Add new testcase to show fast-export does not always exports all tags
[git/jrn.git] / builtin-blame.c
blob101c4162dad22db14ac87257b53496ec558137aa
1 /*
2 * Pickaxe
4 * Copyright (c) 2006, Junio C Hamano
5 */
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"
23 static char blame_usage[] = "git blame [options] [rev-opts] [rev] [--] file";
25 static const char *blame_opt_usage[] = {
26 blame_usage,
27 "",
28 "[rev-opts] are documented in git-rev-list(1)",
29 NULL
32 static int longest_file;
33 static int longest_author;
34 static int max_orig_digits;
35 static int max_digits;
36 static int max_score_digits;
37 static int show_root;
38 static int reverse;
39 static int blank_boundary;
40 static int incremental;
41 static int xdl_opts = XDF_NEED_MINIMAL;
42 static struct string_list mailmap;
44 #ifndef DEBUG
45 #define DEBUG 0
46 #endif
48 /* stats */
49 static int num_read_blob;
50 static int num_get_patch;
51 static int num_commits;
53 #define PICKAXE_BLAME_MOVE 01
54 #define PICKAXE_BLAME_COPY 02
55 #define PICKAXE_BLAME_COPY_HARDER 04
56 #define PICKAXE_BLAME_COPY_HARDEST 010
59 * blame for a blame_entry with score lower than these thresholds
60 * is not passed to the parent using move/copy logic.
62 static unsigned blame_move_score;
63 static unsigned blame_copy_score;
64 #define BLAME_DEFAULT_MOVE_SCORE 20
65 #define BLAME_DEFAULT_COPY_SCORE 40
67 /* bits #0..7 in revision.h, #8..11 used for merge_bases() in commit.c */
68 #define METAINFO_SHOWN (1u<<12)
69 #define MORE_THAN_ONE_PATH (1u<<13)
72 * One blob in a commit that is being suspected
74 struct origin {
75 int refcnt;
76 struct commit *commit;
77 mmfile_t file;
78 unsigned char blob_sha1[20];
79 char path[FLEX_ARRAY];
83 * Given an origin, prepare mmfile_t structure to be used by the
84 * diff machinery
86 static void fill_origin_blob(struct origin *o, mmfile_t *file)
88 if (!o->file.ptr) {
89 enum object_type type;
90 num_read_blob++;
91 file->ptr = read_sha1_file(o->blob_sha1, &type,
92 (unsigned long *)(&(file->size)));
93 if (!file->ptr)
94 die("Cannot read blob %s for path %s",
95 sha1_to_hex(o->blob_sha1),
96 o->path);
97 o->file = *file;
99 else
100 *file = o->file;
104 * Origin is refcounted and usually we keep the blob contents to be
105 * reused.
107 static inline struct origin *origin_incref(struct origin *o)
109 if (o)
110 o->refcnt++;
111 return o;
114 static void origin_decref(struct origin *o)
116 if (o && --o->refcnt <= 0) {
117 free(o->file.ptr);
118 free(o);
122 static void drop_origin_blob(struct origin *o)
124 if (o->file.ptr) {
125 free(o->file.ptr);
126 o->file.ptr = NULL;
131 * Each group of lines is described by a blame_entry; it can be split
132 * as we pass blame to the parents. They form a linked list in the
133 * scoreboard structure, sorted by the target line number.
135 struct blame_entry {
136 struct blame_entry *prev;
137 struct blame_entry *next;
139 /* the first line of this group in the final image;
140 * internally all line numbers are 0 based.
142 int lno;
144 /* how many lines this group has */
145 int num_lines;
147 /* the commit that introduced this group into the final image */
148 struct origin *suspect;
150 /* true if the suspect is truly guilty; false while we have not
151 * checked if the group came from one of its parents.
153 char guilty;
155 /* true if the entry has been scanned for copies in the current parent
157 char scanned;
159 /* the line number of the first line of this group in the
160 * suspect's file; internally all line numbers are 0 based.
162 int s_lno;
164 /* how significant this entry is -- cached to avoid
165 * scanning the lines over and over.
167 unsigned score;
171 * The current state of the blame assignment.
173 struct scoreboard {
174 /* the final commit (i.e. where we started digging from) */
175 struct commit *final;
176 struct rev_info *revs;
177 const char *path;
180 * The contents in the final image.
181 * Used by many functions to obtain contents of the nth line,
182 * indexed with scoreboard.lineno[blame_entry.lno].
184 const char *final_buf;
185 unsigned long final_buf_size;
187 /* linked list of blames */
188 struct blame_entry *ent;
190 /* look-up a line in the final buffer */
191 int num_lines;
192 int *lineno;
195 static inline int same_suspect(struct origin *a, struct origin *b)
197 if (a == b)
198 return 1;
199 if (a->commit != b->commit)
200 return 0;
201 return !strcmp(a->path, b->path);
204 static void sanity_check_refcnt(struct scoreboard *);
207 * If two blame entries that are next to each other came from
208 * contiguous lines in the same origin (i.e. <commit, path> pair),
209 * merge them together.
211 static void coalesce(struct scoreboard *sb)
213 struct blame_entry *ent, *next;
215 for (ent = sb->ent; ent && (next = ent->next); ent = next) {
216 if (same_suspect(ent->suspect, next->suspect) &&
217 ent->guilty == next->guilty &&
218 ent->s_lno + ent->num_lines == next->s_lno) {
219 ent->num_lines += next->num_lines;
220 ent->next = next->next;
221 if (ent->next)
222 ent->next->prev = ent;
223 origin_decref(next->suspect);
224 free(next);
225 ent->score = 0;
226 next = ent; /* again */
230 if (DEBUG) /* sanity */
231 sanity_check_refcnt(sb);
235 * Given a commit and a path in it, create a new origin structure.
236 * The callers that add blame to the scoreboard should use
237 * get_origin() to obtain shared, refcounted copy instead of calling
238 * this function directly.
240 static struct origin *make_origin(struct commit *commit, const char *path)
242 struct origin *o;
243 o = xcalloc(1, sizeof(*o) + strlen(path) + 1);
244 o->commit = commit;
245 o->refcnt = 1;
246 strcpy(o->path, path);
247 return o;
251 * Locate an existing origin or create a new one.
253 static struct origin *get_origin(struct scoreboard *sb,
254 struct commit *commit,
255 const char *path)
257 struct blame_entry *e;
259 for (e = sb->ent; e; e = e->next) {
260 if (e->suspect->commit == commit &&
261 !strcmp(e->suspect->path, path))
262 return origin_incref(e->suspect);
264 return make_origin(commit, path);
268 * Fill the blob_sha1 field of an origin if it hasn't, so that later
269 * call to fill_origin_blob() can use it to locate the data. blob_sha1
270 * for an origin is also used to pass the blame for the entire file to
271 * the parent to detect the case where a child's blob is identical to
272 * that of its parent's.
274 static int fill_blob_sha1(struct origin *origin)
276 unsigned mode;
278 if (!is_null_sha1(origin->blob_sha1))
279 return 0;
280 if (get_tree_entry(origin->commit->object.sha1,
281 origin->path,
282 origin->blob_sha1, &mode))
283 goto error_out;
284 if (sha1_object_info(origin->blob_sha1, NULL) != OBJ_BLOB)
285 goto error_out;
286 return 0;
287 error_out:
288 hashclr(origin->blob_sha1);
289 return -1;
293 * We have an origin -- check if the same path exists in the
294 * parent and return an origin structure to represent it.
296 static struct origin *find_origin(struct scoreboard *sb,
297 struct commit *parent,
298 struct origin *origin)
300 struct origin *porigin = NULL;
301 struct diff_options diff_opts;
302 const char *paths[2];
304 if (parent->util) {
306 * Each commit object can cache one origin in that
307 * commit. This is a freestanding copy of origin and
308 * not refcounted.
310 struct origin *cached = parent->util;
311 if (!strcmp(cached->path, origin->path)) {
313 * The same path between origin and its parent
314 * without renaming -- the most common case.
316 porigin = get_origin(sb, parent, cached->path);
319 * If the origin was newly created (i.e. get_origin
320 * would call make_origin if none is found in the
321 * scoreboard), it does not know the blob_sha1,
322 * so copy it. Otherwise porigin was in the
323 * scoreboard and already knows blob_sha1.
325 if (porigin->refcnt == 1)
326 hashcpy(porigin->blob_sha1, cached->blob_sha1);
327 return porigin;
329 /* otherwise it was not very useful; free it */
330 free(parent->util);
331 parent->util = NULL;
334 /* See if the origin->path is different between parent
335 * and origin first. Most of the time they are the
336 * same and diff-tree is fairly efficient about this.
338 diff_setup(&diff_opts);
339 DIFF_OPT_SET(&diff_opts, RECURSIVE);
340 diff_opts.detect_rename = 0;
341 diff_opts.output_format = DIFF_FORMAT_NO_OUTPUT;
342 paths[0] = origin->path;
343 paths[1] = NULL;
345 diff_tree_setup_paths(paths, &diff_opts);
346 if (diff_setup_done(&diff_opts) < 0)
347 die("diff-setup");
349 if (is_null_sha1(origin->commit->object.sha1))
350 do_diff_cache(parent->tree->object.sha1, &diff_opts);
351 else
352 diff_tree_sha1(parent->tree->object.sha1,
353 origin->commit->tree->object.sha1,
354 "", &diff_opts);
355 diffcore_std(&diff_opts);
357 /* It is either one entry that says "modified", or "created",
358 * or nothing.
360 if (!diff_queued_diff.nr) {
361 /* The path is the same as parent */
362 porigin = get_origin(sb, parent, origin->path);
363 hashcpy(porigin->blob_sha1, origin->blob_sha1);
365 else if (diff_queued_diff.nr != 1)
366 die("internal error in blame::find_origin");
367 else {
368 struct diff_filepair *p = diff_queued_diff.queue[0];
369 switch (p->status) {
370 default:
371 die("internal error in blame::find_origin (%c)",
372 p->status);
373 case 'M':
374 porigin = get_origin(sb, parent, origin->path);
375 hashcpy(porigin->blob_sha1, p->one->sha1);
376 break;
377 case 'A':
378 case 'T':
379 /* Did not exist in parent, or type changed */
380 break;
383 diff_flush(&diff_opts);
384 diff_tree_release_paths(&diff_opts);
385 if (porigin) {
387 * Create a freestanding copy that is not part of
388 * the refcounted origin found in the scoreboard, and
389 * cache it in the commit.
