Teach git-fetch to grab a tag at the same time as a commit
[git] / builtin-blame.c
1 /*
2  * Pickaxe
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 "path-list.h"
20 #include "mailmap.h"
21
22 static char blame_usage[] =
23 "git-blame [-c] [-b] [-l] [--root] [-t] [-f] [-n] [-s] [-p] [-w] [-L n,m] [-S <revs-file>] [-M] [-C] [-C] [--contents <filename>] [--incremental] [commit] [--] file\n"
24 "  -c                  Use the same output mode as git-annotate (Default: off)\n"
25 "  -b                  Show blank SHA-1 for boundary commits (Default: off)\n"
26 "  -l                  Show long commit SHA1 (Default: off)\n"
27 "  --root              Do not treat root commits as boundaries (Default: off)\n"
28 "  -t                  Show raw timestamp (Default: off)\n"
29 "  -f, --show-name     Show original filename (Default: auto)\n"
30 "  -n, --show-number   Show original linenumber (Default: off)\n"
31 "  -s                  Suppress author name and timestamp (Default: off)\n"
32 "  -p, --porcelain     Show in a format designed for machine consumption\n"
33 "  -w                  Ignore whitespace differences\n"
34 "  -L n,m              Process only line range n,m, counting from 1\n"
35 "  -M, -C              Find line movements within and across files\n"
36 "  --incremental       Show blame entries as we find them, incrementally\n"
37 "  --contents file     Use <file>'s contents as the final image\n"
38 "  -S revs-file        Use revisions from revs-file instead of calling git-rev-list\n";
39
40 static int longest_file;
41 static int longest_author;
42 static int max_orig_digits;
43 static int max_digits;
44 static int max_score_digits;
45 static int show_root;
46 static int blank_boundary;
47 static int incremental;
48 static int cmd_is_annotate;
49 static int xdl_opts = XDF_NEED_MINIMAL;
50 static struct path_list mailmap;
51
52 #ifndef DEBUG
53 #define DEBUG 0
54 #endif
55
56 /* stats */
57 static int num_read_blob;
58 static int num_get_patch;
59 static int num_commits;
60
61 #define PICKAXE_BLAME_MOVE              01
62 #define PICKAXE_BLAME_COPY              02
63 #define PICKAXE_BLAME_COPY_HARDER       04
64 #define PICKAXE_BLAME_COPY_HARDEST      010
65
66 /*
67  * blame for a blame_entry with score lower than these thresholds
68  * is not passed to the parent using move/copy logic.
69  */
70 static unsigned blame_move_score;
71 static unsigned blame_copy_score;
72 #define BLAME_DEFAULT_MOVE_SCORE        20
73 #define BLAME_DEFAULT_COPY_SCORE        40
74
75 /* bits #0..7 in revision.h, #8..11 used for merge_bases() in commit.c */
76 #define METAINFO_SHOWN          (1u<<12)
77 #define MORE_THAN_ONE_PATH      (1u<<13)
78
79 /*
80  * One blob in a commit that is being suspected
81  */
82 struct origin {
83         int refcnt;
84         struct commit *commit;
85         mmfile_t file;
86         unsigned char blob_sha1[20];
87         char path[FLEX_ARRAY];
88 };
89
90 /*
91  * Given an origin, prepare mmfile_t structure to be used by the
92  * diff machinery
93  */
94 static char *fill_origin_blob(struct origin *o, mmfile_t *file)
95 {
96         if (!o->file.ptr) {
97                 enum object_type type;
98                 num_read_blob++;
99                 file->ptr = read_sha1_file(o->blob_sha1, &type,
100                                            (unsigned long *)(&(file->size)));
101                 if (!file->ptr)
102                         die("Cannot read blob %s for path %s",
103                             sha1_to_hex(o->blob_sha1),
104                             o->path);
105                 o->file = *file;
106         }
107         else
108                 *file = o->file;
109         return file->ptr;
110 }
111
112 /*
113  * Origin is refcounted and usually we keep the blob contents to be
114  * reused.
115  */
116 static inline struct origin *origin_incref(struct origin *o)
117 {
118         if (o)
119                 o->refcnt++;
120         return o;
121 }
122
123 static void origin_decref(struct origin *o)
124 {
125         if (o && --o->refcnt <= 0) {
126                 free(o->file.ptr);
127                 free(o);
128         }
129 }
130
131 static void drop_origin_blob(struct origin *o)
132 {
133         if (o->file.ptr) {
134                 free(o->file.ptr);
135                 o->file.ptr = NULL;
136         }
137 }
138
139 /*
140  * Each group of lines is described by a blame_entry; it can be split
141  * as we pass blame to the parents.  They form a linked list in the
142  * scoreboard structure, sorted by the target line number.
143  */
144 struct blame_entry {
145         struct blame_entry *prev;
146         struct blame_entry *next;
147
148         /* the first line of this group in the final image;
149          * internally all line numbers are 0 based.
150          */
151         int lno;
152
153         /* how many lines this group has */
154         int num_lines;
155
156         /* the commit that introduced this group into the final image */
157         struct origin *suspect;
158
159         /* true if the suspect is truly guilty; false while we have not
160          * checked if the group came from one of its parents.
161          */
162         char guilty;
163
164         /* the line number of the first line of this group in the
165          * suspect's file; internally all line numbers are 0 based.
166          */
167         int s_lno;
168
169         /* how significant this entry is -- cached to avoid
170          * scanning the lines over and over.
171          */
172         unsigned score;
173 };
174
175 /*
176  * The current state of the blame assignment.
177  */
178 struct scoreboard {
179         /* the final commit (i.e. where we started digging from) */
180         struct commit *final;
181
182         const char *path;
183
184         /*
185          * The contents in the final image.
186          * Used by many functions to obtain contents of the nth line,
187          * indexed with scoreboard.lineno[blame_entry.lno].
188          */
189         const char *final_buf;
190         unsigned long final_buf_size;
191
192         /* linked list of blames */
193         struct blame_entry *ent;
194
195         /* look-up a line in the final buffer */
196         int num_lines;
197         int *lineno;
198 };
199
200 static inline int same_suspect(struct origin *a, struct origin *b)
201 {
202         if (a == b)
203                 return 1;
204         if (a->commit != b->commit)
205                 return 0;
206         return !strcmp(a->path, b->path);
207 }
208
209 static void sanity_check_refcnt(struct scoreboard *);
210
211 /*
212  * If two blame entries that are next to each other came from
213  * contiguous lines in the same origin (i.e. <commit, path> pair),
214  * merge them together.
215  */
216 static void coalesce(struct scoreboard *sb)
217 {
218         struct blame_entry *ent, *next;
219
220         for (ent = sb->ent; ent && (next = ent->next); ent = next) {
221                 if (same_suspect(ent->suspect, next->suspect) &&
222                     ent->guilty == next->guilty &&
223                     ent->s_lno + ent->num_lines == next->s_lno) {
224                         ent->num_lines += next->num_lines;
225                         ent->next = next->next;
226                         if (ent->next)
227                                 ent->next->prev = ent;
228                         origin_decref(next->suspect);
229                         free(next);
230                         ent->score = 0;
231                         next = ent; /* again */
232                 }
233         }
234
235         if (DEBUG) /* sanity */
236                 sanity_check_refcnt(sb);
237 }
238
239 /*
240  * Given a commit and a path in it, create a new origin structure.
241  * The callers that add blame to the scoreboard should use
242  * get_origin() to obtain shared, refcounted copy instead of calling
243  * this function directly.
244  */
245 static struct origin *make_origin(struct commit *commit, const char *path)
246 {
247         struct origin *o;
248         o = xcalloc(1, sizeof(*o) + strlen(path) + 1);
249         o->commit = commit;
250         o->refcnt = 1;
251         strcpy(o->path, path);
252         return o;
253 }
254
255 /*
256  * Locate an existing origin or create a new one.
257  */
258 static struct origin *get_origin(struct scoreboard *sb,
259                                  struct commit *commit,
260                                  const char *path)
261 {
262         struct blame_entry *e;
263
264         for (e = sb->ent; e; e = e->next) {
265                 if (e->suspect->commit == commit &&
266                     !strcmp(e->suspect->path, path))
267                         return origin_incref(e->suspect);
268         }
269         return make_origin(commit, path);
270 }
271
272 /*
273  * Fill the blob_sha1 field of an origin if it hasn't, so that later
274  * call to fill_origin_blob() can use it to locate the data.  blob_sha1
275  * for an origin is also used to pass the blame for the entire file to
276  * the parent to detect the case where a child's blob is identical to
277  * that of its parent's.
278  */
279 static int fill_blob_sha1(struct origin *origin)
280 {
281         unsigned mode;
282
283         if (!is_null_sha1(origin->blob_sha1))
284                 return 0;
285         if (get_tree_entry(origin->commit->object.sha1,
286                            origin->path,
287                            origin->blob_sha1, &mode))
288                 goto error_out;
289         if (sha1_object_info(origin->blob_sha1, NULL) != OBJ_BLOB)
290                 goto error_out;
291         return 0;
292  error_out:
293         hashclr(origin->blob_sha1);
294         return -1;
295 }
296
297 /*
298  * We have an origin -- check if the same path exists in the
299  * parent and return an origin structure to represent it.
300  */
301 static struct origin *find_origin(struct scoreboard *sb,
302                                   struct commit *parent,
303                                   struct origin *origin)
304 {
305         struct origin *porigin = NULL;
306         struct diff_options diff_opts;
307         const char *paths[2];
308
309         if (parent->util) {
310                 /*
311                  * Each commit object can cache one origin in that
312                  * commit.  This is a freestanding copy of origin and
313                  * not refcounted.
314                  */
315                 struct origin *cached = parent->util;
316                 if (!strcmp(cached->path, origin->path)) {
317                         /*
318                          * The same path between origin and its parent
319                          * without renaming -- the most common case.
320                          */
321                         porigin = get_origin(sb, parent, cached->path);
322
323                         /*
324                          * If the origin was newly created (i.e. get_origin
325                          * would call make_origin if none is found in the
326                          * scoreboard), it does not know the blob_sha1,
327                          * so copy it.  Otherwise porigin was in the
328                          * scoreboard and already knows blob_sha1.
329                          */
330                         if (porigin->refcnt == 1)
331                                 hashcpy(porigin->blob_sha1, cached->blob_sha1);
332                         return porigin;
333                 }
334                 /* otherwise it was not very useful; free it */
335                 free(parent->util);
336                 parent->util = NULL;
337         }
338
339         /* See if the origin->path is different between parent
340          * and origin first.  Most of the time they are the
341          * same and diff-tree is fairly efficient about this.
