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
19static char blame_usage[] =
20"git-blame [-c] [-l] [-t] [-f] [-n] [-p] [-L n,m] [-S <revs-file>] [-M] [-C] [-C] [commit] [--] file\n"
21" -c, --compatibility Use the same output mode as git-annotate (Default: off)\n"
22" -b Show blank SHA-1 for boundary commits (Default: off)\n"
23" -l, --long Show long commit SHA1 (Default: off)\n"
24" --root Do not treat root commits as boundaries (Default: off)\n"
25" -t, --time Show raw timestamp (Default: off)\n"
26" -f, --show-name Show original filename (Default: auto)\n"
27" -n, --show-number Show original linenumber (Default: off)\n"
28" -p, --porcelain Show in a format designed for machine consumption\n"
29" -L n,m Process only line range n,m, counting from 1\n"
30" -M, -C Find line movements within and across files\n"
31" --incremental Show blame entries as we find them, incrementally\n"
32" -S revs-file Use revisions from revs-file instead of calling git-rev-list\n";
33
34static int longest_file;
35static int longest_author;
36static int max_orig_digits;
37static int max_digits;
38static int max_score_digits;
39static int show_root;
40static int blank_boundary;
41static int incremental;
42
43#ifndef DEBUG
44#define DEBUG 0
45#endif
46
47/* stats */
48static int num_read_blob;
49static int num_get_patch;
50static int num_commits;
51
52#define PICKAXE_BLAME_MOVE 01
53#define PICKAXE_BLAME_COPY 02
54#define PICKAXE_BLAME_COPY_HARDER 04
55
56/*
57 * blame for a blame_entry with score lower than these thresholds
58 * is not passed to the parent using move/copy logic.
59 */
60static unsigned blame_move_score;
61static unsigned blame_copy_score;
62#define BLAME_DEFAULT_MOVE_SCORE 20
63#define BLAME_DEFAULT_COPY_SCORE 40
64
65/* bits #0..7 in revision.h, #8..11 used for merge_bases() in commit.c */
66#define METAINFO_SHOWN (1u<<12)
67#define MORE_THAN_ONE_PATH (1u<<13)
68
69/*
70 * One blob in a commit that is being suspected
71 */
72struct origin {
73 int refcnt;
74 struct commit *commit;
75 mmfile_t file;
76 unsigned char blob_sha1[20];
77 char path[FLEX_ARRAY];
78};
79
80/*
81 * Given an origin, prepare mmfile_t structure to be used by the
82 * diff machinery
83 */
84static char *fill_origin_blob(struct origin *o, mmfile_t *file)
85{
86 if (!o->file.ptr) {
87 char type[10];
88 num_read_blob++;
89 file->ptr = read_sha1_file(o->blob_sha1, type,
90 (unsigned long *)(&(file->size)));
91 o->file = *file;
92 }
93 else
94 *file = o->file;
95 return file->ptr;
96}
97
98/*
99 * Origin is refcounted and usually we keep the blob contents to be
100 * reused.
101 */
102static inline struct origin *origin_incref(struct origin *o)
103{
104 if (o)
105 o->refcnt++;
106 return o;
107}
108
109static void origin_decref(struct origin *o)
110{
111 if (o && --o->refcnt <= 0) {
112 if (o->file.ptr)
113 free(o->file.ptr);
114 memset(o, 0, sizeof(*o));
115 free(o);
116 }
117}
118
119/*
120 * Each group of lines is described by a blame_entry; it can be split
121 * as we pass blame to the parents. They form a linked list in the
122 * scoreboard structure, sorted by the target line number.
123 */
124struct blame_entry {
125 struct blame_entry *prev;
126 struct blame_entry *next;
127
128 /* the first line of this group in the final image;
129 * internally all line numbers are 0 based.
130 */
131 int lno;
132
133 /* how many lines this group has */
134 int num_lines;
135
136 /* the commit that introduced this group into the final image */
137 struct origin *suspect;
138
139 /* true if the suspect is truly guilty; false while we have not
140 * checked if the group came from one of its parents.
141 */
142 char guilty;
143
144 /* the line number of the first line of this group in the
145 * suspect's file; internally all line numbers are 0 based.
146 */
147 int s_lno;
148
149 /* how significant this entry is -- cached to avoid
150 * scanning the lines over and over.
151 */
152 unsigned score;
153};
154
155/*
156 * The current state of the blame assignment.
157 */
158struct scoreboard {
159 /* the final commit (i.e. where we started digging from) */
160 struct commit *final;
161
162 const char *path;
163
164 /*
165 * The contents in the final image.
166 * Used by many functions to obtain contents of the nth line,
167 * indexed with scoreboard.lineno[blame_entry.lno].
168 */
169 const char *final_buf;
170 unsigned long final_buf_size;
171
172 /* linked list of blames */
173 struct blame_entry *ent;
174
175 /* look-up a line in the final buffer */
176 int num_lines;
177 int *lineno;
178};
179
180static int cmp_suspect(struct origin *a, struct origin *b)
181{
182 int cmp = hashcmp(a->commit->object.sha1, b->commit->object.sha1);
183 if (cmp)
184 return cmp;
185 return strcmp(a->path, b->path);
186}
187
188#define cmp_suspect(a, b) ( ((a)==(b)) ? 0 : cmp_suspect(a,b) )
189
190static void sanity_check_refcnt(struct scoreboard *);
191
192/*
193 * If two blame entries that are next to each other came from
194 * contiguous lines in the same origin (i.e. <commit, path> pair),
195 * merge them together.
196 */
197static void coalesce(struct scoreboard *sb)
198{
199 struct blame_entry *ent, *next;
200
201 for (ent = sb->ent; ent && (next = ent->next); ent = next) {
202 if (!cmp_suspect(ent->suspect, next->suspect) &&
203 ent->guilty == next->guilty &&
204 ent->s_lno + ent->num_lines == next->s_lno) {
205 ent->num_lines += next->num_lines;
206 ent->next = next->next;
207 if (ent->next)
208 ent->next->prev = ent;
209 origin_decref(next->suspect);
210 free(next);
211 ent->score = 0;
212 next = ent; /* again */
213 }
214 }
215
216 if (DEBUG) /* sanity */
217 sanity_check_refcnt(sb);
218}
219
220/*
221 * Given a commit and a path in it, create a new origin structure.
222 * The callers that add blame to the scoreboard should use
223 * get_origin() to obtain shared, refcounted copy instead of calling
224 * this function directly.
225 */
226static struct origin *make_origin(struct commit *commit, const char *path)
227{
228 struct origin *o;
229 o = xcalloc(1, sizeof(*o) + strlen(path) + 1);
230 o->commit = commit;
231 o->refcnt = 1;
232 strcpy(o->path, path);
233 return o;
234}
235
236/*
237 * Locate an existing origin or create a new one.
238 */
239static struct origin *get_origin(struct scoreboard *sb,
240 struct commit *commit,
241 const char *path)
242{
243 struct blame_entry *e;
244
245 for (e = sb->ent; e; e = e->next) {
246 if (e->suspect->commit == commit &&
247 !strcmp(e->suspect->path, path))
248 return origin_incref(e->suspect);
249 }
250 return make_origin(commit, path);
251}
252
253/*
254 * Fill the blob_sha1 field of an origin if it hasn't, so that later
255 * call to fill_origin_blob() can use it to locate the data. blob_sha1
256 * for an origin is also used to pass the blame for the entire file to
257 * the parent to detect the case where a child's blob is identical to
258 * that of its parent's.
259 */
260static int fill_blob_sha1(struct origin *origin)
261{
262 unsigned mode;
263 char type[10];
264
265 if (!is_null_sha1(origin->blob_sha1))
266 return 0;
267 if (get_tree_entry(origin->commit->object.sha1,
268 origin->path,
269 origin->blob_sha1, &mode))
270 goto error_out;
271 if (sha1_object_info(origin->blob_sha1, type, NULL) ||
272 strcmp(type, blob_type))
273 goto error_out;
274 return 0;
275 error_out:
276 hashclr(origin->blob_sha1);
277 return -1;
278}
279
280/*
281 * We have an origin -- check if the same path exists in the
282 * parent and return an origin structure to represent it.
283 */
284static struct origin *find_origin(struct scoreboard *sb,
285 struct commit *parent,
286 struct origin *origin)
287{
288 struct origin *porigin = NULL;
289 struct diff_options diff_opts;
290 const char *paths[2];
291
292 if (parent->util) {
293 /*
294 * Each commit object can cache one origin in that
295 * commit. This is a freestanding copy of origin and
296 * not refcounted.
297 */
298 struct origin *cached = parent->util;
299 if (!strcmp(cached->path, origin->path)) {
300 /*
301 * The same path between origin and its parent
302 * without renaming -- the most common case.
303 */
304 porigin = get_origin(sb, parent, cached->path);
305
306 /*
307 * If the origin was newly created (i.e. get_origin
308 * would call make_origin if none is found in the
309 * scoreboard), it does not know the blob_sha1,
310 * so copy it. Otherwise porigin was in the
311 * scoreboard and already knows blob_sha1.
