Merge branch 'ag/userdiff-go-funcname'
[git/git.git] / blame.c
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
10 void blame_origin_decref(struct blame_origin *o)
11 {
12 if (o && --o->refcnt <= 0) {
13 struct blame_origin *p, *l = NULL;
14 if (o->previous)
15 blame_origin_decref(o->previous);
16 free(o->file.ptr);
17 /* Should be present exactly once in commit chain */
18 for (p = o->commit->util; p; l = p, p = p->next) {
19 if (p == o) {
20 if (l)
21 l->next = p->next;
22 else
23 o->commit->util = p->next;
24 free(o);
25 return;
26 }
27 }
28 die("internal error in blame_origin_decref");
29 }
30 }
31
32 /*
33 * Given a commit and a path in it, create a new origin structure.
34 * The callers that add blame to the scoreboard should use
35 * get_origin() to obtain shared, refcounted copy instead of calling
36 * this function directly.
37 */
38 static struct blame_origin *make_origin(struct commit *commit, const char *path)
39 {
40 struct blame_origin *o;
41 FLEX_ALLOC_STR(o, path, path);
42 o->commit = commit;
43 o->refcnt = 1;
44 o->next = commit->util;
45 commit->util = o;
46 return o;
47 }
48
49 /*
50 * Locate an existing origin or create a new one.
51 * This moves the origin to front position in the commit util list.
52 */
53 static struct blame_origin *get_origin(struct commit *commit, const char *path)
54 {
55 struct blame_origin *o, *l;
56
57 for (o = commit->util, l = NULL; o; l = o, o = o->next) {
58 if (!strcmp(o->path, path)) {
59 /* bump to front */
60 if (l) {
61 l->next = o->next;
62 o->next = commit->util;
63 commit->util = o;
64 }
65 return blame_origin_incref(o);
66 }
67 }
68 return make_origin(commit, path);
69 }
70
71
72
73 static void verify_working_tree_path(struct commit *work_tree, const char *path)
74 {
75 struct commit_list *parents;
76 int pos;
77
78 for (parents = work_tree->parents; parents; parents = parents->next) {
79 const struct object_id *commit_oid = &parents->item->object.oid;
80 struct object_id blob_oid;
81 unsigned mode;
82
83 if (!get_tree_entry(commit_oid->hash, path, blob_oid.hash, &mode) &&
84 sha1_object_info(blob_oid.hash, NULL) == OBJ_BLOB)
85 return;
86 }
87
88 pos = cache_name_pos(path, strlen(path));
89 if (pos >= 0)
90 ; /* path is in the index */
91 else if (-1 - pos < active_nr &&
92 !strcmp(active_cache[-1 - pos]->name, path))
93 ; /* path is in the index, unmerged */
94 else
95 die("no such path '%s' in HEAD", path);
96 }
97
98 static struct commit_list **append_parent(struct commit_list **tail, const struct object_id *oid)
99 {
100 struct commit *parent;
101
102 parent = lookup_commit_reference(oid);
103 if (!parent)
104 die("no such commit %s", oid_to_hex(oid));
105 return &commit_list_insert(parent, tail)->next;
106 }
107
108 static void append_merge_parents(struct commit_list **tail)
109 {
110 int merge_head;
111 struct strbuf line = STRBUF_INIT;
112
113 merge_head = open(git_path_merge_head(), O_RDONLY);
114 if (merge_head < 0) {
115 if (errno == ENOENT)
116 return;
117 die("cannot open '%s' for reading", git_path_merge_head());
118 }
119
120 while (!strbuf_getwholeline_fd(&line, merge_head, '\n')) {
121 struct object_id oid;
122 if (line.len < GIT_SHA1_HEXSZ || get_oid_hex(line.buf, &oid))
123 die("unknown line in '%s': %s", git_path_merge_head(), line.buf);
124 tail = append_parent(tail, &oid);
125 }
126 close(merge_head);
127 strbuf_release(&line);
128 }
129
130 /*
131 * This isn't as simple as passing sb->buf and sb->len, because we
132 * want to transfer ownership of the buffer to the commit (so we
133 * must use detach).
134 */
135 static void set_commit_buffer_from_strbuf(struct commit *c, struct strbuf *sb)
136 {
137 size_t len;
138 void *buf = strbuf_detach(sb, &len);
139 set_commit_buffer(c, buf, len);
140 }
141
142 /*
143 * Prepare a dummy commit that represents the work tree (or staged) item.
144 * Note that annotating work tree item never works in the reverse.
145 */
146 static struct commit *fake_working_tree_commit(struct diff_options *opt,
147 const char *path,
148 const char *contents_from)
149 {
150 struct commit *commit;
151 struct blame_origin *origin;
152 struct commit_list **parent_tail, *parent;
153 struct object_id head_oid;
154 struct strbuf buf = STRBUF_INIT;
155 const char *ident;
156 time_t now;
157 int size, len;
158 struct cache_entry *ce;
159 unsigned mode;
160 struct strbuf msg = STRBUF_INIT;
161
162 read_cache();
163 time(&now);
164 commit = alloc_commit_node();
165 commit->object.parsed = 1;
166 commit->date = now;
167 parent_tail = &commit->parents;
168
169 if (!resolve_ref_unsafe("HEAD", RESOLVE_REF_READING, &head_oid, NULL))
170 die("no such ref: HEAD");
171
172 parent_tail = append_parent(parent_tail, &head_oid);
173 append_merge_parents(parent_tail);
174 verify_working_tree_path(commit, path);
175
176 origin = make_origin(commit, path);
177
178 ident = fmt_ident("Not Committed Yet", "not.committed.yet", NULL, 0);
179 strbuf_addstr(&msg, "tree 0000000000000000000000000000000000000000\n");
180 for (parent = commit->parents; parent; parent = parent->next)
181 strbuf_addf(&msg, "parent %s\n",
182 oid_to_hex(&parent->item->object.oid));
183 strbuf_addf(&msg,
184 "author %s\n"
185 "committer %s\n\n"
186 "Version of %s from %s\n",
187 ident, ident, path,
188 (!contents_from ? path :
189 (!strcmp(contents_from, "-") ? "standard input" : contents_from)));
190 set_commit_buffer_from_strbuf(commit, &msg);
191
192 if (!contents_from || strcmp("-", contents_from)) {
193 struct stat st;
194 const char *read_from;
195 char *buf_ptr;
196 unsigned long buf_len;
197
198 if (contents_from) {
199 if (stat(contents_from, &st) < 0)
200 die_errno("Cannot stat '%s'", contents_from);
201 read_from = contents_from;
202 }
203 else {
204 if (lstat(path, &st) < 0)
205 die_errno("Cannot lstat '%s'", path);
206 read_from = path;
207 }
208 mode = canon_mode(st.st_mode);
209
210 switch (st.st_mode & S_IFMT) {
211 case S_IFREG:
212 if (opt->flags.allow_textconv &&
213 textconv_object(read_from, mode, &null_oid, 0, &buf_ptr, &buf_len))
214 strbuf_attach(&buf, buf_ptr, buf_len, buf_len + 1);
215 else if (strbuf_read_file(&buf, read_from, st.st_size) != st.st_size)
216 die_errno("cannot open or read '%s'", read_from);
217 break;
218 case S_IFLNK:
219 if (strbuf_readlink(&buf, read_from, st.st_size) < 0)
220 die_errno("cannot readlink '%s'", read_from);
221 break;
222 default:
223 die("unsupported file type %s", read_from);
224 }
225 }
226 else {
227 /* Reading from stdin */
228 mode = 0;
229 if (strbuf_read(&buf, 0, 0) < 0)
230 die_errno("failed to read from stdin");
231 }
232 convert_to_git(&the_index, path, buf.buf, buf.len, &buf, 0);
233 origin->file.ptr = buf.buf;
234 origin->file.size = buf.len;
235 pretend_object_file(buf.buf, buf.len, OBJ_BLOB, &origin->blob_oid);
236
237 /*
238 * Read the current index, replace the path entry with
239 * origin->blob_sha1 without mucking with its mode or type
240 * bits; we are not going to write this index out -- we just
241 * want to run "diff-index --cached".
