%.*s", length, hex);
1689                 if (opt & OUTPUT_ANNOTATE_COMPAT) {
1690                         const char *name;
1691                         if (opt & OUTPUT_SHOW_EMAIL)
1692                                 name = ci.author_mail.buf;
1693                         else
1694                                 name = ci.author.buf;
1695                         printf("\t(%10s\t%10s\t%d)", name,
1696                                format_time(ci.author_time, ci.author_tz.buf,
1697                                            show_raw_time),
1698                                ent->lno + 1 + cnt);
1699                 } else {
1700                         if (opt & OUTPUT_SHOW_SCORE)
1701                                 printf(" %*d %02d",
1702                                        max_score_digits, ent->score,
1703                                        ent->suspect->refcnt);
1704                         if (opt & OUTPUT_SHOW_NAME)
1705                                 printf(" %-*.*s", longest_file, longest_file,
1706                                        suspect->path);
1707                         if (opt & OUTPUT_SHOW_NUMBER)
1708                                 printf(" %*d", max_orig_digits,
1709                                        ent->s_lno + 1 + cnt);
1710
1711                         if (!(opt & OUTPUT_NO_AUTHOR)) {
1712                                 const char *name;
1713                                 int pad;
1714                                 if (opt & OUTPUT_SHOW_EMAIL)
1715                                         name = ci.author_mail.buf;
1716                                 else
1717                                         name = ci.author.buf;
1718                                 pad = longest_author - utf8_strwidth(name);
1719                                 printf(" (%s%*s %10s",
1720                                        name, pad, "",
1721                                        format_time(ci.author_time,
1722                                                    ci.author_tz.buf,
1723                                                    show_raw_time));
1724                         }
1725                         printf(" %*d) ",
1726                                max_digits, ent->lno + 1 + cnt);
1727                 }
1728                 do {
1729                         ch = *cp++;
1730                         putchar(ch);
1731                 } while (ch != '\n' &&
1732                          cp < sb->final_buf + sb->final_buf_size);
1733         }
1734
1735         if (sb->final_buf_size && cp[-1] != '\n')
1736                 putchar('\n');
1737
1738         commit_info_destroy(&ci);
1739 }
1740
1741 static void output(struct scoreboard *sb, int option)
1742 {
1743         struct blame_entry *ent;
1744
1745         if (option & OUTPUT_PORCELAIN) {
1746                 for (ent = sb->ent; ent; ent = ent->next) {
1747                         struct blame_entry *oth;
1748                         struct origin *suspect = ent->suspect;
1749                         struct commit *commit = suspect->commit;
1750                         if (commit->object.flags & MORE_THAN_ONE_PATH)
1751                                 continue;
1752                         for (oth = ent->next; oth; oth = oth->next) {
1753                                 if ((oth->suspect->commit != commit) ||
1754                                     !strcmp(oth->suspect->path, suspect->path))
1755                                         continue;
1756                                 commit->object.flags |= MORE_THAN_ONE_PATH;
1757                                 break;
1758                         }
1759                 }
1760         }
1761
1762         for (ent = sb->ent; ent; ent = ent->next) {
1763                 if (option & OUTPUT_PORCELAIN)
1764                         emit_porcelain(sb, ent, option);
1765                 else {
1766                         emit_other(sb, ent, option);
1767                 }
1768         }
1769 }
1770
1771 /*
1772  * To allow quick access to the contents of nth line in the
1773  * final image, prepare an index in the scoreboard.
1774  */
1775 static int prepare_lines(struct scoreboard *sb)
1776 {
1777         const char *buf = sb->final_buf;
1778         unsigned long len = sb->final_buf_size;
1779         int num = 0, incomplete = 0, bol = 1;
1780
1781         if (len && buf[len-1] != '\n')
1782                 incomplete++; /* incomplete line at the end */
1783         while (len--) {
1784                 if (bol) {
1785                         sb->lineno = xrealloc(sb->lineno,
1786                                               sizeof(int *) * (num + 1));
1787                         sb->lineno[num] = buf - sb->final_buf;
1788                         bol = 0;
1789                 }
1790                 if (*buf++ == '\n') {
1791                         num++;
1792                         bol = 1;
1793                 }
1794         }
1795         sb->lineno = xrealloc(sb->lineno,
1796                               sizeof(int *) * (num + incomplete + 1));
1797         sb->lineno[num + incomplete] = buf - sb->final_buf;
1798         sb->num_lines = num + incomplete;
1799         return sb->num_lines;
1800 }
1801
1802 /*
1803  * Add phony grafts for use with -S; this is primarily to
1804  * support git's cvsserver that wants to give a linear history
1805  * to its clients.
