Merge branch 'rr/precompose-utf8-cleanup' into maint
[git] / builtin / blame.c
1 /*
2  * Blame
3  *
4  * Copyright (c) 2006, Junio C Hamano
5  */
6
7 #include "cache.h"
8 #include "builtin.h"
9 #include "blob.h"
10 #include "commit.h"
11 #include "tag.h"
12 #include "tree-walk.h"
13 #include "diff.h"
14 #include "diffcore.h"
15 #include "revision.h"
16 #include "quote.h"
17 #include "xdiff-interface.h"
18 #include "cache-tree.h"
19 #include "string-list.h"
20 #include "mailmap.h"
21 #include "parse-options.h"
22 #include "utf8.h"
23 #include "userdiff.h"
24
25 static char blame_usage[] = "git blame [options] [rev-opts] [rev] [--] file";
26
27 static const char *blame_opt_usage[] = {
28         blame_usage,
29         "",
30         "[rev-opts] are documented in git-rev-list(1)",
31         NULL
32 };
33
34 static int longest_file;
35 static int longest_author;
36 static int max_orig_digits;
37 static int max_digits;
38 static int max_score_digits;
39 static int show_root;
40 static int reverse;
41 static int blank_boundary;
42 static int incremental;
43 static int xdl_opts;
44 static int abbrev = -1;
45
46 static enum date_mode blame_date_mode = DATE_ISO8601;
47 static size_t blame_date_width;
48
49 static struct string_list mailmap;
50
51 #ifndef DEBUG
52 #define DEBUG 0
53 #endif
54
55 /* stats */
56 static int num_read_blob;
57 static int num_get_patch;
58 static int num_commits;
59
60 #define PICKAXE_BLAME_MOVE              01
61 #define PICKAXE_BLAME_COPY              02
62 #define PICKAXE_BLAME_COPY_HARDER       04
63 #define PICKAXE_BLAME_COPY_HARDEST      010
64
65 /*
66  * blame for a blame_entry with score lower than these thresholds
67  * is not passed to the parent using move/copy logic.
68  */
69 static unsigned blame_move_score;
70 static unsigned blame_copy_score;
71 #define BLAME_DEFAULT_MOVE_SCORE        20
72 #define BLAME_DEFAULT_COPY_SCORE        40
73
74 /* bits #0..7 in revision.h, #8..11 used for merge_bases() in commit.c */
75 #define METAINFO_SHOWN          (1u<<12)
76 #define MORE_THAN_ONE_PATH      (1u<<13)
77
78 /*
79  * One blob in a commit that is being suspected
80  */
81 struct origin {
82         int refcnt;
83         struct origin *previous;
84         struct commit *commit;
85         mmfile_t file;
86         unsigned char blob_sha1[20];
87         unsigned mode;
88         char path[FLEX_ARRAY];
89 };
90
91 static int diff_hunks(mmfile_t *file_a, mmfile_t *file_b, long ctxlen,
92                       xdl_emit_hunk_consume_func_t hunk_func, void *cb_data)
93 {
94         xpparam_t xpp = {0};
95         xdemitconf_t xecfg = {0};
96         xdemitcb_t ecb = {NULL};
97
98         xpp.flags = xdl_opts;
99         xecfg.ctxlen = ctxlen;
100         xecfg.hunk_func = hunk_func;
101         ecb.priv = cb_data;
102         return xdi_diff(file_a, file_b, &xpp, &xecfg, &ecb);
103 }
104
105 /*
106  * Prepare diff_filespec and convert it using diff textconv API
107  * if the textconv driver exists.
108  * Return 1 if the conversion succeeds, 0 otherwise.
109  */
110 int textconv_object(const char *path,
111                     unsigned mode,
112                     const unsigned char *sha1,
113                     int sha1_valid,
114                     char **buf,
115                     unsigned long *buf_size)
116 {
117         struct diff_filespec *df;
118         struct userdiff_driver *textconv;
119
120         df = alloc_filespec(path);
121         fill_filespec(df, sha1, sha1_valid, mode);
122         textconv = get_textconv(df);
123         if (!textconv) {
124                 free_filespec(df);
125                 return 0;
126         }
127
128         *buf_size = fill_textconv(textconv, df, buf);
129         free_filespec(df);
130         return 1;
131 }
132
133 /*
134  * Given an origin, prepare mmfile_t structure to be used by the
135  * diff machinery
136  */
137 static void fill_origin_blob(struct diff_options *opt,
138                              struct origin *o, mmfile_t *file)
139 {
140         if (!o->file.ptr) {
141                 enum object_type type;
142                 unsigned long file_size;
143
144                 num_read_blob++;
145                 if (DIFF_OPT_TST(opt, ALLOW_TEXTCONV) &&
146                     textconv_object(o->path, o->mode, o->blob_sha1, 1, &file->ptr, &file_size))
147                         ;
148                 else
149                         file->ptr = read_sha1_file(o->blob_sha1, &type, &file_size);
150                 file->size = file_size;
151
152                 if (!file->ptr)
153                         die("Cannot read blob %s for path %s",
154                             sha1_to_hex(o->blob_sha1),
155                             o->path);
156                 o->file = *file;
157         }
158         else
159                 *file = o->file;
160 }
161
162 /*
163  * Origin is refcounted and usually we keep the blob contents to be
164  * reused.
165  */
166 static inline struct origin *origin_incref(struct origin *o)
167 {
168         if (o)
169                 o->refcnt++;
170         return o;
171 }
172
173 static void origin_decref(struct origin *o)
174 {
175         if (o && --o->refcnt <= 0) {
176                 if (o->previous)
177                         origin_decref(o->previous);
178                 free(o->file.ptr);
179                 free(o);
180         }
181 }
182
183 static void drop_origin_blob(struct origin *o)
184 {
185         if (o->file.ptr) {
186                 free(o->file.ptr);
187                 o->file.ptr = NULL;
188         }
189 }
190
191 /*
192  * Each group of lines is described by a blame_entry; it can be split
193  * as we pass blame to the parents.  They form a linked list in the
194  * scoreboard structure, sorted by the target line number.
195  */
196 struct blame_entry {
197         struct blame_entry *prev;
198         struct blame_entry *next;
199
200         /* the first line of this group in the final image;
201          * internally all line numbers are 0 based.
202          */
203         int lno;
204
205         /* how many lines this group has */
206         int num_lines;
207
208         /* the commit that introduced this group into the final image */
209         struct origin *suspect;
210
211         /* true if the suspect is truly guilty; false while we have not
212          * checked if the group came from one of its parents.
213          */
214         char guilty;
215
216         /* true if the entry has been scanned for copies in the current parent
217          */
218         char scanned;
219
220         /* the line number of the first line of this group in the
221          * suspect's file; internally all line numbers are 0 based.
222          */
223         int s_lno;
224
225         /* how significant this entry is -- cached to avoid
226          * scanning the lines over and over.
227          */
228         unsigned score;
229 };
230
231 /*
232  * The current state of the blame assignment.
233  */
234 struct scoreboard {
235         /* the final commit (i.e. where we started digging from) */
236         struct commit *final;
237         struct rev_info *revs;
238         const char *path;
239
240         /*
241          * The contents in the final image.
242          * Used by many functions to obtain contents of the nth line,
243          * indexed with scoreboard.lineno[blame_entry.lno].
244          */
245         const char *final_buf;
246         unsigned long final_buf_size;
247
248         /* linked list of blames */
249         struct blame_entry *ent;
250
251         /* look-up a line in the final buffer */
252         int num_lines;
253         int *lineno;
254 };
255
256 static inline int same_suspect(struct origin *a, struct origin *b)
257 {
258         if (a == b)
259                 return 1;
260         if (a->commit != b->commit)
261                 return 0;
262         return !strcmp(a->path, b->path);
263 }
264
265 static void sanity_check_refcnt(struct scoreboard *);
266
267 /*
268  * If two blame entries that are next to each other came from
269  * contiguous lines in the same origin (i.e. <commit, path> pair),
270  * merge them together.
271  */
272 static void coalesce(struct scoreboard *sb)
273 {
274         struct blame_entry *ent, *next;
275
276         for (ent = sb->ent; ent && (next = ent->next); ent = next) {
277                 if (same_suspect(ent->suspect, next->suspect) &&
278                     ent->guilty == next->guilty &&
279                     ent->s_lno + ent->num_lines == next->s_lno) {
280                         ent->num_lines += next->num_lines;
281                         ent->next = next->next;
282                         if (ent->next)
283                                 ent->next->prev = ent;
284                         origin_decref(next->suspect);
285                         free(next);
286                         ent->score = 0;
287                         next = ent; /* again */
288                 }
289         }
290
291         if (DEBUG) /* sanity */
292                 sanity_check_refcnt(sb);
293 }
294
295 /*
296  * Given a commit and a path in it, create a new origin structure.
297  * The callers that add blame to the scoreboard should use
298  * get_origin() to obtain shared, refcounted copy instead of calling
299  * this function directly.
300  */
301 static struct origin *make_origin(struct commit *commit, const char *path)
302 {
303         struct origin *o;
304         o = xcalloc(1, sizeof(*o) + strlen(path) + 1);
305         o->commit = commit;
306         o->refcnt = 1;
307         strcpy(o->path, path);
308         return o;
309 }
310
311 /*
312  * Locate an existing origin or create a new one.
313  */
314 static struct origin *get_origin(struct scoreboard *sb,
315                                  struct commit *commit,
316                                  const char *path)
317 {
318         struct blame_entry *e;
319
320         for (e = sb->ent; e; e = e->next) {
321                 if (e->suspect->commit == commit &&
322                     !strcmp(e->suspect->path, path))
323                         return origin_incref(e->suspect);
324         }
325         return make_origin(commit, path);
326 }
327
328 /*
329  * Fill the blob_sha1 field of an origin if it hasn't, so that later
330  * call to fill_origin_blob() can use it to locate the data.  blob_sha1
331  * for an origin is also used to pass the blame for the entire file to
332  * the parent to detect the case where a child's blob is identical to
333  * that of its parent's.
334  *
335  * This also fills origin->mode for corresponding tree path.
336  */
337 static int fill_blob_sha1_and_mode(struct origin *origin)
338 {
339         if (!is_null_sha1(origin->blob_sha1))
340                 return 0;
341         if (get_tree_entry(origin->commit->object.sha1,
342                            origin->path,
343                            origin->blob_sha1, &origin->mode))
344                 goto error_out;
345         if (sha1_object_info(origin->blob_sha1, NULL) != OBJ_BLOB)
346                 goto error_out;
347         return 0;
348  error_out:
349         hashclr(origin->blob_sha1);
350         origin->mode = S_IFINVALID;
351         return -1;
352 }
353
354 /*
355  * We have an origin -- check if the same path exists in the
356  * parent and return an origin structure to represent it.
357  */
358 static struct origin *find_origin(struct scoreboard *sb,
359                                   struct commit *parent,
360                                   struct origin *origin)
361 {
362         struct origin *porigin = NULL;
363         struct diff_options diff_opts;
364         const char *paths[2];
365
366         if (parent->util) {
367                 /*
368                  * Each commit object can cache one origin in that
369                  * commit.  This is a freestanding copy of origin and
370                  * not refcounted.
371                  */
372                 struct origin *cached = parent->util;
373                 if (!strcmp(cached->path, origin->path)) {
374                         /*
375                          * The same path between origin and its parent
376                          * without renaming -- the most common case.
377                          */
378                         porigin = get_origin(sb, parent, cached->path);
379
380                         /*
381                          * If the origin was newly created (i.e. get_origin
382                          * would call make_origin if none is found in the
383                          * scoreboard), it does not know the blob_sha1/mode,
384                          * so copy it.  Otherwise porigin was in the
385                          * scoreboard and already knows blob_sha1/mode.
