Merge branch 'maint-1.6.0' into maint-1.6.1
[git/git.git] / builtin-blame.c
1 /*
2 * Pickaxe
3 *
4 * Copyright (c) 2006, Junio C Hamano
5 */
6
7 #include "cache.h"
8 #include "builtin.h"
9 #include "blob.h"
10 #include "commit.h"
11 #include "tag.h"
12 #include "tree-walk.h"
13 #include "diff.h"
14 #include "diffcore.h"
15 #include "revision.h"
16 #include "quote.h"
17 #include "xdiff-interface.h"
18 #include "cache-tree.h"
19 #include "string-list.h"
20 #include "mailmap.h"
21 #include "parse-options.h"
22 #include "utf8.h"
23
24 static char blame_usage[] = "git blame [options] [rev-opts] [rev] [--] file";
25
26 static const char *blame_opt_usage[] = {
27 blame_usage,
28 "",
29 "[rev-opts] are documented in git-rev-list(1)",
30 NULL
31 };
32
33 static int longest_file;
34 static int longest_author;
35 static int max_orig_digits;
36 static int max_digits;
37 static int max_score_digits;
38 static int show_root;
39 static int reverse;
40 static int blank_boundary;
41 static int incremental;
42 static int xdl_opts = XDF_NEED_MINIMAL;
43 static struct string_list mailmap;
44
45 #ifndef DEBUG
46 #define DEBUG 0
47 #endif
48
49 /* stats */
50 static int num_read_blob;
51 static int num_get_patch;
52 static int num_commits;
53
54 #define PICKAXE_BLAME_MOVE 01
55 #define PICKAXE_BLAME_COPY 02
56 #define PICKAXE_BLAME_COPY_HARDER 04
57 #define PICKAXE_BLAME_COPY_HARDEST 010
58
59 /*
60 * blame for a blame_entry with score lower than these thresholds
61 * is not passed to the parent using move/copy logic.
62 */
63 static unsigned blame_move_score;
64 static unsigned blame_copy_score;
65 #define BLAME_DEFAULT_MOVE_SCORE 20
66 #define BLAME_DEFAULT_COPY_SCORE 40
67
68 /* bits #0..7 in revision.h, #8..11 used for merge_bases() in commit.c */
69 #define METAINFO_SHOWN (1u<<12)
70 #define MORE_THAN_ONE_PATH (1u<<13)
71
72 /*
73 * One blob in a commit that is being suspected
74 */
75 struct origin {
76 int refcnt;
77 struct commit *commit;
78 mmfile_t file;
79 unsigned char blob_sha1[20];
80 char path[FLEX_ARRAY];
81 };
82
83 /*
84 * Given an origin, prepare mmfile_t structure to be used by the
85 * diff machinery
86 */
87 static void fill_origin_blob(struct origin *o, mmfile_t *file)
88 {
89 if (!o->file.ptr) {
90 enum object_type type;
91 num_read_blob++;
92 file->ptr = read_sha1_file(o->blob_sha1, &type,
93 (unsigned long *)(&(file->size)));
94 if (!file->ptr)
95 die("Cannot read blob %s for path %s",
96 sha1_to_hex(o->blob_sha1),
97 o->path);
98 o->file = *file;
99 }
100 else
101 *file = o->file;
102 }
103
104 /*
105 * Origin is refcounted and usually we keep the blob contents to be
106 * reused.
107 */
108 static inline struct origin *origin_incref(struct origin *o)
109 {
110 if (o)
111 o->refcnt++;
112 return o;
113 }
114
115 static void origin_decref(struct origin *o)
116 {
117 if (o && --o->refcnt <= 0) {
118 free(o->file.ptr);
119 free(o);
120 }
121 }
122
123 static void drop_origin_blob(struct origin *o)
124 {
125 if (o->file.ptr) {
126 free(o->file.ptr);
127 o->file.ptr = NULL;
128 }
129 }
130
131 /*
132 * Each group of lines is described by a blame_entry; it can be split
133 * as we pass blame to the parents. They form a linked list in the
134 * scoreboard structure, sorted by the target line number.
135 */
136 struct blame_entry {
137 struct blame_entry *prev;
138 struct blame_entry *next;
139
140 /* the first line of this group in the final image;
141 * internally all line numbers are 0 based.
142 */
143 int lno;
144
145 /* how many lines this group has */
146 int num_lines;
147
148 /* the commit that introduced this group into the final image */
149 struct origin *suspect;
150
151 /* true if the suspect is truly guilty; false while we have not
152 * checked if the group came from one of its parents.
153 */
154 char guilty;
155
156 /* true if the entry has been scanned for copies in the current parent
157 */
158 char scanned;
159
160 /* the line number of the first line of this group in the
161 * suspect's file; internally all line numbers are 0 based.
162 */
163 int s_lno;
164
165 /* how significant this entry is -- cached to avoid
166 * scanning the lines over and over.
167 */
168 unsigned score;
169 };
170
171 /*
172 * The current state of the blame assignment.
173 */
174 struct scoreboard {
175 /* the final commit (i.e. where we started digging from) */
176 struct commit *final;
177 struct rev_info *revs;
178 const char *path;
179
180 /*
181 * The contents in the final image.
182 * Used by many functions to obtain contents of the nth line,
183 * indexed with scoreboard.lineno[blame_entry.lno].
184 */
185 const char *final_buf;
186 unsigned long final_buf_size;
187
188 /* linked list of blames */
189 struct blame_entry *ent;
190
191 /* look-up a line in the final buffer */
192 int num_lines;
193 int *lineno;
194 };
195
196 static inline int same_suspect(struct origin *a, struct origin *b)
197 {
198 if (a == b)
199 return 1;
200 if (a->commit != b->commit)
201 return 0;
202 return !strcmp(a->path, b->path);
203 }
204
205 static void sanity_check_refcnt(struct scoreboard *);
206
207 /*
208 * If two blame entries that are next to each other came from
209 * contiguous lines in the same origin (i.e. <commit, path> pair),
210 * merge them together.
211 */
212 static void coalesce(struct scoreboard *sb)
213 {
214 struct blame_entry *ent, *next;
215
216 for (ent = sb->ent; ent && (next = ent->next); ent = next) {
217 if (same_suspect(ent->suspect, next->suspect) &&
218 ent->guilty == next->guilty &&
219 ent->s_lno + ent->num_lines == next->s_lno) {
220 ent->num_lines += next->num_lines;
221 ent->next = next->next;
222 if (ent->next)
223 ent->next->prev = ent;
224 origin_decref(next->suspect);
225 free(next);
226 ent->score = 0;
227 next = ent; /* again */
228 }
229 }
230
231 if (DEBUG) /* sanity */
232 sanity_check_refcnt(sb);
233 }
234
235 /*
236 * Given a commit and a path in it, create a new origin structure.
237 * The callers that add blame to the scoreboard should use
238 * get_origin() to obtain shared, refcounted copy instead of calling
239 * this function directly.
240 */
241 static struct origin *make_origin(struct commit *commit, const char *path)
242 {
243 struct origin *o;
244 o = xcalloc(1, sizeof(*o) + strlen(path) + 1);
245 o->commit = commit;
246 o->refcnt = 1;
247 strcpy(o->path, path);
248 return o;
249 }
250
251 /*
252 * Locate an existing origin or create a new one.
253 */
254 static struct origin *get_origin(struct scoreboard *sb,
255 struct commit *commit,
256 const char *path)
257 {
258 struct blame_entry *e;
259
260 for (e = sb->ent; e; e = e->next) {
261 if (e->suspect->commit == commit &&
262 !strcmp(e->suspect->path, path))
263 return origin_incref(e->suspect);
264 }
265 return make_origin(commit, path);
266 }
267
268 /*
269 * Fill the blob_sha1 field of an origin if it hasn't, so that later
270 * call to fill_origin_blob() can use it to locate the data. blob_sha1
271 * for an origin is also used to pass the blame for the entire file to
272 * the parent to detect the case where a child's blob is identical to
273 * that of its parent's.
274 */
275 static int fill_blob_sha1(struct origin *origin)
276 {
277 unsigned mode;
278
279 if (!is_null_sha1(origin->blob_sha1))
280 return 0;
281 if (get_tree_entry(origin->commit->object.sha1,
282 origin->path,
283 origin->blob_sha1, &mode))
284 goto error_out;
285 if (sha1_object_info(origin->blob_sha1, NULL) != OBJ_BLOB)
286 goto error_out;
287 return 0;
288 error_out:
289 hashclr(origin->blob_sha1);
290 return -1;
291 }
292
293 /*
294 * We have an origin -- check if the same path exists in the
295 * parent and return an origin structure to represent it.
296 */
297 static struct origin *find_origin(struct scoreboard *sb,
298 struct commit *parent,
299 struct origin *origin)
300 {
301 struct origin *porigin = NULL;
302 struct diff_options diff_opts;
303 const char *paths[2];
304
305 if (parent->util) {
306 /*
307 * Each commit object can cache one origin in that
308 * commit. This is a freestanding copy of origin and
309 * not refcounted.
310 */
311 struct origin *cached = parent->util;
312 if (!strcmp(cached->path, origin->path)) {
313 /*
314 * The same path between origin and its parent
315 * without renaming -- the most common case.
316 */
317 porigin = get_origin(sb, parent, cached->path);
318
319 /*
320 * If the origin was newly created (i.e. get_origin
321 * would call make_origin if none is found in the
322 * scoreboard), it does not know the blob_sha1,
323 * so copy it. Otherwise porigin was in the
324 * scoreboard and already knows blob_sha1.
325 */
326 if (porigin->refcnt == 1)
327 hashcpy(porigin->blob_sha1, cached->blob_sha1);
328 return porigin;
329 }
330 /* otherwise it was not very useful; free it */
331 free(parent->util);
332 parent->util = NULL;
333 }
334
335 /* See if the origin->path is different between parent
336 * and origin first. Most of the time they are the
337 * same and diff-tree is fairly efficient about this.
338 */
339 diff_setup(&diff_opts);
340 DIFF_OPT_SET(&diff_opts, RECURSIVE);
341 diff_opts.detect_rename = 0;
342 diff_opts.output_format = DIFF_FORMAT_NO_OUTPUT;
343 paths[0] = origin->path;
344 paths[1] = NULL;
345
346 diff_tree_setup_paths(paths, &diff_opts);
347 if (diff_setup_done(&diff_opts) < 0)
348 die("diff-setup");
349
350 if (is_null_sha1(origin->commit->object.sha1))
351 do_diff_cache(parent->tree->object.sha1, &diff_opts);
352 else
353 diff_tree_sha1(parent->tree->object.sha1,
354 origin->commit->tree->object.sha1,
355 "", &diff_opts);
356 diffcore_std(&diff_opts);
357
358 /* It is either one entry that says "modified", or "created",
359 * or nothing.