242 */
243 discard_cache();
244 read_cache();
245
246 len = strlen(path);
247 if (!mode) {
248 int pos = cache_name_pos(path, len);
249 if (0 <= pos)
250 mode = active_cache[pos]->ce_mode;
251 else
252 /* Let's not bother reading from HEAD tree */
253 mode = S_IFREG | 0644;
254 }
255 size = cache_entry_size(len);
256 ce = xcalloc(1, size);
257 oidcpy(&ce->oid, &origin->blob_oid);
258 memcpy(ce->name, path, len);
259 ce->ce_flags = create_ce_flags(0);
260 ce->ce_namelen = len;
261 ce->ce_mode = create_ce_mode(mode);
262 add_cache_entry(ce, ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE);
263
264 cache_tree_invalidate_path(&the_index, path);
265
266 return commit;
267 }
268
269
270
271 static int diff_hunks(mmfile_t *file_a, mmfile_t *file_b,
272 xdl_emit_hunk_consume_func_t hunk_func, void *cb_data, int xdl_opts)
273 {
274 xpparam_t xpp = {0};
275 xdemitconf_t xecfg = {0};
276 xdemitcb_t ecb = {NULL};
277
278 xpp.flags = xdl_opts;
279 xecfg.hunk_func = hunk_func;
280 ecb.priv = cb_data;
281 return xdi_diff(file_a, file_b, &xpp, &xecfg, &ecb);
282 }
283
284 /*
285 * Given an origin, prepare mmfile_t structure to be used by the
286 * diff machinery
287 */
288 static void fill_origin_blob(struct diff_options *opt,
289 struct blame_origin *o, mmfile_t *file, int *num_read_blob)
290 {
291 if (!o->file.ptr) {
292 enum object_type type;
293 unsigned long file_size;
294
295 (*num_read_blob)++;
296 if (opt->flags.allow_textconv &&
297 textconv_object(o->path, o->mode, &o->blob_oid, 1, &file->ptr, &file_size))
298 ;
299 else
300 file->ptr = read_sha1_file(o->blob_oid.hash, &type,
301 &file_size);
302 file->size = file_size;
303
304 if (!file->ptr)
305 die("Cannot read blob %s for path %s",
306 oid_to_hex(&o->blob_oid),
307 o->path);
308 o->file = *file;
309 }
310 else
311 *file = o->file;
312 }
313
314 static void drop_origin_blob(struct blame_origin *o)
315 {
316 if (o->file.ptr) {
317 FREE_AND_NULL(o->file.ptr);
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
329 static 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
369 static void *get_next_blame(const void *p)
370 {
371 return ((struct blame_entry *)p)->next;
372 }
373
374 static 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
384 static 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
390 static 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
407 void 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 static int compare_commits_by_reverse_commit_date(const void *a,
414 const void *b,
415 void *c)
416 {
417 return -compare_commits_by_commit_date(a, b, c);
418 }
419
420 /*
421 * For debugging -- origin is refcounted, and this asserts that
422 * we do not underflow.
423 */
424 static void sanity_check_refcnt(struct blame_scoreboard *sb)
425 {
426 int baa = 0;
427 struct blame_entry *ent;
428
429 for (ent = sb->ent; ent; ent = ent->next) {
430 /* Nobody should have zero or negative refcnt */
431 if (ent->suspect->refcnt <= 0) {
432 fprintf(stderr, "%s in %s has negative refcnt %d\n",
433 ent->suspect->path,
434 oid_to_hex(&ent->suspect->commit->object.oid),
435 ent->suspect->refcnt);
436 baa = 1;
437 }
438 }
439 if (baa)
440 sb->on_sanity_fail(sb, baa);
441 }
442
443 /*
444 * If two blame entries that are next to each other came from
445 * contiguous lines in the same origin (i.e. <commit, path> pair),
446 * merge them together.
447 */
448 void blame_coalesce(struct blame_scoreboard *sb)
449 {
450 struct blame_entry *ent, *next;
451
452 for (ent = sb->ent; ent && (next = ent->next); ent = next) {
453 if (ent->suspect == next->suspect &&
454 ent->s_lno + ent->num_lines == next->s_lno) {
455 ent->num_lines += next->num_lines;
456 ent->next = next->next;
457 blame_origin_decref(next->suspect);
458 free(next);
459 ent->score = 0;
460 next = ent; /* again */
461 }
462 }
463
464 if (sb->debug) /* sanity */
465 sanity_check_refcnt(sb);
466 }
467
468 /*
469 * Merge the given sorted list of blames into a preexisting origin.
470 * If there were no previous blames to that commit, it is entered into
471 * the commit priority queue of the score board.
472 */
473
474 static void queue_blames(struct blame_scoreboard *sb, struct blame_origin *porigin,
475 struct blame_entry *sorted)
476 {
477 if (porigin->suspects)
478 porigin->suspects = blame_merge(porigin->suspects, sorted);
479 else {
480 struct blame_origin *o;
481 for (o = porigin->commit->util; o; o = o->next) {
482 if (o->suspects) {
483 porigin->suspects = sorted;
484 return;
485 }
486 }
487 porigin->suspects = sorted;
488 prio_queue_put(&sb->commits, porigin->commit);
489 }
490 }
491
492 /*
493 * Fill the blob_sha1 field of an origin if it hasn't, so that later
494 * call to fill_origin_blob() can use it to locate the data. blob_sha1
495 * for an origin is also used to pass the blame for the entire file to
496 * the parent to detect the case where a child's blob is identical to
497 * that of its parent's.
498 *
499 * This also fills origin->mode for corresponding tree path.
500 */
501 static int fill_blob_sha1_and_mode(struct blame_origin *origin)
502 {
503 if (!is_null_oid(&origin->blob_oid))
504 return 0;
505 if (get_tree_entry(origin->commit->object.oid.hash,
506 origin->path,
507 origin->blob_oid.hash, &origin->mode))
508 goto error_out;
509 if (sha1_object_info(origin->blob_oid.hash, NULL) != OBJ_BLOB)
510 goto error_out;
511 return 0;
512 error_out:
513 oidclr(&origin->blob_oid);
514 origin->mode = S_IFINVALID;
515 return -1;
516 }
517
518 /*
519 * We have an origin -- check if the same path exists in the
520 * parent and return an origin structure to represent it.
521 */
522 static struct blame_origin *find_origin(struct commit *parent,
523 struct blame_origin *origin)
524 {
525 struct blame_origin *porigin;
526 struct diff_options diff_opts;
527 const char *paths[2];
528
529 /* First check any existing origins */
530 for (porigin = parent->util; porigin; porigin = porigin->next)
531 if (!strcmp(porigin->path, origin->path)) {
532 /*
533 * The same path between origin and its parent
534 * without renaming -- the most common case.
535 */
536 return blame_origin_incref (porigin);
537 }
538
539 /* See if the origin->path is different between parent
540 * and origin first. Most of the time they are the
541 * same and diff-tree is fairly efficient about this.