386                          */
387                         if (porigin->refcnt == 1) {
388                                 hashcpy(porigin->blob_sha1, cached->blob_sha1);
389                                 porigin->mode = cached->mode;
390                         }
391                         return porigin;
392                 }
393                 /* otherwise it was not very useful; free it */
394                 free(parent->util);
395                 parent->util = NULL;
396         }
397
398         /* See if the origin->path is different between parent
399          * and origin first.  Most of the time they are the
400          * same and diff-tree is fairly efficient about this.
401          */
402         diff_setup(&diff_opts);
403         DIFF_OPT_SET(&diff_opts, RECURSIVE);
404         diff_opts.detect_rename = 0;
405         diff_opts.output_format = DIFF_FORMAT_NO_OUTPUT;
406         paths[0] = origin->path;
407         paths[1] = NULL;
408
409         diff_tree_setup_paths(paths, &diff_opts);
410         if (diff_setup_done(&diff_opts) < 0)
411                 die("diff-setup");
412
413         if (is_null_sha1(origin->commit->object.sha1))
414                 do_diff_cache(parent->tree->object.sha1, &diff_opts);
415         else
416                 diff_tree_sha1(parent->tree->object.sha1,
417                                origin->commit->tree->object.sha1,
418                                "", &diff_opts);
419         diffcore_std(&diff_opts);
420
421         if (!diff_queued_diff.nr) {
422                 /* The path is the same as parent */
423                 porigin = get_origin(sb, parent, origin->path);
424                 hashcpy(porigin->blob_sha1, origin->blob_sha1);
425                 porigin->mode = origin->mode;
426         } else {
427                 /*
428                  * Since origin->path is a pathspec, if the parent
429                  * commit had it as a directory, we will see a whole
430                  * bunch of deletion of files in the directory that we
431                  * do not care about.
432                  */
433                 int i;
434                 struct diff_filepair *p = NULL;
435                 for (i = 0; i < diff_queued_diff.nr; i++) {
436                         const char *name;
437                         p = diff_queued_diff.queue[i];
438                         name = p->one->path ? p->one->path : p->two->path;
439                         if (!strcmp(name, origin->path))
440                                 break;
441                 }
442                 if (!p)
443                         die("internal error in blame::find_origin");
444                 switch (p->status) {
445                 default:
446                         die("internal error in blame::find_origin (%c)",
447                             p->status);
448                 case 'M':
449                         porigin = get_origin(sb, parent, origin->path);
450                         hashcpy(porigin->blob_sha1, p->one->sha1);
451                         porigin->mode = p->one->mode;
452                         break;
453                 case 'A':
454                 case 'T':
455                         /* Did not exist in parent, or type changed */
456                         break;
457                 }
458         }
459         diff_flush(&diff_opts);
460         diff_tree_release_paths(&diff_opts);
461         if (porigin) {
462                 /*
463                  * Create a freestanding copy that is not part of
464                  * the refcounted origin found in the scoreboard, and
465                  * cache it in the commit.
466                  */
467                 struct origin *cached;
468
469                 cached = make_origin(porigin->commit, porigin->path);
470                 hashcpy(cached->blob_sha1, porigin->blob_sha1);
471                 cached->mode = porigin->mode;
472                 parent->util = cached;
473         }
474         return porigin;
475 }
476
477 /*
478  * We have an origin -- find the path that corresponds to it in its
479  * parent and return an origin structure to represent it.
480  */
481 static struct origin *find_rename(struct scoreboard *sb,
482                                   struct commit *parent,
483                                   struct origin *origin)
484 {
485         struct origin *porigin = NULL;
486         struct diff_options diff_opts;
487         int i;
488         const char *paths[2];
489
490         diff_setup(&diff_opts);
491         DIFF_OPT_SET(&diff_opts, RECURSIVE);
492         diff_opts.detect_rename = DIFF_DETECT_RENAME;
493         diff_opts.output_format = DIFF_FORMAT_NO_OUTPUT;
494         diff_opts.single_follow = origin->path;
495         paths[0] = NULL;
496         diff_tree_setup_paths(paths, &diff_opts);
497         if (diff_setup_done(&diff_opts) < 0)
498                 die("diff-setup");
499
500         if (is_null_sha1(origin->commit->object.sha1))
501                 do_diff_cache(parent->tree->object.sha1, &diff_opts);
502         else
503                 diff_tree_sha1(parent->tree->object.sha1,
504                                origin->commit->tree->object.sha1,
505                                "", &diff_opts);
506         diffcore_std(&diff_opts);
507
508         for (i = 0; i < diff_queued_diff.nr; i++) {
509                 struct diff_filepair *p = diff_queued_diff.queue[i];
510                 if ((p->status == 'R' || p->status == 'C') &&
511                     !strcmp(p->two->path, origin->path)) {
512                         porigin = get_origin(sb, parent, p->one->path);
513                         hashcpy(porigin->blob_sha1, p->one->sha1);
514                         porigin->mode = p->one->mode;
515                         break;
516                 }
517         }
518         diff_flush(&diff_opts);
519         diff_tree_release_paths(&diff_opts);
520         return porigin;
521 }
522
523 /*
524  * Link in a new blame entry to the scoreboard.  Entries that cover the
525  * same line range have been removed from the scoreboard previously.
526  */
527 static void add_blame_entry(struct scoreboard *sb, struct blame_entry *e)
528 {
529         struct blame_entry *ent, *prev = NULL;
530
531         origin_incref(e->suspect);
532
533         for (ent = sb->ent; ent && ent->lno < e->lno; ent = ent->next)
534                 prev = ent;
535
536         /* prev, if not NULL, is the last one that is below e */
537         e->prev = prev;
538         if (prev) {
539                 e->next = prev->next;
540                 prev->next = e;
541         }
542         else {
543                 e->next = sb->ent;
544                 sb->ent = e;
545         }
546         if (e->next)
547                 e->next->prev = e;
548 }
549
550 /*
551  * src typically is on-stack; we want to copy the information in it to
552  * a malloced blame_entry that is already on the linked list of the
553  * scoreboard.  The origin of dst loses a refcnt while the origin of src
554  * gains one.
555  */
556 static void dup_entry(struct blame_entry *dst, struct blame_entry *src)
557 {
558         struct blame_entry *p, *n;
559
560         p = dst->prev;
561         n = dst->next;
562         origin_incref(src->suspect);
563         origin_decref(dst->suspect);
564         memcpy(dst, src, sizeof(*src));
565         dst->prev = p;
566         dst->next = n;
567         dst->score = 0;
568 }
569
570 static const char *nth_line(struct scoreboard *sb, int lno)
571 {
572         return sb->final_buf + sb->lineno[lno];
573 }
574
575 /*
576  * It is known that lines between tlno to same came from parent, and e
577  * has an overlap with that range.  it also is known that parent's
578  * line plno corresponds to e's line tlno.
579  *
580  *                <---- e ----->
581  *                   <------>
582  *                   <------------>
583  *             <------------>
584  *             <------------------>
585  *
586  * Split e into potentially three parts; before this chunk, the chunk
587  * to be blamed for the parent, and after that portion.
588  */
589 static void split_overlap(struct blame_entry *split,
590                           struct blame_entry *e,
591                           int tlno, int plno, int same,
592                           struct origin *parent)
593 {
594         int chunk_end_lno;
595         memset(split, 0, sizeof(struct blame_entry [3]));
596
597         if (e->s_lno < tlno) {
598                 /* there is a pre-chunk part not blamed on parent */
599                 split[0].suspect = origin_incref(e->suspect);
600                 split[0].lno = e->lno;
601                 split[0].s_lno = e->s_lno;
602                 split[0].num_lines = tlno - e->s_lno;
603                 split[1].lno = e->lno + tlno - e->s_lno;
604                 split[1].s_lno = plno;
605         }
606         else {
607                 split[1].lno = e->lno;
608                 split[1].s_lno = plno + (e->s_lno - tlno);
609         }
610
611         if (same < e->s_lno + e->num_lines) {
612                 /* there is a post-chunk part not blamed on parent */
613                 split[2].suspect = origin_incref(e->suspect);
614                 split[2].lno = e->lno + (same - e->s_lno);
615                 split[2].s_lno = e->s_lno + (same - e->s_lno);
616                 split[2].num_lines = e->s_lno + e->num_lines - same;
617                 chunk_end_lno = split[2].lno;
618         }
619         else
620                 chunk_end_lno = e->lno + e->num_lines;
621         split[1].num_lines = chunk_end_lno - split[1].lno;
622
623         /*
624          * if it turns out there is nothing to blame the parent for,
625          * forget about the splitting.  !split[1].suspect signals this.
626          */
627         if (split[1].num_lines < 1)
628                 return;
629         split[1].suspect = origin_incref(parent);
630 }
631
632 /*
633  * split_overlap() divided an existing blame e into up to three parts
634  * in split.  Adjust the linked list of blames in the scoreboard to
635  * reflect the split.
636  */
637 static void split_blame(struct scoreboard *sb,
638                         struct blame_entry *split,
639                         struct blame_entry *e)
640 {
641         struct blame_entry *new_entry;
642
643         if (split[0].suspect && split[2].suspect) {
644                 /* The first part (reuse storage for the existing entry e) */
645                 dup_entry(e, &split[0]);
646
647                 /* The last part -- me */
648                 new_entry = xmalloc(sizeof(*new_entry));
649                 memcpy(new_entry, &(split[2]), sizeof(struct blame_entry));
650                 add_blame_entry(sb, new_entry);
651
652                 /* ... and the middle part -- parent */
653                 new_entry = xmalloc(sizeof(*new_entry));
654                 memcpy(new_entry, &(split[1]), sizeof(struct blame_entry));
655                 add_blame_entry(sb, new_entry);
656         }
657         else if (!split[0].suspect && !split[2].suspect)
658                 /*
659                  * The parent covers the entire area; reuse storage for
660                  * e and replace it with the parent.
661                  */
662                 dup_entry(e, &split[1]);
663         else if (split[0].suspect) {
664                 /* me and then parent */
665                 dup_entry(e, &split[0]);
666
667                 new_entry = xmalloc(sizeof(*new_entry));
668                 memcpy(new_entry, &(split[1]), sizeof(struct blame_entry));
669                 add_blame_entry(sb, new_entry);
670         }
671         else {
672                 /* parent and then me */
673                 dup_entry(e, &split[1]);
674
675                 new_entry = xmalloc(sizeof(*new_entry));
676                 memcpy(new_entry, &(split[2]), sizeof(struct blame_entry));
677                 add_blame_entry(sb, new_entry);
678         }
679
680         if (DEBUG) { /* sanity */
681                 struct blame_entry *ent;
682                 int lno = sb->ent->lno, corrupt = 0;
683
684                 for (ent = sb->ent; ent; ent = ent->next) {
685                         if (lno != ent->lno)
686                                 corrupt = 1;
687                         if (ent->s_lno < 0)
688                                 corrupt = 1;
689                         lno += ent->num_lines;
690                 }
691                 if (corrupt) {
692                         lno = sb->ent->lno;
693                         for (ent = sb->ent; ent; ent = ent->next) {
694                                 printf("L %8d l %8d n %8d\n",
695                                        lno, ent->lno, ent->num_lines);
696                                 lno = ent->lno + ent->num_lines;
697                         }
698                         die("oops");
699                 }
700         }
701 }
702
703 /*
704  * After splitting the blame, the origins used by the
705  * on-stack blame_entry should lose one refcnt each.
706  */
707 static void decref_split(struct blame_entry *split)
708 {
709         int i;
710
711         for (i = 0; i < 3; i++)
712                 origin_decref(split[i].suspect);
713 }
714
715 /*
716  * Helper for blame_chunk().  blame_entry e is known to overlap with
717  * the patch hunk; split it and pass blame to the parent.