360 */
361 if (!diff_queued_diff.nr) {
362 /* The path is the same as parent */
363 porigin = get_origin(sb, parent, origin->path);
364 hashcpy(porigin->blob_sha1, origin->blob_sha1);
365 }
366 else if (diff_queued_diff.nr != 1)
367 die("internal error in blame::find_origin");
368 else {
369 struct diff_filepair *p = diff_queued_diff.queue[0];
370 switch (p->status) {
371 default:
372 die("internal error in blame::find_origin (%c)",
373 p->status);
374 case 'M':
375 porigin = get_origin(sb, parent, origin->path);
376 hashcpy(porigin->blob_sha1, p->one->sha1);
377 break;
378 case 'A':
379 case 'T':
380 /* Did not exist in parent, or type changed */
381 break;
382 }
383 }
384 diff_flush(&diff_opts);
385 diff_tree_release_paths(&diff_opts);
386 if (porigin) {
387 /*
388 * Create a freestanding copy that is not part of
389 * the refcounted origin found in the scoreboard, and
390 * cache it in the commit.
391 */
392 struct origin *cached;
393
394 cached = make_origin(porigin->commit, porigin->path);
395 hashcpy(cached->blob_sha1, porigin->blob_sha1);
396 parent->util = cached;
397 }
398 return porigin;
399 }
400
401 /*
402 * We have an origin -- find the path that corresponds to it in its
403 * parent and return an origin structure to represent it.
404 */
405 static struct origin *find_rename(struct scoreboard *sb,
406 struct commit *parent,
407 struct origin *origin)
408 {
409 struct origin *porigin = NULL;
410 struct diff_options diff_opts;
411 int i;
412 const char *paths[2];
413
414 diff_setup(&diff_opts);
415 DIFF_OPT_SET(&diff_opts, RECURSIVE);
416 diff_opts.detect_rename = DIFF_DETECT_RENAME;
417 diff_opts.output_format = DIFF_FORMAT_NO_OUTPUT;
418 diff_opts.single_follow = origin->path;
419 paths[0] = NULL;
420 diff_tree_setup_paths(paths, &diff_opts);
421 if (diff_setup_done(&diff_opts) < 0)
422 die("diff-setup");
423
424 if (is_null_sha1(origin->commit->object.sha1))
425 do_diff_cache(parent->tree->object.sha1, &diff_opts);
426 else
427 diff_tree_sha1(parent->tree->object.sha1,
428 origin->commit->tree->object.sha1,
429 "", &diff_opts);
430 diffcore_std(&diff_opts);
431
432 for (i = 0; i < diff_queued_diff.nr; i++) {
433 struct diff_filepair *p = diff_queued_diff.queue[i];
434 if ((p->status == 'R' || p->status == 'C') &&
435 !strcmp(p->two->path, origin->path)) {
436 porigin = get_origin(sb, parent, p->one->path);
437 hashcpy(porigin->blob_sha1, p->one->sha1);
438 break;
439 }
440 }
441 diff_flush(&diff_opts);
442 diff_tree_release_paths(&diff_opts);
443 return porigin;
444 }
445
446 /*
447 * Link in a new blame entry to the scoreboard. Entries that cover the
448 * same line range have been removed from the scoreboard previously.
449 */
450 static void add_blame_entry(struct scoreboard *sb, struct blame_entry *e)
451 {
452 struct blame_entry *ent, *prev = NULL;
453
454 origin_incref(e->suspect);
455
456 for (ent = sb->ent; ent && ent->lno < e->lno; ent = ent->next)
457 prev = ent;
458
459 /* prev, if not NULL, is the last one that is below e */
460 e->prev = prev;
461 if (prev) {
462 e->next = prev->next;
463 prev->next = e;
464 }
465 else {
466 e->next = sb->ent;
467 sb->ent = e;
468 }
469 if (e->next)
470 e->next->prev = e;
471 }
472
473 /*
474 * src typically is on-stack; we want to copy the information in it to
475 * a malloced blame_entry that is already on the linked list of the
476 * scoreboard. The origin of dst loses a refcnt while the origin of src
477 * gains one.
478 */
479 static void dup_entry(struct blame_entry *dst, struct blame_entry *src)
480 {
481 struct blame_entry *p, *n;
482
483 p = dst->prev;
484 n = dst->next;
485 origin_incref(src->suspect);
486 origin_decref(dst->suspect);
487 memcpy(dst, src, sizeof(*src));
488 dst->prev = p;
489 dst->next = n;
490 dst->score = 0;
491 }
492
493 static const char *nth_line(struct scoreboard *sb, int lno)
494 {
495 return sb->final_buf + sb->lineno[lno];
496 }
497
498 /*
499 * It is known that lines between tlno to same came from parent, and e
500 * has an overlap with that range. it also is known that parent's
501 * line plno corresponds to e's line tlno.
502 *
503 * <---- e ----->
504 * <------>
505 * <------------>
506 * <------------>
507 * <------------------>
508 *
509 * Split e into potentially three parts; before this chunk, the chunk
510 * to be blamed for the parent, and after that portion.
511 */
512 static void split_overlap(struct blame_entry *split,
513 struct blame_entry *e,
514 int tlno, int plno, int same,
515 struct origin *parent)
516 {
517 int chunk_end_lno;
518 memset(split, 0, sizeof(struct blame_entry [3]));
519
520 if (e->s_lno < tlno) {
521 /* there is a pre-chunk part not blamed on parent */
522 split[0].suspect = origin_incref(e->suspect);
523 split[0].lno = e->lno;
524 split[0].s_lno = e->s_lno;
525 split[0].num_lines = tlno - e->s_lno;
526 split[1].lno = e->lno + tlno - e->s_lno;
527 split[1].s_lno = plno;
528 }
529 else {
530 split[1].lno = e->lno;
531 split[1].s_lno = plno + (e->s_lno - tlno);
532 }
533
534 if (same < e->s_lno + e->num_lines) {
535 /* there is a post-chunk part not blamed on parent */
536 split[2].suspect = origin_incref(e->suspect);
537 split[2].lno = e->lno + (same - e->s_lno);
538 split[2].s_lno = e->s_lno + (same - e->s_lno);
539 split[2].num_lines = e->s_lno + e->num_lines - same;
540 chunk_end_lno = split[2].lno;
541 }
542 else
543 chunk_end_lno = e->lno + e->num_lines;
544 split[1].num_lines = chunk_end_lno - split[1].lno;
545
546 /*
547 * if it turns out there is nothing to blame the parent for,
548 * forget about the splitting. !split[1].suspect signals this.
549 */
550 if (split[1].num_lines < 1)
551 return;
552 split[1].suspect = origin_incref(parent);
553 }
554
555 /*
556 * split_overlap() divided an existing blame e into up to three parts
557 * in split. Adjust the linked list of blames in the scoreboard to
558 * reflect the split.
559 */
560 static void split_blame(struct scoreboard *sb,
561 struct blame_entry *split,
562 struct blame_entry *e)
563 {
564 struct blame_entry *new_entry;
565
566 if (split[0].suspect && split[2].suspect) {
567 /* The first part (reuse storage for the existing entry e) */
568 dup_entry(e, &split[0]);
569
570 /* The last part -- me */
571 new_entry = xmalloc(sizeof(*new_entry));
572 memcpy(new_entry, &(split[2]), sizeof(struct blame_entry));
573 add_blame_entry(sb, new_entry);
574
575 /* ... and the middle part -- parent */
576 new_entry = xmalloc(sizeof(*new_entry));
577 memcpy(new_entry, &(split[1]), sizeof(struct blame_entry));
578 add_blame_entry(sb, new_entry);
579 }
580 else if (!split[0].suspect && !split[2].suspect)
581 /*
582 * The parent covers the entire area; reuse storage for
583 * e and replace it with the parent.
584 */
585 dup_entry(e, &split[1]);
586 else if (split[0].suspect) {
587 /* me and then parent */
588 dup_entry(e, &split[0]);
589
590 new_entry = xmalloc(sizeof(*new_entry));
591 memcpy(new_entry, &(split[1]), sizeof(struct blame_entry));
592 add_blame_entry(sb, new_entry);
593 }
594 else {
595 /* parent and then me */
596 dup_entry(e, &split[1]);
597
598 new_entry = xmalloc(sizeof(*new_entry));
599 memcpy(new_entry, &(split[2]), sizeof(struct blame_entry));
600 add_blame_entry(sb, new_entry);
601 }
602
603 if (DEBUG) { /* sanity */
604 struct blame_entry *ent;
605 int lno = sb->ent->lno, corrupt = 0;
606
607 for (ent = sb->ent; ent; ent = ent->next) {
608 if (lno != ent->lno)
609 corrupt = 1;
610 if (ent->s_lno < 0)
611 corrupt = 1;
612 lno += ent->num_lines;
613 }
614 if (corrupt) {
615 lno = sb->ent->lno;
616 for (ent = sb->ent; ent; ent = ent->next) {
617 printf("L %8d l %8d n %8d\n",
618 lno, ent->lno, ent->num_lines);
619 lno = ent->lno + ent->num_lines;
620 }
621 die("oops");
622 }
623 }
624 }
625
626 /*
627 * After splitting the blame, the origins used by the
628 * on-stack blame_entry should lose one refcnt each.
629 */
630 static void decref_split(struct blame_entry *split)
631 {
632 int i;
633
634 for (i = 0; i < 3; i++)
635 origin_decref(split[i].suspect);
636 }
637
638 /*
639 * Helper for blame_chunk(). blame_entry e is known to overlap with
640 * the patch hunk; split it and pass blame to the parent.
641 */
642 static void blame_overlap(struct scoreboard *sb, struct blame_entry *e,
643 int tlno, int plno, int same,
644 struct origin *parent)
645 {
646 struct blame_entry split[3];
647
648 split_overlap(split, e, tlno, plno, same, parent);
649 if (split[1].suspect)
650 split_blame(sb, split, e);
651 decref_split(split);
652 }
653
654 /*
655 * Find the line number of the last line the target is suspected for.
656 */
657 static int find_last_in_target(struct scoreboard *sb, struct origin *target)
658 {
659 struct blame_entry *e;
660 int last_in_target = -1;
661
662 for (e = sb->ent; e; e = e->next) {
663 if (e->guilty || !same_suspect(e->suspect, target))
664 continue;
665 if (last_in_target < e->s_lno + e->num_lines)
666 last_in_target = e->s_lno + e->num_lines;
667 }
668 return last_in_target;
669 }
670
671 /*
672 * Process one hunk from the patch between the current suspect for
673 * blame_entry e and its parent. Find and split the overlap, and
674 * pass blame to the overlapping part to the parent.
