aa30ec52692558b0d8ed3dc3b6d671185682026e
[git/git.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
45 static enum date_mode blame_date_mode = DATE_ISO8601;
46 static size_t blame_date_width;
47
48 static struct string_list mailmap;
49
50 #ifndef DEBUG
51 #define DEBUG 0
52 #endif
53
54 /* stats */
55 static int num_read_blob;
56 static int num_get_patch;
57 static int num_commits;
58
59 #define PICKAXE_BLAME_MOVE 01
60 #define PICKAXE_BLAME_COPY 02
61 #define PICKAXE_BLAME_COPY_HARDER 04
62 #define PICKAXE_BLAME_COPY_HARDEST 010
63
64 /*
65 * blame for a blame_entry with score lower than these thresholds
66 * is not passed to the parent using move/copy logic.
67 */
68 static unsigned blame_move_score;
69 static unsigned blame_copy_score;
70 #define BLAME_DEFAULT_MOVE_SCORE 20
71 #define BLAME_DEFAULT_COPY_SCORE 40
72
73 /* bits #0..7 in revision.h, #8..11 used for merge_bases() in commit.c */
74 #define METAINFO_SHOWN (1u<<12)
75 #define MORE_THAN_ONE_PATH (1u<<13)
76
77 /*
78 * One blob in a commit that is being suspected
79 */
80 struct origin {
81 int refcnt;
82 struct origin *previous;
83 struct commit *commit;
84 mmfile_t file;
85 unsigned char blob_sha1[20];
86 unsigned mode;
87 char path[FLEX_ARRAY];
88 };
89
90 /*
91 * Prepare diff_filespec and convert it using diff textconv API
92 * if the textconv driver exists.
93 * Return 1 if the conversion succeeds, 0 otherwise.
94 */
95 int textconv_object(const char *path,
96 unsigned mode,
97 const unsigned char *sha1,
98 char **buf,
99 unsigned long *buf_size)
100 {
101 struct diff_filespec *df;
102 struct userdiff_driver *textconv;
103
104 df = alloc_filespec(path);
105 fill_filespec(df, sha1, mode);
106 textconv = get_textconv(df);
107 if (!textconv) {
108 free_filespec(df);
109 return 0;
110 }
111
112 *buf_size = fill_textconv(textconv, df, buf);
113 free_filespec(df);
114 return 1;
115 }
116
117 /*
118 * Given an origin, prepare mmfile_t structure to be used by the
119 * diff machinery
120 */
121 static void fill_origin_blob(struct diff_options *opt,
122 struct origin *o, mmfile_t *file)
123 {
124 if (!o->file.ptr) {
125 enum object_type type;
126 unsigned long file_size;
127
128 num_read_blob++;
129 if (DIFF_OPT_TST(opt, ALLOW_TEXTCONV) &&
130 textconv_object(o->path, o->mode, o->blob_sha1, &file->ptr, &file_size))
131 ;
132 else
133 file->ptr = read_sha1_file(o->blob_sha1, &type, &file_size);
134 file->size = file_size;
135
136 if (!file->ptr)
137 die("Cannot read blob %s for path %s",
138 sha1_to_hex(o->blob_sha1),
139 o->path);
140 o->file = *file;
141 }
142 else
143 *file = o->file;
144 }
145
146 /*
147 * Origin is refcounted and usually we keep the blob contents to be
148 * reused.
149 */
150 static inline struct origin *origin_incref(struct origin *o)
151 {
152 if (o)
153 o->refcnt++;
154 return o;
155 }
156
157 static void origin_decref(struct origin *o)
158 {
159 if (o && --o->refcnt <= 0) {
160 if (o->previous)
161 origin_decref(o->previous);
162 free(o->file.ptr);
163 free(o);
164 }
165 }
166
167 static void drop_origin_blob(struct origin *o)
168 {
169 if (o->file.ptr) {
170 free(o->file.ptr);
171 o->file.ptr = NULL;
172 }
173 }
174
175 /*
176 * Each group of lines is described by a blame_entry; it can be split
177 * as we pass blame to the parents. They form a linked list in the
178 * scoreboard structure, sorted by the target line number.
179 */
180 struct blame_entry {
181 struct blame_entry *prev;
182 struct blame_entry *next;
183
184 /* the first line of this group in the final image;
185 * internally all line numbers are 0 based.
186 */
187 int lno;
188
189 /* how many lines this group has */
190 int num_lines;
191
192 /* the commit that introduced this group into the final image */
193 struct origin *suspect;
194
195 /* true if the suspect is truly guilty; false while we have not
196 * checked if the group came from one of its parents.
197 */
198 char guilty;
199
200 /* true if the entry has been scanned for copies in the current parent
201 */
202 char scanned;
203
204 /* the line number of the first line of this group in the
205 * suspect's file; internally all line numbers are 0 based.
206 */
207 int s_lno;
208
209 /* how significant this entry is -- cached to avoid
210 * scanning the lines over and over.
211 */
212 unsigned score;
213 };
214
215 /*
216 * The current state of the blame assignment.
217 */
218 struct scoreboard {
219 /* the final commit (i.e. where we started digging from) */
220 struct commit *final;
221 struct rev_info *revs;
222 const char *path;
223
224 /*
225 * The contents in the final image.
226 * Used by many functions to obtain contents of the nth line,
227 * indexed with scoreboard.lineno[blame_entry.lno].
228 */
229 const char *final_buf;
230 unsigned long final_buf_size;
231
232 /* linked list of blames */
233 struct blame_entry *ent;
234
235 /* look-up a line in the final buffer */
236 int num_lines;
237 int *lineno;
238 };
239
240 static inline int same_suspect(struct origin *a, struct origin *b)
241 {
242 if (a == b)
243 return 1;
244 if (a->commit != b->commit)
245 return 0;
246 return !strcmp(a->path, b->path);
247 }
248
249 static void sanity_check_refcnt(struct scoreboard *);
250
251 /*
252 * If two blame entries that are next to each other came from
253 * contiguous lines in the same origin (i.e. <commit, path> pair),
254 * merge them together.
255 */
256 static void coalesce(struct scoreboard *sb)
257 {
258 struct blame_entry *ent, *next;
259
260 for (ent = sb->ent; ent && (next = ent->next); ent = next) {
261 if (same_suspect(ent->suspect, next->suspect) &&
262 ent->guilty == next->guilty &&
263 ent->s_lno + ent->num_lines == next->s_lno) {
264 ent->num_lines += next->num_lines;
265 ent->next = next->next;
266 if (ent->next)
267 ent->next->prev = ent;
268 origin_decref(next->suspect);
269 free(next);
270 ent->score = 0;
271 next = ent; /* again */
272 }
273 }
274
275 if (DEBUG) /* sanity */
276 sanity_check_refcnt(sb);
277 }
278
279 /*
280 * Given a commit and a path in it, create a new origin structure.
281 * The callers that add blame to the scoreboard should use
282 * get_origin() to obtain shared, refcounted copy instead of calling
283 * this function directly.
284 */
285 static struct origin *make_origin(struct commit *commit, const char *path)
286 {
287 struct origin *o;
288 o = xcalloc(1, sizeof(*o) + strlen(path) + 1);
289 o->commit = commit;
290 o->refcnt = 1;
291 strcpy(o->path, path);
292 return o;
293 }
294
295 /*
296 * Locate an existing origin or create a new one.
297 */
298 static struct origin *get_origin(struct scoreboard *sb,
299 struct commit *commit,
300 const char *path)
301 {
302 struct blame_entry *e;
303
304 for (e = sb->ent; e; e = e->next) {
305 if (e->suspect->commit == commit &&
306 !strcmp(e->suspect->path, path))
307 return origin_incref(e->suspect);
308 }
309 return make_origin(commit, path);
310 }
311
312 /*
313 * Fill the blob_sha1 field of an origin if it hasn't, so that later
314 * call to fill_origin_blob() can use it to locate the data. blob_sha1
315 * for an origin is also used to pass the blame for the entire file to
316 * the parent to detect the case where a child's blob is identical to
317 * that of its parent's.
318 *
319 * This also fills origin->mode for corresponding tree path.
320 */
321 static int fill_blob_sha1_and_mode(struct origin *origin)
322 {
323 if (!is_null_sha1(origin->blob_sha1))
324 return 0;
325 if (get_tree_entry(origin->commit->object.sha1,
326 origin->path,
327 origin->blob_sha1, &origin->mode))
328 goto error_out;
329 if (sha1_object_info(origin->blob_sha1, NULL) != OBJ_BLOB)
330 goto error_out;
331 return 0;
332 error_out:
333 hashclr(origin->blob_sha1);
334 origin->mode = S_IFINVALID;
335 return -1;
336 }
337
338 /*
339 * We have an origin -- check if the same path exists in the
340 * parent and return an origin structure to represent it.
341 */
342 static struct origin *find_origin(struct scoreboard *sb,
343 struct commit *parent,
344 struct origin *origin)
345 {
346 struct origin *porigin = NULL;
347 struct diff_options diff_opts;
348 const char *paths[2];
349
350 if (parent->util) {
351 /*
352 * Each commit object can cache one origin in that
353 * commit. This is a freestanding copy of origin and
354 * not refcounted.
355 */
356 struct origin *cached = parent->util;
357 if (!strcmp(cached->path, origin->path)) {
358 /*
359 * The same path between origin and its parent
360 * without renaming -- the most common case.
361 */
362 porigin = get_origin(sb, parent, cached->path);
363
364 /*
365 * If the origin was newly created (i.e. get_origin
366 * would call make_origin if none is found in the
367 * scoreboard), it does not know the blob_sha1/mode,
368 * so copy it. Otherwise porigin was in the
369 * scoreboard and already knows blob_sha1/mode.
370 */
371 if (porigin->refcnt == 1) {
372 hashcpy(porigin->blob_sha1, cached->blob_sha1);
373 porigin->mode = cached->mode;
374 }
375 return porigin;
376 }
377 /* otherwise it was not very useful; free it */
378 free(parent->util);
379 parent->util = NULL;
380 }
381
382 /* See if the origin->path is different between parent
383 * and origin first. Most of the time they are the
384 * same and diff-tree is fairly efficient about this.
385 */
386 diff_setup(&diff_opts);
387 DIFF_OPT_SET(&diff_opts, RECURSIVE);
388 diff_opts.detect_rename = 0;
389 diff_opts.output_format = DIFF_FORMAT_NO_OUTPUT;
390 paths[0] = origin->path;
391 paths[1] = NULL;
392
393 diff_tree_setup_paths(paths, &diff_opts);
394 if (diff_setup_done(&diff_opts) < 0)
395 die("diff-setup");
396
397 if (is_null_sha1(origin->commit->object.sha1))
398 do_diff_cache(parent->tree->object.sha1, &diff_opts);
399 else
400 diff_tree_sha1(parent->tree->object.sha1,
401 origin->commit->tree->object.sha1,
402 "", &diff_opts);
403 diffcore_std(&diff_opts);
404
405 if (!diff_queued_diff.nr) {
406 /* The path is the same as parent */
407 porigin = get_origin(sb, parent, origin->path);
408 hashcpy(porigin->blob_sha1, origin->blob_sha1);
409 porigin->mode = origin->mode;
410 } else {
411 /*
412 * Since origin->path is a pathspec, if the parent
413 * commit had it as a directory, we will see a whole
414 * bunch of deletion of files in the directory that we
415 * do not care about.