342          */
343         diff_setup(&diff_opts);
344         DIFF_OPT_SET(&diff_opts, RECURSIVE);
345         diff_opts.detect_rename = 0;
346         diff_opts.output_format = DIFF_FORMAT_NO_OUTPUT;
347         paths[0] = origin->path;
348         paths[1] = NULL;
349
350         diff_tree_setup_paths(paths, &diff_opts);
351         if (diff_setup_done(&diff_opts) < 0)
352                 die("diff-setup");
353
354         if (is_null_sha1(origin->commit->object.sha1))
355                 do_diff_cache(parent->tree->object.sha1, &diff_opts);
356         else
357                 diff_tree_sha1(parent->tree->object.sha1,
358                                origin->commit->tree->object.sha1,
359                                "", &diff_opts);
360         diffcore_std(&diff_opts);
361
362         /* It is either one entry that says "modified", or "created",
363          * or nothing.
364          */
365         if (!diff_queued_diff.nr) {
366                 /* The path is the same as parent */
367                 porigin = get_origin(sb, parent, origin->path);
368                 hashcpy(porigin->blob_sha1, origin->blob_sha1);
369         }
370         else if (diff_queued_diff.nr != 1)
371                 die("internal error in blame::find_origin");
372         else {
373                 struct diff_filepair *p = diff_queued_diff.queue[0];
374                 switch (p->status) {
375                 default:
376                         die("internal error in blame::find_origin (%c)",
377                             p->status);
378                 case 'M':
379                         porigin = get_origin(sb, parent, origin->path);
380                         hashcpy(porigin->blob_sha1, p->one->sha1);
381                         break;
382                 case 'A':
383                 case 'T':
384                         /* Did not exist in parent, or type changed */
385                         break;
386                 }
387         }
388         diff_flush(&diff_opts);
389         diff_tree_release_paths(&diff_opts);
390         if (porigin) {
391                 /*
392                  * Create a freestanding copy that is not part of
393                  * the refcounted origin found in the scoreboard, and
394                  * cache it in the commit.
395                  */
396                 struct origin *cached;
397
398                 cached = make_origin(porigin->commit, porigin->path);
399                 hashcpy(cached->blob_sha1, porigin->blob_sha1);
400                 parent->util = cached;
401         }
402         return porigin;
403 }
404
405 /*
406  * We have an origin -- find the path that corresponds to it in its
407  * parent and return an origin structure to represent it.
408  */
409 static struct origin *find_rename(struct scoreboard *sb,
410                                   struct commit *parent,
411                                   struct origin *origin)
412 {
413         struct origin *porigin = NULL;
414         struct diff_options diff_opts;
415         int i;
416         const char *paths[2];
417
418         diff_setup(&diff_opts);
419         DIFF_OPT_SET(&diff_opts, RECURSIVE);
420         diff_opts.detect_rename = DIFF_DETECT_RENAME;
421         diff_opts.output_format = DIFF_FORMAT_NO_OUTPUT;
422         diff_opts.single_follow = origin->path;
423         paths[0] = NULL;
424         diff_tree_setup_paths(paths, &diff_opts);
425         if (diff_setup_done(&diff_opts) < 0)
426                 die("diff-setup");
427
428         if (is_null_sha1(origin->commit->object.sha1))
429                 do_diff_cache(parent->tree->object.sha1, &diff_opts);
430         else
431                 diff_tree_sha1(parent->tree->object.sha1,
432                                origin->commit->tree->object.sha1,
433                                "", &diff_opts);
434         diffcore_std(&diff_opts);
435
436         for (i = 0; i < diff_queued_diff.nr; i++) {
437                 struct diff_filepair *p = diff_queued_diff.queue[i];
438                 if ((p->status == 'R' || p->status == 'C') &&
439                     !strcmp(p->two->path, origin->path)) {
440                         porigin = get_origin(sb, parent, p->one->path);
441                         hashcpy(porigin->blob_sha1, p->one->sha1);
442                         break;
443                 }
444         }
445         diff_flush(&diff_opts);
446         diff_tree_release_paths(&diff_opts);
447         return porigin;
448 }
449
450 /*
451  * Parsing of patch chunks...
452  */
453 struct chunk {
454         /* line number in postimage; up to but not including this
455          * line is the same as preimage
456          */
457         int same;
458
459         /* preimage line number after this chunk */
460         int p_next;
461
462         /* postimage line number after this chunk */
463         int t_next;
464 };
465
466 struct patch {
467         struct chunk *chunks;
468         int num;
469 };
470
471 struct blame_diff_state {
472         struct xdiff_emit_state xm;
473         struct patch *ret;
474         unsigned hunk_post_context;
475         unsigned hunk_in_pre_context : 1;
476 };
477
478 static void process_u_diff(void *state_, char *line, unsigned long len)
479 {
480         struct blame_diff_state *state = state_;
481         struct chunk *chunk;
482         int off1, off2, len1, len2, num;
483
484         num = state->ret->num;
485         if (len < 4 || line[0] != '@' || line[1] != '@') {
486                 if (state->hunk_in_pre_context && line[0] == ' ')
487                         state->ret->chunks[num - 1].same++;
488                 else {
489                         state->hunk_in_pre_context = 0;
490                         if (line[0] == ' ')
491                                 state->hunk_post_context++;
492                         else
493                                 state->hunk_post_context = 0;
494                 }
495                 return;
496         }
497
498         if (num && state->hunk_post_context) {
499                 chunk = &state->ret->chunks[num - 1];
500                 chunk->p_next -= state->hunk_post_context;
501                 chunk->t_next -= state->hunk_post_context;
502         }
503         state->ret->num = ++num;
504         state->ret->chunks = xrealloc(state->ret->chunks,
505                                       sizeof(struct chunk) * num);
506         chunk = &state->ret->chunks[num - 1];
507         if (parse_hunk_header(line, len, &off1, &len1, &off2, &len2)) {
508                 state->ret->num--;
509                 return;
510         }
511
512         /* Line numbers in patch output are one based. */
513         off1--;
514         off2--;
515
516         chunk->same = len2 ? off2 : (off2 + 1);
517
518         chunk->p_next = off1 + (len1 ? len1 : 1);
519         chunk->t_next = chunk->same + len2;
520         state->hunk_in_pre_context = 1;
521         state->hunk_post_context = 0;
522 }
523
524 static struct patch *compare_buffer(mmfile_t *file_p, mmfile_t *file_o,
525                                     int context)
526 {
527         struct blame_diff_state state;
528         xpparam_t xpp;
529         xdemitconf_t xecfg;
530         xdemitcb_t ecb;
531
532         xpp.flags = xdl_opts;
533         memset(&xecfg, 0, sizeof(xecfg));
534         xecfg.ctxlen = context;
535         ecb.outf = xdiff_outf;
536         ecb.priv = &state;
537         memset(&state, 0, sizeof(state));
538         state.xm.consume = process_u_diff;
539         state.ret = xmalloc(sizeof(struct patch));
540         state.ret->chunks = NULL;
541         state.ret->num = 0;
542
543         xdi_diff(file_p, file_o, &xpp, &xecfg, &ecb);
544
545         if (state.ret->num) {
546                 struct chunk *chunk;
547                 chunk = &state.ret->chunks[state.ret->num - 1];
548                 chunk->p_next -= state.hunk_post_context;
549                 chunk->t_next -= state.hunk_post_context;
550         }
551         return state.ret;
552 }
553
554 /*
555  * Run diff between two origins and grab the patch output, so that
556  * we can pass blame for lines origin is currently suspected for
557  * to its parent.
558  */
559 static struct patch *get_patch(struct origin *parent, struct origin *origin)
560 {
561         mmfile_t file_p, file_o;
562         struct patch *patch;
563
564         fill_origin_blob(parent, &file_p);
565         fill_origin_blob(origin, &file_o);
566         if (!file_p.ptr || !file_o.ptr)
567                 return NULL;
568         patch = compare_buffer(&file_p, &file_o, 0);
569         num_get_patch++;
570         return patch;
571 }
572
573 static void free_patch(struct patch *p)
574 {
575         free(p->chunks);
576         free(p);
577 }
578
579 /*
580  * Link in a new blame entry to the scoreboard.  Entries that cover the
581  * same line range have been removed from the scoreboard previously.
582  */
583 static void add_blame_entry(struct scoreboard *sb, struct blame_entry *e)
584 {
585         struct blame_entry *ent, *prev = NULL;
586
587         origin_incref(e->suspect);
588
589         for (ent = sb->ent; ent && ent->lno < e->lno; ent = ent->next)
590                 prev = ent;
591
592         /* prev, if not NULL, is the last one that is below e */
593         e->prev = prev;
594         if (prev) {
595                 e->next = prev->next;
596                 prev->next = e;
597         }
598         else {
599                 e->next = sb->ent;
600                 sb->ent = e;
601         }
602         if (e->next)
603                 e->next->prev = e;
604 }
605
606 /*
607  * src typically is on-stack; we want to copy the information in it to
608  * a malloced blame_entry that is already on the linked list of the
609  * scoreboard.  The origin of dst loses a refcnt while the origin of src
610  * gains one.
611  */
612 static void dup_entry(struct blame_entry *dst, struct blame_entry *src)
613 {
614         struct blame_entry *p, *n;
615
616         p = dst->prev;
617         n = dst->next;
618         origin_incref(src->suspect);
619         origin_decref(dst->suspect);
620         memcpy(dst, src, sizeof(*src));
621         dst->prev = p;
622         dst->next = n;
623         dst->score = 0;
624 }
625
626 static const char *nth_line(struct scoreboard *sb, int lno)
627 {
628         return sb->final_buf + sb->lineno[lno];
629 }
630
631 /*
632  * It is known that lines between tlno to same came from parent, and e
633  * has an overlap with that range.  it also is known that parent's
634  * line plno corresponds to e's line tlno.
635  *
636  *                <---- e ----->
637  *                   <------>
638  *                   <------------>
639  *             <------------>
640  *             <------------------>
641  *
642  * Split e into potentially three parts; before this chunk, the chunk
643  * to be blamed for the parent, and after that portion.
644  */
645 static void split_overlap(struct blame_entry *split,
646                           struct blame_entry *e,
647                           int tlno, int plno, int same,
648                           struct origin *parent)
649 {
650         int chunk_end_lno;
651         memset(split, 0, sizeof(struct blame_entry [3]));
652
653         if (e->s_lno < tlno) {
654                 /* there is a pre-chunk part not blamed on parent */
655                 split[0].suspect = origin_incref(e->suspect);
656                 split[0].lno = e->lno;
657                 split[0].s_lno = e->s_lno;
658                 split[0].num_lines = tlno - e->s_lno;
659                 split[1].lno = e->lno + tlno - e->s_lno;
660                 split[1].s_lno = plno;
661         }
662         else {
663                 split[1].lno = e->lno;
664                 split[1].s_lno = plno + (e->s_lno - tlno);
665         }
666
667         if (same < e->s_lno + e->num_lines) {
668                 /* there is a post-chunk part not blamed on parent */
669                 split[2].suspect = origin_incref(e->suspect);
670                 split[2].lno = e->lno + (same - e->s_lno);
671                 split[2].s_lno = e->s_lno + (same - e->s_lno);
672                 split[2].num_lines = e->s_lno + e->num_lines - same;
673                 chunk_end_lno = split[2].lno;
674         }
675         else
676                 chunk_end_lno = e->lno + e->num_lines;
677         split[1].num_lines = chunk_end_lno - split[1].lno;
678
679         /*
680          * if it turns out there is nothing to blame the parent for,
681          * forget about the splitting.  !split[1].suspect signals this.