675 */
676 static void blame_chunk(struct scoreboard *sb,
677 int tlno, int plno, int same,
678 struct origin *target, struct origin *parent)
679 {
680 struct blame_entry *e;
681
682 for (e = sb->ent; e; e = e->next) {
683 if (e->guilty || !same_suspect(e->suspect, target))
684 continue;
685 if (same <= e->s_lno)
686 continue;
687 if (tlno < e->s_lno + e->num_lines)
688 blame_overlap(sb, e, tlno, plno, same, parent);
689 }
690 }
691
692 struct blame_chunk_cb_data {
693 struct scoreboard *sb;
694 struct origin *target;
695 struct origin *parent;
696 long plno;
697 long tlno;
698 };
699
700 static void blame_chunk_cb(void *data, long same, long p_next, long t_next)
701 {
702 struct blame_chunk_cb_data *d = data;
703 blame_chunk(d->sb, d->tlno, d->plno, same, d->target, d->parent);
704 d->plno = p_next;
705 d->tlno = t_next;
706 }
707
708 /*
709 * We are looking at the origin 'target' and aiming to pass blame
710 * for the lines it is suspected to its parent. Run diff to find
711 * which lines came from parent and pass blame for them.
712 */
713 static int pass_blame_to_parent(struct scoreboard *sb,
714 struct origin *target,
715 struct origin *parent)
716 {
717 int last_in_target;
718 mmfile_t file_p, file_o;
719 struct blame_chunk_cb_data d = { sb, target, parent, 0, 0 };
720 xpparam_t xpp;
721 xdemitconf_t xecfg;
722
723 last_in_target = find_last_in_target(sb, target);
724 if (last_in_target < 0)
725 return 1; /* nothing remains for this target */
726
727 fill_origin_blob(parent, &file_p);
728 fill_origin_blob(target, &file_o);
729 num_get_patch++;
730
731 memset(&xpp, 0, sizeof(xpp));
732 xpp.flags = xdl_opts;
733 memset(&xecfg, 0, sizeof(xecfg));
734 xecfg.ctxlen = 0;
735 xdi_diff_hunks(&file_p, &file_o, blame_chunk_cb, &d, &xpp, &xecfg);
736 /* The rest (i.e. anything after tlno) are the same as the parent */
737 blame_chunk(sb, d.tlno, d.plno, last_in_target, target, parent);
738
739 return 0;
740 }
741
742 /*
743 * The lines in blame_entry after splitting blames many times can become
744 * very small and trivial, and at some point it becomes pointless to
745 * blame the parents. E.g. "\t\t}\n\t}\n\n" appears everywhere in any
746 * ordinary C program, and it is not worth to say it was copied from
747 * totally unrelated file in the parent.
748 *
749 * Compute how trivial the lines in the blame_entry are.
750 */
751 static unsigned ent_score(struct scoreboard *sb, struct blame_entry *e)
752 {
753 unsigned score;
754 const char *cp, *ep;
755
756 if (e->score)
757 return e->score;
758
759 score = 1;
760 cp = nth_line(sb, e->lno);
761 ep = nth_line(sb, e->lno + e->num_lines);
762 while (cp < ep) {
763 unsigned ch = *((unsigned char *)cp);
764 if (isalnum(ch))
765 score++;
766 cp++;
767 }
768 e->score = score;
769 return score;
770 }
771
772 /*
773 * best_so_far[] and this[] are both a split of an existing blame_entry
774 * that passes blame to the parent. Maintain best_so_far the best split
775 * so far, by comparing this and best_so_far and copying this into
776 * bst_so_far as needed.
777 */
778 static void copy_split_if_better(struct scoreboard *sb,
779 struct blame_entry *best_so_far,
780 struct blame_entry *this)
781 {
782 int i;
783
784 if (!this[1].suspect)
785 return;
786 if (best_so_far[1].suspect) {
787 if (ent_score(sb, &this[1]) < ent_score(sb, &best_so_far[1]))
788 return;
789 }
790
791 for (i = 0; i < 3; i++)
792 origin_incref(this[i].suspect);
793 decref_split(best_so_far);
794 memcpy(best_so_far, this, sizeof(struct blame_entry [3]));
795 }
796
797 /*
798 * We are looking at a part of the final image represented by
799 * ent (tlno and same are offset by ent->s_lno).
800 * tlno is where we are looking at in the final image.
801 * up to (but not including) same match preimage.
802 * plno is where we are looking at in the preimage.
803 *
804 * <-------------- final image ---------------------->
805 * <------ent------>
806 * ^tlno ^same
807 * <---------preimage----->
808 * ^plno
809 *
810 * All line numbers are 0-based.
811 */
812 static void handle_split(struct scoreboard *sb,
813 struct blame_entry *ent,
814 int tlno, int plno, int same,
815 struct origin *parent,
816 struct blame_entry *split)
817 {
818 if (ent->num_lines <= tlno)
819 return;
820 if (tlno < same) {
821 struct blame_entry this[3];
822 tlno += ent->s_lno;
823 same += ent->s_lno;
824 split_overlap(this, ent, tlno, plno, same, parent);
825 copy_split_if_better(sb, split, this);
826 decref_split(this);
827 }
828 }
829
830 struct handle_split_cb_data {
831 struct scoreboard *sb;
832 struct blame_entry *ent;
833 struct origin *parent;
834 struct blame_entry *split;
835 long plno;
836 long tlno;
837 };
838
839 static void handle_split_cb(void *data, long same, long p_next, long t_next)
840 {
841 struct handle_split_cb_data *d = data;
842 handle_split(d->sb, d->ent, d->tlno, d->plno, same, d->parent, d->split);
843 d->plno = p_next;
844 d->tlno = t_next;
845 }
846
847 /*
848 * Find the lines from parent that are the same as ent so that
849 * we can pass blames to it. file_p has the blob contents for
850 * the parent.
851 */
852 static void find_copy_in_blob(struct scoreboard *sb,
853 struct blame_entry *ent,
854 struct origin *parent,
855 struct blame_entry *split,
856 mmfile_t *file_p)
857 {
858 const char *cp;
859 int cnt;
860 mmfile_t file_o;
861 struct handle_split_cb_data d = { sb, ent, parent, split, 0, 0 };
862 xpparam_t xpp;
863 xdemitconf_t xecfg;
864
865 /*
866 * Prepare mmfile that contains only the lines in ent.
867 */
868 cp = nth_line(sb, ent->lno);
869 file_o.ptr = (char*) cp;
870 cnt = ent->num_lines;
871
872 while (cnt && cp < sb->final_buf + sb->final_buf_size) {
873 if (*cp++ == '\n')
874 cnt--;
875 }
876 file_o.size = cp - file_o.ptr;
877
878 /*
879 * file_o is a part of final image we are annotating.
880 * file_p partially may match that image.
881 */
882 memset(&xpp, 0, sizeof(xpp));
883 xpp.flags = xdl_opts;
884 memset(&xecfg, 0, sizeof(xecfg));
885 xecfg.ctxlen = 1;
886 memset(split, 0, sizeof(struct blame_entry [3]));
887 xdi_diff_hunks(file_p, &file_o, handle_split_cb, &d, &xpp, &xecfg);
888 /* remainder, if any, all match the preimage */
889 handle_split(sb, ent, d.tlno, d.plno, ent->num_lines, parent, split);
890 }
891
892 /*
893 * See if lines currently target is suspected for can be attributed to
894 * parent.
895 */
896 static int find_move_in_parent(struct scoreboard *sb,
897 struct origin *target,
898 struct origin *parent)
899 {
900 int last_in_target, made_progress;
901 struct blame_entry *e, split[3];
902 mmfile_t file_p;
903
904 last_in_target = find_last_in_target(sb, target);
905 if (last_in_target < 0)
906 return 1; /* nothing remains for this target */
907
908 fill_origin_blob(parent, &file_p);
909 if (!file_p.ptr)
910 return 0;
911
912 made_progress = 1;
913 while (made_progress) {
914 made_progress = 0;
915 for (e = sb->ent; e; e = e->next) {
916 if (e->guilty || !same_suspect(e->suspect, target) ||
917 ent_score(sb, e) < blame_move_score)
918 continue;
919 find_copy_in_blob(sb, e, parent, split, &file_p);
920 if (split[1].suspect &&
921 blame_move_score < ent_score(sb, &split[1])) {
922 split_blame(sb, split, e);
923 made_progress = 1;
924 }
925 decref_split(split);
926 }
927 }
928 return 0;
929 }
930
931 struct blame_list {
932 struct blame_entry *ent;
933 struct blame_entry split[3];
934 };
935
936 /*
937 * Count the number of entries the target is suspected for,
938 * and prepare a list of entry and the best split.
939 */
940 static struct blame_list *setup_blame_list(struct scoreboard *sb,
941 struct origin *target,
942 int min_score,
943 int *num_ents_p)
944 {
945 struct blame_entry *e;
946 int num_ents, i;
947 struct blame_list *blame_list = NULL;
948
949 for (e = sb->ent, num_ents = 0; e; e = e->next)
950 if (!e->scanned && !e->guilty &&
951 same_suspect(e->suspect, target) &&
952 min_score < ent_score(sb, e))
953 num_ents++;
954 if (num_ents) {
955 blame_list = xcalloc(num_ents, sizeof(struct blame_list));
956 for (e = sb->ent, i = 0; e; e = e->next)
957 if (!e->scanned && !e->guilty &&
958 same_suspect(e->suspect, target) &&
959 min_score < ent_score(sb, e))
960 blame_list[i++].ent = e;
961 }
962 *num_ents_p = num_ents;
963 return blame_list;
964 }
965
966 /*
967 * Reset the scanned status on all entries.
968 */
969 static void reset_scanned_flag(struct scoreboard *sb)
970 {
971 struct blame_entry *e;
972 for (e = sb->ent; e; e = e->next)
973 e->scanned = 0;
974 }
975
976 /*
977 * For lines target is suspected for, see if we can find code movement
978 * across file boundary from the parent commit. porigin is the path
979 * in the parent we already tried.
