Merge branch 'ds/commit-graph-lockfile-fix'
[alt-git.git] / blame.c
blobba1b33c7f4e92c432d4cbdcb43b8d061e436f027
1 #include "cache.h"
2 #include "refs.h"
3 #include "cache-tree.h"
4 #include "mergesort.h"
5 #include "diff.h"
6 #include "diffcore.h"
7 #include "tag.h"
8 #include "blame.h"
9 #include "commit-slab.h"
11 define_commit_slab(blame_suspects, struct blame_origin *);
12 static struct blame_suspects blame_suspects;
14 struct blame_origin *get_blame_suspects(struct commit *commit)
16 struct blame_origin **result;
18 result = blame_suspects_peek(&blame_suspects, commit);
20 return result ? *result : NULL;
23 static void set_blame_suspects(struct commit *commit, struct blame_origin *origin)
25 *blame_suspects_at(&blame_suspects, commit) = origin;
28 void blame_origin_decref(struct blame_origin *o)
30 if (o && --o->refcnt <= 0) {
31 struct blame_origin *p, *l = NULL;
32 if (o->previous)
33 blame_origin_decref(o->previous);
34 free(o->file.ptr);
35 /* Should be present exactly once in commit chain */
36 for (p = get_blame_suspects(o->commit); p; l = p, p = p->next) {
37 if (p == o) {
38 if (l)
39 l->next = p->next;
40 else
41 set_blame_suspects(o->commit, p->next);
42 free(o);
43 return;
46 die("internal error in blame_origin_decref");
51 * Given a commit and a path in it, create a new origin structure.
52 * The callers that add blame to the scoreboard should use
53 * get_origin() to obtain shared, refcounted copy instead of calling
54 * this function directly.
56 static struct blame_origin *make_origin(struct commit *commit, const char *path)
58 struct blame_origin *o;
59 FLEX_ALLOC_STR(o, path, path);
60 o->commit = commit;
61 o->refcnt = 1;
62 o->next = get_blame_suspects(commit);
63 set_blame_suspects(commit, o);
64 return o;
68 * Locate an existing origin or create a new one.
69 * This moves the origin to front position in the commit util list.
71 static struct blame_origin *get_origin(struct commit *commit, const char *path)
73 struct blame_origin *o, *l;
75 for (o = get_blame_suspects(commit), l = NULL; o; l = o, o = o->next) {
76 if (!strcmp(o->path, path)) {
77 /* bump to front */
78 if (l) {
79 l->next = o->next;
80 o->next = get_blame_suspects(commit);
81 set_blame_suspects(commit, o);
83 return blame_origin_incref(o);
86 return make_origin(commit, path);
91 static void verify_working_tree_path(struct commit *work_tree, const char *path)
93 struct commit_list *parents;
94 int pos;
96 for (parents = work_tree->parents; parents; parents = parents->next) {
97 const struct object_id *commit_oid = &parents->item->object.oid;
98 struct object_id blob_oid;
99 unsigned mode;
101 if (!get_tree_entry(commit_oid, path, &blob_oid, &mode) &&
102 oid_object_info(the_repository, &blob_oid, NULL) == OBJ_BLOB)
103 return;
106 pos = cache_name_pos(path, strlen(path));
107 if (pos >= 0)
108 ; /* path is in the index */
109 else if (-1 - pos < active_nr &&
110 !strcmp(active_cache[-1 - pos]->name, path))
111 ; /* path is in the index, unmerged */
112 else
113 die("no such path '%s' in HEAD", path);
116 static struct commit_list **append_parent(struct commit_list **tail, const struct object_id *oid)
118 struct commit *parent;
120 parent = lookup_commit_reference(oid);
121 if (!parent)
122 die("no such commit %s", oid_to_hex(oid));
123 return &commit_list_insert(parent, tail)->next;
126 static void append_merge_parents(struct commit_list **tail)
128 int merge_head;
129 struct strbuf line = STRBUF_INIT;
131 merge_head = open(git_path_merge_head(), O_RDONLY);
132 if (merge_head < 0) {
133 if (errno == ENOENT)
134 return;
135 die("cannot open '%s' for reading", git_path_merge_head());
138 while (!strbuf_getwholeline_fd(&line, merge_head, '\n')) {
139 struct object_id oid;
140 if (line.len < GIT_SHA1_HEXSZ || get_oid_hex(line.buf, &oid))
141 die("unknown line in '%s': %s", git_path_merge_head(), line.buf);
142 tail = append_parent(tail, &oid);
144 close(merge_head);
145 strbuf_release(&line);
149 * This isn't as simple as passing sb->buf and sb->len, because we
150 * want to transfer ownership of the buffer to the commit (so we
151 * must use detach).
153 static void set_commit_buffer_from_strbuf(struct commit *c, struct strbuf *sb)
155 size_t len;
156 void *buf = strbuf_detach(sb, &len);
157 set_commit_buffer(c, buf, len);
161 * Prepare a dummy commit that represents the work tree (or staged) item.
162 * Note that annotating work tree item never works in the reverse.
164 static struct commit *fake_working_tree_commit(struct diff_options *opt,
165 const char *path,
166 const char *contents_from)
168 struct commit *commit;
169 struct blame_origin *origin;
170 struct commit_list **parent_tail, *parent;
171 struct object_id head_oid;
172 struct strbuf buf = STRBUF_INIT;
173 const char *ident;
174 time_t now;
175 int size, len;
176 struct cache_entry *ce;
177 unsigned mode;
178 struct strbuf msg = STRBUF_INIT;
180 read_cache();
181 time(&now);
182 commit = alloc_commit_node();
183 commit->object.parsed = 1;
184 commit->date = now;
185 parent_tail = &commit->parents;
187 if (!resolve_ref_unsafe("HEAD", RESOLVE_REF_READING, &head_oid, NULL))
188 die("no such ref: HEAD");
190 parent_tail = append_parent(parent_tail, &head_oid);
191 append_merge_parents(parent_tail);
192 verify_working_tree_path(commit, path);
194 origin = make_origin(commit, path);
196 ident = fmt_ident("Not Committed Yet", "not.committed.yet", NULL, 0);
197 strbuf_addstr(&msg, "tree 0000000000000000000000000000000000000000\n");
198 for (parent = commit->parents; parent; parent = parent->next)
199 strbuf_addf(&msg, "parent %s\n",
200 oid_to_hex(&parent->item->object.oid));
201 strbuf_addf(&msg,
202 "author %s\n"
203 "committer %s\n\n"
204 "Version of %s from %s\n",
205 ident, ident, path,
206 (!contents_from ? path :
207 (!strcmp(contents_from, "-") ? "standard input" : contents_from)));
208 set_commit_buffer_from_strbuf(commit, &msg);
210 if (!contents_from || strcmp("-", contents_from)) {
211 struct stat st;
212 const char *read_from;
213 char *buf_ptr;
214 unsigned long buf_len;
216 if (contents_from) {
217 if (stat(contents_from, &st) < 0)
218 die_errno("Cannot stat '%s'", contents_from);
219 read_from = contents_from;
221 else {
222 if (lstat(path, &st) < 0)
223 die_errno("Cannot lstat '%s'", path);
224 read_from = path;
226 mode = canon_mode(st.st_mode);
228 switch (st.st_mode & S_IFMT) {
229 case S_IFREG:
230 if (opt->flags.allow_textconv &&
231 textconv_object(read_from, mode, &null_oid, 0, &buf_ptr, &buf_len))
232 strbuf_attach(&buf, buf_ptr, buf_len, buf_len + 1);
233 else if (strbuf_read_file(&buf, read_from, st.st_size) != st.st_size)
234 die_errno("cannot open or read '%s'", read_from);
235 break;
236 case S_IFLNK:
237 if (strbuf_readlink(&buf, read_from, st.st_size) < 0)
238 die_errno("cannot readlink '%s'", read_from);
239 break;
240 default:
241 die("unsupported file type %s", read_from);
244 else {
245 /* Reading from stdin */
246 mode = 0;
247 if (strbuf_read(&buf, 0, 0) < 0)
248 die_errno("failed to read from stdin");
250 convert_to_git(&the_index, path, buf.buf, buf.len, &buf, 0);
251 origin->file.ptr = buf.buf;
252 origin->file.size = buf.len;
253 pretend_object_file(buf.buf, buf.len, OBJ_BLOB, &origin->blob_oid);
256 * Read the current index, replace the path entry with
257 * origin->blob_sha1 without mucking with its mode or type
258 * bits; we are not going to write this index out -- we just
259 * want to run "diff-index --cached".
