alloc: allow arbitrary repositories for alloc functions
[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"
09002f1b 7#include "tag.h"
f5dd754c 8#include "blame.h"
14ba97f8 9#include "alloc.h"
f5dd754c
JS
10
11void blame_origin_decref(struct blame_origin *o)
12{
13 if (o && --o->refcnt <= 0) {
14 struct blame_origin *p, *l = NULL;
15 if (o->previous)
16 blame_origin_decref(o->previous);
17 free(o->file.ptr);
18 /* Should be present exactly once in commit chain */
19 for (p = o->commit->util; p; l = p, p = p->next) {
20 if (p == o) {
21 if (l)
22 l->next = p->next;
23 else
24 o->commit->util = p->next;
25 free(o);
26 return;
27 }
28 }
29 die("internal error in blame_origin_decref");
30 }
31}
32
33/*
34 * Given a commit and a path in it, create a new origin structure.
35 * The callers that add blame to the scoreboard should use
36 * get_origin() to obtain shared, refcounted copy instead of calling
37 * this function directly.
38 */
072bf432 39static struct blame_origin *make_origin(struct commit *commit, const char *path)
f5dd754c
JS
40{
41 struct blame_origin *o;
42 FLEX_ALLOC_STR(o, path, path);
43 o->commit = commit;
44 o->refcnt = 1;
45 o->next = commit->util;
46 commit->util = o;
47 return o;
48}
49
50/*
51 * Locate an existing origin or create a new one.
52 * This moves the origin to front position in the commit util list.
53 */
09002f1b 54static struct blame_origin *get_origin(struct commit *commit, const char *path)
f5dd754c
JS
55{
56 struct blame_origin *o, *l;
57
58 for (o = commit->util, l = NULL; o; l = o, o = o->next) {
59 if (!strcmp(o->path, path)) {
60 /* bump to front */
61 if (l) {
62 l->next = o->next;
63 o->next = commit->util;
64 commit->util = o;
65 }
66 return blame_origin_incref(o);
67 }
68 }
69 return make_origin(commit, path);
70}
072bf432
JS
71
72
73
74static void verify_working_tree_path(struct commit *work_tree, const char *path)
75{
76 struct commit_list *parents;
77 int pos;
78
79 for (parents = work_tree->parents; parents; parents = parents->next) {
80 const struct object_id *commit_oid = &parents->item->object.oid;
81 struct object_id blob_oid;
82 unsigned mode;
83
916bc35b 84 if (!get_tree_entry(commit_oid, path, &blob_oid, &mode) &&
0df8e965 85 oid_object_info(the_repository, &blob_oid, NULL) == OBJ_BLOB)
072bf432
JS
86 return;
87 }
88
89 pos = cache_name_pos(path, strlen(path));
90 if (pos >= 0)
91 ; /* path is in the index */
92 else if (-1 - pos < active_nr &&
93 !strcmp(active_cache[-1 - pos]->name, path))
94 ; /* path is in the index, unmerged */
95 else
96 die("no such path '%s' in HEAD", path);
97}
98
99static struct commit_list **append_parent(struct commit_list **tail, const struct object_id *oid)
100{
101 struct commit *parent;
102
583c6a22 103 parent = lookup_commit_reference(oid);
072bf432
JS
104 if (!parent)
105 die("no such commit %s", oid_to_hex(oid));
106 return &commit_list_insert(parent, tail)->next;
107}
108
109static void append_merge_parents(struct commit_list **tail)
110{
111 int merge_head;
112 struct strbuf line = STRBUF_INIT;
113
114 merge_head = open(git_path_merge_head(), O_RDONLY);
115 if (merge_head < 0) {
116 if (errno == ENOENT)
117 return;
118 die("cannot open '%s' for reading", git_path_merge_head());
119 }
120
121 while (!strbuf_getwholeline_fd(&line, merge_head, '\n')) {
122 struct object_id oid;
123 if (line.len < GIT_SHA1_HEXSZ || get_oid_hex(line.buf, &oid))
124 die("unknown line in '%s': %s", git_path_merge_head(), line.buf);
125 tail = append_parent(tail, &oid);
126 }
127 close(merge_head);
128 strbuf_release(&line);
129}
130
131/*
132 * This isn't as simple as passing sb->buf and sb->len, because we
133 * want to transfer ownership of the buffer to the commit (so we
134 * must use detach).
135 */
136static void set_commit_buffer_from_strbuf(struct commit *c, struct strbuf *sb)
137{
138 size_t len;
139 void *buf = strbuf_detach(sb, &len);
140 set_commit_buffer(c, buf, len);
141}
142
143/*
144 * Prepare a dummy commit that represents the work tree (or staged) item.
