blame: move scoreboard-related methods to libgit
[git/git.git] / blame.c
CommitLineData
072bf432
JS
1#include "cache.h"
2#include "refs.h"
3#include "cache-tree.h"
b543bb1c
JS
4#include "mergesort.h"
5#include "diff.h"
6#include "diffcore.h"
f5dd754c
JS
7#include "blame.h"
8
9void blame_origin_decref(struct blame_origin *o)
10{
11 if (o && --o->refcnt <= 0) {
12 struct blame_origin *p, *l = NULL;
13 if (o->previous)
14 blame_origin_decref(o->previous);
15 free(o->file.ptr);
16 /* Should be present exactly once in commit chain */
17 for (p = o->commit->util; p; l = p, p = p->next) {
18 if (p == o) {
19 if (l)
20 l->next = p->next;
21 else
22 o->commit->util = p->next;
23 free(o);
24 return;
25 }
26 }
27 die("internal error in blame_origin_decref");
28 }
29}
30
31/*
32 * Given a commit and a path in it, create a new origin structure.
33 * The callers that add blame to the scoreboard should use
34 * get_origin() to obtain shared, refcounted copy instead of calling
35 * this function directly.
36 */
072bf432 37static struct blame_origin *make_origin(struct commit *commit, const char *path)
f5dd754c
JS
38{
39 struct blame_origin *o;
40 FLEX_ALLOC_STR(o, path, path);
41 o->commit = commit;
42 o->refcnt = 1;
43 o->next = commit->util;
44 commit->util = o;
45 return o;
46}
47
48/*
49 * Locate an existing origin or create a new one.
50 * This moves the origin to front position in the commit util list.
51 */
52struct blame_origin *get_origin(struct commit *commit, const char *path)
53{
54 struct blame_origin *o, *l;
55
56 for (o = commit->util, l = NULL; o; l = o, o = o->next) {
57 if (!strcmp(o->path, path)) {
58 /* bump to front */
59 if (l) {
60 l->next = o->next;
61 o->next = commit->util;
62 commit->util = o;
63 }
64 return blame_origin_incref(o);
65 }
66 }
67 return make_origin(commit, path);
68}
072bf432
JS
69
70
71
72static void verify_working_tree_path(struct commit *work_tree, const char *path)
73{
74 struct commit_list *parents;
75 int pos;
76
77 for (parents = work_tree->parents; parents; parents = parents->next) {
78 const struct object_id *commit_oid = &parents->item->object.oid;
79 struct object_id blob_oid;
80 unsigned mode;
81
82 if (!get_tree_entry(commit_oid->hash, path, blob_oid.hash, &mode) &&
83 sha1_object_info(blob_oid.hash, NULL) == OBJ_BLOB)
84 return;
85 }
86
87 pos = cache_name_pos(path, strlen(path));
88 if (pos >= 0)
89 ; /* path is in the index */
90 else if (-1 - pos < active_nr &&
91 !strcmp(active_cache[-1 - pos]->name, path))
92 ; /* path is in the index, unmerged */
93 else
94 die("no such path '%s' in HEAD", path);
95}
96
97static struct commit_list **append_parent(struct commit_list **tail, const struct object_id *oid)
98{
99 struct commit *parent;
100
101 parent = lookup_commit_reference(oid->hash);
102 if (!parent)
103 die("no such commit %s", oid_to_hex(oid));
104 return &commit_list_insert(parent, tail)->next;
105}
106
107static void append_merge_parents(struct commit_list **tail)
108{
109 int merge_head;
110 struct strbuf line = STRBUF_INIT;
111
112 merge_head = open(git_path_merge_head(), O_RDONLY);
113 if (merge_head < 0) {
114 if (errno == ENOENT)
115 return;
116 die("cannot open '%s' for reading", git_path_merge_head());
117 }
118
119 while (!strbuf_getwholeline_fd(&line, merge_head, '\n')) {
120 struct object_id oid;
121 if (line.len < GIT_SHA1_HEXSZ || get_oid_hex(line.buf, &oid))
122 die("unknown line in '%s': %s", git_path_merge_head(), line.buf);
123 tail = append_parent(tail, &oid);
124 }
125 close(merge_head);
126 strbuf_release(&line);
127}
128
129/*
130 * This isn't as simple as passing sb->buf and sb->len, because we
131 * want to transfer ownership of the buffer to the commit (so we
132 * must use detach).
133 */
134static void set_commit_buffer_from_strbuf(struct commit *c, struct strbuf *sb)
135{
136 size_t len;
137 void *buf = strbuf_detach(sb, &len);
138 set_commit_buffer(c, buf, len);
139}
140
141/*
142 * Prepare a dummy commit that represents the work tree (or staged) item.
143 * Note that annotating work tree item never works in the reverse.
144 */
145struct commit *fake_working_tree_commit(struct diff_options *opt,
146 const char *path,
147 const char *contents_from)
148{
149 struct commit *commit;
150 struct blame_origin *origin;
151 struct commit_list **parent_tail, *parent;
152 struct object_id head_oid;
153 struct strbuf buf = STRBUF_INIT;
154 const char *ident;
155 time_t now;
156 int size, len;
157 struct cache_entry *ce;
158 unsigned mode;
159 struct strbuf msg = STRBUF_INIT;
160
161 read_cache();
162 time(&now);
163 commit = alloc_commit_node();
164 commit->object.parsed = 1;
165 commit->date = now;
166 parent_tail = &commit->parents;
167
168 if (!resolve_ref_unsafe("HEAD", RESOLVE_REF_READING, head_oid.hash, NULL))
169 die("no such ref: HEAD");
170
171 parent_tail = append_parent(parent_tail, &head_oid);
172 append_merge_parents(parent_tail);
173 verify_working_tree_path(commit, path);
174
175 origin = make_origin(commit, path);
176
177 ident = fmt_ident("Not Committed Yet", "not.committed.yet", NULL, 0);
178 strbuf_addstr(&msg, "tree 0000000000000000000000000000000000000000\n");
179 for (parent = commit->parents; parent; parent = parent->next)
180 strbuf_addf(&msg, "parent %s\n",
181 oid_to_hex(&parent->item->object.oid));
182 strbuf_addf(&msg,
183 "author %s\n"
184 "committer %s\n\n"
185 "Version of %s from %s\n",
186 ident, ident, path,
187 (!contents_from ? path :
188 (!strcmp(contents_from, "-") ? "standard input" : contents_from)));
189 set_commit_buffer_from_strbuf(commit, &msg);
190
191 if (!contents_from || strcmp("-", contents_from)) {
192 struct stat st;
193 const char *read_from;
194 char *buf_ptr;
195 unsigned long buf_len;
196
197 if (contents_from) {
198 if (stat(contents_from, &st) < 0)
199 die_errno("Cannot stat '%s'", contents_from);
200 read_from = contents_from;
201 }
202 else {
203 if (lstat(path, &st) < 0)
204 die_errno("Cannot lstat '%s'", path);
205 read_from = path;
206 }
207 mode = canon_mode(st.st_mode);
208
209 switch (st.st_mode & S_IFMT) {
210 case S_IFREG:
211 if (DIFF_OPT_TST(opt, ALLOW_TEXTCONV) &&
212 textconv_object(read_from, mode, &null_oid, 0, &buf_ptr, &buf_len))
213 strbuf_attach(&buf, buf_ptr, buf_len, buf_len + 1);
214 else if (strbuf_read_file(&buf, read_from, st.st_size) != st.st_size)
215 die_errno("cannot open or read '%s'", read_from);
216 break;
217 case S_IFLNK:
218 if (strbuf_readlink(&buf, read_from, st.st_size) < 0)
219 die_errno("cannot readlink '%s'", read_from);
220 break;
221 default:
222 die("unsupported file type %s", read_from);
223 }
224 }
225 else {
226 /* Reading from stdin */
227 mode = 0;
228 if (strbuf_read(&buf, 0, 0) < 0)
229 die_errno("failed to read from stdin");
230 }
231 convert_to_git(path, buf.buf, buf.len, &buf, 0);
232 origin->file.ptr = buf.buf;
233 origin->file.size = buf.len;
234 pretend_sha1_file(buf.buf, buf.len, OBJ_BLOB, origin->blob_oid.hash);
235
236 /*
237 * Read the current index, replace the path entry with
238 * origin->blob_sha1 without mucking with its mode or type
239 * bits; we are not going to write this index out -- we just
240 * want to run "diff-index --cached".