312 */
313 if (porigin->refcnt == 1)
314 hashcpy(porigin->blob_sha1, cached->blob_sha1);
315 return porigin;
316 }
317 /* otherwise it was not very useful; free it */
318 free(parent->util);
319 parent->util = NULL;
320 }
321
322 /* See if the origin->path is different between parent
323 * and origin first. Most of the time they are the
324 * same and diff-tree is fairly efficient about this.
325 */
326 diff_setup(&diff_opts);
327 diff_opts.recursive = 1;
328 diff_opts.detect_rename = 0;
329 diff_opts.output_format = DIFF_FORMAT_NO_OUTPUT;
330 paths[0] = origin->path;
331 paths[1] = NULL;
332
333 diff_tree_setup_paths(paths, &diff_opts);
334 if (diff_setup_done(&diff_opts) < 0)
335 die("diff-setup");
336 diff_tree_sha1(parent->tree->object.sha1,
337 origin->commit->tree->object.sha1,
338 "", &diff_opts);
339 diffcore_std(&diff_opts);
340
341 /* It is either one entry that says "modified", or "created",
342 * or nothing.
343 */
344 if (!diff_queued_diff.nr) {
345 /* The path is the same as parent */
346 porigin = get_origin(sb, parent, origin->path);
347 hashcpy(porigin->blob_sha1, origin->blob_sha1);
348 }
349 else if (diff_queued_diff.nr != 1)
350 die("internal error in blame::find_origin");
351 else {
352 struct diff_filepair *p = diff_queued_diff.queue[0];
353 switch (p->status) {
354 default:
355 die("internal error in blame::find_origin (%c)",
356 p->status);
357 case 'M':
358 porigin = get_origin(sb, parent, origin->path);
359 hashcpy(porigin->blob_sha1, p->one->sha1);
360 break;
361 case 'A':
362 case 'T':
363 /* Did not exist in parent, or type changed */
364 break;
365 }
366 }
367 diff_flush(&diff_opts);
368 if (porigin) {
369 /*
370 * Create a freestanding copy that is not part of
371 * the refcounted origin found in the scoreboard, and
372 * cache it in the commit.
373 */
374 struct origin *cached;
375
376 cached = make_origin(porigin->commit, porigin->path);
377 hashcpy(cached->blob_sha1, porigin->blob_sha1);
378 parent->util = cached;
379 }
380 return porigin;
381}
382
383/*
384 * We have an origin -- find the path that corresponds to it in its
385 * parent and return an origin structure to represent it.
386 */
387static struct origin *find_rename(struct scoreboard *sb,
388 struct commit *parent,
389 struct origin *origin)
390{
391 struct origin *porigin = NULL;
392 struct diff_options diff_opts;
393 int i;
394 const char *paths[2];
395
396 diff_setup(&diff_opts);
397 diff_opts.recursive = 1;
398 diff_opts.detect_rename = DIFF_DETECT_RENAME;
399 diff_opts.output_format = DIFF_FORMAT_NO_OUTPUT;
400 diff_opts.single_follow = origin->path;
401 paths[0] = NULL;
402 diff_tree_setup_paths(paths, &diff_opts);
403 if (diff_setup_done(&diff_opts) < 0)
404 die("diff-setup");
405 diff_tree_sha1(parent->tree->object.sha1,
406 origin->commit->tree->object.sha1,
407 "", &diff_opts);
408 diffcore_std(&diff_opts);
409
410 for (i = 0; i < diff_queued_diff.nr; i++) {
411 struct diff_filepair *p = diff_queued_diff.queue[i];
412 if ((p->status == 'R' || p->status == 'C') &&
413 !strcmp(p->two->path, origin->path)) {
414 porigin = get_origin(sb, parent, p->one->path);
415 hashcpy(porigin->blob_sha1, p->one->sha1);
416 break;
417 }
418 }
419 diff_flush(&diff_opts);
420 return porigin;
421}
422
423/*
424 * Parsing of patch chunks...
425 */
426struct chunk {
427 /* line number in postimage; up to but not including this
428 * line is the same as preimage
429 */
430 int same;
431
432 /* preimage line number after this chunk */
433 int p_next;
434
435 /* postimage line number after this chunk */
436 int t_next;
437};
438
439struct patch {
440 struct chunk *chunks;
441 int num;
442};
443
444struct blame_diff_state {
445 struct xdiff_emit_state xm;
446 struct patch *ret;
447 unsigned hunk_post_context;
448 unsigned hunk_in_pre_context : 1;
449};
450
451static void process_u_diff(void *state_, char *line, unsigned long len)
452{
453 struct blame_diff_state *state = state_;
454 struct chunk *chunk;
455 int off1, off2, len1, len2, num;
456
457 num = state->ret->num;
458 if (len < 4 || line[0] != '@' || line[1] != '@') {
459 if (state->hunk_in_pre_context && line[0] == ' ')
460 state->ret->chunks[num - 1].same++;
461 else {
462 state->hunk_in_pre_context = 0;
463 if (line[0] == ' ')
464 state->hunk_post_context++;
465 else
466 state->hunk_post_context = 0;
467 }
468 return;
469 }
470
471 if (num && state->hunk_post_context) {
472 chunk = &state->ret->chunks[num - 1];
473 chunk->p_next -= state->hunk_post_context;
474 chunk->t_next -= state->hunk_post_context;
475 }
476 state->ret->num = ++num;
477 state->ret->chunks = xrealloc(state->ret->chunks,
478 sizeof(struct chunk) * num);
479 chunk = &state->ret->chunks[num - 1];
480 if (parse_hunk_header(line, len, &off1, &len1, &off2, &len2)) {
481 state->ret->num--;
482 return;
483 }
484
485 /* Line numbers in patch output are one based. */
486 off1--;
487 off2--;
488
489 chunk->same = len2 ? off2 : (off2 + 1);
490
491 chunk->p_next = off1 + (len1 ? len1 : 1);
492 chunk->t_next = chunk->same + len2;
493 state->hunk_in_pre_context = 1;
494 state->hunk_post_context = 0;
495}
496
497static struct patch *compare_buffer(mmfile_t *file_p, mmfile_t *file_o,
498 int context)
499{
500 struct blame_diff_state state;
501 xpparam_t xpp;
502 xdemitconf_t xecfg;
503 xdemitcb_t ecb;
504
505 xpp.flags = XDF_NEED_MINIMAL;
506 xecfg.ctxlen = context;
507 xecfg.flags = 0;
508 ecb.outf = xdiff_outf;
509 ecb.priv = &state;
510 memset(&state, 0, sizeof(state));
511 state.xm.consume = process_u_diff;
512 state.ret = xmalloc(sizeof(struct patch));
513 state.ret->chunks = NULL;
514 state.ret->num = 0;
515
516 xdl_diff(file_p, file_o, &xpp, &xecfg, &ecb);
517
518 if (state.ret->num) {
519 struct chunk *chunk;
520 chunk = &state.ret->chunks[state.ret->num - 1];
521 chunk->p_next -= state.hunk_post_context;
522 chunk->t_next -= state.hunk_post_context;
523 }
524 return state.ret;
525}
526
527/*
528 * Run diff between two origins and grab the patch output, so that
529 * we can pass blame for lines origin is currently suspected for
530 * to its parent.
531 */
532static struct patch *get_patch(struct origin *parent, struct origin *origin)
533{
534 mmfile_t file_p, file_o;
535 struct patch *patch;
536
537 fill_origin_blob(parent, &file_p);
538 fill_origin_blob(origin, &file_o);
539 if (!file_p.ptr || !file_o.ptr)
540 return NULL;
541 patch = compare_buffer(&file_p, &file_o, 0);
542 num_get_patch++;
543 return patch;
544}
545
546static void free_patch(struct patch *p)
547{
548 free(p->chunks);
549 free(p);
550}
551
552/*
553 * Link in a new blame entry to the scoreboard. Entries that cover the
554 * same line range have been removed from the scoreboard previously.
555 */
556static void add_blame_entry(struct scoreboard *sb, struct blame_entry *e)
557{
558 struct blame_entry *ent, *prev = NULL;
559
560 origin_incref(e->suspect);
561
562 for (ent = sb->ent; ent && ent->lno < e->lno; ent = ent->next)
563 prev = ent;
564
565 /* prev, if not NULL, is the last one that is below e */
566 e->prev = prev;
567 if (prev) {
568 e->next = prev->next;
569 prev->next = e;
570 }
571 else {
572 e->next = sb->ent;
573 sb->ent = e;
574 }
575 if (e->next)
576 e->next->prev = e;
577}
578
579/*
580 * src typically is on-stack; we want to copy the information in it to
581 * an malloced blame_entry that is already on the linked list of the
582 * scoreboard. The origin of dst loses a refcnt while the origin of src
583 * gains one.
584 */
585static void dup_entry(struct blame_entry *dst, struct blame_entry *src)
586{
587 struct blame_entry *p, *n;
588
589 p = dst->prev;
590 n = dst->next;
591 origin_incref(src->suspect);
592 origin_decref(dst->suspect);
593 memcpy(dst, src, sizeof(*src));
594 dst->prev = p;
595 dst->next = n;
596 dst->score = 0;
597}
598
599static const char *nth_line(struct scoreboard *sb, int lno)
600{
601 return sb->final_buf + sb->lineno[lno];
602}
603
604/*
605 * It is known that lines between tlno to same came from parent, and e
606 * has an overlap with that range. it also is known that parent's
607 * line plno corresponds to e's line tlno.