718  */
719 static void blame_overlap(struct scoreboard *sb, struct blame_entry *e,
720                           int tlno, int plno, int same,
721                           struct origin *parent)
722 {
723         struct blame_entry split[3];
724
725         split_overlap(split, e, tlno, plno, same, parent);
726         if (split[1].suspect)
727                 split_blame(sb, split, e);
728         decref_split(split);
729 }
730
731 /*
732  * Find the line number of the last line the target is suspected for.
733  */
734 static int find_last_in_target(struct scoreboard *sb, struct origin *target)
735 {
736         struct blame_entry *e;
737         int last_in_target = -1;
738
739         for (e = sb->ent; e; e = e->next) {
740                 if (e->guilty || !same_suspect(e->suspect, target))
741                         continue;
742                 if (last_in_target < e->s_lno + e->num_lines)
743                         last_in_target = e->s_lno + e->num_lines;
744         }
745         return last_in_target;
746 }
747
748 /*
749  * Process one hunk from the patch between the current suspect for
750  * blame_entry e and its parent.  Find and split the overlap, and
751  * pass blame to the overlapping part to the parent.
752  */
753 static void blame_chunk(struct scoreboard *sb,
754                         int tlno, int plno, int same,
755                         struct origin *target, struct origin *parent)
756 {
757         struct blame_entry *e;
758
759         for (e = sb->ent; e; e = e->next) {
760                 if (e->guilty || !same_suspect(e->suspect, target))
761                         continue;
762                 if (same <= e->s_lno)
763                         continue;
764                 if (tlno < e->s_lno + e->num_lines)
765                         blame_overlap(sb, e, tlno, plno, same, parent);
766         }
767 }
768
769 struct blame_chunk_cb_data {
770         struct scoreboard *sb;
771         struct origin *target;
772         struct origin *parent;
773         long plno;
774         long tlno;
775 };
776
777 static int blame_chunk_cb(long start_a, long count_a,
778                           long start_b, long count_b, void *data)
779 {
780         struct blame_chunk_cb_data *d = data;
781         blame_chunk(d->sb, d->tlno, d->plno, start_b, d->target, d->parent);
782         d->plno = start_a + count_a;
783         d->tlno = start_b + count_b;
784         return 0;
785 }
786
787 /*
788  * We are looking at the origin 'target' and aiming to pass blame
789  * for the lines it is suspected to its parent.  Run diff to find
790  * which lines came from parent and pass blame for them.
791  */
792 static int pass_blame_to_parent(struct scoreboard *sb,
793                                 struct origin *target,
794                                 struct origin *parent)
795 {
796         int last_in_target;
797         mmfile_t file_p, file_o;
798         struct blame_chunk_cb_data d;
799
800         memset(&d, 0, sizeof(d));
801         d.sb = sb; d.target = target; d.parent = parent;
802         last_in_target = find_last_in_target(sb, target);
803         if (last_in_target < 0)
804                 return 1; /* nothing remains for this target */
805
806         fill_origin_blob(&sb->revs->diffopt, parent, &file_p);
807         fill_origin_blob(&sb->revs->diffopt, target, &file_o);
808         num_get_patch++;
809
810         diff_hunks(&file_p, &file_o, 0, blame_chunk_cb, &d);
811         /* The rest (i.e. anything after tlno) are the same as the parent */
812         blame_chunk(sb, d.tlno, d.plno, last_in_target, target, parent);
813
814         return 0;
815 }
816
817 /*
818  * The lines in blame_entry after splitting blames many times can become
819  * very small and trivial, and at some point it becomes pointless to
820  * blame the parents.  E.g. "\t\t}\n\t}\n\n" appears everywhere in any
821  * ordinary C program, and it is not worth to say it was copied from
822  * totally unrelated file in the parent.
823  *
824  * Compute how trivial the lines in the blame_entry are.
825  */
826 static unsigned ent_score(struct scoreboard *sb, struct blame_entry *e)
827 {
828         unsigned score;
829         const char *cp, *ep;
830
831         if (e->score)
832                 return e->score;
833
834         score = 1;
835         cp = nth_line(sb, e->lno);
836         ep = nth_line(sb, e->lno + e->num_lines);
837         while (cp < ep) {
838                 unsigned ch = *((unsigned char *)cp);
839                 if (isalnum(ch))
840                         score++;
841                 cp++;
842         }
843         e->score = score;
844         return score;
845 }
846
847 /*
848  * best_so_far[] and this[] are both a split of an existing blame_entry
849  * that passes blame to the parent.  Maintain best_so_far the best split
850  * so far, by comparing this and best_so_far and copying this into
851  * bst_so_far as needed.
852  */
853 static void copy_split_if_better(struct scoreboard *sb,
854                                  struct blame_entry *best_so_far,
855                                  struct blame_entry *this)
856 {
857         int i;
858
859         if (!this[1].suspect)
860                 return;
861         if (best_so_far[1].suspect) {
862                 if (ent_score(sb, &this[1]) < ent_score(sb, &best_so_far[1]))
863                         return;
864         }
865
866         for (i = 0; i < 3; i++)
867                 origin_incref(this[i].suspect);
868         decref_split(best_so_far);
869         memcpy(best_so_far, this, sizeof(struct blame_entry [3]));
870 }
871
872 /*
873  * We are looking at a part of the final image represented by
874  * ent (tlno and same are offset by ent->s_lno).
875  * tlno is where we are looking at in the final image.
876  * up to (but not including) same match preimage.
877  * plno is where we are looking at in the preimage.
878  *
879  * <-------------- final image ---------------------->
880  *       <------ent------>
881  *         ^tlno ^same
882  *    <---------preimage----->
883  *         ^plno
884  *
885  * All line numbers are 0-based.
886  */
887 static void handle_split(struct scoreboard *sb,
888                          struct blame_entry *ent,
889                          int tlno, int plno, int same,
890                          struct origin *parent,
891                          struct blame_entry *split)
892 {
893         if (ent->num_lines <= tlno)
894                 return;
895         if (tlno < same) {
896                 struct blame_entry this[3];
897                 tlno += ent->s_lno;
898                 same += ent->s_lno;
899                 split_overlap(this, ent, tlno, plno, same, parent);
900                 copy_split_if_better(sb, split, this);
901                 decref_split(this);
902         }
903 }
904
905 struct handle_split_cb_data {
906         struct scoreboard *sb;
907         struct blame_entry *ent;
908         struct origin *parent;
909         struct blame_entry *split;
910         long plno;
911         long tlno;
912 };
913
914 static int handle_split_cb(long start_a, long count_a,
915                            long start_b, long count_b, void *data)
916 {
917         struct handle_split_cb_data *d = data;
918         handle_split(d->sb, d->ent, d->tlno, d->plno, start_b, d->parent,
919                      d->split);
920         d->plno = start_a + count_a;
921         d->tlno = start_b + count_b;
922         return 0;
923 }
924
925 /*
926  * Find the lines from parent that are the same as ent so that
927  * we can pass blames to it.  file_p has the blob contents for
928  * the parent.
929  */
930 static void find_copy_in_blob(struct scoreboard *sb,
931                               struct blame_entry *ent,
932                               struct origin *parent,
933                               struct blame_entry *split,
934                               mmfile_t *file_p)
935 {
936         const char *cp;
937         int cnt;
938         mmfile_t file_o;
939         struct handle_split_cb_data d;
940
941         memset(&d, 0, sizeof(d));
942         d.sb = sb; d.ent = ent; d.parent = parent; d.split = split;
943         /*
944          * Prepare mmfile that contains only the lines in ent.
945          */
946         cp = nth_line(sb, ent->lno);
947         file_o.ptr = (char *) cp;
948         cnt = ent->num_lines;
949
950         while (cnt && cp < sb->final_buf + sb->final_buf_size) {
951                 if (*cp++ == '\n')
952                         cnt--;
953         }
954         file_o.size = cp - file_o.ptr;
955
956         /*
957          * file_o is a part of final image we are annotating.
958          * file_p partially may match that image.
959          */
960         memset(split, 0, sizeof(struct blame_entry [3]));
961         diff_hunks(file_p, &file_o, 1, handle_split_cb, &d);
962         /* remainder, if any, all match the preimage */
963         handle_split(sb, ent, d.tlno, d.plno, ent->num_lines, parent, split);
964 }
965
966 /*
967  * See if lines currently target is suspected for can be attributed to
968  * parent.
969  */
970 static int find_move_in_parent(struct scoreboard *sb,
971                                struct origin *target,
972                                struct origin *parent)
973 {
974         int last_in_target, made_progress;
975         struct blame_entry *e, split[3];
976         mmfile_t file_p;
977
978         last_in_target = find_last_in_target(sb, target);
979         if (last_in_target < 0)
980                 return 1; /* nothing remains for this target */
981
982         fill_origin_blob(&sb->revs->diffopt, parent, &file_p);
983         if (!file_p.ptr)
984                 return 0;
985
986         made_progress = 1;
987         while (made_progress) {
988                 made_progress = 0;
989                 for (e = sb->ent; e; e = e->next) {
990                         if (e->guilty || !same_suspect(e->suspect, target) ||
991                             ent_score(sb, e) < blame_move_score)
992                                 continue;
993                         find_copy_in_blob(sb, e, parent, split, &file_p);
994                         if (split[1].suspect &&
995                             blame_move_score < ent_score(sb, &split[1])) {
996                                 split_blame(sb, split, e);
997                                 made_progress = 1;
998                         }
999                         decref_split(split);
1000                 }
1001         }
1002         return 0;
1003 }
1004
1005 struct blame_list {
1006         struct blame_entry *ent;
1007         struct blame_entry split[3];
1008 };
1009
1010 /*
1011  * Count the number of entries the target is suspected for,
1012  * and prepare a list of entry and the best split.
1013  */
1014 static struct blame_list *setup_blame_list(struct scoreboard *sb,
1015                                            struct origin *target,
1016                                            int min_score,
1017                                            int *num_ents_p)
1018 {
1019         struct blame_entry *e;
1020         int num_ents, i;
1021         struct blame_list *blame_list = NULL;
1022
1023         for (e = sb->ent, num_ents = 0; e; e = e->next)
1024                 if (!e->scanned && !e->guilty &&
1025                     same_suspect(e->suspect, target) &&
1026                     min_score < ent_score(sb, e))
1027                         num_ents++;
1028         if (num_ents) {
1029                 blame_list = xcalloc(num_ents, sizeof(struct blame_list));
1030                 for (e = sb->ent, i = 0; e; e = e->next)
1031                         if (!e->scanned && !e->guilty &&
1032                             same_suspect(e->suspect, target) &&
1033                             min_score < ent_score(sb, e))
1034                                 blame_list[i++].ent = e;
1035         }
1036         *num_ents_p = num_ents;
1037         return blame_list;
1038 }
1039
1040 /*
1041  * Reset the scanned status on all entries.
1042  */
1043 static void reset_scanned_flag(struct scoreboard *sb)
1044 {
1045         struct blame_entry *e;
1046         for (e = sb->ent; e; e = e->next)
1047                 e->scanned = 0;
1048 }
1049
1050 /*
1051  * For lines target is suspected for, see if we can find code movement
1052  * across file boundary from the parent commit.  porigin is the path
1053  * in the parent we already tried.