145 * Note that annotating work tree item never works in the reverse.
146 */
09002f1b
JS
147static struct commit *fake_working_tree_commit(struct diff_options *opt,
148 const char *path,
149 const char *contents_from)
072bf432
JS
150{
151 struct commit *commit;
152 struct blame_origin *origin;
153 struct commit_list **parent_tail, *parent;
154 struct object_id head_oid;
155 struct strbuf buf = STRBUF_INIT;
156 const char *ident;
157 time_t now;
158 int size, len;
159 struct cache_entry *ce;
160 unsigned mode;
161 struct strbuf msg = STRBUF_INIT;
162
163 read_cache();
164 time(&now);
8ba0e5ec 165 commit = alloc_commit_node(the_repository);
072bf432
JS
166 commit->object.parsed = 1;
167 commit->date = now;
168 parent_tail = &commit->parents;
169
49e61479 170 if (!resolve_ref_unsafe("HEAD", RESOLVE_REF_READING, &head_oid, NULL))
072bf432
JS
171 die("no such ref: HEAD");
172
173 parent_tail = append_parent(parent_tail, &head_oid);
174 append_merge_parents(parent_tail);
175 verify_working_tree_path(commit, path);
176
177 origin = make_origin(commit, path);
178
179 ident = fmt_ident("Not Committed Yet", "not.committed.yet", NULL, 0);
180 strbuf_addstr(&msg, "tree 0000000000000000000000000000000000000000\n");
181 for (parent = commit->parents; parent; parent = parent->next)
182 strbuf_addf(&msg, "parent %s\n",
183 oid_to_hex(&parent->item->object.oid));
184 strbuf_addf(&msg,
185 "author %s\n"
186 "committer %s\n\n"
187 "Version of %s from %s\n",
188 ident, ident, path,
189 (!contents_from ? path :
190 (!strcmp(contents_from, "-") ? "standard input" : contents_from)));
191 set_commit_buffer_from_strbuf(commit, &msg);
192
193 if (!contents_from || strcmp("-", contents_from)) {
194 struct stat st;
195 const char *read_from;
196 char *buf_ptr;
197 unsigned long buf_len;
198
199 if (contents_from) {
200 if (stat(contents_from, &st) < 0)
201 die_errno("Cannot stat '%s'", contents_from);
202 read_from = contents_from;
203 }
204 else {
205 if (lstat(path, &st) < 0)
206 die_errno("Cannot lstat '%s'", path);
207 read_from = path;
208 }
209 mode = canon_mode(st.st_mode);
210
211 switch (st.st_mode & S_IFMT) {
212 case S_IFREG:
0d1e0e78 213 if (opt->flags.allow_textconv &&
072bf432
JS
214 textconv_object(read_from, mode, &null_oid, 0, &buf_ptr, &buf_len))
215 strbuf_attach(&buf, buf_ptr, buf_len, buf_len + 1);
216 else if (strbuf_read_file(&buf, read_from, st.st_size) != st.st_size)
217 die_errno("cannot open or read '%s'", read_from);
218 break;
219 case S_IFLNK:
220 if (strbuf_readlink(&buf, read_from, st.st_size) < 0)
221 die_errno("cannot readlink '%s'", read_from);
222 break;
223 default:
224 die("unsupported file type %s", read_from);
225 }
226 }
227 else {
228 /* Reading from stdin */
229 mode = 0;
230 if (strbuf_read(&buf, 0, 0) < 0)
231 die_errno("failed to read from stdin");
232 }
82b474e0 233 convert_to_git(&the_index, path, buf.buf, buf.len, &buf, 0);
072bf432
JS
234 origin->file.ptr = buf.buf;
235 origin->file.size = buf.len;
829e5c3b 236 pretend_object_file(buf.buf, buf.len, OBJ_BLOB, &origin->blob_oid);
072bf432
JS
237
238 /*
239 * Read the current index, replace the path entry with
240 * origin->blob_sha1 without mucking with its mode or type
241 * bits; we are not going to write this index out -- we just
242 * want to run "diff-index --cached".