1806  */
1807 static int read_ancestry(const char *graft_file)
1808 {
1809         FILE *fp = fopen(graft_file, "r");
1810         char buf[1024];
1811         if (!fp)
1812                 return -1;
1813         while (fgets(buf, sizeof(buf), fp)) {
1814                 /* The format is just "Commit Parent1 Parent2 ...\n" */
1815                 int len = strlen(buf);
1816                 struct commit_graft *graft = read_graft_line(buf, len);
1817                 if (graft)
1818                         register_commit_graft(graft, 0);
1819         }
1820         fclose(fp);
1821         return 0;
1822 }
1823
1824 static int update_auto_abbrev(int auto_abbrev, struct origin *suspect)
1825 {
1826         const char *uniq = find_unique_abbrev(suspect->commit->object.sha1,
1827                                               auto_abbrev);
1828         int len = strlen(uniq);
1829         if (auto_abbrev < len)
1830                 return len;
1831         return auto_abbrev;
1832 }
1833
1834 /*
1835  * How many columns do we need to show line numbers, authors,
1836  * and filenames?
1837  */
1838 static void find_alignment(struct scoreboard *sb, int *option)
1839 {
1840         int longest_src_lines = 0;
1841         int longest_dst_lines = 0;
1842         unsigned largest_score = 0;
1843         struct blame_entry *e;
1844         int compute_auto_abbrev = (abbrev < 0);
1845         int auto_abbrev = default_abbrev;
1846
1847         for (e = sb->ent; e; e = e->next) {
1848                 struct origin *suspect = e->suspect;
1849                 struct commit_info ci;
1850                 int num;
1851
1852                 if (compute_auto_abbrev)
1853                         auto_abbrev = update_auto_abbrev(auto_abbrev, suspect);
1854                 if (strcmp(suspect->path, sb->path))
1855                         *option |= OUTPUT_SHOW_NAME;
1856                 num = strlen(suspect->path);
1857                 if (longest_file < num)
1858                         longest_file = num;
1859                 if (!(suspect->commit->object.flags & METAINFO_SHOWN)) {
1860                         suspect->commit->object.flags |= METAINFO_SHOWN;
1861                         get_commit_info(suspect->commit, &ci, 1);
1862                         if (*option & OUTPUT_SHOW_EMAIL)
1863                                 num = utf8_strwidth(ci.author_mail.buf);
1864                         else
1865                                 num = utf8_strwidth(ci.author.buf);
1866                         if (longest_author < num)
1867                                 longest_author = num;
1868                 }
1869                 num = e->s_lno + e->num_lines;
1870                 if (longest_src_lines < num)
1871                         longest_src_lines = num;
1872                 num = e->lno + e->num_lines;
1873                 if (longest_dst_lines < num)
1874                         longest_dst_lines = num;
1875                 if (largest_score < ent_score(sb, e))
1876                         largest_score = ent_score(sb, e);
1877
1878                 commit_info_destroy(&ci);
1879         }
1880         max_orig_digits = decimal_width(longest_src_lines);
1881         max_digits = decimal_width(longest_dst_lines);
1882         max_score_digits = decimal_width(largest_score);
1883
1884         if (compute_auto_abbrev)
1885                 /* one more abbrev length is needed for the boundary commit */
1886                 abbrev = auto_abbrev + 1;
1887 }
1888
1889 /*
1890  * For debugging -- origin is refcounted, and this asserts that
1891  * we do not underflow.
1892  */
1893 static void sanity_check_refcnt(struct scoreboard *sb)
1894 {
1895         int baa = 0;
1896         struct blame_entry *ent;
1897
1898         for (ent = sb->ent; ent; ent = ent->next) {
1899                 /* Nobody should have zero or negative refcnt */
1900                 if (ent->suspect->refcnt <= 0) {
1901                         fprintf(stderr, "%s in %s has negative refcnt %d\n",
1902                                 ent->suspect->path,
1903                                 sha1_to_hex(ent->suspect->commit->object.sha1),
1904                                 ent->suspect->refcnt);
1905                         baa = 1;
1906                 }
1907         }
1908         if (baa) {
1909                 int opt = 0160;
1910                 find_alignment(sb, &opt);
1911                 output(sb, opt);
1912                 die("Baa %d!", baa);
1913         }
1914 }
1915
1916 /*
1917  * Used for the command line parsing; check if the path exists
1918  * in the working tree.