1054  */
1055 static int find_copy_in_parent(struct scoreboard *sb,
1056                                struct origin *target,
1057                                struct commit *parent,
1058                                struct origin *porigin,
1059                                int opt)
1060 {
1061         struct diff_options diff_opts;
1062         const char *paths[1];
1063         int i, j;
1064         int retval;
1065         struct blame_list *blame_list;
1066         int num_ents;
1067
1068         blame_list = setup_blame_list(sb, target, blame_copy_score, &num_ents);
1069         if (!blame_list)
1070                 return 1; /* nothing remains for this target */
1071
1072         diff_setup(&diff_opts);
1073         DIFF_OPT_SET(&diff_opts, RECURSIVE);
1074         diff_opts.output_format = DIFF_FORMAT_NO_OUTPUT;
1075
1076         paths[0] = NULL;
1077         diff_tree_setup_paths(paths, &diff_opts);
1078         if (diff_setup_done(&diff_opts) < 0)
1079                 die("diff-setup");
1080
1081         /* Try "find copies harder" on new path if requested;
1082          * we do not want to use diffcore_rename() actually to
1083          * match things up; find_copies_harder is set only to
1084          * force diff_tree_sha1() to feed all filepairs to diff_queue,
1085          * and this code needs to be after diff_setup_done(), which
1086          * usually makes find-copies-harder imply copy detection.
1087          */
1088         if ((opt & PICKAXE_BLAME_COPY_HARDEST)
1089             || ((opt & PICKAXE_BLAME_COPY_HARDER)
1090                 && (!porigin || strcmp(target->path, porigin->path))))
1091                 DIFF_OPT_SET(&diff_opts, FIND_COPIES_HARDER);
1092
1093         if (is_null_sha1(target->commit->object.sha1))
1094                 do_diff_cache(parent->tree->object.sha1, &diff_opts);
1095         else
1096                 diff_tree_sha1(parent->tree->object.sha1,
1097                                target->commit->tree->object.sha1,
1098                                "", &diff_opts);
1099
1100         if (!DIFF_OPT_TST(&diff_opts, FIND_COPIES_HARDER))
1101                 diffcore_std(&diff_opts);
1102
1103         retval = 0;
1104         while (1) {
1105                 int made_progress = 0;
1106
1107                 for (i = 0; i < diff_queued_diff.nr; i++) {
1108                         struct diff_filepair *p = diff_queued_diff.queue[i];
1109                         struct origin *norigin;
1110                         mmfile_t file_p;
1111                         struct blame_entry this[3];
1112
1113                         if (!DIFF_FILE_VALID(p->one))
1114                                 continue; /* does not exist in parent */
1115                         if (S_ISGITLINK(p->one->mode))
1116                                 continue; /* ignore git links */
1117                         if (porigin && !strcmp(p->one->path, porigin->path))
1118                                 /* find_move already dealt with this path */
1119                                 continue;
1120
1121                         norigin = get_origin(sb, parent, p->one->path);
1122                         hashcpy(norigin->blob_sha1, p->one->sha1);
1123                         norigin->mode = p->one->mode;
1124                         fill_origin_blob(&sb->revs->diffopt, norigin, &file_p);
1125                         if (!file_p.ptr)
1126                                 continue;
1127
1128                         for (j = 0; j < num_ents; j++) {
1129                                 find_copy_in_blob(sb, blame_list[j].ent,
1130                                                   norigin, this, &file_p);
1131                                 copy_split_if_better(sb, blame_list[j].split,
1132                                                      this);
1133                                 decref_split(this);
1134                         }
1135                         origin_decref(norigin);
1136                 }
1137
1138                 for (j = 0; j < num_ents; j++) {
1139                         struct blame_entry *split = blame_list[j].split;
1140                         if (split[1].suspect &&
1141                             blame_copy_score < ent_score(sb, &split[1])) {
1142                                 split_blame(sb, split, blame_list[j].ent);
1143                                 made_progress = 1;
1144                         }
1145                         else
1146                                 blame_list[j].ent->scanned = 1;
1147                         decref_split(split);
1148                 }
1149                 free(blame_list);
1150
1151                 if (!made_progress)
1152                         break;
1153                 blame_list = setup_blame_list(sb, target, blame_copy_score, &num_ents);
1154                 if (!blame_list) {
1155                         retval = 1;
1156                         break;
1157                 }
1158         }
1159         reset_scanned_flag(sb);
1160         diff_flush(&diff_opts);
1161         diff_tree_release_paths(&diff_opts);
1162         return retval;
1163 }
1164
1165 /*
1166  * The blobs of origin and porigin exactly match, so everything
1167  * origin is suspected for can be blamed on the parent.
1168  */
1169 static void pass_whole_blame(struct scoreboard *sb,
1170                              struct origin *origin, struct origin *porigin)
1171 {
1172         struct blame_entry *e;
1173
1174         if (!porigin->file.ptr && origin->file.ptr) {
1175                 /* Steal its file */
1176                 porigin->file = origin->file;
1177                 origin->file.ptr = NULL;
1178         }
1179         for (e = sb->ent; e; e = e->next) {
1180                 if (!same_suspect(e->suspect, origin))
1181                         continue;
1182                 origin_incref(porigin);
1183                 origin_decref(e->suspect);
1184                 e->suspect = porigin;
1185         }
1186 }
1187
1188 /*
1189  * We pass blame from the current commit to its parents.  We keep saying
1190  * "parent" (and "porigin"), but what we mean is to find scapegoat to
1191  * exonerate ourselves.
1192  */
1193 static struct commit_list *first_scapegoat(struct rev_info *revs, struct commit *commit)
1194 {
1195         if (!reverse)
1196                 return commit->parents;
1197         return lookup_decoration(&revs->children, &commit->object);
1198 }
1199
1200 static int num_scapegoats(struct rev_info *revs, struct commit *commit)
1201 {
1202         int cnt;
1203         struct commit_list *l = first_scapegoat(revs, commit);
1204         for (cnt = 0; l; l = l->next)
1205                 cnt++;
1206         return cnt;
1207 }
1208
1209 #define MAXSG 16
1210
1211 static void pass_blame(struct scoreboard *sb, struct origin *origin, int opt)
1212 {
1213         struct rev_info *revs = sb->revs;
1214         int i, pass, num_sg;
1215         struct commit *commit = origin->commit;
1216         struct commit_list *sg;
1217         struct origin *sg_buf[MAXSG];
1218         struct origin *porigin, **sg_origin = sg_buf;
1219
1220         num_sg = num_scapegoats(revs, commit);
1221         if (!num_sg)
1222                 goto finish;
1223         else if (num_sg < ARRAY_SIZE(sg_buf))
1224                 memset(sg_buf, 0, sizeof(sg_buf));
1225         else
1226                 sg_origin = xcalloc(num_sg, sizeof(*sg_origin));
1227
1228         /*
1229          * The first pass looks for unrenamed path to optimize for
1230          * common cases, then we look for renames in the second pass.
1231          */
1232         for (pass = 0; pass < 2; pass++) {
1233                 struct origin *(*find)(struct scoreboard *,
1234                                        struct commit *, struct origin *);
1235                 find = pass ? find_rename : find_origin;
1236
1237                 for (i = 0, sg = first_scapegoat(revs, commit);
1238                      i < num_sg && sg;
1239                      sg = sg->next, i++) {
1240                         struct commit *p = sg->item;
1241                         int j, same;
1242
1243                         if (sg_origin[i])
1244                                 continue;
1245                         if (parse_commit(p))
1246                                 continue;
1247                         porigin = find(sb, p, origin);
1248                         if (!porigin)
1249                                 continue;
1250                         if (!hashcmp(porigin->blob_sha1, origin->blob_sha1)) {
1251                                 pass_whole_blame(sb, origin, porigin);
1252                                 origin_decref(porigin);
1253                                 goto finish;
1254                         }
1255                         for (j = same = 0; j < i; j++)
1256                                 if (sg_origin[j] &&
1257                                     !hashcmp(sg_origin[j]->blob_sha1,
1258                                              porigin->blob_sha1)) {
1259                                         same = 1;
1260                                         break;
1261                                 }
1262                         if (!same)
1263                                 sg_origin[i] = porigin;
1264                         else
1265                                 origin_decref(porigin);
1266                 }
1267         }
1268
1269         num_commits++;
1270         for (i = 0, sg = first_scapegoat(revs, commit);
1271              i < num_sg && sg;
1272              sg = sg->next, i++) {
1273                 struct origin *porigin = sg_origin[i];
1274                 if (!porigin)
1275                         continue;
1276                 if (!origin->previous) {
1277                         origin_incref(porigin);
1278                         origin->previous = porigin;
1279                 }
1280                 if (pass_blame_to_parent(sb, origin, porigin))
1281                         goto finish;
1282         }
1283
1284         /*
1285          * Optionally find moves in parents' files.
1286          */
1287         if (opt & PICKAXE_BLAME_MOVE)
1288                 for (i = 0, sg = first_scapegoat(revs, commit);
1289                      i < num_sg && sg;
1290                      sg = sg->next, i++) {
1291                         struct origin *porigin = sg_origin[i];
1292                         if (!porigin)
1293                                 continue;
1294                         if (find_move_in_parent(sb, origin, porigin))
1295                                 goto finish;
1296                 }
1297
1298         /*
1299          * Optionally find copies from parents' files.
1300          */
1301         if (opt & PICKAXE_BLAME_COPY)
1302                 for (i = 0, sg = first_scapegoat(revs, commit);
1303                      i < num_sg && sg;
1304                      sg = sg->next, i++) {
1305                         struct origin *porigin = sg_origin[i];
1306                         if (find_copy_in_parent(sb, origin, sg->item,
1307                                                 porigin, opt))
1308                                 goto finish;
1309                 }
1310
1311  finish:
1312         for (i = 0; i < num_sg; i++) {
1313                 if (sg_origin[i]) {
1314                         drop_origin_blob(sg_origin[i]);
1315                         origin_decref(sg_origin[i]);
1316                 }
1317         }
1318         drop_origin_blob(origin);
1319         if (sg_buf != sg_origin)
1320                 free(sg_origin);
1321 }
1322
1323 /*
1324  * Information on commits, used for output.
1325  */
1326 struct commit_info {
1327         const char *author;
1328         const char *author_mail;
1329         unsigned long author_time;
1330         const char *author_tz;
1331
1332         /* filled only when asked for details */
1333         const char *committer;
1334         const char *committer_mail;
1335         unsigned long committer_time;
1336         const char *committer_tz;
1337
1338         const char *summary;
1339 };
1340
1341 /*
1342  * Parse author/committer line in the commit object buffer
1343  */
1344 static void get_ac_line(const char *inbuf, const char *what,
1345                         int person_len, char *person,
1346                         int mail_len, char *mail,
1347                         unsigned long *time, const char **tz)
1348 {
1349         int len, tzlen, maillen;
1350         char *tmp, *endp, *timepos, *mailpos;
1351
1352         tmp = strstr(inbuf, what);
1353         if (!tmp)
1354                 goto error_out;
1355         tmp += strlen(what);
1356         endp = strchr(tmp, '\n');
1357         if (!endp)
1358                 len = strlen(tmp);
1359         else
1360                 len = endp - tmp;
1361         if (person_len <= len) {
1362         error_out:
1363                 /* Ugh */
1364                 *tz = "(unknown)";
1365                 strcpy(person, *tz);
1366                 strcpy(mail, *tz);
1367                 *time = 0;
1368                 return;
1369         }
1370         memcpy(person, tmp, len);
1371
1372         tmp = person;
1373         tmp += len;
1374         *tmp = 0;
1375         while (person < tmp && *tmp != ' ')
1376                 tmp--;
1377         if (tmp <= person)
1378                 goto error_out;
1379         *tz = tmp+1;
1380         tzlen = (person+len)-(tmp+1);
1381
1382         *tmp = 0;
1383         while (person < tmp && *tmp != ' ')
1384                 tmp--;
1385         if (tmp <= person)
1386                 goto error_out;
1387         *time = strtoul(tmp, NULL, 10);
1388         timepos = tmp;
1389
1390         *tmp = 0;
1391         while (person < tmp && !(*tmp == ' ' && tmp[1] == '<'))
1392                 tmp--;
1393         if (tmp <= person)
1394                 return;
1395         mailpos = tmp + 1;
1396         *tmp = 0;
1397         maillen = timepos - tmp;
1398         memcpy(mail, mailpos, maillen);
1399
1400         if (!mailmap.nr)
1401                 return;
1402
1403         /*
1404          * mailmap expansion may make the name longer.
