Merge branch 'jc/fix-diff-no-index-diff-opt-parse' into maint
[git] / builtin / blame.c
1 /*
2  * Blame
3  *
4  * Copyright (c) 2006, Junio C Hamano
5  */
6
7 #include "cache.h"
8 #include "builtin.h"
9 #include "blob.h"
10 #include "commit.h"
11 #include "tag.h"
12 #include "tree-walk.h"
13 #include "diff.h"
14 #include "diffcore.h"
15 #include "revision.h"
16 #include "quote.h"
17 #include "xdiff-interface.h"
18 #include "cache-tree.h"
19 #include "string-list.h"
20 #include "mailmap.h"
21 #include "parse-options.h"
22 #include "utf8.h"
23 #include "userdiff.h"
24 #include "line-range.h"
25 #include "line-log.h"
26
27 static char blame_usage[] = N_("git blame [options] [rev-opts] [rev] [--] file");
28
29 static const char *blame_opt_usage[] = {
30         blame_usage,
31         "",
32         N_("[rev-opts] are documented in git-rev-list(1)"),
33         NULL
34 };
35
36 static int longest_file;
37 static int longest_author;
38 static int max_orig_digits;
39 static int max_digits;
40 static int max_score_digits;
41 static int show_root;
42 static int reverse;
43 static int blank_boundary;
44 static int incremental;
45 static int xdl_opts;
46 static int abbrev = -1;
47 static int no_whole_file_rename;
48
49 static enum date_mode blame_date_mode = DATE_ISO8601;
50 static size_t blame_date_width;
51
52 static struct string_list mailmap;
53
54 #ifndef DEBUG
55 #define DEBUG 0
56 #endif
57
58 /* stats */
59 static int num_read_blob;
60 static int num_get_patch;
61 static int num_commits;
62
63 #define PICKAXE_BLAME_MOVE              01
64 #define PICKAXE_BLAME_COPY              02
65 #define PICKAXE_BLAME_COPY_HARDER       04
66 #define PICKAXE_BLAME_COPY_HARDEST      010
67
68 /*
69  * blame for a blame_entry with score lower than these thresholds
70  * is not passed to the parent using move/copy logic.
71  */
72 static unsigned blame_move_score;
73 static unsigned blame_copy_score;
74 #define BLAME_DEFAULT_MOVE_SCORE        20
75 #define BLAME_DEFAULT_COPY_SCORE        40
76
77 /* bits #0..7 in revision.h, #8..11 used for merge_bases() in commit.c */
78 #define METAINFO_SHOWN          (1u<<12)
79 #define MORE_THAN_ONE_PATH      (1u<<13)
80
81 /*
82  * One blob in a commit that is being suspected
83  */
84 struct origin {
85         int refcnt;
86         struct origin *previous;
87         struct commit *commit;
88         mmfile_t file;
89         unsigned char blob_sha1[20];
90         unsigned mode;
91         char path[FLEX_ARRAY];
92 };
93
94 static int diff_hunks(mmfile_t *file_a, mmfile_t *file_b, long ctxlen,
95                       xdl_emit_hunk_consume_func_t hunk_func, void *cb_data)
96 {
97         xpparam_t xpp = {0};
98         xdemitconf_t xecfg = {0};
99         xdemitcb_t ecb = {NULL};
100
101         xpp.flags = xdl_opts;
102         xecfg.ctxlen = ctxlen;
103         xecfg.hunk_func = hunk_func;
104         ecb.priv = cb_data;
105         return xdi_diff(file_a, file_b, &xpp, &xecfg, &ecb);
106 }
107
108 /*
109  * Prepare diff_filespec and convert it using diff textconv API
110  * if the textconv driver exists.
111  * Return 1 if the conversion succeeds, 0 otherwise.
112  */
113 int textconv_object(const char *path,
114                     unsigned mode,
115                     const unsigned char *sha1,
116                     int sha1_valid,
117                     char **buf,
118                     unsigned long *buf_size)
119 {
120         struct diff_filespec *df;
121         struct userdiff_driver *textconv;
122
123         df = alloc_filespec(path);
124         fill_filespec(df, sha1, sha1_valid, mode);
125         textconv = get_textconv(df);
126         if (!textconv) {
127                 free_filespec(df);
128                 return 0;
129         }
130
131         *buf_size = fill_textconv(textconv, df, buf);
132         free_filespec(df);
133         return 1;
134 }
135
136 /*
137  * Given an origin, prepare mmfile_t structure to be used by the
138  * diff machinery
139  */
140 static void fill_origin_blob(struct diff_options *opt,
141                              struct origin *o, mmfile_t *file)
142 {
143         if (!o->file.ptr) {
144                 enum object_type type;
145                 unsigned long file_size;
146
147                 num_read_blob++;
148                 if (DIFF_OPT_TST(opt, ALLOW_TEXTCONV) &&
149                     textconv_object(o->path, o->mode, o->blob_sha1, 1, &file->ptr, &file_size))
150                         ;
151                 else
152                         file->ptr = read_sha1_file(o->blob_sha1, &type, &file_size);
153                 file->size = file_size;
154
155                 if (!file->ptr)
156                         die("Cannot read blob %s for path %s",
157                             sha1_to_hex(o->blob_sha1),
158                             o->path);
159                 o->file = *file;
160         }
161         else
162                 *file = o->file;
163 }
164
165 /*
166  * Origin is refcounted and usually we keep the blob contents to be
167  * reused.
168  */
169 static inline struct origin *origin_incref(struct origin *o)
170 {
171         if (o)
172                 o->refcnt++;
173         return o;
174 }
175
176 static void origin_decref(struct origin *o)
177 {
178         if (o && --o->refcnt <= 0) {
179                 if (o->previous)
180                         origin_decref(o->previous);
181                 free(o->file.ptr);
182                 free(o);
183         }
184 }
185
186 static void drop_origin_blob(struct origin *o)
187 {
188         if (o->file.ptr) {
189                 free(o->file.ptr);
190                 o->file.ptr = NULL;
191         }
192 }
193
194 /*
195  * Each group of lines is described by a blame_entry; it can be split
196  * as we pass blame to the parents.  They form a linked list in the
197  * scoreboard structure, sorted by the target line number.
198  */
199 struct blame_entry {
200         struct blame_entry *prev;
201         struct blame_entry *next;
202
203         /* the first line of this group in the final image;
204          * internally all line numbers are 0 based.
205          */
206         int lno;
207
208         /* how many lines this group has */
209         int num_lines;
210
211         /* the commit that introduced this group into the final image */
212         struct origin *suspect;
213
214         /* true if the suspect is truly guilty; false while we have not
215          * checked if the group came from one of its parents.
216          */
217         char guilty;
218
219         /* true if the entry has been scanned for copies in the current parent
220          */
221         char scanned;
222
223         /* the line number of the first line of this group in the
224          * suspect's file; internally all line numbers are 0 based.
225          */
226         int s_lno;
227
228         /* how significant this entry is -- cached to avoid
229          * scanning the lines over and over.
230          */
231         unsigned score;
232 };
233
234 /*
235  * The current state of the blame assignment.
236  */
237 struct scoreboard {
238         /* the final commit (i.e. where we started digging from) */
239         struct commit *final;
240         struct rev_info *revs;
241         const char *path;
242
243         /*
244          * The contents in the final image.
245          * Used by many functions to obtain contents of the nth line,
246          * indexed with scoreboard.lineno[blame_entry.lno].
247          */
248         const char *final_buf;
249         unsigned long final_buf_size;
250
251         /* linked list of blames */
252         struct blame_entry *ent;
253
254         /* look-up a line in the final buffer */
255         int num_lines;
256         int *lineno;
257 };
258
259 static inline int same_suspect(struct origin *a, struct origin *b)
260 {
261         if (a == b)
262                 return 1;
263         if (a->commit != b->commit)
264                 return 0;
265         return !strcmp(a->path, b->path);
266 }
267
268 static void sanity_check_refcnt(struct scoreboard *);
269
270 /*
271  * If two blame entries that are next to each other came from
272  * contiguous lines in the same origin (i.e. <commit, path> pair),
273  * merge them together.
274  */
275 static void coalesce(struct scoreboard *sb)
276 {
277         struct blame_entry *ent, *next;
278
279         for (ent = sb->ent; ent && (next = ent->next); ent = next) {
280                 if (same_suspect(ent->suspect, next->suspect) &&
281                     ent->guilty == next->guilty &&
282                     ent->s_lno + ent->num_lines == next->s_lno) {
283                         ent->num_lines += next->num_lines;
284                         ent->next = next->next;
285                         if (ent->next)
286                                 ent->next->prev = ent;
287                         origin_decref(next->suspect);
288                         free(next);
289                         ent->score = 0;
290                         next = ent; /* again */
291                 }
292         }
293
294         if (DEBUG) /* sanity */
295                 sanity_check_refcnt(sb);
296 }
297
298 /*
299  * Given a commit and a path in it, create a new origin structure.
300  * The callers that add blame to the scoreboard should use
301  * get_origin() to obtain shared, refcounted copy instead of calling
302  * this function directly.
303  */
304 static struct origin *make_origin(struct commit *commit, const char *path)
305 {
306         struct origin *o;
307         o = xcalloc(1, sizeof(*o) + strlen(path) + 1);
308         o->commit = commit;
309         o->refcnt = 1;
310         strcpy(o->path, path);
311         return o;
312 }
313
314 /*
315  * Locate an existing origin or create a new one.