980 */
981 static int find_copy_in_parent(struct scoreboard *sb,
982 struct origin *target,
983 struct commit *parent,
984 struct origin *porigin,
985 int opt)
986 {
987 struct diff_options diff_opts;
988 const char *paths[1];
989 int i, j;
990 int retval;
991 struct blame_list *blame_list;
992 int num_ents;
993
994 blame_list = setup_blame_list(sb, target, blame_copy_score, &num_ents);
995 if (!blame_list)
996 return 1; /* nothing remains for this target */
997
998 diff_setup(&diff_opts);
999 DIFF_OPT_SET(&diff_opts, RECURSIVE);
1000 diff_opts.output_format = DIFF_FORMAT_NO_OUTPUT;
1001
1002 paths[0] = NULL;
1003 diff_tree_setup_paths(paths, &diff_opts);
1004 if (diff_setup_done(&diff_opts) < 0)
1005 die("diff-setup");
1006
1007 /* Try "find copies harder" on new path if requested;
1008 * we do not want to use diffcore_rename() actually to
1009 * match things up; find_copies_harder is set only to
1010 * force diff_tree_sha1() to feed all filepairs to diff_queue,
1011 * and this code needs to be after diff_setup_done(), which
1012 * usually makes find-copies-harder imply copy detection.
1013 */
1014 if ((opt & PICKAXE_BLAME_COPY_HARDEST)
1015 || ((opt & PICKAXE_BLAME_COPY_HARDER)
1016 && (!porigin || strcmp(target->path, porigin->path))))
1017 DIFF_OPT_SET(&diff_opts, FIND_COPIES_HARDER);
1018
1019 if (is_null_sha1(target->commit->object.sha1))
1020 do_diff_cache(parent->tree->object.sha1, &diff_opts);
1021 else
1022 diff_tree_sha1(parent->tree->object.sha1,
1023 target->commit->tree->object.sha1,
1024 "", &diff_opts);
1025
1026 if (!DIFF_OPT_TST(&diff_opts, FIND_COPIES_HARDER))
1027 diffcore_std(&diff_opts);
1028
1029 retval = 0;
1030 while (1) {
1031 int made_progress = 0;
1032
1033 for (i = 0; i < diff_queued_diff.nr; i++) {
1034 struct diff_filepair *p = diff_queued_diff.queue[i];
1035 struct origin *norigin;
1036 mmfile_t file_p;
1037 struct blame_entry this[3];
1038
1039 if (!DIFF_FILE_VALID(p->one))
1040 continue; /* does not exist in parent */
1041 if (S_ISGITLINK(p->one->mode))
1042 continue; /* ignore git links */
1043 if (porigin && !strcmp(p->one->path, porigin->path))
1044 /* find_move already dealt with this path */
1045 continue;
1046
1047 norigin = get_origin(sb, parent, p->one->path);
1048 hashcpy(norigin->blob_sha1, p->one->sha1);
1049 fill_origin_blob(norigin, &file_p);
1050 if (!file_p.ptr)
1051 continue;
1052
1053 for (j = 0; j < num_ents; j++) {
1054 find_copy_in_blob(sb, blame_list[j].ent,
1055 norigin, this, &file_p);
1056 copy_split_if_better(sb, blame_list[j].split,
1057 this);
1058 decref_split(this);
1059 }
1060 origin_decref(norigin);
1061 }
1062
1063 for (j = 0; j < num_ents; j++) {
1064 struct blame_entry *split = blame_list[j].split;
1065 if (split[1].suspect &&
1066 blame_copy_score < ent_score(sb, &split[1])) {
1067 split_blame(sb, split, blame_list[j].ent);
1068 made_progress = 1;
1069 }
1070 else
1071 blame_list[j].ent->scanned = 1;
1072 decref_split(split);
1073 }
1074 free(blame_list);
1075
1076 if (!made_progress)
1077 break;
1078 blame_list = setup_blame_list(sb, target, blame_copy_score, &num_ents);
1079 if (!blame_list) {
1080 retval = 1;
1081 break;
1082 }
1083 }
1084 reset_scanned_flag(sb);
1085 diff_flush(&diff_opts);
1086 diff_tree_release_paths(&diff_opts);
1087 return retval;
1088 }
1089
1090 /*
1091 * The blobs of origin and porigin exactly match, so everything
1092 * origin is suspected for can be blamed on the parent.
1093 */
1094 static void pass_whole_blame(struct scoreboard *sb,
1095 struct origin *origin, struct origin *porigin)
1096 {
1097 struct blame_entry *e;
1098
1099 if (!porigin->file.ptr && origin->file.ptr) {
1100 /* Steal its file */
1101 porigin->file = origin->file;
1102 origin->file.ptr = NULL;
1103 }
1104 for (e = sb->ent; e; e = e->next) {
1105 if (!same_suspect(e->suspect, origin))
1106 continue;
1107 origin_incref(porigin);
1108 origin_decref(e->suspect);
1109 e->suspect = porigin;
1110 }
1111 }
1112
1113 /*
1114 * We pass blame from the current commit to its parents. We keep saying
1115 * "parent" (and "porigin"), but what we mean is to find scapegoat to
1116 * exonerate ourselves.
1117 */
1118 static struct commit_list *first_scapegoat(struct rev_info *revs, struct commit *commit)
1119 {
1120 if (!reverse)
1121 return commit->parents;
1122 return lookup_decoration(&revs->children, &commit->object);
1123 }
1124
1125 static int num_scapegoats(struct rev_info *revs, struct commit *commit)
1126 {
1127 int cnt;
1128 struct commit_list *l = first_scapegoat(revs, commit);
1129 for (cnt = 0; l; l = l->next)
1130 cnt++;
1131 return cnt;
1132 }
1133
1134 #define MAXSG 16
1135
1136 static void pass_blame(struct scoreboard *sb, struct origin *origin, int opt)
1137 {
1138 struct rev_info *revs = sb->revs;
1139 int i, pass, num_sg;
1140 struct commit *commit = origin->commit;
1141 struct commit_list *sg;
1142 struct origin *sg_buf[MAXSG];
1143 struct origin *porigin, **sg_origin = sg_buf;
1144
1145 num_sg = num_scapegoats(revs, commit);
1146 if (!num_sg)
1147 goto finish;
1148 else if (num_sg < ARRAY_SIZE(sg_buf))
1149 memset(sg_buf, 0, sizeof(sg_buf));
1150 else
1151 sg_origin = xcalloc(num_sg, sizeof(*sg_origin));
1152
1153 /*
1154 * The first pass looks for unrenamed path to optimize for
1155 * common cases, then we look for renames in the second pass.
1156 */
1157 for (pass = 0; pass < 2; pass++) {
1158 struct origin *(*find)(struct scoreboard *,
1159 struct commit *, struct origin *);
1160 find = pass ? find_rename : find_origin;
1161
1162 for (i = 0, sg = first_scapegoat(revs, commit);
1163 i < num_sg && sg;
1164 sg = sg->next, i++) {
1165 struct commit *p = sg->item;
1166 int j, same;
1167
1168 if (sg_origin[i])
1169 continue;
1170 if (parse_commit(p))
1171 continue;
1172 porigin = find(sb, p, origin);
1173 if (!porigin)
1174 continue;
1175 if (!hashcmp(porigin->blob_sha1, origin->blob_sha1)) {
1176 pass_whole_blame(sb, origin, porigin);
1177 origin_decref(porigin);
1178 goto finish;
1179 }
1180 for (j = same = 0; j < i; j++)
1181 if (sg_origin[j] &&
1182 !hashcmp(sg_origin[j]->blob_sha1,
1183 porigin->blob_sha1)) {
1184 same = 1;
1185 break;
1186 }
1187 if (!same)
1188 sg_origin[i] = porigin;
1189 else
1190 origin_decref(porigin);
1191 }
1192 }
1193
1194 num_commits++;
1195 for (i = 0, sg = first_scapegoat(revs, commit);
1196 i < num_sg && sg;
1197 sg = sg->next, i++) {
1198 struct origin *porigin = sg_origin[i];
1199 if (!porigin)
1200 continue;
1201 if (pass_blame_to_parent(sb, origin, porigin))
1202 goto finish;
1203 }
1204
1205 /*
1206 * Optionally find moves in parents' files.
1207 */
1208 if (opt & PICKAXE_BLAME_MOVE)
1209 for (i = 0, sg = first_scapegoat(revs, commit);
1210 i < num_sg && sg;
1211 sg = sg->next, i++) {
1212 struct origin *porigin = sg_origin[i];
1213 if (!porigin)
1214 continue;
1215 if (find_move_in_parent(sb, origin, porigin))
1216 goto finish;
1217 }
1218
1219 /*
1220 * Optionally find copies from parents' files.
1221 */
1222 if (opt & PICKAXE_BLAME_COPY)
1223 for (i = 0, sg = first_scapegoat(revs, commit);
1224 i < num_sg && sg;
1225 sg = sg->next, i++) {
1226 struct origin *porigin = sg_origin[i];
1227 if (find_copy_in_parent(sb, origin, sg->item,
1228 porigin, opt))
1229 goto finish;
1230 }
1231
1232 finish:
1233 for (i = 0; i < num_sg; i++) {
1234 if (sg_origin[i]) {
1235 drop_origin_blob(sg_origin[i]);
1236 origin_decref(sg_origin[i]);
1237 }
1238 }
1239 drop_origin_blob(origin);
1240 if (sg_buf != sg_origin)
1241 free(sg_origin);
1242 }
1243
1244 /*
1245 * Information on commits, used for output.
1246 */
1247 struct commit_info
1248 {
1249 const char *author;
1250 const char *author_mail;
1251 unsigned long author_time;
1252 const char *author_tz;
1253
1254 /* filled only when asked for details */
1255 const char *committer;
1256 const char *committer_mail;
1257 unsigned long committer_time;
1258 const char *committer_tz;
1259
1260 const char *summary;
1261 };
1262
1263 /*
1264 * Parse author/committer line in the commit object buffer
1265 */
1266 static void get_ac_line(const char *inbuf, const char *what,
1267 int bufsz, char *person, const char **mail,
1268 unsigned long *time, const char **tz)
1269 {
1270 int len, tzlen, maillen;
1271 char *tmp, *endp, *timepos;
1272
1273 tmp = strstr(inbuf, what);
1274 if (!tmp)
1275 goto error_out;
1276 tmp += strlen(what);
1277 endp = strchr(tmp, '\n');
1278 if (!endp)
1279 len = strlen(tmp);
1280 else
1281 len = endp - tmp;
1282 if (bufsz <= len) {
1283 error_out:
1284 /* Ugh */
1285 *mail = *tz = "(unknown)";
1286 *time = 0;
1287 return;
1288 }
1289 memcpy(person, tmp, len);
1290
1291 tmp = person;
1292 tmp += len;
1293 *tmp = 0;
1294 while (*tmp != ' ')
1295 tmp--;
1296 *tz = tmp+1;
1297 tzlen = (person+len)-(tmp+1);
1298
1299 *tmp = 0;
1300 while (*tmp != ' ')
1301 tmp--;
1302 *time = strtoul(tmp, NULL, 10);
1303 timepos = tmp;
1304
1305 *tmp = 0;
1306 while (*tmp != ' ')
1307 tmp--;
1308 *mail = tmp + 1;
1309 *tmp = 0;
1310 maillen = timepos - tmp;
1311
1312 if (!mailmap.nr)
1313 return;
1314
1315 /*
1316 * mailmap expansion may make the name longer.