1919  */
1920 static int has_string_in_work_tree(const char *path)
1921 {
1922         struct stat st;
1923         return !lstat(path, &st);
1924 }
1925
1926 static unsigned parse_score(const char *arg)
1927 {
1928         char *end;
1929         unsigned long score = strtoul(arg, &end, 10);
1930         if (*end)
1931                 return 0;
1932         return score;
1933 }
1934
1935 static const char *add_prefix(const char *prefix, const char *path)
1936 {
1937         return prefix_path(prefix, prefix ? strlen(prefix) : 0, path);
1938 }
1939
1940 /*
1941  * Parsing of -L option
1942  */
1943 static void prepare_blame_range(struct scoreboard *sb,
1944                                 const char *bottomtop,
1945                                 long lno,
1946                                 long *bottom, long *top)
1947 {
1948         if (parse_range_arg(bottomtop, nth_line_cb, sb, lno, bottom, top, sb->path))
1949                 usage(blame_usage);
1950 }
1951
1952 static int git_blame_config(const char *var, const char *value, void *cb)
1953 {
1954         if (!strcmp(var, "blame.showroot")) {
1955                 show_root = git_config_bool(var, value);
1956                 return 0;
1957         }
1958         if (!strcmp(var, "blame.blankboundary")) {
1959                 blank_boundary = git_config_bool(var, value);
1960                 return 0;
1961         }
1962         if (!strcmp(var, "blame.date")) {
1963                 if (!value)
1964                         return config_error_nonbool(var);
1965                 blame_date_mode = parse_date_format(value);
1966                 return 0;
1967         }
1968
1969         if (userdiff_config(var, value) < 0)
1970                 return -1;
1971
1972         return git_default_config(var, value, cb);
1973 }
1974
1975 static void verify_working_tree_path(struct commit *work_tree, const char *path)
1976 {
1977         struct commit_list *parents;
1978
1979         for (parents = work_tree->parents; parents; parents = parents->next) {
1980                 const unsigned char *commit_sha1 = parents->item->object.sha1;
1981                 unsigned char blob_sha1[20];
1982                 unsigned mode;
1983
1984                 if (!get_tree_entry(commit_sha1, path, blob_sha1, &mode) &&
1985                     sha1_object_info(blob_sha1, NULL) == OBJ_BLOB)
1986                         return;
1987         }
1988         die("no such path '%s' in HEAD", path);
1989 }
1990
1991 static struct commit_list **append_parent(struct commit_list **tail, const unsigned char *sha1)
1992 {
1993         struct commit *parent;
1994
1995         parent = lookup_commit_reference(sha1);
1996         if (!parent)
1997                 die("no such commit %s", sha1_to_hex(sha1));
1998         return &commit_list_insert(parent, tail)->next;
1999 }
2000
2001 static void append_merge_parents(struct commit_list **tail)
2002 {
2003         int merge_head;
2004         const char *merge_head_file = git_path("MERGE_HEAD");
2005         struct strbuf line = STRBUF_INIT;
2006
2007         merge_head = open(merge_head_file, O_RDONLY);
2008         if (merge_head < 0) {
2009                 if (errno == ENOENT)
2010                         return;
2011                 die("cannot open '%s' for reading", merge_head_file);
2012         }
2013
2014         while (!strbuf_getwholeline_fd(&line, merge_head, '\n')) {
2015                 unsigned char sha1[20];
2016                 if (line.len < 40 || get_sha1_hex(line.buf, sha1))
2017                         die("unknown line in '%s': %s", merge_head_file, line.buf);
2018                 tail = append_parent(tail, sha1);
2019         }
2020         close(merge_head);
2021         strbuf_release(&line);
2022 }
2023
2024 /*
2025  * Prepare a dummy commit that represents the work tree (or staged) item.
2026  * Note that annotating work tree item never works in the reverse.