261 discard_cache();
262 read_cache();
264 len = strlen(path);
265 if (!mode) {
266 int pos = cache_name_pos(path, len);
267 if (0 <= pos)
268 mode = active_cache[pos]->ce_mode;
269 else
270 /* Let's not bother reading from HEAD tree */
271 mode = S_IFREG | 0644;
273 size = cache_entry_size(len);
274 ce = xcalloc(1, size);
275 oidcpy(&ce->oid, &origin->blob_oid);
276 memcpy(ce->name, path, len);
277 ce->ce_flags = create_ce_flags(0);
278 ce->ce_namelen = len;
279 ce->ce_mode = create_ce_mode(mode);
280 add_cache_entry(ce, ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE);
282 cache_tree_invalidate_path(&the_index, path);
284 return commit;
289 static int diff_hunks(mmfile_t *file_a, mmfile_t *file_b,
290 xdl_emit_hunk_consume_func_t hunk_func, void *cb_data, int xdl_opts)
292 xpparam_t xpp = {0};
293 xdemitconf_t xecfg = {0};
294 xdemitcb_t ecb = {NULL};
296 xpp.flags = xdl_opts;
297 xecfg.hunk_func = hunk_func;
298 ecb.priv = cb_data;
299 return xdi_diff(file_a, file_b, &xpp, &xecfg, &ecb);
303 * Given an origin, prepare mmfile_t structure to be used by the
304 * diff machinery
306 static void fill_origin_blob(struct diff_options *opt,
307 struct blame_origin *o, mmfile_t *file, int *num_read_blob)
309 if (!o->file.ptr) {
310 enum object_type type;
311 unsigned long file_size;
313 (*num_read_blob)++;
314 if (opt->flags.allow_textconv &&
315 textconv_object(o->path, o->mode, &o->blob_oid, 1, &file->ptr, &file_size))
317 else
318 file->ptr = read_object_file(&o->blob_oid, &type,
319 &file_size);
320 file->size = file_size;
322 if (!file->ptr)
323 die("Cannot read blob %s for path %s",
324 oid_to_hex(&o->blob_oid),
325 o->path);
326 o->file = *file;
328 else
329 *file = o->file;
332 static void drop_origin_blob(struct blame_origin *o)
334 if (o->file.ptr) {
335 FREE_AND_NULL(o->file.ptr);
340 * Any merge of blames happens on lists of blames that arrived via
341 * different parents in a single suspect. In this case, we want to
342 * sort according to the suspect line numbers as opposed to the final
343 * image line numbers. The function body is somewhat longish because
344 * it avoids unnecessary writes.
347 static struct blame_entry *blame_merge(struct blame_entry *list1,
348 struct blame_entry *list2)
350 struct blame_entry *p1 = list1, *p2 = list2,
351 **tail = &list1;
353 if (!p1)
354 return p2;
355 if (!p2)
356 return p1;
358 if (p1->s_lno <= p2->s_lno) {
359 do {
360 tail = &p1->next;
361 if ((p1 = *tail) == NULL) {
362 *tail = p2;
363 return list1;
365 } while (p1->s_lno <= p2->s_lno);
367 for (;;) {
368 *tail = p2;
369 do {
370 tail = &p2->next;
371 if ((p2 = *tail) == NULL) {
372 *tail = p1;
373 return list1;
375 } while (p1->s_lno > p2->s_lno);
376 *tail = p1;
377 do {
378 tail = &p1->next;
379 if ((p1 = *tail) == NULL) {
380 *tail = p2;
381 return list1;
383 } while (p1->s_lno <= p2->s_lno);
387 static void *get_next_blame(const void *p)
389 return ((struct blame_entry *)p)->next;
392 static void set_next_blame(void *p1, void *p2)
394 ((struct blame_entry *)p1)->next = p2;
398 * Final image line numbers are all different, so we don't need a
399 * three-way comparison here.
402 static int compare_blame_final(const void *p1, const void *p2)
404 return ((struct blame_entry *)p1)->lno > ((struct blame_entry *)p2)->lno
405 ? 1 : -1;
408 static int compare_blame_suspect(const void *p1, const void *p2)
410 const struct blame_entry *s1 = p1, *s2 = p2;
412 * to allow for collating suspects, we sort according to the
413 * respective pointer value as the primary sorting criterion.
414 * The actual relation is pretty unimportant as long as it
415 * establishes a total order. Comparing as integers gives us
416 * that.
418 if (s1->suspect != s2->suspect)
419 return (intptr_t)s1->suspect > (intptr_t)s2->suspect ? 1 : -1;
420 if (s1->s_lno == s2->s_lno)
421 return 0;
422 return s1->s_lno > s2->s_lno ? 1 : -1;
425 void blame_sort_final(struct blame_scoreboard *sb)
427 sb->ent = llist_mergesort(sb->ent, get_next_blame, set_next_blame,
428 compare_blame_final);
431 static int compare_commits_by_reverse_commit_date(const void *a,
432 const void *b,
433 void *c)
435 return -compare_commits_by_commit_date(a, b, c);
439 * For debugging -- origin is refcounted, and this asserts that
440 * we do not underflow.
442 static void sanity_check_refcnt(struct blame_scoreboard *sb)
444 int baa = 0;
445 struct blame_entry *ent;
447 for (ent = sb->ent; ent; ent = ent->next) {
448 /* Nobody should have zero or negative refcnt */
449 if (ent->suspect->refcnt <= 0) {
450 fprintf(stderr, "%s in %s has negative refcnt %d\n",
451 ent->suspect->path,
452 oid_to_hex(&ent->suspect->commit->object.oid),
453 ent->suspect->refcnt);
454 baa = 1;
457 if (baa)
458 sb->on_sanity_fail(sb, baa);
462 * If two blame entries that are next to each other came from
463 * contiguous lines in the same origin (i.e. <commit, path> pair),
464 * merge them together.
466 void blame_coalesce(struct blame_scoreboard *sb)
468 struct blame_entry *ent, *next;
470 for (ent = sb->ent; ent && (next = ent->next); ent = next) {
471 if (ent->suspect == next->suspect &&
472 ent->s_lno + ent->num_lines == next->s_lno) {
473 ent->num_lines += next->num_lines;
474 ent->next = next->next;
475 blame_origin_decref(next->suspect);
476 free(next);
477 ent->score = 0;
478 next = ent; /* again */
482 if (sb->debug) /* sanity */
483 sanity_check_refcnt(sb);
487 * Merge the given sorted list of blames into a preexisting origin.
488 * If there were no previous blames to that commit, it is entered into
489 * the commit priority queue of the score board.