1405          * make room by pushing stuff down.
1406          */
1407         tmp = person + person_len - (tzlen + 1);
1408         memmove(tmp, *tz, tzlen);
1409         tmp[tzlen] = 0;
1410         *tz = tmp;
1411
1412         /*
1413          * Now, convert both name and e-mail using mailmap
1414          */
1415         if (map_user(&mailmap, mail+1, mail_len-1, person, tmp-person-1)) {
1416                 /* Add a trailing '>' to email, since map_user returns plain emails
1417                    Note: It already has '<', since we replace from mail+1 */
1418                 mailpos = memchr(mail, '\0', mail_len);
1419                 if (mailpos && mailpos-mail < mail_len - 1) {
1420                         *mailpos = '>';
1421                         *(mailpos+1) = '\0';
1422                 }
1423         }
1424 }
1425
1426 static void get_commit_info(struct commit *commit,
1427                             struct commit_info *ret,
1428                             int detailed)
1429 {
1430         int len;
1431         const char *subject;
1432         char *reencoded, *message;
1433         static char author_name[1024];
1434         static char author_mail[1024];
1435         static char committer_name[1024];
1436         static char committer_mail[1024];
1437         static char summary_buf[1024];
1438
1439         /*
1440          * We've operated without save_commit_buffer, so
1441          * we now need to populate them for output.
1442          */
1443         if (!commit->buffer) {
1444                 enum object_type type;
1445                 unsigned long size;
1446                 commit->buffer =
1447                         read_sha1_file(commit->object.sha1, &type, &size);
1448                 if (!commit->buffer)
1449                         die("Cannot read commit %s",
1450                             sha1_to_hex(commit->object.sha1));
1451         }
1452         reencoded = reencode_commit_message(commit, NULL);
1453         message   = reencoded ? reencoded : commit->buffer;
1454         ret->author = author_name;
1455         ret->author_mail = author_mail;
1456         get_ac_line(message, "\nauthor ",
1457                     sizeof(author_name), author_name,
1458                     sizeof(author_mail), author_mail,
1459                     &ret->author_time, &ret->author_tz);
1460
1461         if (!detailed) {
1462                 free(reencoded);
1463                 return;
1464         }
1465
1466         ret->committer = committer_name;
1467         ret->committer_mail = committer_mail;
1468         get_ac_line(message, "\ncommitter ",
1469                     sizeof(committer_name), committer_name,
1470                     sizeof(committer_mail), committer_mail,
1471                     &ret->committer_time, &ret->committer_tz);
1472
1473         ret->summary = summary_buf;
1474         len = find_commit_subject(message, &subject);
1475         if (len && len < sizeof(summary_buf)) {
1476                 memcpy(summary_buf, subject, len);
1477                 summary_buf[len] = 0;
1478         } else {
1479                 sprintf(summary_buf, "(%s)", sha1_to_hex(commit->object.sha1));
1480         }
1481         free(reencoded);
1482 }
1483
1484 /*
1485  * To allow LF and other nonportable characters in pathnames,
1486  * they are c-style quoted as needed.
1487  */
1488 static void write_filename_info(const char *path)
1489 {
1490         printf("filename ");
1491         write_name_quoted(path, stdout, '\n');
1492 }
1493
1494 /*
1495  * Porcelain/Incremental format wants to show a lot of details per
1496  * commit.  Instead of repeating this every line, emit it only once,
1497  * the first time each commit appears in the output (unless the
1498  * user has specifically asked for us to repeat).
1499  */
1500 static int emit_one_suspect_detail(struct origin *suspect, int repeat)
1501 {
1502         struct commit_info ci;
1503
1504         if (!repeat && (suspect->commit->object.flags & METAINFO_SHOWN))
1505                 return 0;
1506
1507         suspect->commit->object.flags |= METAINFO_SHOWN;
1508         get_commit_info(suspect->commit, &ci, 1);
1509         printf("author %s\n", ci.author);
1510         printf("author-mail %s\n", ci.author_mail);
1511         printf("author-time %lu\n", ci.author_time);
1512         printf("author-tz %s\n", ci.author_tz);
1513         printf("committer %s\n", ci.committer);
1514         printf("committer-mail %s\n", ci.committer_mail);
1515         printf("committer-time %lu\n", ci.committer_time);
1516         printf("committer-tz %s\n", ci.committer_tz);
1517         printf("summary %s\n", ci.summary);
1518         if (suspect->commit->object.flags & UNINTERESTING)
1519                 printf("boundary\n");
1520         if (suspect->previous) {
1521                 struct origin *prev = suspect->previous;
1522                 printf("previous %s ", sha1_to_hex(prev->commit->object.sha1));
1523                 write_name_quoted(prev->path, stdout, '\n');
1524         }
1525         return 1;
1526 }
1527
1528 /*
1529  * The blame_entry is found to be guilty for the range.  Mark it
1530  * as such, and show it in incremental output.
1531  */
1532 static void found_guilty_entry(struct blame_entry *ent)
1533 {
1534         if (ent->guilty)
1535                 return;
1536         ent->guilty = 1;
1537         if (incremental) {
1538                 struct origin *suspect = ent->suspect;
1539
1540                 printf("%s %d %d %d\n",
1541                        sha1_to_hex(suspect->commit->object.sha1),
1542                        ent->s_lno + 1, ent->lno + 1, ent->num_lines);
1543                 emit_one_suspect_detail(suspect, 0);
1544                 write_filename_info(suspect->path);
1545                 maybe_flush_or_die(stdout, "stdout");
1546         }
1547 }
1548
1549 /*
1550  * The main loop -- while the scoreboard has lines whose true origin
1551  * is still unknown, pick one blame_entry, and allow its current
1552  * suspect to pass blames to its parents.
1553  */
1554 static void assign_blame(struct scoreboard *sb, int opt)
1555 {
1556         struct rev_info *revs = sb->revs;
1557
1558         while (1) {
1559                 struct blame_entry *ent;
1560                 struct commit *commit;
1561                 struct origin *suspect = NULL;
1562
1563                 /* find one suspect to break down */
1564                 for (ent = sb->ent; !suspect && ent; ent = ent->next)
1565                         if (!ent->guilty)
1566                                 suspect = ent->suspect;
1567                 if (!suspect)
1568                         return; /* all done */
1569
1570                 /*
1571                  * We will use this suspect later in the loop,
1572                  * so hold onto it in the meantime.
1573                  */
1574                 origin_incref(suspect);
1575                 commit = suspect->commit;
1576                 if (!commit->object.parsed)
1577                         parse_commit(commit);
1578                 if (reverse ||
1579                     (!(commit->object.flags & UNINTERESTING) &&
1580                      !(revs->max_age != -1 && commit->date < revs->max_age)))
1581                         pass_blame(sb, suspect, opt);
1582                 else {
1583                         commit->object.flags |= UNINTERESTING;
1584                         if (commit->object.parsed)
1585                                 mark_parents_uninteresting(commit);
1586                 }
1587                 /* treat root commit as boundary */
1588                 if (!commit->parents && !show_root)
1589                         commit->object.flags |= UNINTERESTING;
1590
1591                 /* Take responsibility for the remaining entries */
1592                 for (ent = sb->ent; ent; ent = ent->next)
1593                         if (same_suspect(ent->suspect, suspect))
1594                                 found_guilty_entry(ent);
1595                 origin_decref(suspect);
1596
1597                 if (DEBUG) /* sanity */
1598                         sanity_check_refcnt(sb);
1599         }
1600 }
1601
1602 static const char *format_time(unsigned long time, const char *tz_str,
1603                                int show_raw_time)
1604 {
1605         static char time_buf[128];
1606         const char *time_str;
1607         int time_len;
1608         int tz;
1609
1610         if (show_raw_time) {
1611                 snprintf(time_buf, sizeof(time_buf), "%lu %s", time, tz_str);
1612         }
1613         else {
1614                 tz = atoi(tz_str);
1615                 time_str = show_date(time, tz, blame_date_mode);
1616                 time_len = strlen(time_str);
1617                 memcpy(time_buf, time_str, time_len);
1618                 memset(time_buf + time_len, ' ', blame_date_width - time_len);
1619         }
1620         return time_buf;
1621 }
1622
1623 #define OUTPUT_ANNOTATE_COMPAT  001
1624 #define OUTPUT_LONG_OBJECT_NAME 002
1625 #define OUTPUT_RAW_TIMESTAMP    004
1626 #define OUTPUT_PORCELAIN        010
1627 #define OUTPUT_SHOW_NAME        020
1628 #define OUTPUT_SHOW_NUMBER      040
1629 #define OUTPUT_SHOW_SCORE      0100
1630 #define OUTPUT_NO_AUTHOR       0200
1631 #define OUTPUT_SHOW_EMAIL       0400
1632 #define OUTPUT_LINE_PORCELAIN 01000
1633
1634 static void emit_porcelain_details(struct origin *suspect, int repeat)
1635 {
1636         if (emit_one_suspect_detail(suspect, repeat) ||
1637             (suspect->commit->object.flags & MORE_THAN_ONE_PATH))
1638                 write_filename_info(suspect->path);
1639 }
1640
1641 static void emit_porcelain(struct scoreboard *sb, struct blame_entry *ent,
1642                            int opt)
1643 {
1644         int repeat = opt & OUTPUT_LINE_PORCELAIN;
1645         int cnt;
1646         const char *cp;
1647         struct origin *suspect = ent->suspect;
1648         char hex[41];
1649
1650         strcpy(hex, sha1_to_hex(suspect->commit->object.sha1));
1651         printf("%s%c%d %d %d\n",
1652                hex,
1653                ent->guilty ? ' ' : '*', /* purely for debugging */
1654                ent->s_lno + 1,
1655                ent->lno + 1,
1656                ent->num_lines);
1657         emit_porcelain_details(suspect, repeat);
1658
1659         cp = nth_line(sb, ent->lno);
1660         for (cnt = 0; cnt < ent->num_lines; cnt++) {
1661                 char ch;
1662                 if (cnt) {
1663                         printf("%s %d %d\n", hex,
1664                                ent->s_lno + 1 + cnt,
1665                                ent->lno + 1 + cnt);
1666                         if (repeat)
1667                                 emit_porcelain_details(suspect, 1);
1668                 }
1669                 putchar('\t');
1670                 do {
1671                         ch = *cp++;
1672                         putchar(ch);
1673                 } while (ch != '\n' &&
1674                          cp < sb->final_buf + sb->final_buf_size);
1675         }
1676
1677         if (sb->final_buf_size && cp[-1] != '\n')
1678                 putchar('\n');
1679 }
1680
1681 static void emit_other(struct scoreboard *sb, struct blame_entry *ent, int opt)
1682 {
1683         int cnt;
1684         const char *cp;
1685         struct origin *suspect = ent->suspect;
1686         struct commit_info ci;
1687         char hex[41];
1688         int show_raw_time = !!(opt & OUTPUT_RAW_TIMESTAMP);
1689
1690         get_commit_info(suspect->commit, &ci, 1);
1691         strcpy(hex, sha1_to_hex(suspect->commit->object.sha1));
1692
1693         cp = nth_line(sb, ent->lno);