241 */
242 discard_cache();
243 read_cache();
244
245 len = strlen(path);
246 if (!mode) {
247 int pos = cache_name_pos(path, len);
248 if (0 <= pos)
249 mode = active_cache[pos]->ce_mode;
250 else
251 /* Let's not bother reading from HEAD tree */
252 mode = S_IFREG | 0644;
253 }
254 size = cache_entry_size(len);
255 ce = xcalloc(1, size);
256 oidcpy(&ce->oid, &origin->blob_oid);
257 memcpy(ce->name, path, len);
258 ce->ce_flags = create_ce_flags(0);
259 ce->ce_namelen = len;
260 ce->ce_mode = create_ce_mode(mode);
261 add_cache_entry(ce, ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE);
262
263 cache_tree_invalidate_path(&the_index, path);
264
265 return commit;
266}
b543bb1c
JS
267
268
269
270static int diff_hunks(mmfile_t *file_a, mmfile_t *file_b,
271 xdl_emit_hunk_consume_func_t hunk_func, void *cb_data, int xdl_opts)
272{
273 xpparam_t xpp = {0};
274 xdemitconf_t xecfg = {0};
275 xdemitcb_t ecb = {NULL};
276
277 xpp.flags = xdl_opts;
278 xecfg.hunk_func = hunk_func;
279 ecb.priv = cb_data;
280 return xdi_diff(file_a, file_b, &xpp, &xecfg, &ecb);
281}
282
283/*
284 * Given an origin, prepare mmfile_t structure to be used by the
285 * diff machinery
286 */
287static void fill_origin_blob(struct diff_options *opt,
288 struct blame_origin *o, mmfile_t *file, int *num_read_blob)
289{
290 if (!o->file.ptr) {
291 enum object_type type;
292 unsigned long file_size;
293
294 (*num_read_blob)++;
295 if (DIFF_OPT_TST(opt, ALLOW_TEXTCONV) &&
296 textconv_object(o->path, o->mode, &o->blob_oid, 1, &file->ptr, &file_size))
297 ;
298 else
299 file->ptr = read_sha1_file(o->blob_oid.hash, &type,
300 &file_size);
301 file->size = file_size;
302
303 if (!file->ptr)
304 die("Cannot read blob %s for path %s",
305 oid_to_hex(&o->blob_oid),
306 o->path);
307 o->file = *file;
308 }
309 else
310 *file = o->file;
311}
312
313static void drop_origin_blob(struct blame_origin *o)
314{
315 if (o->file.ptr) {
316 free(o->file.ptr);
317 o->file.ptr = NULL;
318 }
319}
320
321/*
322 * Any merge of blames happens on lists of blames that arrived via
323 * different parents in a single suspect. In this case, we want to
324 * sort according to the suspect line numbers as opposed to the final
325 * image line numbers. The function body is somewhat longish because
326 * it avoids unnecessary writes.
327 */
328
329static struct blame_entry *blame_merge(struct blame_entry *list1,
330 struct blame_entry *list2)
331{
332 struct blame_entry *p1 = list1, *p2 = list2,
333 **tail = &list1;
334
335 if (!p1)
336 return p2;
337 if (!p2)
338 return p1;
339
340 if (p1->s_lno <= p2->s_lno) {
341 do {
342 tail = &p1->next;
343 if ((p1 = *tail) == NULL) {
344 *tail = p2;
345 return list1;
346 }
347 } while (p1->s_lno <= p2->s_lno);
348 }
349 for (;;) {
350 *tail = p2;
351 do {
352 tail = &p2->next;
353 if ((p2 = *tail) == NULL) {
354 *tail = p1;
355 return list1;
356 }
357 } while (p1->s_lno > p2->s_lno);
358 *tail = p1;
359 do {
360 tail = &p1->next;
361 if ((p1 = *tail) == NULL) {
362 *tail = p2;
363 return list1;
364 }
365 } while (p1->s_lno <= p2->s_lno);
366 }
367}
368
369static void *get_next_blame(const void *p)
370{
371 return ((struct blame_entry *)p)->next;
372}
373
374static void set_next_blame(void *p1, void *p2)
375{
376 ((struct blame_entry *)p1)->next = p2;
377}
378
379/*
380 * Final image line numbers are all different, so we don't need a
381 * three-way comparison here.
382 */
383
384static int compare_blame_final(const void *p1, const void *p2)
385{
386 return ((struct blame_entry *)p1)->lno > ((struct blame_entry *)p2)->lno
387 ? 1 : -1;
388}
389
390static int compare_blame_suspect(const void *p1, const void *p2)
391{
392 const struct blame_entry *s1 = p1, *s2 = p2;
393 /*
394 * to allow for collating suspects, we sort according to the
395 * respective pointer value as the primary sorting criterion.
396 * The actual relation is pretty unimportant as long as it
397 * establishes a total order. Comparing as integers gives us
398 * that.
399 */
400 if (s1->suspect != s2->suspect)
401 return (intptr_t)s1->suspect > (intptr_t)s2->suspect ? 1 : -1;
402 if (s1->s_lno == s2->s_lno)
403 return 0;
404 return s1->s_lno > s2->s_lno ? 1 : -1;
405}
406
407void blame_sort_final(struct blame_scoreboard *sb)
408{
409 sb->ent = llist_mergesort(sb->ent, get_next_blame, set_next_blame,
410 compare_blame_final);
411}
412
413/*
414 * For debugging -- origin is refcounted, and this asserts that
415 * we do not underflow.