316  */
317 static struct origin *get_origin(struct scoreboard *sb,
318                                  struct commit *commit,
319                                  const char *path)
320 {
321         struct blame_entry *e;
322
323         for (e = sb->ent; e; e = e->next) {
324                 if (e->suspect->commit == commit &&
325                     !strcmp(e->suspect->path, path))
326                         return origin_incref(e->suspect);
327         }
328         return make_origin(commit, path);
329 }
330
331 /*
332  * Fill the blob_sha1 field of an origin if it hasn't, so that later
333  * call to fill_origin_blob() can use it to locate the data.  blob_sha1
334  * for an origin is also used to pass the blame for the entire file to
335  * the parent to detect the case where a child's blob is identical to
336  * that of its parent's.
337  *
338  * This also fills origin->mode for corresponding tree path.
339  */
340 static int fill_blob_sha1_and_mode(struct origin *origin)
341 {
342         if (!is_null_sha1(origin->blob_sha1))
343                 return 0;
344         if (get_tree_entry(origin->commit->object.sha1,
345                            origin->path,
346                            origin->blob_sha1, &origin->mode))
347                 goto error_out;
348         if (sha1_object_info(origin->blob_sha1, NULL) != OBJ_BLOB)
349                 goto error_out;
350         return 0;
351  error_out:
352         hashclr(origin->blob_sha1);
353         origin->mode = S_IFINVALID;
354         return -1;
355 }
356
357 /*
358  * We have an origin -- check if the same path exists in the
359  * parent and return an origin structure to represent it.
360  */
361 static struct origin *find_origin(struct scoreboard *sb,
362                                   struct commit *parent,
363                                   struct origin *origin)
364 {
365         struct origin *porigin = NULL;
366         struct diff_options diff_opts;
367         const char *paths[2];
368
369         if (parent->util) {
370                 /*
371                  * Each commit object can cache one origin in that
372                  * commit.  This is a freestanding copy of origin and
373                  * not refcounted.
374                  */
375                 struct origin *cached = parent->util;
376                 if (!strcmp(cached->path, origin->path)) {
377                         /*
378                          * The same path between origin and its parent
379                          * without renaming -- the most common case.
380                          */
381                         porigin = get_origin(sb, parent, cached->path);
382
383                         /*
384                          * If the origin was newly created (i.e. get_origin
385                          * would call make_origin if none is found in the
386                          * scoreboard), it does not know the blob_sha1/mode,
387                          * so copy it.  Otherwise porigin was in the
388                          * scoreboard and already knows blob_sha1/mode.
389                          */
390                         if (porigin->refcnt == 1) {
391                                 hashcpy(porigin->blob_sha1, cached->blob_sha1);
392                                 porigin->mode = cached->mode;
393                         }
394                         return porigin;
395                 }
396                 /* otherwise it was not very useful; free it */
397                 free(parent->util);
398                 parent->util = NULL;
399         }
400
401         /* See if the origin->path is different between parent
402          * and origin first.  Most of the time they are the
403          * same and diff-tree is fairly efficient about this.
404          */
405         diff_setup(&diff_opts);
406         DIFF_OPT_SET(&diff_opts, RECURSIVE);
407         diff_opts.detect_rename = 0;
408         diff_opts.output_format = DIFF_FORMAT_NO_OUTPUT;
409         paths[0] = origin->path;
410         paths[1] = NULL;
411
412         parse_pathspec(&diff_opts.pathspec,
413                        PATHSPEC_ALL_MAGIC & ~PATHSPEC_LITERAL,
414                        PATHSPEC_LITERAL_PATH, "", paths);
415         diff_setup_done(&diff_opts);
416
417         if (is_null_sha1(origin->commit->object.sha1))
418                 do_diff_cache(parent->tree->object.sha1, &diff_opts);
419         else
420                 diff_tree_sha1(parent->tree->object.sha1,
421                                origin->commit->tree->object.sha1,
422                                "", &diff_opts);
423         diffcore_std(&diff_opts);
424
425         if (!diff_queued_diff.nr) {
426                 /* The path is the same as parent */
427                 porigin = get_origin(sb, parent, origin->path);
428                 hashcpy(porigin->blob_sha1, origin->blob_sha1);
429                 porigin->mode = origin->mode;
430         } else {
431                 /*
432                  * Since origin->path is a pathspec, if the parent
433                  * commit had it as a directory, we will see a whole
434                  * bunch of deletion of files in the directory that we
435                  * do not care about.
436                  */
437                 int i;
438                 struct diff_filepair *p = NULL;
439                 for (i = 0; i < diff_queued_diff.nr; i++) {
440                         const char *name;
441                         p = diff_queued_diff.queue[i];
442                         name = p->one->path ? p->one->path : p->two->path;
443                         if (!strcmp(name, origin->path))
444                                 break;
445                 }
446                 if (!p)
447                         die("internal error in blame::find_origin");
448                 switch (p->status) {
449                 default:
450                         die("internal error in blame::find_origin (%c)",
451                             p->status);
452                 case 'M':
453                         porigin = get_origin(sb, parent, origin->path);
454                         hashcpy(porigin->blob_sha1, p->one->sha1);
455                         porigin->mode = p->one->mode;
456                         break;
457                 case 'A':
458                 case 'T':
459                         /* Did not exist in parent, or type changed */
460                         break;
461                 }
462         }
463         diff_flush(&diff_opts);
464         free_pathspec(&diff_opts.pathspec);
465         if (porigin) {
466                 /*
467                  * Create a freestanding copy that is not part of
468                  * the refcounted origin found in the scoreboard, and
469                  * cache it in the commit.
470                  */
471                 struct origin *cached;
472
473                 cached = make_origin(porigin->commit, porigin->path);
474                 hashcpy(cached->blob_sha1, porigin->blob_sha1);
475                 cached->mode = porigin->mode;
476                 parent->util = cached;
477         }
478         return porigin;
479 }
480
481 /*
482  * We have an origin -- find the path that corresponds to it in its
483  * parent and return an origin structure to represent it.
484  */
485 static struct origin *find_rename(struct scoreboard *sb,
486                                   struct commit *parent,
487                                   struct origin *origin)
488 {
489         struct origin *porigin = NULL;
490         struct diff_options diff_opts;
491         int i;
492
493         diff_setup(&diff_opts);
494         DIFF_OPT_SET(&diff_opts, RECURSIVE);
495         diff_opts.detect_rename = DIFF_DETECT_RENAME;
496         diff_opts.output_format = DIFF_FORMAT_NO_OUTPUT;
497         diff_opts.single_follow = origin->path;
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         free_pathspec(&diff_opts.pathspec);
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         int i, j;
1068         int retval;
1069         struct blame_list *blame_list;
1070         int num_ents;
1071
1072         blame_list = setup_blame_list(sb, target, blame_copy_score, &num_ents);
1073         if (!blame_list)
1074                 return 1; /* nothing remains for this target */
1075
1076         diff_setup(&diff_opts);
1077         DIFF_OPT_SET(&diff_opts, RECURSIVE);
1078         diff_opts.output_format = DIFF_FORMAT_NO_OUTPUT;
1079
1080         diff_setup_done(&diff_opts);
1081
1082         /* Try "find copies harder" on new path if requested;
1083          * we do not want to use diffcore_rename() actually to
1084          * match things up; find_copies_harder is set only to
1085          * force diff_tree_sha1() to feed all filepairs to diff_queue,
1086          * and this code needs to be after diff_setup_done(), which
1087          * usually makes find-copies-harder imply copy detection.
1088          */
1089         if ((opt & PICKAXE_BLAME_COPY_HARDEST)
1090             || ((opt & PICKAXE_BLAME_COPY_HARDER)
1091                 && (!porigin || strcmp(target->path, porigin->path))))
1092                 DIFF_OPT_SET(&diff_opts, FIND_COPIES_HARDER);
1093
1094         if (is_null_sha1(target->commit->object.sha1))
1095                 do_diff_cache(parent->tree->object.sha1, &diff_opts);
1096         else
1097                 diff_tree_sha1(parent->tree->object.sha1,
1098                                target->commit->tree->object.sha1,
1099                                "", &diff_opts);
1100
1101         if (!DIFF_OPT_TST(&diff_opts, FIND_COPIES_HARDER))
1102                 diffcore_std(&diff_opts);
1103
1104         retval = 0;
1105         while (1) {
1106                 int made_progress = 0;
1107
1108                 for (i = 0; i < diff_queued_diff.nr; i++) {
1109                         struct diff_filepair *p = diff_queued_diff.queue[i];
1110                         struct origin *norigin;
1111                         mmfile_t file_p;
1112                         struct blame_entry this[3];
1113
1114                         if (!DIFF_FILE_VALID(p->one))
1115                                 continue; /* does not exist in parent */
1116                         if (S_ISGITLINK(p->one->mode))
1117                                 continue; /* ignore git links */
1118                         if (porigin && !strcmp(p->one->path, porigin->path))
1119                                 /* find_move already dealt with this path */
1120                                 continue;
1121
1122                         norigin = get_origin(sb, parent, p->one->path);
1123                         hashcpy(norigin->blob_sha1, p->one->sha1);
1124                         norigin->mode = p->one->mode;
1125                         fill_origin_blob(&sb->revs->diffopt, norigin, &file_p);
1126                         if (!file_p.ptr)
1127                                 continue;
1128
1129                         for (j = 0; j < num_ents; j++) {
1130                                 find_copy_in_blob(sb, blame_list[j].ent,
1131                                                   norigin, this, &file_p);
1132                                 copy_split_if_better(sb, blame_list[j].split,
1133                                                      this);
1134                                 decref_split(this);
1135                         }
1136                         origin_decref(norigin);
1137                 }
1138
1139                 for (j = 0; j < num_ents; j++) {
1140                         struct blame_entry *split = blame_list[j].split;
1141                         if (split[1].suspect &&
1142                             blame_copy_score < ent_score(sb, &split[1])) {
1143                                 split_blame(sb, split, blame_list[j].ent);
1144                                 made_progress = 1;
1145                         }
1146                         else
1147                                 blame_list[j].ent->scanned = 1;
1148                         decref_split(split);
1149                 }
1150                 free(blame_list);
1151
1152                 if (!made_progress)
1153                         break;
1154                 blame_list = setup_blame_list(sb, target, blame_copy_score, &num_ents);
1155                 if (!blame_list) {
1156                         retval = 1;
1157                         break;
1158                 }
1159         }
1160         reset_scanned_flag(sb);
1161         diff_flush(&diff_opts);
1162         free_pathspec(&diff_opts.pathspec);
1163         return retval;
1164 }
1165
1166 /*
1167  * The blobs of origin and porigin exactly match, so everything
1168  * origin is suspected for can be blamed on the parent.