1694         for (cnt = 0; cnt < ent->num_lines; cnt++) {
1695                 char ch;
1696                 int length = (opt & OUTPUT_LONG_OBJECT_NAME) ? 40 : abbrev;
1697
1698                 if (suspect->commit->object.flags & UNINTERESTING) {
1699                         if (blank_boundary)
1700                                 memset(hex, ' ', length);
1701                         else if (!(opt & OUTPUT_ANNOTATE_COMPAT)) {
1702                                 length--;
1703                                 putchar('^');
1704                         }
1705                 }
1706
1707                 printf("%.*s", length, hex);
1708                 if (opt & OUTPUT_ANNOTATE_COMPAT) {
1709                         const char *name;
1710                         if (opt & OUTPUT_SHOW_EMAIL)
1711                                 name = ci.author_mail;
1712                         else
1713                                 name = ci.author;
1714                         printf("\t(%10s\t%10s\t%d)", name,
1715                                format_time(ci.author_time, ci.author_tz,
1716                                            show_raw_time),
1717                                ent->lno + 1 + cnt);
1718                 } else {
1719                         if (opt & OUTPUT_SHOW_SCORE)
1720                                 printf(" %*d %02d",
1721                                        max_score_digits, ent->score,
1722                                        ent->suspect->refcnt);
1723                         if (opt & OUTPUT_SHOW_NAME)
1724                                 printf(" %-*.*s", longest_file, longest_file,
1725                                        suspect->path);
1726                         if (opt & OUTPUT_SHOW_NUMBER)
1727                                 printf(" %*d", max_orig_digits,
1728                                        ent->s_lno + 1 + cnt);
1729
1730                         if (!(opt & OUTPUT_NO_AUTHOR)) {
1731                                 const char *name;
1732                                 int pad;
1733                                 if (opt & OUTPUT_SHOW_EMAIL)
1734                                         name = ci.author_mail;
1735                                 else
1736                                         name = ci.author;
1737                                 pad = longest_author - utf8_strwidth(name);
1738                                 printf(" (%s%*s %10s",
1739                                        name, pad, "",
1740                                        format_time(ci.author_time,
1741                                                    ci.author_tz,
1742                                                    show_raw_time));
1743                         }
1744                         printf(" %*d) ",
1745                                max_digits, ent->lno + 1 + cnt);
1746                 }
1747                 do {
1748                         ch = *cp++;
1749                         putchar(ch);
1750                 } while (ch != '\n' &&
1751                          cp < sb->final_buf + sb->final_buf_size);
1752         }
1753
1754         if (sb->final_buf_size && cp[-1] != '\n')
1755                 putchar('\n');
1756 }
1757
1758 static void output(struct scoreboard *sb, int option)
1759 {
1760         struct blame_entry *ent;
1761
1762         if (option & OUTPUT_PORCELAIN) {
1763                 for (ent = sb->ent; ent; ent = ent->next) {
1764                         struct blame_entry *oth;
1765                         struct origin *suspect = ent->suspect;
1766                         struct commit *commit = suspect->commit;
1767                         if (commit->object.flags & MORE_THAN_ONE_PATH)
1768                                 continue;
1769                         for (oth = ent->next; oth; oth = oth->next) {
1770                                 if ((oth->suspect->commit != commit) ||
1771                                     !strcmp(oth->suspect->path, suspect->path))
1772                                         continue;
1773                                 commit->object.flags |= MORE_THAN_ONE_PATH;
1774                                 break;
1775                         }
1776                 }
1777         }
1778
1779         for (ent = sb->ent; ent; ent = ent->next) {
1780                 if (option & OUTPUT_PORCELAIN)
1781                         emit_porcelain(sb, ent, option);
1782                 else {
1783                         emit_other(sb, ent, option);
1784                 }
1785         }
1786 }
1787
1788 /*
1789  * To allow quick access to the contents of nth line in the
1790  * final image, prepare an index in the scoreboard.
1791  */
1792 static int prepare_lines(struct scoreboard *sb)
1793 {
1794         const char *buf = sb->final_buf;
1795         unsigned long len = sb->final_buf_size;
1796         int num = 0, incomplete = 0, bol = 1;
1797
1798         if (len && buf[len-1] != '\n')
1799                 incomplete++; /* incomplete line at the end */
1800         while (len--) {
1801                 if (bol) {
1802                         sb->lineno = xrealloc(sb->lineno,
1803                                               sizeof(int *) * (num + 1));
1804                         sb->lineno[num] = buf - sb->final_buf;
1805                         bol = 0;
1806                 }
1807                 if (*buf++ == '\n') {
1808                         num++;
1809                         bol = 1;
1810                 }
1811         }
1812         sb->lineno = xrealloc(sb->lineno,
1813                               sizeof(int *) * (num + incomplete + 1));
1814         sb->lineno[num + incomplete] = buf - sb->final_buf;
1815         sb->num_lines = num + incomplete;
1816         return sb->num_lines;
1817 }
1818
1819 /*
1820  * Add phony grafts for use with -S; this is primarily to
1821  * support git's cvsserver that wants to give a linear history
1822  * to its clients.
1823  */
1824 static int read_ancestry(const char *graft_file)
1825 {
1826         FILE *fp = fopen(graft_file, "r");
1827         char buf[1024];
1828         if (!fp)
1829                 return -1;
1830         while (fgets(buf, sizeof(buf), fp)) {
1831                 /* The format is just "Commit Parent1 Parent2 ...\n" */
1832                 int len = strlen(buf);
1833                 struct commit_graft *graft = read_graft_line(buf, len);
1834                 if (graft)
1835                         register_commit_graft(graft, 0);
1836         }
1837         fclose(fp);
1838         return 0;
1839 }
1840
1841 static int update_auto_abbrev(int auto_abbrev, struct origin *suspect)
1842 {
1843         const char *uniq = find_unique_abbrev(suspect->commit->object.sha1,
1844                                               auto_abbrev);
1845         int len = strlen(uniq);
1846         if (auto_abbrev < len)
1847                 return len;
1848         return auto_abbrev;
1849 }
1850
1851 /*
1852  * How many columns do we need to show line numbers, authors,
1853  * and filenames?
1854  */
1855 static void find_alignment(struct scoreboard *sb, int *option)
1856 {
1857         int longest_src_lines = 0;
1858         int longest_dst_lines = 0;
1859         unsigned largest_score = 0;
1860         struct blame_entry *e;
1861         int compute_auto_abbrev = (abbrev < 0);
1862         int auto_abbrev = default_abbrev;
1863
1864         for (e = sb->ent; e; e = e->next) {
1865                 struct origin *suspect = e->suspect;
1866                 struct commit_info ci;
1867                 int num;
1868
1869                 if (compute_auto_abbrev)
1870                         auto_abbrev = update_auto_abbrev(auto_abbrev, suspect);
1871                 if (strcmp(suspect->path, sb->path))
1872                         *option |= OUTPUT_SHOW_NAME;
1873                 num = strlen(suspect->path);
1874                 if (longest_file < num)
1875                         longest_file = num;
1876                 if (!(suspect->commit->object.flags & METAINFO_SHOWN)) {
1877                         suspect->commit->object.flags |= METAINFO_SHOWN;
1878                         get_commit_info(suspect->commit, &ci, 1);
1879                         if (*option & OUTPUT_SHOW_EMAIL)
1880                                 num = utf8_strwidth(ci.author_mail);
1881                         else
1882                                 num = utf8_strwidth(ci.author);
1883                         if (longest_author < num)
1884                                 longest_author = num;
1885                 }
1886                 num = e->s_lno + e->num_lines;
1887                 if (longest_src_lines < num)
1888                         longest_src_lines = num;
1889                 num = e->lno + e->num_lines;
1890                 if (longest_dst_lines < num)
1891                         longest_dst_lines = num;
1892                 if (largest_score < ent_score(sb, e))
1893                         largest_score = ent_score(sb, e);
1894         }
1895         max_orig_digits = decimal_width(longest_src_lines);
1896         max_digits = decimal_width(longest_dst_lines);
1897         max_score_digits = decimal_width(largest_score);
1898
1899         if (compute_auto_abbrev)
1900                 /* one more abbrev length is needed for the boundary commit */
1901                 abbrev = auto_abbrev + 1;
1902 }
1903
1904 /*
1905  * For debugging -- origin is refcounted, and this asserts that
1906  * we do not underflow.
1907  */
1908 static void sanity_check_refcnt(struct scoreboard *sb)
1909 {
1910         int baa = 0;
1911         struct blame_entry *ent;
1912
1913         for (ent = sb->ent; ent; ent = ent->next) {
1914                 /* Nobody should have zero or negative refcnt */
1915                 if (ent->suspect->refcnt <= 0) {
1916                         fprintf(stderr, "%s in %s has negative refcnt %d\n",
1917                                 ent->suspect->path,
1918                                 sha1_to_hex(ent->suspect->commit->object.sha1),
1919                                 ent->suspect->refcnt);
1920                         baa = 1;
1921                 }
1922         }
1923         if (baa) {
1924                 int opt = 0160;
1925                 find_alignment(sb, &opt);
1926                 output(sb, opt);
1927                 die("Baa %d!", baa);
1928         }
1929 }
1930
1931 /*
1932  * Used for the command line parsing; check if the path exists
1933  * in the working tree.
1934  */
1935 static int has_string_in_work_tree(const char *path)
1936 {
1937         struct stat st;
1938         return !lstat(path, &st);
1939 }
1940
1941 static unsigned parse_score(const char *arg)
1942 {
1943         char *end;
1944         unsigned long score = strtoul(arg, &end, 10);
1945         if (*end)
1946                 return 0;
1947         return score;
1948 }
1949
1950 static const char *add_prefix(const char *prefix, const char *path)
1951 {
1952         return prefix_path(prefix, prefix ? strlen(prefix) : 0, path);
1953 }
1954
1955 /*
1956  * Parsing of (comma separated) one item in the -L option
1957  */
1958 static const char *parse_loc(const char *spec,
1959                              struct scoreboard *sb, long lno,
1960                              long begin, long *ret)
1961 {
1962         char *term;
1963         const char *line;
1964         long num;
1965         int reg_error;
1966         regex_t regexp;
1967         regmatch_t match[1];
1968
1969         /* Allow "-L <something>,+20" to mean starting at <something>
1970          * for 20 lines, or "-L <something>,-5" for 5 lines ending at
1971          * <something>.