682          */
683         if (split[1].num_lines < 1)
684                 return;
685         split[1].suspect = origin_incref(parent);
686 }
687
688 /*
689  * split_overlap() divided an existing blame e into up to three parts
690  * in split.  Adjust the linked list of blames in the scoreboard to
691  * reflect the split.
692  */
693 static void split_blame(struct scoreboard *sb,
694                         struct blame_entry *split,
695                         struct blame_entry *e)
696 {
697         struct blame_entry *new_entry;
698
699         if (split[0].suspect && split[2].suspect) {
700                 /* The first part (reuse storage for the existing entry e) */
701                 dup_entry(e, &split[0]);
702
703                 /* The last part -- me */
704                 new_entry = xmalloc(sizeof(*new_entry));
705                 memcpy(new_entry, &(split[2]), sizeof(struct blame_entry));
706                 add_blame_entry(sb, new_entry);
707
708                 /* ... and the middle part -- parent */
709                 new_entry = xmalloc(sizeof(*new_entry));
710                 memcpy(new_entry, &(split[1]), sizeof(struct blame_entry));
711                 add_blame_entry(sb, new_entry);
712         }
713         else if (!split[0].suspect && !split[2].suspect)
714                 /*
715                  * The parent covers the entire area; reuse storage for
716                  * e and replace it with the parent.
717                  */
718                 dup_entry(e, &split[1]);
719         else if (split[0].suspect) {
720                 /* me and then parent */
721                 dup_entry(e, &split[0]);
722
723                 new_entry = xmalloc(sizeof(*new_entry));
724                 memcpy(new_entry, &(split[1]), sizeof(struct blame_entry));
725                 add_blame_entry(sb, new_entry);
726         }
727         else {
728                 /* parent and then me */
729                 dup_entry(e, &split[1]);
730
731                 new_entry = xmalloc(sizeof(*new_entry));
732                 memcpy(new_entry, &(split[2]), sizeof(struct blame_entry));
733                 add_blame_entry(sb, new_entry);
734         }
735
736         if (DEBUG) { /* sanity */
737                 struct blame_entry *ent;
738                 int lno = sb->ent->lno, corrupt = 0;
739
740                 for (ent = sb->ent; ent; ent = ent->next) {
741                         if (lno != ent->lno)
742                                 corrupt = 1;
743                         if (ent->s_lno < 0)
744                                 corrupt = 1;
745                         lno += ent->num_lines;
746                 }
747                 if (corrupt) {
748                         lno = sb->ent->lno;
749                         for (ent = sb->ent; ent; ent = ent->next) {
750                                 printf("L %8d l %8d n %8d\n",
751                                        lno, ent->lno, ent->num_lines);
752                                 lno = ent->lno + ent->num_lines;
753                         }
754                         die("oops");
755                 }
756         }
757 }
758
759 /*
760  * After splitting the blame, the origins used by the
761  * on-stack blame_entry should lose one refcnt each.
762  */
763 static void decref_split(struct blame_entry *split)
764 {
765         int i;
766
767         for (i = 0; i < 3; i++)
768                 origin_decref(split[i].suspect);
769 }
770
771 /*
772  * Helper for blame_chunk().  blame_entry e is known to overlap with
773  * the patch hunk; split it and pass blame to the parent.
774  */
775 static void blame_overlap(struct scoreboard *sb, struct blame_entry *e,
776                           int tlno, int plno, int same,
777                           struct origin *parent)
778 {
779         struct blame_entry split[3];
780
781         split_overlap(split, e, tlno, plno, same, parent);
782         if (split[1].suspect)
783                 split_blame(sb, split, e);
784         decref_split(split);
785 }
786
787 /*
788  * Find the line number of the last line the target is suspected for.
789  */
790 static int find_last_in_target(struct scoreboard *sb, struct origin *target)
791 {
792         struct blame_entry *e;
793         int last_in_target = -1;
794
795         for (e = sb->ent; e; e = e->next) {
796                 if (e->guilty || !same_suspect(e->suspect, target))
797                         continue;
798                 if (last_in_target < e->s_lno + e->num_lines)
799                         last_in_target = e->s_lno + e->num_lines;
800         }
801         return last_in_target;
802 }
803
804 /*
805  * Process one hunk from the patch between the current suspect for
806  * blame_entry e and its parent.  Find and split the overlap, and
807  * pass blame to the overlapping part to the parent.
808  */
809 static void blame_chunk(struct scoreboard *sb,
810                         int tlno, int plno, int same,
811                         struct origin *target, struct origin *parent)
812 {
813         struct blame_entry *e;
814
815         for (e = sb->ent; e; e = e->next) {
816                 if (e->guilty || !same_suspect(e->suspect, target))
817                         continue;
818                 if (same <= e->s_lno)
819                         continue;
820                 if (tlno < e->s_lno + e->num_lines)
821                         blame_overlap(sb, e, tlno, plno, same, parent);
822         }
823 }
824
825 /*
826  * We are looking at the origin 'target' and aiming to pass blame
827  * for the lines it is suspected to its parent.  Run diff to find
828  * which lines came from parent and pass blame for them.
829  */
830 static int pass_blame_to_parent(struct scoreboard *sb,
831                                 struct origin *target,
832                                 struct origin *parent)
833 {
834         int i, last_in_target, plno, tlno;
835         struct patch *patch;
836
837         last_in_target = find_last_in_target(sb, target);
838         if (last_in_target < 0)
839                 return 1; /* nothing remains for this target */
840
841         patch = get_patch(parent, target);
842         plno = tlno = 0;
843         for (i = 0; i < patch->num; i++) {
844                 struct chunk *chunk = &patch->chunks[i];
845
846                 blame_chunk(sb, tlno, plno, chunk->same, target, parent);
847                 plno = chunk->p_next;
848                 tlno = chunk->t_next;
849         }
850         /* The rest (i.e. anything after tlno) are the same as the parent */
851         blame_chunk(sb, tlno, plno, last_in_target, target, parent);
852
853         free_patch(patch);
854         return 0;
855 }
856
857 /*
858  * The lines in blame_entry after splitting blames many times can become
859  * very small and trivial, and at some point it becomes pointless to
860  * blame the parents.  E.g. "\t\t}\n\t}\n\n" appears everywhere in any
861  * ordinary C program, and it is not worth to say it was copied from
862  * totally unrelated file in the parent.
863  *
864  * Compute how trivial the lines in the blame_entry are.
865  */
866 static unsigned ent_score(struct scoreboard *sb, struct blame_entry *e)
867 {
868         unsigned score;
869         const char *cp, *ep;
870
871         if (e->score)
872                 return e->score;
873
874         score = 1;
875         cp = nth_line(sb, e->lno);
876         ep = nth_line(sb, e->lno + e->num_lines);
877         while (cp < ep) {
878                 unsigned ch = *((unsigned char *)cp);
879                 if (isalnum(ch))
880                         score++;
881                 cp++;
882         }
883         e->score = score;
884         return score;
885 }
886
887 /*
888  * best_so_far[] and this[] are both a split of an existing blame_entry
889  * that passes blame to the parent.  Maintain best_so_far the best split
890  * so far, by comparing this and best_so_far and copying this into
891  * bst_so_far as needed.
892  */
893 static void copy_split_if_better(struct scoreboard *sb,
894                                  struct blame_entry *best_so_far,
895                                  struct blame_entry *this)
896 {
897         int i;
898
899         if (!this[1].suspect)
900                 return;
901         if (best_so_far[1].suspect) {
902                 if (ent_score(sb, &this[1]) < ent_score(sb, &best_so_far[1]))
903                         return;
904         }
905
906         for (i = 0; i < 3; i++)
907                 origin_incref(this[i].suspect);
908         decref_split(best_so_far);
909         memcpy(best_so_far, this, sizeof(struct blame_entry [3]));
910 }
911
912 /*
913  * We are looking at a part of the final image represented by
914  * ent (tlno and same are offset by ent->s_lno).
915  * tlno is where we are looking at in the final image.
916  * up to (but not including) same match preimage.
917  * plno is where we are looking at in the preimage.
918  *
919  * <-------------- final image ---------------------->
920  *       <------ent------>
921  *         ^tlno ^same
922  *    <---------preimage----->
923  *         ^plno
924  *
925  * All line numbers are 0-based.
926  */
927 static void handle_split(struct scoreboard *sb,
928                          struct blame_entry *ent,
929                          int tlno, int plno, int same,
930                          struct origin *parent,
931                          struct blame_entry *split)
932 {
933         if (ent->num_lines <= tlno)
934                 return;
935         if (tlno < same) {
936                 struct blame_entry this[3];
937                 tlno += ent->s_lno;
938                 same += ent->s_lno;
939                 split_overlap(this, ent, tlno, plno, same, parent);
940                 copy_split_if_better(sb, split, this);
941                 decref_split(this);
942         }
943 }
944
945 /*
946  * Find the lines from parent that are the same as ent so that
947  * we can pass blames to it.  file_p has the blob contents for
948  * the parent.
949  */
950 static void find_copy_in_blob(struct scoreboard *sb,
951                               struct blame_entry *ent,
952                               struct origin *parent,
953                               struct blame_entry *split,
954                               mmfile_t *file_p)
955 {
956         const char *cp;
957         int cnt;
958         mmfile_t file_o;
959         struct patch *patch;
960         int i, plno, tlno;
961
962         /*
963          * Prepare mmfile that contains only the lines in ent.
964          */
965         cp = nth_line(sb, ent->lno);
966         file_o.ptr = (char*) cp;
967         cnt = ent->num_lines;
968
969         while (cnt && cp < sb->final_buf + sb->final_buf_size) {
970                 if (*cp++ == '\n')
971                         cnt--;
972         }
973         file_o.size = cp - file_o.ptr;
974
975         patch = compare_buffer(file_p, &file_o, 1);
976
977         /*
978          * file_o is a part of final image we are annotating.
979          * file_p partially may match that image.
980          */
981         memset(split, 0, sizeof(struct blame_entry [3]));
982         plno = tlno = 0;
983         for (i = 0; i < patch->num; i++) {
984                 struct chunk *chunk = &patch->chunks[i];
985
986                 handle_split(sb, ent, tlno, plno, chunk->same, parent, split);
987                 plno = chunk->p_next;
988                 tlno = chunk->t_next;
989         }
990         /* remainder, if any, all match the preimage */
991         handle_split(sb, ent, tlno, plno, ent->num_lines, parent, split);
992         free_patch(patch);
993 }
994
995 /*
996  * See if lines currently target is suspected for can be attributed to
997  * parent.