391 struct origin *cached;
393 cached = make_origin(porigin->commit, porigin->path);
394 hashcpy(cached->blob_sha1, porigin->blob_sha1);
395 parent->util = cached;
397 return porigin;
401 * We have an origin -- find the path that corresponds to it in its
402 * parent and return an origin structure to represent it.
404 static struct origin *find_rename(struct scoreboard *sb,
405 struct commit *parent,
406 struct origin *origin)
408 struct origin *porigin = NULL;
409 struct diff_options diff_opts;
410 int i;
411 const char *paths[2];
413 diff_setup(&diff_opts);
414 DIFF_OPT_SET(&diff_opts, RECURSIVE);
415 diff_opts.detect_rename = DIFF_DETECT_RENAME;
416 diff_opts.output_format = DIFF_FORMAT_NO_OUTPUT;
417 diff_opts.single_follow = origin->path;
418 paths[0] = NULL;
419 diff_tree_setup_paths(paths, &diff_opts);
420 if (diff_setup_done(&diff_opts) < 0)
421 die("diff-setup");
423 if (is_null_sha1(origin->commit->object.sha1))
424 do_diff_cache(parent->tree->object.sha1, &diff_opts);
425 else
426 diff_tree_sha1(parent->tree->object.sha1,
427 origin->commit->tree->object.sha1,
428 "", &diff_opts);
429 diffcore_std(&diff_opts);
431 for (i = 0; i < diff_queued_diff.nr; i++) {
432 struct diff_filepair *p = diff_queued_diff.queue[i];
433 if ((p->status == 'R' || p->status == 'C') &&
434 !strcmp(p->two->path, origin->path)) {
435 porigin = get_origin(sb, parent, p->one->path);
436 hashcpy(porigin->blob_sha1, p->one->sha1);
437 break;
440 diff_flush(&diff_opts);
441 diff_tree_release_paths(&diff_opts);
442 return porigin;
446 * Parsing of patch chunks...
448 struct chunk {
449 /* line number in postimage; up to but not including this
450 * line is the same as preimage
452 int same;
454 /* preimage line number after this chunk */
455 int p_next;
457 /* postimage line number after this chunk */
458 int t_next;
461 struct patch {
462 struct chunk *chunks;
463 int num;
466 struct blame_diff_state {
467 struct xdiff_emit_state xm;
468 struct patch *ret;
469 unsigned hunk_post_context;
470 unsigned hunk_in_pre_context : 1;
473 static void process_u_diff(void *state_, char *line, unsigned long len)
475 struct blame_diff_state *state = state_;
476 struct chunk *chunk;
477 int off1, off2, len1, len2, num;
479 num = state->ret->num;
480 if (len < 4 || line[0] != '@' || line[1] != '@') {
481 if (state->hunk_in_pre_context && line[0] == ' ')
482 state->ret->chunks[num - 1].same++;
483 else {
484 state->hunk_in_pre_context = 0;
485 if (line[0] == ' ')
486 state->hunk_post_context++;
487 else
488 state->hunk_post_context = 0;
490 return;
493 if (num && state->hunk_post_context) {
494 chunk = &state->ret->chunks[num - 1];
495 chunk->p_next -= state->hunk_post_context;
496 chunk->t_next -= state->hunk_post_context;
498 state->ret->num = ++num;
499 state->ret->chunks = xrealloc(state->ret->chunks,
500 sizeof(struct chunk) * num);
501 chunk = &state->ret->chunks[num - 1];
502 if (parse_hunk_header(line, len, &off1, &len1, &off2, &len2)) {
503 state->ret->num--;
504 return;
507 /* Line numbers in patch output are one based. */
508 off1--;
509 off2--;
511 chunk->same = len2 ? off2 : (off2 + 1);
513 chunk->p_next = off1 + (len1 ? len1 : 1);
514 chunk->t_next = chunk->same + len2;
515 state->hunk_in_pre_context = 1;
516 state->hunk_post_context = 0;
519 static struct patch *compare_buffer(mmfile_t *file_p, mmfile_t *file_o,
520 int context)
522 struct blame_diff_state state;
523 xpparam_t xpp;
524 xdemitconf_t xecfg;
525 xdemitcb_t ecb;
527 xpp.flags = xdl_opts;
528 memset(&xecfg, 0, sizeof(xecfg));
529 xecfg.ctxlen = context;
530 ecb.outf = xdiff_outf;
531 ecb.priv = &state;
532 memset(&state, 0, sizeof(state));
533 state.xm.consume = process_u_diff;
534 state.ret = xmalloc(sizeof(struct patch));
535 state.ret->chunks = NULL;
536 state.ret->num = 0;
538 xdi_diff(file_p, file_o, &xpp, &xecfg, &ecb);
540 if (state.ret->num) {
541 struct chunk *chunk;
542 chunk = &state.ret->chunks[state.ret->num - 1];
543 chunk->p_next -= state.hunk_post_context;
544 chunk->t_next -= state.hunk_post_context;
546 return state.ret;
550 * Run diff between two origins and grab the patch output, so that
551 * we can pass blame for lines origin is currently suspected for
552 * to its parent.
554 static struct patch *get_patch(struct origin *parent, struct origin *origin)
556 mmfile_t file_p, file_o;
557 struct patch *patch;
559 fill_origin_blob(parent, &file_p);
560 fill_origin_blob(origin, &file_o);
561 if (!file_p.ptr || !file_o.ptr)
562 return NULL;
563 patch = compare_buffer(&file_p, &file_o, 0);
564 num_get_patch++;
565 return patch;
568 static void free_patch(struct patch *p)
570 free(p->chunks);
571 free(p);
575 * Link in a new blame entry to the scoreboard. Entries that cover the
576 * same line range have been removed from the scoreboard previously.
578 static void add_blame_entry(struct scoreboard *sb, struct blame_entry *e)
580 struct blame_entry *ent, *prev = NULL;
582 origin_incref(e->suspect);
584 for (ent = sb->ent; ent && ent->lno < e->lno; ent = ent->next)
585 prev = ent;
587 /* prev, if not NULL, is the last one that is below e */
588 e->prev = prev;
589 if (prev) {
590 e->next = prev->next;
591 prev->next = e;
593 else {
594 e->next = sb->ent;
595 sb->ent = e;
597 if (e->next)
598 e->next->prev = e;
602 * src typically is on-stack; we want to copy the information in it to
603 * a malloced blame_entry that is already on the linked list of the
604 * scoreboard. The origin of dst loses a refcnt while the origin of src
605 * gains one.
607 static void dup_entry(struct blame_entry *dst, struct blame_entry *src)
609 struct blame_entry *p, *n;
611 p = dst->prev;
612 n = dst->next;
613 origin_incref(src->suspect);
614 origin_decref(dst->suspect);
615 memcpy(dst, src, sizeof(*src));
616 dst->prev = p;
617 dst->next = n;
618 dst->score = 0;
621 static const char *nth_line(struct scoreboard *sb, int lno)
623 return sb->final_buf + sb->lineno[lno];
627 * It is known that lines between tlno to same came from parent, and e
628 * has an overlap with that range. it also is known that parent's
629 * line plno corresponds to e's line tlno.
631 * <---- e ----->
632 * <------>
633 * <------------>
634 * <------------>
635 * <------------------>
637 * Split e into potentially three parts; before this chunk, the chunk
638 * to be blamed for the parent, and after that portion.
640 static void split_overlap(struct blame_entry *split,
641 struct blame_entry *e,
642 int tlno, int plno, int same,
643 struct origin *parent)
645 int chunk_end_lno;
646 memset(split, 0, sizeof(struct blame_entry [3]));
648 if (e->s_lno < tlno) {
649 /* there is a pre-chunk part not blamed on parent */
650 split[0].suspect = origin_incref(e->suspect);
651 split[0].lno = e->lno;
652 split[0].s_lno = e->s_lno;
653 split[0].num_lines = tlno - e->s_lno;
654 split[1].lno = e->lno + tlno - e->s_lno;
655 split[1].s_lno = plno;
657 else {
658 split[1].lno = e->lno;
659 split[1].s_lno = plno + (e->s_lno - tlno);
662 if (same < e->s_lno + e->num_lines) {
663 /* there is a post-chunk part not blamed on parent */
664 split[2].suspect = origin_incref(e->suspect);
665 split[2].lno = e->lno + (same - e->s_lno);
666 split[2].s_lno = e->s_lno + (same - e->s_lno);
667 split[2].num_lines = e->s_lno + e->num_lines - same;
668 chunk_end_lno = split[2].lno;
670 else
671 chunk_end_lno = e->lno + e->num_lines;
672 split[1].num_lines = chunk_end_lno - split[1].lno;
675 * if it turns out there is nothing to blame the parent for,
676 * forget about the splitting. !split[1].suspect signals this.
678 if (split[1].num_lines < 1)
679 return;
680 split[1].suspect = origin_incref(parent);
684 * split_overlap() divided an existing blame e into up to three parts
685 * in split. Adjust the linked list of blames in the scoreboard to
686 * reflect the split.
688 static void split_blame(struct scoreboard *sb,
689 struct blame_entry *split,
690 struct blame_entry *e)
692 struct blame_entry *new_entry;
694 if (split[0].suspect && split[2].suspect) {
695 /* The first part (reuse storage for the existing entry e) */
696 dup_entry(e, &split[0]);
698 /* The last part -- me */
699 new_entry = xmalloc(sizeof(*new_entry));
700 memcpy(new_entry, &(split[2]), sizeof(struct blame_entry));
701 add_blame_entry(sb, new_entry);
703 /* ... and the middle part -- parent */
704 new_entry = xmalloc(sizeof(*new_entry));
705 memcpy(new_entry, &(split[1]), sizeof(struct blame_entry));
706 add_blame_entry(sb, new_entry);
708 else if (!split[0].suspect && !split[2].suspect)
710 * The parent covers the entire area; reuse storage for
711 * e and replace it with the parent.