416 */
417 int i;
418 struct diff_filepair *p = NULL;
419 for (i = 0; i < diff_queued_diff.nr; i++) {
420 const char *name;
421 p = diff_queued_diff.queue[i];
422 name = p->one->path ? p->one->path : p->two->path;
423 if (!strcmp(name, origin->path))
424 break;
425 }
426 if (!p)
427 die("internal error in blame::find_origin");
428 switch (p->status) {
429 default:
430 die("internal error in blame::find_origin (%c)",
431 p->status);
432 case 'M':
433 porigin = get_origin(sb, parent, origin->path);
434 hashcpy(porigin->blob_sha1, p->one->sha1);
435 porigin->mode = p->one->mode;
436 break;
437 case 'A':
438 case 'T':
439 /* Did not exist in parent, or type changed */
440 break;
441 }
442 }
443 diff_flush(&diff_opts);
444 diff_tree_release_paths(&diff_opts);
445 if (porigin) {
446 /*
447 * Create a freestanding copy that is not part of
448 * the refcounted origin found in the scoreboard, and
449 * cache it in the commit.
450 */
451 struct origin *cached;
452
453 cached = make_origin(porigin->commit, porigin->path);
454 hashcpy(cached->blob_sha1, porigin->blob_sha1);
455 cached->mode = porigin->mode;
456 parent->util = cached;
457 }
458 return porigin;
459 }
460
461 /*
462 * We have an origin -- find the path that corresponds to it in its
463 * parent and return an origin structure to represent it.
464 */
465 static struct origin *find_rename(struct scoreboard *sb,
466 struct commit *parent,
467 struct origin *origin)
468 {
469 struct origin *porigin = NULL;
470 struct diff_options diff_opts;
471 int i;
472 const char *paths[2];
473
474 diff_setup(&diff_opts);
475 DIFF_OPT_SET(&diff_opts, RECURSIVE);
476 diff_opts.detect_rename = DIFF_DETECT_RENAME;
477 diff_opts.output_format = DIFF_FORMAT_NO_OUTPUT;
478 diff_opts.single_follow = origin->path;
479 paths[0] = NULL;
480 diff_tree_setup_paths(paths, &diff_opts);
481 if (diff_setup_done(&diff_opts) < 0)
482 die("diff-setup");
483
484 if (is_null_sha1(origin->commit->object.sha1))
485 do_diff_cache(parent->tree->object.sha1, &diff_opts);
486 else
487 diff_tree_sha1(parent->tree->object.sha1,
488 origin->commit->tree->object.sha1,
489 "", &diff_opts);
490 diffcore_std(&diff_opts);
491
492 for (i = 0; i < diff_queued_diff.nr; i++) {
493 struct diff_filepair *p = diff_queued_diff.queue[i];
494 if ((p->status == 'R' || p->status == 'C') &&
495 !strcmp(p->two->path, origin->path)) {
496 porigin = get_origin(sb, parent, p->one->path);
497 hashcpy(porigin->blob_sha1, p->one->sha1);
498 porigin->mode = p->one->mode;
499 break;
500 }
501 }
502 diff_flush(&diff_opts);
503 diff_tree_release_paths(&diff_opts);
504 return porigin;
505 }
506
507 /*
508 * Link in a new blame entry to the scoreboard. Entries that cover the
509 * same line range have been removed from the scoreboard previously.
510 */
511 static void add_blame_entry(struct scoreboard *sb, struct blame_entry *e)
512 {
513 struct blame_entry *ent, *prev = NULL;
514
515 origin_incref(e->suspect);
516
517 for (ent = sb->ent; ent && ent->lno < e->lno; ent = ent->next)
518 prev = ent;
519
520 /* prev, if not NULL, is the last one that is below e */
521 e->prev = prev;
522 if (prev) {
523 e->next = prev->next;
524 prev->next = e;
525 }
526 else {
527 e->next = sb->ent;
528 sb->ent = e;
529 }
530 if (e->next)
531 e->next->prev = e;
532 }
533
534 /*
535 * src typically is on-stack; we want to copy the information in it to
536 * a malloced blame_entry that is already on the linked list of the
537 * scoreboard. The origin of dst loses a refcnt while the origin of src
538 * gains one.
539 */
540 static void dup_entry(struct blame_entry *dst, struct blame_entry *src)
541 {
542 struct blame_entry *p, *n;
543
544 p = dst->prev;
545 n = dst->next;
546 origin_incref(src->suspect);
547 origin_decref(dst->suspect);
548 memcpy(dst, src, sizeof(*src));
549 dst->prev = p;
550 dst->next = n;
551 dst->score = 0;
552 }
553
554 static const char *nth_line(struct scoreboard *sb, int lno)
555 {
556 return sb->final_buf + sb->lineno[lno];
557 }
558
559 /*
560 * It is known that lines between tlno to same came from parent, and e
561 * has an overlap with that range. it also is known that parent's
562 * line plno corresponds to e's line tlno.
563 *
564 * <---- e ----->
565 * <------>
566 * <------------>
567 * <------------>
568 * <------------------>
569 *
570 * Split e into potentially three parts; before this chunk, the chunk
571 * to be blamed for the parent, and after that portion.
572 */
573 static void split_overlap(struct blame_entry *split,
574 struct blame_entry *e,
575 int tlno, int plno, int same,
576 struct origin *parent)
577 {
578 int chunk_end_lno;
579 memset(split, 0, sizeof(struct blame_entry [3]));
580
581 if (e->s_lno < tlno) {
582 /* there is a pre-chunk part not blamed on parent */
583 split[0].suspect = origin_incref(e->suspect);
584 split[0].lno = e->lno;
585 split[0].s_lno = e->s_lno;
586 split[0].num_lines = tlno - e->s_lno;
587 split[1].lno = e->lno + tlno - e->s_lno;
588 split[1].s_lno = plno;
589 }
590 else {
591 split[1].lno = e->lno;
592 split[1].s_lno = plno + (e->s_lno - tlno);
593 }
594
595 if (same < e->s_lno + e->num_lines) {
596 /* there is a post-chunk part not blamed on parent */
597 split[2].suspect = origin_incref(e->suspect);
598 split[2].lno = e->lno + (same - e->s_lno);
599 split[2].s_lno = e->s_lno + (same - e->s_lno);
600 split[2].num_lines = e->s_lno + e->num_lines - same;
601 chunk_end_lno = split[2].lno;
602 }
603 else
604 chunk_end_lno = e->lno + e->num_lines;
605 split[1].num_lines = chunk_end_lno - split[1].lno;
606
607 /*
608 * if it turns out there is nothing to blame the parent for,
609 * forget about the splitting. !split[1].suspect signals this.
610 */
611 if (split[1].num_lines < 1)
612 return;
613 split[1].suspect = origin_incref(parent);
614 }
615
616 /*
617 * split_overlap() divided an existing blame e into up to three parts
618 * in split. Adjust the linked list of blames in the scoreboard to
619 * reflect the split.
620 */
621 static void split_blame(struct scoreboard *sb,
622 struct blame_entry *split,
623 struct blame_entry *e)
624 {
625 struct blame_entry *new_entry;
626
627 if (split[0].suspect && split[2].suspect) {
628 /* The first part (reuse storage for the existing entry e) */
629 dup_entry(e, &split[0]);
630
631 /* The last part -- me */
632 new_entry = xmalloc(sizeof(*new_entry));
633 memcpy(new_entry, &(split[2]), sizeof(struct blame_entry));
634 add_blame_entry(sb, new_entry);
635
636 /* ... and the middle part -- parent */
637 new_entry = xmalloc(sizeof(*new_entry));
638 memcpy(new_entry, &(split[1]), sizeof(struct blame_entry));
639 add_blame_entry(sb, new_entry);
640 }
641 else if (!split[0].suspect && !split[2].suspect)
642 /*
643 * The parent covers the entire area; reuse storage for
644 * e and replace it with the parent.
645 */
646 dup_entry(e, &split[1]);
647 else if (split[0].suspect) {
648 /* me and then parent */
649 dup_entry(e, &split[0]);
650
651 new_entry = xmalloc(sizeof(*new_entry));
652 memcpy(new_entry, &(split[1]), sizeof(struct blame_entry));
653 add_blame_entry(sb, new_entry);
654 }
655 else {
656 /* parent and then me */
657 dup_entry(e, &split[1]);
658
659 new_entry = xmalloc(sizeof(*new_entry));
660 memcpy(new_entry, &(split[2]), sizeof(struct blame_entry));
661 add_blame_entry(sb, new_entry);
662 }
663
664 if (DEBUG) { /* sanity */
665 struct blame_entry *ent;
666 int lno = sb->ent->lno, corrupt = 0;
667
668 for (ent = sb->ent; ent; ent = ent->next) {
669 if (lno != ent->lno)
670 corrupt = 1;
671 if (ent->s_lno < 0)
672 corrupt = 1;
673 lno += ent->num_lines;
674 }
675 if (corrupt) {
676 lno = sb->ent->lno;
677 for (ent = sb->ent; ent; ent = ent->next) {
678 printf("L %8d l %8d n %8d\n",
679 lno, ent->lno, ent->num_lines);
680 lno = ent->lno + ent->num_lines;
681 }
682 die("oops");
683 }
684 }
685 }
686
687 /*
688 * After splitting the blame, the origins used by the
689 * on-stack blame_entry should lose one refcnt each.
690 */
691 static void decref_split(struct blame_entry *split)
692 {
693 int i;
694
695 for (i = 0; i < 3; i++)
696 origin_decref(split[i].suspect);
697 }
698
699 /*
700 * Helper for blame_chunk(). blame_entry e is known to overlap with
701 * the patch hunk; split it and pass blame to the parent.
702 */
703 static void blame_overlap(struct scoreboard *sb, struct blame_entry *e,
704 int tlno, int plno, int same,
705 struct origin *parent)
706 {
707 struct blame_entry split[3];
708
709 split_overlap(split, e, tlno, plno, same, parent);
710 if (split[1].suspect)
711 split_blame(sb, split, e);
712 decref_split(split);
713 }
714
715 /*
716 * Find the line number of the last line the target is suspected for.
717 */
718 static int find_last_in_target(struct scoreboard *sb, struct origin *target)
719 {
720 struct blame_entry *e;
721 int last_in_target = -1;
722
723 for (e = sb->ent; e; e = e->next) {
724 if (e->guilty || !same_suspect(e->suspect, target))
725 continue;
726 if (last_in_target < e->s_lno + e->num_lines)
727 last_in_target = e->s_lno + e->num_lines;
728 }
729 return last_in_target;
730 }
731
732 /*
733 * Process one hunk from the patch between the current suspect for
734 * blame_entry e and its parent. Find and split the overlap, and
735 * pass blame to the overlapping part to the parent.