2027  */
2028 static struct commit *fake_working_tree_commit(struct diff_options *opt,
2029                                                const char *path,
2030                                                const char *contents_from)
2031 {
2032         struct commit *commit;
2033         struct origin *origin;
2034         struct commit_list **parent_tail, *parent;
2035         unsigned char head_sha1[20];
2036         struct strbuf buf = STRBUF_INIT;
2037         const char *ident;
2038         time_t now;
2039         int size, len;
2040         struct cache_entry *ce;
2041         unsigned mode;
2042         struct strbuf msg = STRBUF_INIT;
2043
2044         time(&now);
2045         commit = xcalloc(1, sizeof(*commit));
2046         commit->object.parsed = 1;
2047         commit->date = now;
2048         commit->object.type = OBJ_COMMIT;
2049         parent_tail = &commit->parents;
2050
2051         if (!resolve_ref_unsafe("HEAD", head_sha1, 1, NULL))
2052                 die("no such ref: HEAD");
2053
2054         parent_tail = append_parent(parent_tail, head_sha1);
2055         append_merge_parents(parent_tail);
2056         verify_working_tree_path(commit, path);
2057
2058         origin = make_origin(commit, path);
2059
2060         ident = fmt_ident("Not Committed Yet", "not.committed.yet", NULL, 0);
2061         strbuf_addstr(&msg, "tree 0000000000000000000000000000000000000000\n");
2062         for (parent = commit->parents; parent; parent = parent->next)
2063                 strbuf_addf(&msg, "parent %s\n",
2064                             sha1_to_hex(parent->item->object.sha1));
2065         strbuf_addf(&msg,
2066                     "author %s\n"
2067                     "committer %s\n\n"
2068                     "Version of %s from %s\n",
2069                     ident, ident, path,
2070                     (!contents_from ? path :
2071                      (!strcmp(contents_from, "-") ? "standard input" : contents_from)));
2072         commit->buffer = strbuf_detach(&msg, NULL);
2073
2074         if (!contents_from || strcmp("-", contents_from)) {
2075                 struct stat st;
2076                 const char *read_from;
2077                 char *buf_ptr;
2078                 unsigned long buf_len;
2079
2080                 if (contents_from) {
2081                         if (stat(contents_from, &st) < 0)
2082                                 die_errno("Cannot stat '%s'", contents_from);
2083                         read_from = contents_from;
2084                 }
2085                 else {
2086                         if (lstat(path, &st) < 0)
2087                                 die_errno("Cannot lstat '%s'", path);
2088                         read_from = path;
2089                 }
2090                 mode = canon_mode(st.st_mode);
2091
2092                 switch (st.st_mode & S_IFMT) {
2093                 case S_IFREG:
2094                         if (DIFF_OPT_TST(opt, ALLOW_TEXTCONV) &&
2095                             textconv_object(read_from, mode, null_sha1, 0, &buf_ptr, &buf_len))
2096                                 strbuf_attach(&buf, buf_ptr, buf_len, buf_len + 1);
2097                         else if (strbuf_read_file(&buf, read_from, st.st_size) != st.st_size)
2098                                 die_errno("cannot open or read '%s'", read_from);
2099                         break;
2100                 case S_IFLNK:
2101                         if (strbuf_readlink(&buf, read_from, st.st_size) < 0)
2102                                 die_errno("cannot readlink '%s'", read_from);
2103                         break;
2104                 default:
2105                         die("unsupported file type %s", read_from);
2106                 }
2107         }
2108         else {
2109                 /* Reading from stdin */
2110                 mode = 0;
2111                 if (strbuf_read(&buf, 0, 0) < 0)
2112                         die_errno("failed to read from stdin");
2113         }
2114         convert_to_git(path, buf.buf, buf.len, &buf, 0);
2115         origin->file.ptr = buf.buf;
2116         origin->file.size = buf.len;
2117         pretend_sha1_file(buf.buf, buf.len, OBJ_BLOB, origin->blob_sha1);
2118         commit->util = origin;
2119
2120         /*
2121          * Read the current index, replace the path entry with
2122          * origin->blob_sha1 without mucking with its mode or type
2123          * bits; we are not going to write this index out -- we just
2124          * want to run "diff-index --cached".
2125          */
2126         discard_cache();
2127         read_cache();
2128
2129         len = strlen(path);
2130         if (!mode) {
2131                 int pos = cache_name_pos(path, len);
2132                 if (0 <= pos)
2133                         mode = active_cache[pos]->ce_mode;
2134                 else
2135                         /* Let's not bother reading from HEAD tree */
2136                         mode = S_IFREG | 0644;
2137         }
2138         size = cache_entry_size(len);
2139         ce = xcalloc(1, size);
2140         hashcpy(ce->sha1, origin->blob_sha1);
2141         memcpy(ce->name, path, len);
2142         ce->ce_flags = create_ce_flags(0);
2143         ce->ce_namelen = len;
2144         ce->ce_mode = create_ce_mode(mode);
2145         add_cache_entry(ce, ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE);
2146
2147         /*
2148          * We are not going to write this out, so this does not matter
2149          * right now, but someday we might optimize diff-index --cached
2150          * with cache-tree information.