492 static void queue_blames(struct blame_scoreboard *sb, struct blame_origin *porigin,
493 struct blame_entry *sorted)
495 if (porigin->suspects)
496 porigin->suspects = blame_merge(porigin->suspects, sorted);
497 else {
498 struct blame_origin *o;
499 for (o = get_blame_suspects(porigin->commit); o; o = o->next) {
500 if (o->suspects) {
501 porigin->suspects = sorted;
502 return;
505 porigin->suspects = sorted;
506 prio_queue_put(&sb->commits, porigin->commit);
511 * Fill the blob_sha1 field of an origin if it hasn't, so that later
512 * call to fill_origin_blob() can use it to locate the data. blob_sha1
513 * for an origin is also used to pass the blame for the entire file to
514 * the parent to detect the case where a child's blob is identical to
515 * that of its parent's.
517 * This also fills origin->mode for corresponding tree path.
519 static int fill_blob_sha1_and_mode(struct blame_origin *origin)
521 if (!is_null_oid(&origin->blob_oid))
522 return 0;
523 if (get_tree_entry(&origin->commit->object.oid, origin->path, &origin->blob_oid, &origin->mode))
524 goto error_out;
525 if (oid_object_info(the_repository, &origin->blob_oid, NULL) != OBJ_BLOB)
526 goto error_out;
527 return 0;
528 error_out:
529 oidclr(&origin->blob_oid);
530 origin->mode = S_IFINVALID;
531 return -1;
535 * We have an origin -- check if the same path exists in the
536 * parent and return an origin structure to represent it.
538 static struct blame_origin *find_origin(struct commit *parent,
539 struct blame_origin *origin)
541 struct blame_origin *porigin;
542 struct diff_options diff_opts;
543 const char *paths[2];
545 /* First check any existing origins */
546 for (porigin = get_blame_suspects(parent); porigin; porigin = porigin->next)
547 if (!strcmp(porigin->path, origin->path)) {
549 * The same path between origin and its parent
550 * without renaming -- the most common case.
552 return blame_origin_incref (porigin);
555 /* See if the origin->path is different between parent
556 * and origin first. Most of the time they are the
557 * same and diff-tree is fairly efficient about this.
559 diff_setup(&diff_opts);
560 diff_opts.flags.recursive = 1;
561 diff_opts.detect_rename = 0;
562 diff_opts.output_format = DIFF_FORMAT_NO_OUTPUT;
563 paths[0] = origin->path;
564 paths[1] = NULL;
566 parse_pathspec(&diff_opts.pathspec,
567 PATHSPEC_ALL_MAGIC & ~PATHSPEC_LITERAL,
568 PATHSPEC_LITERAL_PATH, "", paths);
569 diff_setup_done(&diff_opts);
571 if (is_null_oid(&origin->commit->object.oid))
572 do_diff_cache(get_commit_tree_oid(parent), &diff_opts);
573 else
574 diff_tree_oid(get_commit_tree_oid(parent),
575 get_commit_tree_oid(origin->commit),
576 "", &diff_opts);
577 diffcore_std(&diff_opts);
579 if (!diff_queued_diff.nr) {
580 /* The path is the same as parent */
581 porigin = get_origin(parent, origin->path);
582 oidcpy(&porigin->blob_oid, &origin->blob_oid);
583 porigin->mode = origin->mode;
584 } else {
586 * Since origin->path is a pathspec, if the parent
587 * commit had it as a directory, we will see a whole
588 * bunch of deletion of files in the directory that we
589 * do not care about.
591 int i;
592 struct diff_filepair *p = NULL;
593 for (i = 0; i < diff_queued_diff.nr; i++) {
594 const char *name;
595 p = diff_queued_diff.queue[i];
596 name = p->one->path ? p->one->path : p->two->path;
597 if (!strcmp(name, origin->path))
598 break;
600 if (!p)
601 die("internal error in blame::find_origin");
602 switch (p->status) {
603 default:
604 die("internal error in blame::find_origin (%c)",
605 p->status);
606 case 'M':
607 porigin = get_origin(parent, origin->path);
608 oidcpy(&porigin->blob_oid, &p->one->oid);
609 porigin->mode = p->one->mode;
610 break;
611 case 'A':
612 case 'T':
613 /* Did not exist in parent, or type changed */
614 break;
617 diff_flush(&diff_opts);
618 clear_pathspec(&diff_opts.pathspec);
619 return porigin;
623 * We have an origin -- find the path that corresponds to it in its
624 * parent and return an origin structure to represent it.
626 static struct blame_origin *find_rename(struct commit *parent,
627 struct blame_origin *origin)
629 struct blame_origin *porigin = NULL;
630 struct diff_options diff_opts;
631 int i;
633 diff_setup(&diff_opts);
634 diff_opts.flags.recursive = 1;
635 diff_opts.detect_rename = DIFF_DETECT_RENAME;
636 diff_opts.output_format = DIFF_FORMAT_NO_OUTPUT;
637 diff_opts.single_follow = origin->path;
638 diff_setup_done(&diff_opts);
640 if (is_null_oid(&origin->commit->object.oid))
641 do_diff_cache(get_commit_tree_oid(parent), &diff_opts);
642 else
643 diff_tree_oid(get_commit_tree_oid(parent),
644 get_commit_tree_oid(origin->commit),
645 "", &diff_opts);
646 diffcore_std(&diff_opts);
648 for (i = 0; i < diff_queued_diff.nr; i++) {
649 struct diff_filepair *p = diff_queued_diff.queue[i];
650 if ((p->status == 'R' || p->status == 'C') &&
651 !strcmp(p->two->path, origin->path)) {
652 porigin = get_origin(parent, p->one->path);
653 oidcpy(&porigin->blob_oid, &p->one->oid);
654 porigin->mode = p->one->mode;
655 break;
658 diff_flush(&diff_opts);
659 clear_pathspec(&diff_opts.pathspec);
660 return porigin;
664 * Append a new blame entry to a given output queue.
666 static void add_blame_entry(struct blame_entry ***queue,
667 const struct blame_entry *src)
669 struct blame_entry *e = xmalloc(sizeof(*e));
670 memcpy(e, src, sizeof(*e));
671 blame_origin_incref(e->suspect);
673 e->next = **queue;
674 **queue = e;
675 *queue = &e->next;
679 * src typically is on-stack; we want to copy the information in it to
680 * a malloced blame_entry that gets added to the given queue. The
681 * origin of dst loses a refcnt.
683 static void dup_entry(struct blame_entry ***queue,
684 struct blame_entry *dst, struct blame_entry *src)
686 blame_origin_incref(src->suspect);
687 blame_origin_decref(dst->suspect);
688 memcpy(dst, src, sizeof(*src));
689 dst->next = **queue;
690 **queue = dst;
691 *queue = &dst->next;
694 const char *blame_nth_line(struct blame_scoreboard *sb, long lno)
696 return sb->final_buf + sb->lineno[lno];
700 * It is known that lines between tlno to same came from parent, and e
701 * has an overlap with that range. it also is known that parent's
702 * line plno corresponds to e's line tlno.
704 * <---- e ----->
705 * <------>
706 * <------------>
707 * <------------>
708 * <------------------>
710 * Split e into potentially three parts; before this chunk, the chunk
711 * to be blamed for the parent, and after that portion.