1972          */
1973         if (1 < begin && (spec[0] == '+' || spec[0] == '-')) {
1974                 num = strtol(spec + 1, &term, 10);
1975                 if (term != spec + 1) {
1976                         if (spec[0] == '-')
1977                                 num = 0 - num;
1978                         if (0 < num)
1979                                 *ret = begin + num - 2;
1980                         else if (!num)
1981                                 *ret = begin;
1982                         else
1983                                 *ret = begin + num;
1984                         return term;
1985                 }
1986                 return spec;
1987         }
1988         num = strtol(spec, &term, 10);
1989         if (term != spec) {
1990                 *ret = num;
1991                 return term;
1992         }
1993         if (spec[0] != '/')
1994                 return spec;
1995
1996         /* it could be a regexp of form /.../ */
1997         for (term = (char *) spec + 1; *term && *term != '/'; term++) {
1998                 if (*term == '\\')
1999                         term++;
2000         }
2001         if (*term != '/')
2002                 return spec;
2003
2004         /* try [spec+1 .. term-1] as regexp */
2005         *term = 0;
2006         begin--; /* input is in human terms */
2007         line = nth_line(sb, begin);
2008
2009         if (!(reg_error = regcomp(&regexp, spec + 1, REG_NEWLINE)) &&
2010             !(reg_error = regexec(&regexp, line, 1, match, 0))) {
2011                 const char *cp = line + match[0].rm_so;
2012                 const char *nline;
2013
2014                 while (begin++ < lno) {
2015                         nline = nth_line(sb, begin);
2016                         if (line <= cp && cp < nline)
2017                                 break;
2018                         line = nline;
2019                 }
2020                 *ret = begin;
2021                 regfree(&regexp);
2022                 *term++ = '/';
2023                 return term;
2024         }
2025         else {
2026                 char errbuf[1024];
2027                 regerror(reg_error, &regexp, errbuf, 1024);
2028                 die("-L parameter '%s': %s", spec + 1, errbuf);
2029         }
2030 }
2031
2032 /*
2033  * Parsing of -L option
2034  */
2035 static void prepare_blame_range(struct scoreboard *sb,
2036                                 const char *bottomtop,
2037                                 long lno,
2038                                 long *bottom, long *top)
2039 {
2040         const char *term;
2041
2042         term = parse_loc(bottomtop, sb, lno, 1, bottom);
2043         if (*term == ',') {
2044                 term = parse_loc(term + 1, sb, lno, *bottom + 1, top);
2045                 if (*term)
2046                         usage(blame_usage);
2047         }
2048         if (*term)
2049                 usage(blame_usage);
2050 }
2051
2052 static int git_blame_config(const char *var, const char *value, void *cb)
2053 {
2054         if (!strcmp(var, "blame.showroot")) {
2055                 show_root = git_config_bool(var, value);
2056                 return 0;
2057         }
2058         if (!strcmp(var, "blame.blankboundary")) {
2059                 blank_boundary = git_config_bool(var, value);
2060                 return 0;
2061         }
2062         if (!strcmp(var, "blame.date")) {
2063                 if (!value)
2064                         return config_error_nonbool(var);
2065                 blame_date_mode = parse_date_format(value);
2066                 return 0;
2067         }
2068
2069         if (userdiff_config(var, value) < 0)
2070                 return -1;
2071
2072         return git_default_config(var, value, cb);
2073 }
2074
2075 /*
2076  * Prepare a dummy commit that represents the work tree (or staged) item.
2077  * Note that annotating work tree item never works in the reverse.
2078  */
2079 static struct commit *fake_working_tree_commit(struct diff_options *opt,
2080                                                const char *path,
2081                                                const char *contents_from)
2082 {
2083         struct commit *commit;
2084         struct origin *origin;
2085         unsigned char head_sha1[20];
2086         struct strbuf buf = STRBUF_INIT;
2087         const char *ident;
2088         time_t now;
2089         int size, len;
2090         struct cache_entry *ce;
2091         unsigned mode;
2092
2093         if (get_sha1("HEAD", head_sha1))
2094                 die("No such ref: HEAD");
2095
2096         time(&now);
2097         commit = xcalloc(1, sizeof(*commit));
2098         commit->parents = xcalloc(1, sizeof(*commit->parents));
2099         commit->parents->item = lookup_commit_reference(head_sha1);
2100         commit->object.parsed = 1;
2101         commit->date = now;
2102         commit->object.type = OBJ_COMMIT;
2103
2104         origin = make_origin(commit, path);
2105
2106         if (!contents_from || strcmp("-", contents_from)) {
2107                 struct stat st;
2108                 const char *read_from;
2109                 char *buf_ptr;
2110                 unsigned long buf_len;
2111
2112                 if (contents_from) {
2113                         if (stat(contents_from, &st) < 0)
2114                                 die_errno("Cannot stat '%s'", contents_from);
2115                         read_from = contents_from;
2116                 }
2117                 else {
2118                         if (lstat(path, &st) < 0)
2119                                 die_errno("Cannot lstat '%s'", path);
2120                         read_from = path;
2121                 }
2122                 mode = canon_mode(st.st_mode);
2123
2124                 switch (st.st_mode & S_IFMT) {
2125                 case S_IFREG:
2126                         if (DIFF_OPT_TST(opt, ALLOW_TEXTCONV) &&
2127                             textconv_object(read_from, mode, null_sha1, 0, &buf_ptr, &buf_len))
2128                                 strbuf_attach(&buf, buf_ptr, buf_len, buf_len + 1);
2129                         else if (strbuf_read_file(&buf, read_from, st.st_size) != st.st_size)
2130                                 die_errno("cannot open or read '%s'", read_from);
2131                         break;
2132                 case S_IFLNK:
2133                         if (strbuf_readlink(&buf, read_from, st.st_size) < 0)
2134                                 die_errno("cannot readlink '%s'", read_from);
2135                         break;
2136                 default:
2137                         die("unsupported file type %s", read_from);
2138                 }
2139         }
2140         else {
2141                 /* Reading from stdin */
2142                 contents_from = "standard input";
2143                 mode = 0;
2144                 if (strbuf_read(&buf, 0, 0) < 0)
2145                         die_errno("failed to read from stdin");
2146         }
2147         convert_to_git(path, buf.buf, buf.len, &buf, 0);
2148         origin->file.ptr = buf.buf;
2149         origin->file.size = buf.len;
2150         pretend_sha1_file(buf.buf, buf.len, OBJ_BLOB, origin->blob_sha1);
2151         commit->util = origin;
2152
2153         /*
2154          * Read the current index, replace the path entry with
2155          * origin->blob_sha1 without mucking with its mode or type
2156          * bits; we are not going to write this index out -- we just
2157          * want to run "diff-index --cached".
2158          */
2159         discard_cache();
2160         read_cache();
2161
2162         len = strlen(path);
2163         if (!mode) {
2164                 int pos = cache_name_pos(path, len);
2165                 if (0 <= pos)
2166                         mode = active_cache[pos]->ce_mode;
2167                 else
2168                         /* Let's not bother reading from HEAD tree */
2169                         mode = S_IFREG | 0644;
2170         }
2171         size = cache_entry_size(len);
2172         ce = xcalloc(1, size);
2173         hashcpy(ce->sha1, origin->blob_sha1);
2174         memcpy(ce->name, path, len);
2175         ce->ce_flags = create_ce_flags(0);
2176         ce->ce_namelen = len;
2177         ce->ce_mode = create_ce_mode(mode);
2178         add_cache_entry(ce, ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE);
2179
2180         /*
2181          * We are not going to write this out, so this does not matter
2182          * right now, but someday we might optimize diff-index --cached
2183          * with cache-tree information.
2184          */
2185         cache_tree_invalidate_path(active_cache_tree, path);
2186
2187         commit->buffer = xmalloc(400);
2188         ident = fmt_ident("Not Committed Yet", "not.committed.yet", NULL, 0);
2189         snprintf(commit->buffer, 400,
2190                 "tree 0000000000000000000000000000000000000000\n"
2191                 "parent %s\n"
2192                 "author %s\n"
2193                 "committer %s\n\n"
2194                 "Version of %s from %s\n",
2195                 sha1_to_hex(head_sha1),
2196                 ident, ident, path, contents_from ? contents_from : path);
2197         return commit;
2198 }
2199
2200 static const char *prepare_final(struct scoreboard *sb)
2201 {
2202         int i;
2203         const char *final_commit_name = NULL;
2204         struct rev_info *revs = sb->revs;
2205
2206         /*
2207          * There must be one and only one positive commit in the
2208          * revs->pending array.
2209          */
2210         for (i = 0; i < revs->pending.nr; i++) {
2211                 struct object *obj = revs->pending.objects[i].item;
2212                 if (obj->flags & UNINTERESTING)
2213                         continue;
2214                 while (obj->type == OBJ_TAG)
2215                         obj = deref_tag(obj, NULL, 0);
2216                 if (obj->type != OBJ_COMMIT)
2217                         die("Non commit %s?", revs->pending.objects[i].name);
2218                 if (sb->final)
2219                         die("More than one commit to dig from %s and %s?",
2220                             revs->pending.objects[i].name,
2221                             final_commit_name);
2222                 sb->final = (struct commit *) obj;
2223                 final_commit_name = revs->pending.objects[i].name;
2224         }
2225         return final_commit_name;
2226 }
2227
2228 static const char *prepare_initial(struct scoreboard *sb)
2229 {
2230         int i;
2231         const char *final_commit_name = NULL;
2232         struct rev_info *revs = sb->revs;
2233
2234         /*
2235          * There must be one and only one negative commit, and it must be
2236          * the boundary.
2237          */
2238         for (i = 0; i < revs->pending.nr; i++) {
2239                 struct object *obj = revs->pending.objects[i].item;
2240                 if (!(obj->flags & UNINTERESTING))
2241                         continue;
2242                 while (obj->type == OBJ_TAG)
2243                         obj = deref_tag(obj, NULL, 0);
2244                 if (obj->type != OBJ_COMMIT)
2245                         die("Non commit %s?", revs->pending.objects[i].name);
2246                 if (sb->final)
2247                         die("More than one commit to dig down to %s and %s?",
2248                             revs->pending.objects[i].name,
2249                             final_commit_name);
2250                 sb->final = (struct commit *) obj;
2251                 final_commit_name = revs->pending.objects[i].name;
2252         }
2253         if (!final_commit_name)
2254                 die("No commit to dig down to?");
2255         return final_commit_name;
2256 }
2257
2258 static int blame_copy_callback(const struct option *option, const char *arg, int unset)
2259 {
2260         int *opt = option->value;
2261
2262         /*
2263          * -C enables copy from removed files;
2264          * -C -C enables copy from existing files, but only
2265          *       when blaming a new file;
2266          * -C -C -C enables copy from existing files for
2267          *          everybody
2268          */
2269         if (*opt & PICKAXE_BLAME_COPY_HARDER)
2270                 *opt |= PICKAXE_BLAME_COPY_HARDEST;
2271         if (*opt & PICKAXE_BLAME_COPY)
2272                 *opt |= PICKAXE_BLAME_COPY_HARDER;
2273         *opt |= PICKAXE_BLAME_COPY | PICKAXE_BLAME_MOVE;
2274
2275         if (arg)
2276                 blame_copy_score = parse_score(arg);
2277         return 0;
2278 }
2279
2280 static int blame_move_callback(const struct option *option, const char *arg, int unset)
2281 {
2282         int *opt = option->value;
2283
2284         *opt |= PICKAXE_BLAME_MOVE;
2285
2286         if (arg)
2287                 blame_move_score = parse_score(arg);
2288         return 0;
2289 }
2290
2291 static int blame_bottomtop_callback(const struct option *option, const char *arg, int unset)
2292 {
2293         const char **bottomtop = option->value;
2294         if (!arg)
2295                 return -1;
2296         if (*bottomtop)
2297                 die("More than one '-L n,m' option given");
2298         *bottomtop = arg;
2299         return 0;
2300 }
2301
2302 int cmd_blame(int argc, const char **argv, const char *prefix)