736 */
737 static void blame_chunk(struct scoreboard *sb,
738 int tlno, int plno, int same,
739 struct origin *target, struct origin *parent)
740 {
741 struct blame_entry *e;
742
743 for (e = sb->ent; e; e = e->next) {
744 if (e->guilty || !same_suspect(e->suspect, target))
745 continue;
746 if (same <= e->s_lno)
747 continue;
748 if (tlno < e->s_lno + e->num_lines)
749 blame_overlap(sb, e, tlno, plno, same, parent);
750 }
751 }
752
753 struct blame_chunk_cb_data {
754 struct scoreboard *sb;
755 struct origin *target;
756 struct origin *parent;
757 long plno;
758 long tlno;
759 };
760
761 static void blame_chunk_cb(void *data, long same, long p_next, long t_next)
762 {
763 struct blame_chunk_cb_data *d = data;
764 blame_chunk(d->sb, d->tlno, d->plno, same, d->target, d->parent);
765 d->plno = p_next;
766 d->tlno = t_next;
767 }
768
769 /*
770 * We are looking at the origin 'target' and aiming to pass blame
771 * for the lines it is suspected to its parent. Run diff to find
772 * which lines came from parent and pass blame for them.
773 */
774 static int pass_blame_to_parent(struct scoreboard *sb,
775 struct origin *target,
776 struct origin *parent)
777 {
778 int last_in_target;
779 mmfile_t file_p, file_o;
780 struct blame_chunk_cb_data d;
781 xpparam_t xpp;
782 xdemitconf_t xecfg;
783 memset(&d, 0, sizeof(d));
784 d.sb = sb; d.target = target; d.parent = parent;
785 last_in_target = find_last_in_target(sb, target);
786 if (last_in_target < 0)
787 return 1; /* nothing remains for this target */
788
789 fill_origin_blob(&sb->revs->diffopt, parent, &file_p);
790 fill_origin_blob(&sb->revs->diffopt, target, &file_o);
791 num_get_patch++;
792
793 memset(&xpp, 0, sizeof(xpp));
794 xpp.flags = xdl_opts;
795 memset(&xecfg, 0, sizeof(xecfg));
796 xecfg.ctxlen = 0;
797 xdi_diff_hunks(&file_p, &file_o, blame_chunk_cb, &d, &xpp, &xecfg);
798 /* The rest (i.e. anything after tlno) are the same as the parent */
799 blame_chunk(sb, d.tlno, d.plno, last_in_target, target, parent);
800
801 return 0;
802 }
803
804 /*
805 * The lines in blame_entry after splitting blames many times can become
806 * very small and trivial, and at some point it becomes pointless to
807 * blame the parents. E.g. "\t\t}\n\t}\n\n" appears everywhere in any
808 * ordinary C program, and it is not worth to say it was copied from
809 * totally unrelated file in the parent.
810 *
811 * Compute how trivial the lines in the blame_entry are.
812 */
813 static unsigned ent_score(struct scoreboard *sb, struct blame_entry *e)
814 {
815 unsigned score;
816 const char *cp, *ep;
817
818 if (e->score)
819 return e->score;
820
821 score = 1;
822 cp = nth_line(sb, e->lno);
823 ep = nth_line(sb, e->lno + e->num_lines);
824 while (cp < ep) {
825 unsigned ch = *((unsigned char *)cp);
826 if (isalnum(ch))
827 score++;
828 cp++;
829 }
830 e->score = score;
831 return score;
832 }
833
834 /*
835 * best_so_far[] and this[] are both a split of an existing blame_entry
836 * that passes blame to the parent. Maintain best_so_far the best split
837 * so far, by comparing this and best_so_far and copying this into
838 * bst_so_far as needed.
839 */
840 static void copy_split_if_better(struct scoreboard *sb,
841 struct blame_entry *best_so_far,
842 struct blame_entry *this)
843 {
844 int i;
845
846 if (!this[1].suspect)
847 return;
848 if (best_so_far[1].suspect) {
849 if (ent_score(sb, &this[1]) < ent_score(sb, &best_so_far[1]))
850 return;
851 }
852
853 for (i = 0; i < 3; i++)
854 origin_incref(this[i].suspect);
855 decref_split(best_so_far);
856 memcpy(best_so_far, this, sizeof(struct blame_entry [3]));
857 }
858
859 /*
860 * We are looking at a part of the final image represented by
861 * ent (tlno and same are offset by ent->s_lno).
862 * tlno is where we are looking at in the final image.
863 * up to (but not including) same match preimage.
864 * plno is where we are looking at in the preimage.
865 *
866 * <-------------- final image ---------------------->
867 * <------ent------>
868 * ^tlno ^same
869 * <---------preimage----->
870 * ^plno
871 *
872 * All line numbers are 0-based.
873 */
874 static void handle_split(struct scoreboard *sb,
875 struct blame_entry *ent,
876 int tlno, int plno, int same,
877 struct origin *parent,
878 struct blame_entry *split)
879 {
880 if (ent->num_lines <= tlno)
881 return;
882 if (tlno < same) {
883 struct blame_entry this[3];
884 tlno += ent->s_lno;
885 same += ent->s_lno;
886 split_overlap(this, ent, tlno, plno, same, parent);
887 copy_split_if_better(sb, split, this);
888 decref_split(this);
889 }
890 }
891
892 struct handle_split_cb_data {
893 struct scoreboard *sb;
894 struct blame_entry *ent;
895 struct origin *parent;
896 struct blame_entry *split;
897 long plno;
898 long tlno;
899 };
900
901 static void handle_split_cb(void *data, long same, long p_next, long t_next)
902 {
903 struct handle_split_cb_data *d = data;
904 handle_split(d->sb, d->ent, d->tlno, d->plno, same, d->parent, d->split);
905 d->plno = p_next;
906 d->tlno = t_next;
907 }
908
909 /*
910 * Find the lines from parent that are the same as ent so that
911 * we can pass blames to it. file_p has the blob contents for
912 * the parent.
913 */
914 static void find_copy_in_blob(struct scoreboard *sb,
915 struct blame_entry *ent,
916 struct origin *parent,
917 struct blame_entry *split,
918 mmfile_t *file_p)
919 {
920 const char *cp;
921 int cnt;
922 mmfile_t file_o;
923 struct handle_split_cb_data d;
924 xpparam_t xpp;
925 xdemitconf_t xecfg;
926 memset(&d, 0, sizeof(d));
927 d.sb = sb; d.ent = ent; d.parent = parent; d.split = split;
928 /*
929 * Prepare mmfile that contains only the lines in ent.
930 */
931 cp = nth_line(sb, ent->lno);
932 file_o.ptr = (char *) cp;
933 cnt = ent->num_lines;
934
935 while (cnt && cp < sb->final_buf + sb->final_buf_size) {
936 if (*cp++ == '\n')
937 cnt--;
938 }
939 file_o.size = cp - file_o.ptr;
940
941 /*
942 * file_o is a part of final image we are annotating.
943 * file_p partially may match that image.
944 */
945 memset(&xpp, 0, sizeof(xpp));
946 xpp.flags = xdl_opts;
947 memset(&xecfg, 0, sizeof(xecfg));
948 xecfg.ctxlen = 1;
949 memset(split, 0, sizeof(struct blame_entry [3]));
950 xdi_diff_hunks(file_p, &file_o, handle_split_cb, &d, &xpp, &xecfg);
951 /* remainder, if any, all match the preimage */
952 handle_split(sb, ent, d.tlno, d.plno, ent->num_lines, parent, split);
953 }
954
955 /*
956 * See if lines currently target is suspected for can be attributed to
957 * parent.
958 */
959 static int find_move_in_parent(struct scoreboard *sb,
960 struct origin *target,
961 struct origin *parent)
962 {
963 int last_in_target, made_progress;
964 struct blame_entry *e, split[3];
965 mmfile_t file_p;
966
967 last_in_target = find_last_in_target(sb, target);
968 if (last_in_target < 0)
969 return 1; /* nothing remains for this target */
970
971 fill_origin_blob(&sb->revs->diffopt, parent, &file_p);
972 if (!file_p.ptr)
973 return 0;
974
975 made_progress = 1;
976 while (made_progress) {
977 made_progress = 0;
978 for (e = sb->ent; e; e = e->next) {
979 if (e->guilty || !same_suspect(e->suspect, target) ||
980 ent_score(sb, e) < blame_move_score)
981 continue;
982 find_copy_in_blob(sb, e, parent, split, &file_p);
983 if (split[1].suspect &&
984 blame_move_score < ent_score(sb, &split[1])) {
985 split_blame(sb, split, e);
986 made_progress = 1;
987 }
988 decref_split(split);
989 }
990 }
991 return 0;
992 }
993
994 struct blame_list {
995 struct blame_entry *ent;
996 struct blame_entry split[3];
997 };
998
999 /*
1000 * Count the number of entries the target is suspected for,
1001 * and prepare a list of entry and the best split.
1002 */
1003 static struct blame_list *setup_blame_list(struct scoreboard *sb,
1004 struct origin *target,
1005 int min_score,
1006 int *num_ents_p)
1007 {
1008 struct blame_entry *e;
1009 int num_ents, i;
1010 struct blame_list *blame_list = NULL;
1011
1012 for (e = sb->ent, num_ents = 0; e; e = e->next)
1013 if (!e->scanned && !e->guilty &&
1014 same_suspect(e->suspect, target) &&
1015 min_score < ent_score(sb, e))
1016 num_ents++;
1017 if (num_ents) {
1018 blame_list = xcalloc(num_ents, sizeof(struct blame_list));
1019 for (e = sb->ent, i = 0; e; e = e->next)
1020 if (!e->scanned && !e->guilty &&
1021 same_suspect(e->suspect, target) &&
1022 min_score < ent_score(sb, e))
1023 blame_list[i++].ent = e;
1024 }
1025 *num_ents_p = num_ents;
1026 return blame_list;
1027 }
1028
1029 /*
1030 * Reset the scanned status on all entries.
1031 */
1032 static void reset_scanned_flag(struct scoreboard *sb)
1033 {
1034 struct blame_entry *e;
1035 for (e = sb->ent; e; e = e->next)
1036 e->scanned = 0;
1037 }
1038
1039 /*
1040 * For lines target is suspected for, see if we can find code movement
1041 * across file boundary from the parent commit. porigin is the path
1042 * in the parent we already tried.