713 static void split_overlap(struct blame_entry *split,
714 struct blame_entry *e,
715 int tlno, int plno, int same,
716 struct blame_origin *parent)
718 int chunk_end_lno;
719 memset(split, 0, sizeof(struct blame_entry [3]));
721 if (e->s_lno < tlno) {
722 /* there is a pre-chunk part not blamed on parent */
723 split[0].suspect = blame_origin_incref(e->suspect);
724 split[0].lno = e->lno;
725 split[0].s_lno = e->s_lno;
726 split[0].num_lines = tlno - e->s_lno;
727 split[1].lno = e->lno + tlno - e->s_lno;
728 split[1].s_lno = plno;
730 else {
731 split[1].lno = e->lno;
732 split[1].s_lno = plno + (e->s_lno - tlno);
735 if (same < e->s_lno + e->num_lines) {
736 /* there is a post-chunk part not blamed on parent */
737 split[2].suspect = blame_origin_incref(e->suspect);
738 split[2].lno = e->lno + (same - e->s_lno);
739 split[2].s_lno = e->s_lno + (same - e->s_lno);
740 split[2].num_lines = e->s_lno + e->num_lines - same;
741 chunk_end_lno = split[2].lno;
743 else
744 chunk_end_lno = e->lno + e->num_lines;
745 split[1].num_lines = chunk_end_lno - split[1].lno;
748 * if it turns out there is nothing to blame the parent for,
749 * forget about the splitting. !split[1].suspect signals this.
751 if (split[1].num_lines < 1)
752 return;
753 split[1].suspect = blame_origin_incref(parent);
757 * split_overlap() divided an existing blame e into up to three parts
758 * in split. Any assigned blame is moved to queue to
759 * reflect the split.
761 static void split_blame(struct blame_entry ***blamed,
762 struct blame_entry ***unblamed,
763 struct blame_entry *split,
764 struct blame_entry *e)
766 if (split[0].suspect && split[2].suspect) {
767 /* The first part (reuse storage for the existing entry e) */
768 dup_entry(unblamed, e, &split[0]);
770 /* The last part -- me */
771 add_blame_entry(unblamed, &split[2]);
773 /* ... and the middle part -- parent */
774 add_blame_entry(blamed, &split[1]);
776 else if (!split[0].suspect && !split[2].suspect)
778 * The parent covers the entire area; reuse storage for
779 * e and replace it with the parent.
781 dup_entry(blamed, e, &split[1]);
782 else if (split[0].suspect) {
783 /* me and then parent */
784 dup_entry(unblamed, e, &split[0]);
785 add_blame_entry(blamed, &split[1]);
787 else {
788 /* parent and then me */
789 dup_entry(blamed, e, &split[1]);
790 add_blame_entry(unblamed, &split[2]);
795 * After splitting the blame, the origins used by the
796 * on-stack blame_entry should lose one refcnt each.
798 static void decref_split(struct blame_entry *split)
800 int i;
802 for (i = 0; i < 3; i++)
803 blame_origin_decref(split[i].suspect);
807 * reverse_blame reverses the list given in head, appending tail.
808 * That allows us to build lists in reverse order, then reverse them
809 * afterwards. This can be faster than building the list in proper
810 * order right away. The reason is that building in proper order
811 * requires writing a link in the _previous_ element, while building
812 * in reverse order just requires placing the list head into the
813 * _current_ element.
816 static struct blame_entry *reverse_blame(struct blame_entry *head,
817 struct blame_entry *tail)
819 while (head) {
820 struct blame_entry *next = head->next;
821 head->next = tail;
822 tail = head;
823 head = next;
825 return tail;
829 * Process one hunk from the patch between the current suspect for
830 * blame_entry e and its parent. This first blames any unfinished
831 * entries before the chunk (which is where target and parent start
832 * differing) on the parent, and then splits blame entries at the
833 * start and at the end of the difference region. Since use of -M and
834 * -C options may lead to overlapping/duplicate source line number
835 * ranges, all we can rely on from sorting/merging is the order of the
836 * first suspect line number.
838 static void blame_chunk(struct blame_entry ***dstq, struct blame_entry ***srcq,
839 int tlno, int offset, int same,
840 struct blame_origin *parent)
842 struct blame_entry *e = **srcq;
843 struct blame_entry *samep = NULL, *diffp = NULL;
845 while (e && e->s_lno < tlno) {
846 struct blame_entry *next = e->next;
848 * current record starts before differing portion. If
849 * it reaches into it, we need to split it up and
850 * examine the second part separately.
852 if (e->s_lno + e->num_lines > tlno) {
853 /* Move second half to a new record */
854 int len = tlno - e->s_lno;
855 struct blame_entry *n = xcalloc(1, sizeof (struct blame_entry));
856 n->suspect = e->suspect;
857 n->lno = e->lno + len;
858 n->s_lno = e->s_lno + len;
859 n->num_lines = e->num_lines - len;
860 e->num_lines = len;
861 e->score = 0;
862 /* Push new record to diffp */
863 n->next = diffp;
864 diffp = n;
865 } else
866 blame_origin_decref(e->suspect);
867 /* Pass blame for everything before the differing
868 * chunk to the parent */
869 e->suspect = blame_origin_incref(parent);
870 e->s_lno += offset;
871 e->next = samep;
872 samep = e;
873 e = next;
876 * As we don't know how much of a common stretch after this
877 * diff will occur, the currently blamed parts are all that we
878 * can assign to the parent for now.
881 if (samep) {
882 **dstq = reverse_blame(samep, **dstq);
883 *dstq = &samep->next;
886 * Prepend the split off portions: everything after e starts
887 * after the blameable portion.
889 e = reverse_blame(diffp, e);
892 * Now retain records on the target while parts are different
893 * from the parent.
895 samep = NULL;
896 diffp = NULL;
897 while (e && e->s_lno < same) {
898 struct blame_entry *next = e->next;
901 * If current record extends into sameness, need to split.
903 if (e->s_lno + e->num_lines > same) {
905 * Move second half to a new record to be
906 * processed by later chunks
908 int len = same - e->s_lno;
909 struct blame_entry *n = xcalloc(1, sizeof (struct blame_entry));
910 n->suspect = blame_origin_incref(e->suspect);
911 n->lno = e->lno + len;
912 n->s_lno = e->s_lno + len;
913 n->num_lines = e->num_lines - len;
914 e->num_lines = len;
915 e->score = 0;
916 /* Push new record to samep */
917 n->next = samep;
918 samep = n;
920 e->next = diffp;
921 diffp = e;
922 e = next;
924 **srcq = reverse_blame(diffp, reverse_blame(samep, e));
925 /* Move across elements that are in the unblamable portion */
926 if (diffp)
927 *srcq = &diffp->next;
930 struct blame_chunk_cb_data {
931 struct blame_origin *parent;
932 long offset;
933 struct blame_entry **dstq;
934 struct blame_entry **srcq;
937 /* diff chunks are from parent to target */
938 static int blame_chunk_cb(long start_a, long count_a,
939 long start_b, long count_b, void *data)
941 struct blame_chunk_cb_data *d = data;
942 if (start_a - start_b != d->offset)
943 die("internal error in blame::blame_chunk_cb");
944 blame_chunk(&d->dstq, &d->srcq, start_b, start_a - start_b,
945 start_b + count_b, d->parent);
946 d->offset = start_a + count_a - (start_b + count_b);
947 return 0;
951 * We are looking at the origin 'target' and aiming to pass blame
952 * for the lines it is suspected to its parent. Run diff to find
953 * which lines came from parent and pass blame for them.