1169  */
1170 static void pass_whole_blame(struct scoreboard *sb,
1171                              struct origin *origin, struct origin *porigin)
1172 {
1173         struct blame_entry *e;
1174
1175         if (!porigin->file.ptr && origin->file.ptr) {
1176                 /* Steal its file */
1177                 porigin->file = origin->file;
1178                 origin->file.ptr = NULL;
1179         }
1180         for (e = sb->ent; e; e = e->next) {
1181                 if (!same_suspect(e->suspect, origin))
1182                         continue;
1183                 origin_incref(porigin);
1184                 origin_decref(e->suspect);
1185                 e->suspect = porigin;
1186         }
1187 }
1188
1189 /*
1190  * We pass blame from the current commit to its parents.  We keep saying
1191  * "parent" (and "porigin"), but what we mean is to find scapegoat to
1192  * exonerate ourselves.
1193  */
1194 static struct commit_list *first_scapegoat(struct rev_info *revs, struct commit *commit)
1195 {
1196         if (!reverse)
1197                 return commit->parents;
1198         return lookup_decoration(&revs->children, &commit->object);
1199 }
1200
1201 static int num_scapegoats(struct rev_info *revs, struct commit *commit)
1202 {
1203         int cnt;
1204         struct commit_list *l = first_scapegoat(revs, commit);
1205         for (cnt = 0; l; l = l->next)
1206                 cnt++;
1207         return cnt;
1208 }
1209
1210 #define MAXSG 16
1211
1212 static void pass_blame(struct scoreboard *sb, struct origin *origin, int opt)
1213 {
1214         struct rev_info *revs = sb->revs;
1215         int i, pass, num_sg;
1216         struct commit *commit = origin->commit;
1217         struct commit_list *sg;
1218         struct origin *sg_buf[MAXSG];
1219         struct origin *porigin, **sg_origin = sg_buf;
1220
1221         num_sg = num_scapegoats(revs, commit);
1222         if (!num_sg)
1223                 goto finish;
1224         else if (num_sg < ARRAY_SIZE(sg_buf))
1225                 memset(sg_buf, 0, sizeof(sg_buf));
1226         else
1227                 sg_origin = xcalloc(num_sg, sizeof(*sg_origin));
1228
1229         /*
1230          * The first pass looks for unrenamed path to optimize for
1231          * common cases, then we look for renames in the second pass.
1232          */
1233         for (pass = 0; pass < 2 - no_whole_file_rename; pass++) {
1234                 struct origin *(*find)(struct scoreboard *,
1235                                        struct commit *, struct origin *);
1236                 find = pass ? find_rename : find_origin;
1237
1238                 for (i = 0, sg = first_scapegoat(revs, commit);
1239                      i < num_sg && sg;
1240                      sg = sg->next, i++) {
1241                         struct commit *p = sg->item;
1242                         int j, same;
1243
1244                         if (sg_origin[i])
1245                                 continue;
1246                         if (parse_commit(p))
1247                                 continue;
1248                         porigin = find(sb, p, origin);
1249                         if (!porigin)
1250                                 continue;
1251                         if (!hashcmp(porigin->blob_sha1, origin->blob_sha1)) {
1252                                 pass_whole_blame(sb, origin, porigin);
1253                                 origin_decref(porigin);
1254                                 goto finish;
1255                         }
1256                         for (j = same = 0; j < i; j++)
1257                                 if (sg_origin[j] &&
1258                                     !hashcmp(sg_origin[j]->blob_sha1,
1259                                              porigin->blob_sha1)) {
1260                                         same = 1;
1261                                         break;
1262                                 }
1263                         if (!same)
1264                                 sg_origin[i] = porigin;
1265                         else
1266                                 origin_decref(porigin);
1267                 }
1268         }
1269
1270         num_commits++;
1271         for (i = 0, sg = first_scapegoat(revs, commit);
1272              i < num_sg && sg;
1273              sg = sg->next, i++) {
1274                 struct origin *porigin = sg_origin[i];
1275                 if (!porigin)
1276                         continue;
1277                 if (!origin->previous) {
1278                         origin_incref(porigin);
1279                         origin->previous = porigin;
1280                 }
1281                 if (pass_blame_to_parent(sb, origin, porigin))
1282                         goto finish;
1283         }
1284
1285         /*
1286          * Optionally find moves in parents' files.
1287          */
1288         if (opt & PICKAXE_BLAME_MOVE)
1289                 for (i = 0, sg = first_scapegoat(revs, commit);
1290                      i < num_sg && sg;
1291                      sg = sg->next, i++) {
1292                         struct origin *porigin = sg_origin[i];
1293                         if (!porigin)
1294                                 continue;
1295                         if (find_move_in_parent(sb, origin, porigin))
1296                                 goto finish;
1297                 }
1298
1299         /*
1300          * Optionally find copies from parents' files.
1301          */
1302         if (opt & PICKAXE_BLAME_COPY)
1303                 for (i = 0, sg = first_scapegoat(revs, commit);
1304                      i < num_sg && sg;
1305                      sg = sg->next, i++) {
1306                         struct origin *porigin = sg_origin[i];
1307                         if (find_copy_in_parent(sb, origin, sg->item,
1308                                                 porigin, opt))
1309                                 goto finish;
1310                 }
1311
1312  finish:
1313         for (i = 0; i < num_sg; i++) {
1314                 if (sg_origin[i]) {
1315                         drop_origin_blob(sg_origin[i]);
1316                         origin_decref(sg_origin[i]);
1317                 }
1318         }
1319         drop_origin_blob(origin);
1320         if (sg_buf != sg_origin)
1321                 free(sg_origin);
1322 }
1323
1324 /*
1325  * Information on commits, used for output.