416 */
417static void sanity_check_refcnt(struct blame_scoreboard *sb)
418{
419 int baa = 0;
420 struct blame_entry *ent;
421
422 for (ent = sb->ent; ent; ent = ent->next) {
423 /* Nobody should have zero or negative refcnt */
424 if (ent->suspect->refcnt <= 0) {
425 fprintf(stderr, "%s in %s has negative refcnt %d\n",
426 ent->suspect->path,
427 oid_to_hex(&ent->suspect->commit->object.oid),
428 ent->suspect->refcnt);
429 baa = 1;
430 }
431 }
432 if (baa)
433 sb->on_sanity_fail(sb, baa);
434}
435
436/*
437 * If two blame entries that are next to each other came from
438 * contiguous lines in the same origin (i.e. <commit, path> pair),
439 * merge them together.
440 */
441void blame_coalesce(struct blame_scoreboard *sb)
442{
443 struct blame_entry *ent, *next;
444
445 for (ent = sb->ent; ent && (next = ent->next); ent = next) {
446 if (ent->suspect == next->suspect &&
447 ent->s_lno + ent->num_lines == next->s_lno) {
448 ent->num_lines += next->num_lines;
449 ent->next = next->next;
450 blame_origin_decref(next->suspect);
451 free(next);
452 ent->score = 0;
453 next = ent; /* again */
454 }
455 }
456
457 if (sb->debug) /* sanity */
458 sanity_check_refcnt(sb);
459}
460
461/*
462 * Merge the given sorted list of blames into a preexisting origin.
463 * If there were no previous blames to that commit, it is entered into
464 * the commit priority queue of the score board.
465 */
466
467static void queue_blames(struct blame_scoreboard *sb, struct blame_origin *porigin,
468 struct blame_entry *sorted)
469{
470 if (porigin->suspects)
471 porigin->suspects = blame_merge(porigin->suspects, sorted);
472 else {
473 struct blame_origin *o;
474 for (o = porigin->commit->util; o; o = o->next) {
475 if (o->suspects) {
476 porigin->suspects = sorted;
477 return;
478 }
479 }
480 porigin->suspects = sorted;
481 prio_queue_put(&sb->commits, porigin->commit);
482 }
483}
484
485/*
486 * We have an origin -- check if the same path exists in the
487 * parent and return an origin structure to represent it.
488 */
489static struct blame_origin *find_origin(struct commit *parent,
490 struct blame_origin *origin)
491{
492 struct blame_origin *porigin;
493 struct diff_options diff_opts;
494 const char *paths[2];
495
496 /* First check any existing origins */
497 for (porigin = parent->util; porigin; porigin = porigin->next)
498 if (!strcmp(porigin->path, origin->path)) {
499 /*
500 * The same path between origin and its parent
501 * without renaming -- the most common case.
502 */
503 return blame_origin_incref (porigin);
504 }
505
506 /* See if the origin->path is different between parent
507 * and origin first. Most of the time they are the
508 * same and diff-tree is fairly efficient about this.
509 */
510 diff_setup(&diff_opts);
511 DIFF_OPT_SET(&diff_opts, RECURSIVE);
512 diff_opts.detect_rename = 0;
513 diff_opts.output_format = DIFF_FORMAT_NO_OUTPUT;
514 paths[0] = origin->path;
515 paths[1] = NULL;
516
517 parse_pathspec(&diff_opts.pathspec,
518 PATHSPEC_ALL_MAGIC & ~PATHSPEC_LITERAL,
519 PATHSPEC_LITERAL_PATH, "", paths);
520 diff_setup_done(&diff_opts);
521
522 if (is_null_oid(&origin->commit->object.oid))
523 do_diff_cache(parent->tree->object.oid.hash, &diff_opts);
524 else
525 diff_tree_sha1(parent->tree->object.oid.hash,
526 origin->commit->tree->object.oid.hash,
527 "", &diff_opts);
528 diffcore_std(&diff_opts);
529
530 if (!diff_queued_diff.nr) {
531 /* The path is the same as parent */
532 porigin = get_origin(parent, origin->path);
533 oidcpy(&porigin->blob_oid, &origin->blob_oid);
534 porigin->mode = origin->mode;
535 } else {
536 /*
537 * Since origin->path is a pathspec, if the parent
538 * commit had it as a directory, we will see a whole
539 * bunch of deletion of files in the directory that we
540 * do not care about.
541 */
542 int i;
543 struct diff_filepair *p = NULL;
544 for (i = 0; i < diff_queued_diff.nr; i++) {
545 const char *name;
546 p = diff_queued_diff.queue[i];
547 name = p->one->path ? p->one->path : p->two->path;
548 if (!strcmp(name, origin->path))
549 break;
550 }
551 if (!p)
552 die("internal error in blame::find_origin");
553 switch (p->status) {
554 default:
555 die("internal error in blame::find_origin (%c)",
556 p->status);
557 case 'M':
558 porigin = get_origin(parent, origin->path);
559 oidcpy(&porigin->blob_oid, &p->one->oid);
560 porigin->mode = p->one->mode;
561 break;
562 case 'A':
563 case 'T':
564 /* Did not exist in parent, or type changed */
565 break;
566 }
567 }
568 diff_flush(&diff_opts);
569 clear_pathspec(&diff_opts.pathspec);
570 return porigin;
571}
572
573/*
574 * We have an origin -- find the path that corresponds to it in its
575 * parent and return an origin structure to represent it.
576 */
577static struct blame_origin *find_rename(struct commit *parent,
578 struct blame_origin *origin)
579{
580 struct blame_origin *porigin = NULL;
581 struct diff_options diff_opts;
582 int i;
583
584 diff_setup(&diff_opts);
585 DIFF_OPT_SET(&diff_opts, RECURSIVE);
586 diff_opts.detect_rename = DIFF_DETECT_RENAME;
587 diff_opts.output_format = DIFF_FORMAT_NO_OUTPUT;
588 diff_opts.single_follow = origin->path;
589 diff_setup_done(&diff_opts);
590
591 if (is_null_oid(&origin->commit->object.oid))
592 do_diff_cache(parent->tree->object.oid.hash, &diff_opts);
593 else
594 diff_tree_sha1(parent->tree->object.oid.hash,
595 origin->commit->tree->object.oid.hash,
596 "", &diff_opts);
597 diffcore_std(&diff_opts);
598
599 for (i = 0; i < diff_queued_diff.nr; i++) {
600 struct diff_filepair *p = diff_queued_diff.queue[i];
601 if ((p->status == 'R' || p->status == 'C') &&
602 !strcmp(p->two->path, origin->path)) {
603 porigin = get_origin(parent, p->one->path);
604 oidcpy(&porigin->blob_oid, &p->one->oid);
605 porigin->mode = p->one->mode;
606 break;
607 }
608 }
609 diff_flush(&diff_opts);
610 clear_pathspec(&diff_opts.pathspec);
611 return porigin;
612}
613
614/*
615 * Append a new blame entry to a given output queue.