1317 * make room by pushing stuff down.
1318 */
1319 tmp = person + bufsz - (tzlen + 1);
1320 memmove(tmp, *tz, tzlen);
1321 tmp[tzlen] = 0;
1322 *tz = tmp;
1323
1324 tmp = tmp - (maillen + 1);
1325 memmove(tmp, *mail, maillen);
1326 tmp[maillen] = 0;
1327 *mail = tmp;
1328
1329 /*
1330 * Now, convert e-mail using mailmap
1331 */
1332 map_email(&mailmap, tmp + 1, person, tmp-person-1);
1333 }
1334
1335 static void get_commit_info(struct commit *commit,
1336 struct commit_info *ret,
1337 int detailed)
1338 {
1339 int len;
1340 char *tmp, *endp, *reencoded, *message;
1341 static char author_buf[1024];
1342 static char committer_buf[1024];
1343 static char summary_buf[1024];
1344
1345 /*
1346 * We've operated without save_commit_buffer, so
1347 * we now need to populate them for output.
1348 */
1349 if (!commit->buffer) {
1350 enum object_type type;
1351 unsigned long size;
1352 commit->buffer =
1353 read_sha1_file(commit->object.sha1, &type, &size);
1354 if (!commit->buffer)
1355 die("Cannot read commit %s",
1356 sha1_to_hex(commit->object.sha1));
1357 }
1358 reencoded = reencode_commit_message(commit, NULL);
1359 message = reencoded ? reencoded : commit->buffer;
1360 ret->author = author_buf;
1361 get_ac_line(message, "\nauthor ",
1362 sizeof(author_buf), author_buf, &ret->author_mail,
1363 &ret->author_time, &ret->author_tz);
1364
1365 if (!detailed) {
1366 free(reencoded);
1367 return;
1368 }
1369
1370 ret->committer = committer_buf;
1371 get_ac_line(message, "\ncommitter ",
1372 sizeof(committer_buf), committer_buf, &ret->committer_mail,
1373 &ret->committer_time, &ret->committer_tz);
1374
1375 ret->summary = summary_buf;
1376 tmp = strstr(message, "\n\n");
1377 if (!tmp) {
1378 error_out:
1379 sprintf(summary_buf, "(%s)", sha1_to_hex(commit->object.sha1));
1380 free(reencoded);
1381 return;
1382 }
1383 tmp += 2;
1384 endp = strchr(tmp, '\n');
1385 if (!endp)
1386 endp = tmp + strlen(tmp);
1387 len = endp - tmp;
1388 if (len >= sizeof(summary_buf) || len == 0)
1389 goto error_out;
1390 memcpy(summary_buf, tmp, len);
1391 summary_buf[len] = 0;
1392 free(reencoded);
1393 }
1394
1395 /*
1396 * To allow LF and other nonportable characters in pathnames,
1397 * they are c-style quoted as needed.
1398 */
1399 static void write_filename_info(const char *path)
1400 {
1401 printf("filename ");
1402 write_name_quoted(path, stdout, '\n');
1403 }
1404
1405 /*
1406 * The blame_entry is found to be guilty for the range. Mark it
1407 * as such, and show it in incremental output.
1408 */
1409 static void found_guilty_entry(struct blame_entry *ent)
1410 {
1411 if (ent->guilty)
1412 return;
1413 ent->guilty = 1;
1414 if (incremental) {
1415 struct origin *suspect = ent->suspect;
1416
1417 printf("%s %d %d %d\n",
1418 sha1_to_hex(suspect->commit->object.sha1),
1419 ent->s_lno + 1, ent->lno + 1, ent->num_lines);
1420 if (!(suspect->commit->object.flags & METAINFO_SHOWN)) {
1421 struct commit_info ci;
1422 suspect->commit->object.flags |= METAINFO_SHOWN;
1423 get_commit_info(suspect->commit, &ci, 1);
1424 printf("author %s\n", ci.author);
1425 printf("author-mail %s\n", ci.author_mail);
1426 printf("author-time %lu\n", ci.author_time);
1427 printf("author-tz %s\n", ci.author_tz);
1428 printf("committer %s\n", ci.committer);
1429 printf("committer-mail %s\n", ci.committer_mail);
1430 printf("committer-time %lu\n", ci.committer_time);
1431 printf("committer-tz %s\n", ci.committer_tz);
1432 printf("summary %s\n", ci.summary);
1433 if (suspect->commit->object.flags & UNINTERESTING)
1434 printf("boundary\n");
1435 }
1436 write_filename_info(suspect->path);
1437 maybe_flush_or_die(stdout, "stdout");
1438 }
1439 }
1440
1441 /*
1442 * The main loop -- while the scoreboard has lines whose true origin
1443 * is still unknown, pick one blame_entry, and allow its current
1444 * suspect to pass blames to its parents.
1445 */
1446 static void assign_blame(struct scoreboard *sb, int opt)
1447 {
1448 struct rev_info *revs = sb->revs;
1449
1450 while (1) {
1451 struct blame_entry *ent;
1452 struct commit *commit;
1453 struct origin *suspect = NULL;
1454
1455 /* find one suspect to break down */
1456 for (ent = sb->ent; !suspect && ent; ent = ent->next)
1457 if (!ent->guilty)
1458 suspect = ent->suspect;
1459 if (!suspect)
1460 return; /* all done */
1461
1462 /*
1463 * We will use this suspect later in the loop,
1464 * so hold onto it in the meantime.
1465 */
1466 origin_incref(suspect);
1467 commit = suspect->commit;
1468 if (!commit->object.parsed)
1469 parse_commit(commit);
1470 if (reverse ||
1471 (!(commit->object.flags & UNINTERESTING) &&
1472 !(revs->max_age != -1 && commit->date < revs->max_age)))
1473 pass_blame(sb, suspect, opt);
1474 else {
1475 commit->object.flags |= UNINTERESTING;
1476 if (commit->object.parsed)
1477 mark_parents_uninteresting(commit);
1478 }
1479 /* treat root commit as boundary */
1480 if (!commit->parents && !show_root)
1481 commit->object.flags |= UNINTERESTING;
1482
1483 /* Take responsibility for the remaining entries */
1484 for (ent = sb->ent; ent; ent = ent->next)
1485 if (same_suspect(ent->suspect, suspect))
1486 found_guilty_entry(ent);
1487 origin_decref(suspect);
1488
1489 if (DEBUG) /* sanity */
1490 sanity_check_refcnt(sb);
1491 }
1492 }
1493
1494 static const char *format_time(unsigned long time, const char *tz_str,
1495 int show_raw_time)
1496 {
1497 static char time_buf[128];
1498 time_t t = time;
1499 int minutes, tz;
1500 struct tm *tm;
1501
1502 if (show_raw_time) {
1503 sprintf(time_buf, "%lu %s", time, tz_str);
1504 return time_buf;
1505 }
1506
1507 tz = atoi(tz_str);
1508 minutes = tz < 0 ? -tz : tz;
1509 minutes = (minutes / 100)*60 + (minutes % 100);
1510 minutes = tz < 0 ? -minutes : minutes;
1511 t = time + minutes * 60;
1512 tm = gmtime(&t);
1513
1514 strftime(time_buf, sizeof(time_buf), "%Y-%m-%d %H:%M:%S ", tm);
1515 strcat(time_buf, tz_str);
1516 return time_buf;
1517 }
1518
1519 #define OUTPUT_ANNOTATE_COMPAT 001
1520 #define OUTPUT_LONG_OBJECT_NAME 002
1521 #define OUTPUT_RAW_TIMESTAMP 004
1522 #define OUTPUT_PORCELAIN 010
1523 #define OUTPUT_SHOW_NAME 020
1524 #define OUTPUT_SHOW_NUMBER 040
1525 #define OUTPUT_SHOW_SCORE 0100
1526 #define OUTPUT_NO_AUTHOR 0200
1527
1528 static void emit_porcelain(struct scoreboard *sb, struct blame_entry *ent)
1529 {
1530 int cnt;
1531 const char *cp;
1532 struct origin *suspect = ent->suspect;
1533 char hex[41];
1534
1535 strcpy(hex, sha1_to_hex(suspect->commit->object.sha1));
1536 printf("%s%c%d %d %d\n",
1537 hex,
1538 ent->guilty ? ' ' : '*', // purely for debugging
1539 ent->s_lno + 1,
1540 ent->lno + 1,
1541 ent->num_lines);
1542 if (!(suspect->commit->object.flags & METAINFO_SHOWN)) {
1543 struct commit_info ci;
1544 suspect->commit->object.flags |= METAINFO_SHOWN;
1545 get_commit_info(suspect->commit, &ci, 1);
1546 printf("author %s\n", ci.author);
1547 printf("author-mail %s\n", ci.author_mail);
1548 printf("author-time %lu\n", ci.author_time);
1549 printf("author-tz %s\n", ci.author_tz);
1550 printf("committer %s\n", ci.committer);
1551 printf("committer-mail %s\n", ci.committer_mail);
1552 printf("committer-time %lu\n", ci.committer_time);
1553 printf("committer-tz %s\n", ci.committer_tz);
1554 write_filename_info(suspect->path);
1555 printf("summary %s\n", ci.summary);
1556 if (suspect->commit->object.flags & UNINTERESTING)
1557 printf("boundary\n");
1558 }
1559 else if (suspect->commit->object.flags & MORE_THAN_ONE_PATH)
1560 write_filename_info(suspect->path);
1561
1562 cp = nth_line(sb, ent->lno);
1563 for (cnt = 0; cnt < ent->num_lines; cnt++) {
1564 char ch;
1565 if (cnt)
1566 printf("%s %d %d\n", hex,
1567 ent->s_lno + 1 + cnt,
1568 ent->lno + 1 + cnt);
1569 putchar('\t');
1570 do {
1571 ch = *cp++;
1572 putchar(ch);
1573 } while (ch != '\n' &&
1574 cp < sb->final_buf + sb->final_buf_size);
1575 }
1576 }
1577
1578 static void emit_other(struct scoreboard *sb, struct blame_entry *ent, int opt)
1579 {
1580 int cnt;
1581 const char *cp;
1582 struct origin *suspect = ent->suspect;
1583 struct commit_info ci;
1584 char hex[41];
