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