616 */
617static void add_blame_entry(struct blame_entry ***queue,
618 const struct blame_entry *src)
619{
620 struct blame_entry *e = xmalloc(sizeof(*e));
621 memcpy(e, src, sizeof(*e));
622 blame_origin_incref(e->suspect);
623
624 e->next = **queue;
625 **queue = e;
626 *queue = &e->next;
627}
628
629/*
630 * src typically is on-stack; we want to copy the information in it to
631 * a malloced blame_entry that gets added to the given queue. The
632 * origin of dst loses a refcnt.
633 */
634static void dup_entry(struct blame_entry ***queue,
635 struct blame_entry *dst, struct blame_entry *src)
636{
637 blame_origin_incref(src->suspect);
638 blame_origin_decref(dst->suspect);
639 memcpy(dst, src, sizeof(*src));
640 dst->next = **queue;
641 **queue = dst;
642 *queue = &dst->next;
643}
644
645const char *blame_nth_line(struct blame_scoreboard *sb, long lno)
646{
647 return sb->final_buf + sb->lineno[lno];
648}
649
650/*
651 * It is known that lines between tlno to same came from parent, and e
652 * has an overlap with that range. it also is known that parent's
653 * line plno corresponds to e's line tlno.
654 *
655 * <---- e ----->
656 * <------>
657 * <------------>
658 * <------------>
659 * <------------------>
660 *
661 * Split e into potentially three parts; before this chunk, the chunk
662 * to be blamed for the parent, and after that portion.
663 */
664static void split_overlap(struct blame_entry *split,
665 struct blame_entry *e,
666 int tlno, int plno, int same,
667 struct blame_origin *parent)
668{
669 int chunk_end_lno;
670 memset(split, 0, sizeof(struct blame_entry [3]));
671
672 if (e->s_lno < tlno) {
673 /* there is a pre-chunk part not blamed on parent */
674 split[0].suspect = blame_origin_incref(e->suspect);
675 split[0].lno = e->lno;
676 split[0].s_lno = e->s_lno;
677 split[0].num_lines = tlno - e->s_lno;
678 split[1].lno = e->lno + tlno - e->s_lno;
679 split[1].s_lno = plno;
680 }
681 else {
682 split[1].lno = e->lno;
683 split[1].s_lno = plno + (e->s_lno - tlno);
684 }
685
686 if (same < e->s_lno + e->num_lines) {
687 /* there is a post-chunk part not blamed on parent */
688 split[2].suspect = blame_origin_incref(e->suspect);
689 split[2].lno = e->lno + (same - e->s_lno);
690 split[2].s_lno = e->s_lno + (same - e->s_lno);
691 split[2].num_lines = e->s_lno + e->num_lines - same;
692 chunk_end_lno = split[2].lno;
693 }
694 else
695 chunk_end_lno = e->lno + e->num_lines;
696 split[1].num_lines = chunk_end_lno - split[1].lno;
697
698 /*
699 * if it turns out there is nothing to blame the parent for,
700 * forget about the splitting. !split[1].suspect signals this.
701 */
702 if (split[1].num_lines < 1)
703 return;
704 split[1].suspect = blame_origin_incref(parent);
705}
706
707/*
708 * split_overlap() divided an existing blame e into up to three parts
709 * in split. Any assigned blame is moved to queue to
710 * reflect the split.
711 */
712static void split_blame(struct blame_entry ***blamed,
713 struct blame_entry ***unblamed,
714 struct blame_entry *split,
715 struct blame_entry *e)
716{
717 if (split[0].suspect && split[2].suspect) {
718 /* The first part (reuse storage for the existing entry e) */
719 dup_entry(unblamed, e, &split[0]);
720
721 /* The last part -- me */
722 add_blame_entry(unblamed, &split[2]);
723
724 /* ... and the middle part -- parent */
725 add_blame_entry(blamed, &split[1]);
726 }
727 else if (!split[0].suspect && !split[2].suspect)
728 /*
729 * The parent covers the entire area; reuse storage for
730 * e and replace it with the parent.
731 */
732 dup_entry(blamed, e, &split[1]);
733 else if (split[0].suspect) {
734 /* me and then parent */
735 dup_entry(unblamed, e, &split[0]);
736 add_blame_entry(blamed, &split[1]);
737 }
738 else {
739 /* parent and then me */
740 dup_entry(blamed, e, &split[1]);
741 add_blame_entry(unblamed, &split[2]);
742 }
743}
744
745/*
746 * After splitting the blame, the origins used by the
747 * on-stack blame_entry should lose one refcnt each.
748 */
749static void decref_split(struct blame_entry *split)
750{
751 int i;
752
753 for (i = 0; i < 3; i++)
754 blame_origin_decref(split[i].suspect);
755}
756
757/*
758 * reverse_blame reverses the list given in head, appending tail.
759 * That allows us to build lists in reverse order, then reverse them
760 * afterwards. This can be faster than building the list in proper
761 * order right away. The reason is that building in proper order
762 * requires writing a link in the _previous_ element, while building
763 * in reverse order just requires placing the list head into the
764 * _current_ element.
765 */
766
767static struct blame_entry *reverse_blame(struct blame_entry *head,
768 struct blame_entry *tail)
769{
770 while (head) {
771 struct blame_entry *next = head->next;
772 head->next = tail;
773 tail = head;
774 head = next;
775 }
776 return tail;
777}
778
779/*
780 * Process one hunk from the patch between the current suspect for
781 * blame_entry e and its parent. This first blames any unfinished
782 * entries before the chunk (which is where target and parent start
783 * differing) on the parent, and then splits blame entries at the
784 * start and at the end of the difference region. Since use of -M and
785 * -C options may lead to overlapping/duplicate source line number
786 * ranges, all we can rely on from sorting/merging is the order of the
787 * first suspect line number.
788 */
789static void blame_chunk(struct blame_entry ***dstq, struct blame_entry ***srcq,
790 int tlno, int offset, int same,
791 struct blame_origin *parent)
792{
793 struct blame_entry *e = **srcq;
794 struct blame_entry *samep = NULL, *diffp = NULL;
795
796 while (e && e->s_lno < tlno) {
797 struct blame_entry *next = e->next;
798 /*
799 * current record starts before differing portion. If
800 * it reaches into it, we need to split it up and
801 * examine the second part separately.