608 *
609 * <---- e ----->
610 * <------>
611 * <------------>
612 * <------------>
613 * <------------------>
614 *
615 * Split e into potentially three parts; before this chunk, the chunk
616 * to be blamed for the parent, and after that portion.
617 */
618static void split_overlap(struct blame_entry *split,
619 struct blame_entry *e,
620 int tlno, int plno, int same,
621 struct origin *parent)
622{
623 int chunk_end_lno;
624 memset(split, 0, sizeof(struct blame_entry [3]));
625
626 if (e->s_lno < tlno) {
627 /* there is a pre-chunk part not blamed on parent */
628 split[0].suspect = origin_incref(e->suspect);
629 split[0].lno = e->lno;
630 split[0].s_lno = e->s_lno;
631 split[0].num_lines = tlno - e->s_lno;
632 split[1].lno = e->lno + tlno - e->s_lno;
633 split[1].s_lno = plno;
634 }
635 else {
636 split[1].lno = e->lno;
637 split[1].s_lno = plno + (e->s_lno - tlno);
638 }
639
640 if (same < e->s_lno + e->num_lines) {
641 /* there is a post-chunk part not blamed on parent */
642 split[2].suspect = origin_incref(e->suspect);
643 split[2].lno = e->lno + (same - e->s_lno);
644 split[2].s_lno = e->s_lno + (same - e->s_lno);
645 split[2].num_lines = e->s_lno + e->num_lines - same;
646 chunk_end_lno = split[2].lno;
647 }
648 else
649 chunk_end_lno = e->lno + e->num_lines;
650 split[1].num_lines = chunk_end_lno - split[1].lno;
651
652 /*
653 * if it turns out there is nothing to blame the parent for,
654 * forget about the splitting. !split[1].suspect signals this.
655 */
656 if (split[1].num_lines < 1)
657 return;
658 split[1].suspect = origin_incref(parent);
659}
660
661/*
662 * split_overlap() divided an existing blame e into up to three parts
663 * in split. Adjust the linked list of blames in the scoreboard to
664 * reflect the split.
665 */
666static void split_blame(struct scoreboard *sb,
667 struct blame_entry *split,
668 struct blame_entry *e)
669{
670 struct blame_entry *new_entry;
671
672 if (split[0].suspect && split[2].suspect) {
673 /* The first part (reuse storage for the existing entry e) */
674 dup_entry(e, &split[0]);
675
676 /* The last part -- me */
677 new_entry = xmalloc(sizeof(*new_entry));
678 memcpy(new_entry, &(split[2]), sizeof(struct blame_entry));
679 add_blame_entry(sb, new_entry);
680
681 /* ... and the middle part -- parent */
682 new_entry = xmalloc(sizeof(*new_entry));
683 memcpy(new_entry, &(split[1]), sizeof(struct blame_entry));
684 add_blame_entry(sb, new_entry);
685 }
686 else if (!split[0].suspect && !split[2].suspect)
687 /*
688 * The parent covers the entire area; reuse storage for
689 * e and replace it with the parent.
690 */
691 dup_entry(e, &split[1]);
692 else if (split[0].suspect) {
693 /* me and then parent */
694 dup_entry(e, &split[0]);
695
696 new_entry = xmalloc(sizeof(*new_entry));
697 memcpy(new_entry, &(split[1]), sizeof(struct blame_entry));
698 add_blame_entry(sb, new_entry);
699 }
700 else {
701 /* parent and then me */
702 dup_entry(e, &split[1]);
703
704 new_entry = xmalloc(sizeof(*new_entry));
705 memcpy(new_entry, &(split[2]), sizeof(struct blame_entry));
706 add_blame_entry(sb, new_entry);
707 }
708
709 if (DEBUG) { /* sanity */
710 struct blame_entry *ent;
711 int lno = sb->ent->lno, corrupt = 0;
712
713 for (ent = sb->ent; ent; ent = ent->next) {
714 if (lno != ent->lno)
715 corrupt = 1;
716 if (ent->s_lno < 0)
717 corrupt = 1;
718 lno += ent->num_lines;
719 }
720 if (corrupt) {
721 lno = sb->ent->lno;
722 for (ent = sb->ent; ent; ent = ent->next) {
723 printf("L %8d l %8d n %8d\n",
724 lno, ent->lno, ent->num_lines);
725 lno = ent->lno + ent->num_lines;
726 }
727 die("oops");
728 }
729 }
730}
731
732/*
733 * After splitting the blame, the origins used by the
734 * on-stack blame_entry should lose one refcnt each.
735 */
736static void decref_split(struct blame_entry *split)
737{
738 int i;
739
740 for (i = 0; i < 3; i++)
741 origin_decref(split[i].suspect);
742}
743
744/*
745 * Helper for blame_chunk(). blame_entry e is known to overlap with
746 * the patch hunk; split it and pass blame to the parent.
747 */
748static void blame_overlap(struct scoreboard *sb, struct blame_entry *e,
749 int tlno, int plno, int same,
750 struct origin *parent)
751{
752 struct blame_entry split[3];
753
754 split_overlap(split, e, tlno, plno, same, parent);
755 if (split[1].suspect)
756 split_blame(sb, split, e);
757 decref_split(split);
758}
759
760/*
761 * Find the line number of the last line the target is suspected for.
762 */
763static int find_last_in_target(struct scoreboard *sb, struct origin *target)
764{
765 struct blame_entry *e;
766 int last_in_target = -1;
767
768 for (e = sb->ent; e; e = e->next) {
769 if (e->guilty || cmp_suspect(e->suspect, target))
770 continue;
771 if (last_in_target < e->s_lno + e->num_lines)
772 last_in_target = e->s_lno + e->num_lines;
773 }
774 return last_in_target;
775}
776
777/*
778 * Process one hunk from the patch between the current suspect for
779 * blame_entry e and its parent. Find and split the overlap, and
780 * pass blame to the overlapping part to the parent.
781 */
782static void blame_chunk(struct scoreboard *sb,
783 int tlno, int plno, int same,
784 struct origin *target, struct origin *parent)
785{
786 struct blame_entry *e;
787
788 for (e = sb->ent; e; e = e->next) {
789 if (e->guilty || cmp_suspect(e->suspect, target))
790 continue;
791 if (same <= e->s_lno)
792 continue;
793 if (tlno < e->s_lno + e->num_lines)
794 blame_overlap(sb, e, tlno, plno, same, parent);
795 }
796}
797
798/*
799 * We are looking at the origin 'target' and aiming to pass blame
800 * for the lines it is suspected to its parent. Run diff to find
801 * which lines came from parent and pass blame for them.
802 */
803static int pass_blame_to_parent(struct scoreboard *sb,
804 struct origin *target,
805 struct origin *parent)
806{
807 int i, last_in_target, plno, tlno;
808 struct patch *patch;
809
810 last_in_target = find_last_in_target(sb, target);
811 if (last_in_target < 0)
812 return 1; /* nothing remains for this target */
813
814 patch = get_patch(parent, target);
815 plno = tlno = 0;
816 for (i = 0; i < patch->num; i++) {
817 struct chunk *chunk = &patch->chunks[i];
818
819 blame_chunk(sb, tlno, plno, chunk->same, target, parent);
820 plno = chunk->p_next;
821 tlno = chunk->t_next;
822 }
823 /* The rest (i.e. anything after tlno) are the same as the parent */
824 blame_chunk(sb, tlno, plno, last_in_target, target, parent);
825
826 free_patch(patch);
827 return 0;
828}
829
830/*
831 * The lines in blame_entry after splitting blames many times can become
832 * very small and trivial, and at some point it becomes pointless to
833 * blame the parents. E.g. "\t\t}\n\t}\n\n" appears everywhere in any
834 * ordinary C program, and it is not worth to say it was copied from
835 * totally unrelated file in the parent.
836 *
837 * Compute how trivial the lines in the blame_entry are.
838 */
839static unsigned ent_score(struct scoreboard *sb, struct blame_entry *e)
840{
841 unsigned score;
842 const char *cp, *ep;
843
844 if (e->score)
845 return e->score;
846
847 score = 1;
848 cp = nth_line(sb, e->lno);
849 ep = nth_line(sb, e->lno + e->num_lines);
850 while (cp < ep) {
851 unsigned ch = *((unsigned char *)cp);
852 if (isalnum(ch))
853 score++;
854 cp++;
855 }
856 e->score = score;
857 return score;
858}
859
860/*
861 * best_so_far[] and this[] are both a split of an existing blame_entry
862 * that passes blame to the parent. Maintain best_so_far the best split
863 * so far, by comparing this and best_so_far and copying this into
864 * bst_so_far as needed.