1585 int show_raw_time = !!(opt & OUTPUT_RAW_TIMESTAMP);
1586
1587 get_commit_info(suspect->commit, &ci, 1);
1588 strcpy(hex, sha1_to_hex(suspect->commit->object.sha1));
1589
1590 cp = nth_line(sb, ent->lno);
1591 for (cnt = 0; cnt < ent->num_lines; cnt++) {
1592 char ch;
1593 int length = (opt & OUTPUT_LONG_OBJECT_NAME) ? 40 : 8;
1594
1595 if (suspect->commit->object.flags & UNINTERESTING) {
1596 if (blank_boundary)
1597 memset(hex, ' ', length);
1598 else if (!(opt & OUTPUT_ANNOTATE_COMPAT)) {
1599 length--;
1600 putchar('^');
1601 }
1602 }
1603
1604 printf("%.*s", length, hex);
1605 if (opt & OUTPUT_ANNOTATE_COMPAT)
1606 printf("\t(%10s\t%10s\t%d)", ci.author,
1607 format_time(ci.author_time, ci.author_tz,
1608 show_raw_time),
1609 ent->lno + 1 + cnt);
1610 else {
1611 if (opt & OUTPUT_SHOW_SCORE)
1612 printf(" %*d %02d",
1613 max_score_digits, ent->score,
1614 ent->suspect->refcnt);
1615 if (opt & OUTPUT_SHOW_NAME)
1616 printf(" %-*.*s", longest_file, longest_file,
1617 suspect->path);
1618 if (opt & OUTPUT_SHOW_NUMBER)
1619 printf(" %*d", max_orig_digits,
1620 ent->s_lno + 1 + cnt);
1621
1622 if (!(opt & OUTPUT_NO_AUTHOR)) {
1623 int pad = longest_author - utf8_strwidth(ci.author);
1624 printf(" (%s%*s %10s",
1625 ci.author, pad, "",
1626 format_time(ci.author_time,
1627 ci.author_tz,
1628 show_raw_time));
1629 }
1630 printf(" %*d) ",
1631 max_digits, ent->lno + 1 + cnt);
1632 }
1633 do {
1634 ch = *cp++;
1635 putchar(ch);
1636 } while (ch != '\n' &&
1637 cp < sb->final_buf + sb->final_buf_size);
1638 }
1639 }
1640
1641 static void output(struct scoreboard *sb, int option)
1642 {
1643 struct blame_entry *ent;
1644
1645 if (option & OUTPUT_PORCELAIN) {
1646 for (ent = sb->ent; ent; ent = ent->next) {
1647 struct blame_entry *oth;
1648 struct origin *suspect = ent->suspect;
1649 struct commit *commit = suspect->commit;
1650 if (commit->object.flags & MORE_THAN_ONE_PATH)
1651 continue;
1652 for (oth = ent->next; oth; oth = oth->next) {
1653 if ((oth->suspect->commit != commit) ||
1654 !strcmp(oth->suspect->path, suspect->path))
1655 continue;
1656 commit->object.flags |= MORE_THAN_ONE_PATH;
1657 break;
1658 }
1659 }
1660 }
1661
1662 for (ent = sb->ent; ent; ent = ent->next) {
1663 if (option & OUTPUT_PORCELAIN)
1664 emit_porcelain(sb, ent);
1665 else {
1666 emit_other(sb, ent, option);
1667 }
1668 }
1669 }
1670
1671 /*
1672 * To allow quick access to the contents of nth line in the
1673 * final image, prepare an index in the scoreboard.
1674 */
1675 static int prepare_lines(struct scoreboard *sb)
1676 {
1677 const char *buf = sb->final_buf;
1678 unsigned long len = sb->final_buf_size;
1679 int num = 0, incomplete = 0, bol = 1;
1680
1681 if (len && buf[len-1] != '\n')
1682 incomplete++; /* incomplete line at the end */
1683 while (len--) {
1684 if (bol) {
1685 sb->lineno = xrealloc(sb->lineno,
1686 sizeof(int* ) * (num + 1));
1687 sb->lineno[num] = buf - sb->final_buf;
1688 bol = 0;
1689 }
1690 if (*buf++ == '\n') {
1691 num++;
1692 bol = 1;
1693 }
1694 }
1695 sb->lineno = xrealloc(sb->lineno,
1696 sizeof(int* ) * (num + incomplete + 1));
1697 sb->lineno[num + incomplete] = buf - sb->final_buf;
1698 sb->num_lines = num + incomplete;
1699 return sb->num_lines;
1700 }
1701
1702 /*
1703 * Add phony grafts for use with -S; this is primarily to
1704 * support git's cvsserver that wants to give a linear history
1705 * to its clients.
1706 */
1707 static int read_ancestry(const char *graft_file)
1708 {
1709 FILE *fp = fopen(graft_file, "r");
1710 char buf[1024];
1711 if (!fp)
1712 return -1;
1713 while (fgets(buf, sizeof(buf), fp)) {
1714 /* The format is just "Commit Parent1 Parent2 ...\n" */
1715 int len = strlen(buf);
1716 struct commit_graft *graft = read_graft_line(buf, len);
1717 if (graft)
1718 register_commit_graft(graft, 0);
1719 }
1720 fclose(fp);
1721 return 0;
1722 }
1723
1724 /*
1725 * How many columns do we need to show line numbers in decimal?
1726 */
1727 static int lineno_width(int lines)
1728 {
1729 int i, width;
1730
1731 for (width = 1, i = 10; i <= lines + 1; width++)
1732 i *= 10;
1733 return width;
1734 }
1735
1736 /*
1737 * How many columns do we need to show line numbers, authors,
1738 * and filenames?
1739 */
1740 static void find_alignment(struct scoreboard *sb, int *option)
1741 {
1742 int longest_src_lines = 0;
1743 int longest_dst_lines = 0;
1744 unsigned largest_score = 0;
1745 struct blame_entry *e;
1746
1747 for (e = sb->ent; e; e = e->next) {
1748 struct origin *suspect = e->suspect;
1749 struct commit_info ci;
1750 int num;
1751
1752 if (strcmp(suspect->path, sb->path))
1753 *option |= OUTPUT_SHOW_NAME;
1754 num = strlen(suspect->path);
1755 if (longest_file < num)
1756 longest_file = num;
1757 if (!(suspect->commit->object.flags & METAINFO_SHOWN)) {
1758 suspect->commit->object.flags |= METAINFO_SHOWN;
1759 get_commit_info(suspect->commit, &ci, 1);
1760 num = utf8_strwidth(ci.author);
1761 if (longest_author < num)
1762 longest_author = num;
1763 }
1764 num = e->s_lno + e->num_lines;
1765 if (longest_src_lines < num)
1766 longest_src_lines = num;
1767 num = e->lno + e->num_lines;
1768 if (longest_dst_lines < num)
1769 longest_dst_lines = num;
1770 if (largest_score < ent_score(sb, e))
1771 largest_score = ent_score(sb, e);
1772 }
1773 max_orig_digits = lineno_width(longest_src_lines);
1774 max_digits = lineno_width(longest_dst_lines);
1775 max_score_digits = lineno_width(largest_score);
1776 }
1777
1778 /*
1779 * For debugging -- origin is refcounted, and this asserts that
1780 * we do not underflow.
1781 */
1782 static void sanity_check_refcnt(struct scoreboard *sb)
1783 {
1784 int baa = 0;
1785 struct blame_entry *ent;
1786
1787 for (ent = sb->ent; ent; ent = ent->next) {
1788 /* Nobody should have zero or negative refcnt */
1789 if (ent->suspect->refcnt <= 0) {
1790 fprintf(stderr, "%s in %s has negative refcnt %d\n",
1791 ent->suspect->path,
1792 sha1_to_hex(ent->suspect->commit->object.sha1),
1793 ent->suspect->refcnt);
1794 baa = 1;
1795 }
1796 }
1797 for (ent = sb->ent; ent; ent = ent->next) {
1798 /* Mark the ones that haven't been checked */
1799 if (0 < ent->suspect->refcnt)
1800 ent->suspect->refcnt = -ent->suspect->refcnt;
1801 }
1802 for (ent = sb->ent; ent; ent = ent->next) {
1803 /*
1804 * ... then pick each and see if they have the the
1805 * correct refcnt.
1806 */
1807 int found;
1808 struct blame_entry *e;
1809 struct origin *suspect = ent->suspect;
1810
1811 if (0 < suspect->refcnt)
1812 continue;
1813 suspect->refcnt = -suspect->refcnt; /* Unmark */
1814 for (found = 0, e = sb->ent; e; e = e->next) {
1815 if (e->suspect != suspect)
1816 continue;
1817 found++;
1818 }
1819 if (suspect->refcnt != found) {
1820 fprintf(stderr, "%s in %s has refcnt %d, not %d\n",
1821 ent->suspect->path,
1822 sha1_to_hex(ent->suspect->commit->object.sha1),
1823 ent->suspect->refcnt, found);
1824 baa = 2;
1825 }
1826 }
1827 if (baa) {
1828 int opt = 0160;
1829 find_alignment(sb, &opt);
1830 output(sb, opt);
1831 die("Baa %d!", baa);
1832 }
1833 }
1834
1835 /*
1836 * Used for the command line parsing; check if the path exists
1837 * in the working tree.
1838 */
1839 static int has_string_in_work_tree(const char *path)
1840 {
1841 struct stat st;
1842 return !lstat(path, &st);
1843 }
1844
1845 static unsigned parse_score(const char *arg)
1846 {
1847 char *end;
1848 unsigned long score = strtoul(arg, &end, 10);
1849 if (*end)
1850 return 0;
1851 return score;
1852 }
1853
1854 static const char *add_prefix(const char *prefix, const char *path)
1855 {
1856 return prefix_path(prefix, prefix ? strlen(prefix) : 0, path);
1857 }
1858
1859 /*
1860 * Parsing of (comma separated) one item in the -L option
1861 */
1862 static const char *parse_loc(const char *spec,
1863 struct scoreboard *sb, long lno,
1864 long begin, long *ret)
1865 {
1866 char *term;
1867 const char *line;
1868 long num;
1869 int reg_error;
1870 regex_t regexp;
1871 regmatch_t match[1];
1872
1873 /* Allow "-L <something>,+20" to mean starting at <something>
1874 * for 20 lines, or "-L <something>,-5" for 5 lines ending at
1875 * <something>.