2151          */
2152         cache_tree_invalidate_path(active_cache_tree, path);
2153
2154         return commit;
2155 }
2156
2157 static const char *prepare_final(struct scoreboard *sb)
2158 {
2159         int i;
2160         const char *final_commit_name = NULL;
2161         struct rev_info *revs = sb->revs;
2162
2163         /*
2164          * There must be one and only one positive commit in the
2165          * revs->pending array.
2166          */
2167         for (i = 0; i < revs->pending.nr; i++) {
2168                 struct object *obj = revs->pending.objects[i].item;
2169                 if (obj->flags & UNINTERESTING)
2170                         continue;
2171                 while (obj->type == OBJ_TAG)
2172                         obj = deref_tag(obj, NULL, 0);
2173                 if (obj->type != OBJ_COMMIT)
2174                         die("Non commit %s?", revs->pending.objects[i].name);
2175                 if (sb->final)
2176                         die("More than one commit to dig from %s and %s?",
2177                             revs->pending.objects[i].name,
2178                             final_commit_name);
2179                 sb->final = (struct commit *) obj;
2180                 final_commit_name = revs->pending.objects[i].name;
2181         }
2182         return final_commit_name;
2183 }
2184
2185 static const char *prepare_initial(struct scoreboard *sb)
2186 {
2187         int i;
2188         const char *final_commit_name = NULL;
2189         struct rev_info *revs = sb->revs;
2190
2191         /*
2192          * There must be one and only one negative commit, and it must be
2193          * the boundary.
2194          */
2195         for (i = 0; i < revs->pending.nr; i++) {
2196                 struct object *obj = revs->pending.objects[i].item;
2197                 if (!(obj->flags & UNINTERESTING))
2198                         continue;
2199                 while (obj->type == OBJ_TAG)
2200                         obj = deref_tag(obj, NULL, 0);
2201                 if (obj->type != OBJ_COMMIT)
2202                         die("Non commit %s?", revs->pending.objects[i].name);
2203                 if (sb->final)
2204                         die("More than one commit to dig down to %s and %s?",
2205                             revs->pending.objects[i].name,
2206                             final_commit_name);
2207                 sb->final = (struct commit *) obj;
2208                 final_commit_name = revs->pending.objects[i].name;
2209         }
2210         if (!final_commit_name)
2211                 die("No commit to dig down to?");
2212         return final_commit_name;
2213 }
2214
2215 static int blame_copy_callback(const struct option *option, const char *arg, int unset)
2216 {
2217         int *opt = option->value;
2218
2219         /*
2220          * -C enables copy from removed files;
2221          * -C -C enables copy from existing files, but only
2222          *       when blaming a new file;
2223          * -C -C -C enables copy from existing files for
2224          *          everybody
2225          */
2226         if (*opt & PICKAXE_BLAME_COPY_HARDER)
2227                 *opt |= PICKAXE_BLAME_COPY_HARDEST;
2228         if (*opt & PICKAXE_BLAME_COPY)
2229                 *opt |= PICKAXE_BLAME_COPY_HARDER;
2230         *opt |= PICKAXE_BLAME_COPY | PICKAXE_BLAME_MOVE;
2231
2232         if (arg)
2233                 blame_copy_score = parse_score(arg);
2234         return 0;
2235 }
2236
2237 static int blame_move_callback(const struct option *option, const char *arg, int unset)
2238 {
2239         int *opt = option->value;
2240
2241         *opt |= PICKAXE_BLAME_MOVE;
2242
2243         if (arg)
2244                 blame_move_score = parse_score(arg);
2245         return 0;
2246 }
2247
2248 static int blame_bottomtop_callback(const struct option *option, const char *arg, int unset)
2249 {
2250         const char **bottomtop = option->value;