2303 {
2304         struct rev_info revs;
2305         const char *path;
2306         struct scoreboard sb;
2307         struct origin *o;
2308         struct blame_entry *ent;
2309         long dashdash_pos, bottom, top, lno;
2310         const char *final_commit_name = NULL;
2311         enum object_type type;
2312
2313         static const char *bottomtop = NULL;
2314         static int output_option = 0, opt = 0;
2315         static int show_stats = 0;
2316         static const char *revs_file = NULL;
2317         static const char *contents_from = NULL;
2318         static const struct option options[] = {
2319                 OPT_BOOLEAN(0, "incremental", &incremental, "Show blame entries as we find them, incrementally"),
2320                 OPT_BOOLEAN('b', NULL, &blank_boundary, "Show blank SHA-1 for boundary commits (Default: off)"),
2321                 OPT_BOOLEAN(0, "root", &show_root, "Do not treat root commits as boundaries (Default: off)"),
2322                 OPT_BOOLEAN(0, "show-stats", &show_stats, "Show work cost statistics"),
2323                 OPT_BIT(0, "score-debug", &output_option, "Show output score for blame entries", OUTPUT_SHOW_SCORE),
2324                 OPT_BIT('f', "show-name", &output_option, "Show original filename (Default: auto)", OUTPUT_SHOW_NAME),
2325                 OPT_BIT('n', "show-number", &output_option, "Show original linenumber (Default: off)", OUTPUT_SHOW_NUMBER),
2326                 OPT_BIT('p', "porcelain", &output_option, "Show in a format designed for machine consumption", OUTPUT_PORCELAIN),
2327                 OPT_BIT(0, "line-porcelain", &output_option, "Show porcelain format with per-line commit information", OUTPUT_PORCELAIN|OUTPUT_LINE_PORCELAIN),
2328                 OPT_BIT('c', NULL, &output_option, "Use the same output mode as git-annotate (Default: off)", OUTPUT_ANNOTATE_COMPAT),
2329                 OPT_BIT('t', NULL, &output_option, "Show raw timestamp (Default: off)", OUTPUT_RAW_TIMESTAMP),
2330                 OPT_BIT('l', NULL, &output_option, "Show long commit SHA1 (Default: off)", OUTPUT_LONG_OBJECT_NAME),
2331                 OPT_BIT('s', NULL, &output_option, "Suppress author name and timestamp (Default: off)", OUTPUT_NO_AUTHOR),
2332                 OPT_BIT('e', "show-email", &output_option, "Show author email instead of name (Default: off)", OUTPUT_SHOW_EMAIL),
2333                 OPT_BIT('w', NULL, &xdl_opts, "Ignore whitespace differences", XDF_IGNORE_WHITESPACE),
2334                 OPT_BIT(0, "minimal", &xdl_opts, "Spend extra cycles to find better match", XDF_NEED_MINIMAL),
2335                 OPT_STRING('S', NULL, &revs_file, "file", "Use revisions from <file> instead of calling git-rev-list"),
2336                 OPT_STRING(0, "contents", &contents_from, "file", "Use <file>'s contents as the final image"),
2337                 { OPTION_CALLBACK, 'C', NULL, &opt, "score", "Find line copies within and across files", PARSE_OPT_OPTARG, blame_copy_callback },
2338                 { OPTION_CALLBACK, 'M', NULL, &opt, "score", "Find line movements within and across files", PARSE_OPT_OPTARG, blame_move_callback },
2339                 OPT_CALLBACK('L', NULL, &bottomtop, "n,m", "Process only line range n,m, counting from 1", blame_bottomtop_callback),
2340                 OPT__ABBREV(&abbrev),
2341                 OPT_END()
2342         };
2343
2344         struct parse_opt_ctx_t ctx;
2345         int cmd_is_annotate = !strcmp(argv[0], "annotate");
2346
2347         git_config(git_blame_config, NULL);
2348         init_revisions(&revs, NULL);
2349         revs.date_mode = blame_date_mode;
2350         DIFF_OPT_SET(&revs.diffopt, ALLOW_TEXTCONV);
2351
2352         save_commit_buffer = 0;
2353         dashdash_pos = 0;
2354
2355         parse_options_start(&ctx, argc, argv, prefix, options,
2356                             PARSE_OPT_KEEP_DASHDASH | PARSE_OPT_KEEP_ARGV0);
2357         for (;;) {
2358                 switch (parse_options_step(&ctx, options, blame_opt_usage)) {
2359                 case PARSE_OPT_HELP:
2360                         exit(129);
2361                 case PARSE_OPT_DONE:
2362                         if (ctx.argv[0])
2363                                 dashdash_pos = ctx.cpidx;
2364                         goto parse_done;
2365                 }
2366
2367                 if (!strcmp(ctx.argv[0], "--reverse")) {
2368                         ctx.argv[0] = "--children";
2369                         reverse = 1;
2370                 }
2371                 parse_revision_opt(&revs, &ctx, options, blame_opt_usage);
2372         }
2373 parse_done:
2374         argc = parse_options_end(&ctx);
2375
2376         if (0 < abbrev)
2377                 /* one more abbrev length is needed for the boundary commit */
2378                 abbrev++;
2379
2380         if (revs_file && read_ancestry(revs_file))
2381                 die_errno("reading graft file '%s' failed", revs_file);
2382
2383         if (cmd_is_annotate) {
2384                 output_option |= OUTPUT_ANNOTATE_COMPAT;
2385                 blame_date_mode = DATE_ISO8601;
2386         } else {
2387                 blame_date_mode = revs.date_mode;
2388         }
2389
2390         /* The maximum width used to show the dates */
2391         switch (blame_date_mode) {
2392         case DATE_RFC2822:
2393                 blame_date_width = sizeof("Thu, 19 Oct 2006 16:00:04 -0700");
2394                 break;
2395         case DATE_ISO8601:
2396                 blame_date_width = sizeof("2006-10-19 16:00:04 -0700");
2397                 break;
2398         case DATE_RAW:
2399                 blame_date_width = sizeof("1161298804 -0700");
2400                 break;
2401         case DATE_SHORT:
2402                 blame_date_width = sizeof("2006-10-19");
2403                 break;
2404         case DATE_RELATIVE:
2405                 /* "normal" is used as the fallback for "relative" */
2406         case DATE_LOCAL:
2407         case DATE_NORMAL:
2408                 blame_date_width = sizeof("Thu Oct 19 16:00:04 2006 -0700");
2409                 break;
2410         }
2411         blame_date_width -= 1; /* strip the null */
2412
2413         if (DIFF_OPT_TST(&revs.diffopt, FIND_COPIES_HARDER))
2414                 opt |= (PICKAXE_BLAME_COPY | PICKAXE_BLAME_MOVE |
2415                         PICKAXE_BLAME_COPY_HARDER);
2416
2417         if (!blame_move_score)
2418                 blame_move_score = BLAME_DEFAULT_MOVE_SCORE;
2419         if (!blame_copy_score)
2420                 blame_copy_score = BLAME_DEFAULT_COPY_SCORE;
2421
2422         /*
2423          * We have collected options unknown to us in argv[1..unk]
2424          * which are to be passed to revision machinery if we are
2425          * going to do the "bottom" processing.
2426          *
2427          * The remaining are:
2428          *
2429          * (1) if dashdash_pos != 0, it is either
2430          *     "blame [revisions] -- <path>" or
2431          *     "blame -- <path> <rev>"
2432          *
2433          * (2) otherwise, it is one of the two:
2434          *     "blame [revisions] <path>"
2435          *     "blame <path> <rev>"
2436          *
2437          * Note that we must strip out <path> from the arguments: we do not
2438          * want the path pruning but we may want "bottom" processing.
2439          */
2440         if (dashdash_pos) {
2441                 switch (argc - dashdash_pos - 1) {
2442                 case 2: /* (1b) */
2443                         if (argc != 4)
2444                                 usage_with_options(blame_opt_usage, options);
2445                         /* reorder for the new way: <rev> -- <path> */
2446                         argv[1] = argv[3];
2447                         argv[3] = argv[2];
2448                         argv[2] = "--";
2449                         /* FALLTHROUGH */
2450                 case 1: /* (1a) */
2451                         path = add_prefix(prefix, argv[--argc]);
2452                         argv[argc] = NULL;
2453                         break;
2454                 default:
2455                         usage_with_options(blame_opt_usage, options);
2456                 }
2457         } else {
2458                 if (argc < 2)
2459                         usage_with_options(blame_opt_usage, options);
2460                 path = add_prefix(prefix, argv[argc - 1]);
2461                 if (argc == 3 && !has_string_in_work_tree(path)) { /* (2b) */
2462                         path = add_prefix(prefix, argv[1]);
2463                         argv[1] = argv[2];
2464                 }
2465                 argv[argc - 1] = "--";
2466
2467                 setup_work_tree();
2468                 if (!has_string_in_work_tree(path))
2469                         die_errno("cannot stat path '%s'", path);
2470         }
2471
2472         revs.disable_stdin = 1;
2473         setup_revisions(argc, argv, &revs, NULL);
2474         memset(&sb, 0, sizeof(sb));
2475
2476         sb.revs = &revs;
2477         if (!reverse)
2478                 final_commit_name = prepare_final(&sb);
2479         else if (contents_from)
2480                 die("--contents and --children do not blend well.");
2481         else
2482                 final_commit_name = prepare_initial(&sb);
2483
2484         if (!sb.final) {
2485                 /*
2486                  * "--not A B -- path" without anything positive;
2487                  * do not default to HEAD, but use the working tree
2488                  * or "--contents".
2489                  */
2490                 setup_work_tree();
2491                 sb.final = fake_working_tree_commit(&sb.revs->diffopt,
2492                                                     path, contents_from);
2493                 add_pending_object(&revs, &(sb.final->object), ":");
2494         }
2495         else if (contents_from)
2496                 die("Cannot use --contents with final commit object name");
2497
2498         /*
2499          * If we have bottom, this will mark the ancestors of the
2500          * bottom commits we would reach while traversing as
2501          * uninteresting.
2502          */
2503         if (prepare_revision_walk(&revs))
2504                 die("revision walk setup failed");
2505
2506         if (is_null_sha1(sb.final->object.sha1)) {
2507                 char *buf;
2508                 o = sb.final->util;
2509                 buf = xmalloc(o->file.size + 1);
2510                 memcpy(buf, o->file.ptr, o->file.size + 1);
2511                 sb.final_buf = buf;
2512                 sb.final_buf_size = o->file.size;
2513         }
2514         else {
2515                 o = get_origin(&sb, sb.final, path);
2516                 if (fill_blob_sha1_and_mode(o))
2517                         die("no such path %s in %s", path, final_commit_name);
2518
2519                 if (DIFF_OPT_TST(&sb.revs->diffopt, ALLOW_TEXTCONV) &&
2520                     textconv_object(path, o->mode, o->blob_sha1, 1, (char **) &sb.final_buf,
2521                                     &sb.final_buf_size))
2522                         ;
2523                 else
2524                         sb.final_buf = read_sha1_file(o->blob_sha1, &type,
2525                                                       &sb.final_buf_size);
2526
2527                 if (!sb.final_buf)
2528                         die("Cannot read blob %s for path %s",
2529                             sha1_to_hex(o->blob_sha1),
2530                             path);
2531         }
2532         num_read_blob++;
2533         lno = prepare_lines(&sb);
2534
2535         bottom = top = 0;
2536         if (bottomtop)
2537                 prepare_blame_range(&sb, bottomtop, lno, &bottom, &top);
2538         if (bottom && top && top < bottom) {
2539                 long tmp;
2540                 tmp = top; top = bottom; bottom = tmp;
2541         }
2542         if (bottom < 1)
2543                 bottom = 1;
2544         if (top < 1)
2545                 top = lno;
2546         bottom--;
2547         if (lno < top || lno < bottom)
2548                 die("file %s has only %lu lines", path, lno);
2549
2550         ent = xcalloc(1, sizeof(*ent));
2551         ent->lno = bottom;
2552         ent->num_lines = top - bottom;
2553         ent->suspect = o;
2554         ent->s_lno = bottom;
2555
2556         sb.ent = ent;
2557         sb.path = path;
2558
2559         read_mailmap(&mailmap, NULL);
2560
2561         if (!incremental)
2562                 setup_pager();
2563
2564         assign_blame(&sb, opt);
2565
2566         if (incremental)
2567                 return 0;
2568
2569         coalesce(&sb);
2570
2571         if (!(output_option & OUTPUT_PORCELAIN))
2572                 find_alignment(&sb, &output_option);
2573
2574         output(&sb, output_option);
2575         free((void *)sb.final_buf);
2576         for (ent = sb.ent; ent; ) {
2577                 struct blame_entry *e = ent->next;
2578                 free(ent);
2579                 ent = e;
2580         }
2581
2582         if (show_stats) {
2583                 printf("num read blob: %d\n", num_read_blob);
2584                 printf("num get patch: %d\n", num_get_patch);
2585                 printf("num commits: %d\n", num_commits);
2586         }
2587         return 0;
2588 }