1876 */
1877 if (1 < begin && (spec[0] == '+' || spec[0] == '-')) {
1878 num = strtol(spec + 1, &term, 10);
1879 if (term != spec + 1) {
1880 if (spec[0] == '-')
1881 num = 0 - num;
1882 if (0 < num)
1883 *ret = begin + num - 2;
1884 else if (!num)
1885 *ret = begin;
1886 else
1887 *ret = begin + num;
1888 return term;
1889 }
1890 return spec;
1891 }
1892 num = strtol(spec, &term, 10);
1893 if (term != spec) {
1894 *ret = num;
1895 return term;
1896 }
1897 if (spec[0] != '/')
1898 return spec;
1899
1900 /* it could be a regexp of form /.../ */
1901 for (term = (char*) spec + 1; *term && *term != '/'; term++) {
1902 if (*term == '\\')
1903 term++;
1904 }
1905 if (*term != '/')
1906 return spec;
1907
1908 /* try [spec+1 .. term-1] as regexp */
1909 *term = 0;
1910 begin--; /* input is in human terms */
1911 line = nth_line(sb, begin);
1912
1913 if (!(reg_error = regcomp(&regexp, spec + 1, REG_NEWLINE)) &&
1914 !(reg_error = regexec(&regexp, line, 1, match, 0))) {
1915 const char *cp = line + match[0].rm_so;
1916 const char *nline;
1917
1918 while (begin++ < lno) {
1919 nline = nth_line(sb, begin);
1920 if (line <= cp && cp < nline)
1921 break;
1922 line = nline;
1923 }
1924 *ret = begin;
1925 regfree(&regexp);
1926 *term++ = '/';
1927 return term;
1928 }
1929 else {
1930 char errbuf[1024];
1931 regerror(reg_error, &regexp, errbuf, 1024);
1932 die("-L parameter '%s': %s", spec + 1, errbuf);
1933 }
1934 }
1935
1936 /*
1937 * Parsing of -L option
1938 */
1939 static void prepare_blame_range(struct scoreboard *sb,
1940 const char *bottomtop,
1941 long lno,
1942 long *bottom, long *top)
1943 {
1944 const char *term;
1945
1946 term = parse_loc(bottomtop, sb, lno, 1, bottom);
1947 if (*term == ',') {
1948 term = parse_loc(term + 1, sb, lno, *bottom + 1, top);
1949 if (*term)
1950 usage(blame_usage);
1951 }
1952 if (*term)
1953 usage(blame_usage);
1954 }
1955
1956 static int git_blame_config(const char *var, const char *value, void *cb)
1957 {
1958 if (!strcmp(var, "blame.showroot")) {
1959 show_root = git_config_bool(var, value);
1960 return 0;
1961 }
1962 if (!strcmp(var, "blame.blankboundary")) {
1963 blank_boundary = git_config_bool(var, value);
1964 return 0;
1965 }
1966 return git_default_config(var, value, cb);
1967 }
1968
1969 /*
1970 * Prepare a dummy commit that represents the work tree (or staged) item.
1971 * Note that annotating work tree item never works in the reverse.
1972 */
1973 static struct commit *fake_working_tree_commit(const char *path, const char *contents_from)
1974 {
1975 struct commit *commit;
1976 struct origin *origin;
1977 unsigned char head_sha1[20];
1978 struct strbuf buf = STRBUF_INIT;
1979 const char *ident;
1980 time_t now;
1981 int size, len;
1982 struct cache_entry *ce;
1983 unsigned mode;
1984
1985 if (get_sha1("HEAD", head_sha1))
1986 die("No such ref: HEAD");
1987
1988 time(&now);
1989 commit = xcalloc(1, sizeof(*commit));
1990 commit->parents = xcalloc(1, sizeof(*commit->parents));
1991 commit->parents->item = lookup_commit_reference(head_sha1);
1992 commit->object.parsed = 1;
1993 commit->date = now;
1994 commit->object.type = OBJ_COMMIT;
1995
1996 origin = make_origin(commit, path);
1997
1998 if (!contents_from || strcmp("-", contents_from)) {
1999 struct stat st;
2000 const char *read_from;
2001
2002 if (contents_from) {
2003 if (stat(contents_from, &st) < 0)
2004 die("Cannot stat %s", contents_from);
2005 read_from = contents_from;
2006 }
2007 else {
2008 if (lstat(path, &st) < 0)
2009 die("Cannot lstat %s", path);
2010 read_from = path;
2011 }
2012 mode = canon_mode(st.st_mode);
2013 switch (st.st_mode & S_IFMT) {
2014 case S_IFREG:
2015 if (strbuf_read_file(&buf, read_from, st.st_size) != st.st_size)
2016 die("cannot open or read %s", read_from);
2017 break;
2018 case S_IFLNK:
2019 if (strbuf_readlink(&buf, read_from, st.st_size) < 0)
2020 die("cannot readlink %s", read_from);
2021 break;
2022 default:
2023 die("unsupported file type %s", read_from);
2024 }
2025 }
2026 else {
2027 /* Reading from stdin */
2028 contents_from = "standard input";
2029 mode = 0;
2030 if (strbuf_read(&buf, 0, 0) < 0)
2031 die("read error %s from stdin", strerror(errno));
2032 }
2033 convert_to_git(path, buf.buf, buf.len, &buf, 0);
2034 origin->file.ptr = buf.buf;
2035 origin->file.size = buf.len;
2036 pretend_sha1_file(buf.buf, buf.len, OBJ_BLOB, origin->blob_sha1);
2037 commit->util = origin;
2038
2039 /*
2040 * Read the current index, replace the path entry with
2041 * origin->blob_sha1 without mucking with its mode or type
2042 * bits; we are not going to write this index out -- we just
2043 * want to run "diff-index --cached".
2044 */
2045 discard_cache();
2046 read_cache();
2047
2048 len = strlen(path);
2049 if (!mode) {
2050 int pos = cache_name_pos(path, len);
2051 if (0 <= pos)
2052 mode = active_cache[pos]->ce_mode;
2053 else
2054 /* Let's not bother reading from HEAD tree */
2055 mode = S_IFREG | 0644;
2056 }
2057 size = cache_entry_size(len);
2058 ce = xcalloc(1, size);
2059 hashcpy(ce->sha1, origin->blob_sha1);
2060 memcpy(ce->name, path, len);
2061 ce->ce_flags = create_ce_flags(len, 0);
2062 ce->ce_mode = create_ce_mode(mode);
2063 add_cache_entry(ce, ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE);
2064
2065 /*
2066 * We are not going to write this out, so this does not matter
2067 * right now, but someday we might optimize diff-index --cached
2068 * with cache-tree information.
2069 */
2070 cache_tree_invalidate_path(active_cache_tree, path);
2071
2072 commit->buffer = xmalloc(400);
2073 ident = fmt_ident("Not Committed Yet", "not.committed.yet", NULL, 0);
2074 snprintf(commit->buffer, 400,
2075 "tree 0000000000000000000000000000000000000000\n"
2076 "parent %s\n"
2077 "author %s\n"
2078 "committer %s\n\n"
2079 "Version of %s from %s\n",
2080 sha1_to_hex(head_sha1),
2081 ident, ident, path, contents_from ? contents_from : path);
2082 return commit;
2083 }
2084
2085 static const char *prepare_final(struct scoreboard *sb)
2086 {
2087 int i;
2088 const char *final_commit_name = NULL;
2089 struct rev_info *revs = sb->revs;
2090
2091 /*
2092 * There must be one and only one positive commit in the
2093 * revs->pending array.
2094 */
2095 for (i = 0; i < revs->pending.nr; i++) {
2096 struct object *obj = revs->pending.objects[i].item;
2097 if (obj->flags & UNINTERESTING)
2098 continue;
2099 while (obj->type == OBJ_TAG)
2100 obj = deref_tag(obj, NULL, 0);
2101 if (obj->type != OBJ_COMMIT)
2102 die("Non commit %s?", revs->pending.objects[i].name);
2103 if (sb->final)
2104 die("More than one commit to dig from %s and %s?",
2105 revs->pending.objects[i].name,
2106 final_commit_name);
2107 sb->final = (struct commit *) obj;
2108 final_commit_name = revs->pending.objects[i].name;
2109 }
2110 return final_commit_name;
2111 }
2112
2113 static const char *prepare_initial(struct scoreboard *sb)
2114 {
2115 int i;
2116 const char *final_commit_name = NULL;
2117 struct rev_info *revs = sb->revs;
2118
2119 /*
2120 * There must be one and only one negative commit, and it must be
2121 * the boundary.