2251         if (!arg)
2252                 return -1;
2253         if (*bottomtop)
2254                 die("More than one '-L n,m' option given");
2255         *bottomtop = arg;
2256         return 0;
2257 }
2258
2259 int cmd_blame(int argc, const char **argv, const char *prefix)
2260 {
2261         struct rev_info revs;
2262         const char *path;
2263         struct scoreboard sb;
2264         struct origin *o;
2265         struct blame_entry *ent;
2266         long dashdash_pos, bottom, top, lno;
2267         const char *final_commit_name = NULL;
2268         enum object_type type;
2269
2270         static const char *bottomtop = NULL;
2271         static int output_option = 0, opt = 0;
2272         static int show_stats = 0;
2273         static const char *revs_file = NULL;
2274         static const char *contents_from = NULL;
2275         static const struct option options[] = {
2276                 OPT_BOOLEAN(0, "incremental", &incremental, N_("Show blame entries as we find them, incrementally")),
2277                 OPT_BOOLEAN('b', NULL, &blank_boundary, N_("Show blank SHA-1 for boundary commits (Default: off)")),
2278                 OPT_BOOLEAN(0, "root", &show_root, N_("Do not treat root commits as boundaries (Default: off)")),
2279                 OPT_BOOLEAN(0, "show-stats", &show_stats, N_("Show work cost statistics")),
2280                 OPT_BIT(0, "score-debug", &output_option, N_("Show output score for blame entries"), OUTPUT_SHOW_SCORE),
2281                 OPT_BIT('f', "show-name", &output_option, N_("Show original filename (Default: auto)"), OUTPUT_SHOW_NAME),
2282                 OPT_BIT('n', "show-number", &output_option, N_("Show original linenumber (Default: off)"), OUTPUT_SHOW_NUMBER),
2283                 OPT_BIT('p', "porcelain", &output_option, N_("Show in a format designed for machine consumption"), OUTPUT_PORCELAIN),
2284                 OPT_BIT(0, "line-porcelain", &output_option, N_("Show porcelain format with per-line commit information"), OUTPUT_PORCELAIN|OUTPUT_LINE_PORCELAIN),
2285                 OPT_BIT('c', NULL, &output_option, N_("Use the same output mode as git-annotate (Default: off)"), OUTPUT_ANNOTATE_COMPAT),
2286                 OPT_BIT('t', NULL, &output_option, N_("Show raw timestamp (Default: off)"), OUTPUT_RAW_TIMESTAMP),
2287                 OPT_BIT('l', NULL, &output_option, N_("Show long commit SHA1 (Default: off)"), OUTPUT_LONG_OBJECT_NAME),
2288                 OPT_BIT('s', NULL, &output_option, N_("Suppress author name and timestamp (Default: off)"), OUTPUT_NO_AUTHOR),
2289                 OPT_BIT('e', "show-email", &output_option, N_("Show author email instead of name (Default: off)"), OUTPUT_SHOW_EMAIL),
2290                 OPT_BIT('w', NULL, &xdl_opts, N_("Ignore whitespace differences"), XDF_IGNORE_WHITESPACE),
2291                 OPT_BIT(0, "minimal", &xdl_opts, N_("Spend extra cycles to find better match"), XDF_NEED_MINIMAL),
2292                 OPT_STRING('S', NULL, &revs_file, N_("file"), N_("Use revisions from <file> instead of calling git-rev-list")),
2293                 OPT_STRING(0, "contents", &contents_from, N_("file"), N_("Use <file>'s contents as the final image")),
2294                 { OPTION_CALLBACK, 'C', NULL, &opt, N_("score"), N_("Find line copies within and across files"), PARSE_OPT_OPTARG, blame_copy_callback },
2295                 { OPTION_CALLBACK, 'M', NULL, &opt, N_("score"), N_("Find line movements within and across files"), PARSE_OPT_OPTARG, blame_move_callback },
2296                 OPT_CALLBACK('L', NULL, &bottomtop, N_("n,m"), N_("Process only line range n,m, counting from 1"), blame_bottomtop_callback),
2297                 OPT__ABBREV(&abbrev),
2298                 OPT_END()
2299         };
2300
2301         struct parse_opt_ctx_t ctx;
2302         int cmd_is_annotate = !strcmp(argv[0], "annotate");
2303
2304         git_config(git_blame_config, NULL);
2305         init_revisions(&revs, NULL);