713 dup_entry(e, &split[1]);
714 else if (split[0].suspect) {
715 /* me and then parent */
716 dup_entry(e, &split[0]);
718 new_entry = xmalloc(sizeof(*new_entry));
719 memcpy(new_entry, &(split[1]), sizeof(struct blame_entry));
720 add_blame_entry(sb, new_entry);
722 else {
723 /* parent and then me */
724 dup_entry(e, &split[1]);
726 new_entry = xmalloc(sizeof(*new_entry));
727 memcpy(new_entry, &(split[2]), sizeof(struct blame_entry));
728 add_blame_entry(sb, new_entry);
731 if (DEBUG) { /* sanity */
732 struct blame_entry *ent;
733 int lno = sb->ent->lno, corrupt = 0;
735 for (ent = sb->ent; ent; ent = ent->next) {
736 if (lno != ent->lno)
737 corrupt = 1;
738 if (ent->s_lno < 0)
739 corrupt = 1;
740 lno += ent->num_lines;
742 if (corrupt) {
743 lno = sb->ent->lno;
744 for (ent = sb->ent; ent; ent = ent->next) {
745 printf("L %8d l %8d n %8d\n",
746 lno, ent->lno, ent->num_lines);
747 lno = ent->lno + ent->num_lines;
749 die("oops");
755 * After splitting the blame, the origins used by the
756 * on-stack blame_entry should lose one refcnt each.
758 static void decref_split(struct blame_entry *split)
760 int i;
762 for (i = 0; i < 3; i++)
763 origin_decref(split[i].suspect);
767 * Helper for blame_chunk(). blame_entry e is known to overlap with
768 * the patch hunk; split it and pass blame to the parent.
770 static void blame_overlap(struct scoreboard *sb, struct blame_entry *e,
771 int tlno, int plno, int same,
772 struct origin *parent)
774 struct blame_entry split[3];
776 split_overlap(split, e, tlno, plno, same, parent);
777 if (split[1].suspect)
778 split_blame(sb, split, e);
779 decref_split(split);
783 * Find the line number of the last line the target is suspected for.
785 static int find_last_in_target(struct scoreboard *sb, struct origin *target)
787 struct blame_entry *e;
788 int last_in_target = -1;
790 for (e = sb->ent; e; e = e->next) {
791 if (e->guilty || !same_suspect(e->suspect, target))
792 continue;
793 if (last_in_target < e->s_lno + e->num_lines)
794 last_in_target = e->s_lno + e->num_lines;
796 return last_in_target;
800 * Process one hunk from the patch between the current suspect for
801 * blame_entry e and its parent. Find and split the overlap, and
802 * pass blame to the overlapping part to the parent.
804 static void blame_chunk(struct scoreboard *sb,
805 int tlno, int plno, int same,
806 struct origin *target, struct origin *parent)
808 struct blame_entry *e;
810 for (e = sb->ent; e; e = e->next) {
811 if (e->guilty || !same_suspect(e->suspect, target))
812 continue;
813 if (same <= e->s_lno)
814 continue;
815 if (tlno < e->s_lno + e->num_lines)
816 blame_overlap(sb, e, tlno, plno, same, parent);
821 * We are looking at the origin 'target' and aiming to pass blame
822 * for the lines it is suspected to its parent. Run diff to find
823 * which lines came from parent and pass blame for them.
825 static int pass_blame_to_parent(struct scoreboard *sb,
826 struct origin *target,
827 struct origin *parent)
829 int i, last_in_target, plno, tlno;
830 struct patch *patch;
832 last_in_target = find_last_in_target(sb, target);
833 if (last_in_target < 0)
834 return 1; /* nothing remains for this target */
836 patch = get_patch(parent, target);
837 plno = tlno = 0;
838 for (i = 0; i < patch->num; i++) {
839 struct chunk *chunk = &patch->chunks[i];
841 blame_chunk(sb, tlno, plno, chunk->same, target, parent);
842 plno = chunk->p_next;
843 tlno = chunk->t_next;
845 /* The rest (i.e. anything after tlno) are the same as the parent */
846 blame_chunk(sb, tlno, plno, last_in_target, target, parent);
848 free_patch(patch);
849 return 0;
853 * The lines in blame_entry after splitting blames many times can become
854 * very small and trivial, and at some point it becomes pointless to
855 * blame the parents. E.g. "\t\t}\n\t}\n\n" appears everywhere in any
856 * ordinary C program, and it is not worth to say it was copied from
857 * totally unrelated file in the parent.
859 * Compute how trivial the lines in the blame_entry are.
861 static unsigned ent_score(struct scoreboard *sb, struct blame_entry *e)
863 unsigned score;
864 const char *cp, *ep;
866 if (e->score)
867 return e->score;
869 score = 1;
870 cp = nth_line(sb, e->lno);
871 ep = nth_line(sb, e->lno + e->num_lines);
872 while (cp < ep) {
873 unsigned ch = *((unsigned char *)cp);
874 if (isalnum(ch))
875 score++;
876 cp++;
878 e->score = score;
879 return score;
883 * best_so_far[] and this[] are both a split of an existing blame_entry
884 * that passes blame to the parent. Maintain best_so_far the best split
885 * so far, by comparing this and best_so_far and copying this into
886 * bst_so_far as needed.
888 static void copy_split_if_better(struct scoreboard *sb,
889 struct blame_entry *best_so_far,
890 struct blame_entry *this)
892 int i;
894 if (!this[1].suspect)
895 return;
896 if (best_so_far[1].suspect) {
897 if (ent_score(sb, &this[1]) < ent_score(sb, &best_so_far[1]))
898 return;
901 for (i = 0; i < 3; i++)
902 origin_incref(this[i].suspect);
903 decref_split(best_so_far);
904 memcpy(best_so_far, this, sizeof(struct blame_entry [3]));
908 * We are looking at a part of the final image represented by
909 * ent (tlno and same are offset by ent->s_lno).
910 * tlno is where we are looking at in the final image.
911 * up to (but not including) same match preimage.
912 * plno is where we are looking at in the preimage.
914 * <-------------- final image ---------------------->
915 * <------ent------>
916 * ^tlno ^same
917 * <---------preimage----->
918 * ^plno
920 * All line numbers are 0-based.
922 static void handle_split(struct scoreboard *sb,
923 struct blame_entry *ent,
924 int tlno, int plno, int same,
925 struct origin *parent,
926 struct blame_entry *split)
928 if (ent->num_lines <= tlno)
929 return;
930 if (tlno < same) {
931 struct blame_entry this[3];
932 tlno += ent->s_lno;
933 same += ent->s_lno;
934 split_overlap(this, ent, tlno, plno, same, parent);
935 copy_split_if_better(sb, split, this);
936 decref_split(this);
941 * Find the lines from parent that are the same as ent so that
942 * we can pass blames to it. file_p has the blob contents for
943 * the parent.
945 static void find_copy_in_blob(struct scoreboard *sb,
946 struct blame_entry *ent,
947 struct origin *parent,
948 struct blame_entry *split,
949 mmfile_t *file_p)
951 const char *cp;
952 int cnt;
953 mmfile_t file_o;
954 struct patch *patch;
955 int i, plno, tlno;
958 * Prepare mmfile that contains only the lines in ent.
960 cp = nth_line(sb, ent->lno);
961 file_o.ptr = (char*) cp;
962 cnt = ent->num_lines;
964 while (cnt && cp < sb->final_buf + sb->final_buf_size) {
965 if (*cp++ == '\n')
966 cnt--;
968 file_o.size = cp - file_o.ptr;
970 patch = compare_buffer(file_p, &file_o, 1);
973 * file_o is a part of final image we are annotating.
974 * file_p partially may match that image.
976 memset(split, 0, sizeof(struct blame_entry [3]));
977 plno = tlno = 0;
978 for (i = 0; i < patch->num; i++) {
979 struct chunk *chunk = &patch->chunks[i];
981 handle_split(sb, ent, tlno, plno, chunk->same, parent, split);
982 plno = chunk->p_next;
983 tlno = chunk->t_next;
985 /* remainder, if any, all match the preimage */
986 handle_split(sb, ent, tlno, plno, ent->num_lines, parent, split);
987 free_patch(patch);
991 * See if lines currently target is suspected for can be attributed to
992 * parent.
994 static int find_move_in_parent(struct scoreboard *sb,
995 struct origin *target,
996 struct origin *parent)
998 int last_in_target, made_progress;
999 struct blame_entry *e, split[3];
1000 mmfile_t file_p;
1002 last_in_target = find_last_in_target(sb, target);
1003 if (last_in_target < 0)
1004 return 1; /* nothing remains for this target */
1006 fill_origin_blob(parent, &file_p);
1007 if (!file_p.ptr)
1008 return 0;
1010 made_progress = 1;
1011 while (made_progress) {
1012 made_progress = 0;
1013 for (e = sb->ent; e; e = e->next) {
1014 if (e->guilty || !same_suspect(e->suspect, target) ||
1015 ent_score(sb, e) < blame_move_score)
1016 continue;
1017 find_copy_in_blob(sb, e, parent, split, &file_p);
1018 if (split[1].suspect &&
1019 blame_move_score < ent_score(sb, &split[1])) {
1020 split_blame(sb, split, e);
1021 made_progress = 1;
1023 decref_split(split);
1026 return 0;
1029 struct blame_list {
1030 struct blame_entry *ent;
1031 struct blame_entry split[3];
1035 * Count the number of entries the target is suspected for,
1036 * and prepare a list of entry and the best split.