865 */
866static void copy_split_if_better(struct scoreboard *sb,
867 struct blame_entry *best_so_far,
868 struct blame_entry *this)
869{
870 int i;
871
872 if (!this[1].suspect)
873 return;
874 if (best_so_far[1].suspect) {
875 if (ent_score(sb, &this[1]) < ent_score(sb, &best_so_far[1]))
876 return;
877 }
878
879 for (i = 0; i < 3; i++)
880 origin_incref(this[i].suspect);
881 decref_split(best_so_far);
882 memcpy(best_so_far, this, sizeof(struct blame_entry [3]));
883}
884
885/*
886 * Find the lines from parent that are the same as ent so that
887 * we can pass blames to it. file_p has the blob contents for
888 * the parent.
889 */
890static void find_copy_in_blob(struct scoreboard *sb,
891 struct blame_entry *ent,
892 struct origin *parent,
893 struct blame_entry *split,
894 mmfile_t *file_p)
895{
896 const char *cp;
897 int cnt;
898 mmfile_t file_o;
899 struct patch *patch;
900 int i, plno, tlno;
901
902 /*
903 * Prepare mmfile that contains only the lines in ent.
904 */
905 cp = nth_line(sb, ent->lno);
906 file_o.ptr = (char*) cp;
907 cnt = ent->num_lines;
908
909 while (cnt && cp < sb->final_buf + sb->final_buf_size) {
910 if (*cp++ == '\n')
911 cnt--;
912 }
913 file_o.size = cp - file_o.ptr;
914
915 patch = compare_buffer(file_p, &file_o, 1);
916
917 memset(split, 0, sizeof(struct blame_entry [3]));
918 plno = tlno = 0;
919 for (i = 0; i < patch->num; i++) {
920 struct chunk *chunk = &patch->chunks[i];
921
922 /* tlno to chunk->same are the same as ent */
923 if (ent->num_lines <= tlno)
924 break;
925 if (tlno < chunk->same) {
926 struct blame_entry this[3];
927 split_overlap(this, ent,
928 tlno + ent->s_lno, plno,
929 chunk->same + ent->s_lno,
930 parent);
931 copy_split_if_better(sb, split, this);
932 decref_split(this);
933 }
934 plno = chunk->p_next;
935 tlno = chunk->t_next;
936 }
937 free_patch(patch);
938}
939
940/*
941 * See if lines currently target is suspected for can be attributed to
942 * parent.
943 */
944static int find_move_in_parent(struct scoreboard *sb,
945 struct origin *target,
946 struct origin *parent)
947{
948 int last_in_target, made_progress;
949 struct blame_entry *e, split[3];
950 mmfile_t file_p;
951
952 last_in_target = find_last_in_target(sb, target);
953 if (last_in_target < 0)
954 return 1; /* nothing remains for this target */
955
956 fill_origin_blob(parent, &file_p);
957 if (!file_p.ptr)
958 return 0;
959
960 made_progress = 1;
961 while (made_progress) {
962 made_progress = 0;
963 for (e = sb->ent; e; e = e->next) {
964 if (e->guilty || cmp_suspect(e->suspect, target))
965 continue;
966 find_copy_in_blob(sb, e, parent, split, &file_p);
967 if (split[1].suspect &&
968 blame_move_score < ent_score(sb, &split[1])) {
969 split_blame(sb, split, e);
970 made_progress = 1;
971 }
972 decref_split(split);
973 }
974 }
975 return 0;
976}
977
978struct blame_list {
979 struct blame_entry *ent;
980 struct blame_entry split[3];
981};
982
983/*
984 * Count the number of entries the target is suspected for,
985 * and prepare a list of entry and the best split.
986 */
987static struct blame_list *setup_blame_list(struct scoreboard *sb,
988 struct origin *target,
989 int *num_ents_p)
990{
991 struct blame_entry *e;
992 int num_ents, i;
993 struct blame_list *blame_list = NULL;
994
995 for (e = sb->ent, num_ents = 0; e; e = e->next)
996 if (!e->guilty && !cmp_suspect(e->suspect, target))
997 num_ents++;
998 if (num_ents) {
999 blame_list = xcalloc(num_ents, sizeof(struct blame_list));
1000 for (e = sb->ent, i = 0; e; e = e->next)
1001 if (!e->guilty && !cmp_suspect(e->suspect, target))
1002 blame_list[i++].ent = e;
1003 }
1004 *num_ents_p = num_ents;
1005 return blame_list;
1006}
1007
1008/*
1009 * For lines target is suspected for, see if we can find code movement
1010 * across file boundary from the parent commit. porigin is the path
1011 * in the parent we already tried.
1012 */
1013static int find_copy_in_parent(struct scoreboard *sb,
1014 struct origin *target,
1015 struct commit *parent,
1016 struct origin *porigin,
1017 int opt)
1018{
1019 struct diff_options diff_opts;
1020 const char *paths[1];
1021 int i, j;
1022 int retval;
1023 struct blame_list *blame_list;
1024 int num_ents;
1025
1026 blame_list = setup_blame_list(sb, target, &num_ents);
1027 if (!blame_list)
1028 return 1; /* nothing remains for this target */
1029
1030 diff_setup(&diff_opts);
1031 diff_opts.recursive = 1;
1032 diff_opts.output_format = DIFF_FORMAT_NO_OUTPUT;
1033
1034 paths[0] = NULL;
1035 diff_tree_setup_paths(paths, &diff_opts);
1036 if (diff_setup_done(&diff_opts) < 0)
1037 die("diff-setup");
1038
1039 /* Try "find copies harder" on new path if requested;
1040 * we do not want to use diffcore_rename() actually to
1041 * match things up; find_copies_harder is set only to
1042 * force diff_tree_sha1() to feed all filepairs to diff_queue,
1043 * and this code needs to be after diff_setup_done(), which
1044 * usually makes find-copies-harder imply copy detection.
1045 */
1046 if ((opt & PICKAXE_BLAME_COPY_HARDER) &&
1047 (!porigin || strcmp(target->path, porigin->path)))
1048 diff_opts.find_copies_harder = 1;
1049
1050 diff_tree_sha1(parent->tree->object.sha1,
1051 target->commit->tree->object.sha1,
1052 "", &diff_opts);
1053
1054 if (!diff_opts.find_copies_harder)
1055 diffcore_std(&diff_opts);
1056
1057 retval = 0;
1058 while (1) {
1059 int made_progress = 0;
1060
1061 for (i = 0; i < diff_queued_diff.nr; i++) {
1062 struct diff_filepair *p = diff_queued_diff.queue[i];
1063 struct origin *norigin;
1064 mmfile_t file_p;
1065 struct blame_entry this[3];
1066
1067 if (!DIFF_FILE_VALID(p->one))
1068 continue; /* does not exist in parent */
1069 if (porigin && !strcmp(p->one->path, porigin->path))
1070 /* find_move already dealt with this path */
1071 continue;
1072
1073 norigin = get_origin(sb, parent, p->one->path);
1074 hashcpy(norigin->blob_sha1, p->one->sha1);
1075 fill_origin_blob(norigin, &file_p);
1076 if (!file_p.ptr)
1077 continue;
1078
1079 for (j = 0; j < num_ents; j++) {
1080 find_copy_in_blob(sb, blame_list[j].ent,
1081 norigin, this, &file_p);
1082 copy_split_if_better(sb, blame_list[j].split,
1083 this);
1084 decref_split(this);
1085 }
1086 origin_decref(norigin);
1087 }
1088
1089 for (j = 0; j < num_ents; j++) {
1090 struct blame_entry *split = blame_list[j].split;
1091 if (split[1].suspect &&
1092 blame_copy_score < ent_score(sb, &split[1])) {
1093 split_blame(sb, split, blame_list[j].ent);
1094 made_progress = 1;
1095 }
1096 decref_split(split);
1097 }
1098 free(blame_list);
1099
1100 if (!made_progress)
1101 break;
1102 blame_list = setup_blame_list(sb, target, &num_ents);
1103 if (!blame_list) {
1104 retval = 1;
1105 break;
1106 }
1107 }
1108 diff_flush(&diff_opts);
1109
1110 return retval;
1111}
1112
1113/*
1114 * The blobs of origin and porigin exactly match, so everything
1115 * origin is suspected for can be blamed on the parent.
1116 */
1117static void pass_whole_blame(struct scoreboard *sb,
1118 struct origin *origin, struct origin *porigin)
1119{
1120 struct blame_entry *e;
1121
1122 if (!porigin->file.ptr && origin->file.ptr) {
1123 /* Steal its file */
1124 porigin->file = origin->file;
1125 origin->file.ptr = NULL;
1126 }
1127 for (e = sb->ent; e; e = e->next) {
1128 if (cmp_suspect(e->suspect, origin))
1129 continue;
1130 origin_incref(porigin);
1131 origin_decref(e->suspect);
1132 e->suspect = porigin;
1133 }
1134}
1135
1136#define MAXPARENT 16
1137
1138static void pass_blame(struct scoreboard *sb, struct origin *origin, int opt)
1139{
1140 int i, pass;
1141 struct commit *commit = origin->commit;
1142 struct commit_list *parent;
1143 struct origin *parent_origin[MAXPARENT], *porigin;
1144
1145 memset(parent_origin, 0, sizeof(parent_origin));
1146
1147 /* The first pass looks for unrenamed path to optimize for
1148 * common cases, then we look for renames in the second pass.