2306         revs.date_mode = blame_date_mode;
2307         DIFF_OPT_SET(&revs.diffopt, ALLOW_TEXTCONV);
2308         DIFF_OPT_SET(&revs.diffopt, FOLLOW_RENAMES);
2309
2310         save_commit_buffer = 0;
2311         dashdash_pos = 0;
2312
2313         parse_options_start(&ctx, argc, argv, prefix, options,
2314                             PARSE_OPT_KEEP_DASHDASH | PARSE_OPT_KEEP_ARGV0);
2315         for (;;) {
2316                 switch (parse_options_step(&ctx, options, blame_opt_usage)) {
2317                 case PARSE_OPT_HELP:
2318                         exit(129);
2319                 case PARSE_OPT_DONE:
2320                         if (ctx.argv[0])
2321                                 dashdash_pos = ctx.cpidx;
2322                         goto parse_done;
2323                 }
2324
2325                 if (!strcmp(ctx.argv[0], "--reverse")) {
2326                         ctx.argv[0] = "--children";
2327                         reverse = 1;
2328                 }
2329                 parse_revision_opt(&revs, &ctx, options, blame_opt_usage);
2330         }
2331 parse_done:
2332         no_whole_file_rename = !DIFF_OPT_TST(&revs.diffopt, FOLLOW_RENAMES);
2333         DIFF_OPT_CLR(&revs.diffopt, FOLLOW_RENAMES);
2334         argc = parse_options_end(&ctx);
2335
2336         if (0 < abbrev)
2337                 /* one more abbrev length is needed for the boundary commit */
2338                 abbrev++;
2339
2340         if (revs_file && read_ancestry(revs_file))
2341                 die_errno("reading graft file '%s' failed", revs_file);
2342
2343         if (cmd_is_annotate) {
2344                 output_option |= OUTPUT_ANNOTATE_COMPAT;
2345                 blame_date_mode = DATE_ISO8601;
2346         } else {
2347                 blame_date_mode = revs.date_mode;
2348         }
2349
2350         /* The maximum width used to show the dates */
2351         switch (blame_date_mode) {
2352         case DATE_RFC2822:
2353                 blame_date_width = sizeof("Thu, 19 Oct 2006 16:00:04 -0700");
2354                 break;
2355         case DATE_ISO8601:
2356                 blame_date_width = sizeof("2006-10-19 16:00:04 -0700");
2357                 break;
2358         case DATE_RAW:
2359                 blame_date_width = sizeof("1161298804 -0700");
2360                 break;
2361         case DATE_SHORT:
2362                 blame_date_width = sizeof("2006-10-19");
2363                 break;
2364         case DATE_RELATIVE:
2365                 /* "normal" is used as the fallback for "relative" */
2366         case DATE_LOCAL:
2367         case DATE_NORMAL:
2368                 blame_date_width = sizeof("Thu Oct 19 16:00:04 2006 -0700");
2369                 break;
2370         }
2371         blame_date_width -= 1; /* strip the null */
2372
2373         if (DIFF_OPT_TST(&revs.diffopt, FIND_COPIES_HARDER))
2374                 opt |= (PICKAXE_BLAME_COPY | PICKAXE_BLAME_MOVE |
2375                         PICKAXE_BLAME_COPY_HARDER);
2376
2377         if (!blame_move_score)
2378                 blame_move_score = BLAME_DEFAULT_MOVE_SCORE;
2379         if (!blame_copy_score)
2380                 blame_copy_score = BLAME_DEFAULT_COPY_SCORE;
2381
2382         /*
2383          * We have collected options unknown to us in argv[1..unk]
2384          * which are to be passed to revision machinery if we are
2385          * going to do the "bottom" processing.
2386          *
2387          * The remaining are:
2388          *
2389          * (1) if dashdash_pos != 0, it is either
2390          *     "blame [revisions] -- <path>" or
2391          *     "blame -- <path> <rev>"
2392          *
2393          * (2) otherwise, it is one of the two:
2394          *     "blame [revisions] <path>"
2395          *     "blame <path> <rev>"
2396          *
2397          * Note that we must strip out <path> from the arguments: we do not
2398          * want the path pruning but we may want "bottom" processing.