542 */
543 diff_setup(&diff_opts);
544 diff_opts.flags.recursive = 1;
545 diff_opts.detect_rename = 0;
546 diff_opts.output_format = DIFF_FORMAT_NO_OUTPUT;
547 paths[0] = origin->path;
548 paths[1] = NULL;
549
550 parse_pathspec(&diff_opts.pathspec,
551 PATHSPEC_ALL_MAGIC & ~PATHSPEC_LITERAL,
552 PATHSPEC_LITERAL_PATH, "", paths);
553 diff_setup_done(&diff_opts);
554
555 if (is_null_oid(&origin->commit->object.oid))
556 do_diff_cache(&parent->tree->object.oid, &diff_opts);
557 else
558 diff_tree_oid(&parent->tree->object.oid,
559 &origin->commit->tree->object.oid,
560 "", &diff_opts);
561 diffcore_std(&diff_opts);
562
563 if (!diff_queued_diff.nr) {
564 /* The path is the same as parent */
565 porigin = get_origin(parent, origin->path);
566 oidcpy(&porigin->blob_oid, &origin->blob_oid);
567 porigin->mode = origin->mode;
568 } else {
569 /*
570 * Since origin->path is a pathspec, if the parent
571 * commit had it as a directory, we will see a whole
572 * bunch of deletion of files in the directory that we
573 * do not care about.
574 */
575 int i;
576 struct diff_filepair *p = NULL;
577 for (i = 0; i < diff_queued_diff.nr; i++) {
578 const char *name;
579 p = diff_queued_diff.queue[i];
580 name = p->one->path ? p->one->path : p->two->path;
581 if (!strcmp(name, origin->path))
582 break;
583 }
584 if (!p)
585 die("internal error in blame::find_origin");
586 switch (p->status) {
587 default:
588 die("internal error in blame::find_origin (%c)",
589 p->status);
590 case 'M':
591 porigin = get_origin(parent, origin->path);
592 oidcpy(&porigin->blob_oid, &p->one->oid);
593 porigin->mode = p->one->mode;
594 break;
595 case 'A':
596 case 'T':
597 /* Did not exist in parent, or type changed */
598 break;
599 }
600 }
601 diff_flush(&diff_opts);
602 clear_pathspec(&diff_opts.pathspec);
603 return porigin;
604 }
605
606 /*
607 * We have an origin -- find the path that corresponds to it in its
608 * parent and return an origin structure to represent it.
609 */
610 static struct blame_origin *find_rename(struct commit *parent,
611 struct blame_origin *origin)
612 {
613 struct blame_origin *porigin = NULL;
614 struct diff_options diff_opts;
615 int i;
616
617 diff_setup(&diff_opts);
618 diff_opts.flags.recursive = 1;
619 diff_opts.detect_rename = DIFF_DETECT_RENAME;
620 diff_opts.output_format = DIFF_FORMAT_NO_OUTPUT;
621 diff_opts.single_follow = origin->path;
622 diff_setup_done(&diff_opts);
623
624 if (is_null_oid(&origin->commit->object.oid))
625 do_diff_cache(&parent->tree->object.oid, &diff_opts);
626 else
627 diff_tree_oid(&parent->tree->object.oid,
628 &origin->commit->tree->object.oid,
629 "", &diff_opts);
630 diffcore_std(&diff_opts);
631
632 for (i = 0; i < diff_queued_diff.nr; i++) {
633 struct diff_filepair *p = diff_queued_diff.queue[i];
634 if ((p->status == 'R' || p->status == 'C') &&
635 !strcmp(p->two->path, origin->path)) {
636 porigin = get_origin(parent, p->one->path);
637 oidcpy(&porigin->blob_oid, &p->one->oid);
638 porigin->mode = p->one->mode;
639 break;
640 }
641 }
642 diff_flush(&diff_opts);
643 clear_pathspec(&diff_opts.pathspec);
644 return porigin;
645 }
646
647 /*
648 * Append a new blame entry to a given output queue.
649 */
650 static void add_blame_entry(struct blame_entry ***queue,
651 const struct blame_entry *src)
652 {
653 struct blame_entry *e = xmalloc(sizeof(*e));
654 memcpy(e, src, sizeof(*e));
655 blame_origin_incref(e->suspect);
656
657 e->next = **queue;
658 **queue = e;
659 *queue = &e->next;
660 }
661
662 /*
663 * src typically is on-stack; we want to copy the information in it to
664 * a malloced blame_entry that gets added to the given queue. The
665 * origin of dst loses a refcnt.
666 */
667 static void dup_entry(struct blame_entry ***queue,
668 struct blame_entry *dst, struct blame_entry *src)
669 {
670 blame_origin_incref(src->suspect);
671 blame_origin_decref(dst->suspect);
672 memcpy(dst, src, sizeof(*src));
673 dst->next = **queue;
674 **queue = dst;
675 *queue = &dst->next;
676 }
677
678 const char *blame_nth_line(struct blame_scoreboard *sb, long lno)
679 {
680 return sb->final_buf + sb->lineno[lno];
681 }
682
683 /*
684 * It is known that lines between tlno to same came from parent, and e
685 * has an overlap with that range. it also is known that parent's
686 * line plno corresponds to e's line tlno.
687 *
688 * <---- e ----->
689 * <------>
690 * <------------>
691 * <------------>
692 * <------------------>
693 *
694 * Split e into potentially three parts; before this chunk, the chunk
695 * to be blamed for the parent, and after that portion.
696 */
697 static void split_overlap(struct blame_entry *split,
698 struct blame_entry *e,
699 int tlno, int plno, int same,
700 struct blame_origin *parent)
701 {
702 int chunk_end_lno;
703 memset(split, 0, sizeof(struct blame_entry [3]));
704
705 if (e->s_lno < tlno) {
706 /* there is a pre-chunk part not blamed on parent */
707 split[0].suspect = blame_origin_incref(e->suspect);
708 split[0].lno = e->lno;
709 split[0].s_lno = e->s_lno;
710 split[0].num_lines = tlno - e->s_lno;
711 split[1].lno = e->lno + tlno - e->s_lno;
712 split[1].s_lno = plno;
713 }
714 else {
715 split[1].lno = e->lno;
716 split[1].s_lno = plno + (e->s_lno - tlno);
717 }
718
719 if (same < e->s_lno + e->num_lines) {
720 /* there is a post-chunk part not blamed on parent */
721 split[2].suspect = blame_origin_incref(e->suspect);
722 split[2].lno = e->lno + (same - e->s_lno);
723 split[2].s_lno = e->s_lno + (same - e->s_lno);
724 split[2].num_lines = e->s_lno + e->num_lines - same;
725 chunk_end_lno = split[2].lno;
726 }
727 else
728 chunk_end_lno = e->lno + e->num_lines;
729 split[1].num_lines = chunk_end_lno - split[1].lno;
730
731 /*
732 * if it turns out there is nothing to blame the parent for,
733 * forget about the splitting. !split[1].suspect signals this.
734 */
735 if (split[1].num_lines < 1)
736 return;
737 split[1].suspect = blame_origin_incref(parent);
738 }
739
740 /*
741 * split_overlap() divided an existing blame e into up to three parts
742 * in split. Any assigned blame is moved to queue to
743 * reflect the split.
744 */
745 static void split_blame(struct blame_entry ***blamed,
746 struct blame_entry ***unblamed,
747 struct blame_entry *split,
748 struct blame_entry *e)
749 {
750 if (split[0].suspect && split[2].suspect) {
751 /* The first part (reuse storage for the existing entry e) */
752 dup_entry(unblamed, e, &split[0]);
753
754 /* The last part -- me */
755 add_blame_entry(unblamed, &split[2]);
756
757 /* ... and the middle part -- parent */
758 add_blame_entry(blamed, &split[1]);
759 }
760 else if (!split[0].suspect && !split[2].suspect)
761 /*
762 * The parent covers the entire area; reuse storage for
763 * e and replace it with the parent.
764 */
765 dup_entry(blamed, e, &split[1]);
766 else if (split[0].suspect) {
767 /* me and then parent */
768 dup_entry(unblamed, e, &split[0]);
769 add_blame_entry(blamed, &split[1]);
770 }
771 else {
772 /* parent and then me */
773 dup_entry(blamed, e, &split[1]);
774 add_blame_entry(unblamed, &split[2]);
775 }
776 }
777
778 /*
779 * After splitting the blame, the origins used by the
780 * on-stack blame_entry should lose one refcnt each.