802 */
803 if (e->s_lno + e->num_lines > tlno) {
804 /* Move second half to a new record */
805 int len = tlno - e->s_lno;
806 struct blame_entry *n = xcalloc(1, sizeof (struct blame_entry));
807 n->suspect = e->suspect;
808 n->lno = e->lno + len;
809 n->s_lno = e->s_lno + len;
810 n->num_lines = e->num_lines - len;
811 e->num_lines = len;
812 e->score = 0;
813 /* Push new record to diffp */
814 n->next = diffp;
815 diffp = n;
816 } else
817 blame_origin_decref(e->suspect);
818 /* Pass blame for everything before the differing
819 * chunk to the parent */
820 e->suspect = blame_origin_incref(parent);
821 e->s_lno += offset;
822 e->next = samep;
823 samep = e;
824 e = next;
825 }
826 /*
827 * As we don't know how much of a common stretch after this
828 * diff will occur, the currently blamed parts are all that we
829 * can assign to the parent for now.
830 */
831
832 if (samep) {
833 **dstq = reverse_blame(samep, **dstq);
834 *dstq = &samep->next;
835 }
836 /*
837 * Prepend the split off portions: everything after e starts
838 * after the blameable portion.
839 */
840 e = reverse_blame(diffp, e);
841
842 /*
843 * Now retain records on the target while parts are different
844 * from the parent.
845 */
846 samep = NULL;
847 diffp = NULL;
848 while (e && e->s_lno < same) {
849 struct blame_entry *next = e->next;
850
851 /*
852 * If current record extends into sameness, need to split.
853 */
854 if (e->s_lno + e->num_lines > same) {
855 /*
856 * Move second half to a new record to be
857 * processed by later chunks
858 */
859 int len = same - e->s_lno;
860 struct blame_entry *n = xcalloc(1, sizeof (struct blame_entry));
861 n->suspect = blame_origin_incref(e->suspect);
862 n->lno = e->lno + len;
863 n->s_lno = e->s_lno + len;
864 n->num_lines = e->num_lines - len;
865 e->num_lines = len;
866 e->score = 0;
867 /* Push new record to samep */
868 n->next = samep;
869 samep = n;
870 }
871 e->next = diffp;
872 diffp = e;
873 e = next;
874 }
875 **srcq = reverse_blame(diffp, reverse_blame(samep, e));
876 /* Move across elements that are in the unblamable portion */
877 if (diffp)
878 *srcq = &diffp->next;
879}
880
881struct blame_chunk_cb_data {
882 struct blame_origin *parent;
883 long offset;
884 struct blame_entry **dstq;
885 struct blame_entry **srcq;
886};
887
888/* diff chunks are from parent to target */
889static int blame_chunk_cb(long start_a, long count_a,
890 long start_b, long count_b, void *data)
891{
892 struct blame_chunk_cb_data *d = data;
893 if (start_a - start_b != d->offset)
894 die("internal error in blame::blame_chunk_cb");
895 blame_chunk(&d->dstq, &d->srcq, start_b, start_a - start_b,
896 start_b + count_b, d->parent);
897 d->offset = start_a + count_a - (start_b + count_b);
898 return 0;
899}
900
901/*
902 * We are looking at the origin 'target' and aiming to pass blame
903 * for the lines it is suspected to its parent. Run diff to find
904 * which lines came from parent and pass blame for them.
905 */
906static void pass_blame_to_parent(struct blame_scoreboard *sb,
907 struct blame_origin *target,
908 struct blame_origin *parent)
909{
910 mmfile_t file_p, file_o;
911 struct blame_chunk_cb_data d;
912 struct blame_entry *newdest = NULL;
913
914 if (!target->suspects)
915 return; /* nothing remains for this target */
916
917 d.parent = parent;
918 d.offset = 0;
919 d.dstq = &newdest; d.srcq = &target->suspects;
920
921 fill_origin_blob(&sb->revs->diffopt, parent, &file_p, &sb->num_read_blob);
922 fill_origin_blob(&sb->revs->diffopt, target, &file_o, &sb->num_read_blob);
923 sb->num_get_patch++;
924
925 if (diff_hunks(&file_p, &file_o, blame_chunk_cb, &d, sb->xdl_opts))
926 die("unable to generate diff (%s -> %s)",
927 oid_to_hex(&parent->commit->object.oid),
928 oid_to_hex(&target->commit->object.oid));
929 /* The rest are the same as the parent */
930 blame_chunk(&d.dstq, &d.srcq, INT_MAX, d.offset, INT_MAX, parent);
931 *d.dstq = NULL;
932 queue_blames(sb, parent, newdest);
933
934 return;
935}
936
937/*
938 * The lines in blame_entry after splitting blames many times can become
939 * very small and trivial, and at some point it becomes pointless to
940 * blame the parents. E.g. "\t\t}\n\t}\n\n" appears everywhere in any
941 * ordinary C program, and it is not worth to say it was copied from
942 * totally unrelated file in the parent.
943 *
944 * Compute how trivial the lines in the blame_entry are.
945 */
946unsigned blame_entry_score(struct blame_scoreboard *sb, struct blame_entry *e)
947{
948 unsigned score;
949 const char *cp, *ep;
950
951 if (e->score)
952 return e->score;
953
954 score = 1;
955 cp = blame_nth_line(sb, e->lno);
956 ep = blame_nth_line(sb, e->lno + e->num_lines);
957 while (cp < ep) {
958 unsigned ch = *((unsigned char *)cp);
959 if (isalnum(ch))
960 score++;
961 cp++;
962 }
963 e->score = score;
964 return score;
965}
966
967/*
968 * best_so_far[] and this[] are both a split of an existing blame_entry
969 * that passes blame to the parent. Maintain best_so_far the best split
970 * so far, by comparing this and best_so_far and copying this into
971 * bst_so_far as needed.
972 */
973static void copy_split_if_better(struct blame_scoreboard *sb,
974 struct blame_entry *best_so_far,
975 struct blame_entry *this)
976{
977 int i;
978
979 if (!this[1].suspect)
980 return;
981 if (best_so_far[1].suspect) {
982 if (blame_entry_score(sb, &this[1]) < blame_entry_score(sb, &best_so_far[1]))
983 return;
984 }
985
986 for (i = 0; i < 3; i++)
987 blame_origin_incref(this[i].suspect);
988 decref_split(best_so_far);
989 memcpy(best_so_far, this, sizeof(struct blame_entry [3]));
990}
991
992/*
993 * We are looking at a part of the final image represented by
994 * ent (tlno and same are offset by ent->s_lno).
995 * tlno is where we are looking at in the final image.
996 * up to (but not including) same match preimage.
997 * plno is where we are looking at in the preimage.
998 *
999 * <-------------- final image ---------------------->
1000 * <------ent------>
1001 * ^tlno ^same
1002 * <---------preimage----->
1003 * ^plno
1004 *
1005 * All line numbers are 0-based.