1326  */
1327 struct commit_info {
1328         struct strbuf author;
1329         struct strbuf author_mail;
1330         unsigned long author_time;
1331         struct strbuf author_tz;
1332
1333         /* filled only when asked for details */
1334         struct strbuf committer;
1335         struct strbuf committer_mail;
1336         unsigned long committer_time;
1337         struct strbuf committer_tz;
1338
1339         struct strbuf summary;
1340 };
1341
1342 /*
1343  * Parse author/committer line in the commit object buffer
1344  */
1345 static void get_ac_line(const char *inbuf, const char *what,
1346         struct strbuf *name, struct strbuf *mail,
1347         unsigned long *time, struct strbuf *tz)
1348 {
1349         struct ident_split ident;
1350         size_t len, maillen, namelen;
1351         char *tmp, *endp;
1352         const char *namebuf, *mailbuf;
1353
1354         tmp = strstr(inbuf, what);
1355         if (!tmp)
1356                 goto error_out;
1357         tmp += strlen(what);
1358         endp = strchr(tmp, '\n');
1359         if (!endp)
1360                 len = strlen(tmp);
1361         else
1362                 len = endp - tmp;
1363
1364         if (split_ident_line(&ident, tmp, len)) {
1365         error_out:
1366                 /* Ugh */
1367                 tmp = "(unknown)";
1368                 strbuf_addstr(name, tmp);
1369                 strbuf_addstr(mail, tmp);
1370                 strbuf_addstr(tz, tmp);
1371                 *time = 0;
1372                 return;
1373         }
1374
1375         namelen = ident.name_end - ident.name_begin;
1376         namebuf = ident.name_begin;
1377
1378         maillen = ident.mail_end - ident.mail_begin;
1379         mailbuf = ident.mail_begin;
1380
1381         if (ident.date_begin && ident.date_end)
1382                 *time = strtoul(ident.date_begin, NULL, 10);
1383         else
1384                 *time = 0;
1385
1386         if (ident.tz_begin && ident.tz_end)
1387                 strbuf_add(tz, ident.tz_begin, ident.tz_end - ident.tz_begin);
1388         else
1389                 strbuf_addstr(tz, "(unknown)");
1390
1391         /*
1392          * Now, convert both name and e-mail using mailmap
1393          */
1394         map_user(&mailmap, &mailbuf, &maillen,
1395                  &namebuf, &namelen);
1396
1397         strbuf_addf(mail, "<%.*s>", (int)maillen, mailbuf);
1398         strbuf_add(name, namebuf, namelen);
1399 }
1400
1401 static void commit_info_init(struct commit_info *ci)
1402 {
1403
1404         strbuf_init(&ci->author, 0);
1405         strbuf_init(&ci->author_mail, 0);
1406         strbuf_init(&ci->author_tz, 0);
1407         strbuf_init(&ci->committer, 0);
1408         strbuf_init(&ci->committer_mail, 0);
1409         strbuf_init(&ci->committer_tz, 0);
1410         strbuf_init(&ci->summary, 0);
1411 }
1412
1413 static void commit_info_destroy(struct commit_info *ci)
1414 {
1415
1416         strbuf_release(&ci->author);
1417         strbuf_release(&ci->author_mail);
1418         strbuf_release(&ci->author_tz);
1419         strbuf_release(&ci->committer);
1420         strbuf_release(&ci->committer_mail);
1421         strbuf_release(&ci->committer_tz);
1422         strbuf_release(&ci->summary);
1423 }
1424
1425 static void get_commit_info(struct commit *commit,
1426                             struct commit_info *ret,
1427                             int detailed)
1428 {
1429         int len;
1430         const char *subject, *encoding;
1431         char *message;
1432
1433         commit_info_init(ret);
1434
1435         encoding = get_log_output_encoding();
1436         message = logmsg_reencode(commit, NULL, encoding);
1437         get_ac_line(message, "\nauthor ",
1438                     &ret->author, &ret->author_mail,
1439                     &ret->author_time, &ret->author_tz);
1440
1441         if (!detailed) {
1442                 logmsg_free(message, commit);
1443                 return;
1444         }
1445
1446         get_ac_line(message, "\ncommitter ",
1447                     &ret->committer, &ret->committer_mail,
1448                     &ret->committer_time, &ret->committer_tz);
1449
1450         len = find_commit_subject(message, &subject);
1451         if (len)
1452                 strbuf_add(&ret->summary, subject, len);
1453         else
1454                 strbuf_addf(&ret->summary, "(%s)", sha1_to_hex(commit->object.sha1));
1455
1456         logmsg_free(message, commit);
1457 }
1458
1459 /*
1460  * To allow LF and other nonportable characters in pathnames,
1461  * they are c-style quoted as needed.
1462  */
1463 static void write_filename_info(const char *path)
1464 {
1465         printf("filename ");
1466         write_name_quoted(path, stdout, '\n');
1467 }
1468
1469 /*
1470  * Porcelain/Incremental format wants to show a lot of details per
1471  * commit.  Instead of repeating this every line, emit it only once,
1472  * the first time each commit appears in the output (unless the
1473  * user has specifically asked for us to repeat).
1474  */
1475 static int emit_one_suspect_detail(struct origin *suspect, int repeat)
1476 {
1477         struct commit_info ci;
1478
1479         if (!repeat && (suspect->commit->object.flags & METAINFO_SHOWN))
1480                 return 0;
1481
1482         suspect->commit->object.flags |= METAINFO_SHOWN;
1483         get_commit_info(suspect->commit, &ci, 1);
1484         printf("author %s\n", ci.author.buf);
1485         printf("author-mail %s\n", ci.author_mail.buf);
1486         printf("author-time %lu\n", ci.author_time);
1487         printf("author-tz %s\n", ci.author_tz.buf);
1488         printf("committer %s\n", ci.committer.buf);
1489         printf("committer-mail %s\n", ci.committer_mail.buf);
1490         printf("committer-time %lu\n", ci.committer_time);
1491         printf("committer-tz %s\n", ci.committer_tz.buf);
1492         printf("summary %s\n", ci.summary.buf);
1493         if (suspect->commit->object.flags & UNINTERESTING)
1494                 printf("boundary\n");
1495         if (suspect->previous) {
1496                 struct origin *prev = suspect->previous;
1497                 printf("previous %s ", sha1_to_hex(prev->commit->object.sha1));
1498                 write_name_quoted(prev->path, stdout, '\n');
1499         }
1500
1501         commit_info_destroy(&ci);
1502
1503         return 1;
1504 }
1505
1506 /*
1507  * The blame_entry is found to be guilty for the range.  Mark it
1508  * as such, and show it in incremental output.
1509  */
1510 static void found_guilty_entry(struct blame_entry *ent)
1511 {
1512         if (ent->guilty)
1513                 return;
1514         ent->guilty = 1;
1515         if (incremental) {
1516                 struct origin *suspect = ent->suspect;
1517
1518                 printf("%s %d %d %d\n",
1519                        sha1_to_hex(suspect->commit->object.sha1),
1520                        ent->s_lno + 1, ent->lno + 1, ent->num_lines);
1521                 emit_one_suspect_detail(suspect, 0);
1522                 write_filename_info(suspect->path);
1523                 maybe_flush_or_die(stdout, "stdout");
1524         }
1525 }
1526
1527 /*
1528  * The main loop -- while the scoreboard has lines whose true origin
1529  * is still unknown, pick one blame_entry, and allow its current
1530  * suspect to pass blames to its parents.
1531  */
1532 static void assign_blame(struct scoreboard *sb, int opt)
1533 {
1534         struct rev_info *revs = sb->revs;
1535
1536         while (1) {
1537                 struct blame_entry *ent;
1538                 struct commit *commit;
1539                 struct origin *suspect = NULL;
1540
1541                 /* find one suspect to break down */
1542                 for (ent = sb->ent; !suspect && ent; ent = ent->next)
1543                         if (!ent->guilty)
1544                                 suspect = ent->suspect;
1545                 if (!suspect)
1546                         return; /* all done */
1547
1548                 /*
1549                  * We will use this suspect later in the loop,
1550                  * so hold onto it in the meantime.
1551                  */
1552                 origin_incref(suspect);
1553                 commit = suspect->commit;
1554                 parse_commit(commit);
1555                 if (reverse ||
1556                     (!(commit->object.flags & UNINTERESTING) &&
1557                      !(revs->max_age != -1 && commit->date < revs->max_age)))
1558                         pass_blame(sb, suspect, opt);
1559                 else {
1560                         commit->object.flags |= UNINTERESTING;
1561                         if (commit->object.parsed)
1562                                 mark_parents_uninteresting(commit);
1563                 }
1564                 /* treat root commit as boundary */
1565                 if (!commit->parents && !show_root)
1566                         commit->object.flags |= UNINTERESTING;
1567
1568                 /* Take responsibility for the remaining entries */
1569                 for (ent = sb->ent; ent; ent = ent->next)
1570                         if (same_suspect(ent->suspect, suspect))
1571                                 found_guilty_entry(ent);
1572                 origin_decref(suspect);
1573
1574                 if (DEBUG) /* sanity */
1575                         sanity_check_refcnt(sb);
1576         }
1577 }
1578
1579 static const char *format_time(unsigned long time, const char *tz_str,
1580                                int show_raw_time)
1581 {
1582         static char time_buf[128];
1583         const char *time_str;
1584         int time_len;
1585         int tz;
1586
1587         if (show_raw_time) {
1588                 snprintf(time_buf, sizeof(time_buf), "%lu %s", time, tz_str);
1589         }
1590         else {
1591                 tz = atoi(tz_str);
1592                 time_str = show_date(time, tz, blame_date_mode);
1593                 time_len = strlen(time_str);
1594                 memcpy(time_buf, time_str, time_len);
1595                 memset(time_buf + time_len, ' ', blame_date_width - time_len);
1596         }
1597         return time_buf;
1598 }
1599
1600 #define OUTPUT_ANNOTATE_COMPAT  001
1601 #define OUTPUT_LONG_OBJECT_NAME 002
1602 #define OUTPUT_RAW_TIMESTAMP    004
1603 #define OUTPUT_PORCELAIN        010
1604 #define OUTPUT_SHOW_NAME        020
1605 #define OUTPUT_SHOW_NUMBER      040
1606 #define OUTPUT_SHOW_SCORE      0100
1607 #define OUTPUT_NO_AUTHOR       0200
1608 #define OUTPUT_SHOW_EMAIL       0400
1609 #define OUTPUT_LINE_PORCELAIN 01000
1610
1611 static void emit_porcelain_details(struct origin *suspect, int repeat)
1612 {
1613         if (emit_one_suspect_detail(suspect, repeat) ||
1614             (suspect->commit->object.flags & MORE_THAN_ONE_PATH))
1615                 write_filename_info(suspect->path);
1616 }
1617
1618 static void emit_porcelain(struct scoreboard *sb, struct blame_entry *ent,
1619                            int opt)
1620 {
1621         int repeat = opt & OUTPUT_LINE_PORCELAIN;
1622         int cnt;
1623         const char *cp;
1624         struct origin *suspect = ent->suspect;
1625         char hex[41];
1626
1627         strcpy(hex, sha1_to_hex(suspect->commit->object.sha1));