781 */
782 static void decref_split(struct blame_entry *split)
783 {
784 int i;
785
786 for (i = 0; i < 3; i++)
787 blame_origin_decref(split[i].suspect);
788 }
789
790 /*
791 * reverse_blame reverses the list given in head, appending tail.
792 * That allows us to build lists in reverse order, then reverse them
793 * afterwards. This can be faster than building the list in proper
794 * order right away. The reason is that building in proper order
795 * requires writing a link in the _previous_ element, while building
796 * in reverse order just requires placing the list head into the
797 * _current_ element.
798 */
799
800 static struct blame_entry *reverse_blame(struct blame_entry *head,
801 struct blame_entry *tail)
802 {
803 while (head) {
804 struct blame_entry *next = head->next;
805 head->next = tail;
806 tail = head;
807 head = next;
808 }
809 return tail;
810 }
811
812 /*
813 * Process one hunk from the patch between the current suspect for
814 * blame_entry e and its parent. This first blames any unfinished
815 * entries before the chunk (which is where target and parent start
816 * differing) on the parent, and then splits blame entries at the
817 * start and at the end of the difference region. Since use of -M and
818 * -C options may lead to overlapping/duplicate source line number
819 * ranges, all we can rely on from sorting/merging is the order of the
820 * first suspect line number.
821 */
822 static void blame_chunk(struct blame_entry ***dstq, struct blame_entry ***srcq,
823 int tlno, int offset, int same,
824 struct blame_origin *parent)
825 {
826 struct blame_entry *e = **srcq;
827 struct blame_entry *samep = NULL, *diffp = NULL;
828
829 while (e && e->s_lno < tlno) {
830 struct blame_entry *next = e->next;
831 /*
832 * current record starts before differing portion. If
833 * it reaches into it, we need to split it up and
834 * examine the second part separately.
835 */
836 if (e->s_lno + e->num_lines > tlno) {
837 /* Move second half to a new record */
838 int len = tlno - e->s_lno;
839 struct blame_entry *n = xcalloc(1, sizeof (struct blame_entry));
840 n->suspect = e->suspect;
841 n->lno = e->lno + len;
842 n->s_lno = e->s_lno + len;
843 n->num_lines = e->num_lines - len;
844 e->num_lines = len;
845 e->score = 0;
846 /* Push new record to diffp */
847 n->next = diffp;
848 diffp = n;
849 } else
850 blame_origin_decref(e->suspect);
851 /* Pass blame for everything before the differing
852 * chunk to the parent */
853 e->suspect = blame_origin_incref(parent);
854 e->s_lno += offset;
855 e->next = samep;
856 samep = e;
857 e = next;
858 }
859 /*
860 * As we don't know how much of a common stretch after this
861 * diff will occur, the currently blamed parts are all that we
862 * can assign to the parent for now.
863 */
864
865 if (samep) {
866 **dstq = reverse_blame(samep, **dstq);
867 *dstq = &samep->next;
868 }
869 /*
870 * Prepend the split off portions: everything after e starts
871 * after the blameable portion.
872 */
873 e = reverse_blame(diffp, e);
874
875 /*
876 * Now retain records on the target while parts are different
877 * from the parent.
878 */
879 samep = NULL;
880 diffp = NULL;
881 while (e && e->s_lno < same) {
882 struct blame_entry *next = e->next;
883
884 /*
885 * If current record extends into sameness, need to split.
886 */
887 if (e->s_lno + e->num_lines > same) {
888 /*
889 * Move second half to a new record to be
890 * processed by later chunks
891 */
892 int len = same - e->s_lno;
893 struct blame_entry *n = xcalloc(1, sizeof (struct blame_entry));
894 n->suspect = blame_origin_incref(e->suspect);
895 n->lno = e->lno + len;
896 n->s_lno = e->s_lno + len;
897 n->num_lines = e->num_lines - len;
898 e->num_lines = len;
899 e->score = 0;
900 /* Push new record to samep */
901 n->next = samep;
902 samep = n;
903 }
904 e->next = diffp;
905 diffp = e;
906 e = next;
907 }
908 **srcq = reverse_blame(diffp, reverse_blame(samep, e));
909 /* Move across elements that are in the unblamable portion */
910 if (diffp)
911 *srcq = &diffp->next;
912 }
913
914 struct blame_chunk_cb_data {
915 struct blame_origin *parent;
916 long offset;
917 struct blame_entry **dstq;
918 struct blame_entry **srcq;
919 };
920
921 /* diff chunks are from parent to target */
922 static int blame_chunk_cb(long start_a, long count_a,
923 long start_b, long count_b, void *data)
924 {
925 struct blame_chunk_cb_data *d = data;
926 if (start_a - start_b != d->offset)
927 die("internal error in blame::blame_chunk_cb");
928 blame_chunk(&d->dstq, &d->srcq, start_b, start_a - start_b,
929 start_b + count_b, d->parent);
930 d->offset = start_a + count_a - (start_b + count_b);
931 return 0;
932 }
933
934 /*
935 * We are looking at the origin 'target' and aiming to pass blame
936 * for the lines it is suspected to its parent. Run diff to find
937 * which lines came from parent and pass blame for them.
938 */
939 static void pass_blame_to_parent(struct blame_scoreboard *sb,
940 struct blame_origin *target,
941 struct blame_origin *parent)
942 {
943 mmfile_t file_p, file_o;
944 struct blame_chunk_cb_data d;
945 struct blame_entry *newdest = NULL;
946
947 if (!target->suspects)
948 return; /* nothing remains for this target */
949
950 d.parent = parent;
951 d.offset = 0;
952 d.dstq = &newdest; d.srcq = &target->suspects;
953
954 fill_origin_blob(&sb->revs->diffopt, parent, &file_p, &sb->num_read_blob);
955 fill_origin_blob(&sb->revs->diffopt, target, &file_o, &sb->num_read_blob);
956 sb->num_get_patch++;
957
958 if (diff_hunks(&file_p, &file_o, blame_chunk_cb, &d, sb->xdl_opts))
959 die("unable to generate diff (%s -> %s)",
960 oid_to_hex(&parent->commit->object.oid),
961 oid_to_hex(&target->commit->object.oid));
962 /* The rest are the same as the parent */
963 blame_chunk(&d.dstq, &d.srcq, INT_MAX, d.offset, INT_MAX, parent);
964 *d.dstq = NULL;
965 queue_blames(sb, parent, newdest);
966
967 return;
968 }
969
970 /*
971 * The lines in blame_entry after splitting blames many times can become
972 * very small and trivial, and at some point it becomes pointless to
973 * blame the parents. E.g. "\t\t}\n\t}\n\n" appears everywhere in any
974 * ordinary C program, and it is not worth to say it was copied from
975 * totally unrelated file in the parent.
976 *
977 * Compute how trivial the lines in the blame_entry are.
978 */
979 unsigned blame_entry_score(struct blame_scoreboard *sb, struct blame_entry *e)
980 {
981 unsigned score;
982 const char *cp, *ep;
983
984 if (e->score)
985 return e->score;
986
987 score = 1;
988 cp = blame_nth_line(sb, e->lno);
989 ep = blame_nth_line(sb, e->lno + e->num_lines);
990 while (cp < ep) {
991 unsigned ch = *((unsigned char *)cp);
992 if (isalnum(ch))
993 score++;
994 cp++;
995 }
996 e->score = score;
997 return score;
998 }
999
1000 /*
1001 * best_so_far[] and potential[] are both a split of an existing blame_entry
1002 * that passes blame to the parent. Maintain best_so_far the best split so
1003 * far, by comparing potential and best_so_far and copying potential into
1004 * bst_so_far as needed.