998  */
999 static int find_move_in_parent(struct scoreboard *sb,
1000                                struct origin *target,
1001                                struct origin *parent)
1002 {
1003         int last_in_target, made_progress;
1004         struct blame_entry *e, split[3];
1005         mmfile_t file_p;
1006
1007         last_in_target = find_last_in_target(sb, target);
1008         if (last_in_target < 0)
1009                 return 1; /* nothing remains for this target */
1010
1011         fill_origin_blob(parent, &file_p);
1012         if (!file_p.ptr)
1013                 return 0;
1014
1015         made_progress = 1;
1016         while (made_progress) {
1017                 made_progress = 0;
1018                 for (e = sb->ent; e; e = e->next) {
1019                         if (e->guilty || !same_suspect(e->suspect, target))
1020                                 continue;
1021                         find_copy_in_blob(sb, e, parent, split, &file_p);
1022                         if (split[1].suspect &&
1023                             blame_move_score < ent_score(sb, &split[1])) {
1024                                 split_blame(sb, split, e);
1025                                 made_progress = 1;
1026                         }
1027                         decref_split(split);
1028                 }
1029         }
1030         return 0;
1031 }
1032
1033 struct blame_list {
1034         struct blame_entry *ent;
1035         struct blame_entry split[3];
1036 };
1037
1038 /*
1039  * Count the number of entries the target is suspected for,
1040  * and prepare a list of entry and the best split.
1041  */
1042 static struct blame_list *setup_blame_list(struct scoreboard *sb,
1043                                            struct origin *target,
1044                                            int *num_ents_p)
1045 {
1046         struct blame_entry *e;
1047         int num_ents, i;
1048         struct blame_list *blame_list = NULL;
1049
1050         for (e = sb->ent, num_ents = 0; e; e = e->next)
1051                 if (!e->guilty && same_suspect(e->suspect, target))
1052                         num_ents++;
1053         if (num_ents) {
1054                 blame_list = xcalloc(num_ents, sizeof(struct blame_list));
1055                 for (e = sb->ent, i = 0; e; e = e->next)
1056                         if (!e->guilty && same_suspect(e->suspect, target))
1057                                 blame_list[i++].ent = e;
1058         }
1059         *num_ents_p = num_ents;
1060         return blame_list;
1061 }
1062
1063 /*
1064  * For lines target is suspected for, see if we can find code movement
1065  * across file boundary from the parent commit.  porigin is the path
1066  * in the parent we already tried.
1067  */
1068 static int find_copy_in_parent(struct scoreboard *sb,
1069                                struct origin *target,
1070                                struct commit *parent,
1071                                struct origin *porigin,
1072                                int opt)
1073 {
1074         struct diff_options diff_opts;
1075         const char *paths[1];
1076         int i, j;
1077         int retval;
1078         struct blame_list *blame_list;
1079         int num_ents;
1080
1081         blame_list = setup_blame_list(sb, target, &num_ents);
1082         if (!blame_list)
1083                 return 1; /* nothing remains for this target */
1084
1085         diff_setup(&diff_opts);
1086         DIFF_OPT_SET(&diff_opts, RECURSIVE);
1087         diff_opts.output_format = DIFF_FORMAT_NO_OUTPUT;
1088
1089         paths[0] = NULL;
1090         diff_tree_setup_paths(paths, &diff_opts);
1091         if (diff_setup_done(&diff_opts) < 0)
1092                 die("diff-setup");
1093
1094         /* Try "find copies harder" on new path if requested;
1095          * we do not want to use diffcore_rename() actually to
1096          * match things up; find_copies_harder is set only to
1097          * force diff_tree_sha1() to feed all filepairs to diff_queue,
1098          * and this code needs to be after diff_setup_done(), which
1099          * usually makes find-copies-harder imply copy detection.
1100          */
1101         if ((opt & PICKAXE_BLAME_COPY_HARDEST)
1102             || ((opt & PICKAXE_BLAME_COPY_HARDER)
1103                 && (!porigin || strcmp(target->path, porigin->path))))
1104                 DIFF_OPT_SET(&diff_opts, FIND_COPIES_HARDER);
1105
1106         if (is_null_sha1(target->commit->object.sha1))
1107                 do_diff_cache(parent->tree->object.sha1, &diff_opts);
1108         else
1109                 diff_tree_sha1(parent->tree->object.sha1,
1110                                target->commit->tree->object.sha1,
1111                                "", &diff_opts);
1112
1113         if (!DIFF_OPT_TST(&diff_opts, FIND_COPIES_HARDER))
1114                 diffcore_std(&diff_opts);
1115
1116         retval = 0;
1117         while (1) {
1118                 int made_progress = 0;
1119
1120                 for (i = 0; i < diff_queued_diff.nr; i++) {
1121                         struct diff_filepair *p = diff_queued_diff.queue[i];
1122                         struct origin *norigin;
1123                         mmfile_t file_p;
1124                         struct blame_entry this[3];
1125
1126                         if (!DIFF_FILE_VALID(p->one))
1127                                 continue; /* does not exist in parent */
1128                         if (porigin && !strcmp(p->one->path, porigin->path))
1129                                 /* find_move already dealt with this path */
1130                                 continue;
1131
1132                         norigin = get_origin(sb, parent, p->one->path);
1133                         hashcpy(norigin->blob_sha1, p->one->sha1);
1134                         fill_origin_blob(norigin, &file_p);
1135                         if (!file_p.ptr)
1136                                 continue;
1137
1138                         for (j = 0; j < num_ents; j++) {
1139                                 find_copy_in_blob(sb, blame_list[j].ent,
1140                                                   norigin, this, &file_p);
1141                                 copy_split_if_better(sb, blame_list[j].split,
1142                                                      this);
1143                                 decref_split(this);
1144                         }
1145                         origin_decref(norigin);
1146                 }
1147
1148                 for (j = 0; j < num_ents; j++) {
1149                         struct blame_entry *split = blame_list[j].split;
1150                         if (split[1].suspect &&
1151                             blame_copy_score < ent_score(sb, &split[1])) {
1152                                 split_blame(sb, split, blame_list[j].ent);
1153                                 made_progress = 1;
1154                         }
1155                         decref_split(split);
1156                 }
1157                 free(blame_list);
1158
1159                 if (!made_progress)
1160                         break;
1161                 blame_list = setup_blame_list(sb, target, &num_ents);
1162                 if (!blame_list) {
1163                         retval = 1;
1164                         break;
1165                 }
1166         }
1167         diff_flush(&diff_opts);
1168         diff_tree_release_paths(&diff_opts);
1169         return retval;
1170 }
1171
1172 /*
1173  * The blobs of origin and porigin exactly match, so everything
1174  * origin is suspected for can be blamed on the parent.
1175  */
1176 static void pass_whole_blame(struct scoreboard *sb,
1177                              struct origin *origin, struct origin *porigin)
1178 {
1179         struct blame_entry *e;
1180
1181         if (!porigin->file.ptr && origin->file.ptr) {
1182                 /* Steal its file */
1183                 porigin->file = origin->file;
1184                 origin->file.ptr = NULL;
1185         }
1186         for (e = sb->ent; e; e = e->next) {
1187                 if (!same_suspect(e->suspect, origin))
1188                         continue;
1189                 origin_incref(porigin);
1190                 origin_decref(e->suspect);
1191                 e->suspect = porigin;
1192         }
1193 }
1194
1195 #define MAXPARENT 16
1196
1197 static void pass_blame(struct scoreboard *sb, struct origin *origin, int opt)
1198 {
1199         int i, pass;
1200         struct commit *commit = origin->commit;
1201         struct commit_list *parent;
1202         struct origin *parent_origin[MAXPARENT], *porigin;
1203
1204         memset(parent_origin, 0, sizeof(parent_origin));
1205
1206         /* The first pass looks for unrenamed path to optimize for
1207          * common cases, then we look for renames in the second pass.
1208          */
1209         for (pass = 0; pass < 2; pass++) {
1210                 struct origin *(*find)(struct scoreboard *,
1211                                        struct commit *, struct origin *);
1212                 find = pass ? find_rename : find_origin;
1213
1214                 for (i = 0, parent = commit->parents;
1215                      i < MAXPARENT && parent;
1216                      parent = parent->next, i++) {
1217                         struct commit *p = parent->item;
1218                         int j, same;
1219
1220                         if (parent_origin[i])
1221                                 continue;
1222                         if (parse_commit(p))
1223                                 continue;
1224                         porigin = find(sb, p, origin);
1225                         if (!porigin)
1226                                 continue;
1227                         if (!hashcmp(porigin->blob_sha1, origin->blob_sha1)) {
1228                                 pass_whole_blame(sb, origin, porigin);
1229                                 origin_decref(porigin);
1230                                 goto finish;
1231                         }
1232                         for (j = same = 0; j < i; j++)
1233                                 if (parent_origin[j] &&
1234                                     !hashcmp(parent_origin[j]->blob_sha1,
1235                                              porigin->blob_sha1)) {
1236                                         same = 1;
1237                                         break;
1238                                 }
1239                         if (!same)
1240                                 parent_origin[i] = porigin;
1241                         else
1242                                 origin_decref(porigin);
1243                 }
1244         }
1245
1246         num_commits++;
1247         for (i = 0, parent = commit->parents;
1248              i < MAXPARENT && parent;
1249              parent = parent->next, i++) {
1250                 struct origin *porigin = parent_origin[i];
1251                 if (!porigin)
1252                         continue;
1253                 if (pass_blame_to_parent(sb, origin, porigin))
1254                         goto finish;
1255         }
1256
1257         /*
1258          * Optionally find moves in parents' files.
1259          */
1260         if (opt & PICKAXE_BLAME_MOVE)
1261                 for (i = 0, parent = commit->parents;
1262                      i < MAXPARENT && parent;
1263                      parent = parent->next, i++) {
1264                         struct origin *porigin = parent_origin[i];
1265                         if (!porigin)
1266                                 continue;
1267                         if (find_move_in_parent(sb, origin, porigin))
1268                                 goto finish;
1269                 }
1270
1271         /*
1272          * Optionally find copies from parents' files.
1273          */
1274         if (opt & PICKAXE_BLAME_COPY)
1275                 for (i = 0, parent = commit->parents;
1276                      i < MAXPARENT && parent;
1277                      parent = parent->next, i++) {
1278                         struct origin *porigin = parent_origin[i];
1279                         if (find_copy_in_parent(sb, origin, parent->item,
1280                                                 porigin, opt))
1281                                 goto finish;
1282                 }
1283
1284  finish:
1285         for (i = 0; i < MAXPARENT; i++) {
1286                 if (parent_origin[i]) {
1287                         drop_origin_blob(parent_origin[i]);
1288                         origin_decref(parent_origin[i]);
1289                 }
1290         }
1291         drop_origin_blob(origin);
1292 }
1293
1294 /*
1295  * Information on commits, used for output.