2122 */
2123 for (i = 0; i < revs->pending.nr; i++) {
2124 struct object *obj = revs->pending.objects[i].item;
2125 if (!(obj->flags & UNINTERESTING))
2126 continue;
2127 while (obj->type == OBJ_TAG)
2128 obj = deref_tag(obj, NULL, 0);
2129 if (obj->type != OBJ_COMMIT)
2130 die("Non commit %s?", revs->pending.objects[i].name);
2131 if (sb->final)
2132 die("More than one commit to dig down to %s and %s?",
2133 revs->pending.objects[i].name,
2134 final_commit_name);
2135 sb->final = (struct commit *) obj;
2136 final_commit_name = revs->pending.objects[i].name;
2137 }
2138 if (!final_commit_name)
2139 die("No commit to dig down to?");
2140 return final_commit_name;
2141 }
2142
2143 static int blame_copy_callback(const struct option *option, const char *arg, int unset)
2144 {
2145 int *opt = option->value;
2146
2147 /*
2148 * -C enables copy from removed files;
2149 * -C -C enables copy from existing files, but only
2150 * when blaming a new file;
2151 * -C -C -C enables copy from existing files for
2152 * everybody
2153 */
2154 if (*opt & PICKAXE_BLAME_COPY_HARDER)
2155 *opt |= PICKAXE_BLAME_COPY_HARDEST;
2156 if (*opt & PICKAXE_BLAME_COPY)
2157 *opt |= PICKAXE_BLAME_COPY_HARDER;
2158 *opt |= PICKAXE_BLAME_COPY | PICKAXE_BLAME_MOVE;
2159
2160 if (arg)
2161 blame_copy_score = parse_score(arg);
2162 return 0;
2163 }
2164
2165 static int blame_move_callback(const struct option *option, const char *arg, int unset)
2166 {
2167 int *opt = option->value;
2168
2169 *opt |= PICKAXE_BLAME_MOVE;
2170
2171 if (arg)
2172 blame_move_score = parse_score(arg);
2173 return 0;
2174 }
2175
2176 static int blame_bottomtop_callback(const struct option *option, const char *arg, int unset)
2177 {
2178 const char **bottomtop = option->value;
2179 if (!arg)
2180 return -1;
2181 if (*bottomtop)
2182 die("More than one '-L n,m' option given");
2183 *bottomtop = arg;
2184 return 0;
2185 }
2186
2187 int cmd_blame(int argc, const char **argv, const char *prefix)
2188 {
2189 struct rev_info revs;
2190 const char *path;
2191 struct scoreboard sb;
2192 struct origin *o;
2193 struct blame_entry *ent;
2194 long dashdash_pos, bottom, top, lno;
2195 const char *final_commit_name = NULL;
2196 enum object_type type;
2197
2198 static const char *bottomtop = NULL;
2199 static int output_option = 0, opt = 0;
2200 static int show_stats = 0;
2201 static const char *revs_file = NULL;
2202 static const char *contents_from = NULL;
2203 static const struct option options[] = {
2204 OPT_BOOLEAN(0, "incremental", &incremental, "Show blame entries as we find them, incrementally"),
2205 OPT_BOOLEAN('b', NULL, &blank_boundary, "Show blank SHA-1 for boundary commits (Default: off)"),
2206 OPT_BOOLEAN(0, "root", &show_root, "Do not treat root commits as boundaries (Default: off)"),
2207 OPT_BOOLEAN(0, "show-stats", &show_stats, "Show work cost statistics"),
2208 OPT_BIT(0, "score-debug", &output_option, "Show output score for blame entries", OUTPUT_SHOW_SCORE),
2209 OPT_BIT('f', "show-name", &output_option, "Show original filename (Default: auto)", OUTPUT_SHOW_NAME),
2210 OPT_BIT('n', "show-number", &output_option, "Show original linenumber (Default: off)", OUTPUT_SHOW_NUMBER),
2211 OPT_BIT('p', "porcelain", &output_option, "Show in a format designed for machine consumption", OUTPUT_PORCELAIN),
2212 OPT_BIT('c', NULL, &output_option, "Use the same output mode as git-annotate (Default: off)", OUTPUT_ANNOTATE_COMPAT),
2213 OPT_BIT('t', NULL, &output_option, "Show raw timestamp (Default: off)", OUTPUT_RAW_TIMESTAMP),
2214 OPT_BIT('l', NULL, &output_option, "Show long commit SHA1 (Default: off)", OUTPUT_LONG_OBJECT_NAME),
2215 OPT_BIT('s', NULL, &output_option, "Suppress author name and timestamp (Default: off)", OUTPUT_NO_AUTHOR),
2216 OPT_BIT('w', NULL, &xdl_opts, "Ignore whitespace differences", XDF_IGNORE_WHITESPACE),
2217 OPT_STRING('S', NULL, &revs_file, "file", "Use revisions from <file> instead of calling git-rev-list"),
2218 OPT_STRING(0, "contents", &contents_from, "file", "Use <file>'s contents as the final image"),
2219 { OPTION_CALLBACK, 'C', NULL, &opt, "score", "Find line copies within and across files", PARSE_OPT_OPTARG, blame_copy_callback },
2220 { OPTION_CALLBACK, 'M', NULL, &opt, "score", "Find line movements within and across files", PARSE_OPT_OPTARG, blame_move_callback },
2221 OPT_CALLBACK('L', NULL, &bottomtop, "n,m", "Process only line range n,m, counting from 1", blame_bottomtop_callback),
2222 OPT_END()
2223 };
2224
2225 struct parse_opt_ctx_t ctx;
2226 int cmd_is_annotate = !strcmp(argv[0], "annotate");
2227
2228 git_config(git_blame_config, NULL);
2229 init_revisions(&revs, NULL);
2230 save_commit_buffer = 0;
2231 dashdash_pos = 0;
2232
2233 parse_options_start(&ctx, argc, argv, PARSE_OPT_KEEP_DASHDASH |
2234 PARSE_OPT_KEEP_ARGV0);
2235 for (;;) {
2236 switch (parse_options_step(&ctx, options, blame_opt_usage)) {
2237 case PARSE_OPT_HELP:
2238 exit(129);
2239 case PARSE_OPT_DONE:
2240 if (ctx.argv[0])
2241 dashdash_pos = ctx.cpidx;
2242 goto parse_done;
2243 }
2244
2245 if (!strcmp(ctx.argv[0], "--reverse")) {
2246 ctx.argv[0] = "--children";
2247 reverse = 1;
2248 }
2249 parse_revision_opt(&revs, &ctx, options, blame_opt_usage);
2250 }
2251 parse_done:
2252 argc = parse_options_end(&ctx);
2253
2254 if (revs_file && read_ancestry(revs_file))
2255 die("reading graft file %s failed: %s",
2256 revs_file, strerror(errno));
2257
2258 if (cmd_is_annotate)
2259 output_option |= OUTPUT_ANNOTATE_COMPAT;
2260
2261 if (DIFF_OPT_TST(&revs.diffopt, FIND_COPIES_HARDER))
2262 opt |= (PICKAXE_BLAME_COPY | PICKAXE_BLAME_MOVE |
2263 PICKAXE_BLAME_COPY_HARDER);
2264
2265 if (!blame_move_score)
2266 blame_move_score = BLAME_DEFAULT_MOVE_SCORE;
2267 if (!blame_copy_score)
2268 blame_copy_score = BLAME_DEFAULT_COPY_SCORE;
2269
2270 /*
2271 * We have collected options unknown to us in argv[1..unk]
2272 * which are to be passed to revision machinery if we are
2273 * going to do the "bottom" processing.
2274 *
2275 * The remaining are:
2276 *
2277 * (1) if dashdash_pos != 0, its either
2278 * "blame [revisions] -- <path>" or
2279 * "blame -- <path> <rev>"
2280 *
2281 * (2) otherwise, its one of the two:
2282 * "blame [revisions] <path>"
2283 * "blame <path> <rev>"
2284 *
2285 * Note that we must strip out <path> from the arguments: we do not
2286 * want the path pruning but we may want "bottom" processing.
2287 */
2288 if (dashdash_pos) {
2289 switch (argc - dashdash_pos - 1) {
2290 case 2: /* (1b) */
2291 if (argc != 4)
2292 usage_with_options(blame_opt_usage, options);
2293 /* reorder for the new way: <rev> -- <path> */
2294 argv[1] = argv[3];
2295 argv[3] = argv[2];
2296 argv[2] = "--";
2297 /* FALLTHROUGH */
2298 case 1: /* (1a) */
2299 path = add_prefix(prefix, argv[--argc]);
2300 argv[argc] = NULL;
2301 break;
2302 default:
2303 usage_with_options(blame_opt_usage, options);
2304 }
2305 } else {
2306 if (argc < 2)
2307 usage_with_options(blame_opt_usage, options);
2308 path = add_prefix(prefix, argv[argc - 1]);
2309 if (argc == 3 && !has_string_in_work_tree(path)) { /* (2b) */
2310 path = add_prefix(prefix, argv[1]);
2311 argv[1] = argv[2];
2312 }
2313 argv[argc - 1] = "--";
2314
2315 setup_work_tree();
2316 if (!has_string_in_work_tree(path))
2317 die("cannot stat path %s: %s", path, strerror(errno));
2318 }
2319
2320 setup_revisions(argc, argv, &revs, NULL);
2321 memset(&sb, 0, sizeof(sb));
2322
2323 sb.revs = &revs;
2324 if (!reverse)
2325 final_commit_name = prepare_final(&sb);
2326 else if (contents_from)
2327 die("--contents and --children do not blend well.");
2328 else
2329 final_commit_name = prepare_initial(&sb);
2330
2331 if (!sb.final) {
2332 /*
2333 * "--not A B -- path" without anything positive;
2334 * do not default to HEAD, but use the working tree
2335 * or "--contents".
2336 */
2337 setup_work_tree();
2338 sb.final = fake_working_tree_commit(path, contents_from);
2339 add_pending_object(&revs, &(sb.final->object), ":");
2340 }
2341 else if (contents_from)
2342 die("Cannot use --contents with final commit object name");
2343
2344 /*
2345 * If we have bottom, this will mark the ancestors of the
2346 * bottom commits we would reach while traversing as
2347 * uninteresting.
2348 */
2349 if (prepare_revision_walk(&revs))
2350 die("revision walk setup failed");
2351
2352 if (is_null_sha1(sb.final->object.sha1)) {
2353 char *buf;
2354 o = sb.final->util;
2355 buf = xmalloc(o->file.size + 1);
2356 memcpy(buf, o->file.ptr, o->file.size + 1);
2357 sb.final_buf = buf;
2358 sb.final_buf_size = o->file.size;
2359 }
2360 else {
2361 o = get_origin(&sb, sb.final, path);
2362 if (fill_blob_sha1(o))
2363 die("no such path %s in %s", path, final_commit_name);
2364
2365 sb.final_buf = read_sha1_file(o->blob_sha1, &type,
2366 &sb.final_buf_size);
2367 if (!sb.final_buf)
2368 die("Cannot read blob %s for path %s",
2369 sha1_to_hex(o->blob_sha1),
2370 path);
2371 }
2372 num_read_blob++;
2373 lno = prepare_lines(&sb);
2374
2375 bottom = top = 0;
2376 if (bottomtop)
2377 prepare_blame_range(&sb, bottomtop, lno, &bottom, &top);
2378 if (bottom && top && top < bottom) {
2379 long tmp;
2380 tmp = top; top = bottom; bottom = tmp;
2381 }
2382 if (bottom < 1)
2383 bottom = 1;
2384 if (top < 1)
2385 top = lno;
2386 bottom--;
2387 if (lno < top)
2388 die("file %s has only %lu lines", path, lno);
2389
2390 ent = xcalloc(1, sizeof(*ent));
2391 ent->lno = bottom;
2392 ent->num_lines = top - bottom;
2393 ent->suspect = o;
2394 ent->s_lno = bottom;
2395
2396 sb.ent = ent;
2397 sb.path = path;
2398
2399 read_mailmap(&mailmap, ".mailmap", NULL);
2400
2401 if (!incremental)
2402 setup_pager();
2403
2404 assign_blame(&sb, opt);
2405
2406 if (incremental)
2407 return 0;
2408
2409 coalesce(&sb);
2410
2411 if (!(output_option & OUTPUT_PORCELAIN))
2412 find_alignment(&sb, &output_option);
2413
2414 output(&sb, output_option);
2415 free((void *)sb.final_buf);
2416 for (ent = sb.ent; ent; ) {
2417 struct blame_entry *e = ent->next;
2418 free(ent);
2419 ent = e;
2420 }
2421
2422 if (show_stats) {
2423 printf("num read blob: %d\n", num_read_blob);
2424 printf("num get patch: %d\n", num_get_patch);
2425 printf("num commits: %d\n", num_commits);
2426 }
2427 return 0;
2428 }