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