1038 static struct blame_list *setup_blame_list(struct scoreboard *sb,
1039 struct origin *target,
1040 int min_score,
1041 int *num_ents_p)
1043 struct blame_entry *e;
1044 int num_ents, i;
1045 struct blame_list *blame_list = NULL;
1047 for (e = sb->ent, num_ents = 0; e; e = e->next)
1048 if (!e->scanned && !e->guilty &&
1049 same_suspect(e->suspect, target) &&
1050 min_score < ent_score(sb, e))
1051 num_ents++;
1052 if (num_ents) {
1053 blame_list = xcalloc(num_ents, sizeof(struct blame_list));
1054 for (e = sb->ent, i = 0; e; e = e->next)
1055 if (!e->scanned && !e->guilty &&
1056 same_suspect(e->suspect, target) &&
1057 min_score < ent_score(sb, e))
1058 blame_list[i++].ent = e;
1060 *num_ents_p = num_ents;
1061 return blame_list;
1065 * Reset the scanned status on all entries.
1067 static void reset_scanned_flag(struct scoreboard *sb)
1069 struct blame_entry *e;
1070 for (e = sb->ent; e; e = e->next)
1071 e->scanned = 0;
1075 * For lines target is suspected for, see if we can find code movement
1076 * across file boundary from the parent commit. porigin is the path
1077 * in the parent we already tried.
1079 static int find_copy_in_parent(struct scoreboard *sb,
1080 struct origin *target,
1081 struct commit *parent,
1082 struct origin *porigin,
1083 int opt)
1085 struct diff_options diff_opts;
1086 const char *paths[1];
1087 int i, j;
1088 int retval;
1089 struct blame_list *blame_list;
1090 int num_ents;
1092 blame_list = setup_blame_list(sb, target, blame_copy_score, &num_ents);
1093 if (!blame_list)
1094 return 1; /* nothing remains for this target */
1096 diff_setup(&diff_opts);
1097 DIFF_OPT_SET(&diff_opts, RECURSIVE);
1098 diff_opts.output_format = DIFF_FORMAT_NO_OUTPUT;
1100 paths[0] = NULL;
1101 diff_tree_setup_paths(paths, &diff_opts);
1102 if (diff_setup_done(&diff_opts) < 0)
1103 die("diff-setup");
1105 /* Try "find copies harder" on new path if requested;
1106 * we do not want to use diffcore_rename() actually to
1107 * match things up; find_copies_harder is set only to
1108 * force diff_tree_sha1() to feed all filepairs to diff_queue,
1109 * and this code needs to be after diff_setup_done(), which
1110 * usually makes find-copies-harder imply copy detection.
1112 if ((opt & PICKAXE_BLAME_COPY_HARDEST)
1113 || ((opt & PICKAXE_BLAME_COPY_HARDER)
1114 && (!porigin || strcmp(target->path, porigin->path))))
1115 DIFF_OPT_SET(&diff_opts, FIND_COPIES_HARDER);
1117 if (is_null_sha1(target->commit->object.sha1))
1118 do_diff_cache(parent->tree->object.sha1, &diff_opts);
1119 else
1120 diff_tree_sha1(parent->tree->object.sha1,
1121 target->commit->tree->object.sha1,
1122 "", &diff_opts);
1124 if (!DIFF_OPT_TST(&diff_opts, FIND_COPIES_HARDER))
1125 diffcore_std(&diff_opts);
1127 retval = 0;
1128 while (1) {
1129 int made_progress = 0;
1131 for (i = 0; i < diff_queued_diff.nr; i++) {
1132 struct diff_filepair *p = diff_queued_diff.queue[i];
1133 struct origin *norigin;
1134 mmfile_t file_p;
1135 struct blame_entry this[3];
1137 if (!DIFF_FILE_VALID(p->one))
1138 continue; /* does not exist in parent */
1139 if (S_ISGITLINK(p->one->mode))
1140 continue; /* ignore git links */
1141 if (porigin && !strcmp(p->one->path, porigin->path))
1142 /* find_move already dealt with this path */
1143 continue;
1145 norigin = get_origin(sb, parent, p->one->path);
1146 hashcpy(norigin->blob_sha1, p->one->sha1);
1147 fill_origin_blob(norigin, &file_p);
1148 if (!file_p.ptr)
1149 continue;
1151 for (j = 0; j < num_ents; j++) {
1152 find_copy_in_blob(sb, blame_list[j].ent,
1153 norigin, this, &file_p);
1154 copy_split_if_better(sb, blame_list[j].split,
1155 this);
1156 decref_split(this);
1158 origin_decref(norigin);
1161 for (j = 0; j < num_ents; j++) {
1162 struct blame_entry *split = blame_list[j].split;
1163 if (split[1].suspect &&
1164 blame_copy_score < ent_score(sb, &split[1])) {
1165 split_blame(sb, split, blame_list[j].ent);
1166 made_progress = 1;
1168 else
1169 blame_list[j].ent->scanned = 1;
1170 decref_split(split);
1172 free(blame_list);
1174 if (!made_progress)
1175 break;
1176 blame_list = setup_blame_list(sb, target, blame_copy_score, &num_ents);
1177 if (!blame_list) {
1178 retval = 1;
1179 break;
1182 reset_scanned_flag(sb);
1183 diff_flush(&diff_opts);
1184 diff_tree_release_paths(&diff_opts);
1185 return retval;
1189 * The blobs of origin and porigin exactly match, so everything
1190 * origin is suspected for can be blamed on the parent.
1192 static void pass_whole_blame(struct scoreboard *sb,
1193 struct origin *origin, struct origin *porigin)
1195 struct blame_entry *e;
1197 if (!porigin->file.ptr && origin->file.ptr) {
1198 /* Steal its file */
1199 porigin->file = origin->file;
1200 origin->file.ptr = NULL;
1202 for (e = sb->ent; e; e = e->next) {
1203 if (!same_suspect(e->suspect, origin))
1204 continue;
1205 origin_incref(porigin);
1206 origin_decref(e->suspect);
1207 e->suspect = porigin;
1212 * We pass blame from the current commit to its parents. We keep saying
1213 * "parent" (and "porigin"), but what we mean is to find scapegoat to
1214 * exonerate ourselves.
1216 static struct commit_list *first_scapegoat(struct rev_info *revs, struct commit *commit)
1218 if (!reverse)
1219 return commit->parents;
1220 return lookup_decoration(&revs->children, &commit->object);
1223 static int num_scapegoats(struct rev_info *revs, struct commit *commit)
1225 int cnt;
1226 struct commit_list *l = first_scapegoat(revs, commit);
1227 for (cnt = 0; l; l = l->next)
1228 cnt++;
1229 return cnt;
1232 #define MAXSG 16
1234 static void pass_blame(struct scoreboard *sb, struct origin *origin, int opt)
1236 struct rev_info *revs = sb->revs;
1237 int i, pass, num_sg;
1238 struct commit *commit = origin->commit;
1239 struct commit_list *sg;
1240 struct origin *sg_buf[MAXSG];
1241 struct origin *porigin, **sg_origin = sg_buf;
1243 num_sg = num_scapegoats(revs, commit);
1244 if (!num_sg)
1245 goto finish;
1246 else if (num_sg < ARRAY_SIZE(sg_buf))
1247 memset(sg_buf, 0, sizeof(sg_buf));
1248 else
1249 sg_origin = xcalloc(num_sg, sizeof(*sg_origin));
1252 * The first pass looks for unrenamed path to optimize for
1253 * common cases, then we look for renames in the second pass.
1255 for (pass = 0; pass < 2; pass++) {
1256 struct origin *(*find)(struct scoreboard *,
1257 struct commit *, struct origin *);
1258 find = pass ? find_rename : find_origin;
1260 for (i = 0, sg = first_scapegoat(revs, commit);
1261 i < num_sg && sg;
1262 sg = sg->next, i++) {
1263 struct commit *p = sg->item;
1264 int j, same;
1266 if (sg_origin[i])
1267 continue;
1268 if (parse_commit(p))
1269 continue;
1270 porigin = find(sb, p, origin);
1271 if (!porigin)
1272 continue;
1273 if (!hashcmp(porigin->blob_sha1, origin->blob_sha1)) {
1274 pass_whole_blame(sb, origin, porigin);
1275 origin_decref(porigin);
1276 goto finish;
1278 for (j = same = 0; j < i; j++)
1279 if (sg_origin[j] &&
1280 !hashcmp(sg_origin[j]->blob_sha1,
1281 porigin->blob_sha1)) {
1282 same = 1;
1283 break;
1285 if (!same)
1286 sg_origin[i] = porigin;
1287 else
1288 origin_decref(porigin);
1292 num_commits++;
1293 for (i = 0, sg = first_scapegoat(revs, commit);
1294 i < num_sg && sg;
1295 sg = sg->next, i++) {
1296 struct origin *porigin = sg_origin[i];
1297 if (!porigin)
1298 continue;
1299 if (pass_blame_to_parent(sb, origin, porigin))
1300 goto finish;
1304 * Optionally find moves in parents' files.
1306 if (opt & PICKAXE_BLAME_MOVE)
1307 for (i = 0, sg = first_scapegoat(revs, commit);
1308 i < num_sg && sg;
1309 sg = sg->next, i++) {
1310 struct origin *porigin = sg_origin[i];
1311 if (!porigin)
1312 continue;
1313 if (find_move_in_parent(sb, origin, porigin))
1314 goto finish;
1318 * Optionally find copies from parents' files.
1320 if (opt & PICKAXE_BLAME_COPY)
1321 for (i = 0, sg = first_scapegoat(revs, commit);
1322 i < num_sg && sg;
1323 sg = sg->next, i++) {
1324 struct origin *porigin = sg_origin[i];
1325 if (find_copy_in_parent(sb, origin, sg->item,
1326 porigin, opt))
1327 goto finish;
1330 finish:
1331 for (i = 0; i < num_sg; i++) {
1332 if (sg_origin[i]) {
1333 drop_origin_blob(sg_origin[i]);
1334 origin_decref(sg_origin[i]);
1337 drop_origin_blob(origin);
1338 if (sg_buf != sg_origin)
1339 free(sg_origin);
1343 * Information on commits, used for output.