955 static void pass_blame_to_parent(struct blame_scoreboard *sb,
956 struct blame_origin *target,
957 struct blame_origin *parent)
959 mmfile_t file_p, file_o;
960 struct blame_chunk_cb_data d;
961 struct blame_entry *newdest = NULL;
963 if (!target->suspects)
964 return; /* nothing remains for this target */
966 d.parent = parent;
967 d.offset = 0;
968 d.dstq = &newdest; d.srcq = &target->suspects;
970 fill_origin_blob(&sb->revs->diffopt, parent, &file_p, &sb->num_read_blob);
971 fill_origin_blob(&sb->revs->diffopt, target, &file_o, &sb->num_read_blob);
972 sb->num_get_patch++;
974 if (diff_hunks(&file_p, &file_o, blame_chunk_cb, &d, sb->xdl_opts))
975 die("unable to generate diff (%s -> %s)",
976 oid_to_hex(&parent->commit->object.oid),
977 oid_to_hex(&target->commit->object.oid));
978 /* The rest are the same as the parent */
979 blame_chunk(&d.dstq, &d.srcq, INT_MAX, d.offset, INT_MAX, parent);
980 *d.dstq = NULL;
981 queue_blames(sb, parent, newdest);
983 return;
987 * The lines in blame_entry after splitting blames many times can become
988 * very small and trivial, and at some point it becomes pointless to
989 * blame the parents. E.g. "\t\t}\n\t}\n\n" appears everywhere in any
990 * ordinary C program, and it is not worth to say it was copied from
991 * totally unrelated file in the parent.
993 * Compute how trivial the lines in the blame_entry are.
995 unsigned blame_entry_score(struct blame_scoreboard *sb, struct blame_entry *e)
997 unsigned score;
998 const char *cp, *ep;
1000 if (e->score)
1001 return e->score;
1003 score = 1;
1004 cp = blame_nth_line(sb, e->lno);
1005 ep = blame_nth_line(sb, e->lno + e->num_lines);
1006 while (cp < ep) {
1007 unsigned ch = *((unsigned char *)cp);
1008 if (isalnum(ch))
1009 score++;
1010 cp++;
1012 e->score = score;
1013 return score;
1017 * best_so_far[] and potential[] are both a split of an existing blame_entry
1018 * that passes blame to the parent. Maintain best_so_far the best split so
1019 * far, by comparing potential and best_so_far and copying potential into
1020 * bst_so_far as needed.
1022 static void copy_split_if_better(struct blame_scoreboard *sb,
1023 struct blame_entry *best_so_far,
1024 struct blame_entry *potential)
1026 int i;
1028 if (!potential[1].suspect)
1029 return;
1030 if (best_so_far[1].suspect) {
1031 if (blame_entry_score(sb, &potential[1]) <
1032 blame_entry_score(sb, &best_so_far[1]))
1033 return;
1036 for (i = 0; i < 3; i++)
1037 blame_origin_incref(potential[i].suspect);
1038 decref_split(best_so_far);
1039 memcpy(best_so_far, potential, sizeof(struct blame_entry[3]));
1043 * We are looking at a part of the final image represented by
1044 * ent (tlno and same are offset by ent->s_lno).
1045 * tlno is where we are looking at in the final image.
1046 * up to (but not including) same match preimage.
1047 * plno is where we are looking at in the preimage.
1049 * <-------------- final image ---------------------->
1050 * <------ent------>
1051 * ^tlno ^same
1052 * <---------preimage----->
1053 * ^plno
1055 * All line numbers are 0-based.
1057 static void handle_split(struct blame_scoreboard *sb,
1058 struct blame_entry *ent,
1059 int tlno, int plno, int same,
1060 struct blame_origin *parent,
1061 struct blame_entry *split)
1063 if (ent->num_lines <= tlno)
1064 return;
1065 if (tlno < same) {
1066 struct blame_entry potential[3];
1067 tlno += ent->s_lno;
1068 same += ent->s_lno;
1069 split_overlap(potential, ent, tlno, plno, same, parent);
1070 copy_split_if_better(sb, split, potential);
1071 decref_split(potential);
1075 struct handle_split_cb_data {
1076 struct blame_scoreboard *sb;
1077 struct blame_entry *ent;
1078 struct blame_origin *parent;
1079 struct blame_entry *split;
1080 long plno;
1081 long tlno;
1084 static int handle_split_cb(long start_a, long count_a,
1085 long start_b, long count_b, void *data)
1087 struct handle_split_cb_data *d = data;
1088 handle_split(d->sb, d->ent, d->tlno, d->plno, start_b, d->parent,
1089 d->split);
1090 d->plno = start_a + count_a;
1091 d->tlno = start_b + count_b;
1092 return 0;
1096 * Find the lines from parent that are the same as ent so that
1097 * we can pass blames to it. file_p has the blob contents for
1098 * the parent.
1100 static void find_copy_in_blob(struct blame_scoreboard *sb,
1101 struct blame_entry *ent,
1102 struct blame_origin *parent,
1103 struct blame_entry *split,
1104 mmfile_t *file_p)
1106 const char *cp;
1107 mmfile_t file_o;
1108 struct handle_split_cb_data d;
1110 memset(&d, 0, sizeof(d));
1111 d.sb = sb; d.ent = ent; d.parent = parent; d.split = split;
1113 * Prepare mmfile that contains only the lines in ent.
1115 cp = blame_nth_line(sb, ent->lno);
1116 file_o.ptr = (char *) cp;
1117 file_o.size = blame_nth_line(sb, ent->lno + ent->num_lines) - cp;
1120 * file_o is a part of final image we are annotating.
1121 * file_p partially may match that image.
1123 memset(split, 0, sizeof(struct blame_entry [3]));
1124 if (diff_hunks(file_p, &file_o, handle_split_cb, &d, sb->xdl_opts))
1125 die("unable to generate diff (%s)",
1126 oid_to_hex(&parent->commit->object.oid));
1127 /* remainder, if any, all match the preimage */
1128 handle_split(sb, ent, d.tlno, d.plno, ent->num_lines, parent, split);
1131 /* Move all blame entries from list *source that have a score smaller
1132 * than score_min to the front of list *small.
1133 * Returns a pointer to the link pointing to the old head of the small list.
1136 static struct blame_entry **filter_small(struct blame_scoreboard *sb,
1137 struct blame_entry **small,
1138 struct blame_entry **source,
1139 unsigned score_min)
1141 struct blame_entry *p = *source;
1142 struct blame_entry *oldsmall = *small;
1143 while (p) {
1144 if (blame_entry_score(sb, p) <= score_min) {
1145 *small = p;
1146 small = &p->next;
1147 p = *small;
1148 } else {
1149 *source = p;
1150 source = &p->next;
1151 p = *source;
1154 *small = oldsmall;
1155 *source = NULL;
1156 return small;
1160 * See if lines currently target is suspected for can be attributed to
1161 * parent.
1163 static void find_move_in_parent(struct blame_scoreboard *sb,
1164 struct blame_entry ***blamed,
1165 struct blame_entry **toosmall,
1166 struct blame_origin *target,
1167 struct blame_origin *parent)
1169 struct blame_entry *e, split[3];
1170 struct blame_entry *unblamed = target->suspects;
1171 struct blame_entry *leftover = NULL;
1172 mmfile_t file_p;
1174 if (!unblamed)
1175 return; /* nothing remains for this target */
1177 fill_origin_blob(&sb->revs->diffopt, parent, &file_p, &sb->num_read_blob);
1178 if (!file_p.ptr)
1179 return;
1181 /* At each iteration, unblamed has a NULL-terminated list of
1182 * entries that have not yet been tested for blame. leftover
1183 * contains the reversed list of entries that have been tested
1184 * without being assignable to the parent.
1186 do {
1187 struct blame_entry **unblamedtail = &unblamed;
1188 struct blame_entry *next;
1189 for (e = unblamed; e; e = next) {
1190 next = e->next;
1191 find_copy_in_blob(sb, e, parent, split, &file_p);
1192 if (split[1].suspect &&
1193 sb->move_score < blame_entry_score(sb, &split[1])) {
1194 split_blame(blamed, &unblamedtail, split, e);
1195 } else {
1196 e->next = leftover;
1197 leftover = e;
1199 decref_split(split);
1201 *unblamedtail = NULL;
1202 toosmall = filter_small(sb, toosmall, &unblamed, sb->move_score);
1203 } while (unblamed);
1204 target->suspects = reverse_blame(leftover, NULL);
1207 struct blame_list {
1208 struct blame_entry *ent;
1209 struct blame_entry split[3];
1213 * Count the number of entries the target is suspected for,
1214 * and prepare a list of entry and the best split.