1149 */
1150 for (pass = 0; pass < 2; pass++) {
1151 struct origin *(*find)(struct scoreboard *,
1152 struct commit *, struct origin *);
1153 find = pass ? find_rename : find_origin;
1154
1155 for (i = 0, parent = commit->parents;
1156 i < MAXPARENT && parent;
1157 parent = parent->next, i++) {
1158 struct commit *p = parent->item;
1159 int j, same;
1160
1161 if (parent_origin[i])
1162 continue;
1163 if (parse_commit(p))
1164 continue;
1165 porigin = find(sb, p, origin);
1166 if (!porigin)
1167 continue;
1168 if (!hashcmp(porigin->blob_sha1, origin->blob_sha1)) {
1169 pass_whole_blame(sb, origin, porigin);
1170 origin_decref(porigin);
1171 goto finish;
1172 }
1173 for (j = same = 0; j < i; j++)
1174 if (parent_origin[j] &&
1175 !hashcmp(parent_origin[j]->blob_sha1,
1176 porigin->blob_sha1)) {
1177 same = 1;
1178 break;
1179 }
1180 if (!same)
1181 parent_origin[i] = porigin;
1182 else
1183 origin_decref(porigin);
1184 }
1185 }
1186
1187 num_commits++;
1188 for (i = 0, parent = commit->parents;
1189 i < MAXPARENT && parent;
1190 parent = parent->next, i++) {
1191 struct origin *porigin = parent_origin[i];
1192 if (!porigin)
1193 continue;
1194 if (pass_blame_to_parent(sb, origin, porigin))
1195 goto finish;
1196 }
1197
1198 /*
1199 * Optionally find moves in parents' files.
1200 */
1201 if (opt & PICKAXE_BLAME_MOVE)
1202 for (i = 0, parent = commit->parents;
1203 i < MAXPARENT && parent;
1204 parent = parent->next, i++) {
1205 struct origin *porigin = parent_origin[i];
1206 if (!porigin)
1207 continue;
1208 if (find_move_in_parent(sb, origin, porigin))
1209 goto finish;
1210 }
1211
1212 /*
1213 * Optionally find copies from parents' files.
1214 */
1215 if (opt & PICKAXE_BLAME_COPY)
1216 for (i = 0, parent = commit->parents;
1217 i < MAXPARENT && parent;
1218 parent = parent->next, i++) {
1219 struct origin *porigin = parent_origin[i];
1220 if (find_copy_in_parent(sb, origin, parent->item,
1221 porigin, opt))
1222 goto finish;
1223 }
1224
1225 finish:
1226 for (i = 0; i < MAXPARENT; i++)
1227 origin_decref(parent_origin[i]);
1228}
1229
1230/*
1231 * Information on commits, used for output.
1232 */
1233struct commit_info
1234{
1235 char *author;
1236 char *author_mail;
1237 unsigned long author_time;
1238 char *author_tz;
1239
1240 /* filled only when asked for details */
1241 char *committer;
1242 char *committer_mail;
1243 unsigned long committer_time;
1244 char *committer_tz;
1245
1246 char *summary;
1247};
1248
1249/*
1250 * Parse author/committer line in the commit object buffer
1251 */
1252static void get_ac_line(const char *inbuf, const char *what,
1253 int bufsz, char *person, char **mail,
1254 unsigned long *time, char **tz)
1255{
1256 int len;
1257 char *tmp, *endp;
1258
1259 tmp = strstr(inbuf, what);
1260 if (!tmp)
1261 goto error_out;
1262 tmp += strlen(what);
1263 endp = strchr(tmp, '\n');
1264 if (!endp)
1265 len = strlen(tmp);
1266 else
1267 len = endp - tmp;
1268 if (bufsz <= len) {
1269 error_out:
1270 /* Ugh */
1271 person = *mail = *tz = "(unknown)";
1272 *time = 0;
1273 return;
1274 }
1275 memcpy(person, tmp, len);
1276
1277 tmp = person;
1278 tmp += len;
1279 *tmp = 0;
1280 while (*tmp != ' ')
1281 tmp--;
1282 *tz = tmp+1;
1283
1284 *tmp = 0;
1285 while (*tmp != ' ')
1286 tmp--;
1287 *time = strtoul(tmp, NULL, 10);
1288
1289 *tmp = 0;
1290 while (*tmp != ' ')
1291 tmp--;
1292 *mail = tmp + 1;
1293 *tmp = 0;
1294}
1295
1296static void get_commit_info(struct commit *commit,
1297 struct commit_info *ret,
1298 int detailed)
1299{
1300 int len;
1301 char *tmp, *endp;
1302 static char author_buf[1024];
1303 static char committer_buf[1024];
1304 static char summary_buf[1024];
1305
1306 /*
1307 * We've operated without save_commit_buffer, so
1308 * we now need to populate them for output.
1309 */
1310 if (!commit->buffer) {
1311 char type[20];
1312 unsigned long size;
1313 commit->buffer =
1314 read_sha1_file(commit->object.sha1, type, &size);
1315 }
1316 ret->author = author_buf;
1317 get_ac_line(commit->buffer, "\nauthor ",
1318 sizeof(author_buf), author_buf, &ret->author_mail,
1319 &ret->author_time, &ret->author_tz);
1320
1321 if (!detailed)
1322 return;
1323
1324 ret->committer = committer_buf;
1325 get_ac_line(commit->buffer, "\ncommitter ",
1326 sizeof(committer_buf), committer_buf, &ret->committer_mail,
1327 &ret->committer_time, &ret->committer_tz);
1328
1329 ret->summary = summary_buf;
1330 tmp = strstr(commit->buffer, "\n\n");
1331 if (!tmp) {
1332 error_out:
1333 sprintf(summary_buf, "(%s)", sha1_to_hex(commit->object.sha1));
1334 return;
1335 }
1336 tmp += 2;
1337 endp = strchr(tmp, '\n');
1338 if (!endp)
1339 goto error_out;
1340 len = endp - tmp;
1341 if (len >= sizeof(summary_buf))
1342 goto error_out;
1343 memcpy(summary_buf, tmp, len);
1344 summary_buf[len] = 0;
1345}
1346
1347/*
1348 * To allow LF and other nonportable characters in pathnames,
1349 * they are c-style quoted as needed.
1350 */
1351static void write_filename_info(const char *path)
1352{
1353 printf("filename ");
1354 write_name_quoted(NULL, 0, path, 1, stdout);
1355 putchar('\n');
1356}
1357
1358/*
1359 * The blame_entry is found to be guilty for the range. Mark it
1360 * as such, and show it in incremental output.
1361 */
1362static void found_guilty_entry(struct blame_entry *ent)
1363{
1364 if (ent->guilty)
1365 return;
1366 ent->guilty = 1;
1367 if (incremental) {
1368 struct origin *suspect = ent->suspect;
1369
1370 printf("%s %d %d %d\n",
1371 sha1_to_hex(suspect->commit->object.sha1),
1372 ent->s_lno + 1, ent->lno + 1, ent->num_lines);
1373 if (!(suspect->commit->object.flags & METAINFO_SHOWN)) {
1374 struct commit_info ci;
1375 suspect->commit->object.flags |= METAINFO_SHOWN;
1376 get_commit_info(suspect->commit, &ci, 1);
1377 printf("author %s\n", ci.author);
1378 printf("author-mail %s\n", ci.author_mail);
1379 printf("author-time %lu\n", ci.author_time);
1380 printf("author-tz %s\n", ci.author_tz);
1381 printf("committer %s\n", ci.committer);
1382 printf("committer-mail %s\n", ci.committer_mail);
1383 printf("committer-time %lu\n", ci.committer_time);
1384 printf("committer-tz %s\n", ci.committer_tz);
1385 printf("summary %s\n", ci.summary);
1386 if (suspect->commit->object.flags & UNINTERESTING)
1387 printf("boundary\n");
1388 }
1389 write_filename_info(suspect->path);
1390 }
1391}
1392
1393/*
1394 * The main loop -- while the scoreboard has lines whose true origin
1395 * is still unknown, pick one blame_entry, and allow its current
1396 * suspect to pass blames to its parents.
1397 */
1398static void assign_blame(struct scoreboard *sb, struct rev_info *revs, int opt)
1399{
1400 while (1) {
1401 struct blame_entry *ent;
1402 struct commit *commit;
1403 struct origin *suspect = NULL;
1404
1405 /* find one suspect to break down */
1406 for (ent = sb->ent; !suspect && ent; ent = ent->next)
1407 if (!ent->guilty)
1408 suspect = ent->suspect;
1409 if (!suspect)
1410 return; /* all done */
1411
1412 /*
1413 * We will use this suspect later in the loop,
1414 * so hold onto it in the meantime.