1628         printf("%s%c%d %d %d\n",
1629                hex,
1630                ent->guilty ? ' ' : '*', /* purely for debugging */
1631                ent->s_lno + 1,
1632                ent->lno + 1,
1633                ent->num_lines);
1634         emit_porcelain_details(suspect, repeat);
1635
1636         cp = nth_line(sb, ent->lno);
1637         for (cnt = 0; cnt < ent->num_lines; cnt++) {
1638                 char ch;
1639                 if (cnt) {
1640                         printf("%s %d %d\n", hex,
1641                                ent->s_lno + 1 + cnt,
1642                                ent->lno + 1 + cnt);
1643                         if (repeat)
1644                                 emit_porcelain_details(suspect, 1);
1645                 }
1646                 putchar('\t');
1647                 do {
1648                         ch = *cp++;
1649                         putchar(ch);
1650                 } while (ch != '\n' &&
1651                          cp < sb->final_buf + sb->final_buf_size);
1652         }
1653
1654         if (sb->final_buf_size && cp[-1] != '\n')
1655                 putchar('\n');
1656 }
1657
1658 static void emit_other(struct scoreboard *sb, struct blame_entry *ent, int opt)
1659 {
1660         int cnt;
1661         const char *cp;
1662         struct origin *suspect = ent->suspect;
1663         struct commit_info ci;
1664         char hex[41];
1665         int show_raw_time = !!(opt & OUTPUT_RAW_TIMESTAMP);
1666
1667         get_commit_info(suspect->commit, &ci, 1);
1668         strcpy(hex, sha1_to_hex(suspect->commit->object.sha1));
1669
1670         cp = nth_line(sb, ent->lno);
1671         for (cnt = 0; cnt < ent->num_lines; cnt++) {
1672                 char ch;
1673                 int length = (opt & OUTPUT_LONG_OBJECT_NAME) ? 40 : abbrev;
1674
1675                 if (suspect->commit->object.flags & UNINTERESTING) {
1676                         if (blank_boundary)
1677                                 memset(hex, ' ', length);
1678                         else if (!(opt & OUTPUT_ANNOTATE_COMPAT)) {
1679                                 length--;
1680                                 putchar('^');
1681                         }
1682                 }
1683
1684                 printf("%.*s", length, hex);
1685                 if (opt & OUTPUT_ANNOTATE_COMPAT) {
1686                         const char *name;
1687                         if (opt & OUTPUT_SHOW_EMAIL)
1688                                 name = ci.author_mail.buf;
1689                         else
1690                                 name = ci.author.buf;
1691                         printf("\t(%10s\t%10s\t%d)", name,
1692                                format_time(ci.author_time, ci.author_tz.buf,
1693                                            show_raw_time),
1694                                ent->lno + 1 + cnt);
1695                 } else {
1696                         if (opt & OUTPUT_SHOW_SCORE)
1697                                 printf(" %*d %02d",
1698                                        max_score_digits, ent->score,
1699                                        ent->suspect->refcnt);
1700                         if (opt & OUTPUT_SHOW_NAME)
1701                                 printf(" %-*.*s", longest_file, longest_file,
1702                                        suspect->path);
1703                         if (opt & OUTPUT_SHOW_NUMBER)
1704                                 printf(" %*d", max_orig_digits,
1705                                        ent->s_lno + 1 + cnt);
1706
1707                         if (!(opt & OUTPUT_NO_AUTHOR)) {
1708                                 const char *name;
1709                                 int pad;
1710                                 if (opt & OUTPUT_SHOW_EMAIL)
1711                                         name = ci.author_mail.buf;
1712                                 else
1713                                         name = ci.author.buf;
1714                                 pad = longest_author - utf8_strwidth(name);
1715                                 printf(" (%s%*s %10s",
1716                                        name, pad, "",
1717                                        format_time(ci.author_time,
1718                                                    ci.author_tz.buf,
1719                                                    show_raw_time));
1720                         }
1721                         printf(" %*d) ",
1722                                max_digits, ent->lno + 1 + cnt);
1723                 }
1724                 do {
1725                         ch = *cp++;
1726                         putchar(ch);
1727                 } while (ch != '\n' &&
1728                          cp < sb->final_buf + sb->final_buf_size);
1729         }
1730
1731         if (sb->final_buf_size && cp[-1] != '\n')
1732                 putchar('\n');
1733
1734         commit_info_destroy(&ci);
1735 }
1736
1737 static void output(struct scoreboard *sb, int option)
1738 {
1739         struct blame_entry *ent;
1740
1741         if (option & OUTPUT_PORCELAIN) {
1742                 for (ent = sb->ent; ent; ent = ent->next) {
1743                         struct blame_entry *oth;
1744                         struct origin *suspect = ent->suspect;
1745                         struct commit *commit = suspect->commit;
1746                         if (commit->object.flags & MORE_THAN_ONE_PATH)
1747                                 continue;
1748                         for (oth = ent->next; oth; oth = oth->next) {
1749                                 if ((oth->suspect->commit != commit) ||
1750                                     !strcmp(oth->suspect->path, suspect->path))
1751                                         continue;
1752                                 commit->object.flags |= MORE_THAN_ONE_PATH;
1753                                 break;
1754                         }
1755                 }
1756         }
1757
1758         for (ent = sb->ent; ent; ent = ent->next) {
1759                 if (option & OUTPUT_PORCELAIN)
1760                         emit_porcelain(sb, ent, option);
1761                 else {
1762                         emit_other(sb, ent, option);
1763                 }
1764         }
1765 }
1766
1767 /*
1768  * To allow quick access to the contents of nth line in the
1769  * final image, prepare an index in the scoreboard.
1770  */
1771 static int prepare_lines(struct scoreboard *sb)
1772 {
1773         const char *buf = sb->final_buf;
1774         unsigned long len = sb->final_buf_size;
1775         int num = 0, incomplete = 0, bol = 1;
1776
1777         if (len && buf[len-1] != '\n')
1778                 incomplete++; /* incomplete line at the end */
1779         while (len--) {
1780                 if (bol) {
1781                         sb->lineno = xrealloc(sb->lineno,
1782                                               sizeof(int *) * (num + 1));
1783                         sb->lineno[num] = buf - sb->final_buf;
1784                         bol = 0;
1785                 }
1786                 if (*buf++ == '\n') {
1787                         num++;
1788                         bol = 1;
1789                 }
1790         }
1791         sb->lineno = xrealloc(sb->lineno,
1792                               sizeof(int *) * (num + incomplete + 1));
1793         sb->lineno[num + incomplete] = buf - sb->final_buf;
1794         sb->num_lines = num + incomplete;
1795         return sb->num_lines;
1796 }
1797
1798 /*
1799  * Add phony grafts for use with -S; this is primarily to
1800  * support git's cvsserver that wants to give a linear history
1801  * to its clients.
1802  */
1803 static int read_ancestry(const char *graft_file)
1804 {
1805         FILE *fp = fopen(graft_file, "r");
1806         struct strbuf buf = STRBUF_INIT;
1807         if (!fp)
1808                 return -1;
1809         while (!strbuf_getwholeline(&buf, fp, '\n')) {
1810                 /* The format is just "Commit Parent1 Parent2 ...\n" */
1811                 struct commit_graft *graft = read_graft_line(buf.buf, buf.len);
1812                 if (graft)
1813                         register_commit_graft(graft, 0);
1814         }
1815         fclose(fp);
1816         strbuf_release(&buf);
1817         return 0;
1818 }
1819
1820 static int update_auto_abbrev(int auto_abbrev, struct origin *suspect)
1821 {
1822         const char *uniq = find_unique_abbrev(suspect->commit->object.sha1,
1823                                               auto_abbrev);
1824         int len = strlen(uniq);
1825         if (auto_abbrev < len)
1826                 return len;
1827         return auto_abbrev;
1828 }
1829
1830 /*
1831  * How many columns do we need to show line numbers, authors,
1832  * and filenames?
1833  */
1834 static void find_alignment(struct scoreboard *sb, int *option)
1835 {
1836         int longest_src_lines = 0;
1837         int longest_dst_lines = 0;
1838         unsigned largest_score = 0;
1839         struct blame_entry *e;
1840         int compute_auto_abbrev = (abbrev < 0);
1841         int auto_abbrev = default_abbrev;
1842
1843         for (e = sb->ent; e; e = e->next) {
1844                 struct origin *suspect = e->suspect;
1845                 struct commit_info ci;
1846                 int num;
1847
1848                 if (compute_auto_abbrev)
1849                         auto_abbrev = update_auto_abbrev(auto_abbrev, suspect);
1850                 if (strcmp(suspect->path, sb->path))
1851                         *option |= OUTPUT_SHOW_NAME;
1852                 num = strlen(suspect->path);
1853                 if (longest_file < num)
1854                         longest_file = num;
1855                 if (!(suspect->commit->object.flags & METAINFO_SHOWN)) {
1856                         suspect->commit->object.flags |= METAINFO_SHOWN;
1857                         get_commit_info(suspect->commit, &ci, 1);
1858                         if (*option & OUTPUT_SHOW_EMAIL)
1859                                 num = utf8_strwidth(ci.author_mail.buf);
1860                         else
1861                                 num = utf8_strwidth(ci.author.buf);
1862                         if (longest_author < num)
1863                                 longest_author = num;
1864                 }
1865                 num = e->s_lno + e->num_lines;
1866                 if (longest_src_lines < num)
1867                         longest_src_lines = num;
1868                 num = e->lno + e->num_lines;
1869                 if (longest_dst_lines < num)
1870                         longest_dst_lines = num;
1871                 if (largest_score < ent_score(sb, e))
1872                         largest_score = ent_score(sb, e);
1873
1874                 commit_info_destroy(&ci);
1875         }
1876         max_orig_digits = decimal_width(longest_src_lines);
1877         max_digits = decimal_width(longest_dst_lines);
1878         max_score_digits = decimal_width(largest_score);
1879
1880         if (compute_auto_abbrev)
1881                 /* one more abbrev length is needed for the boundary commit */
1882                 abbrev = auto_abbrev + 1;
1883 }
1884
1885 /*
1886  * For debugging -- origin is refcounted, and this asserts that
1887  * we do not underflow.