1005 */
1006 static void copy_split_if_better(struct blame_scoreboard *sb,
1007 struct blame_entry *best_so_far,
1008 struct blame_entry *potential)
1009 {
1010 int i;
1011
1012 if (!potential[1].suspect)
1013 return;
1014 if (best_so_far[1].suspect) {
1015 if (blame_entry_score(sb, &potential[1]) <
1016 blame_entry_score(sb, &best_so_far[1]))
1017 return;
1018 }
1019
1020 for (i = 0; i < 3; i++)
1021 blame_origin_incref(potential[i].suspect);
1022 decref_split(best_so_far);
1023 memcpy(best_so_far, potential, sizeof(struct blame_entry[3]));
1024 }
1025
1026 /*
1027 * We are looking at a part of the final image represented by
1028 * ent (tlno and same are offset by ent->s_lno).
1029 * tlno is where we are looking at in the final image.
1030 * up to (but not including) same match preimage.
1031 * plno is where we are looking at in the preimage.
1032 *
1033 * <-------------- final image ---------------------->
1034 * <------ent------>
1035 * ^tlno ^same
1036 * <---------preimage----->
1037 * ^plno
1038 *
1039 * All line numbers are 0-based.
1040 */
1041 static void handle_split(struct blame_scoreboard *sb,
1042 struct blame_entry *ent,
1043 int tlno, int plno, int same,
1044 struct blame_origin *parent,
1045 struct blame_entry *split)
1046 {
1047 if (ent->num_lines <= tlno)
1048 return;
1049 if (tlno < same) {
1050 struct blame_entry potential[3];
1051 tlno += ent->s_lno;
1052 same += ent->s_lno;
1053 split_overlap(potential, ent, tlno, plno, same, parent);
1054 copy_split_if_better(sb, split, potential);
1055 decref_split(potential);
1056 }
1057 }
1058
1059 struct handle_split_cb_data {
1060 struct blame_scoreboard *sb;
1061 struct blame_entry *ent;
1062 struct blame_origin *parent;
1063 struct blame_entry *split;
1064 long plno;
1065 long tlno;
1066 };
1067
1068 static int handle_split_cb(long start_a, long count_a,
1069 long start_b, long count_b, void *data)
1070 {
1071 struct handle_split_cb_data *d = data;
1072 handle_split(d->sb, d->ent, d->tlno, d->plno, start_b, d->parent,
1073 d->split);
1074 d->plno = start_a + count_a;
1075 d->tlno = start_b + count_b;
1076 return 0;
1077 }
1078
1079 /*
1080 * Find the lines from parent that are the same as ent so that
1081 * we can pass blames to it. file_p has the blob contents for
1082 * the parent.
1083 */
1084 static void find_copy_in_blob(struct blame_scoreboard *sb,
1085 struct blame_entry *ent,
1086 struct blame_origin *parent,
1087 struct blame_entry *split,
1088 mmfile_t *file_p)
1089 {
1090 const char *cp;
1091 mmfile_t file_o;
1092 struct handle_split_cb_data d;
1093
1094 memset(&d, 0, sizeof(d));
1095 d.sb = sb; d.ent = ent; d.parent = parent; d.split = split;
1096 /*
1097 * Prepare mmfile that contains only the lines in ent.
1098 */
1099 cp = blame_nth_line(sb, ent->lno);
1100 file_o.ptr = (char *) cp;
1101 file_o.size = blame_nth_line(sb, ent->lno + ent->num_lines) - cp;
1102
1103 /*
1104 * file_o is a part of final image we are annotating.
1105 * file_p partially may match that image.
1106 */
1107 memset(split, 0, sizeof(struct blame_entry [3]));
1108 if (diff_hunks(file_p, &file_o, handle_split_cb, &d, sb->xdl_opts))
1109 die("unable to generate diff (%s)",
1110 oid_to_hex(&parent->commit->object.oid));
1111 /* remainder, if any, all match the preimage */
1112 handle_split(sb, ent, d.tlno, d.plno, ent->num_lines, parent, split);
1113 }
1114
1115 /* Move all blame entries from list *source that have a score smaller
1116 * than score_min to the front of list *small.
1117 * Returns a pointer to the link pointing to the old head of the small list.
1118 */
1119
1120 static struct blame_entry **filter_small(struct blame_scoreboard *sb,
1121 struct blame_entry **small,
1122 struct blame_entry **source,
1123 unsigned score_min)
1124 {
1125 struct blame_entry *p = *source;
1126 struct blame_entry *oldsmall = *small;
1127 while (p) {
1128 if (blame_entry_score(sb, p) <= score_min) {
1129 *small = p;
1130 small = &p->next;
1131 p = *small;
1132 } else {
1133 *source = p;
1134 source = &p->next;
1135 p = *source;
1136 }
1137 }
1138 *small = oldsmall;
1139 *source = NULL;
1140 return small;
1141 }
1142
1143 /*
1144 * See if lines currently target is suspected for can be attributed to
1145 * parent.
1146 */
1147 static void find_move_in_parent(struct blame_scoreboard *sb,
1148 struct blame_entry ***blamed,
1149 struct blame_entry **toosmall,
1150 struct blame_origin *target,
1151 struct blame_origin *parent)
1152 {
1153 struct blame_entry *e, split[3];
1154 struct blame_entry *unblamed = target->suspects;
1155 struct blame_entry *leftover = NULL;
1156 mmfile_t file_p;
1157
1158 if (!unblamed)
1159 return; /* nothing remains for this target */
1160
1161 fill_origin_blob(&sb->revs->diffopt, parent, &file_p, &sb->num_read_blob);
1162 if (!file_p.ptr)
1163 return;
1164
1165 /* At each iteration, unblamed has a NULL-terminated list of
1166 * entries that have not yet been tested for blame. leftover
1167 * contains the reversed list of entries that have been tested
1168 * without being assignable to the parent.
1169 */
1170 do {
1171 struct blame_entry **unblamedtail = &unblamed;
1172 struct blame_entry *next;
1173 for (e = unblamed; e; e = next) {
1174 next = e->next;
1175 find_copy_in_blob(sb, e, parent, split, &file_p);
1176 if (split[1].suspect &&
1177 sb->move_score < blame_entry_score(sb, &split[1])) {
1178 split_blame(blamed, &unblamedtail, split, e);
1179 } else {
1180 e->next = leftover;
1181 leftover = e;
1182 }
1183 decref_split(split);
1184 }
1185 *unblamedtail = NULL;
1186 toosmall = filter_small(sb, toosmall, &unblamed, sb->move_score);
1187 } while (unblamed);
1188 target->suspects = reverse_blame(leftover, NULL);
1189 }
1190
1191 struct blame_list {
1192 struct blame_entry *ent;
1193 struct blame_entry split[3];
1194 };
1195
1196 /*
1197 * Count the number of entries the target is suspected for,
1198 * and prepare a list of entry and the best split.
1199 */
1200 static struct blame_list *setup_blame_list(struct blame_entry *unblamed,
1201 int *num_ents_p)
1202 {
1203 struct blame_entry *e;
1204 int num_ents, i;
1205 struct blame_list *blame_list = NULL;
1206
1207 for (e = unblamed, num_ents = 0; e; e = e->next)
1208 num_ents++;
1209 if (num_ents) {
1210 blame_list = xcalloc(num_ents, sizeof(struct blame_list));
1211 for (e = unblamed, i = 0; e; e = e->next)
1212 blame_list[i++].ent = e;
1213 }
1214 *num_ents_p = num_ents;
1215 return blame_list;
1216 }
1217
1218 /*
1219 * For lines target is suspected for, see if we can find code movement
1220 * across file boundary from the parent commit. porigin is the path
1221 * in the parent we already tried.