1043 */
1044 static int find_copy_in_parent(struct scoreboard *sb,
1045 struct origin *target,
1046 struct commit *parent,
1047 struct origin *porigin,
1048 int opt)
1049 {
1050 struct diff_options diff_opts;
1051 const char *paths[1];
1052 int i, j;
1053 int retval;
1054 struct blame_list *blame_list;
1055 int num_ents;
1056
1057 blame_list = setup_blame_list(sb, target, blame_copy_score, &num_ents);
1058 if (!blame_list)
1059 return 1; /* nothing remains for this target */
1060
1061 diff_setup(&diff_opts);
1062 DIFF_OPT_SET(&diff_opts, RECURSIVE);
1063 diff_opts.output_format = DIFF_FORMAT_NO_OUTPUT;
1064
1065 paths[0] = NULL;
1066 diff_tree_setup_paths(paths, &diff_opts);
1067 if (diff_setup_done(&diff_opts) < 0)
1068 die("diff-setup");
1069
1070 /* Try "find copies harder" on new path if requested;
1071 * we do not want to use diffcore_rename() actually to
1072 * match things up; find_copies_harder is set only to
1073 * force diff_tree_sha1() to feed all filepairs to diff_queue,
1074 * and this code needs to be after diff_setup_done(), which
1075 * usually makes find-copies-harder imply copy detection.
1076 */
1077 if ((opt & PICKAXE_BLAME_COPY_HARDEST)
1078 || ((opt & PICKAXE_BLAME_COPY_HARDER)
1079 && (!porigin || strcmp(target->path, porigin->path))))
1080 DIFF_OPT_SET(&diff_opts, FIND_COPIES_HARDER);
1081
1082 if (is_null_sha1(target->commit->object.sha1))
1083 do_diff_cache(parent->tree->object.sha1, &diff_opts);
1084 else
1085 diff_tree_sha1(parent->tree->object.sha1,
1086 target->commit->tree->object.sha1,
1087 "", &diff_opts);
1088
1089 if (!DIFF_OPT_TST(&diff_opts, FIND_COPIES_HARDER))
1090 diffcore_std(&diff_opts);
1091
1092 retval = 0;
1093 while (1) {
1094 int made_progress = 0;
1095
1096 for (i = 0; i < diff_queued_diff.nr; i++) {
1097 struct diff_filepair *p = diff_queued_diff.queue[i];
1098 struct origin *norigin;
1099 mmfile_t file_p;
1100 struct blame_entry this[3];
1101
1102 if (!DIFF_FILE_VALID(p->one))
1103 continue; /* does not exist in parent */
1104 if (S_ISGITLINK(p->one->mode))
1105 continue; /* ignore git links */
1106 if (porigin && !strcmp(p->one->path, porigin->path))
1107 /* find_move already dealt with this path */
1108 continue;
1109
1110 norigin = get_origin(sb, parent, p->one->path);
1111 hashcpy(norigin->blob_sha1, p->one->sha1);
1112 norigin->mode = p->one->mode;
1113 fill_origin_blob(&sb->revs->diffopt, norigin, &file_p);
1114 if (!file_p.ptr)
1115 continue;
1116
1117 for (j = 0; j < num_ents; j++) {
1118 find_copy_in_blob(sb, blame_list[j].ent,
1119 norigin, this, &file_p);
1120 copy_split_if_better(sb, blame_list[j].split,
1121 this);
1122 decref_split(this);
1123 }
1124 origin_decref(norigin);
1125 }
1126
1127 for (j = 0; j < num_ents; j++) {
1128 struct blame_entry *split = blame_list[j].split;
1129 if (split[1].suspect &&
1130 blame_copy_score < ent_score(sb, &split[1])) {
1131 split_blame(sb, split, blame_list[j].ent);
1132 made_progress = 1;
1133 }
1134 else
1135 blame_list[j].ent->scanned = 1;
1136 decref_split(split);
1137 }
1138 free(blame_list);
1139
1140 if (!made_progress)
1141 break;
1142 blame_list = setup_blame_list(sb, target, blame_copy_score, &num_ents);
1143 if (!blame_list) {
1144 retval = 1;
1145 break;
1146 }
1147 }
1148 reset_scanned_flag(sb);
1149 diff_flush(&diff_opts);
1150 diff_tree_release_paths(&diff_opts);
1151 return retval;
1152 }
1153
1154 /*
1155 * The blobs of origin and porigin exactly match, so everything
1156 * origin is suspected for can be blamed on the parent.
1157 */
1158 static void pass_whole_blame(struct scoreboard *sb,
1159 struct origin *origin, struct origin *porigin)
1160 {
1161 struct blame_entry *e;
1162
1163 if (!porigin->file.ptr && origin->file.ptr) {
1164 /* Steal its file */
1165 porigin->file = origin->file;
1166 origin->file.ptr = NULL;
1167 }
1168 for (e = sb->ent; e; e = e->next) {
1169 if (!same_suspect(e->suspect, origin))
1170 continue;
1171 origin_incref(porigin);
1172 origin_decref(e->suspect);
1173 e->suspect = porigin;
1174 }
1175 }
1176
1177 /*
1178 * We pass blame from the current commit to its parents. We keep saying
1179 * "parent" (and "porigin"), but what we mean is to find scapegoat to
1180 * exonerate ourselves.
1181 */
1182 static struct commit_list *first_scapegoat(struct rev_info *revs, struct commit *commit)
1183 {
1184 if (!reverse)
1185 return commit->parents;
1186 return lookup_decoration(&revs->children, &commit->object);
1187 }
1188
1189 static int num_scapegoats(struct rev_info *revs, struct commit *commit)
1190 {
1191 int cnt;
1192 struct commit_list *l = first_scapegoat(revs, commit);
1193 for (cnt = 0; l; l = l->next)
1194 cnt++;
1195 return cnt;
1196 }
1197
1198 #define MAXSG 16
1199
1200 static void pass_blame(struct scoreboard *sb, struct origin *origin, int opt)
1201 {
1202 struct rev_info *revs = sb->revs;
1203 int i, pass, num_sg;
1204 struct commit *commit = origin->commit;
1205 struct commit_list *sg;
1206 struct origin *sg_buf[MAXSG];
1207 struct origin *porigin, **sg_origin = sg_buf;
1208
1209 num_sg = num_scapegoats(revs, commit);
1210 if (!num_sg)
1211 goto finish;
1212 else if (num_sg < ARRAY_SIZE(sg_buf))
1213 memset(sg_buf, 0, sizeof(sg_buf));
1214 else
1215 sg_origin = xcalloc(num_sg, sizeof(*sg_origin));
1216
1217 /*
1218 * The first pass looks for unrenamed path to optimize for
1219 * common cases, then we look for renames in the second pass.
1220 */
1221 for (pass = 0; pass < 2; pass++) {
1222 struct origin *(*find)(struct scoreboard *,
1223 struct commit *, struct origin *);
1224 find = pass ? find_rename : find_origin;
1225
1226 for (i = 0, sg = first_scapegoat(revs, commit);
1227 i < num_sg && sg;
1228 sg = sg->next, i++) {
1229 struct commit *p = sg->item;
1230 int j, same;
1231
1232 if (sg_origin[i])
1233 continue;
1234 if (parse_commit(p))
1235 continue;
1236 porigin = find(sb, p, origin);
1237 if (!porigin)
1238 continue;
1239 if (!hashcmp(porigin->blob_sha1, origin->blob_sha1)) {
1240 pass_whole_blame(sb, origin, porigin);
1241 origin_decref(porigin);
1242 goto finish;
1243 }
1244 for (j = same = 0; j < i; j++)
1245 if (sg_origin[j] &&
1246 !hashcmp(sg_origin[j]->blob_sha1,
1247 porigin->blob_sha1)) {
1248 same = 1;
1249 break;
1250 }
1251 if (!same)
1252 sg_origin[i] = porigin;
1253 else
1254 origin_decref(porigin);
1255 }
1256 }
1257
1258 num_commits++;
1259 for (i = 0, sg = first_scapegoat(revs, commit);
1260 i < num_sg && sg;
1261 sg = sg->next, i++) {
1262 struct origin *porigin = sg_origin[i];
1263 if (!porigin)
1264 continue;
1265 if (!origin->previous) {
1266 origin_incref(porigin);
1267 origin->previous = porigin;
1268 }
1269 if (pass_blame_to_parent(sb, origin, porigin))
1270 goto finish;
1271 }
1272
1273 /*
1274 * Optionally find moves in parents' files.
1275 */
1276 if (opt & PICKAXE_BLAME_MOVE)
1277 for (i = 0, sg = first_scapegoat(revs, commit);
1278 i < num_sg && sg;
1279 sg = sg->next, i++) {
1280 struct origin *porigin = sg_origin[i];
1281 if (!porigin)
1282 continue;
1283 if (find_move_in_parent(sb, origin, porigin))
1284 goto finish;
1285 }
1286
1287 /*
1288 * Optionally find copies from parents' files.
1289 */
1290 if (opt & PICKAXE_BLAME_COPY)
1291 for (i = 0, sg = first_scapegoat(revs, commit);
1292 i < num_sg && sg;
1293 sg = sg->next, i++) {
1294 struct origin *porigin = sg_origin[i];
1295 if (find_copy_in_parent(sb, origin, sg->item,
1296 porigin, opt))
1297 goto finish;
1298 }
1299
1300 finish:
1301 for (i = 0; i < num_sg; i++) {
1302 if (sg_origin[i]) {
1303 drop_origin_blob(sg_origin[i]);
1304 origin_decref(sg_origin[i]);
1305 }
1306 }
1307 drop_origin_blob(origin);
1308 if (sg_buf != sg_origin)
1309 free(sg_origin);
1310 }
1311
1312 /*
1313 * Information on commits, used for output.
1314 */
1315 struct commit_info
1316 {
1317 const char *author;
1318 const char *author_mail;
1319 unsigned long author_time;
1320 const char *author_tz;
1321
1322 /* filled only when asked for details */
1323 const char *committer;
1324 const char *committer_mail;
1325 unsigned long committer_time;
1326 const char *committer_tz;
1327
1328 const char *summary;
1329 };
1330
1331 /*
1332 * Parse author/committer line in the commit object buffer
1333 */
1334 static void get_ac_line(const char *inbuf, const char *what,
1335 int person_len, char *person,
1336 int mail_len, char *mail,
1337 unsigned long *time, const char **tz)
1338 {
1339 int len, tzlen, maillen;
1340 char *tmp, *endp, *timepos, *mailpos;
1341
1342 tmp = strstr(inbuf, what);
1343 if (!tmp)
1344 goto error_out;
1345 tmp += strlen(what);
1346 endp = strchr(tmp, '\n');
1347 if (!endp)
1348 len = strlen(tmp);
1349 else
1350 len = endp - tmp;
1351 if (person_len <= len) {
1352 error_out:
1353 /* Ugh */
1354 *tz = "(unknown)";
1355 strcpy(person, *tz);
1356 strcpy(mail, *tz);
1357 *time = 0;
1358 return;
1359 }
1360 memcpy(person, tmp, len);
1361
1362 tmp = person;
1363 tmp += len;
1364 *tmp = 0;
1365 while (person < tmp && *tmp != ' ')
1366 tmp--;
1367 if (tmp <= person)
1368 goto error_out;
1369 *tz = tmp+1;
1370 tzlen = (person+len)-(tmp+1);
1371
1372 *tmp = 0;
1373 while (person < tmp && *tmp != ' ')
1374 tmp--;
1375 if (tmp <= person)
1376 goto error_out;
1377 *time = strtoul(tmp, NULL, 10);
1378 timepos = tmp;
1379
1380 *tmp = 0;
1381 while (person < tmp && *tmp != ' ')
1382 tmp--;
1383 if (tmp <= person)
1384 return;
1385 mailpos = tmp + 1;
1386 *tmp = 0;
1387 maillen = timepos - tmp;
1388 memcpy(mail, mailpos, maillen);
1389
1390 if (!mailmap.nr)
1391 return;
1392
1393 /*
1394 * mailmap expansion may make the name longer.