1006 */
1007static void handle_split(struct blame_scoreboard *sb,
1008 struct blame_entry *ent,
1009 int tlno, int plno, int same,
1010 struct blame_origin *parent,
1011 struct blame_entry *split)
1012{
1013 if (ent->num_lines <= tlno)
1014 return;
1015 if (tlno < same) {
1016 struct blame_entry this[3];
1017 tlno += ent->s_lno;
1018 same += ent->s_lno;
1019 split_overlap(this, ent, tlno, plno, same, parent);
1020 copy_split_if_better(sb, split, this);
1021 decref_split(this);
1022 }
1023}
1024
1025struct handle_split_cb_data {
1026 struct blame_scoreboard *sb;
1027 struct blame_entry *ent;
1028 struct blame_origin *parent;
1029 struct blame_entry *split;
1030 long plno;
1031 long tlno;
1032};
1033
1034static int handle_split_cb(long start_a, long count_a,
1035 long start_b, long count_b, void *data)
1036{
1037 struct handle_split_cb_data *d = data;
1038 handle_split(d->sb, d->ent, d->tlno, d->plno, start_b, d->parent,
1039 d->split);
1040 d->plno = start_a + count_a;
1041 d->tlno = start_b + count_b;
1042 return 0;
1043}
1044
1045/*
1046 * Find the lines from parent that are the same as ent so that
1047 * we can pass blames to it. file_p has the blob contents for
1048 * the parent.
1049 */
1050static void find_copy_in_blob(struct blame_scoreboard *sb,
1051 struct blame_entry *ent,
1052 struct blame_origin *parent,
1053 struct blame_entry *split,
1054 mmfile_t *file_p)
1055{
1056 const char *cp;
1057 mmfile_t file_o;
1058 struct handle_split_cb_data d;
1059
1060 memset(&d, 0, sizeof(d));
1061 d.sb = sb; d.ent = ent; d.parent = parent; d.split = split;
1062 /*
1063 * Prepare mmfile that contains only the lines in ent.
1064 */
1065 cp = blame_nth_line(sb, ent->lno);
1066 file_o.ptr = (char *) cp;
1067 file_o.size = blame_nth_line(sb, ent->lno + ent->num_lines) - cp;
1068
1069 /*
1070 * file_o is a part of final image we are annotating.
1071 * file_p partially may match that image.
1072 */
1073 memset(split, 0, sizeof(struct blame_entry [3]));
1074 if (diff_hunks(file_p, &file_o, handle_split_cb, &d, sb->xdl_opts))
1075 die("unable to generate diff (%s)",
1076 oid_to_hex(&parent->commit->object.oid));
1077 /* remainder, if any, all match the preimage */
1078 handle_split(sb, ent, d.tlno, d.plno, ent->num_lines, parent, split);
1079}
1080
1081/* Move all blame entries from list *source that have a score smaller
1082 * than score_min to the front of list *small.
1083 * Returns a pointer to the link pointing to the old head of the small list.
1084 */
1085
1086static struct blame_entry **filter_small(struct blame_scoreboard *sb,
1087 struct blame_entry **small,
1088 struct blame_entry **source,
1089 unsigned score_min)
1090{
1091 struct blame_entry *p = *source;
1092 struct blame_entry *oldsmall = *small;
1093 while (p) {
1094 if (blame_entry_score(sb, p) <= score_min) {
1095 *small = p;
1096 small = &p->next;
1097 p = *small;
1098 } else {
1099 *source = p;
1100 source = &p->next;
1101 p = *source;
1102 }
1103 }
1104 *small = oldsmall;
1105 *source = NULL;
1106 return small;
1107}
1108
1109/*
1110 * See if lines currently target is suspected for can be attributed to
1111 * parent.
1112 */
1113static void find_move_in_parent(struct blame_scoreboard *sb,
1114 struct blame_entry ***blamed,
1115 struct blame_entry **toosmall,
1116 struct blame_origin *target,
1117 struct blame_origin *parent)
1118{
1119 struct blame_entry *e, split[3];
1120 struct blame_entry *unblamed = target->suspects;
1121 struct blame_entry *leftover = NULL;
1122 mmfile_t file_p;
1123
1124 if (!unblamed)
1125 return; /* nothing remains for this target */
1126
1127 fill_origin_blob(&sb->revs->diffopt, parent, &file_p, &sb->num_read_blob);
1128 if (!file_p.ptr)
1129 return;
1130
1131 /* At each iteration, unblamed has a NULL-terminated list of
1132 * entries that have not yet been tested for blame. leftover
1133 * contains the reversed list of entries that have been tested
1134 * without being assignable to the parent.
1135 */
1136 do {
1137 struct blame_entry **unblamedtail = &unblamed;
1138 struct blame_entry *next;
1139 for (e = unblamed; e; e = next) {
1140 next = e->next;
1141 find_copy_in_blob(sb, e, parent, split, &file_p);
1142 if (split[1].suspect &&
1143 sb->move_score < blame_entry_score(sb, &split[1])) {
1144 split_blame(blamed, &unblamedtail, split, e);
1145 } else {
1146 e->next = leftover;
1147 leftover = e;
1148 }
1149 decref_split(split);
1150 }
1151 *unblamedtail = NULL;
1152 toosmall = filter_small(sb, toosmall, &unblamed, sb->move_score);
1153 } while (unblamed);
1154 target->suspects = reverse_blame(leftover, NULL);
1155}
1156
1157struct blame_list {
1158 struct blame_entry *ent;
1159 struct blame_entry split[3];
1160};
1161
1162/*
1163 * Count the number of entries the target is suspected for,
1164 * and prepare a list of entry and the best split.
1165 */
1166static struct blame_list *setup_blame_list(struct blame_entry *unblamed,
1167 int *num_ents_p)
1168{
1169 struct blame_entry *e;
1170 int num_ents, i;
1171 struct blame_list *blame_list = NULL;
1172
1173 for (e = unblamed, num_ents = 0; e; e = e->next)
1174 num_ents++;
1175 if (num_ents) {
1176 blame_list = xcalloc(num_ents, sizeof(struct blame_list));
1177 for (e = unblamed, i = 0; e; e = e->next)
1178 blame_list[i++].ent = e;
1179 }
1180 *num_ents_p = num_ents;
1181 return blame_list;
1182}
1183
1184/*
1185 * For lines target is suspected for, see if we can find code movement
1186 * across file boundary from the parent commit. porigin is the path
1187 * in the parent we already tried.
1188 */
1189static void find_copy_in_parent(struct blame_scoreboard *sb,
1190 struct blame_entry ***blamed,
1191 struct blame_entry **toosmall,
1192 struct blame_origin *target,
1193 struct commit *parent,
1194 struct blame_origin *porigin,
1195 int opt)
1196{
1197 struct diff_options diff_opts;
1198 int i, j;
1199 struct blame_list *blame_list;
1200 int num_ents;
1201 struct blame_entry *unblamed = target->suspects;
1202 struct blame_entry *leftover = NULL;
1203
1204 if (!unblamed)
1205 return; /* nothing remains for this target */
1206
1207 diff_setup(&diff_opts);
1208 DIFF_OPT_SET(&diff_opts, RECURSIVE);
1209 diff_opts.output_format = DIFF_FORMAT_NO_OUTPUT;
1210
1211 diff_setup_done(&diff_opts);
1212
1213 /* Try "find copies harder" on new path if requested;
1214 * we do not want to use diffcore_rename() actually to
1215 * match things up; find_copies_harder is set only to
1216 * force diff_tree_sha1() to feed all filepairs to diff_queue,
1217 * and this code needs to be after diff_setup_done(), which
1218 * usually makes find-copies-harder imply copy detection.