1222 */
1223 static void find_copy_in_parent(struct blame_scoreboard *sb,
1224 struct blame_entry ***blamed,
1225 struct blame_entry **toosmall,
1226 struct blame_origin *target,
1227 struct commit *parent,
1228 struct blame_origin *porigin,
1229 int opt)
1230 {
1231 struct diff_options diff_opts;
1232 int i, j;
1233 struct blame_list *blame_list;
1234 int num_ents;
1235 struct blame_entry *unblamed = target->suspects;
1236 struct blame_entry *leftover = NULL;
1237
1238 if (!unblamed)
1239 return; /* nothing remains for this target */
1240
1241 diff_setup(&diff_opts);
1242 diff_opts.flags.recursive = 1;
1243 diff_opts.output_format = DIFF_FORMAT_NO_OUTPUT;
1244
1245 diff_setup_done(&diff_opts);
1246
1247 /* Try "find copies harder" on new path if requested;
1248 * we do not want to use diffcore_rename() actually to
1249 * match things up; find_copies_harder is set only to
1250 * force diff_tree_oid() to feed all filepairs to diff_queue,
1251 * and this code needs to be after diff_setup_done(), which
1252 * usually makes find-copies-harder imply copy detection.
1253 */
1254 if ((opt & PICKAXE_BLAME_COPY_HARDEST)
1255 || ((opt & PICKAXE_BLAME_COPY_HARDER)
1256 && (!porigin || strcmp(target->path, porigin->path))))
1257 diff_opts.flags.find_copies_harder = 1;
1258
1259 if (is_null_oid(&target->commit->object.oid))
1260 do_diff_cache(&parent->tree->object.oid, &diff_opts);
1261 else
1262 diff_tree_oid(&parent->tree->object.oid,
1263 &target->commit->tree->object.oid,
1264 "", &diff_opts);
1265
1266 if (!diff_opts.flags.find_copies_harder)
1267 diffcore_std(&diff_opts);
1268
1269 do {
1270 struct blame_entry **unblamedtail = &unblamed;
1271 blame_list = setup_blame_list(unblamed, &num_ents);
1272
1273 for (i = 0; i < diff_queued_diff.nr; i++) {
1274 struct diff_filepair *p = diff_queued_diff.queue[i];
1275 struct blame_origin *norigin;
1276 mmfile_t file_p;
1277 struct blame_entry potential[3];
1278
1279 if (!DIFF_FILE_VALID(p->one))
1280 continue; /* does not exist in parent */
1281 if (S_ISGITLINK(p->one->mode))
1282 continue; /* ignore git links */
1283 if (porigin && !strcmp(p->one->path, porigin->path))
1284 /* find_move already dealt with this path */
1285 continue;
1286
1287 norigin = get_origin(parent, p->one->path);
1288 oidcpy(&norigin->blob_oid, &p->one->oid);
1289 norigin->mode = p->one->mode;
1290 fill_origin_blob(&sb->revs->diffopt, norigin, &file_p, &sb->num_read_blob);
1291 if (!file_p.ptr)
1292 continue;
1293
1294 for (j = 0; j < num_ents; j++) {
1295 find_copy_in_blob(sb, blame_list[j].ent,
1296 norigin, potential, &file_p);
1297 copy_split_if_better(sb, blame_list[j].split,
1298 potential);
1299 decref_split(potential);
1300 }
1301 blame_origin_decref(norigin);
1302 }
1303
1304 for (j = 0; j < num_ents; j++) {
1305 struct blame_entry *split = blame_list[j].split;
1306 if (split[1].suspect &&
1307 sb->copy_score < blame_entry_score(sb, &split[1])) {
1308 split_blame(blamed, &unblamedtail, split,
1309 blame_list[j].ent);
1310 } else {
1311 blame_list[j].ent->next = leftover;
1312 leftover = blame_list[j].ent;
1313 }
1314 decref_split(split);
1315 }
1316 free(blame_list);
1317 *unblamedtail = NULL;
1318 toosmall = filter_small(sb, toosmall, &unblamed, sb->copy_score);
1319 } while (unblamed);
1320 target->suspects = reverse_blame(leftover, NULL);
1321 diff_flush(&diff_opts);
1322 clear_pathspec(&diff_opts.pathspec);
1323 }
1324
1325 /*
1326 * The blobs of origin and porigin exactly match, so everything
1327 * origin is suspected for can be blamed on the parent.
1328 */
1329 static void pass_whole_blame(struct blame_scoreboard *sb,
1330 struct blame_origin *origin, struct blame_origin *porigin)
1331 {
1332 struct blame_entry *e, *suspects;
1333
1334 if (!porigin->file.ptr && origin->file.ptr) {
1335 /* Steal its file */
1336 porigin->file = origin->file;
1337 origin->file.ptr = NULL;
1338 }
1339 suspects = origin->suspects;
1340 origin->suspects = NULL;
1341 for (e = suspects; e; e = e->next) {
1342 blame_origin_incref(porigin);
1343 blame_origin_decref(e->suspect);
1344 e->suspect = porigin;
1345 }
1346 queue_blames(sb, porigin, suspects);
1347 }
1348
1349 /*
1350 * We pass blame from the current commit to its parents. We keep saying
1351 * "parent" (and "porigin"), but what we mean is to find scapegoat to
1352 * exonerate ourselves.
1353 */
1354 static struct commit_list *first_scapegoat(struct rev_info *revs, struct commit *commit,
1355 int reverse)
1356 {
1357 if (!reverse) {
1358 if (revs->first_parent_only &&
1359 commit->parents &&
1360 commit->parents->next) {
1361 free_commit_list(commit->parents->next);
1362 commit->parents->next = NULL;
1363 }
1364 return commit->parents;
1365 }
1366 return lookup_decoration(&revs->children, &commit->object);
1367 }
1368
1369 static int num_scapegoats(struct rev_info *revs, struct commit *commit, int reverse)
1370 {
1371 struct commit_list *l = first_scapegoat(revs, commit, reverse);
1372 return commit_list_count(l);
1373 }
1374
1375 /* Distribute collected unsorted blames to the respected sorted lists
1376 * in the various origins.
1377 */
1378 static void distribute_blame(struct blame_scoreboard *sb, struct blame_entry *blamed)
1379 {
1380 blamed = llist_mergesort(blamed, get_next_blame, set_next_blame,
1381 compare_blame_suspect);
1382 while (blamed)
1383 {
1384 struct blame_origin *porigin = blamed->suspect;
1385 struct blame_entry *suspects = NULL;
1386 do {
1387 struct blame_entry *next = blamed->next;
1388 blamed->next = suspects;
1389 suspects = blamed;
1390 blamed = next;
1391 } while (blamed && blamed->suspect == porigin);
1392 suspects = reverse_blame(suspects, NULL);
1393 queue_blames(sb, porigin, suspects);
1394 }
1395 }
1396
1397 #define MAXSG 16
1398
1399 static void pass_blame(struct blame_scoreboard *sb, struct blame_origin *origin, int opt)
1400 {
1401 struct rev_info *revs = sb->revs;
1402 int i, pass, num_sg;
1403 struct commit *commit = origin->commit;
1404 struct commit_list *sg;
1405 struct blame_origin *sg_buf[MAXSG];
1406 struct blame_origin *porigin, **sg_origin = sg_buf;
1407 struct blame_entry *toosmall = NULL;
1408 struct blame_entry *blames, **blametail = &blames;
1409
1410 num_sg = num_scapegoats(revs, commit, sb->reverse);
1411 if (!num_sg)
1412 goto finish;
1413 else if (num_sg < ARRAY_SIZE(sg_buf))
1414 memset(sg_buf, 0, sizeof(sg_buf));
1415 else
1416 sg_origin = xcalloc(num_sg, sizeof(*sg_origin));
1417
1418 /*
1419 * The first pass looks for unrenamed path to optimize for
1420 * common cases, then we look for renames in the second pass.