1296  */
1297 struct commit_info
1298 {
1299         const char *author;
1300         const char *author_mail;
1301         unsigned long author_time;
1302         const char *author_tz;
1303
1304         /* filled only when asked for details */
1305         const char *committer;
1306         const char *committer_mail;
1307         unsigned long committer_time;
1308         const char *committer_tz;
1309
1310         const char *summary;
1311 };
1312
1313 /*
1314  * Parse author/committer line in the commit object buffer
1315  */
1316 static void get_ac_line(const char *inbuf, const char *what,
1317                         int bufsz, char *person, const char **mail,
1318                         unsigned long *time, const char **tz)
1319 {
1320         int len, tzlen, maillen;
1321         char *tmp, *endp, *timepos;
1322
1323         tmp = strstr(inbuf, what);
1324         if (!tmp)
1325                 goto error_out;
1326         tmp += strlen(what);
1327         endp = strchr(tmp, '\n');
1328         if (!endp)
1329                 len = strlen(tmp);
1330         else
1331                 len = endp - tmp;
1332         if (bufsz <= len) {
1333         error_out:
1334                 /* Ugh */
1335                 *mail = *tz = "(unknown)";
1336                 *time = 0;
1337                 return;
1338         }
1339         memcpy(person, tmp, len);
1340
1341         tmp = person;
1342         tmp += len;
1343         *tmp = 0;
1344         while (*tmp != ' ')
1345                 tmp--;
1346         *tz = tmp+1;
1347         tzlen = (person+len)-(tmp+1);
1348
1349         *tmp = 0;
1350         while (*tmp != ' ')
1351                 tmp--;
1352         *time = strtoul(tmp, NULL, 10);
1353         timepos = tmp;
1354
1355         *tmp = 0;
1356         while (*tmp != ' ')
1357                 tmp--;
1358         *mail = tmp + 1;
1359         *tmp = 0;
1360         maillen = timepos - tmp;
1361
1362         if (!mailmap.nr)
1363                 return;
1364
1365         /*
1366          * mailmap expansion may make the name longer.
1367          * make room by pushing stuff down.
1368          */
1369         tmp = person + bufsz - (tzlen + 1);
1370         memmove(tmp, *tz, tzlen);
1371         tmp[tzlen] = 0;
1372         *tz = tmp;
1373
1374         tmp = tmp - (maillen + 1);
1375         memmove(tmp, *mail, maillen);
1376         tmp[maillen] = 0;
1377         *mail = tmp;
1378
1379         /*
1380          * Now, convert e-mail using mailmap
1381          */
1382         map_email(&mailmap, tmp + 1, person, tmp-person-1);
1383 }
1384
1385 static void get_commit_info(struct commit *commit,
1386                             struct commit_info *ret,
1387                             int detailed)
1388 {
1389         int len;
1390         char *tmp, *endp;
1391         static char author_buf[1024];
1392         static char committer_buf[1024];
1393         static char summary_buf[1024];
1394
1395         /*
1396          * We've operated without save_commit_buffer, so
1397          * we now need to populate them for output.
1398          */
1399         if (!commit->buffer) {
1400                 enum object_type type;
1401                 unsigned long size;
1402                 commit->buffer =
1403                         read_sha1_file(commit->object.sha1, &type, &size);
1404                 if (!commit->buffer)
1405                         die("Cannot read commit %s",
1406                             sha1_to_hex(commit->object.sha1));
1407         }
1408         ret->author = author_buf;
1409         get_ac_line(commit->buffer, "\nauthor ",
1410                     sizeof(author_buf), author_buf, &ret->author_mail,
1411                     &ret->author_time, &ret->author_tz);
1412
1413         if (!detailed)
1414                 return;
1415
1416         ret->committer = committer_buf;
1417         get_ac_line(commit->buffer, "\ncommitter ",
1418                     sizeof(committer_buf), committer_buf, &ret->committer_mail,
1419                     &ret->committer_time, &ret->committer_tz);
1420
1421         ret->summary = summary_buf;
1422         tmp = strstr(commit->buffer, "\n\n");
1423         if (!tmp) {
1424         error_out:
1425                 sprintf(summary_buf, "(%s)", sha1_to_hex(commit->object.sha1));
1426                 return;
1427         }
1428         tmp += 2;
1429         endp = strchr(tmp, '\n');
1430         if (!endp)
1431                 endp = tmp + strlen(tmp);
1432         len = endp - tmp;
1433         if (len >= sizeof(summary_buf) || len == 0)
1434                 goto error_out;
1435         memcpy(summary_buf, tmp, len);
1436         summary_buf[len] = 0;
1437 }
1438
1439 /*
1440  * To allow LF and other nonportable characters in pathnames,
1441  * they are c-style quoted as needed.
1442  */
1443 static void write_filename_info(const char *path)
1444 {
1445         printf("filename ");
1446         write_name_quoted(path, stdout, '\n');
1447 }
1448
1449 /*
1450  * The blame_entry is found to be guilty for the range.  Mark it
1451  * as such, and show it in incremental output.
1452  */
1453 static void found_guilty_entry(struct blame_entry *ent)
1454 {
1455         if (ent->guilty)
1456                 return;
1457         ent->guilty = 1;
1458         if (incremental) {
1459                 struct origin *suspect = ent->suspect;
1460
1461                 printf("%s %d %d %d\n",
1462                        sha1_to_hex(suspect->commit->object.sha1),
1463                        ent->s_lno + 1, ent->lno + 1, ent->num_lines);
1464                 if (!(suspect->commit->object.flags & METAINFO_SHOWN)) {
1465                         struct commit_info ci;
1466                         suspect->commit->object.flags |= METAINFO_SHOWN;
1467                         get_commit_info(suspect->commit, &ci, 1);
1468                         printf("author %s\n", ci.author);
1469                         printf("author-mail %s\n", ci.author_mail);
1470                         printf("author-time %lu\n", ci.author_time);
1471                         printf("author-tz %s\n", ci.author_tz);
1472                         printf("committer %s\n", ci.committer);
1473                         printf("committer-mail %s\n", ci.committer_mail);
1474                         printf("committer-time %lu\n", ci.committer_time);
1475                         printf("committer-tz %s\n", ci.committer_tz);
1476                         printf("summary %s\n", ci.summary);
1477                         if (suspect->commit->object.flags & UNINTERESTING)
1478                                 printf("boundary\n");
1479                 }
1480                 write_filename_info(suspect->path);
1481                 maybe_flush_or_die(stdout, "stdout");
1482         }
1483 }
1484
1485 /*
1486  * The main loop -- while the scoreboard has lines whose true origin
1487  * is still unknown, pick one blame_entry, and allow its current
1488  * suspect to pass blames to its parents.
1489  */
1490 static void assign_blame(struct scoreboard *sb, struct rev_info *revs, int opt)
1491 {
1492         while (1) {
1493                 struct blame_entry *ent;
1494                 struct commit *commit;
1495                 struct origin *suspect = NULL;
1496
1497                 /* find one suspect to break down */
1498                 for (ent = sb->ent; !suspect && ent; ent = ent->next)
1499                         if (!ent->guilty)
1500                                 suspect = ent->suspect;
1501                 if (!suspect)
1502                         return; /* all done */
1503
1504                 /*
1505                  * We will use this suspect later in the loop,
1506                  * so hold onto it in the meantime.