1395 * make room by pushing stuff down.
1396 */
1397 tmp = person + person_len - (tzlen + 1);
1398 memmove(tmp, *tz, tzlen);
1399 tmp[tzlen] = 0;
1400 *tz = tmp;
1401
1402 /*
1403 * Now, convert both name and e-mail using mailmap
1404 */
1405 if (map_user(&mailmap, mail+1, mail_len-1, person, tmp-person-1)) {
1406 /* Add a trailing '>' to email, since map_user returns plain emails
1407 Note: It already has '<', since we replace from mail+1 */
1408 mailpos = memchr(mail, '\0', mail_len);
1409 if (mailpos && mailpos-mail < mail_len - 1) {
1410 *mailpos = '>';
1411 *(mailpos+1) = '\0';
1412 }
1413 }
1414 }
1415
1416 static void get_commit_info(struct commit *commit,
1417 struct commit_info *ret,
1418 int detailed)
1419 {
1420 int len;
1421 const char *subject;
1422 char *reencoded, *message;
1423 static char author_name[1024];
1424 static char author_mail[1024];
1425 static char committer_name[1024];
1426 static char committer_mail[1024];
1427 static char summary_buf[1024];
1428
1429 /*
1430 * We've operated without save_commit_buffer, so
1431 * we now need to populate them for output.
1432 */
1433 if (!commit->buffer) {
1434 enum object_type type;
1435 unsigned long size;
1436 commit->buffer =
1437 read_sha1_file(commit->object.sha1, &type, &size);
1438 if (!commit->buffer)
1439 die("Cannot read commit %s",
1440 sha1_to_hex(commit->object.sha1));
1441 }
1442 reencoded = reencode_commit_message(commit, NULL);
1443 message = reencoded ? reencoded : commit->buffer;
1444 ret->author = author_name;
1445 ret->author_mail = author_mail;
1446 get_ac_line(message, "\nauthor ",
1447 sizeof(author_name), author_name,
1448 sizeof(author_mail), author_mail,
1449 &ret->author_time, &ret->author_tz);
1450
1451 if (!detailed) {
1452 free(reencoded);
1453 return;
1454 }
1455
1456 ret->committer = committer_name;
1457 ret->committer_mail = committer_mail;
1458 get_ac_line(message, "\ncommitter ",
1459 sizeof(committer_name), committer_name,
1460 sizeof(committer_mail), committer_mail,
1461 &ret->committer_time, &ret->committer_tz);
1462
1463 ret->summary = summary_buf;
1464 len = find_commit_subject(message, &subject);
1465 if (len && len < sizeof(summary_buf)) {
1466 memcpy(summary_buf, subject, len);
1467 summary_buf[len] = 0;
1468 } else {
1469 sprintf(summary_buf, "(%s)", sha1_to_hex(commit->object.sha1));
1470 }
1471 free(reencoded);
1472 }
1473
1474 /*
1475 * To allow LF and other nonportable characters in pathnames,
1476 * they are c-style quoted as needed.
1477 */
1478 static void write_filename_info(const char *path)
1479 {
1480 printf("filename ");
1481 write_name_quoted(path, stdout, '\n');
1482 }
1483
1484 /*
1485 * Porcelain/Incremental format wants to show a lot of details per
1486 * commit. Instead of repeating this every line, emit it only once,
1487 * the first time each commit appears in the output.
1488 */
1489 static int emit_one_suspect_detail(struct origin *suspect)
1490 {
1491 struct commit_info ci;
1492
1493 if (suspect->commit->object.flags & METAINFO_SHOWN)
1494 return 0;
1495
1496 suspect->commit->object.flags |= METAINFO_SHOWN;
1497 get_commit_info(suspect->commit, &ci, 1);
1498 printf("author %s\n", ci.author);
1499 printf("author-mail %s\n", ci.author_mail);
1500 printf("author-time %lu\n", ci.author_time);
1501 printf("author-tz %s\n", ci.author_tz);
1502 printf("committer %s\n", ci.committer);
1503 printf("committer-mail %s\n", ci.committer_mail);
1504 printf("committer-time %lu\n", ci.committer_time);
1505 printf("committer-tz %s\n", ci.committer_tz);
1506 printf("summary %s\n", ci.summary);
1507 if (suspect->commit->object.flags & UNINTERESTING)
1508 printf("boundary\n");
1509 if (suspect->previous) {
1510 struct origin *prev = suspect->previous;
1511 printf("previous %s ", sha1_to_hex(prev->commit->object.sha1));
1512 write_name_quoted(prev->path, stdout, '\n');
1513 }
1514 return 1;
1515 }
1516
1517 /*
1518 * The blame_entry is found to be guilty for the range. Mark it
1519 * as such, and show it in incremental output.
1520 */
1521 static void found_guilty_entry(struct blame_entry *ent)
1522 {
1523 if (ent->guilty)
1524 return;
1525 ent->guilty = 1;
1526 if (incremental) {
1527 struct origin *suspect = ent->suspect;
1528
1529 printf("%s %d %d %d\n",
1530 sha1_to_hex(suspect->commit->object.sha1),
1531 ent->s_lno + 1, ent->lno + 1, ent->num_lines);
1532 emit_one_suspect_detail(suspect);
1533 write_filename_info(suspect->path);
1534 maybe_flush_or_die(stdout, "stdout");
1535 }
1536 }
1537
1538 /*
1539 * The main loop -- while the scoreboard has lines whose true origin
1540 * is still unknown, pick one blame_entry, and allow its current
1541 * suspect to pass blames to its parents.
1542 */
1543 static void assign_blame(struct scoreboard *sb, int opt)
1544 {
1545 struct rev_info *revs = sb->revs;
1546
1547 while (1) {
1548 struct blame_entry *ent;
1549 struct commit *commit;
1550 struct origin *suspect = NULL;
1551
1552 /* find one suspect to break down */
1553 for (ent = sb->ent; !suspect && ent; ent = ent->next)
1554 if (!ent->guilty)
1555 suspect = ent->suspect;
1556 if (!suspect)
1557 return; /* all done */
1558
1559 /*
1560 * We will use this suspect later in the loop,
1561 * so hold onto it in the meantime.
1562 */
1563 origin_incref(suspect);
1564 commit = suspect->commit;
1565 if (!commit->object.parsed)
1566 parse_commit(commit);
1567 if (reverse ||
1568 (!(commit->object.flags & UNINTERESTING) &&
1569 !(revs->max_age != -1 && commit->date < revs->max_age)))
1570 pass_blame(sb, suspect, opt);
1571 else {
1572 commit->object.flags |= UNINTERESTING;
1573 if (commit->object.parsed)
1574 mark_parents_uninteresting(commit);
1575 }
1576 /* treat root commit as boundary */
1577 if (!commit->parents && !show_root)
1578 commit->object.flags |= UNINTERESTING;
1579
1580 /* Take responsibility for the remaining entries */
1581 for (ent = sb->ent; ent; ent = ent->next)
1582 if (same_suspect(ent->suspect, suspect))
1583 found_guilty_entry(ent);
1584 origin_decref(suspect);
1585
1586 if (DEBUG) /* sanity */
1587 sanity_check_refcnt(sb);
1588 }
1589 }
1590
1591 static const char *format_time(unsigned long time, const char *tz_str,
1592 int show_raw_time)
1593 {
1594 static char time_buf[128];
1595 const char *time_str;
1596 int time_len;
1597 int tz;
1598
1599 if (show_raw_time) {
1600 sprintf(time_buf, "%lu %s", time, tz_str);
1601 }
1602 else {
1603 tz = atoi(tz_str);
1604 time_str = show_date(time, tz, blame_date_mode);
1605 time_len = strlen(time_str);
1606 memcpy(time_buf, time_str, time_len);
1607 memset(time_buf + time_len, ' ', blame_date_width - time_len);
1608 }
1609 return time_buf;
1610 }
1611
1612 #define OUTPUT_ANNOTATE_COMPAT 001
1613 #define OUTPUT_LONG_OBJECT_NAME 002
1614 #define OUTPUT_RAW_TIMESTAMP 004
1615 #define OUTPUT_PORCELAIN 010
1616 #define OUTPUT_SHOW_NAME 020
1617 #define OUTPUT_SHOW_NUMBER 040
1618 #define OUTPUT_SHOW_SCORE 0100
1619 #define OUTPUT_NO_AUTHOR 0200
1620 #define OUTPUT_SHOW_EMAIL 0400
1621
1622 static void emit_porcelain(struct scoreboard *sb, struct blame_entry *ent)
1623 {
1624 int cnt;
1625 const char *cp;
1626 struct origin *suspect = ent->suspect;
1627 char hex[41];
1628
1629 strcpy(hex, sha1_to_hex(suspect->commit->object.sha1));
1630 printf("%s%c%d %d %d\n",
1631 hex,
1632 ent->guilty ? ' ' : '*', /* purely for debugging */
1633 ent->s_lno + 1,
1634 ent->lno + 1,
1635 ent->num_lines);
1636 if (emit_one_suspect_detail(suspect) ||
1637 (suspect->commit->object.flags & MORE_THAN_ONE_PATH))
1638 write_filename_info(suspect->path);
1639
1640 cp = nth_line(sb, ent->lno);
1641 for (cnt = 0; cnt < ent->num_lines; cnt++) {
1642 char ch;
1643 if (cnt)
1644 printf("%s %d %d\n", hex,
1645 ent->s_lno + 1 + cnt,
1646 ent->lno + 1 + cnt);
1647 putchar('\t');
1648 do {
1649 ch = *cp++;
1650 putchar(ch);
1651 } while (ch != '\n' &&
1652 cp < sb->final_buf + sb->final_buf_size);
1653 }
1654
1655 if (sb->final_buf_size && cp[-1] != '\n')
1656 putchar('\n');
1657 }
1658
1659 static void emit_other(struct scoreboard *sb, struct blame_entry *ent, int opt)
1660 {
1661 int cnt;
1662 const char *cp;
1663 struct origin *suspect = ent->suspect;
1664 struct commit_info ci;
1665 char hex[41];
1666 int show_raw_time = !!(opt & OUTPUT_RAW_TIMESTAMP);
1667
1668 get_commit_info(suspect->commit, &ci, 1);
1669 strcpy(hex, sha1_to_hex(suspect->commit->object.sha1));
1670
1671 cp = nth_line(sb, ent->lno);
1672 for (cnt = 0; cnt < ent->num_lines; cnt++) {
1673 char ch;
1674 int length = (opt & OUTPUT_LONG_OBJECT_NAME) ? 40 : 8;
1675
1676 if (suspect->commit->object.flags & UNINTERESTING) {
1677 if (blank_boundary)
1678 memset(hex, ' ', length);
1679 else if (!(opt & OUTPUT_ANNOTATE_COMPAT)) {
1680 length--;
1681 putchar('^');
1682 }
1683 }
1684
1685 printf("%.*s", length, hex);
1686 if (opt & OUTPUT_ANNOTATE_COMPAT) {
1687 const char *name;
1688 if (opt & OUTPUT_SHOW_EMAIL)
1689 name = ci.author_mail;
1690 else
1691 name = ci.author;
1692 printf("\t(%10s\t%10s\t%d)", name,
1693 format_time(ci.author_time, ci.author_tz,
1694 show_raw_time),
1695 ent->lno + 1 + cnt);
1696 } else {
1697 if (opt & OUTPUT_SHOW_SCORE)
1698 printf(" %*d %02d",
1699 max_score_digits, ent->score,
1700 ent->suspect->refcnt);
1701 if (opt & OUTPUT_SHOW_NAME)