1345 struct commit_info
1347 const char *author;
1348 const char *author_mail;
1349 unsigned long author_time;
1350 const char *author_tz;
1352 /* filled only when asked for details */
1353 const char *committer;
1354 const char *committer_mail;
1355 unsigned long committer_time;
1356 const char *committer_tz;
1358 const char *summary;
1362 * Parse author/committer line in the commit object buffer
1364 static void get_ac_line(const char *inbuf, const char *what,
1365 int bufsz, char *person, const char **mail,
1366 unsigned long *time, const char **tz)
1368 int len, tzlen, maillen;
1369 char *tmp, *endp, *timepos;
1371 tmp = strstr(inbuf, what);
1372 if (!tmp)
1373 goto error_out;
1374 tmp += strlen(what);
1375 endp = strchr(tmp, '\n');
1376 if (!endp)
1377 len = strlen(tmp);
1378 else
1379 len = endp - tmp;
1380 if (bufsz <= len) {
1381 error_out:
1382 /* Ugh */
1383 *mail = *tz = "(unknown)";
1384 *time = 0;
1385 return;
1387 memcpy(person, tmp, len);
1389 tmp = person;
1390 tmp += len;
1391 *tmp = 0;
1392 while (*tmp != ' ')
1393 tmp--;
1394 *tz = tmp+1;
1395 tzlen = (person+len)-(tmp+1);
1397 *tmp = 0;
1398 while (*tmp != ' ')
1399 tmp--;
1400 *time = strtoul(tmp, NULL, 10);
1401 timepos = tmp;
1403 *tmp = 0;
1404 while (*tmp != ' ')
1405 tmp--;
1406 *mail = tmp + 1;
1407 *tmp = 0;
1408 maillen = timepos - tmp;
1410 if (!mailmap.nr)
1411 return;
1414 * mailmap expansion may make the name longer.
1415 * make room by pushing stuff down.
1417 tmp = person + bufsz - (tzlen + 1);
1418 memmove(tmp, *tz, tzlen);
1419 tmp[tzlen] = 0;
1420 *tz = tmp;
1422 tmp = tmp - (maillen + 1);
1423 memmove(tmp, *mail, maillen);
1424 tmp[maillen] = 0;
1425 *mail = tmp;
1428 * Now, convert e-mail using mailmap
1430 map_email(&mailmap, tmp + 1, person, tmp-person-1);
1433 static void get_commit_info(struct commit *commit,
1434 struct commit_info *ret,
1435 int detailed)
1437 int len;
1438 char *tmp, *endp;
1439 static char author_buf[1024];
1440 static char committer_buf[1024];
1441 static char summary_buf[1024];
1444 * We've operated without save_commit_buffer, so
1445 * we now need to populate them for output.
1447 if (!commit->buffer) {
1448 enum object_type type;
1449 unsigned long size;
1450 commit->buffer =
1451 read_sha1_file(commit->object.sha1, &type, &size);
1452 if (!commit->buffer)
1453 die("Cannot read commit %s",
1454 sha1_to_hex(commit->object.sha1));
1456 ret->author = author_buf;
1457 get_ac_line(commit->buffer, "\nauthor ",
1458 sizeof(author_buf), author_buf, &ret->author_mail,
1459 &ret->author_time, &ret->author_tz);
1461 if (!detailed)
1462 return;
1464 ret->committer = committer_buf;
1465 get_ac_line(commit->buffer, "\ncommitter ",
1466 sizeof(committer_buf), committer_buf, &ret->committer_mail,
1467 &ret->committer_time, &ret->committer_tz);
1469 ret->summary = summary_buf;
1470 tmp = strstr(commit->buffer, "\n\n");
1471 if (!tmp) {
1472 error_out:
1473 sprintf(summary_buf, "(%s)", sha1_to_hex(commit->object.sha1));
1474 return;
1476 tmp += 2;
1477 endp = strchr(tmp, '\n');
1478 if (!endp)
1479 endp = tmp + strlen(tmp);
1480 len = endp - tmp;
1481 if (len >= sizeof(summary_buf) || len == 0)
1482 goto error_out;
1483 memcpy(summary_buf, tmp, len);
1484 summary_buf[len] = 0;
1488 * To allow LF and other nonportable characters in pathnames,
1489 * they are c-style quoted as needed.
1491 static void write_filename_info(const char *path)
1493 printf("filename ");
1494 write_name_quoted(path, stdout, '\n');
1498 * The blame_entry is found to be guilty for the range. Mark it
1499 * as such, and show it in incremental output.
1501 static void found_guilty_entry(struct blame_entry *ent)
1503 if (ent->guilty)
1504 return;
1505 ent->guilty = 1;
1506 if (incremental) {
1507 struct origin *suspect = ent->suspect;
1509 printf("%s %d %d %d\n",
1510 sha1_to_hex(suspect->commit->object.sha1),
1511 ent->s_lno + 1, ent->lno + 1, ent->num_lines);
1512 if (!(suspect->commit->object.flags & METAINFO_SHOWN)) {
1513 struct commit_info ci;
1514 suspect->commit->object.flags |= METAINFO_SHOWN;
1515 get_commit_info(suspect->commit, &ci, 1);
1516 printf("author %s\n", ci.author);
1517 printf("author-mail %s\n", ci.author_mail);
1518 printf("author-time %lu\n", ci.author_time);
1519 printf("author-tz %s\n", ci.author_tz);
1520 printf("committer %s\n", ci.committer);
1521 printf("committer-mail %s\n", ci.committer_mail);
1522 printf("committer-time %lu\n", ci.committer_time);
1523 printf("committer-tz %s\n", ci.committer_tz);
1524 printf("summary %s\n", ci.summary);
1525 if (suspect->commit->object.flags & UNINTERESTING)
1526 printf("boundary\n");
1528 write_filename_info(suspect->path);
1529 maybe_flush_or_die(stdout, "stdout");
1534 * The main loop -- while the scoreboard has lines whose true origin
1535 * is still unknown, pick one blame_entry, and allow its current
1536 * suspect to pass blames to its parents.
1538 static void assign_blame(struct scoreboard *sb, int opt)
1540 struct rev_info *revs = sb->revs;
1542 while (1) {
1543 struct blame_entry *ent;
1544 struct commit *commit;
1545 struct origin *suspect = NULL;
1547 /* find one suspect to break down */
1548 for (ent = sb->ent; !suspect && ent; ent = ent->next)
1549 if (!ent->guilty)
1550 suspect = ent->suspect;
1551 if (!suspect)
1552 return; /* all done */
1555 * We will use this suspect later in the loop,
1556 * so hold onto it in the meantime.
1558 origin_incref(suspect);
1559 commit = suspect->commit;
1560 if (!commit->object.parsed)
1561 parse_commit(commit);
1562 if (reverse ||
1563 (!(commit->object.flags & UNINTERESTING) &&
1564 !(revs->max_age != -1 && commit->date < revs->max_age)))
1565 pass_blame(sb, suspect, opt);
1566 else {
1567 commit->object.flags |= UNINTERESTING;
1568 if (commit->object.parsed)
1569 mark_parents_uninteresting(commit);
1571 /* treat root commit as boundary */
1572 if (!commit->parents && !show_root)
1573 commit->object.flags |= UNINTERESTING;
1575 /* Take responsibility for the remaining entries */
1576 for (ent = sb->ent; ent; ent = ent->next)
1577 if (same_suspect(ent->suspect, suspect))
1578 found_guilty_entry(ent);
1579 origin_decref(suspect);
1581 if (DEBUG) /* sanity */
1582 sanity_check_refcnt(sb);
1586 static const char *format_time(unsigned long time, const char *tz_str,
1587 int show_raw_time)
1589 static char time_buf[128];
1590 time_t t = time;
1591 int minutes, tz;
1592 struct tm *tm;
1594 if (show_raw_time) {
1595 sprintf(time_buf, "%lu %s", time, tz_str);
1596 return time_buf;
1599 tz = atoi(tz_str);
1600 minutes = tz < 0 ? -tz : tz;
1601 minutes = (minutes / 100)*60 + (minutes % 100);
1602 minutes = tz < 0 ? -minutes : minutes;
1603 t = time + minutes * 60;
1604 tm = gmtime(&t);
1606 strftime(time_buf, sizeof(time_buf), "%Y-%m-%d %H:%M:%S ", tm);
1607 strcat(time_buf, tz_str);
1608 return time_buf;
1611 #define OUTPUT_ANNOTATE_COMPAT 001
1612 #define OUTPUT_LONG_OBJECT_NAME 002
1613 #define OUTPUT_RAW_TIMESTAMP 004
1614 #define OUTPUT_PORCELAIN 010
1615 #define OUTPUT_SHOW_NAME 020
1616 #define OUTPUT_SHOW_NUMBER 040
1617 #define OUTPUT_SHOW_SCORE 0100
1618 #define OUTPUT_NO_AUTHOR 0200
1620 static void emit_porcelain(struct scoreboard *sb, struct blame_entry *ent)
1622 int cnt;
1623 const char *cp;
1624 struct origin *suspect = ent->suspect;
1625 char hex[41];
1627 strcpy(hex, sha1_to_hex(suspect->commit->object.sha1));
1628 printf("%s%c%d %d %d\n",
1629 hex,
1630 ent->guilty ? ' ' : '*', // purely for debugging
1631 ent->s_lno + 1,
1632 ent->lno + 1,
1633 ent->num_lines);
1634 if (!(suspect->commit->object.flags & METAINFO_SHOWN)) {
1635 struct commit_info ci;
1636 suspect->commit->object.flags |= METAINFO_SHOWN;
1637 get_commit_info(suspect->commit, &ci, 1);
1638 printf("author %s\n", ci.author);
1639 printf("author-mail %s\n", ci.author_mail);
1640 printf("author-time %lu\n", ci.author_time);
1641 printf("author-tz %s\n", ci.author_tz);
1642 printf("committer %s\n", ci.committer);
1643 printf("committer-mail %s\n", ci.committer_mail);
1644 printf("committer-time %lu\n", ci.committer_time);
1645 printf("committer-tz %s\n", ci.committer_tz);
1646 write_filename_info(suspect->path);
1647 printf("summary %s\n", ci.summary);
1648 if (suspect->commit->object.flags & UNINTERESTING)