1507                  */
1508                 origin_incref(suspect);
1509                 commit = suspect->commit;
1510                 if (!commit->object.parsed)
1511                         parse_commit(commit);
1512                 if (!(commit->object.flags & UNINTERESTING) &&
1513                     !(revs->max_age != -1 && commit->date < revs->max_age))
1514                         pass_blame(sb, suspect, opt);
1515                 else {
1516                         commit->object.flags |= UNINTERESTING;
1517                         if (commit->object.parsed)
1518                                 mark_parents_uninteresting(commit);
1519                 }
1520                 /* treat root commit as boundary */
1521                 if (!commit->parents && !show_root)
1522                         commit->object.flags |= UNINTERESTING;
1523
1524                 /* Take responsibility for the remaining entries */
1525                 for (ent = sb->ent; ent; ent = ent->next)
1526                         if (same_suspect(ent->suspect, suspect))
1527                                 found_guilty_entry(ent);
1528                 origin_decref(suspect);
1529
1530                 if (DEBUG) /* sanity */
1531                         sanity_check_refcnt(sb);
1532         }
1533 }
1534
1535 static const char *format_time(unsigned long time, const char *tz_str,
1536                                int show_raw_time)
1537 {
1538         static char time_buf[128];
1539         time_t t = time;
1540         int minutes, tz;
1541         struct tm *tm;
1542
1543         if (show_raw_time) {
1544                 sprintf(time_buf, "%lu %s", time, tz_str);
1545                 return time_buf;
1546         }
1547
1548         tz = atoi(tz_str);
1549         minutes = tz < 0 ? -tz : tz;
1550         minutes = (minutes / 100)*60 + (minutes % 100);
1551         minutes = tz < 0 ? -minutes : minutes;
1552         t = time + minutes * 60;
1553         tm = gmtime(&t);
1554
1555         strftime(time_buf, sizeof(time_buf), "%Y-%m-%d %H:%M:%S ", tm);
1556         strcat(time_buf, tz_str);
1557         return time_buf;
1558 }
1559
1560 #define OUTPUT_ANNOTATE_COMPAT  001
1561 #define OUTPUT_LONG_OBJECT_NAME 002
1562 #define OUTPUT_RAW_TIMESTAMP    004
1563 #define OUTPUT_PORCELAIN        010
1564 #define OUTPUT_SHOW_NAME        020
1565 #define OUTPUT_SHOW_NUMBER      040
1566 #define OUTPUT_SHOW_SCORE      0100
1567 #define OUTPUT_NO_AUTHOR       0200
1568
1569 static void emit_porcelain(struct scoreboard *sb, struct blame_entry *ent)
1570 {
1571         int cnt;
1572         const char *cp;
1573         struct origin *suspect = ent->suspect;
1574         char hex[41];
1575
1576         strcpy(hex, sha1_to_hex(suspect->commit->object.sha1));
1577         printf("%s%c%d %d %d\n",
1578                hex,
1579                ent->guilty ? ' ' : '*', // purely for debugging
1580                ent->s_lno + 1,
1581                ent->lno + 1,
1582                ent->num_lines);
1583         if (!(suspect->commit->object.flags & METAINFO_SHOWN)) {
1584                 struct commit_info ci;
1585                 suspect->commit->object.flags |= METAINFO_SHOWN;
1586                 get_commit_info(suspect->commit, &ci, 1);
1587                 printf("author %s\n", ci.author);
1588                 printf("author-mail %s\n", ci.author_mail);
1589                 printf("author-time %lu\n", ci.author_time);
1590                 printf("author-tz %s\n", ci.author_tz);
1591                 printf("committer %s\n", ci.committer);
1592                 printf("committer-mail %s\n", ci.committer_mail);
1593                 printf("committer-time %lu\n", ci.committer_time);
1594                 printf("committer-tz %s\n", ci.committer_tz);
1595                 write_filename_info(suspect->path);
1596                 printf("summary %s\n", ci.summary);
1597                 if (suspect->commit->object.flags & UNINTERESTING)
1598                         printf("boundary\n");
1599         }
1600         else if (suspect->commit->object.flags & MORE_THAN_ONE_PATH)
1601                 write_filename_info(suspect->path);
1602
1603         cp = nth_line(sb, ent->lno);
1604         for (cnt = 0; cnt < ent->num_lines; cnt++) {
1605                 char ch;
1606                 if (cnt)
1607                         printf("%s %d %d\n", hex,
1608                                ent->s_lno + 1 + cnt,
1609                                ent->lno + 1 + cnt);
1610                 putchar('\t');
1611                 do {
1612                         ch = *cp++;
1613                         putchar(ch);
1614                 } while (ch != '\n' &&
1615                          cp < sb->final_buf + sb->final_buf_size);
1616         }
1617 }
1618
1619 static void emit_other(struct scoreboard *sb, struct blame_entry *ent, int opt)
1620 {
1621         int cnt;
1622         const char *cp;
1623         struct origin *suspect = ent->suspect;
1624         struct commit_info ci;
1625         char hex[41];
1626         int show_raw_time = !!(opt & OUTPUT_RAW_TIMESTAMP);
1627
1628         get_commit_info(suspect->commit, &ci, 1);
1629         strcpy(hex, sha1_to_hex(suspect->commit->object.sha1));
1630
1631         cp = nth_line(sb, ent->lno);
1632         for (cnt = 0; cnt < ent->num_lines; cnt++) {
1633                 char ch;
1634                 int length = (opt & OUTPUT_LONG_OBJECT_NAME) ? 40 : 8;
1635
1636                 if (suspect->commit->object.flags & UNINTERESTING) {
1637                         if (blank_boundary)
1638                                 memset(hex, ' ', length);
1639                         else if (!cmd_is_annotate) {
1640                                 length--;
1641                                 putchar('^');
1642                         }
1643                 }
1644
1645                 printf("%.*s", length, hex);
1646                 if (opt & OUTPUT_ANNOTATE_COMPAT)
1647                         printf("\t(%10s\t%10s\t%d)", ci.author,
1648                                format_time(ci.author_time, ci.author_tz,
1649                                            show_raw_time),
1650                                ent->lno + 1 + cnt);
1651                 else {
1652                         if (opt & OUTPUT_SHOW_SCORE)
1653                                 printf(" %*d %02d",
1654                                        max_score_digits, ent->score,
1655                                        ent->suspect->refcnt);
1656                         if (opt & OUTPUT_SHOW_NAME)
1657                                 printf(" %-*.*s", longest_file, longest_file,
1658                                        suspect->path);
1659                         if (opt & OUTPUT_SHOW_NUMBER)
1660                                 printf(" %*d", max_orig_digits,
1661                                        ent->s_lno + 1 + cnt);
1662
1663                         if (!(opt & OUTPUT_NO_AUTHOR))
1664                                 printf(" (%-*.*s %10s",
1665                                        longest_author, longest_author,
1666                                        ci.author,
1667                                        format_time(ci.author_time,
1668                                                    ci.author_tz,
1669                                                    show_raw_time));
1670                         printf(" %*d) ",
1671                                max_digits, ent->lno + 1 + cnt);
1672                 }
1673                 do {
1674                         ch = *cp++;
1675                         putchar(ch);
1676                 } while (ch != '\n' &&
1677                          cp < sb->final_buf + sb->final_buf_size);
1678         }
1679 }
1680
1681 static void output(struct scoreboard *sb, int option)
1682 {
1683         struct blame_entry *ent;
1684
1685         if (option & OUTPUT_PORCELAIN) {
1686                 for (ent = sb->ent; ent; ent = ent->next) {
1687                         struct blame_entry *oth;
1688                         struct origin *suspect = ent->suspect;
1689                         struct commit *commit = suspect->commit;
1690                         if (commit->object.flags & MORE_THAN_ONE_PATH)
1691                                 continue;
1692                         for (oth = ent->next; oth; oth = oth->next) {
1693                                 if ((oth->suspect->commit != commit) ||
1694                                     !strcmp(oth->suspect->path, suspect->path))
1695                                         continue;
1696                                 commit->object.flags |= MORE_THAN_ONE_PATH;
1697                                 break;
1698                         }
1699                 }
1700         }
1701
1702         for (ent = sb->ent; ent; ent = ent->next) {
1703                 if (option & OUTPUT_PORCELAIN)
1704                         emit_porcelain(sb, ent);
1705                 else {
1706                         emit_other(sb, ent, option);
1707                 }
1708         }
1709 }
1710
1711 /*
1712  * To allow quick access to the contents of nth line in the
1713  * final image, prepare an index in the scoreboard.
1714  */
1715 static int prepare_lines(struct scoreboard *sb)
1716 {
1717         const char *buf = sb->final_buf;
1718         unsigned long len = sb->final_buf_size;
1719         int num = 0, incomplete = 0, bol = 1;
1720
1721         if (len && buf[len-1] != '\n')
1722                 incomplete++; /* incomplete line at the end */
1723         while (len--) {
1724                 if (bol) {
1725                         sb->lineno = xrealloc(sb->lineno,
1726                                               sizeof(int* ) * (num + 1));
1727                         sb->lineno[num] = buf - sb->final_buf;
1728                         bol = 0;
1729                 }
1730                 if (*buf++ == '\n') {
1731                         num++;
1732                         bol = 1;
1733                 }
1734         }
1735         sb->lineno = xrealloc(sb->lineno,
1736                               sizeof(int* ) * (num + incomplete + 1));
1737         sb->lineno[num + incomplete] = buf - sb->final_buf;
1738         sb->num_lines = num + incomplete;
1739         return sb->num_lines;
1740 }
1741
1742 /*
1743  * Add phony grafts for use with -S; this is primarily to
1744  * support git-cvsserver that wants to give a linear history
1745  * to its clients.
1746  */
1747 static int read_ancestry(const char *graft_file)
1748 {
1749         FILE *fp = fopen(graft_file, "r");
1750         char buf[1024];
1751         if (!fp)
1752                 return -1;
1753         while (fgets(buf, sizeof(buf), fp)) {
1754                 /* The format is just "Commit Parent1 Parent2 ...\n" */
1755                 int len = strlen(buf);
1756                 struct commit_graft *graft = read_graft_line(buf, len);
1757                 if (graft)
1758                         register_commit_graft(graft, 0);
1759         }
1760         fclose(fp);
1761         return 0;
1762 }
1763
1764 /*
1765  * How many columns do we need to show line numbers in decimal?
1766  */
1767 static int lineno_width(int lines)
1768 {
1769         int i, width;
1770
1771         for (width = 1, i = 10; i <= lines + 1; width++)
1772                 i *= 10;
1773         return width;
1774 }
1775
1776 /*
1777  * How many columns do we need to show line numbers, authors,
1778  * and filenames?
1779  */
1780 static void find_alignment(struct scoreboard *sb, int *option)
1781 {
1782         int longest_src_lines = 0;
1783         int longest_dst_lines = 0;
1784         unsigned largest_score = 0;
1785         struct blame_entry *e;
1786
1787         for (e = sb->ent; e; e = e->next) {
1788                 struct origin *suspect = e->suspect;
1789                 struct commit_info ci;
1790                 int num;
1791
1792                 if (strcmp(suspect->path, sb->path))
1793                         *option |= OUTPUT_SHOW_NAME;
1794                 num = strlen(suspect->path);
1795                 if (longest_file < num)
1796                         longest_file = num;
1797                 if (!(suspect->commit->object.flags & METAINFO_SHOWN)) {
1798                         suspect->commit->object.flags |= METAINFO_SHOWN;
1799                         get_commit_info(suspect->commit, &ci, 1);
1800                         num = strlen(ci.author);
1801                         if (longest_author < num)
1802                                 longest_author = num;
1803                 }
1804                 num = e->s_lno + e->num_lines;
1805                 if (longest_src_lines < num)
1806                         longest_src_lines = num;
1807                 num = e->lno + e->num_lines;
1808                 if (longest_dst_lines < num)
1809                         longest_dst_lines = num;
1810                 if (largest_score < ent_score(sb, e))
1811                         largest_score = ent_score(sb, e);
1812         }
1813         max_orig_digits = lineno_width(longest_src_lines);
1814         max_digits = lineno_width(longest_dst_lines);
1815         max_score_digits = lineno_width(largest_score);
1816 }
1817
1818 /*
1819  * For debugging -- origin is refcounted, and this asserts that
1820  * we do not underflow.
1821  */
1822 static void sanity_check_refcnt(struct scoreboard *sb)
1823 {
1824         int baa = 0;
1825         struct blame_entry *ent;
1826
1827         for (ent = sb->ent; ent; ent = ent->next) {
1828                 /* Nobody should have zero or negative refcnt */
1829                 if (ent->suspect->refcnt <= 0) {
1830                         fprintf(stderr, "%s in %s has negative refcnt %d\n",
1831                                 ent->suspect->path,
1832                                 sha1_to_hex(ent->suspect->commit->object.sha1),
1833                                 ent->suspect->refcnt);
1834                         baa = 1;
1835                 }
1836         }
1837         for (ent = sb->ent; ent; ent = ent->next) {
1838                 /* Mark the ones that haven't been checked */
1839                 if (0 < ent->suspect->refcnt)
1840                         ent->suspect->refcnt = -ent->suspect->refcnt;
1841         }
1842         for (ent = sb->ent; ent; ent = ent->next) {
1843                 /*
1844                  * ... then pick each and see if they have the the
1845                  * correct refcnt.
1846                  */
1847                 int found;
1848                 struct blame_entry *e;
1849                 struct origin *suspect = ent->suspect;
1850
1851                 if (0 < suspect->refcnt)
1852                         continue;
1853                 suspect->refcnt = -suspect->refcnt; /* Unmark */
1854                 for (found = 0, e = sb->ent; e; e = e->next) {
1855                         if (e->suspect != suspect)
1856                                 continue;
1857                         found++;
1858                 }
1859                 if (suspect->refcnt != found) {
1860                         fprintf(stderr, "%s in %s has refcnt %d, not %d\n",
1861                                 ent->suspect->path,
1862                                 sha1_to_hex(ent->suspect->commit->object.sha1),
1863                                 ent->suspect->refcnt, found);
1864                         baa = 2;
1865                 }
1866         }
1867         if (baa) {
1868                 int opt = 0160;
1869                 find_alignment(sb, &opt);
1870                 output(sb, opt);
1871                 die("Baa %d!", baa);
1872         }
1873 }
1874
1875 /*
1876  * Used for the command line parsing; check if the path exists
1877  * in the working tree.