1219 */
1220 if ((opt & PICKAXE_BLAME_COPY_HARDEST)
1221 || ((opt & PICKAXE_BLAME_COPY_HARDER)
1222 && (!porigin || strcmp(target->path, porigin->path))))
1223 DIFF_OPT_SET(&diff_opts, FIND_COPIES_HARDER);
1224
1225 if (is_null_oid(&target->commit->object.oid))
1226 do_diff_cache(parent->tree->object.oid.hash, &diff_opts);
1227 else
1228 diff_tree_sha1(parent->tree->object.oid.hash,
1229 target->commit->tree->object.oid.hash,
1230 "", &diff_opts);
1231
1232 if (!DIFF_OPT_TST(&diff_opts, FIND_COPIES_HARDER))
1233 diffcore_std(&diff_opts);
1234
1235 do {
1236 struct blame_entry **unblamedtail = &unblamed;
1237 blame_list = setup_blame_list(unblamed, &num_ents);
1238
1239 for (i = 0; i < diff_queued_diff.nr; i++) {
1240 struct diff_filepair *p = diff_queued_diff.queue[i];
1241 struct blame_origin *norigin;
1242 mmfile_t file_p;
1243 struct blame_entry this[3];
1244
1245 if (!DIFF_FILE_VALID(p->one))
1246 continue; /* does not exist in parent */
1247 if (S_ISGITLINK(p->one->mode))
1248 continue; /* ignore git links */
1249 if (porigin && !strcmp(p->one->path, porigin->path))
1250 /* find_move already dealt with this path */
1251 continue;
1252
1253 norigin = get_origin(parent, p->one->path);
1254 oidcpy(&norigin->blob_oid, &p->one->oid);
1255 norigin->mode = p->one->mode;
1256 fill_origin_blob(&sb->revs->diffopt, norigin, &file_p, &sb->num_read_blob);
1257 if (!file_p.ptr)
1258 continue;
1259
1260 for (j = 0; j < num_ents; j++) {
1261 find_copy_in_blob(sb, blame_list[j].ent,
1262 norigin, this, &file_p);
1263 copy_split_if_better(sb, blame_list[j].split,
1264 this);
1265 decref_split(this);
1266 }
1267 blame_origin_decref(norigin);
1268 }
1269
1270 for (j = 0; j < num_ents; j++) {
1271 struct blame_entry *split = blame_list[j].split;
1272 if (split[1].suspect &&
1273 sb->copy_score < blame_entry_score(sb, &split[1])) {
1274 split_blame(blamed, &unblamedtail, split,
1275 blame_list[j].ent);
1276 } else {
1277 blame_list[j].ent->next = leftover;
1278 leftover = blame_list[j].ent;
1279 }
1280 decref_split(split);
1281 }
1282 free(blame_list);
1283 *unblamedtail = NULL;
1284 toosmall = filter_small(sb, toosmall, &unblamed, sb->copy_score);
1285 } while (unblamed);
1286 target->suspects = reverse_blame(leftover, NULL);
1287 diff_flush(&diff_opts);
1288 clear_pathspec(&diff_opts.pathspec);
1289}
1290
1291/*
1292 * The blobs of origin and porigin exactly match, so everything
1293 * origin is suspected for can be blamed on the parent.
1294 */
1295static void pass_whole_blame(struct blame_scoreboard *sb,
1296 struct blame_origin *origin, struct blame_origin *porigin)
1297{
1298 struct blame_entry *e, *suspects;
1299
1300 if (!porigin->file.ptr && origin->file.ptr) {
1301 /* Steal its file */
1302 porigin->file = origin->file;
1303 origin->file.ptr = NULL;
1304 }
1305 suspects = origin->suspects;
1306 origin->suspects = NULL;
1307 for (e = suspects; e; e = e->next) {
1308 blame_origin_incref(porigin);
1309 blame_origin_decref(e->suspect);
1310 e->suspect = porigin;
1311 }
1312 queue_blames(sb, porigin, suspects);
1313}
1314
1315/*
1316 * We pass blame from the current commit to its parents. We keep saying
1317 * "parent" (and "porigin"), but what we mean is to find scapegoat to
1318 * exonerate ourselves.
1319 */
1320static struct commit_list *first_scapegoat(struct rev_info *revs, struct commit *commit,
1321 int reverse)
1322{
1323 if (!reverse) {
1324 if (revs->first_parent_only &&
1325 commit->parents &&
1326 commit->parents->next) {
1327 free_commit_list(commit->parents->next);
1328 commit->parents->next = NULL;
1329 }
1330 return commit->parents;
1331 }
1332 return lookup_decoration(&revs->children, &commit->object);
1333}
1334
1335static int num_scapegoats(struct rev_info *revs, struct commit *commit, int reverse)
1336{
1337 struct commit_list *l = first_scapegoat(revs, commit, reverse);
1338 return commit_list_count(l);
1339}
1340
1341/* Distribute collected unsorted blames to the respected sorted lists
1342 * in the various origins.
1343 */
1344static void distribute_blame(struct blame_scoreboard *sb, struct blame_entry *blamed)
1345{
1346 blamed = llist_mergesort(blamed, get_next_blame, set_next_blame,
1347 compare_blame_suspect);
1348 while (blamed)
1349 {
1350 struct blame_origin *porigin = blamed->suspect;
1351 struct blame_entry *suspects = NULL;
1352 do {
1353 struct blame_entry *next = blamed->next;
1354 blamed->next = suspects;
1355 suspects = blamed;
1356 blamed = next;
1357 } while (blamed && blamed->suspect == porigin);
1358 suspects = reverse_blame(suspects, NULL);
1359 queue_blames(sb, porigin, suspects);
1360 }
1361}
1362
1363#define MAXSG 16
1364
1365static void pass_blame(struct blame_scoreboard *sb, struct blame_origin *origin, int opt)
1366{
1367 struct rev_info *revs = sb->revs;
1368 int i, pass, num_sg;
1369 struct commit *commit = origin->commit;
1370 struct commit_list *sg;
1371 struct blame_origin *sg_buf[MAXSG];
1372 struct blame_origin *porigin, **sg_origin = sg_buf;
1373 struct blame_entry *toosmall = NULL;
1374 struct blame_entry *blames, **blametail = &blames;
1375
1376 num_sg = num_scapegoats(revs, commit, sb->reverse);
1377 if (!num_sg)
1378 goto finish;
1379 else if (num_sg < ARRAY_SIZE(sg_buf))
1380 memset(sg_buf, 0, sizeof(sg_buf));
1381 else
1382 sg_origin = xcalloc(num_sg, sizeof(*sg_origin));
1383
1384 /*
1385 * The first pass looks for unrenamed path to optimize for
1386 * common cases, then we look for renames in the second pass.