1888  */
1889 static void sanity_check_refcnt(struct scoreboard *sb)
1890 {
1891         int baa = 0;
1892         struct blame_entry *ent;
1893
1894         for (ent = sb->ent; ent; ent = ent->next) {
1895                 /* Nobody should have zero or negative refcnt */
1896                 if (ent->suspect->refcnt <= 0) {
1897                         fprintf(stderr, "%s in %s has negative refcnt %d\n",
1898                                 ent->suspect->path,
1899                                 sha1_to_hex(ent->suspect->commit->object.sha1),
1900                                 ent->suspect->refcnt);
1901                         baa = 1;
1902                 }
1903         }
1904         if (baa) {
1905                 int opt = 0160;
1906                 find_alignment(sb, &opt);
1907                 output(sb, opt);
1908                 die("Baa %d!", baa);
1909         }
1910 }
1911
1912 /*
1913  * Used for the command line parsing; check if the path exists
1914  * in the working tree.
1915  */
1916 static int has_string_in_work_tree(const char *path)
1917 {
1918         struct stat st;
1919         return !lstat(path, &st);
1920 }
1921
1922 static unsigned parse_score(const char *arg)
1923 {
1924         char *end;
1925         unsigned long score = strtoul(arg, &end, 10);
1926         if (*end)
1927                 return 0;
1928         return score;
1929 }
1930
1931 static const char *add_prefix(const char *prefix, const char *path)
1932 {
1933         return prefix_path(prefix, prefix ? strlen(prefix) : 0, path);
1934 }
1935
1936 static int git_blame_config(const char *var, const char *value, void *cb)
1937 {
1938         if (!strcmp(var, "blame.showroot")) {
1939                 show_root = git_config_bool(var, value);
1940                 return 0;
1941         }
1942         if (!strcmp(var, "blame.blankboundary")) {
1943                 blank_boundary = git_config_bool(var, value);
1944                 return 0;
1945         }
1946         if (!strcmp(var, "blame.date")) {
1947                 if (!value)
1948                         return config_error_nonbool(var);
1949                 blame_date_mode = parse_date_format(value);
1950                 return 0;
1951         }
1952
1953         if (userdiff_config(var, value) < 0)
1954                 return -1;
1955
1956         return git_default_config(var, value, cb);
1957 }
1958
1959 static void verify_working_tree_path(struct commit *work_tree, const char *path)
1960 {
1961         struct commit_list *parents;
1962
1963         for (parents = work_tree->parents; parents; parents = parents->next) {
1964                 const unsigned char *commit_sha1 = parents->item->object.sha1;
1965                 unsigned char blob_sha1[20];
1966                 unsigned mode;
1967
1968                 if (!get_tree_entry(commit_sha1, path, blob_sha1, &mode) &&
1969                     sha1_object_info(blob_sha1, NULL) == OBJ_BLOB)
1970                         return;
1971         }
1972         die("no such path '%s' in HEAD", path);
1973 }
1974
1975 static struct commit_list **append_parent(struct commit_list **tail, const unsigned char *sha1)
1976 {
1977         struct commit *parent;
1978
1979         parent = lookup_commit_reference(sha1);
1980         if (!parent)
1981                 die("no such commit %s", sha1_to_hex(sha1));
1982         return &commit_list_insert(parent, tail)->next;
1983 }
1984
1985 static void append_merge_parents(struct commit_list **tail)
1986 {
1987         int merge_head;
1988         const char *merge_head_file = git_path("MERGE_HEAD");
1989         struct strbuf line = STRBUF_INIT;
1990
1991         merge_head = open(merge_head_file, O_RDONLY);
1992         if (merge_head < 0) {
1993                 if (errno == ENOENT)
1994                         return;
1995                 die("cannot open '%s' for reading", merge_head_file);
1996         }
1997
1998         while (!strbuf_getwholeline_fd(&line, merge_head, '\n')) {
1999                 unsigned char sha1[20];
2000                 if (line.len < 40 || get_sha1_hex(line.buf, sha1))
2001                         die("unknown line in '%s': %s", merge_head_file, line.buf);
2002                 tail = append_parent(tail, sha1);
2003         }
2004         close(merge_head);
2005         strbuf_release(&line);
2006 }
2007
2008 /*
2009  * Prepare a dummy commit that represents the work tree (or staged) item.
2010  * Note that annotating work tree item never works in the reverse.
2011  */
2012 static struct commit *fake_working_tree_commit(struct diff_options *opt,
2013                                                const char *path,
2014                                                const char *contents_from)
2015 {
2016         struct commit *commit;
2017         struct origin *origin;
2018         struct commit_list **parent_tail, *parent;
2019         unsigned char head_sha1[20];
2020         struct strbuf buf = STRBUF_INIT;
2021         const char *ident;
2022         time_t now;
2023         int size, len;
2024         struct cache_entry *ce;
2025         unsigned mode;
2026         struct strbuf msg = STRBUF_INIT;
2027
2028         time(&now);
2029         commit = xcalloc(1, sizeof(*commit));
2030         commit->object.parsed = 1;
2031         commit->date = now;
2032         commit->object.type = OBJ_COMMIT;
2033         parent_tail = &commit->parents;
2034
2035         if (!resolve_ref_unsafe("HEAD", head_sha1, 1, NULL))
2036                 die("no such ref: HEAD");
2037
2038         parent_tail = append_parent(parent_tail, head_sha1);
2039         append_merge_parents(parent_tail);
2040         verify_working_tree_path(commit, path);
2041
2042         origin = make_origin(commit, path);
2043
2044         ident = fmt_ident("Not Committed Yet", "not.committed.yet", NULL, 0);
2045         strbuf_addstr(&msg, "tree 0000000000000000000000000000000000000000\n");
2046         for (parent = commit->parents; parent; parent = parent->next)
2047                 strbuf_addf(&msg, "parent %s\n",
2048                             sha1_to_hex(parent->item->object.sha1));
2049         strbuf_addf(&msg,
2050                     "author %s\n"
2051                     "committer %s\n\n"
2052                     "Version of %s from %s\n",
2053                     ident, ident, path,
2054                     (!contents_from ? path :
2055                      (!strcmp(contents_from, "-") ? "standard input" : contents_from)));
2056         commit->buffer = strbuf_detach(&msg, NULL);
2057
2058         if (!contents_from || strcmp("-", contents_from)) {
2059                 struct stat st;
2060                 const char *read_from;
2061                 char *buf_ptr;
2062                 unsigned long buf_len;
2063
2064                 if (contents_from) {
2065                         if (stat(contents_from, &st) < 0)
2066                                 die_errno("Cannot stat '%s'", contents_from);
2067                         read_from = contents_from;
2068                 }
2069                 else {
2070                         if (lstat(path, &st) < 0)
2071                                 die_errno("Cannot lstat '%s'", path);
2072                         read_from = path;
2073                 }
2074                 mode = canon_mode(st.st_mode);
2075
2076                 switch (st.st_mode & S_IFMT) {
2077                 case S_IFREG:
2078                         if (DIFF_OPT_TST(opt, ALLOW_TEXTCONV) &&
2079                             textconv_object(read_from, mode, null_sha1, 0, &buf_ptr, &buf_len))
2080                                 strbuf_attach(&buf, buf_ptr, buf_len, buf_len + 1);
2081                         else if (strbuf_read_file(&buf, read_from, st.st_size) != st.st_size)
2082                                 die_errno("cannot open or read '%s'", read_from);
2083                         break;
2084                 case S_IFLNK:
2085                         if (strbuf_readlink(&buf, read_from, st.st_size) < 0)
2086                                 die_errno("cannot readlink '%s'", read_from);
2087                         break;
2088                 default:
2089                         die("unsupported file type %s", read_from);
2090                 }
2091         }
2092         else {
2093                 /* Reading from stdin */
2094                 mode = 0;
2095                 if (strbuf_read(&buf, 0, 0) < 0)
2096                         die_errno("failed to read from stdin");
2097         }
2098         convert_to_git(path, buf.buf, buf.len, &buf, 0);
2099         origin->file.ptr = buf.buf;
2100         origin->file.size = buf.len;
2101         pretend_sha1_file(buf.buf, buf.len, OBJ_BLOB, origin->blob_sha1);
2102         commit->util = origin;
2103
2104         /*
2105          * Read the current index, replace the path entry with
2106          * origin->blob_sha1 without mucking with its mode or type
2107          * bits; we are not going to write this index out -- we just
2108          * want to run "diff-index --cached".