1878  */
1879 static int has_path_in_work_tree(const char *path)
1880 {
1881         struct stat st;
1882         return !lstat(path, &st);
1883 }
1884
1885 static unsigned parse_score(const char *arg)
1886 {
1887         char *end;
1888         unsigned long score = strtoul(arg, &end, 10);
1889         if (*end)
1890                 return 0;
1891         return score;
1892 }
1893
1894 static const char *add_prefix(const char *prefix, const char *path)
1895 {
1896         return prefix_path(prefix, prefix ? strlen(prefix) : 0, path);
1897 }
1898
1899 /*
1900  * Parsing of (comma separated) one item in the -L option
1901  */
1902 static const char *parse_loc(const char *spec,
1903                              struct scoreboard *sb, long lno,
1904                              long begin, long *ret)
1905 {
1906         char *term;
1907         const char *line;
1908         long num;
1909         int reg_error;
1910         regex_t regexp;
1911         regmatch_t match[1];
1912
1913         /* Allow "-L <something>,+20" to mean starting at <something>
1914          * for 20 lines, or "-L <something>,-5" for 5 lines ending at
1915          * <something>.
1916          */
1917         if (1 < begin && (spec[0] == '+' || spec[0] == '-')) {
1918                 num = strtol(spec + 1, &term, 10);
1919                 if (term != spec + 1) {
1920                         if (spec[0] == '-')
1921                                 num = 0 - num;
1922                         if (0 < num)
1923                                 *ret = begin + num - 2;
1924                         else if (!num)
1925                                 *ret = begin;
1926                         else
1927                                 *ret = begin + num;
1928                         return term;
1929                 }
1930                 return spec;
1931         }
1932         num = strtol(spec, &term, 10);
1933         if (term != spec) {
1934                 *ret = num;
1935                 return term;
1936         }
1937         if (spec[0] != '/')
1938                 return spec;
1939
1940         /* it could be a regexp of form /.../ */
1941         for (term = (char*) spec + 1; *term && *term != '/'; term++) {
1942                 if (*term == '\\')
1943                         term++;
1944         }
1945         if (*term != '/')
1946                 return spec;
1947
1948         /* try [spec+1 .. term-1] as regexp */
1949         *term = 0;
1950         begin--; /* input is in human terms */
1951         line = nth_line(sb, begin);
1952
1953         if (!(reg_error = regcomp(&regexp, spec + 1, REG_NEWLINE)) &&
1954             !(reg_error = regexec(&regexp, line, 1, match, 0))) {
1955                 const char *cp = line + match[0].rm_so;
1956                 const char *nline;
1957
1958                 while (begin++ < lno) {
1959                         nline = nth_line(sb, begin);
1960                         if (line <= cp && cp < nline)
1961                                 break;
1962                         line = nline;
1963                 }
1964                 *ret = begin;
1965                 regfree(&regexp);
1966                 *term++ = '/';
1967                 return term;
1968         }
1969         else {
1970                 char errbuf[1024];
1971                 regerror(reg_error, &regexp, errbuf, 1024);
1972                 die("-L parameter '%s': %s", spec + 1, errbuf);
1973         }
1974 }
1975
1976 /*
1977  * Parsing of -L option
1978  */
1979 static void prepare_blame_range(struct scoreboard *sb,
1980                                 const char *bottomtop,
1981                                 long lno,
1982                                 long *bottom, long *top)
1983 {
1984         const char *term;
1985
1986         term = parse_loc(bottomtop, sb, lno, 1, bottom);
1987         if (*term == ',') {
1988                 term = parse_loc(term + 1, sb, lno, *bottom + 1, top);
1989                 if (*term)
1990                         usage(blame_usage);
1991         }
1992         if (*term)
1993                 usage(blame_usage);
1994 }
1995
1996 static int git_blame_config(const char *var, const char *value)
1997 {
1998         if (!strcmp(var, "blame.showroot")) {
1999                 show_root = git_config_bool(var, value);
2000                 return 0;
2001         }
2002         if (!strcmp(var, "blame.blankboundary")) {
2003                 blank_boundary = git_config_bool(var, value);
2004                 return 0;
2005         }
2006         return git_default_config(var, value);
2007 }
2008
2009 static struct commit *fake_working_tree_commit(const char *path, const char *contents_from)
2010 {
2011         struct commit *commit;
2012         struct origin *origin;
2013         unsigned char head_sha1[20];
2014         struct strbuf buf;
2015         const char *ident;
2016         time_t now;
2017         int size, len;
2018         struct cache_entry *ce;
2019         unsigned mode;
2020
2021         if (get_sha1("HEAD", head_sha1))
2022                 die("No such ref: HEAD");
2023
2024         time(&now);
2025         commit = xcalloc(1, sizeof(*commit));
2026         commit->parents = xcalloc(1, sizeof(*commit->parents));
2027         commit->parents->item = lookup_commit_reference(head_sha1);
2028         commit->object.parsed = 1;
2029         commit->date = now;
2030         commit->object.type = OBJ_COMMIT;
2031
2032         origin = make_origin(commit, path);
2033
2034         strbuf_init(&buf, 0);
2035         if (!contents_from || strcmp("-", contents_from)) {
2036                 struct stat st;
2037                 const char *read_from;
2038                 unsigned long fin_size;
2039
2040                 if (contents_from) {
2041                         if (stat(contents_from, &st) < 0)
2042                                 die("Cannot stat %s", contents_from);
2043                         read_from = contents_from;
2044                 }
2045                 else {
2046                         if (lstat(path, &st) < 0)
2047                                 die("Cannot lstat %s", path);
2048                         read_from = path;
2049                 }
2050                 fin_size = xsize_t(st.st_size);
2051                 mode = canon_mode(st.st_mode);
2052                 switch (st.st_mode & S_IFMT) {
2053                 case S_IFREG:
2054                         if (strbuf_read_file(&buf, read_from, st.st_size) != st.st_size)
2055                                 die("cannot open or read %s", read_from);
2056                         break;
2057                 case S_IFLNK:
2058                         if (readlink(read_from, buf.buf, buf.alloc) != fin_size)
2059                                 die("cannot readlink %s", read_from);
2060                         buf.len = fin_size;
2061                         break;
2062                 default:
2063                         die("unsupported file type %s", read_from);
2064                 }
2065         }
2066         else {
2067                 /* Reading from stdin */
2068                 contents_from = "standard input";
2069                 mode = 0;
2070                 if (strbuf_read(&buf, 0, 0) < 0)
2071                         die("read error %s from stdin", strerror(errno));
2072         }
2073         convert_to_git(path, buf.buf, buf.len, &buf, 0);
2074         origin->file.ptr = buf.buf;
2075         origin->file.size = buf.len;
2076         pretend_sha1_file(buf.buf, buf.len, OBJ_BLOB, origin->blob_sha1);
2077         commit->util = origin;
2078
2079         /*
2080          * Read the current index, replace the path entry with
2081          * origin->blob_sha1 without mucking with its mode or type
2082          * bits; we are not going to write this index out -- we just
2083          * want to run "diff-index --cached".
2084          */
2085         discard_cache();
2086         read_cache();
2087
2088         len = strlen(path);
2089         if (!mode) {
2090                 int pos = cache_name_pos(path, len);
2091                 if (0 <= pos)
2092                         mode = active_cache[pos]->ce_mode;
2093                 else
2094                         /* Let's not bother reading from HEAD tree */
2095                         mode = S_IFREG | 0644;
2096         }
2097         size = cache_entry_size(len);
2098         ce = xcalloc(1, size);
2099         hashcpy(ce->sha1, origin->blob_sha1);
2100         memcpy(ce->name, path, len);
2101         ce->ce_flags = create_ce_flags(len, 0);
2102         ce->ce_mode = create_ce_mode(mode);
2103         add_cache_entry(ce, ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE);
2104
2105         /*
2106          * We are not going to write this out, so this does not matter
2107          * right now, but someday we might optimize diff-index --cached
2108          * with cache-tree information.