1649 printf("boundary\n");
1651 else if (suspect->commit->object.flags & MORE_THAN_ONE_PATH)
1652 write_filename_info(suspect->path);
1654 cp = nth_line(sb, ent->lno);
1655 for (cnt = 0; cnt < ent->num_lines; cnt++) {
1656 char ch;
1657 if (cnt)
1658 printf("%s %d %d\n", hex,
1659 ent->s_lno + 1 + cnt,
1660 ent->lno + 1 + cnt);
1661 putchar('\t');
1662 do {
1663 ch = *cp++;
1664 putchar(ch);
1665 } while (ch != '\n' &&
1666 cp < sb->final_buf + sb->final_buf_size);
1670 static void emit_other(struct scoreboard *sb, struct blame_entry *ent, int opt)
1672 int cnt;
1673 const char *cp;
1674 struct origin *suspect = ent->suspect;
1675 struct commit_info ci;
1676 char hex[41];
1677 int show_raw_time = !!(opt & OUTPUT_RAW_TIMESTAMP);
1679 get_commit_info(suspect->commit, &ci, 1);
1680 strcpy(hex, sha1_to_hex(suspect->commit->object.sha1));
1682 cp = nth_line(sb, ent->lno);
1683 for (cnt = 0; cnt < ent->num_lines; cnt++) {
1684 char ch;
1685 int length = (opt & OUTPUT_LONG_OBJECT_NAME) ? 40 : 8;
1687 if (suspect->commit->object.flags & UNINTERESTING) {
1688 if (blank_boundary)
1689 memset(hex, ' ', length);
1690 else if (!(opt & OUTPUT_ANNOTATE_COMPAT)) {
1691 length--;
1692 putchar('^');
1696 printf("%.*s", length, hex);
1697 if (opt & OUTPUT_ANNOTATE_COMPAT)
1698 printf("\t(%10s\t%10s\t%d)", ci.author,
1699 format_time(ci.author_time, ci.author_tz,
1700 show_raw_time),
1701 ent->lno + 1 + cnt);
1702 else {
1703 if (opt & OUTPUT_SHOW_SCORE)
1704 printf(" %*d %02d",
1705 max_score_digits, ent->score,
1706 ent->suspect->refcnt);
1707 if (opt & OUTPUT_SHOW_NAME)
1708 printf(" %-*.*s", longest_file, longest_file,
1709 suspect->path);
1710 if (opt & OUTPUT_SHOW_NUMBER)
1711 printf(" %*d", max_orig_digits,
1712 ent->s_lno + 1 + cnt);
1714 if (!(opt & OUTPUT_NO_AUTHOR))
1715 printf(" (%-*.*s %10s",
1716 longest_author, longest_author,
1717 ci.author,
1718 format_time(ci.author_time,
1719 ci.author_tz,
1720 show_raw_time));
1721 printf(" %*d) ",
1722 max_digits, ent->lno + 1 + cnt);
1724 do {
1725 ch = *cp++;
1726 putchar(ch);
1727 } while (ch != '\n' &&
1728 cp < sb->final_buf + sb->final_buf_size);
1732 static void output(struct scoreboard *sb, int option)
1734 struct blame_entry *ent;
1736 if (option & OUTPUT_PORCELAIN) {
1737 for (ent = sb->ent; ent; ent = ent->next) {
1738 struct blame_entry *oth;
1739 struct origin *suspect = ent->suspect;
1740 struct commit *commit = suspect->commit;
1741 if (commit->object.flags & MORE_THAN_ONE_PATH)
1742 continue;
1743 for (oth = ent->next; oth; oth = oth->next) {
1744 if ((oth->suspect->commit != commit) ||
1745 !strcmp(oth->suspect->path, suspect->path))
1746 continue;
1747 commit->object.flags |= MORE_THAN_ONE_PATH;
1748 break;
1753 for (ent = sb->ent; ent; ent = ent->next) {
1754 if (option & OUTPUT_PORCELAIN)
1755 emit_porcelain(sb, ent);
1756 else {
1757 emit_other(sb, ent, option);
1763 * To allow quick access to the contents of nth line in the
1764 * final image, prepare an index in the scoreboard.
1766 static int prepare_lines(struct scoreboard *sb)
1768 const char *buf = sb->final_buf;
1769 unsigned long len = sb->final_buf_size;
1770 int num = 0, incomplete = 0, bol = 1;
1772 if (len && buf[len-1] != '\n')
1773 incomplete++; /* incomplete line at the end */
1774 while (len--) {
1775 if (bol) {
1776 sb->lineno = xrealloc(sb->lineno,
1777 sizeof(int* ) * (num + 1));
1778 sb->lineno[num] = buf - sb->final_buf;
1779 bol = 0;
1781 if (*buf++ == '\n') {
1782 num++;
1783 bol = 1;
1786 sb->lineno = xrealloc(sb->lineno,
1787 sizeof(int* ) * (num + incomplete + 1));
1788 sb->lineno[num + incomplete] = buf - sb->final_buf;
1789 sb->num_lines = num + incomplete;
1790 return sb->num_lines;
1794 * Add phony grafts for use with -S; this is primarily to
1795 * support git's cvsserver that wants to give a linear history
1796 * to its clients.
1798 static int read_ancestry(const char *graft_file)
1800 FILE *fp = fopen(graft_file, "r");
1801 char buf[1024];
1802 if (!fp)
1803 return -1;
1804 while (fgets(buf, sizeof(buf), fp)) {
1805 /* The format is just "Commit Parent1 Parent2 ...\n" */
1806 int len = strlen(buf);
1807 struct commit_graft *graft = read_graft_line(buf, len);
1808 if (graft)
1809 register_commit_graft(graft, 0);
1811 fclose(fp);
1812 return 0;
1816 * How many columns do we need to show line numbers in decimal?
1818 static int lineno_width(int lines)
1820 int i, width;
1822 for (width = 1, i = 10; i <= lines + 1; width++)
1823 i *= 10;
1824 return width;
1828 * How many columns do we need to show line numbers, authors,
1829 * and filenames?
1831 static void find_alignment(struct scoreboard *sb, int *option)
1833 int longest_src_lines = 0;
1834 int longest_dst_lines = 0;
1835 unsigned largest_score = 0;
1836 struct blame_entry *e;
1838 for (e = sb->ent; e; e = e->next) {
1839 struct origin *suspect = e->suspect;
1840 struct commit_info ci;
1841 int num;
1843 if (strcmp(suspect->path, sb->path))
1844 *option |= OUTPUT_SHOW_NAME;
1845 num = strlen(suspect->path);
1846 if (longest_file < num)
1847 longest_file = num;
1848 if (!(suspect->commit->object.flags & METAINFO_SHOWN)) {
1849 suspect->commit->object.flags |= METAINFO_SHOWN;
1850 get_commit_info(suspect->commit, &ci, 1);
1851 num = strlen(ci.author);
1852 if (longest_author < num)
1853 longest_author = num;
1855 num = e->s_lno + e->num_lines;
1856 if (longest_src_lines < num)
1857 longest_src_lines = num;
1858 num = e->lno + e->num_lines;
1859 if (longest_dst_lines < num)
1860 longest_dst_lines = num;
1861 if (largest_score < ent_score(sb, e))
1862 largest_score = ent_score(sb, e);
1864 max_orig_digits = lineno_width(longest_src_lines);
1865 max_digits = lineno_width(longest_dst_lines);
1866 max_score_digits = lineno_width(largest_score);
1870 * For debugging -- origin is refcounted, and this asserts that
1871 * we do not underflow.
1873 static void sanity_check_refcnt(struct scoreboard *sb)
1875 int baa = 0;
1876 struct blame_entry *ent;
1878 for (ent = sb->ent; ent; ent = ent->next) {
1879 /* Nobody should have zero or negative refcnt */
1880 if (ent->suspect->refcnt <= 0) {
1881 fprintf(stderr, "%s in %s has negative refcnt %d\n",
1882 ent->suspect->path,
1883 sha1_to_hex(ent->suspect->commit->object.sha1),
1884 ent->suspect->refcnt);
1885 baa = 1;
1888 for (ent = sb->ent; ent; ent = ent->next) {
1889 /* Mark the ones that haven't been checked */
1890 if (0 < ent->suspect->refcnt)
1891 ent->suspect->refcnt = -ent->suspect->refcnt;
1893 for (ent = sb->ent; ent; ent = ent->next) {
1895 * ... then pick each and see if they have the the
1896 * correct refcnt.
1898 int found;
1899 struct blame_entry *e;
1900 struct origin *suspect = ent->suspect;
1902 if (0 < suspect->refcnt)
1903 continue;
1904 suspect->refcnt = -suspect->refcnt; /* Unmark */
1905 for (found = 0, e = sb->ent; e; e = e->next) {
1906 if (e->suspect != suspect)
1907 continue;
1908 found++;
1910 if (suspect->refcnt != found) {
1911 fprintf(stderr, "%s in %s has refcnt %d, not %d\n",
1912 ent->suspect->path,
1913 sha1_to_hex(ent->suspect->commit->object.sha1),
1914 ent->suspect->refcnt, found);
1915 baa = 2;
1918 if (baa) {
1919 int opt = 0160;
1920 find_alignment(sb, &opt);
1921 output(sb, opt);
1922 die("Baa %d!", baa);
1927 * Used for the command line parsing; check if the path exists
1928 * in the working tree.
1930 static int has_string_in_work_tree(const char *path)
1932 struct stat st;
1933 return !lstat(path, &st);
1936 static unsigned parse_score(const char *arg)
1938 char *end;
1939 unsigned long score = strtoul(arg, &end, 10);
1940 if (*end)
1941 return 0;
1942 return score;
1945 static const char *add_prefix(const char *prefix, const char *path)
1947 return prefix_path(prefix, prefix ? strlen(prefix) : 0, path);
1951 * Parsing of (comma separated) one item in the -L option
1953 static const char *parse_loc(const char *spec,
1954 struct scoreboard *sb, long lno,
1955 long begin, long *ret)
1957 char *term;
1958 const char *line;
1959 long num;
1960 int reg_error;
1961 regex_t regexp;
1962 regmatch_t match[1];
1964 /* Allow "-L <something>,+20" to mean starting at <something>
1965 * for 20 lines, or "-L <something>,-5" for 5 lines ending at
1966 * <something>.