1216 static struct blame_list *setup_blame_list(struct blame_entry *unblamed,
1217 int *num_ents_p)
1219 struct blame_entry *e;
1220 int num_ents, i;
1221 struct blame_list *blame_list = NULL;
1223 for (e = unblamed, num_ents = 0; e; e = e->next)
1224 num_ents++;
1225 if (num_ents) {
1226 blame_list = xcalloc(num_ents, sizeof(struct blame_list));
1227 for (e = unblamed, i = 0; e; e = e->next)
1228 blame_list[i++].ent = e;
1230 *num_ents_p = num_ents;
1231 return blame_list;
1235 * For lines target is suspected for, see if we can find code movement
1236 * across file boundary from the parent commit. porigin is the path
1237 * in the parent we already tried.
1239 static void find_copy_in_parent(struct blame_scoreboard *sb,
1240 struct blame_entry ***blamed,
1241 struct blame_entry **toosmall,
1242 struct blame_origin *target,
1243 struct commit *parent,
1244 struct blame_origin *porigin,
1245 int opt)
1247 struct diff_options diff_opts;
1248 int i, j;
1249 struct blame_list *blame_list;
1250 int num_ents;
1251 struct blame_entry *unblamed = target->suspects;
1252 struct blame_entry *leftover = NULL;
1254 if (!unblamed)
1255 return; /* nothing remains for this target */
1257 diff_setup(&diff_opts);
1258 diff_opts.flags.recursive = 1;
1259 diff_opts.output_format = DIFF_FORMAT_NO_OUTPUT;
1261 diff_setup_done(&diff_opts);
1263 /* Try "find copies harder" on new path if requested;
1264 * we do not want to use diffcore_rename() actually to
1265 * match things up; find_copies_harder is set only to
1266 * force diff_tree_oid() to feed all filepairs to diff_queue,
1267 * and this code needs to be after diff_setup_done(), which
1268 * usually makes find-copies-harder imply copy detection.
1270 if ((opt & PICKAXE_BLAME_COPY_HARDEST)
1271 || ((opt & PICKAXE_BLAME_COPY_HARDER)
1272 && (!porigin || strcmp(target->path, porigin->path))))
1273 diff_opts.flags.find_copies_harder = 1;
1275 if (is_null_oid(&target->commit->object.oid))
1276 do_diff_cache(get_commit_tree_oid(parent), &diff_opts);
1277 else
1278 diff_tree_oid(get_commit_tree_oid(parent),
1279 get_commit_tree_oid(target->commit),
1280 "", &diff_opts);
1282 if (!diff_opts.flags.find_copies_harder)
1283 diffcore_std(&diff_opts);
1285 do {
1286 struct blame_entry **unblamedtail = &unblamed;
1287 blame_list = setup_blame_list(unblamed, &num_ents);
1289 for (i = 0; i < diff_queued_diff.nr; i++) {
1290 struct diff_filepair *p = diff_queued_diff.queue[i];
1291 struct blame_origin *norigin;
1292 mmfile_t file_p;
1293 struct blame_entry potential[3];
1295 if (!DIFF_FILE_VALID(p->one))
1296 continue; /* does not exist in parent */
1297 if (S_ISGITLINK(p->one->mode))
1298 continue; /* ignore git links */
1299 if (porigin && !strcmp(p->one->path, porigin->path))
1300 /* find_move already dealt with this path */
1301 continue;
1303 norigin = get_origin(parent, p->one->path);
1304 oidcpy(&norigin->blob_oid, &p->one->oid);
1305 norigin->mode = p->one->mode;
1306 fill_origin_blob(&sb->revs->diffopt, norigin, &file_p, &sb->num_read_blob);
1307 if (!file_p.ptr)
1308 continue;
1310 for (j = 0; j < num_ents; j++) {
1311 find_copy_in_blob(sb, blame_list[j].ent,
1312 norigin, potential, &file_p);
1313 copy_split_if_better(sb, blame_list[j].split,
1314 potential);
1315 decref_split(potential);
1317 blame_origin_decref(norigin);
1320 for (j = 0; j < num_ents; j++) {
1321 struct blame_entry *split = blame_list[j].split;
1322 if (split[1].suspect &&
1323 sb->copy_score < blame_entry_score(sb, &split[1])) {
1324 split_blame(blamed, &unblamedtail, split,
1325 blame_list[j].ent);
1326 } else {
1327 blame_list[j].ent->next = leftover;
1328 leftover = blame_list[j].ent;
1330 decref_split(split);
1332 free(blame_list);
1333 *unblamedtail = NULL;
1334 toosmall = filter_small(sb, toosmall, &unblamed, sb->copy_score);
1335 } while (unblamed);
1336 target->suspects = reverse_blame(leftover, NULL);
1337 diff_flush(&diff_opts);
1338 clear_pathspec(&diff_opts.pathspec);
1342 * The blobs of origin and porigin exactly match, so everything
1343 * origin is suspected for can be blamed on the parent.
1345 static void pass_whole_blame(struct blame_scoreboard *sb,
1346 struct blame_origin *origin, struct blame_origin *porigin)
1348 struct blame_entry *e, *suspects;
1350 if (!porigin->file.ptr && origin->file.ptr) {
1351 /* Steal its file */
1352 porigin->file = origin->file;
1353 origin->file.ptr = NULL;
1355 suspects = origin->suspects;
1356 origin->suspects = NULL;
1357 for (e = suspects; e; e = e->next) {
1358 blame_origin_incref(porigin);
1359 blame_origin_decref(e->suspect);
1360 e->suspect = porigin;
1362 queue_blames(sb, porigin, suspects);
1366 * We pass blame from the current commit to its parents. We keep saying
1367 * "parent" (and "porigin"), but what we mean is to find scapegoat to
1368 * exonerate ourselves.
1370 static struct commit_list *first_scapegoat(struct rev_info *revs, struct commit *commit,
1371 int reverse)
1373 if (!reverse) {
1374 if (revs->first_parent_only &&
1375 commit->parents &&
1376 commit->parents->next) {
1377 free_commit_list(commit->parents->next);
1378 commit->parents->next = NULL;
1380 return commit->parents;
1382 return lookup_decoration(&revs->children, &commit->object);
1385 static int num_scapegoats(struct rev_info *revs, struct commit *commit, int reverse)
1387 struct commit_list *l = first_scapegoat(revs, commit, reverse);
1388 return commit_list_count(l);
1391 /* Distribute collected unsorted blames to the respected sorted lists
1392 * in the various origins.
1394 static void distribute_blame(struct blame_scoreboard *sb, struct blame_entry *blamed)
1396 blamed = llist_mergesort(blamed, get_next_blame, set_next_blame,
1397 compare_blame_suspect);
1398 while (blamed)
1400 struct blame_origin *porigin = blamed->suspect;
1401 struct blame_entry *suspects = NULL;
1402 do {
1403 struct blame_entry *next = blamed->next;
1404 blamed->next = suspects;
1405 suspects = blamed;
1406 blamed = next;
1407 } while (blamed && blamed->suspect == porigin);
1408 suspects = reverse_blame(suspects, NULL);
1409 queue_blames(sb, porigin, suspects);
1413 #define MAXSG 16
1415 static void pass_blame(struct blame_scoreboard *sb, struct blame_origin *origin, int opt)
1417 struct rev_info *revs = sb->revs;
1418 int i, pass, num_sg;
1419 struct commit *commit = origin->commit;
1420 struct commit_list *sg;
1421 struct blame_origin *sg_buf[MAXSG];
1422 struct blame_origin *porigin, **sg_origin = sg_buf;
1423 struct blame_entry *toosmall = NULL;
1424 struct blame_entry *blames, **blametail = &blames;
1426 num_sg = num_scapegoats(revs, commit, sb->reverse);
1427 if (!num_sg)
1428 goto finish;
1429 else if (num_sg < ARRAY_SIZE(sg_buf))
1430 memset(sg_buf, 0, sizeof(sg_buf));
1431 else
1432 sg_origin = xcalloc(num_sg, sizeof(*sg_origin));
1435 * The first pass looks for unrenamed path to optimize for
1436 * common cases, then we look for renames in the second pass.