1415 */
1416 origin_incref(suspect);
1417 commit = suspect->commit;
1418 if (!commit->object.parsed)
1419 parse_commit(commit);
1420 if (!(commit->object.flags & UNINTERESTING) &&
1421 !(revs->max_age != -1 && commit->date < revs->max_age))
1422 pass_blame(sb, suspect, opt);
1423 else {
1424 commit->object.flags |= UNINTERESTING;
1425 if (commit->object.parsed)
1426 mark_parents_uninteresting(commit);
1427 }
1428 /* treat root commit as boundary */
1429 if (!commit->parents && !show_root)
1430 commit->object.flags |= UNINTERESTING;
1431
1432 /* Take responsibility for the remaining entries */
1433 for (ent = sb->ent; ent; ent = ent->next)
1434 if (!cmp_suspect(ent->suspect, suspect))
1435 found_guilty_entry(ent);
1436 origin_decref(suspect);
1437
1438 if (DEBUG) /* sanity */
1439 sanity_check_refcnt(sb);
1440 }
1441}
1442
1443static const char *format_time(unsigned long time, const char *tz_str,
1444 int show_raw_time)
1445{
1446 static char time_buf[128];
1447 time_t t = time;
1448 int minutes, tz;
1449 struct tm *tm;
1450
1451 if (show_raw_time) {
1452 sprintf(time_buf, "%lu %s", time, tz_str);
1453 return time_buf;
1454 }
1455
1456 tz = atoi(tz_str);
1457 minutes = tz < 0 ? -tz : tz;
1458 minutes = (minutes / 100)*60 + (minutes % 100);
1459 minutes = tz < 0 ? -minutes : minutes;
1460 t = time + minutes * 60;
1461 tm = gmtime(&t);
1462
1463 strftime(time_buf, sizeof(time_buf), "%Y-%m-%d %H:%M:%S ", tm);
1464 strcat(time_buf, tz_str);
1465 return time_buf;
1466}
1467
1468#define OUTPUT_ANNOTATE_COMPAT 001
1469#define OUTPUT_LONG_OBJECT_NAME 002
1470#define OUTPUT_RAW_TIMESTAMP 004
1471#define OUTPUT_PORCELAIN 010
1472#define OUTPUT_SHOW_NAME 020
1473#define OUTPUT_SHOW_NUMBER 040
1474#define OUTPUT_SHOW_SCORE 0100
1475
1476static void emit_porcelain(struct scoreboard *sb, struct blame_entry *ent)
1477{
1478 int cnt;
1479 const char *cp;
1480 struct origin *suspect = ent->suspect;
1481 char hex[41];
1482
1483 strcpy(hex, sha1_to_hex(suspect->commit->object.sha1));
1484 printf("%s%c%d %d %d\n",
1485 hex,
1486 ent->guilty ? ' ' : '*', // purely for debugging
1487 ent->s_lno + 1,
1488 ent->lno + 1,
1489 ent->num_lines);
1490 if (!(suspect->commit->object.flags & METAINFO_SHOWN)) {
1491 struct commit_info ci;
1492 suspect->commit->object.flags |= METAINFO_SHOWN;
1493 get_commit_info(suspect->commit, &ci, 1);
1494 printf("author %s\n", ci.author);
1495 printf("author-mail %s\n", ci.author_mail);
1496 printf("author-time %lu\n", ci.author_time);
1497 printf("author-tz %s\n", ci.author_tz);
1498 printf("committer %s\n", ci.committer);
1499 printf("committer-mail %s\n", ci.committer_mail);
1500 printf("committer-time %lu\n", ci.committer_time);
1501 printf("committer-tz %s\n", ci.committer_tz);
1502 write_filename_info(suspect->path);
1503 printf("summary %s\n", ci.summary);
1504 if (suspect->commit->object.flags & UNINTERESTING)
1505 printf("boundary\n");
1506 }
1507 else if (suspect->commit->object.flags & MORE_THAN_ONE_PATH)
1508 write_filename_info(suspect->path);
1509
1510 cp = nth_line(sb, ent->lno);
1511 for (cnt = 0; cnt < ent->num_lines; cnt++) {
1512 char ch;
1513 if (cnt)
1514 printf("%s %d %d\n", hex,
1515 ent->s_lno + 1 + cnt,
1516 ent->lno + 1 + cnt);
1517 putchar('\t');
1518 do {
1519 ch = *cp++;
1520 putchar(ch);
1521 } while (ch != '\n' &&
1522 cp < sb->final_buf + sb->final_buf_size);
1523 }
1524}
1525
1526static void emit_other(struct scoreboard *sb, struct blame_entry *ent, int opt)
1527{
1528 int cnt;
1529 const char *cp;
1530 struct origin *suspect = ent->suspect;
1531 struct commit_info ci;
1532 char hex[41];
1533 int show_raw_time = !!(opt & OUTPUT_RAW_TIMESTAMP);
1534
1535 get_commit_info(suspect->commit, &ci, 1);
1536 strcpy(hex, sha1_to_hex(suspect->commit->object.sha1));
1537
1538 cp = nth_line(sb, ent->lno);
1539 for (cnt = 0; cnt < ent->num_lines; cnt++) {
1540 char ch;
1541 int length = (opt & OUTPUT_LONG_OBJECT_NAME) ? 40 : 8;
1542
1543 if (suspect->commit->object.flags & UNINTERESTING) {
1544 if (!blank_boundary) {
1545 length--;
1546 putchar('^');
1547 }
1548 else
1549 memset(hex, ' ', length);
1550 }
1551
1552 printf("%.*s", length, hex);
1553 if (opt & OUTPUT_ANNOTATE_COMPAT)
1554 printf("\t(%10s\t%10s\t%d)", ci.author,
1555 format_time(ci.author_time, ci.author_tz,
1556 show_raw_time),
1557 ent->lno + 1 + cnt);
1558 else {
1559 if (opt & OUTPUT_SHOW_SCORE)
1560 printf(" %*d %02d",
1561 max_score_digits, ent->score,
1562 ent->suspect->refcnt);
1563 if (opt & OUTPUT_SHOW_NAME)
1564 printf(" %-*.*s", longest_file, longest_file,
1565 suspect->path);
1566 if (opt & OUTPUT_SHOW_NUMBER)
1567 printf(" %*d", max_orig_digits,
1568 ent->s_lno + 1 + cnt);
1569 printf(" (%-*.*s %10s %*d) ",
1570 longest_author, longest_author, ci.author,
1571 format_time(ci.author_time, ci.author_tz,
1572 show_raw_time),
1573 max_digits, ent->lno + 1 + cnt);
1574 }
1575 do {
1576 ch = *cp++;
1577 putchar(ch);
1578 } while (ch != '\n' &&
1579 cp < sb->final_buf + sb->final_buf_size);
1580 }
1581}
1582
1583static void output(struct scoreboard *sb, int option)
1584{
1585 struct blame_entry *ent;
1586
1587 if (option & OUTPUT_PORCELAIN) {
1588 for (ent = sb->ent; ent; ent = ent->next) {
1589 struct blame_entry *oth;
1590 struct origin *suspect = ent->suspect;
1591 struct commit *commit = suspect->commit;
1592 if (commit->object.flags & MORE_THAN_ONE_PATH)
1593 continue;
1594 for (oth = ent->next; oth; oth = oth->next) {
1595 if ((oth->suspect->commit != commit) ||
1596 !strcmp(oth->suspect->path, suspect->path))
1597 continue;
1598 commit->object.flags |= MORE_THAN_ONE_PATH;
1599 break;
1600 }
1601 }
1602 }
1603
1604 for (ent = sb->ent; ent; ent = ent->next) {
1605 if (option & OUTPUT_PORCELAIN)
1606 emit_porcelain(sb, ent);
1607 else {
1608 emit_other(sb, ent, option);
1609 }
1610 }
1611}
1612
1613/*
1614 * To allow quick access to the contents of nth line in the
1615 * final image, prepare an index in the scoreboard.
1616 */
1617static int prepare_lines(struct scoreboard *sb)
1618{
1619 const char *buf = sb->final_buf;
1620 unsigned long len = sb->final_buf_size;
1621 int num = 0, incomplete = 0, bol = 1;
1622
1623 if (len && buf[len-1] != '\n')
1624 incomplete++; /* incomplete line at the end */
1625 while (len--) {
1626 if (bol) {
1627 sb->lineno = xrealloc(sb->lineno,
1628 sizeof(int* ) * (num + 1));
1629 sb->lineno[num] = buf - sb->final_buf;
1630 bol = 0;
1631 }
1632 if (*buf++ == '\n') {
1633 num++;
1634 bol = 1;
1635 }
1636 }
1637 sb->lineno = xrealloc(sb->lineno,
1638 sizeof(int* ) * (num + incomplete + 1));
1639 sb->lineno[num + incomplete] = buf - sb->final_buf;
1640 sb->num_lines = num + incomplete;
1641 return sb->num_lines;
1642}
1643
1644/*
1645 * Add phony grafts for use with -S; this is primarily to
1646 * support git-cvsserver that wants to give a linear history
1647 * to its clients.
1648 */
1649static int read_ancestry(const char *graft_file)
1650{
1651 FILE *fp = fopen(graft_file, "r");
1652 char buf[1024];
1653 if (!fp)
1654 return -1;
1655 while (fgets(buf, sizeof(buf), fp)) {
1656 /* The format is just "Commit Parent1 Parent2 ...\n" */
1657 int len = strlen(buf);
1658 struct commit_graft *graft = read_graft_line(buf, len);
1659 if (graft)
1660 register_commit_graft(graft, 0);
1661 }
1662 fclose(fp);
1663 return 0;
1664}
1665
1666/*
1667 * How many columns do we need to show line numbers in decimal?