1387 */
1388 for (pass = 0; pass < 2 - sb->no_whole_file_rename; pass++) {
1389 struct blame_origin *(*find)(struct commit *, struct blame_origin *);
1390 find = pass ? find_rename : find_origin;
1391
1392 for (i = 0, sg = first_scapegoat(revs, commit, sb->reverse);
1393 i < num_sg && sg;
1394 sg = sg->next, i++) {
1395 struct commit *p = sg->item;
1396 int j, same;
1397
1398 if (sg_origin[i])
1399 continue;
1400 if (parse_commit(p))
1401 continue;
1402 porigin = find(p, origin);
1403 if (!porigin)
1404 continue;
1405 if (!oidcmp(&porigin->blob_oid, &origin->blob_oid)) {
1406 pass_whole_blame(sb, origin, porigin);
1407 blame_origin_decref(porigin);
1408 goto finish;
1409 }
1410 for (j = same = 0; j < i; j++)
1411 if (sg_origin[j] &&
1412 !oidcmp(&sg_origin[j]->blob_oid, &porigin->blob_oid)) {
1413 same = 1;
1414 break;
1415 }
1416 if (!same)
1417 sg_origin[i] = porigin;
1418 else
1419 blame_origin_decref(porigin);
1420 }
1421 }
1422
1423 sb->num_commits++;
1424 for (i = 0, sg = first_scapegoat(revs, commit, sb->reverse);
1425 i < num_sg && sg;
1426 sg = sg->next, i++) {
1427 struct blame_origin *porigin = sg_origin[i];
1428 if (!porigin)
1429 continue;
1430 if (!origin->previous) {
1431 blame_origin_incref(porigin);
1432 origin->previous = porigin;
1433 }
1434 pass_blame_to_parent(sb, origin, porigin);
1435 if (!origin->suspects)
1436 goto finish;
1437 }
1438
1439 /*
1440 * Optionally find moves in parents' files.
1441 */
1442 if (opt & PICKAXE_BLAME_MOVE) {
1443 filter_small(sb, &toosmall, &origin->suspects, sb->move_score);
1444 if (origin->suspects) {
1445 for (i = 0, sg = first_scapegoat(revs, commit, sb->reverse);
1446 i < num_sg && sg;
1447 sg = sg->next, i++) {
1448 struct blame_origin *porigin = sg_origin[i];
1449 if (!porigin)
1450 continue;
1451 find_move_in_parent(sb, &blametail, &toosmall, origin, porigin);
1452 if (!origin->suspects)
1453 break;
1454 }
1455 }
1456 }
1457
1458 /*
1459 * Optionally find copies from parents' files.
1460 */
1461 if (opt & PICKAXE_BLAME_COPY) {
1462 if (sb->copy_score > sb->move_score)
1463 filter_small(sb, &toosmall, &origin->suspects, sb->copy_score);
1464 else if (sb->copy_score < sb->move_score) {
1465 origin->suspects = blame_merge(origin->suspects, toosmall);
1466 toosmall = NULL;
1467 filter_small(sb, &toosmall, &origin->suspects, sb->copy_score);
1468 }
1469 if (!origin->suspects)
1470 goto finish;
1471
1472 for (i = 0, sg = first_scapegoat(revs, commit, sb->reverse);
1473 i < num_sg && sg;
1474 sg = sg->next, i++) {
1475 struct blame_origin *porigin = sg_origin[i];
1476 find_copy_in_parent(sb, &blametail, &toosmall,
1477 origin, sg->item, porigin, opt);
1478 if (!origin->suspects)
1479 goto finish;
1480 }
1481 }
1482
1483finish:
1484 *blametail = NULL;
1485 distribute_blame(sb, blames);
1486 /*
1487 * prepend toosmall to origin->suspects
1488 *
1489 * There is no point in sorting: this ends up on a big
1490 * unsorted list in the caller anyway.
1491 */
1492 if (toosmall) {
1493 struct blame_entry **tail = &toosmall;
1494 while (*tail)
1495 tail = &(*tail)->next;
1496 *tail = origin->suspects;
1497 origin->suspects = toosmall;
1498 }
1499 for (i = 0; i < num_sg; i++) {
1500 if (sg_origin[i]) {
1501 drop_origin_blob(sg_origin[i]);
1502 blame_origin_decref(sg_origin[i]);
1503 }
1504 }
1505 drop_origin_blob(origin);
1506 if (sg_buf != sg_origin)
1507 free(sg_origin);
1508}
1509
1510/*
1511 * The main loop -- while we have blobs with lines whose true origin
1512 * is still unknown, pick one blob, and allow its lines to pass blames
1513 * to its parents. */
1514void assign_blame(struct blame_scoreboard *sb, int opt)
1515{
1516 struct rev_info *revs = sb->revs;
1517 struct commit *commit = prio_queue_get(&sb->commits);
1518
1519 while (commit) {
1520 struct blame_entry *ent;
1521 struct blame_origin *suspect = commit->util;
1522
1523 /* find one suspect to break down */
1524 while (suspect && !suspect->suspects)
1525 suspect = suspect->next;
1526
1527 if (!suspect) {
1528 commit = prio_queue_get(&sb->commits);
1529 continue;
1530 }
1531
1532 assert(commit == suspect->commit);
1533
1534 /*
1535 * We will use this suspect later in the loop,
1536 * so hold onto it in the meantime.
1537 */
1538 blame_origin_incref(suspect);
1539 parse_commit(commit);
1540 if (sb->reverse ||
1541 (!(commit->object.flags & UNINTERESTING) &&
1542 !(revs->max_age != -1 && commit->date < revs->max_age)))
1543 pass_blame(sb, suspect, opt);
1544 else {
1545 commit->object.flags |= UNINTERESTING;
1546 if (commit->object.parsed)
1547 mark_parents_uninteresting(commit);
1548 }
1549 /* treat root commit as boundary */
1550 if (!commit->parents && !sb->show_root)
1551 commit->object.flags |= UNINTERESTING;
1552
1553 /* Take responsibility for the remaining entries */
1554 ent = suspect->suspects;
1555 if (ent) {
1556 suspect->guilty = 1;
1557 for (;;) {
1558 struct blame_entry *next = ent->next;
1559 if (sb->found_guilty_entry)
1560 sb->found_guilty_entry(ent, sb->found_guilty_entry_data);
1561 if (next) {
1562 ent = next;
1563 continue;
1564 }
1565 ent->next = sb->ent;
1566 sb->ent = suspect->suspects;
1567 suspect->suspects = NULL;
1568 break;
1569 }
1570 }
1571 blame_origin_decref(suspect);
1572
1573 if (sb->debug) /* sanity */
1574 sanity_check_refcnt(sb);
1575 }
1576}