1438 for (pass = 0; pass < 2 - sb->no_whole_file_rename; pass++) {
1439 struct blame_origin *(*find)(struct commit *, struct blame_origin *);
1440 find = pass ? find_rename : find_origin;
1442 for (i = 0, sg = first_scapegoat(revs, commit, sb->reverse);
1443 i < num_sg && sg;
1444 sg = sg->next, i++) {
1445 struct commit *p = sg->item;
1446 int j, same;
1448 if (sg_origin[i])
1449 continue;
1450 if (parse_commit(p))
1451 continue;
1452 porigin = find(p, origin);
1453 if (!porigin)
1454 continue;
1455 if (!oidcmp(&porigin->blob_oid, &origin->blob_oid)) {
1456 pass_whole_blame(sb, origin, porigin);
1457 blame_origin_decref(porigin);
1458 goto finish;
1460 for (j = same = 0; j < i; j++)
1461 if (sg_origin[j] &&
1462 !oidcmp(&sg_origin[j]->blob_oid, &porigin->blob_oid)) {
1463 same = 1;
1464 break;
1466 if (!same)
1467 sg_origin[i] = porigin;
1468 else
1469 blame_origin_decref(porigin);
1473 sb->num_commits++;
1474 for (i = 0, sg = first_scapegoat(revs, commit, sb->reverse);
1475 i < num_sg && sg;
1476 sg = sg->next, i++) {
1477 struct blame_origin *porigin = sg_origin[i];
1478 if (!porigin)
1479 continue;
1480 if (!origin->previous) {
1481 blame_origin_incref(porigin);
1482 origin->previous = porigin;
1484 pass_blame_to_parent(sb, origin, porigin);
1485 if (!origin->suspects)
1486 goto finish;
1490 * Optionally find moves in parents' files.
1492 if (opt & PICKAXE_BLAME_MOVE) {
1493 filter_small(sb, &toosmall, &origin->suspects, sb->move_score);
1494 if (origin->suspects) {
1495 for (i = 0, sg = first_scapegoat(revs, commit, sb->reverse);
1496 i < num_sg && sg;
1497 sg = sg->next, i++) {
1498 struct blame_origin *porigin = sg_origin[i];
1499 if (!porigin)
1500 continue;
1501 find_move_in_parent(sb, &blametail, &toosmall, origin, porigin);
1502 if (!origin->suspects)
1503 break;
1509 * Optionally find copies from parents' files.
1511 if (opt & PICKAXE_BLAME_COPY) {
1512 if (sb->copy_score > sb->move_score)
1513 filter_small(sb, &toosmall, &origin->suspects, sb->copy_score);
1514 else if (sb->copy_score < sb->move_score) {
1515 origin->suspects = blame_merge(origin->suspects, toosmall);
1516 toosmall = NULL;
1517 filter_small(sb, &toosmall, &origin->suspects, sb->copy_score);
1519 if (!origin->suspects)
1520 goto finish;
1522 for (i = 0, sg = first_scapegoat(revs, commit, sb->reverse);
1523 i < num_sg && sg;
1524 sg = sg->next, i++) {
1525 struct blame_origin *porigin = sg_origin[i];
1526 find_copy_in_parent(sb, &blametail, &toosmall,
1527 origin, sg->item, porigin, opt);
1528 if (!origin->suspects)
1529 goto finish;
1533 finish:
1534 *blametail = NULL;
1535 distribute_blame(sb, blames);
1537 * prepend toosmall to origin->suspects
1539 * There is no point in sorting: this ends up on a big
1540 * unsorted list in the caller anyway.
1542 if (toosmall) {
1543 struct blame_entry **tail = &toosmall;
1544 while (*tail)
1545 tail = &(*tail)->next;
1546 *tail = origin->suspects;
1547 origin->suspects = toosmall;
1549 for (i = 0; i < num_sg; i++) {
1550 if (sg_origin[i]) {
1551 drop_origin_blob(sg_origin[i]);
1552 blame_origin_decref(sg_origin[i]);
1555 drop_origin_blob(origin);
1556 if (sg_buf != sg_origin)
1557 free(sg_origin);
1561 * The main loop -- while we have blobs with lines whose true origin
1562 * is still unknown, pick one blob, and allow its lines to pass blames
1563 * to its parents. */
1564 void assign_blame(struct blame_scoreboard *sb, int opt)
1566 struct rev_info *revs = sb->revs;
1567 struct commit *commit = prio_queue_get(&sb->commits);
1569 while (commit) {
1570 struct blame_entry *ent;
1571 struct blame_origin *suspect = get_blame_suspects(commit);
1573 /* find one suspect to break down */
1574 while (suspect && !suspect->suspects)
1575 suspect = suspect->next;
1577 if (!suspect) {
1578 commit = prio_queue_get(&sb->commits);
1579 continue;
1582 assert(commit == suspect->commit);
1585 * We will use this suspect later in the loop,
1586 * so hold onto it in the meantime.
1588 blame_origin_incref(suspect);
1589 parse_commit(commit);
1590 if (sb->reverse ||
1591 (!(commit->object.flags & UNINTERESTING) &&
1592 !(revs->max_age != -1 && commit->date < revs->max_age)))
1593 pass_blame(sb, suspect, opt);
1594 else {
1595 commit->object.flags |= UNINTERESTING;
1596 if (commit->object.parsed)
1597 mark_parents_uninteresting(commit);
1599 /* treat root commit as boundary */
1600 if (!commit->parents && !sb->show_root)
1601 commit->object.flags |= UNINTERESTING;
1603 /* Take responsibility for the remaining entries */
1604 ent = suspect->suspects;
1605 if (ent) {
1606 suspect->guilty = 1;
1607 for (;;) {
1608 struct blame_entry *next = ent->next;
1609 if (sb->found_guilty_entry)
1610 sb->found_guilty_entry(ent, sb->found_guilty_entry_data);
1611 if (next) {
1612 ent = next;
1613 continue;
1615 ent->next = sb->ent;
1616 sb->ent = suspect->suspects;
1617 suspect->suspects = NULL;
1618 break;
1621 blame_origin_decref(suspect);
1623 if (sb->debug) /* sanity */
1624 sanity_check_refcnt(sb);
1628 static const char *get_next_line(const char *start, const char *end)
1630 const char *nl = memchr(start, '\n', end - start);
1631 return nl ? nl + 1 : end;
1635 * To allow quick access to the contents of nth line in the
1636 * final image, prepare an index in the scoreboard.