2399          */
2400         if (dashdash_pos) {
2401                 switch (argc - dashdash_pos - 1) {
2402                 case 2: /* (1b) */
2403                         if (argc != 4)
2404                                 usage_with_options(blame_opt_usage, options);
2405                         /* reorder for the new way: <rev> -- <path> */
2406                         argv[1] = argv[3];
2407                         argv[3] = argv[2];
2408                         argv[2] = "--";
2409                         /* FALLTHROUGH */
2410                 case 1: /* (1a) */
2411                         path = add_prefix(prefix, argv[--argc]);
2412                         argv[argc] = NULL;
2413                         break;
2414                 default:
2415                         usage_with_options(blame_opt_usage, options);
2416                 }
2417         } else {
2418                 if (argc < 2)
2419                         usage_with_options(blame_opt_usage, options);
2420                 path = add_prefix(prefix, argv[argc - 1]);
2421                 if (argc == 3 && !has_string_in_work_tree(path)) { /* (2b) */
2422                         path = add_prefix(prefix, argv[1]);
2423                         argv[1] = argv[2];
2424                 }
2425                 argv[argc - 1] = "--";
2426
2427                 setup_work_tree();
2428                 if (!has_string_in_work_tree(path))
2429                         die_errno("cannot stat path '%s'", path);
2430         }
2431
2432         revs.disable_stdin = 1;
2433         setup_revisions(argc, argv, &revs, NULL);
2434         memset(&sb, 0, sizeof(sb));
2435
2436         sb.revs = &revs;
2437         if (!reverse)
2438                 final_commit_name = prepare_final(&sb);
2439         else if (contents_from)
2440                 die("--contents and --children do not blend well.");
2441         else
2442                 final_commit_name = prepare_initial(&sb);
2443
2444         if (!sb.final) {
2445                 /*
2446                  * "--not A B -- path" without anything positive;
2447                  * do not default to HEAD, but use the working tree
2448                  * or "--contents".
2449                  */
2450                 setup_work_tree();
2451                 sb.final = fake_working_tree_commit(&sb.revs->diffopt,
2452                                                     path, contents_from);
2453                 add_pending_object(&revs, &(sb.final->object), ":");
2454         }
2455         else if (contents_from)
2456                 die("Cannot use --contents with final commit object name");
2457
2458         /*
2459          * If we have bottom, this will mark the ancestors of the
2460          * bottom commits we would reach while traversing as
2461          * uninteresting.
2462          */
2463         if (prepare_revision_walk(&revs))
2464                 die("revision walk setup failed");
2465
2466         if (is_null_sha1(sb.final->object.sha1)) {
2467                 char *buf;
2468                 o = sb.final->util;
2469                 buf = xmalloc(o->file.size + 1);
2470                 memcpy(buf, o->file.ptr, o->file.size + 1);
2471                 sb.final_buf = buf;
2472                 sb.final_buf_size = o->file.size;
2473         }
2474         else {
2475                 o = get_origin(&sb, sb.final, path);
2476                 if (fill_blob_sha1_and_mode(o))
2477                         die("no such path %s in %s", path, final_commit_name);
2478
2479                 if (DIFF_OPT_TST(&sb.revs->diffopt, ALLOW_TEXTCONV) &&
2480                     textconv_object(path, o->mode, o->blob_sha1, 1, (char **) &sb.final_buf,
2481                                     &sb.final_buf_size))
2482                         ;
2483                 else
2484                         sb.final_buf = read_sha1_file(o->blob_sha1, &type,
2485                                                       &sb.final_buf_size);
2486
2487                 if (!sb.final_buf)
2488                         die("Cannot read blob %s for path %s",
2489                             sha1_to_hex(o->blob_sha1),
2490                             path);
2491         }
2492         num_read_blob++;
2493         lno = prepare_lines(&sb);
2494
2495         bottom = top = 0;
2496         if (bottomtop)
2497                 prepare_blame_range(&sb, bottomtop, lno, &bottom, &top);
2498         if (bottom < 1)
2499                 bottom = 1;
2500         if (top < 1)
2501                 top = lno;
2502         bottom--;
2503         if (lno < top || lno < bottom)
2504                 die("file %s has only %lu lines", path, lno);
2505
2506         ent = xcalloc(1, sizeof(*ent));
2507         ent->lno = bottom;
2508         ent->num_lines = top - bottom;
2509         ent->suspect = o;
2510         ent->s_lno = bottom;
2511
2512         sb.ent = ent;
2513         sb.path = path;
2514
2515         read_mailmap(&mailmap, NULL);
2516
2517         if (!incremental)
2518                 setup_pager();
2519
2520         assign_blame(&sb, opt);
2521
2522         if (incremental)
2523                 return 0;
2524
2525         coalesce(&sb);
2526
2527         if (!(output_option & OUTPUT_PORCELAIN))
2528                 find_alignment(&sb, &output_option);
2529
2530         output(&sb, output_option);
2531         free((void *)sb.final_buf);
2532         for (ent = sb.ent; ent; ) {
2533                 struct blame_entry *e = ent->next;
2534                 free(ent);
2535                 ent = e;
2536         }
2537
2538         if (show_stats) {
2539                 printf("num read blob: %d\n", num_read_blob);
2540                 printf("num get patch: %d\n", num_get_patch);
2541                 printf("num commits: %d\n", num_commits);
2542         }
2543         return 0;
2544 }