1702 printf(" %-*.*s", longest_file, longest_file,
1703 suspect->path);
1704 if (opt & OUTPUT_SHOW_NUMBER)
1705 printf(" %*d", max_orig_digits,
1706 ent->s_lno + 1 + cnt);
1707
1708 if (!(opt & OUTPUT_NO_AUTHOR)) {
1709 const char *name;
1710 int pad;
1711 if (opt & OUTPUT_SHOW_EMAIL)
1712 name = ci.author_mail;
1713 else
1714 name = ci.author;
1715 pad = longest_author - utf8_strwidth(name);
1716 printf(" (%s%*s %10s",
1717 name, pad, "",
1718 format_time(ci.author_time,
1719 ci.author_tz,
1720 show_raw_time));
1721 }
1722 printf(" %*d) ",
1723 max_digits, ent->lno + 1 + cnt);
1724 }
1725 do {
1726 ch = *cp++;
1727 putchar(ch);
1728 } while (ch != '\n' &&
1729 cp < sb->final_buf + sb->final_buf_size);
1730 }
1731
1732 if (sb->final_buf_size && cp[-1] != '\n')
1733 putchar('\n');
1734 }
1735
1736 static void output(struct scoreboard *sb, int option)
1737 {
1738 struct blame_entry *ent;
1739
1740 if (option & OUTPUT_PORCELAIN) {
1741 for (ent = sb->ent; ent; ent = ent->next) {
1742 struct blame_entry *oth;
1743 struct origin *suspect = ent->suspect;
1744 struct commit *commit = suspect->commit;
1745 if (commit->object.flags & MORE_THAN_ONE_PATH)
1746 continue;
1747 for (oth = ent->next; oth; oth = oth->next) {
1748 if ((oth->suspect->commit != commit) ||
1749 !strcmp(oth->suspect->path, suspect->path))
1750 continue;
1751 commit->object.flags |= MORE_THAN_ONE_PATH;
1752 break;
1753 }
1754 }
1755 }
1756
1757 for (ent = sb->ent; ent; ent = ent->next) {
1758 if (option & OUTPUT_PORCELAIN)
1759 emit_porcelain(sb, ent);
1760 else {
1761 emit_other(sb, ent, option);
1762 }
1763 }
1764 }
1765
1766 /*
1767 * To allow quick access to the contents of nth line in the
1768 * final image, prepare an index in the scoreboard.
1769 */
1770 static int prepare_lines(struct scoreboard *sb)
1771 {
1772 const char *buf = sb->final_buf;
1773 unsigned long len = sb->final_buf_size;
1774 int num = 0, incomplete = 0, bol = 1;
1775
1776 if (len && buf[len-1] != '\n')
1777 incomplete++; /* incomplete line at the end */
1778 while (len--) {
1779 if (bol) {
1780 sb->lineno = xrealloc(sb->lineno,
1781 sizeof(int *) * (num + 1));
1782 sb->lineno[num] = buf - sb->final_buf;
1783 bol = 0;
1784 }
1785 if (*buf++ == '\n') {
1786 num++;
1787 bol = 1;
1788 }
1789 }
1790 sb->lineno = xrealloc(sb->lineno,
1791 sizeof(int *) * (num + incomplete + 1));
1792 sb->lineno[num + incomplete] = buf - sb->final_buf;
1793 sb->num_lines = num + incomplete;
1794 return sb->num_lines;
1795 }
1796
1797 /*
1798 * Add phony grafts for use with -S; this is primarily to
1799 * support git's cvsserver that wants to give a linear history
1800 * to its clients.
1801 */
1802 static int read_ancestry(const char *graft_file)
1803 {
1804 FILE *fp = fopen(graft_file, "r");
1805 char buf[1024];
1806 if (!fp)
1807 return -1;
1808 while (fgets(buf, sizeof(buf), fp)) {
1809 /* The format is just "Commit Parent1 Parent2 ...\n" */
1810 int len = strlen(buf);
1811 struct commit_graft *graft = read_graft_line(buf, len);
1812 if (graft)
1813 register_commit_graft(graft, 0);
1814 }
1815 fclose(fp);
1816 return 0;
1817 }
1818
1819 /*
1820 * How many columns do we need to show line numbers in decimal?
1821 */
1822 static int lineno_width(int lines)
1823 {
1824 int i, width;
1825
1826 for (width = 1, i = 10; i <= lines; width++)
1827 i *= 10;
1828 return width;
1829 }
1830
1831 /*
1832 * How many columns do we need to show line numbers, authors,
1833 * and filenames?
1834 */
1835 static void find_alignment(struct scoreboard *sb, int *option)
1836 {
1837 int longest_src_lines = 0;
1838 int longest_dst_lines = 0;
1839 unsigned largest_score = 0;
1840 struct blame_entry *e;
1841
1842 for (e = sb->ent; e; e = e->next) {
1843 struct origin *suspect = e->suspect;
1844 struct commit_info ci;
1845 int num;
1846
1847 if (strcmp(suspect->path, sb->path))
1848 *option |= OUTPUT_SHOW_NAME;
1849 num = strlen(suspect->path);
1850 if (longest_file < num)
1851 longest_file = num;
1852 if (!(suspect->commit->object.flags & METAINFO_SHOWN)) {
1853 suspect->commit->object.flags |= METAINFO_SHOWN;
1854 get_commit_info(suspect->commit, &ci, 1);
1855 if (*option & OUTPUT_SHOW_EMAIL)
1856 num = utf8_strwidth(ci.author_mail);
1857 else
1858 num = utf8_strwidth(ci.author);
1859 if (longest_author < num)
1860 longest_author = num;
1861 }
1862 num = e->s_lno + e->num_lines;
1863 if (longest_src_lines < num)
1864 longest_src_lines = num;
1865 num = e->lno + e->num_lines;
1866 if (longest_dst_lines < num)
1867 longest_dst_lines = num;
1868 if (largest_score < ent_score(sb, e))
1869 largest_score = ent_score(sb, e);
1870 }
1871 max_orig_digits = lineno_width(longest_src_lines);
1872 max_digits = lineno_width(longest_dst_lines);
1873 max_score_digits = lineno_width(largest_score);
1874 }
1875
1876 /*
1877 * For debugging -- origin is refcounted, and this asserts that
1878 * we do not underflow.
1879 */
1880 static void sanity_check_refcnt(struct scoreboard *sb)
1881 {
1882 int baa = 0;
1883 struct blame_entry *ent;
1884
1885 for (ent = sb->ent; ent; ent = ent->next) {
1886 /* Nobody should have zero or negative refcnt */
1887 if (ent->suspect->refcnt <= 0) {
1888 fprintf(stderr, "%s in %s has negative refcnt %d\n",
1889 ent->suspect->path,
1890 sha1_to_hex(ent->suspect->commit->object.sha1),
1891 ent->suspect->refcnt);
1892 baa = 1;
1893 }
1894 }
1895 if (baa) {
1896 int opt = 0160;
1897 find_alignment(sb, &opt);
1898 output(sb, opt);
1899 die("Baa %d!", baa);
1900 }
1901 }
1902
1903 /*
1904 * Used for the command line parsing; check if the path exists
1905 * in the working tree.
1906 */
1907 static int has_string_in_work_tree(const char *path)
1908 {
1909 struct stat st;
1910 return !lstat(path, &st);
1911 }
1912
1913 static unsigned parse_score(const char *arg)
1914 {
1915 char *end;
1916 unsigned long score = strtoul(arg, &end, 10);
1917 if (*end)
1918 return 0;
1919 return score;
1920 }
1921
1922 static const char *add_prefix(const char *prefix, const char *path)
1923 {
1924 return prefix_path(prefix, prefix ? strlen(prefix) : 0, path);
1925 }
1926
1927 /*
1928 * Parsing of (comma separated) one item in the -L option
1929 */
1930 static const char *parse_loc(const char *spec,
1931 struct scoreboard *sb, long lno,
1932 long begin, long *ret)
1933 {
1934 char *term;
1935 const char *line;
1936 long num;
1937 int reg_error;
1938 regex_t regexp;
1939 regmatch_t match[1];
1940
1941 /* Allow "-L <something>,+20" to mean starting at <something>
1942 * for 20 lines, or "-L <something>,-5" for 5 lines ending at
1943 * <something>.
1944 */
1945 if (1 < begin && (spec[0] == '+' || spec[0] == '-')) {
1946 num = strtol(spec + 1, &term, 10);
1947 if (term != spec + 1) {
1948 if (spec[0] == '-')
1949 num = 0 - num;
1950 if (0 < num)
1951 *ret = begin + num - 2;
1952 else if (!num)
1953 *ret = begin;
1954 else
1955 *ret = begin + num;
1956 return term;
1957 }
1958 return spec;
1959 }
1960 num = strtol(spec, &term, 10);
1961 if (term != spec) {
1962 *ret = num;
1963 return term;
1964 }
1965 if (spec[0] != '/')
1966 return spec;
1967
1968 /* it could be a regexp of form /.../ */
1969 for (term = (char *) spec + 1; *term && *term != '/'; term++) {
1970 if (*term == '\\')
1971 term++;
1972 }
1973 if (*term != '/')
1974 return spec;
1975
1976 /* try [spec+1 .. term-1] as regexp */
1977 *term = 0;
1978 begin--; /* input is in human terms */
1979 line = nth_line(sb, begin);
1980
1981 if (!(reg_error = regcomp(&regexp, spec + 1, REG_NEWLINE)) &&
1982 !(reg_error = regexec(&regexp, line, 1, match, 0))) {
1983 const char *cp = line + match[0].rm_so;
1984 const char *nline;
1985
1986 while (begin++ < lno) {
1987 nline = nth_line(sb, begin);
1988 if (line <= cp && cp < nline)
1989 break;
1990 line = nline;
1991 }
1992 *ret = begin;
1993 regfree(&regexp);
1994 *term++ = '/';
1995 return term;
1996 }
1997 else {
1998 char errbuf[1024];
1999 regerror(reg_error, &regexp, errbuf, 1024);
2000 die("-L parameter '%s': %s", spec + 1, errbuf);
2001 }
2002 }
2003
2004 /*
2005 * Parsing of -L option
2006 */
2007 static void prepare_blame_range(struct scoreboard *sb,
2008 const char *bottomtop,
2009 long lno,
2010 long *bottom, long *top)
2011 {
2012 const char *term;
2013
2014 term = parse_loc(bottomtop, sb, lno, 1, bottom);
2015 if (*term == ',') {
2016 term = parse_loc(term + 1, sb, lno, *bottom + 1, top);
2017 if (*term)
2018 usage(blame_usage);
2019 }
2020 if (*term)
2021 usage(blame_usage);
2022 }
2023
2024 static int git_blame_config(const char *var, const char *value, void *cb)
2025 {
2026 if (!strcmp(var, "blame.showroot")) {
2027 show_root = git_config_bool(var, value);
2028 return 0;
2029 }
2030 if (!strcmp(var, "blame.blankboundary")) {
2031 blank_boundary = git_config_bool(var, value);
2032 return 0;
2033 }
2034 if (!strcmp(var, "blame.date")) {
2035 if (!value)
2036 return config_error_nonbool(var);
2037 blame_date_mode = parse_date_format(value);
2038 return 0;
2039 }
2040
2041 switch (userdiff_config(var, value)) {
2042 case 0:
2043 break;
2044 case -1:
2045 return -1;
2046 default:
2047 return 0;
2048 }
2049
2050 return git_default_config(var, value, cb);
2051 }
2052
2053 /*
2054 * Prepare a dummy commit that represents the work tree (or staged) item.