1421 */
1422 for (pass = 0; pass < 2 - sb->no_whole_file_rename; pass++) {
1423 struct blame_origin *(*find)(struct commit *, struct blame_origin *);
1424 find = pass ? find_rename : find_origin;
1425
1426 for (i = 0, sg = first_scapegoat(revs, commit, sb->reverse);
1427 i < num_sg && sg;
1428 sg = sg->next, i++) {
1429 struct commit *p = sg->item;
1430 int j, same;
1431
1432 if (sg_origin[i])
1433 continue;
1434 if (parse_commit(p))
1435 continue;
1436 porigin = find(p, origin);
1437 if (!porigin)
1438 continue;
1439 if (!oidcmp(&porigin->blob_oid, &origin->blob_oid)) {
1440 pass_whole_blame(sb, origin, porigin);
1441 blame_origin_decref(porigin);
1442 goto finish;
1443 }
1444 for (j = same = 0; j < i; j++)
1445 if (sg_origin[j] &&
1446 !oidcmp(&sg_origin[j]->blob_oid, &porigin->blob_oid)) {
1447 same = 1;
1448 break;
1449 }
1450 if (!same)
1451 sg_origin[i] = porigin;
1452 else
1453 blame_origin_decref(porigin);
1454 }
1455 }
1456
1457 sb->num_commits++;
1458 for (i = 0, sg = first_scapegoat(revs, commit, sb->reverse);
1459 i < num_sg && sg;
1460 sg = sg->next, i++) {
1461 struct blame_origin *porigin = sg_origin[i];
1462 if (!porigin)
1463 continue;
1464 if (!origin->previous) {
1465 blame_origin_incref(porigin);
1466 origin->previous = porigin;
1467 }
1468 pass_blame_to_parent(sb, origin, porigin);
1469 if (!origin->suspects)
1470 goto finish;
1471 }
1472
1473 /*
1474 * Optionally find moves in parents' files.
1475 */
1476 if (opt & PICKAXE_BLAME_MOVE) {
1477 filter_small(sb, &toosmall, &origin->suspects, sb->move_score);
1478 if (origin->suspects) {
1479 for (i = 0, sg = first_scapegoat(revs, commit, sb->reverse);
1480 i < num_sg && sg;
1481 sg = sg->next, i++) {
1482 struct blame_origin *porigin = sg_origin[i];
1483 if (!porigin)
1484 continue;
1485 find_move_in_parent(sb, &blametail, &toosmall, origin, porigin);
1486 if (!origin->suspects)
1487 break;
1488 }
1489 }
1490 }
1491
1492 /*
1493 * Optionally find copies from parents' files.
1494 */
1495 if (opt & PICKAXE_BLAME_COPY) {
1496 if (sb->copy_score > sb->move_score)
1497 filter_small(sb, &toosmall, &origin->suspects, sb->copy_score);
1498 else if (sb->copy_score < sb->move_score) {
1499 origin->suspects = blame_merge(origin->suspects, toosmall);
1500 toosmall = NULL;
1501 filter_small(sb, &toosmall, &origin->suspects, sb->copy_score);
1502 }
1503 if (!origin->suspects)
1504 goto finish;
1505
1506 for (i = 0, sg = first_scapegoat(revs, commit, sb->reverse);
1507 i < num_sg && sg;
1508 sg = sg->next, i++) {
1509 struct blame_origin *porigin = sg_origin[i];
1510 find_copy_in_parent(sb, &blametail, &toosmall,
1511 origin, sg->item, porigin, opt);
1512 if (!origin->suspects)
1513 goto finish;
1514 }
1515 }
1516
1517 finish:
1518 *blametail = NULL;
1519 distribute_blame(sb, blames);
1520 /*
1521 * prepend toosmall to origin->suspects
1522 *
1523 * There is no point in sorting: this ends up on a big
1524 * unsorted list in the caller anyway.
1525 */
1526 if (toosmall) {
1527 struct blame_entry **tail = &toosmall;
1528 while (*tail)
1529 tail = &(*tail)->next;
1530 *tail = origin->suspects;
1531 origin->suspects = toosmall;
1532 }
1533 for (i = 0; i < num_sg; i++) {
1534 if (sg_origin[i]) {
1535 drop_origin_blob(sg_origin[i]);
1536 blame_origin_decref(sg_origin[i]);
1537 }
1538 }
1539 drop_origin_blob(origin);
1540 if (sg_buf != sg_origin)
1541 free(sg_origin);
1542 }
1543
1544 /*
1545 * The main loop -- while we have blobs with lines whose true origin
1546 * is still unknown, pick one blob, and allow its lines to pass blames
1547 * to its parents. */
1548 void assign_blame(struct blame_scoreboard *sb, int opt)
1549 {
1550 struct rev_info *revs = sb->revs;
1551 struct commit *commit = prio_queue_get(&sb->commits);
1552
1553 while (commit) {
1554 struct blame_entry *ent;
1555 struct blame_origin *suspect = commit->util;
1556
1557 /* find one suspect to break down */
1558 while (suspect && !suspect->suspects)
1559 suspect = suspect->next;
1560
1561 if (!suspect) {
1562 commit = prio_queue_get(&sb->commits);
1563 continue;
1564 }
1565
1566 assert(commit == suspect->commit);
1567
1568 /*
1569 * We will use this suspect later in the loop,
1570 * so hold onto it in the meantime.
1571 */
1572 blame_origin_incref(suspect);
1573 parse_commit(commit);
1574 if (sb->reverse ||
1575 (!(commit->object.flags & UNINTERESTING) &&
1576 !(revs->max_age != -1 && commit->date < revs->max_age)))
1577 pass_blame(sb, suspect, opt);
1578 else {
1579 commit->object.flags |= UNINTERESTING;
1580 if (commit->object.parsed)
1581 mark_parents_uninteresting(commit);
1582 }
1583 /* treat root commit as boundary */
1584 if (!commit->parents && !sb->show_root)
1585 commit->object.flags |= UNINTERESTING;
1586
1587 /* Take responsibility for the remaining entries */
1588 ent = suspect->suspects;
1589 if (ent) {
1590 suspect->guilty = 1;
1591 for (;;) {
1592 struct blame_entry *next = ent->next;
1593 if (sb->found_guilty_entry)
1594 sb->found_guilty_entry(ent, sb->found_guilty_entry_data);
1595 if (next) {
1596 ent = next;
1597 continue;
1598 }
1599 ent->next = sb->ent;
1600 sb->ent = suspect->suspects;
1601 suspect->suspects = NULL;
1602 break;
1603 }
1604 }
1605 blame_origin_decref(suspect);
1606
1607 if (sb->debug) /* sanity */
1608 sanity_check_refcnt(sb);
1609 }
1610 }
1611
1612 static const char *get_next_line(const char *start, const char *end)
1613 {
1614 const char *nl = memchr(start, '\n', end - start);
1615 return nl ? nl + 1 : end;
1616 }
1617
1618 /*
1619 * To allow quick access to the contents of nth line in the
1620 * final image, prepare an index in the scoreboard.
1621 */
1622 static int prepare_lines(struct blame_scoreboard *sb)
1623 {
1624 const char *buf = sb->final_buf;
1625 unsigned long len = sb->final_buf_size;
1626 const char *end = buf + len;
1627 const char *p;
1628 int *lineno;
1629 int num = 0;
1630
1631 for (p = buf; p < end; p = get_next_line(p, end))
1632 num++;
1633
1634 ALLOC_ARRAY(sb->lineno, num + 1);
1635 lineno = sb->lineno;
1636
1637 for (p = buf; p < end; p = get_next_line(p, end))
1638 *lineno++ = p - buf;
1639
1640 *lineno = len;
1641
1642 sb->num_lines = num;
1643 return sb->num_lines;
1644 }
1645
1646 static struct commit *find_single_final(struct rev_info *revs,
1647 const char **name_p)
1648 {
1649 int i;
1650 struct commit *found = NULL;
1651 const char *name = NULL;
1652
1653 for (i = 0; i < revs->pending.nr; i++) {
1654 struct object *obj = revs->pending.objects[i].item;
1655 if (obj->flags & UNINTERESTING)
1656 continue;
1657 obj = deref_tag(obj, NULL, 0);
1658 if (obj->type != OBJ_COMMIT)
1659 die("Non commit %s?", revs->pending.objects[i].name);
1660 if (found)
1661 die("More than one commit to dig from %s and %s?",
1662 revs->pending.objects[i].name, name);
1663 found = (struct commit *)obj;
1664 name = revs->pending.objects[i].name;
1665 }
1666 if (name_p)
1667 *name_p = xstrdup_or_null(name);
1668 return found;
1669 }
1670
1671 static struct commit *dwim_reverse_initial(struct rev_info *revs,
1672 const char **name_p)
1673 {
1674 /*
1675 * DWIM "git blame --reverse ONE -- PATH" as
1676 * "git blame --reverse ONE..HEAD -- PATH" but only do so
1677 * when it makes sense.