1668 */
1669static int lineno_width(int lines)
1670{
1671 int i, width;
1672
1673 for (width = 1, i = 10; i <= lines + 1; width++)
1674 i *= 10;
1675 return width;
1676}
1677
1678/*
1679 * How many columns do we need to show line numbers, authors,
1680 * and filenames?
1681 */
1682static void find_alignment(struct scoreboard *sb, int *option)
1683{
1684 int longest_src_lines = 0;
1685 int longest_dst_lines = 0;
1686 unsigned largest_score = 0;
1687 struct blame_entry *e;
1688
1689 for (e = sb->ent; e; e = e->next) {
1690 struct origin *suspect = e->suspect;
1691 struct commit_info ci;
1692 int num;
1693
1694 if (strcmp(suspect->path, sb->path))
1695 *option |= OUTPUT_SHOW_NAME;
1696 num = strlen(suspect->path);
1697 if (longest_file < num)
1698 longest_file = num;
1699 if (!(suspect->commit->object.flags & METAINFO_SHOWN)) {
1700 suspect->commit->object.flags |= METAINFO_SHOWN;
1701 get_commit_info(suspect->commit, &ci, 1);
1702 num = strlen(ci.author);
1703 if (longest_author < num)
1704 longest_author = num;
1705 }
1706 num = e->s_lno + e->num_lines;
1707 if (longest_src_lines < num)
1708 longest_src_lines = num;
1709 num = e->lno + e->num_lines;
1710 if (longest_dst_lines < num)
1711 longest_dst_lines = num;
1712 if (largest_score < ent_score(sb, e))
1713 largest_score = ent_score(sb, e);
1714 }
1715 max_orig_digits = lineno_width(longest_src_lines);
1716 max_digits = lineno_width(longest_dst_lines);
1717 max_score_digits = lineno_width(largest_score);
1718}
1719
1720/*
1721 * For debugging -- origin is refcounted, and this asserts that
1722 * we do not underflow.
1723 */
1724static void sanity_check_refcnt(struct scoreboard *sb)
1725{
1726 int baa = 0;
1727 struct blame_entry *ent;
1728
1729 for (ent = sb->ent; ent; ent = ent->next) {
1730 /* Nobody should have zero or negative refcnt */
1731 if (ent->suspect->refcnt <= 0) {
1732 fprintf(stderr, "%s in %s has negative refcnt %d\n",
1733 ent->suspect->path,
1734 sha1_to_hex(ent->suspect->commit->object.sha1),
1735 ent->suspect->refcnt);
1736 baa = 1;
1737 }
1738 }
1739 for (ent = sb->ent; ent; ent = ent->next) {
1740 /* Mark the ones that haven't been checked */
1741 if (0 < ent->suspect->refcnt)
1742 ent->suspect->refcnt = -ent->suspect->refcnt;
1743 }
1744 for (ent = sb->ent; ent; ent = ent->next) {
1745 /*
1746 * ... then pick each and see if they have the the
1747 * correct refcnt.
1748 */
1749 int found;
1750 struct blame_entry *e;
1751 struct origin *suspect = ent->suspect;
1752
1753 if (0 < suspect->refcnt)
1754 continue;
1755 suspect->refcnt = -suspect->refcnt; /* Unmark */
1756 for (found = 0, e = sb->ent; e; e = e->next) {
1757 if (e->suspect != suspect)
1758 continue;
1759 found++;
1760 }
1761 if (suspect->refcnt != found) {
1762 fprintf(stderr, "%s in %s has refcnt %d, not %d\n",
1763 ent->suspect->path,
1764 sha1_to_hex(ent->suspect->commit->object.sha1),
1765 ent->suspect->refcnt, found);
1766 baa = 2;
1767 }
1768 }
1769 if (baa) {
1770 int opt = 0160;
1771 find_alignment(sb, &opt);
1772 output(sb, opt);
1773 die("Baa %d!", baa);
1774 }
1775}
1776
1777/*
1778 * Used for the command line parsing; check if the path exists
1779 * in the working tree.
1780 */
1781static int has_path_in_work_tree(const char *path)
1782{
1783 struct stat st;
1784 return !lstat(path, &st);
1785}
1786
1787static unsigned parse_score(const char *arg)
1788{
1789 char *end;
1790 unsigned long score = strtoul(arg, &end, 10);
1791 if (*end)
1792 return 0;
1793 return score;
1794}
1795
1796static const char *add_prefix(const char *prefix, const char *path)
1797{
1798 if (!prefix || !prefix[0])
1799 return path;
1800 return prefix_path(prefix, strlen(prefix), path);
1801}
1802
1803/*
1804 * Parsing of (comma separated) one item in the -L option
1805 */
1806static const char *parse_loc(const char *spec,
1807 struct scoreboard *sb, long lno,
1808 long begin, long *ret)
1809{
1810 char *term;
1811 const char *line;
1812 long num;
1813 int reg_error;
1814 regex_t regexp;
1815 regmatch_t match[1];
1816
1817 /* Allow "-L <something>,+20" to mean starting at <something>
1818 * for 20 lines, or "-L <something>,-5" for 5 lines ending at
1819 * <something>.
1820 */
1821 if (1 < begin && (spec[0] == '+' || spec[0] == '-')) {
1822 num = strtol(spec + 1, &term, 10);
1823 if (term != spec + 1) {
1824 if (spec[0] == '-')
1825 num = 0 - num;
1826 if (0 < num)
1827 *ret = begin + num - 2;
1828 else if (!num)
1829 *ret = begin;
1830 else
1831 *ret = begin + num;
1832 return term;
1833 }
1834 return spec;
1835 }
1836 num = strtol(spec, &term, 10);
1837 if (term != spec) {
1838 *ret = num;
1839 return term;
1840 }
1841 if (spec[0] != '/')
1842 return spec;
1843
1844 /* it could be a regexp of form /.../ */
1845 for (term = (char*) spec + 1; *term && *term != '/'; term++) {
1846 if (*term == '\\')
1847 term++;
1848 }
1849 if (*term != '/')
1850 return spec;
1851
1852 /* try [spec+1 .. term-1] as regexp */
1853 *term = 0;
1854 begin--; /* input is in human terms */
1855 line = nth_line(sb, begin);
1856
1857 if (!(reg_error = regcomp(®exp, spec + 1, REG_NEWLINE)) &&
1858 !(reg_error = regexec(®exp, line, 1, match, 0))) {
1859 const char *cp = line + match[0].rm_so;
1860 const char *nline;
1861
1862 while (begin++ < lno) {
1863 nline = nth_line(sb, begin);
1864 if (line <= cp && cp < nline)
1865 break;
1866 line = nline;
1867 }
1868 *ret = begin;
1869 regfree(®exp);
1870 *term++ = '/';
1871 return term;
1872 }
1873 else {
1874 char errbuf[1024];
1875 regerror(reg_error, ®exp, errbuf, 1024);
1876 die("-L parameter '%s': %s", spec + 1, errbuf);
1877 }
1878}
1879
1880/*
1881 * Parsing of -L option
1882 */
1883static void prepare_blame_range(struct scoreboard *sb,
1884 const char *bottomtop,
1885 long lno,
1886 long *bottom, long *top)
1887{
1888 const char *term;
1889
1890 term = parse_loc(bottomtop, sb, lno, 1, bottom);
1891 if (*term == ',') {
1892 term = parse_loc(term + 1, sb, lno, *bottom + 1, top);
1893 if (*term)
1894 usage(blame_usage);
1895 }
1896 if (*term)
1897 usage(blame_usage);
1898}
1899
1900static int git_blame_config(const char *var, const char *value)
1901{
1902 if (!strcmp(var, "blame.showroot")) {
1903 show_root = git_config_bool(var, value);
1904 return 0;
1905 }
1906 if (!strcmp(var, "blame.blankboundary")) {
1907 blank_boundary = git_config_bool(var, value);
1908 return 0;
1909 }
1910 return git_default_config(var, value);
1911}
1912
1913int cmd_blame(int argc, const char **argv, const char *prefix)
1914{
1915 struct rev_info revs;
1916 const char *path;
1917 struct scoreboard sb;
1918 struct origin *o;
1919 struct blame_entry *ent;
1920 int i, seen_dashdash, unk, opt;
1921 long bottom, top, lno;
1922 int output_option = 0;
1923 const char *revs_file = NULL;
1924 const char *final_commit_name = NULL;
1925 char type[10];
1926 const char *bottomtop = NULL;
1927
1928 git_config(git_blame_config);
1929 save_commit_buffer = 0;
1930
1931 opt = 0;
1932 seen_dashdash = 0;
1933 for (unk = i = 1; i < argc; i++) {
1934 const char *arg = argv[i];
1935 if (*arg != '-')
1936 break;
1937 else if (!strcmp("-b", arg))
1938 blank_boundary = 1;
1939 else if (!strcmp("--root", arg))
1940 show_root = 1;
1941 else if (!strcmp("-c", arg))
1942 output_option |= OUTPUT_ANNOTATE_COMPAT;
1943 else if (!strcmp("-t", arg))
1944 output_option |= OUTPUT_RAW_TIMESTAMP;
1945 else if (!strcmp("-l", arg))
1946 output_option |= OUTPUT_LONG_OBJECT_NAME;
1947 else if (!strcmp("-S", arg) && ++i < argc)
1948 revs_file = argv[i];
1949 else if (!strncmp("-M", arg, 2)) {
1950 opt |= PICKAXE_BLAME_MOVE;
1951 blame_move_score = parse_score(arg+2);
1952 }
1953 else if (!strncmp("-C", arg, 2)) {
1954 if (opt & PICKAXE_BLAME_COPY)
1955 opt |= PICKAXE_BLAME_COPY_HARDER;
1956 opt |= PICKAXE_BLAME_COPY | PICKAXE_BLAME_MOVE;
1957 blame_copy_score = parse_score(arg+2);
1958 }
1959 else if (!strncmp("-L", arg, 2)) {
1960 if (!arg[2]) {
1961 if (++i >= argc)
1962 usage(blame_usage);
1963 arg = argv[i];
1964 }
1965 else
1966 arg += 2;
1967 if (bottomtop)
1968 die("More than one '-L n,m' option given");
1969 bottomtop = arg;
1970 }
1971 else if (!strcmp("--incremental", arg))
1972 incremental = 1;
1973 else if (!strcmp("--score-debug", arg))
1974 output_option |= OUTPUT_SHOW_SCORE;
1975 else if (!strcmp("-f", arg) ||
1976 !strcmp("--show-name", arg))
1977 output_option |= OUTPUT_SHOW_NAME;
1978 else if (!strcmp("-n", arg) ||
1979 !strcmp("--show-number", arg))
1980 output_option |= OUTPUT_SHOW_NUMBER;
1981 else if (!strcmp("-p", arg) ||
1982 !strcmp("--porcelain", arg))
1983 output_option |= OUTPUT_PORCELAIN;
1984 else if (!strcmp("--", arg)) {
1985 seen_dashdash = 1;
1986 i++;
1987 break;
1988 }
1989 else
1990 argv[unk++] = arg;
1991 }
1992
1993 if (!incremental)
1994 setup_pager();
1995
1996 if (!blame_move_score)
1997 blame_move_score = BLAME_DEFAULT_MOVE_SCORE;
1998 if (!blame_copy_score)
1999 blame_copy_score = BLAME_DEFAULT_COPY_SCORE;
2000
2001 /*
2002 * We have collected options unknown to us in argv[1..unk]
2003 * which are to be passed to revision machinery if we are
2004 * going to do the "bottom" processing.