2109          */
2110         cache_tree_invalidate_path(active_cache_tree, path);
2111
2112         commit->buffer = xmalloc(400);
2113         ident = fmt_ident("Not Committed Yet", "not.committed.yet", NULL, 0);
2114         snprintf(commit->buffer, 400,
2115                 "tree 0000000000000000000000000000000000000000\n"
2116                 "parent %s\n"
2117                 "author %s\n"
2118                 "committer %s\n\n"
2119                 "Version of %s from %s\n",
2120                 sha1_to_hex(head_sha1),
2121                 ident, ident, path, contents_from ? contents_from : path);
2122         return commit;
2123 }
2124
2125 int cmd_blame(int argc, const char **argv, const char *prefix)
2126 {
2127         struct rev_info revs;
2128         const char *path;
2129         struct scoreboard sb;
2130         struct origin *o;
2131         struct blame_entry *ent;
2132         int i, seen_dashdash, unk, opt;
2133         long bottom, top, lno;
2134         int output_option = 0;
2135         int show_stats = 0;
2136         const char *revs_file = NULL;
2137         const char *final_commit_name = NULL;
2138         enum object_type type;
2139         const char *bottomtop = NULL;
2140         const char *contents_from = NULL;
2141
2142         cmd_is_annotate = !strcmp(argv[0], "annotate");
2143
2144         git_config(git_blame_config);
2145         save_commit_buffer = 0;
2146
2147         opt = 0;
2148         seen_dashdash = 0;
2149         for (unk = i = 1; i < argc; i++) {
2150                 const char *arg = argv[i];
2151                 if (*arg != '-')
2152                         break;
2153                 else if (!strcmp("-b", arg))
2154                         blank_boundary = 1;
2155                 else if (!strcmp("--root", arg))
2156                         show_root = 1;
2157                 else if (!strcmp(arg, "--show-stats"))
2158                         show_stats = 1;
2159                 else if (!strcmp("-c", arg))
2160                         output_option |= OUTPUT_ANNOTATE_COMPAT;
2161                 else if (!strcmp("-t", arg))
2162                         output_option |= OUTPUT_RAW_TIMESTAMP;
2163                 else if (!strcmp("-l", arg))
2164                         output_option |= OUTPUT_LONG_OBJECT_NAME;
2165                 else if (!strcmp("-s", arg))
2166                         output_option |= OUTPUT_NO_AUTHOR;
2167                 else if (!strcmp("-w", arg))
2168                         xdl_opts |= XDF_IGNORE_WHITESPACE;
2169                 else if (!strcmp("-S", arg) && ++i < argc)
2170                         revs_file = argv[i];
2171                 else if (!prefixcmp(arg, "-M")) {
2172                         opt |= PICKAXE_BLAME_MOVE;
2173                         blame_move_score = parse_score(arg+2);
2174                 }
2175                 else if (!prefixcmp(arg, "-C")) {
2176                         /*
2177                          * -C enables copy from removed files;
2178                          * -C -C enables copy from existing files, but only
2179                          *       when blaming a new file;
2180                          * -C -C -C enables copy from existing files for
2181                          *          everybody
2182                          */
2183                         if (opt & PICKAXE_BLAME_COPY_HARDER)
2184                                 opt |= PICKAXE_BLAME_COPY_HARDEST;
2185                         if (opt & PICKAXE_BLAME_COPY)
2186                                 opt |= PICKAXE_BLAME_COPY_HARDER;
2187                         opt |= PICKAXE_BLAME_COPY | PICKAXE_BLAME_MOVE;
2188                         blame_copy_score = parse_score(arg+2);
2189                 }
2190                 else if (!prefixcmp(arg, "-L")) {
2191                         if (!arg[2]) {
2192                                 if (++i >= argc)
2193                                         usage(blame_usage);
2194                                 arg = argv[i];
2195                         }
2196                         else
2197                                 arg += 2;
2198                         if (bottomtop)
2199                                 die("More than one '-L n,m' option given");
2200                         bottomtop = arg;
2201                 }
2202                 else if (!strcmp("--contents", arg)) {
2203                         if (++i >= argc)
2204                                 usage(blame_usage);
2205                         contents_from = argv[i];
2206                 }
2207                 else if (!strcmp("--incremental", arg))
2208                         incremental = 1;
2209                 else if (!strcmp("--score-debug", arg))
2210                         output_option |= OUTPUT_SHOW_SCORE;
2211                 else if (!strcmp("-f", arg) ||
2212                          !strcmp("--show-name", arg))
2213                         output_option |= OUTPUT_SHOW_NAME;
2214                 else if (!strcmp("-n", arg) ||
2215                          !strcmp("--show-number", arg))
2216                         output_option |= OUTPUT_SHOW_NUMBER;
2217                 else if (!strcmp("-p", arg) ||
2218                          !strcmp("--porcelain", arg))
2219                         output_option |= OUTPUT_PORCELAIN;
2220                 else if (!strcmp("--", arg)) {
2221                         seen_dashdash = 1;
2222                         i++;
2223                         break;
2224                 }
2225                 else
2226                         argv[unk++] = arg;
2227         }
2228
2229         if (!blame_move_score)
2230                 blame_move_score = BLAME_DEFAULT_MOVE_SCORE;
2231         if (!blame_copy_score)
2232                 blame_copy_score = BLAME_DEFAULT_COPY_SCORE;
2233
2234         /*
2235          * We have collected options unknown to us in argv[1..unk]
2236          * which are to be passed to revision machinery if we are
2237          * going to do the "bottom" processing.
2238          *
2239          * The remaining are:
2240          *
2241          * (1) if seen_dashdash, its either
2242          *     "-options -- <path>" or
2243          *     "-options -- <path> <rev>".
2244          *     but the latter is allowed only if there is no
2245          *     options that we passed to revision machinery.
2246          *
2247          * (2) otherwise, we may have "--" somewhere later and
2248          *     might be looking at the first one of multiple 'rev'
2249          *     parameters (e.g. " master ^next ^maint -- path").
2250          *     See if there is a dashdash first, and give the
2251          *     arguments before that to revision machinery.
2252          *     After that there must be one 'path'.
2253          *
2254          * (3) otherwise, its one of the three:
2255          *     "-options <path> <rev>"
2256          *     "-options <rev> <path>"
2257          *     "-options <path>"
2258          *     but again the first one is allowed only if
2259          *     there is no options that we passed to revision
2260          *     machinery.
2261          */
2262
2263         if (seen_dashdash) {
2264                 /* (1) */
2265                 if (argc <= i)
2266                         usage(blame_usage);
2267                 path = add_prefix(prefix, argv[i]);
2268                 if (i + 1 == argc - 1) {
2269                         if (unk != 1)
2270                                 usage(blame_usage);
2271                         argv[unk++] = argv[i + 1];
2272                 }
2273                 else if (i + 1 != argc)
2274                         /* garbage at end */
2275                         usage(blame_usage);
2276         }
2277         else {
2278                 int j;
2279                 for (j = i; !seen_dashdash && j < argc; j++)
2280                         if (!strcmp(argv[j], "--"))
2281                                 seen_dashdash = j;
2282                 if (seen_dashdash) {
2283                         /* (2) */
2284                         if (seen_dashdash + 1 != argc - 1)
2285                                 usage(blame_usage);
2286                         path = add_prefix(prefix, argv[seen_dashdash + 1]);
2287                         for (j = i; j < seen_dashdash; j++)
2288                                 argv[unk++] = argv[j];
2289                 }
2290                 else {
2291                         /* (3) */
2292                         if (argc <= i)
2293                                 usage(blame_usage);
2294                         path = add_prefix(prefix, argv[i]);
2295                         if (i + 1 == argc - 1) {
2296                                 final_commit_name = argv[i + 1];
2297
2298                                 /* if (unk == 1) we could be getting
2299                                  * old-style
2300                                  */
2301                                 if (unk == 1 && !has_path_in_work_tree(path)) {
2302                                         path = add_prefix(prefix, argv[i + 1]);
2303                                         final_commit_name = argv[i];
2304                                 }
2305                         }
2306                         else if (i != argc - 1)
2307                                 usage(blame_usage); /* garbage at end */
2308
2309                         setup_work_tree();
2310                         if (!has_path_in_work_tree(path))
2311                                 die("cannot stat path %s: %s",
2312                                     path, strerror(errno));
2313                 }
2314         }
2315
2316         if (final_commit_name)
2317                 argv[unk++] = final_commit_name;
2318
2319         /*
2320          * Now we got rev and path.  We do not want the path pruning
2321          * but we may want "bottom" processing.
2322          */
2323         argv[unk++] = "--"; /* terminate the rev name */
2324         argv[unk] = NULL;
2325
2326         init_revisions(&revs, NULL);
2327         setup_revisions(unk, argv, &revs, NULL);
2328         memset(&sb, 0, sizeof(sb));
2329
2330         /*
2331          * There must be one and only one positive commit in the
2332          * revs->pending array.
2333          */
2334         for (i = 0; i < revs.pending.nr; i++) {
2335                 struct object *obj = revs.pending.objects[i].item;
2336                 if (obj->flags & UNINTERESTING)
2337                         continue;
2338                 while (obj->type == OBJ_TAG)
2339                         obj = deref_tag(obj, NULL, 0);
2340                 if (obj->type != OBJ_COMMIT)
2341                         die("Non commit %s?",
2342                             revs.pending.objects[i].name);
2343                 if (sb.final)
2344                         die("More than one commit to dig from %s and %s?",
2345                             revs.pending.objects[i].name,
2346                             final_commit_name);
2347                 sb.final = (struct commit *) obj;
2348                 final_commit_name = revs.pending.objects[i].name;
2349         }
2350
2351         if (!sb.final) {
2352                 /*
2353                  * "--not A B -- path" without anything positive;
2354                  * do not default to HEAD, but use the working tree
2355                  * or "--contents".
2356                  */
2357                 setup_work_tree();
2358                 sb.final = fake_working_tree_commit(path, contents_from);
2359                 add_pending_object(&revs, &(sb.final->object), ":");
2360         }
2361         else if (contents_from)
2362                 die("Cannot use --contents with final commit object name");
2363
2364         /*
2365          * If we have bottom, this will mark the ancestors of the
2366          * bottom commits we would reach while traversing as
2367          * uninteresting.
2368          */
2369         if (prepare_revision_walk(&revs))
2370                 die("revision walk setup failed");
2371
2372         if (is_null_sha1(sb.final->object.sha1)) {
2373                 char *buf;
2374                 o = sb.final->util;
2375                 buf = xmalloc(o->file.size + 1);
2376                 memcpy(buf, o->file.ptr, o->file.size + 1);
2377                 sb.final_buf = buf;
2378                 sb.final_buf_size = o->file.size;
2379         }
2380         else {
2381                 o = get_origin(&sb, sb.final, path);
2382                 if (fill_blob_sha1(o))
2383                         die("no such path %s in %s", path, final_commit_name);
2384
2385                 sb.final_buf = read_sha1_file(o->blob_sha1, &type,
2386                                               &sb.final_buf_size);
2387                 if (!sb.final_buf)
2388                         die("Cannot read blob %s for path %s",
2389                             sha1_to_hex(o->blob_sha1),
2390                             path);
2391         }
2392         num_read_blob++;
2393         lno = prepare_lines(&sb);
2394
2395         bottom = top = 0;
2396         if (bottomtop)
2397                 prepare_blame_range(&sb, bottomtop, lno, &bottom, &top);
2398         if (bottom && top && top < bottom) {
2399                 long tmp;
2400                 tmp = top; top = bottom; bottom = tmp;
2401         }
2402         if (bottom < 1)
2403                 bottom = 1;
2404         if (top < 1)
2405                 top = lno;
2406         bottom--;
2407         if (lno < top)
2408                 die("file %s has only %lu lines", path, lno);
2409
2410         ent = xcalloc(1, sizeof(*ent));
2411         ent->lno = bottom;
2412         ent->num_lines = top - bottom;
2413         ent->suspect = o;
2414         ent->s_lno = bottom;
2415
2416         sb.ent = ent;
2417         sb.path = path;
2418
2419         if (revs_file && read_ancestry(revs_file))
2420                 die("reading graft file %s failed: %s",
2421                     revs_file, strerror(errno));
2422
2423         read_mailmap(&mailmap, ".mailmap", NULL);
2424
2425         if (!incremental)
2426                 setup_pager();
2427
2428         assign_blame(&sb, &revs, opt);
2429
2430         if (incremental)
2431                 return 0;
2432
2433         coalesce(&sb);
2434
2435         if (!(output_option & OUTPUT_PORCELAIN))
2436                 find_alignment(&sb, &output_option);
2437
2438         output(&sb, output_option);
2439         free((void *)sb.final_buf);
2440         for (ent = sb.ent; ent; ) {
2441                 struct blame_entry *e = ent->next;
2442                 free(ent);
2443                 ent = e;
2444         }
2445
2446         if (show_stats) {
2447                 printf("num read blob: %d\n", num_read_blob);
2448                 printf("num get patch: %d\n", num_get_patch);
2449                 printf("num commits: %d\n", num_commits);
2450         }
2451         return 0;
2452 }