2055 * Note that annotating work tree item never works in the reverse.
2056 */
2057 static struct commit *fake_working_tree_commit(struct diff_options *opt,
2058 const char *path,
2059 const char *contents_from)
2060 {
2061 struct commit *commit;
2062 struct origin *origin;
2063 unsigned char head_sha1[20];
2064 struct strbuf buf = STRBUF_INIT;
2065 const char *ident;
2066 time_t now;
2067 int size, len;
2068 struct cache_entry *ce;
2069 unsigned mode;
2070
2071 if (get_sha1("HEAD", head_sha1))
2072 die("No such ref: HEAD");
2073
2074 time(&now);
2075 commit = xcalloc(1, sizeof(*commit));
2076 commit->parents = xcalloc(1, sizeof(*commit->parents));
2077 commit->parents->item = lookup_commit_reference(head_sha1);
2078 commit->object.parsed = 1;
2079 commit->date = now;
2080 commit->object.type = OBJ_COMMIT;
2081
2082 origin = make_origin(commit, path);
2083
2084 if (!contents_from || strcmp("-", contents_from)) {
2085 struct stat st;
2086 const char *read_from;
2087 unsigned long buf_len;
2088
2089 if (contents_from) {
2090 if (stat(contents_from, &st) < 0)
2091 die_errno("Cannot stat '%s'", contents_from);
2092 read_from = contents_from;
2093 }
2094 else {
2095 if (lstat(path, &st) < 0)
2096 die_errno("Cannot lstat '%s'", path);
2097 read_from = path;
2098 }
2099 mode = canon_mode(st.st_mode);
2100
2101 switch (st.st_mode & S_IFMT) {
2102 case S_IFREG:
2103 if (DIFF_OPT_TST(opt, ALLOW_TEXTCONV) &&
2104 textconv_object(read_from, mode, null_sha1, &buf.buf, &buf_len))
2105 buf.len = buf_len;
2106 else if (strbuf_read_file(&buf, read_from, st.st_size) != st.st_size)
2107 die_errno("cannot open or read '%s'", read_from);
2108 break;
2109 case S_IFLNK:
2110 if (strbuf_readlink(&buf, read_from, st.st_size) < 0)
2111 die_errno("cannot readlink '%s'", read_from);
2112 break;
2113 default:
2114 die("unsupported file type %s", read_from);
2115 }
2116 }
2117 else {
2118 /* Reading from stdin */
2119 contents_from = "standard input";
2120 mode = 0;
2121 if (strbuf_read(&buf, 0, 0) < 0)
2122 die_errno("failed to read from stdin");
2123 }
2124 convert_to_git(path, buf.buf, buf.len, &buf, 0);
2125 origin->file.ptr = buf.buf;
2126 origin->file.size = buf.len;
2127 pretend_sha1_file(buf.buf, buf.len, OBJ_BLOB, origin->blob_sha1);
2128 commit->util = origin;
2129
2130 /*
2131 * Read the current index, replace the path entry with
2132 * origin->blob_sha1 without mucking with its mode or type
2133 * bits; we are not going to write this index out -- we just
2134 * want to run "diff-index --cached".
2135 */
2136 discard_cache();
2137 read_cache();
2138
2139 len = strlen(path);
2140 if (!mode) {
2141 int pos = cache_name_pos(path, len);
2142 if (0 <= pos)
2143 mode = active_cache[pos]->ce_mode;
2144 else
2145 /* Let's not bother reading from HEAD tree */
2146 mode = S_IFREG | 0644;
2147 }
2148 size = cache_entry_size(len);
2149 ce = xcalloc(1, size);
2150 hashcpy(ce->sha1, origin->blob_sha1);
2151 memcpy(ce->name, path, len);
2152 ce->ce_flags = create_ce_flags(len, 0);
2153 ce->ce_mode = create_ce_mode(mode);
2154 add_cache_entry(ce, ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE);
2155
2156 /*
2157 * We are not going to write this out, so this does not matter
2158 * right now, but someday we might optimize diff-index --cached
2159 * with cache-tree information.
2160 */
2161 cache_tree_invalidate_path(active_cache_tree, path);
2162
2163 commit->buffer = xmalloc(400);
2164 ident = fmt_ident("Not Committed Yet", "not.committed.yet", NULL, 0);
2165 snprintf(commit->buffer, 400,
2166 "tree 0000000000000000000000000000000000000000\n"
2167 "parent %s\n"
2168 "author %s\n"
2169 "committer %s\n\n"
2170 "Version of %s from %s\n",
2171 sha1_to_hex(head_sha1),
2172 ident, ident, path, contents_from ? contents_from : path);
2173 return commit;
2174 }
2175
2176 static const char *prepare_final(struct scoreboard *sb)
2177 {
2178 int i;
2179 const char *final_commit_name = NULL;
2180 struct rev_info *revs = sb->revs;
2181
2182 /*
2183 * There must be one and only one positive commit in the
2184 * revs->pending array.
2185 */
2186 for (i = 0; i < revs->pending.nr; i++) {
2187 struct object *obj = revs->pending.objects[i].item;
2188 if (obj->flags & UNINTERESTING)
2189 continue;
2190 while (obj->type == OBJ_TAG)
2191 obj = deref_tag(obj, NULL, 0);
2192 if (obj->type != OBJ_COMMIT)
2193 die("Non commit %s?", revs->pending.objects[i].name);
2194 if (sb->final)
2195 die("More than one commit to dig from %s and %s?",
2196 revs->pending.objects[i].name,
2197 final_commit_name);
2198 sb->final = (struct commit *) obj;
2199 final_commit_name = revs->pending.objects[i].name;
2200 }
2201 return final_commit_name;
2202 }
2203
2204 static const char *prepare_initial(struct scoreboard *sb)
2205 {
2206 int i;
2207 const char *final_commit_name = NULL;
2208 struct rev_info *revs = sb->revs;
2209
2210 /*
2211 * There must be one and only one negative commit, and it must be
2212 * the boundary.
2213 */
2214 for (i = 0; i < revs->pending.nr; i++) {
2215 struct object *obj = revs->pending.objects[i].item;
2216 if (!(obj->flags & UNINTERESTING))
2217 continue;
2218 while (obj->type == OBJ_TAG)
2219 obj = deref_tag(obj, NULL, 0);
2220 if (obj->type != OBJ_COMMIT)
2221 die("Non commit %s?", revs->pending.objects[i].name);
2222 if (sb->final)
2223 die("More than one commit to dig down to %s and %s?",
2224 revs->pending.objects[i].name,
2225 final_commit_name);
2226 sb->final = (struct commit *) obj;
2227 final_commit_name = revs->pending.objects[i].name;
2228 }
2229 if (!final_commit_name)
2230 die("No commit to dig down to?");
2231 return final_commit_name;
2232 }
2233
2234 static int blame_copy_callback(const struct option *option, const char *arg, int unset)
2235 {
2236 int *opt = option->value;
2237
2238 /*
2239 * -C enables copy from removed files;
2240 * -C -C enables copy from existing files, but only
2241 * when blaming a new file;
2242 * -C -C -C enables copy from existing files for
2243 * everybody
2244 */
2245 if (*opt & PICKAXE_BLAME_COPY_HARDER)
2246 *opt |= PICKAXE_BLAME_COPY_HARDEST;
2247 if (*opt & PICKAXE_BLAME_COPY)
2248 *opt |= PICKAXE_BLAME_COPY_HARDER;
2249 *opt |= PICKAXE_BLAME_COPY | PICKAXE_BLAME_MOVE;
2250
2251 if (arg)
2252 blame_copy_score = parse_score(arg);
2253 return 0;
2254 }
2255
2256 static int blame_move_callback(const struct option *option, const char *arg, int unset)
2257 {
2258 int *opt = option->value;
2259
2260 *opt |= PICKAXE_BLAME_MOVE;
2261
2262 if (arg)
2263 blame_move_score = parse_score(arg);
2264 return 0;
2265 }
2266
2267 static int blame_bottomtop_callback(const struct option *option, const char *arg, int unset)
2268 {
2269 const char **bottomtop = option->value;
2270 if (!arg)
2271 return -1;
2272 if (*bottomtop)
2273 die("More than one '-L n,m' option given");
2274 *bottomtop = arg;
2275 return 0;
2276 }
2277
2278 int cmd_blame(int argc, const char **argv, const char *prefix)
2279 {
2280 struct rev_info revs;
2281 const char *path;
2282 struct scoreboard sb;
2283 struct origin *o;
2284 struct blame_entry *ent;
2285 long dashdash_pos, bottom, top, lno;
2286 const char *final_commit_name = NULL;
2287 enum object_type type;
2288
2289 static const char *bottomtop = NULL;
2290 static int output_option = 0, opt = 0;
2291 static int show_stats = 0;
2292 static const char *revs_file = NULL;
2293 static const char *contents_from = NULL;
2294 static const struct option options[] = {
2295 OPT_BOOLEAN(0, "incremental", &incremental, "Show blame entries as we find them, incrementally"),
2296 OPT_BOOLEAN('b', NULL, &blank_boundary, "Show blank SHA-1 for boundary commits (Default: off)"),
2297 OPT_BOOLEAN(0, "root", &show_root, "Do not treat root commits as boundaries (Default: off)"),
2298 OPT_BOOLEAN(0, "show-stats", &show_stats, "Show work cost statistics"),
2299 OPT_BIT(0, "score-debug", &output_option, "Show output score for blame entries", OUTPUT_SHOW_SCORE),
2300 OPT_BIT('f', "show-name", &output_option, "Show original filename (Default: auto)", OUTPUT_SHOW_NAME),
2301 OPT_BIT('n', "show-number", &output_option, "Show original linenumber (Default: off)", OUTPUT_SHOW_NUMBER),
2302 OPT_BIT('p', "porcelain", &output_option, "Show in a format designed for machine consumption", OUTPUT_PORCELAIN),