1968 if (1 < begin && (spec[0] == '+' || spec[0] == '-')) {
1969 num = strtol(spec + 1, &term, 10);
1970 if (term != spec + 1) {
1971 if (spec[0] == '-')
1972 num = 0 - num;
1973 if (0 < num)
1974 *ret = begin + num - 2;
1975 else if (!num)
1976 *ret = begin;
1977 else
1978 *ret = begin + num;
1979 return term;
1981 return spec;
1983 num = strtol(spec, &term, 10);
1984 if (term != spec) {
1985 *ret = num;
1986 return term;
1988 if (spec[0] != '/')
1989 return spec;
1991 /* it could be a regexp of form /.../ */
1992 for (term = (char*) spec + 1; *term && *term != '/'; term++) {
1993 if (*term == '\\')
1994 term++;
1996 if (*term != '/')
1997 return spec;
1999 /* try [spec+1 .. term-1] as regexp */
2000 *term = 0;
2001 begin--; /* input is in human terms */
2002 line = nth_line(sb, begin);
2004 if (!(reg_error = regcomp(&regexp, spec + 1, REG_NEWLINE)) &&
2005 !(reg_error = regexec(&regexp, line, 1, match, 0))) {
2006 const char *cp = line + match[0].rm_so;
2007 const char *nline;
2009 while (begin++ < lno) {
2010 nline = nth_line(sb, begin);
2011 if (line <= cp && cp < nline)
2012 break;
2013 line = nline;
2015 *ret = begin;
2016 regfree(&regexp);
2017 *term++ = '/';
2018 return term;
2020 else {
2021 char errbuf[1024];
2022 regerror(reg_error, &regexp, errbuf, 1024);
2023 die("-L parameter '%s': %s", spec + 1, errbuf);
2028 * Parsing of -L option
2030 static void prepare_blame_range(struct scoreboard *sb,
2031 const char *bottomtop,
2032 long lno,
2033 long *bottom, long *top)
2035 const char *term;
2037 term = parse_loc(bottomtop, sb, lno, 1, bottom);
2038 if (*term == ',') {
2039 term = parse_loc(term + 1, sb, lno, *bottom + 1, top);
2040 if (*term)
2041 usage(blame_usage);
2043 if (*term)
2044 usage(blame_usage);
2047 static int git_blame_config(const char *var, const char *value, void *cb)
2049 if (!strcmp(var, "blame.showroot")) {
2050 show_root = git_config_bool(var, value);
2051 return 0;
2053 if (!strcmp(var, "blame.blankboundary")) {
2054 blank_boundary = git_config_bool(var, value);
2055 return 0;
2057 return git_default_config(var, value, cb);
2061 * Prepare a dummy commit that represents the work tree (or staged) item.
2062 * Note that annotating work tree item never works in the reverse.
2064 static struct commit *fake_working_tree_commit(const char *path, const char *contents_from)
2066 struct commit *commit;
2067 struct origin *origin;
2068 unsigned char head_sha1[20];
2069 struct strbuf buf;
2070 const char *ident;
2071 time_t now;
2072 int size, len;
2073 struct cache_entry *ce;
2074 unsigned mode;
2076 if (get_sha1("HEAD", head_sha1))
2077 die("No such ref: HEAD");
2079 time(&now);
2080 commit = xcalloc(1, sizeof(*commit));
2081 commit->parents = xcalloc(1, sizeof(*commit->parents));
2082 commit->parents->item = lookup_commit_reference(head_sha1);
2083 commit->object.parsed = 1;
2084 commit->date = now;
2085 commit->object.type = OBJ_COMMIT;
2087 origin = make_origin(commit, path);
2089 strbuf_init(&buf, 0);
2090 if (!contents_from || strcmp("-", contents_from)) {
2091 struct stat st;
2092 const char *read_from;
2093 unsigned long fin_size;
2095 if (contents_from) {
2096 if (stat(contents_from, &st) < 0)
2097 die("Cannot stat %s", contents_from);
2098 read_from = contents_from;
2100 else {
2101 if (lstat(path, &st) < 0)
2102 die("Cannot lstat %s", path);
2103 read_from = path;
2105 fin_size = xsize_t(st.st_size);
2106 mode = canon_mode(st.st_mode);
2107 switch (st.st_mode & S_IFMT) {
2108 case S_IFREG:
2109 if (strbuf_read_file(&buf, read_from, st.st_size) != st.st_size)
2110 die("cannot open or read %s", read_from);
2111 break;
2112 case S_IFLNK:
2113 if (readlink(read_from, buf.buf, buf.alloc) != fin_size)
2114 die("cannot readlink %s", read_from);
2115 buf.len = fin_size;
2116 break;
2117 default:
2118 die("unsupported file type %s", read_from);
2121 else {
2122 /* Reading from stdin */
2123 contents_from = "standard input";
2124 mode = 0;
2125 if (strbuf_read(&buf, 0, 0) < 0)
2126 die("read error %s from stdin", strerror(errno));
2128 convert_to_git(path, buf.buf, buf.len, &buf, 0);
2129 origin->file.ptr = buf.buf;
2130 origin->file.size = buf.len;
2131 pretend_sha1_file(buf.buf, buf.len, OBJ_BLOB, origin->blob_sha1);
2132 commit->util = origin;
2135 * Read the current index, replace the path entry with
2136 * origin->blob_sha1 without mucking with its mode or type
2137 * bits; we are not going to write this index out -- we just
2138 * want to run "diff-index --cached".
2140 discard_cache();
2141 read_cache();
2143 len = strlen(path);
2144 if (!mode) {
2145 int pos = cache_name_pos(path, len);
2146 if (0 <= pos)
2147 mode = active_cache[pos]->ce_mode;
2148 else
2149 /* Let's not bother reading from HEAD tree */
2150 mode = S_IFREG | 0644;
2152 size = cache_entry_size(len);
2153 ce = xcalloc(1, size);
2154 hashcpy(ce->sha1, origin->blob_sha1);
2155 memcpy(ce->name, path, len);
2156 ce->ce_flags = create_ce_flags(len, 0);
2157 ce->ce_mode = create_ce_mode(mode);
2158 add_cache_entry(ce, ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE);
2161 * We are not going to write this out, so this does not matter
2162 * right now, but someday we might optimize diff-index --cached
2163 * with cache-tree information.
2165 cache_tree_invalidate_path(active_cache_tree, path);
2167 commit->buffer = xmalloc(400);
2168 ident = fmt_ident("Not Committed Yet", "not.committed.yet", NULL, 0);
2169 snprintf(commit->buffer, 400,
2170 "tree 0000000000000000000000000000000000000000\n"
2171 "parent %s\n"
2172 "author %s\n"
2173 "committer %s\n\n"
2174 "Version of %s from %s\n",
2175 sha1_to_hex(head_sha1),
2176 ident, ident, path, contents_from ? contents_from : path);
2177 return commit;
2180 static const char *prepare_final(struct scoreboard *sb)
2182 int i;
2183 const char *final_commit_name = NULL;
2184 struct rev_info *revs = sb->revs;
2187 * There must be one and only one positive commit in the
2188 * revs->pending array.
2190 for (i = 0; i < revs->pending.nr; i++) {
2191 struct object *obj = revs->pending.objects[i].item;
2192 if (obj->flags & UNINTERESTING)
2193 continue;
2194 while (obj->type == OBJ_TAG)
2195 obj = deref_tag(obj, NULL, 0);
2196 if (obj->type != OBJ_COMMIT)
2197 die("Non commit %s?", revs->pending.objects[i].name);
2198 if (sb->final)
2199 die("More than one commit to dig from %s and %s?",
2200 revs->pending.objects[i].name,
2201 final_commit_name);
2202 sb->final = (struct commit *) obj;
2203 final_commit_name = revs->pending.objects[i].name;
2205 return final_commit_name;
2208 static const char *prepare_initial(struct scoreboard *sb)
2210 int i;
2211 const char *final_commit_name = NULL;
2212 struct rev_info *revs = sb->revs;
2215 * There must be one and only one negative commit, and it must be
2216 * the boundary.