1638 static int prepare_lines(struct blame_scoreboard *sb)
1640 const char *buf = sb->final_buf;
1641 unsigned long len = sb->final_buf_size;
1642 const char *end = buf + len;
1643 const char *p;
1644 int *lineno;
1645 int num = 0;
1647 for (p = buf; p < end; p = get_next_line(p, end))
1648 num++;
1650 ALLOC_ARRAY(sb->lineno, num + 1);
1651 lineno = sb->lineno;
1653 for (p = buf; p < end; p = get_next_line(p, end))
1654 *lineno++ = p - buf;
1656 *lineno = len;
1658 sb->num_lines = num;
1659 return sb->num_lines;
1662 static struct commit *find_single_final(struct rev_info *revs,
1663 const char **name_p)
1665 int i;
1666 struct commit *found = NULL;
1667 const char *name = NULL;
1669 for (i = 0; i < revs->pending.nr; i++) {
1670 struct object *obj = revs->pending.objects[i].item;
1671 if (obj->flags & UNINTERESTING)
1672 continue;
1673 obj = deref_tag(obj, NULL, 0);
1674 if (obj->type != OBJ_COMMIT)
1675 die("Non commit %s?", revs->pending.objects[i].name);
1676 if (found)
1677 die("More than one commit to dig from %s and %s?",
1678 revs->pending.objects[i].name, name);
1679 found = (struct commit *)obj;
1680 name = revs->pending.objects[i].name;
1682 if (name_p)
1683 *name_p = xstrdup_or_null(name);
1684 return found;
1687 static struct commit *dwim_reverse_initial(struct rev_info *revs,
1688 const char **name_p)
1691 * DWIM "git blame --reverse ONE -- PATH" as
1692 * "git blame --reverse ONE..HEAD -- PATH" but only do so
1693 * when it makes sense.
1695 struct object *obj;
1696 struct commit *head_commit;
1697 struct object_id head_oid;
1699 if (revs->pending.nr != 1)
1700 return NULL;
1702 /* Is that sole rev a committish? */
1703 obj = revs->pending.objects[0].item;
1704 obj = deref_tag(obj, NULL, 0);
1705 if (obj->type != OBJ_COMMIT)
1706 return NULL;
1708 /* Do we have HEAD? */
1709 if (!resolve_ref_unsafe("HEAD", RESOLVE_REF_READING, &head_oid, NULL))
1710 return NULL;
1711 head_commit = lookup_commit_reference_gently(&head_oid, 1);
1712 if (!head_commit)
1713 return NULL;
1715 /* Turn "ONE" into "ONE..HEAD" then */
1716 obj->flags |= UNINTERESTING;
1717 add_pending_object(revs, &head_commit->object, "HEAD");
1719 if (name_p)
1720 *name_p = revs->pending.objects[0].name;
1721 return (struct commit *)obj;
1724 static struct commit *find_single_initial(struct rev_info *revs,
1725 const char **name_p)
1727 int i;
1728 struct commit *found = NULL;
1729 const char *name = NULL;
1732 * There must be one and only one negative commit, and it must be
1733 * the boundary.
1735 for (i = 0; i < revs->pending.nr; i++) {
1736 struct object *obj = revs->pending.objects[i].item;
1737 if (!(obj->flags & UNINTERESTING))
1738 continue;
1739 obj = deref_tag(obj, NULL, 0);
1740 if (obj->type != OBJ_COMMIT)
1741 die("Non commit %s?", revs->pending.objects[i].name);
1742 if (found)
1743 die("More than one commit to dig up from, %s and %s?",
1744 revs->pending.objects[i].name, name);
1745 found = (struct commit *) obj;
1746 name = revs->pending.objects[i].name;
1749 if (!name)
1750 found = dwim_reverse_initial(revs, &name);
1751 if (!name)
1752 die("No commit to dig up from?");
1754 if (name_p)
1755 *name_p = xstrdup(name);
1756 return found;
1759 void init_scoreboard(struct blame_scoreboard *sb)
1761 memset(sb, 0, sizeof(struct blame_scoreboard));
1762 sb->move_score = BLAME_DEFAULT_MOVE_SCORE;
1763 sb->copy_score = BLAME_DEFAULT_COPY_SCORE;
1766 void setup_scoreboard(struct blame_scoreboard *sb, const char *path, struct blame_origin **orig)
1768 const char *final_commit_name = NULL;
1769 struct blame_origin *o;
1770 struct commit *final_commit = NULL;
1771 enum object_type type;
1773 init_blame_suspects(&blame_suspects);
1775 if (sb->reverse && sb->contents_from)
1776 die(_("--contents and --reverse do not blend well."));
1778 if (!sb->reverse) {
1779 sb->final = find_single_final(sb->revs, &final_commit_name);
1780 sb->commits.compare = compare_commits_by_commit_date;
1781 } else {
1782 sb->final = find_single_initial(sb->revs, &final_commit_name);
1783 sb->commits.compare = compare_commits_by_reverse_commit_date;
1786 if (sb->final && sb->contents_from)
1787 die(_("cannot use --contents with final commit object name"));
1789 if (sb->reverse && sb->revs->first_parent_only)
1790 sb->revs->children.name = NULL;
1792 if (!sb->final) {
1794 * "--not A B -- path" without anything positive;
1795 * do not default to HEAD, but use the working tree
1796 * or "--contents".
1798 setup_work_tree();
1799 sb->final = fake_working_tree_commit(&sb->revs->diffopt,
1800 path, sb->contents_from);
1801 add_pending_object(sb->revs, &(sb->final->object), ":");
1804 if (sb->reverse && sb->revs->first_parent_only) {
1805 final_commit = find_single_final(sb->revs, NULL);
1806 if (!final_commit)
1807 die(_("--reverse and --first-parent together require specified latest commit"));
1811 * If we have bottom, this will mark the ancestors of the
1812 * bottom commits we would reach while traversing as
1813 * uninteresting.
1815 if (prepare_revision_walk(sb->revs))
1816 die(_("revision walk setup failed"));
1818 if (sb->reverse && sb->revs->first_parent_only) {
1819 struct commit *c = final_commit;
1821 sb->revs->children.name = "children";
1822 while (c->parents &&
1823 oidcmp(&c->object.oid, &sb->final->object.oid)) {
1824 struct commit_list *l = xcalloc(1, sizeof(*l));
1826 l->item = c;
1827 if (add_decoration(&sb->revs->children,
1828 &c->parents->item->object, l))
1829 BUG("not unique item in first-parent chain");
1830 c = c->parents->item;
1833 if (oidcmp(&c->object.oid, &sb->final->object.oid))
1834 die(_("--reverse --first-parent together require range along first-parent chain"));
1837 if (is_null_oid(&sb->final->object.oid)) {
1838 o = get_blame_suspects(sb->final);
1839 sb->final_buf = xmemdupz(o->file.ptr, o->file.size);
1840 sb->final_buf_size = o->file.size;
1842 else {
1843 o = get_origin(sb->final, path);
1844 if (fill_blob_sha1_and_mode(o))
1845 die(_("no such path %s in %s"), path, final_commit_name);
1847 if (sb->revs->diffopt.flags.allow_textconv &&
1848 textconv_object(path, o->mode, &o->blob_oid, 1, (char **) &sb->final_buf,
1849 &sb->final_buf_size))
1851 else
1852 sb->final_buf = read_object_file(&o->blob_oid, &type,
1853 &sb->final_buf_size);
1855 if (!sb->final_buf)
1856 die(_("cannot read blob %s for path %s"),
1857 oid_to_hex(&o->blob_oid),
1858 path);
1860 sb->num_read_blob++;
1861 prepare_lines(sb);
1863 if (orig)
1864 *orig = o;
1866 free((char *)final_commit_name);
1871 struct blame_entry *blame_entry_prepend(struct blame_entry *head,
1872 long start, long end,
1873 struct blame_origin *o)
1875 struct blame_entry *new_head = xcalloc(1, sizeof(struct blame_entry));
1876 new_head->lno = start;
1877 new_head->num_lines = end - start;
1878 new_head->suspect = o;
1879 new_head->s_lno = start;
1880 new_head->next = head;
1881 blame_origin_incref(o);
1882 return new_head;