2005 *
2006 * The remaining are:
2007 *
2008 * (1) if seen_dashdash, its either
2009 * "-options -- <path>" or
2010 * "-options -- <path> <rev>".
2011 * but the latter is allowed only if there is no
2012 * options that we passed to revision machinery.
2013 *
2014 * (2) otherwise, we may have "--" somewhere later and
2015 * might be looking at the first one of multiple 'rev'
2016 * parameters (e.g. " master ^next ^maint -- path").
2017 * See if there is a dashdash first, and give the
2018 * arguments before that to revision machinery.
2019 * After that there must be one 'path'.
2020 *
2021 * (3) otherwise, its one of the three:
2022 * "-options <path> <rev>"
2023 * "-options <rev> <path>"
2024 * "-options <path>"
2025 * but again the first one is allowed only if
2026 * there is no options that we passed to revision
2027 * machinery.
2028 */
2029
2030 if (seen_dashdash) {
2031 /* (1) */
2032 if (argc <= i)
2033 usage(blame_usage);
2034 path = add_prefix(prefix, argv[i]);
2035 if (i + 1 == argc - 1) {
2036 if (unk != 1)
2037 usage(blame_usage);
2038 argv[unk++] = argv[i + 1];
2039 }
2040 else if (i + 1 != argc)
2041 /* garbage at end */
2042 usage(blame_usage);
2043 }
2044 else {
2045 int j;
2046 for (j = i; !seen_dashdash && j < argc; j++)
2047 if (!strcmp(argv[j], "--"))
2048 seen_dashdash = j;
2049 if (seen_dashdash) {
2050 if (seen_dashdash + 1 != argc - 1)
2051 usage(blame_usage);
2052 path = add_prefix(prefix, argv[seen_dashdash + 1]);
2053 for (j = i; j < seen_dashdash; j++)
2054 argv[unk++] = argv[j];
2055 }
2056 else {
2057 /* (3) */
2058 path = add_prefix(prefix, argv[i]);
2059 if (i + 1 == argc - 1) {
2060 final_commit_name = argv[i + 1];
2061
2062 /* if (unk == 1) we could be getting
2063 * old-style
2064 */
2065 if (unk == 1 && !has_path_in_work_tree(path)) {
2066 path = add_prefix(prefix, argv[i + 1]);
2067 final_commit_name = argv[i];
2068 }
2069 }
2070 else if (i != argc - 1)
2071 usage(blame_usage); /* garbage at end */
2072
2073 if (!has_path_in_work_tree(path))
2074 die("cannot stat path %s: %s",
2075 path, strerror(errno));
2076 }
2077 }
2078
2079 if (final_commit_name)
2080 argv[unk++] = final_commit_name;
2081
2082 /*
2083 * Now we got rev and path. We do not want the path pruning
2084 * but we may want "bottom" processing.
2085 */
2086 argv[unk++] = "--"; /* terminate the rev name */
2087 argv[unk] = NULL;
2088
2089 init_revisions(&revs, NULL);
2090 setup_revisions(unk, argv, &revs, "HEAD");
2091 memset(&sb, 0, sizeof(sb));
2092
2093 /*
2094 * There must be one and only one positive commit in the
2095 * revs->pending array.
2096 */
2097 for (i = 0; i < revs.pending.nr; i++) {
2098 struct object *obj = revs.pending.objects[i].item;
2099 if (obj->flags & UNINTERESTING)
2100 continue;
2101 while (obj->type == OBJ_TAG)
2102 obj = deref_tag(obj, NULL, 0);
2103 if (obj->type != OBJ_COMMIT)
2104 die("Non commit %s?",
2105 revs.pending.objects[i].name);
2106 if (sb.final)
2107 die("More than one commit to dig from %s and %s?",
2108 revs.pending.objects[i].name,
2109 final_commit_name);
2110 sb.final = (struct commit *) obj;
2111 final_commit_name = revs.pending.objects[i].name;
2112 }
2113
2114 if (!sb.final) {
2115 /*
2116 * "--not A B -- path" without anything positive;
2117 * default to HEAD.
2118 */
2119 unsigned char head_sha1[20];
2120
2121 final_commit_name = "HEAD";
2122 if (get_sha1(final_commit_name, head_sha1))
2123 die("No such ref: HEAD");
2124 sb.final = lookup_commit_reference(head_sha1);
2125 add_pending_object(&revs, &(sb.final->object), "HEAD");
2126 }
2127
2128 /*
2129 * If we have bottom, this will mark the ancestors of the
2130 * bottom commits we would reach while traversing as
2131 * uninteresting.
2132 */
2133 prepare_revision_walk(&revs);
2134
2135 o = get_origin(&sb, sb.final, path);
2136 if (fill_blob_sha1(o))
2137 die("no such path %s in %s", path, final_commit_name);
2138
2139 sb.final_buf = read_sha1_file(o->blob_sha1, type, &sb.final_buf_size);
2140 num_read_blob++;
2141 lno = prepare_lines(&sb);
2142
2143 bottom = top = 0;
2144 if (bottomtop)
2145 prepare_blame_range(&sb, bottomtop, lno, &bottom, &top);
2146 if (bottom && top && top < bottom) {
2147 long tmp;
2148 tmp = top; top = bottom; bottom = tmp;
2149 }
2150 if (bottom < 1)
2151 bottom = 1;
2152 if (top < 1)
2153 top = lno;
2154 bottom--;
2155 if (lno < top)
2156 die("file %s has only %lu lines", path, lno);
2157
2158 ent = xcalloc(1, sizeof(*ent));
2159 ent->lno = bottom;
2160 ent->num_lines = top - bottom;
2161 ent->suspect = o;
2162 ent->s_lno = bottom;
2163
2164 sb.ent = ent;
2165 sb.path = path;
2166
2167 if (revs_file && read_ancestry(revs_file))
2168 die("reading graft file %s failed: %s",
2169 revs_file, strerror(errno));
2170
2171 assign_blame(&sb, &revs, opt);
2172
2173 if (incremental)
2174 return 0;
2175
2176 coalesce(&sb);
2177
2178 if (!(output_option & OUTPUT_PORCELAIN))
2179 find_alignment(&sb, &output_option);
2180
2181 output(&sb, output_option);
2182 free((void *)sb.final_buf);
2183 for (ent = sb.ent; ent; ) {
2184 struct blame_entry *e = ent->next;
2185 free(ent);
2186 ent = e;
2187 }
2188
2189 if (DEBUG) {
2190 printf("num read blob: %d\n", num_read_blob);
2191 printf("num get patch: %d\n", num_get_patch);
2192 printf("num commits: %d\n", num_commits);
2193 }
2194 return 0;
2195}