2303 OPT_BIT('c', NULL, &output_option, "Use the same output mode as git-annotate (Default: off)", OUTPUT_ANNOTATE_COMPAT),
2304 OPT_BIT('t', NULL, &output_option, "Show raw timestamp (Default: off)", OUTPUT_RAW_TIMESTAMP),
2305 OPT_BIT('l', NULL, &output_option, "Show long commit SHA1 (Default: off)", OUTPUT_LONG_OBJECT_NAME),
2306 OPT_BIT('s', NULL, &output_option, "Suppress author name and timestamp (Default: off)", OUTPUT_NO_AUTHOR),
2307 OPT_BIT('e', "show-email", &output_option, "Show author email instead of name (Default: off)", OUTPUT_SHOW_EMAIL),
2308 OPT_BIT('w', NULL, &xdl_opts, "Ignore whitespace differences", XDF_IGNORE_WHITESPACE),
2309 OPT_STRING('S', NULL, &revs_file, "file", "Use revisions from <file> instead of calling git-rev-list"),
2310 OPT_STRING(0, "contents", &contents_from, "file", "Use <file>'s contents as the final image"),
2311 { OPTION_CALLBACK, 'C', NULL, &opt, "score", "Find line copies within and across files", PARSE_OPT_OPTARG, blame_copy_callback },
2312 { OPTION_CALLBACK, 'M', NULL, &opt, "score", "Find line movements within and across files", PARSE_OPT_OPTARG, blame_move_callback },
2313 OPT_CALLBACK('L', NULL, &bottomtop, "n,m", "Process only line range n,m, counting from 1", blame_bottomtop_callback),
2314 OPT_END()
2315 };
2316
2317 struct parse_opt_ctx_t ctx;
2318 int cmd_is_annotate = !strcmp(argv[0], "annotate");
2319
2320 git_config(git_blame_config, NULL);
2321 init_revisions(&revs, NULL);
2322 revs.date_mode = blame_date_mode;
2323 DIFF_OPT_SET(&revs.diffopt, ALLOW_TEXTCONV);
2324
2325 save_commit_buffer = 0;
2326 dashdash_pos = 0;
2327
2328 parse_options_start(&ctx, argc, argv, prefix, options,
2329 PARSE_OPT_KEEP_DASHDASH | PARSE_OPT_KEEP_ARGV0);
2330 for (;;) {
2331 switch (parse_options_step(&ctx, options, blame_opt_usage)) {
2332 case PARSE_OPT_HELP:
2333 exit(129);
2334 case PARSE_OPT_DONE:
2335 if (ctx.argv[0])
2336 dashdash_pos = ctx.cpidx;
2337 goto parse_done;
2338 }
2339
2340 if (!strcmp(ctx.argv[0], "--reverse")) {
2341 ctx.argv[0] = "--children";
2342 reverse = 1;
2343 }
2344 parse_revision_opt(&revs, &ctx, options, blame_opt_usage);
2345 }
2346 parse_done:
2347 argc = parse_options_end(&ctx);
2348
2349 if (revs_file && read_ancestry(revs_file))
2350 die_errno("reading graft file '%s' failed", revs_file);
2351
2352 if (cmd_is_annotate) {
2353 output_option |= OUTPUT_ANNOTATE_COMPAT;
2354 blame_date_mode = DATE_ISO8601;
2355 } else {
2356 blame_date_mode = revs.date_mode;
2357 }
2358
2359 /* The maximum width used to show the dates */
2360 switch (blame_date_mode) {
2361 case DATE_RFC2822:
2362 blame_date_width = sizeof("Thu, 19 Oct 2006 16:00:04 -0700");
2363 break;
2364 case DATE_ISO8601:
2365 blame_date_width = sizeof("2006-10-19 16:00:04 -0700");
2366 break;
2367 case DATE_RAW:
2368 blame_date_width = sizeof("1161298804 -0700");
2369 break;
2370 case DATE_SHORT:
2371 blame_date_width = sizeof("2006-10-19");
2372 break;
2373 case DATE_RELATIVE:
2374 /* "normal" is used as the fallback for "relative" */
2375 case DATE_LOCAL:
2376 case DATE_NORMAL:
2377 blame_date_width = sizeof("Thu Oct 19 16:00:04 2006 -0700");
2378 break;
2379 }
2380 blame_date_width -= 1; /* strip the null */
2381
2382 if (DIFF_OPT_TST(&revs.diffopt, FIND_COPIES_HARDER))
2383 opt |= (PICKAXE_BLAME_COPY | PICKAXE_BLAME_MOVE |
2384 PICKAXE_BLAME_COPY_HARDER);
2385
2386 if (!blame_move_score)
2387 blame_move_score = BLAME_DEFAULT_MOVE_SCORE;
2388 if (!blame_copy_score)
2389 blame_copy_score = BLAME_DEFAULT_COPY_SCORE;
2390
2391 /*
2392 * We have collected options unknown to us in argv[1..unk]
2393 * which are to be passed to revision machinery if we are
2394 * going to do the "bottom" processing.
2395 *
2396 * The remaining are:
2397 *
2398 * (1) if dashdash_pos != 0, it is either
2399 * "blame [revisions] -- <path>" or
2400 * "blame -- <path> <rev>"
2401 *
2402 * (2) otherwise, it is one of the two:
2403 * "blame [revisions] <path>"
2404 * "blame <path> <rev>"
2405 *
2406 * Note that we must strip out <path> from the arguments: we do not
2407 * want the path pruning but we may want "bottom" processing.
2408 */
2409 if (dashdash_pos) {
2410 switch (argc - dashdash_pos - 1) {
2411 case 2: /* (1b) */
2412 if (argc != 4)
2413 usage_with_options(blame_opt_usage, options);
2414 /* reorder for the new way: <rev> -- <path> */
2415 argv[1] = argv[3];
2416 argv[3] = argv[2];
2417 argv[2] = "--";
2418 /* FALLTHROUGH */
2419 case 1: /* (1a) */
2420 path = add_prefix(prefix, argv[--argc]);
2421 argv[argc] = NULL;
2422 break;
2423 default:
2424 usage_with_options(blame_opt_usage, options);
2425 }
2426 } else {
2427 if (argc < 2)
2428 usage_with_options(blame_opt_usage, options);
2429 path = add_prefix(prefix, argv[argc - 1]);
2430 if (argc == 3 && !has_string_in_work_tree(path)) { /* (2b) */
2431 path = add_prefix(prefix, argv[1]);
2432 argv[1] = argv[2];
2433 }
2434 argv[argc - 1] = "--";
2435
2436 setup_work_tree();
2437 if (!has_string_in_work_tree(path))
2438 die_errno("cannot stat path '%s'", path);
2439 }
2440
2441 revs.disable_stdin = 1;
2442 setup_revisions(argc, argv, &revs, NULL);
2443 memset(&sb, 0, sizeof(sb));
2444
2445 sb.revs = &revs;
2446 if (!reverse)
2447 final_commit_name = prepare_final(&sb);
2448 else if (contents_from)
2449 die("--contents and --children do not blend well.");
2450 else
2451 final_commit_name = prepare_initial(&sb);
2452
2453 if (!sb.final) {
2454 /*
2455 * "--not A B -- path" without anything positive;
2456 * do not default to HEAD, but use the working tree
2457 * or "--contents".
2458 */
2459 setup_work_tree();
2460 sb.final = fake_working_tree_commit(&sb.revs->diffopt,
2461 path, contents_from);
2462 add_pending_object(&revs, &(sb.final->object), ":");
2463 }
2464 else if (contents_from)
2465 die("Cannot use --contents with final commit object name");
2466
2467 /*
2468 * If we have bottom, this will mark the ancestors of the
2469 * bottom commits we would reach while traversing as
2470 * uninteresting.
2471 */
2472 if (prepare_revision_walk(&revs))
2473 die("revision walk setup failed");
2474
2475 if (is_null_sha1(sb.final->object.sha1)) {
2476 char *buf;
2477 o = sb.final->util;
2478 buf = xmalloc(o->file.size + 1);
2479 memcpy(buf, o->file.ptr, o->file.size + 1);
2480 sb.final_buf = buf;
2481 sb.final_buf_size = o->file.size;
2482 }
2483 else {
2484 o = get_origin(&sb, sb.final, path);
2485 if (fill_blob_sha1_and_mode(o))
2486 die("no such path %s in %s", path, final_commit_name);
2487
2488 if (DIFF_OPT_TST(&sb.revs->diffopt, ALLOW_TEXTCONV) &&
2489 textconv_object(path, o->mode, o->blob_sha1, (char **) &sb.final_buf,
2490 &sb.final_buf_size))
2491 ;
2492 else
2493 sb.final_buf = read_sha1_file(o->blob_sha1, &type,
2494 &sb.final_buf_size);
2495
2496 if (!sb.final_buf)
2497 die("Cannot read blob %s for path %s",
2498 sha1_to_hex(o->blob_sha1),
2499 path);
2500 }
2501 num_read_blob++;
2502 lno = prepare_lines(&sb);
2503
2504 bottom = top = 0;
2505 if (bottomtop)
2506 prepare_blame_range(&sb, bottomtop, lno, &bottom, &top);
2507 if (bottom && top && top < bottom) {
2508 long tmp;
2509 tmp = top; top = bottom; bottom = tmp;
2510 }
2511 if (bottom < 1)
2512 bottom = 1;
2513 if (top < 1)
2514 top = lno;
2515 bottom--;
2516 if (lno < top || lno < bottom)
2517 die("file %s has only %lu lines", path, lno);
2518
2519 ent = xcalloc(1, sizeof(*ent));
2520 ent->lno = bottom;
2521 ent->num_lines = top - bottom;
2522 ent->suspect = o;
2523 ent->s_lno = bottom;
2524
2525 sb.ent = ent;
2526 sb.path = path;
2527
2528 read_mailmap(&mailmap, NULL);
2529
2530 if (!incremental)
2531 setup_pager();
2532
2533 assign_blame(&sb, opt);
2534
2535 if (incremental)
2536 return 0;
2537
2538 coalesce(&sb);
2539
2540 if (!(output_option & OUTPUT_PORCELAIN))
2541 find_alignment(&sb, &output_option);
2542
2543 output(&sb, output_option);
2544 free((void *)sb.final_buf);
2545 for (ent = sb.ent; ent; ) {
2546 struct blame_entry *e = ent->next;
2547 free(ent);
2548 ent = e;
2549 }
2550
2551 if (show_stats) {
2552 printf("num read blob: %d\n", num_read_blob);
2553 printf("num get patch: %d\n", num_get_patch);
2554 printf("num commits: %d\n", num_commits);
2555 }
2556 return 0;
2557 }