2109          */
2110         discard_cache();
2111         read_cache();
2112
2113         len = strlen(path);
2114         if (!mode) {
2115                 int pos = cache_name_pos(path, len);
2116                 if (0 <= pos)
2117                         mode = active_cache[pos]->ce_mode;
2118                 else
2119                         /* Let's not bother reading from HEAD tree */
2120                         mode = S_IFREG | 0644;
2121         }
2122         size = cache_entry_size(len);
2123         ce = xcalloc(1, size);
2124         hashcpy(ce->sha1, origin->blob_sha1);
2125         memcpy(ce->name, path, len);
2126         ce->ce_flags = create_ce_flags(0);
2127         ce->ce_namelen = len;
2128         ce->ce_mode = create_ce_mode(mode);
2129         add_cache_entry(ce, ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE);
2130
2131         /*
2132          * We are not going to write this out, so this does not matter
2133          * right now, but someday we might optimize diff-index --cached
2134          * with cache-tree information.
2135          */
2136         cache_tree_invalidate_path(active_cache_tree, path);
2137
2138         return commit;
2139 }
2140
2141 static const char *prepare_final(struct scoreboard *sb)
2142 {
2143         int i;
2144         const char *final_commit_name = NULL;
2145         struct rev_info *revs = sb->revs;
2146
2147         /*
2148          * There must be one and only one positive commit in the
2149          * revs->pending array.
2150          */
2151         for (i = 0; i < revs->pending.nr; i++) {
2152                 struct object *obj = revs->pending.objects[i].item;
2153                 if (obj->flags & UNINTERESTING)
2154                         continue;
2155                 while (obj->type == OBJ_TAG)
2156                         obj = deref_tag(obj, NULL, 0);
2157                 if (obj->type != OBJ_COMMIT)
2158                         die("Non commit %s?", revs->pending.objects[i].name);
2159                 if (sb->final)
2160                         die("More than one commit to dig from %s and %s?",
2161                             revs->pending.objects[i].name,
2162                             final_commit_name);
2163                 sb->final = (struct commit *) obj;
2164                 final_commit_name = revs->pending.objects[i].name;
2165         }
2166         return final_commit_name;
2167 }
2168
2169 static const char *prepare_initial(struct scoreboard *sb)
2170 {
2171         int i;
2172         const char *final_commit_name = NULL;
2173         struct rev_info *revs = sb->revs;
2174
2175         /*
2176          * There must be one and only one negative commit, and it must be
2177          * the boundary.
2178          */
2179         for (i = 0; i < revs->pending.nr; i++) {
2180                 struct object *obj = revs->pending.objects[i].item;
2181                 if (!(obj->flags & UNINTERESTING))
2182                         continue;
2183                 while (obj->type == OBJ_TAG)
2184                         obj = deref_tag(obj, NULL, 0);
2185                 if (obj->type != OBJ_COMMIT)
2186                         die("Non commit %s?", revs->pending.objects[i].name);
2187                 if (sb->final)
2188                         die("More than one commit to dig down to %s and %s?",
2189                             revs->pending.objects[i].name,
2190                             final_commit_name);
2191                 sb->final = (struct commit *) obj;
2192                 final_commit_name = revs->pending.objects[i].name;
2193         }
2194         if (!final_commit_name)
2195                 die("No commit to dig down to?");
2196         return final_commit_name;
2197 }
2198
2199 static int blame_copy_callback(const struct option *option, const char *arg, int unset)
2200 {
2201         int *opt = option->value;
2202
2203         /*
2204          * -C enables copy from removed files;
2205          * -C -C enables copy from existing files, but only
2206          *       when blaming a new file;
2207          * -C -C -C enables copy from existing files for
2208          *          everybody
2209          */
2210         if (*opt & PICKAXE_BLAME_COPY_HARDER)
2211                 *opt |= PICKAXE_BLAME_COPY_HARDEST;
2212         if (*opt & PICKAXE_BLAME_COPY)
2213                 *opt |= PICKAXE_BLAME_COPY_HARDER;
2214         *opt |= PICKAXE_BLAME_COPY | PICKAXE_BLAME_MOVE;
2215
2216         if (arg)
2217                 blame_copy_score = parse_score(arg);
2218         return 0;
2219 }
2220
2221 static int blame_move_callback(const struct option *option, const char *arg, int unset)
2222 {
2223         int *opt = option->value;
2224
2225         *opt |= PICKAXE_BLAME_MOVE;
2226
2227         if (arg)
2228                 blame_move_score = parse_score(arg);
2229         return 0;
2230 }
2231
2232 int cmd_blame(int argc, const char **argv, const char *prefix)
2233 {
2234         struct rev_info revs;
2235         const char *path;
2236         struct scoreboard sb;
2237         struct origin *o;
2238         struct blame_entry *ent = NULL;
2239         long dashdash_pos, lno;
2240         const char *final_commit_name = NULL;
2241         enum object_type type;
2242
2243         static struct string_list range_list;
2244         static int output_option = 0, opt = 0;
2245         static int show_stats = 0;
2246         static const char *revs_file = NULL;
2247         static const char *contents_from = NULL;
2248         static const struct option options[] = {
2249                 OPT_BOOL(0, "incremental", &incremental, N_("Show blame entries as we find them, incrementally")),
2250                 OPT_BOOL('b', NULL, &blank_boundary, N_("Show blank SHA-1 for boundary commits (Default: off)")),
2251                 OPT_BOOL(0, "root", &show_root, N_("Do not treat root commits as boundaries (Default: off)")),
2252                 OPT_BOOL(0, "show-stats", &show_stats, N_("Show work cost statistics")),
2253                 OPT_BIT(0, "score-debug", &output_option, N_("Show output score for blame entries"), OUTPUT_SHOW_SCORE),
2254                 OPT_BIT('f', "show-name", &output_option, N_("Show original filename (Default: auto)"), OUTPUT_SHOW_NAME),
2255                 OPT_BIT('n', "show-number", &output_option, N_("Show original linenumber (Default: off)"), OUTPUT_SHOW_NUMBER),
2256                 OPT_BIT('p', "porcelain", &output_option, N_("Show in a format designed for machine consumption"), OUTPUT_PORCELAIN),
2257                 OPT_BIT(0, "line-porcelain", &output_option, N_("Show porcelain format with per-line commit information"), OUTPUT_PORCELAIN|OUTPUT_LINE_PORCELAIN),
2258                 OPT_BIT('c', NULL, &output_option, N_("Use the same output mode as git-annotate (Default: off)"), OUTPUT_ANNOTATE_COMPAT),
2259                 OPT_BIT('t', NULL, &output_option, N_("Show raw timestamp (Default: off)"), OUTPUT_RAW_TIMESTAMP),
2260                 OPT_BIT('l', NULL, &output_option, N_("Show long commit SHA1 (Default: off)"), OUTPUT_LONG_OBJECT_NAME),
2261                 OPT_BIT('s', NULL, &output_option, N_("Suppress author name and timestamp (Default: off)"), OUTPUT_NO_AUTHOR),
2262                 OPT_BIT('e', "show-email", &output_option, N_("Show author email instead of name (Default: off)"), OUTPUT_SHOW_EMAIL),
2263                 OPT_BIT('w', NULL, &xdl_opts, N_("Ignore whitespace differences"), XDF_IGNORE_WHITESPACE),
2264                 OPT_BIT(0, "minimal", &xdl_opts, N_("Spend extra cycles to find better match"), XDF_NEED_MINIMAL),
2265                 OPT_STRING('S', NULL, &revs_file, N_("file"), N_("Use revisions from <file> instead of calling git-rev-list")),
2266                 OPT_STRING(0, "contents", &contents_from, N_("file"), N_("Use <file>'s contents as the final image")),
2267                 { OPTION_CALLBACK, 'C', NULL, &opt, N_("score"), N_("Find line copies within and across files"), PARSE_OPT_OPTARG, blame_copy_callback },
2268                 { OPTION_CALLBACK, 'M', NULL, &opt, N_("score"), N_("Find line movements within and across files"), PARSE_OPT_OPTARG, blame_move_callback },
2269                 OPT_STRING_LIST('L', NULL, &range_list, N_("n,m"), N_("Process only line range n,m, counting from 1")),
2270                 OPT__ABBREV(&abbrev),
2271                 OPT_END()
2272         };
2273
2274         struct parse_opt_ctx_t ctx;
2275         int cmd_is_annotate = !strcmp(argv[0], "annotate");
2276         struct range_set ranges;
2277         unsigned int range_i;
2278         long anchor;
2279
2280         git_config(git_blame_config, NULL);
2281         init_revisions(&revs, NULL);
2282         revs.date_mode = blame_date_mode;
2283         DIFF_OPT_SET(&revs.diffopt, ALLOW_TEXTCONV);
2284         DIFF_OPT_SET(&revs.diffopt, FOLLOW_RENAMES);
2285
2286         save_commit_buffer = 0;
2287         dashdash_pos = 0;
2288
2289         parse_options_start(&ctx, argc, argv, prefix, options,
2290                             PARSE_OPT_KEEP_DASHDASH | PARSE_OPT_KEEP_ARGV0);
2291         for (;;) {
2292                 switch (parse_options_step(&ctx, options, blame_opt_usage)) {
2293                 case PARSE_OPT_HELP:
2294                         exit(129);
2295                 case PARSE_OPT_DONE:
2296                         if (ctx.argv[0])
2297                                 dashdash_pos = ctx.cpidx;
2298                         goto parse_done;
2299                 }
2300
2301                 if (!strcmp(ctx.argv[0], "--reverse")) {
2302                         ctx.argv[0] = "--children";
2303                         reverse = 1;
2304                 }
2305                 parse_revision_opt(&revs, &ctx, options, blame_opt_usage);
2306         }
2307 parse_done:
2308         no_whole_file_rename = !DIFF_OPT_TST(&revs.diffopt, FOLLOW_RENAMES);
2309         DIFF_OPT_CLR(&revs.diffopt, FOLLOW_RENAMES);
2310         argc = parse_options_end(&ctx);
2311
2312         if (0 < abbrev)
2313                 /* one more abbrev length is needed for the boundary commit */
2314                 abbrev++;
2315
2316         if (revs_file && read_ancestry(revs_file))
2317                 die_errno("reading graft file '%s' failed", revs_file);
2318
2319         if (cmd_is_annotate) {
2320                 output_option |= OUTPUT_ANNOTATE_COMPAT;
2321                 blame_date_mode = DATE_ISO8601;
2322         } else {
2323                 blame_date_mode = revs.date_mode;
2324         }
2325
2326         /* The maximum width used to show the dates */
2327         switch (blame_date_mode) {
2328         case DATE_RFC2822:
2329                 blame_date_width = sizeof("Thu, 19 Oct 2006 16:00:04 -0700");
2330                 break;
2331         case DATE_ISO8601:
2332                 blame_date_width = sizeof("2006-10-19 16:00:04 -0700");
2333                 break;
2334         case DATE_RAW:
2335                 blame_date_width = sizeof("1161298804 -0700");
2336                 break;
2337         case DATE_SHORT:
2338                 blame_date_width = sizeof("2006-10-19");
2339                 break;
2340         case DATE_RELATIVE:
2341                 /* "normal" is used as the fallback for "relative" */
2342         case DATE_LOCAL:
2343         case DATE_NORMAL:
2344                 blame_date_width = sizeof("Thu Oct 19 16:00:04 2006 -0700");
2345                 break;
2346         }
2347         blame_date_width -= 1; /* strip the null */
2348
2349         if (DIFF_OPT_TST(&revs.diffopt, FIND_COPIES_HARDER))
2350                 opt |= (PICKAXE_BLAME_COPY | PICKAXE_BLAME_MOVE |
2351                         PICKAXE_BLAME_COPY_HARDER);
2352
2353         if (!blame_move_score)
2354                 blame_move_score = BLAME_DEFAULT_MOVE_SCORE;
2355         if (!blame_copy_score)
2356                 blame_copy_score = BLAME_DEFAULT_COPY_SCORE;
2357
2358         /*
2359          * We have collected options unknown to us in argv[1..unk]
2360          * which are to be passed to revision machinery if we are
2361          * going to do the "bottom" processing.