2218 for (i = 0; i < revs->pending.nr; i++) {
2219 struct object *obj = revs->pending.objects[i].item;
2220 if (!(obj->flags & UNINTERESTING))
2221 continue;
2222 while (obj->type == OBJ_TAG)
2223 obj = deref_tag(obj, NULL, 0);
2224 if (obj->type != OBJ_COMMIT)
2225 die("Non commit %s?", revs->pending.objects[i].name);
2226 if (sb->final)
2227 die("More than one commit to dig down to %s and %s?",
2228 revs->pending.objects[i].name,
2229 final_commit_name);
2230 sb->final = (struct commit *) obj;
2231 final_commit_name = revs->pending.objects[i].name;
2233 if (!final_commit_name)
2234 die("No commit to dig down to?");
2235 return final_commit_name;
2238 static int blame_copy_callback(const struct option *option, const char *arg, int unset)
2240 int *opt = option->value;
2243 * -C enables copy from removed files;
2244 * -C -C enables copy from existing files, but only
2245 * when blaming a new file;
2246 * -C -C -C enables copy from existing files for
2247 * everybody
2249 if (*opt & PICKAXE_BLAME_COPY_HARDER)
2250 *opt |= PICKAXE_BLAME_COPY_HARDEST;
2251 if (*opt & PICKAXE_BLAME_COPY)
2252 *opt |= PICKAXE_BLAME_COPY_HARDER;
2253 *opt |= PICKAXE_BLAME_COPY | PICKAXE_BLAME_MOVE;
2255 if (arg)
2256 blame_copy_score = parse_score(arg);
2257 return 0;
2260 static int blame_move_callback(const struct option *option, const char *arg, int unset)
2262 int *opt = option->value;
2264 *opt |= PICKAXE_BLAME_MOVE;
2266 if (arg)
2267 blame_move_score = parse_score(arg);
2268 return 0;
2271 static int blame_bottomtop_callback(const struct option *option, const char *arg, int unset)
2273 const char **bottomtop = option->value;
2274 if (!arg)
2275 return -1;
2276 if (*bottomtop)
2277 die("More than one '-L n,m' option given");
2278 *bottomtop = arg;
2279 return 0;
2282 int cmd_blame(int argc, const char **argv, const char *prefix)
2284 struct rev_info revs;
2285 const char *path;
2286 struct scoreboard sb;
2287 struct origin *o;
2288 struct blame_entry *ent;
2289 long dashdash_pos, bottom, top, lno;
2290 const char *final_commit_name = NULL;
2291 enum object_type type;
2293 static const char *bottomtop = NULL;
2294 static int output_option = 0, opt = 0;
2295 static int show_stats = 0;
2296 static const char *revs_file = NULL;
2297 static const char *contents_from = NULL;
2298 static const struct option options[] = {
2299 OPT_BOOLEAN(0, "incremental", &incremental, "Show blame entries as we find them, incrementally"),
2300 OPT_BOOLEAN('b', NULL, &blank_boundary, "Show blank SHA-1 for boundary commits (Default: off)"),
2301 OPT_BOOLEAN(0, "root", &show_root, "Do not treat root commits as boundaries (Default: off)"),
2302 OPT_BOOLEAN(0, "show-stats", &show_stats, "Show work cost statistics"),
2303 OPT_BIT(0, "score-debug", &output_option, "Show output score for blame entries", OUTPUT_SHOW_SCORE),
2304 OPT_BIT('f', "show-name", &output_option, "Show original filename (Default: auto)", OUTPUT_SHOW_NAME),
2305 OPT_BIT('n', "show-number", &output_option, "Show original linenumber (Default: off)", OUTPUT_SHOW_NUMBER),
2306 OPT_BIT('p', "porcelain", &output_option, "Show in a format designed for machine consumption", OUTPUT_PORCELAIN),
2307 OPT_BIT('c', NULL, &output_option, "Use the same output mode as git-annotate (Default: off)", OUTPUT_ANNOTATE_COMPAT),
2308 OPT_BIT('t', NULL, &output_option, "Show raw timestamp (Default: off)", OUTPUT_RAW_TIMESTAMP),
2309 OPT_BIT('l', NULL, &output_option, "Show long commit SHA1 (Default: off)", OUTPUT_LONG_OBJECT_NAME),
2310 OPT_BIT('s', NULL, &output_option, "Suppress author name and timestamp (Default: off)", OUTPUT_NO_AUTHOR),
2311 OPT_BIT('w', NULL, &xdl_opts, "Ignore whitespace differences", XDF_IGNORE_WHITESPACE),
2312 OPT_STRING('S', NULL, &revs_file, "file", "Use revisions from <file> instead of calling git-rev-list"),
2313 OPT_STRING(0, "contents", &contents_from, "file", "Use <file>'s contents as the final image"),
2314 { OPTION_CALLBACK, 'C', NULL, &opt, "score", "Find line copies within and across files", PARSE_OPT_OPTARG, blame_copy_callback },
2315 { OPTION_CALLBACK, 'M', NULL, &opt, "score", "Find line movements within and across files", PARSE_OPT_OPTARG, blame_move_callback },
2316 OPT_CALLBACK('L', NULL, &bottomtop, "n,m", "Process only line range n,m, counting from 1", blame_bottomtop_callback),
2317 OPT_END()
2320 struct parse_opt_ctx_t ctx;
2321 int cmd_is_annotate = !strcmp(argv[0], "annotate");
2323 git_config(git_blame_config, NULL);
2324 init_revisions(&revs, NULL);
2325 save_commit_buffer = 0;
2326 dashdash_pos = 0;
2328 parse_options_start(&ctx, argc, argv, PARSE_OPT_KEEP_DASHDASH |
2329 PARSE_OPT_KEEP_ARGV0);
2330 for (;;) {
2331 switch (parse_options_step(&ctx, options, blame_opt_usage)) {
2332 case PARSE_OPT_HELP:
2333 exit(129);
2334 case PARSE_OPT_DONE:
2335 if (ctx.argv[0])
2336 dashdash_pos = ctx.cpidx;
2337 goto parse_done;
2340 if (!strcmp(ctx.argv[0], "--reverse")) {
2341 ctx.argv[0] = "--children";
2342 reverse = 1;
2344 parse_revision_opt(&revs, &ctx, options, blame_opt_usage);
2346 parse_done:
2347 argc = parse_options_end(&ctx);
2349 if (cmd_is_annotate)
2350 output_option |= OUTPUT_ANNOTATE_COMPAT;
2352 if (DIFF_OPT_TST(&revs.diffopt, FIND_COPIES_HARDER))
2353 opt |= (PICKAXE_BLAME_COPY | PICKAXE_BLAME_MOVE |
2354 PICKAXE_BLAME_COPY_HARDER);
2356 if (!blame_move_score)
2357 blame_move_score = BLAME_DEFAULT_MOVE_SCORE;
2358 if (!blame_copy_score)
2359 blame_copy_score = BLAME_DEFAULT_COPY_SCORE;
2362 * We have collected options unknown to us in argv[1..unk]
2363 * which are to be passed to revision machinery if we are
2364 * going to do the "bottom" processing.
2366 * The remaining are:
2368 * (1) if dashdash_pos != 0, its either
2369 * "blame [revisions] -- <path>" or
2370 * "blame -- <path> <rev>"
2372 * (2) otherwise, its one of the two:
2373 * "blame [revisions] <path>"
2374 * "blame <path> <rev>"
2376 * Note that we must strip out <path> from the arguments: we do not
2377 * want the path pruning but we may want "bottom" processing.
2379 if (dashdash_pos) {
2380 switch (argc - dashdash_pos - 1) {
2381 case 2: /* (1b) */
2382 if (argc != 4)
2383 usage_with_options(blame_opt_usage, options);
2384 /* reorder for the new way: <rev> -- <path> */
2385 argv[1] = argv[3];
2386 argv[3] = argv[2];
2387 argv[2] = "--";
2388 /* FALLTHROUGH */
2389 case 1: /* (1a) */
2390 path = add_prefix(prefix, argv[--argc]);
2391 argv[argc] = NULL;
2392 break;
2393 default:
2394 usage_with_options(blame_opt_usage, options);
2396 } else {
2397 if (argc < 2)
2398 usage_with_options(blame_opt_usage, options);
2399 path = add_prefix(prefix, argv[argc - 1]);
2400 if (argc == 3 && !has_string_in_work_tree(path)) { /* (2b) */
2401 path = add_prefix(prefix, argv[1]);
2402 argv[1] = argv[2];
2404 argv[argc - 1] = "--";
2406 setup_work_tree();
2407 if (!has_string_in_work_tree(path))
2408 die("cannot stat path %s: %s", path, strerror(errno));
2411 setup_revisions(argc, argv, &revs, NULL);
2412 memset(&sb, 0, sizeof(sb));
2414 sb.revs = &revs;
2415 if (!reverse)
2416 final_commit_name = prepare_final(&sb);
2417 else if (contents_from)
2418 die("--contents and --children do not blend well.");
2419 else
2420 final_commit_name = prepare_initial(&sb);
2422 if (!sb.final) {
2424 * "--not A B -- path" without anything positive;
2425 * do not default to HEAD, but use the working tree
2426 * or "--contents".
2428 setup_work_tree();
2429 sb.final = fake_working_tree_commit(path, contents_from);
2430 add_pending_object(&revs, &(sb.final->object), ":");
2432 else if (contents_from)
2433 die("Cannot use --contents with final commit object name");
2436 * If we have bottom, this will mark the ancestors of the
2437 * bottom commits we would reach while traversing as
2438 * uninteresting.
2440 if (prepare_revision_walk(&revs))
2441 die("revision walk setup failed");
2443 if (is_null_sha1(sb.final->object.sha1)) {
2444 char *buf;
2445 o = sb.final->util;
2446 buf = xmalloc(o->file.size + 1);
2447 memcpy(buf, o->file.ptr, o->file.size + 1);
2448 sb.final_buf = buf;
2449 sb.final_buf_size = o->file.size;
2451 else {
2452 o = get_origin(&sb, sb.final, path);
2453 if (fill_blob_sha1(o))
2454 die("no such path %s in %s", path, final_commit_name);
2456 sb.final_buf = read_sha1_file(o->blob_sha1, &type,
2457 &sb.final_buf_size);
2458 if (!sb.final_buf)
2459 die("Cannot read blob %s for path %s",
2460 sha1_to_hex(o->blob_sha1),
2461 path);
2463 num_read_blob++;
2464 lno = prepare_lines(&sb);
2466 bottom = top = 0;
2467 if (bottomtop)
2468 prepare_blame_range(&sb, bottomtop, lno, &bottom, &top);
2469 if (bottom && top && top < bottom) {
2470 long tmp;
2471 tmp = top; top = bottom; bottom = tmp;
2473 if (bottom < 1)
2474 bottom = 1;
2475 if (top < 1)
2476 top = lno;
2477 bottom--;
2478 if (lno < top)
2479 die("file %s has only %lu lines", path, lno);
2481 ent = xcalloc(1, sizeof(*ent));
2482 ent->lno = bottom;
2483 ent->num_lines = top - bottom;
2484 ent->suspect = o;
2485 ent->s_lno = bottom;
2487 sb.ent = ent;
2488 sb.path = path;
2490 if (revs_file && read_ancestry(revs_file))
2491 die("reading graft file %s failed: %s",
2492 revs_file, strerror(errno));
2494 read_mailmap(&mailmap, ".mailmap", NULL);
2496 if (!incremental)
2497 setup_pager();
2499 assign_blame(&sb, opt);
2501 if (incremental)
2502 return 0;
2504 coalesce(&sb);
2506 if (!(output_option & OUTPUT_PORCELAIN))
2507 find_alignment(&sb, &output_option);
2509 output(&sb, output_option);
2510 free((void *)sb.final_buf);
2511 for (ent = sb.ent; ent; ) {
2512 struct blame_entry *e = ent->next;
2513 free(ent);
2514 ent = e;
2517 if (show_stats) {
2518 printf("num read blob: %d\n", num_read_blob);
2519 printf("num get patch: %d\n", num_get_patch);
2520 printf("num commits: %d\n", num_commits);
2522 return 0;