1678 */
1679 struct object *obj;
1680 struct commit *head_commit;
1681 struct object_id head_oid;
1682
1683 if (revs->pending.nr != 1)
1684 return NULL;
1685
1686 /* Is that sole rev a committish? */
1687 obj = revs->pending.objects[0].item;
1688 obj = deref_tag(obj, NULL, 0);
1689 if (obj->type != OBJ_COMMIT)
1690 return NULL;
1691
1692 /* Do we have HEAD? */
1693 if (!resolve_ref_unsafe("HEAD", RESOLVE_REF_READING, &head_oid, NULL))
1694 return NULL;
1695 head_commit = lookup_commit_reference_gently(&head_oid, 1);
1696 if (!head_commit)
1697 return NULL;
1698
1699 /* Turn "ONE" into "ONE..HEAD" then */
1700 obj->flags |= UNINTERESTING;
1701 add_pending_object(revs, &head_commit->object, "HEAD");
1702
1703 if (name_p)
1704 *name_p = revs->pending.objects[0].name;
1705 return (struct commit *)obj;
1706 }
1707
1708 static struct commit *find_single_initial(struct rev_info *revs,
1709 const char **name_p)
1710 {
1711 int i;
1712 struct commit *found = NULL;
1713 const char *name = NULL;
1714
1715 /*
1716 * There must be one and only one negative commit, and it must be
1717 * the boundary.
1718 */
1719 for (i = 0; i < revs->pending.nr; i++) {
1720 struct object *obj = revs->pending.objects[i].item;
1721 if (!(obj->flags & UNINTERESTING))
1722 continue;
1723 obj = deref_tag(obj, NULL, 0);
1724 if (obj->type != OBJ_COMMIT)
1725 die("Non commit %s?", revs->pending.objects[i].name);
1726 if (found)
1727 die("More than one commit to dig up from, %s and %s?",
1728 revs->pending.objects[i].name, name);
1729 found = (struct commit *) obj;
1730 name = revs->pending.objects[i].name;
1731 }
1732
1733 if (!name)
1734 found = dwim_reverse_initial(revs, &name);
1735 if (!name)
1736 die("No commit to dig up from?");
1737
1738 if (name_p)
1739 *name_p = xstrdup(name);
1740 return found;
1741 }
1742
1743 void init_scoreboard(struct blame_scoreboard *sb)
1744 {
1745 memset(sb, 0, sizeof(struct blame_scoreboard));
1746 sb->move_score = BLAME_DEFAULT_MOVE_SCORE;
1747 sb->copy_score = BLAME_DEFAULT_COPY_SCORE;
1748 }
1749
1750 void setup_scoreboard(struct blame_scoreboard *sb, const char *path, struct blame_origin **orig)
1751 {
1752 const char *final_commit_name = NULL;
1753 struct blame_origin *o;
1754 struct commit *final_commit = NULL;
1755 enum object_type type;
1756
1757 if (sb->reverse && sb->contents_from)
1758 die(_("--contents and --reverse do not blend well."));
1759
1760 if (!sb->reverse) {
1761 sb->final = find_single_final(sb->revs, &final_commit_name);
1762 sb->commits.compare = compare_commits_by_commit_date;
1763 } else {
1764 sb->final = find_single_initial(sb->revs, &final_commit_name);
1765 sb->commits.compare = compare_commits_by_reverse_commit_date;
1766 }
1767
1768 if (sb->final && sb->contents_from)
1769 die(_("cannot use --contents with final commit object name"));
1770
1771 if (sb->reverse && sb->revs->first_parent_only)
1772 sb->revs->children.name = NULL;
1773
1774 if (!sb->final) {
1775 /*
1776 * "--not A B -- path" without anything positive;
1777 * do not default to HEAD, but use the working tree
1778 * or "--contents".
1779 */
1780 setup_work_tree();
1781 sb->final = fake_working_tree_commit(&sb->revs->diffopt,
1782 path, sb->contents_from);
1783 add_pending_object(sb->revs, &(sb->final->object), ":");
1784 }
1785
1786 if (sb->reverse && sb->revs->first_parent_only) {
1787 final_commit = find_single_final(sb->revs, NULL);
1788 if (!final_commit)
1789 die(_("--reverse and --first-parent together require specified latest commit"));
1790 }
1791
1792 /*
1793 * If we have bottom, this will mark the ancestors of the
1794 * bottom commits we would reach while traversing as
1795 * uninteresting.
1796 */
1797 if (prepare_revision_walk(sb->revs))
1798 die(_("revision walk setup failed"));
1799
1800 if (sb->reverse && sb->revs->first_parent_only) {
1801 struct commit *c = final_commit;
1802
1803 sb->revs->children.name = "children";
1804 while (c->parents &&
1805 oidcmp(&c->object.oid, &sb->final->object.oid)) {
1806 struct commit_list *l = xcalloc(1, sizeof(*l));
1807
1808 l->item = c;
1809 if (add_decoration(&sb->revs->children,
1810 &c->parents->item->object, l))
1811 die("BUG: not unique item in first-parent chain");
1812 c = c->parents->item;
1813 }
1814
1815 if (oidcmp(&c->object.oid, &sb->final->object.oid))
1816 die(_("--reverse --first-parent together require range along first-parent chain"));
1817 }
1818
1819 if (is_null_oid(&sb->final->object.oid)) {
1820 o = sb->final->util;
1821 sb->final_buf = xmemdupz(o->file.ptr, o->file.size);
1822 sb->final_buf_size = o->file.size;
1823 }
1824 else {
1825 o = get_origin(sb->final, path);
1826 if (fill_blob_sha1_and_mode(o))
1827 die(_("no such path %s in %s"), path, final_commit_name);
1828
1829 if (sb->revs->diffopt.flags.allow_textconv &&
1830 textconv_object(path, o->mode, &o->blob_oid, 1, (char **) &sb->final_buf,
1831 &sb->final_buf_size))
1832 ;
1833 else
1834 sb->final_buf = read_sha1_file(o->blob_oid.hash, &type,
1835 &sb->final_buf_size);
1836
1837 if (!sb->final_buf)
1838 die(_("cannot read blob %s for path %s"),
1839 oid_to_hex(&o->blob_oid),
1840 path);
1841 }
1842 sb->num_read_blob++;
1843 prepare_lines(sb);
1844
1845 if (orig)
1846 *orig = o;
1847
1848 free((char *)final_commit_name);
1849 }
1850
1851
1852
1853 struct blame_entry *blame_entry_prepend(struct blame_entry *head,
1854 long start, long end,
1855 struct blame_origin *o)
1856 {
1857 struct blame_entry *new_head = xcalloc(1, sizeof(struct blame_entry));
1858 new_head->lno = start;
1859 new_head->num_lines = end - start;
1860 new_head->suspect = o;
1861 new_head->s_lno = start;
1862 new_head->next = head;
1863 blame_origin_incref(o);
1864 return new_head;
1865 }