2362          *
2363          * The remaining are:
2364          *
2365          * (1) if dashdash_pos != 0, it is either
2366          *     "blame [revisions] -- <path>" or
2367          *     "blame -- <path> <rev>"
2368          *
2369          * (2) otherwise, it is one of the two:
2370          *     "blame [revisions] <path>"
2371          *     "blame <path> <rev>"
2372          *
2373          * Note that we must strip out <path> from the arguments: we do not
2374          * want the path pruning but we may want "bottom" processing.
2375          */
2376         if (dashdash_pos) {
2377                 switch (argc - dashdash_pos - 1) {
2378                 case 2: /* (1b) */
2379                         if (argc != 4)
2380                                 usage_with_options(blame_opt_usage, options);
2381                         /* reorder for the new way: <rev> -- <path> */
2382                         argv[1] = argv[3];
2383                         argv[3] = argv[2];
2384                         argv[2] = "--";
2385                         /* FALLTHROUGH */
2386                 case 1: /* (1a) */
2387                         path = add_prefix(prefix, argv[--argc]);
2388                         argv[argc] = NULL;
2389                         break;
2390                 default:
2391                         usage_with_options(blame_opt_usage, options);
2392                 }
2393         } else {
2394                 if (argc < 2)
2395                         usage_with_options(blame_opt_usage, options);
2396                 path = add_prefix(prefix, argv[argc - 1]);
2397                 if (argc == 3 && !has_string_in_work_tree(path)) { /* (2b) */
2398                         path = add_prefix(prefix, argv[1]);
2399                         argv[1] = argv[2];
2400                 }
2401                 argv[argc - 1] = "--";
2402
2403                 setup_work_tree();
2404                 if (!has_string_in_work_tree(path))
2405                         die_errno("cannot stat path '%s'", path);
2406         }
2407
2408         revs.disable_stdin = 1;
2409         setup_revisions(argc, argv, &revs, NULL);
2410         memset(&sb, 0, sizeof(sb));
2411
2412         sb.revs = &revs;
2413         if (!reverse)
2414                 final_commit_name = prepare_final(&sb);
2415         else if (contents_from)
2416                 die("--contents and --children do not blend well.");
2417         else
2418                 final_commit_name = prepare_initial(&sb);
2419
2420         if (!sb.final) {
2421                 /*
2422                  * "--not A B -- path" without anything positive;
2423                  * do not default to HEAD, but use the working tree
2424                  * or "--contents".
2425                  */
2426                 setup_work_tree();
2427                 sb.final = fake_working_tree_commit(&sb.revs->diffopt,
2428                                                     path, contents_from);
2429                 add_pending_object(&revs, &(sb.final->object), ":");
2430         }
2431         else if (contents_from)
2432                 die("Cannot use --contents with final commit object name");
2433
2434         /*
2435          * If we have bottom, this will mark the ancestors of the
2436          * bottom commits we would reach while traversing as
2437          * uninteresting.
2438          */
2439         if (prepare_revision_walk(&revs))
2440                 die("revision walk setup failed");
2441
2442         if (is_null_sha1(sb.final->object.sha1)) {
2443                 char *buf;
2444                 o = sb.final->util;
2445                 buf = xmalloc(o->file.size + 1);
2446                 memcpy(buf, o->file.ptr, o->file.size + 1);
2447                 sb.final_buf = buf;
2448                 sb.final_buf_size = o->file.size;
2449         }
2450         else {
2451                 o = get_origin(&sb, sb.final, path);
2452                 if (fill_blob_sha1_and_mode(o))
2453                         die("no such path %s in %s", path, final_commit_name);
2454
2455                 if (DIFF_OPT_TST(&sb.revs->diffopt, ALLOW_TEXTCONV) &&
2456                     textconv_object(path, o->mode, o->blob_sha1, 1, (char **) &sb.final_buf,
2457                                     &sb.final_buf_size))
2458                         ;
2459                 else
2460                         sb.final_buf = read_sha1_file(o->blob_sha1, &type,
2461                                                       &sb.final_buf_size);
2462
2463                 if (!sb.final_buf)
2464                         die("Cannot read blob %s for path %s",
2465                             sha1_to_hex(o->blob_sha1),
2466                             path);
2467         }
2468         num_read_blob++;
2469         lno = prepare_lines(&sb);
2470
2471         if (lno && !range_list.nr)
2472                 string_list_append(&range_list, xstrdup("1"));
2473
2474         anchor = 1;
2475         range_set_init(&ranges, range_list.nr);
2476         for (range_i = 0; range_i < range_list.nr; ++range_i) {
2477                 long bottom, top;
2478                 if (parse_range_arg(range_list.items[range_i].string,
2479                                     nth_line_cb, &sb, lno, anchor,
2480                                     &bottom, &top, sb.path))
2481                         usage(blame_usage);
2482                 if (lno < top || ((lno || bottom) && lno < bottom))
2483                         die("file %s has only %lu lines", path, lno);
2484                 if (bottom < 1)
2485                         bottom = 1;
2486                 if (top < 1)
2487                         top = lno;
2488                 bottom--;
2489                 range_set_append_unsafe(&ranges, bottom, top);
2490                 anchor = top + 1;
2491         }
2492         sort_and_merge_range_set(&ranges);
2493
2494         for (range_i = ranges.nr; range_i > 0; --range_i) {
2495                 const struct range *r = &ranges.ranges[range_i - 1];
2496                 long bottom = r->start;
2497                 long top = r->end;
2498                 struct blame_entry *next = ent;
2499                 ent = xcalloc(1, sizeof(*ent));
2500                 ent->lno = bottom;
2501                 ent->num_lines = top - bottom;
2502                 ent->suspect = o;
2503                 ent->s_lno = bottom;
2504                 ent->next = next;
2505                 if (next)
2506                         next->prev = ent;
2507                 origin_incref(o);
2508         }
2509         origin_decref(o);
2510
2511         range_set_release(&ranges);
2512         string_list_clear(&range_list, 0);
2513
2514         sb.ent = ent;
2515         sb.path = path;
2516
2517         read_mailmap(&mailmap, NULL);
2518
2519         if (!incremental)
2520                 setup_pager();
2521
2522         assign_blame(&sb, opt);
2523
2524         if (incremental)
2525                 return 0;
2526
2527         coalesce(&sb);
2528
2529         if (!(output_option & OUTPUT_PORCELAIN))
2530                 find_alignment(&sb, &output_option);
2531
2532         output(&sb, output_option);
2533         free((void *)sb.final_buf);
2534         for (ent = sb.ent; ent; ) {
2535                 struct blame_entry *e = ent->next;
2536                 free(ent);
2537                 ent = e;
2538         }
2539
2540         if (show_stats) {
2541                 printf("num read blob: %d\n", num_read_blob);
2542                 printf("num get patch: %d\n", num_get_patch);
2543                 printf("num commits: %d\n", num_commits);
2544         }
2545         return 0;
2546 }