243 */
244 discard_cache();
245 read_cache();
246
247 len = strlen(path);
248 if (!mode) {
249 int pos = cache_name_pos(path, len);
250 if (0 <= pos)
251 mode = active_cache[pos]->ce_mode;
252 else
253 /* Let's not bother reading from HEAD tree */
254 mode = S_IFREG | 0644;
255 }
256 size = cache_entry_size(len);
257 ce = xcalloc(1, size);
258 oidcpy(&ce->oid, &origin->blob_oid);
259 memcpy(ce->name, path, len);
260 ce->ce_flags = create_ce_flags(0);
261 ce->ce_namelen = len;
262 ce->ce_mode = create_ce_mode(mode);
263 add_cache_entry(ce, ADD_CACHE_OK_TO_ADD|ADD_CACHE_OK_TO_REPLACE);
264
265 cache_tree_invalidate_path(&the_index, path);
266
267 return commit;
268}
b543bb1c
JS
269
270
271
272static int diff_hunks(mmfile_t *file_a, mmfile_t *file_b,
273 xdl_emit_hunk_consume_func_t hunk_func, void *cb_data, int xdl_opts)
274{
275 xpparam_t xpp = {0};
276 xdemitconf_t xecfg = {0};
277 xdemitcb_t ecb = {NULL};
278
279 xpp.flags = xdl_opts;
280 xecfg.hunk_func = hunk_func;
281 ecb.priv = cb_data;
282 return xdi_diff(file_a, file_b, &xpp, &xecfg, &ecb);
283}
284
285/*
286 * Given an origin, prepare mmfile_t structure to be used by the
287 * diff machinery
288 */
289static void fill_origin_blob(struct diff_options *opt,
290 struct blame_origin *o, mmfile_t *file, int *num_read_blob)
291{
292 if (!o->file.ptr) {
293 enum object_type type;
294 unsigned long file_size;
295
296 (*num_read_blob)++;
0d1e0e78 297 if (opt->flags.allow_textconv &&
b543bb1c
JS
298 textconv_object(o->path, o->mode, &o->blob_oid, 1, &file->ptr, &file_size))
299 ;
300 else
b4f5aca4 301 file->ptr = read_object_file(&o->blob_oid, &type,
302 &file_size);
b543bb1c
JS
303 file->size = file_size;
304
305 if (!file->ptr)
306 die("Cannot read blob %s for path %s",
307 oid_to_hex(&o->blob_oid),
308 o->path);
309 o->file = *file;
310 }
311 else
312 *file = o->file;
313}
314
315static void drop_origin_blob(struct blame_origin *o)
316{
317 if (o->file.ptr) {
e140f7af 318 FREE_AND_NULL(o->file.ptr);
b543bb1c
JS
319 }
320}
321
322/*
323 * Any merge of blames happens on lists of blames that arrived via
324 * different parents in a single suspect. In this case, we want to
325 * sort according to the suspect line numbers as opposed to the final
326 * image line numbers. The function body is somewhat longish because
327 * it avoids unnecessary writes.
328 */
329
330static struct blame_entry *blame_merge(struct blame_entry *list1,
331 struct blame_entry *list2)
332{
333 struct blame_entry *p1 = list1, *p2 = list2,
334 **tail = &list1;
335
336 if (!p1)
337 return p2;
338 if (!p2)
339 return p1;
340
341 if (p1->s_lno <= p2->s_lno) {
342 do {
343 tail = &p1->next;
344 if ((p1 = *tail) == NULL) {
345 *tail = p2;
346 return list1;
347 }
348 } while (p1->s_lno <= p2->s_lno);
349 }
350 for (;;) {
351 *tail = p2;
352 do {
353 tail = &p2->next;
354 if ((p2 = *tail) == NULL) {
355 *tail = p1;
356 return list1;
357 }
358 } while (p1->s_lno > p2->s_lno);
359 *tail = p1;
360 do {
361 tail = &p1->next;
362 if ((p1 = *tail) == NULL) {
363 *tail = p2;
364 return list1;
365 }
366 } while (p1->s_lno <= p2->s_lno);
367 }
368}
369
370static void *get_next_blame(const void *p)
371{
372 return ((struct blame_entry *)p)->next;
373}
374
375static void set_next_blame(void *p1, void *p2)
376{
377 ((struct blame_entry *)p1)->next = p2;
378}
379
380/*
381 * Final image line numbers are all different, so we don't need a
382 * three-way comparison here.
383 */
384
385static int compare_blame_final(const void *p1, const void *p2)
386{
387 return ((struct blame_entry *)p1)->lno > ((struct blame_entry *)p2)->lno
388 ? 1 : -1;
389}
390
391static int compare_blame_suspect(const void *p1, const void *p2)
392{
393 const struct blame_entry *s1 = p1, *s2 = p2;
394 /*
395 * to allow for collating suspects, we sort according to the
396 * respective pointer value as the primary sorting criterion.
397 * The actual relation is pretty unimportant as long as it
398 * establishes a total order. Comparing as integers gives us
399 * that.
400 */
401 if (s1->suspect != s2->suspect)
402 return (intptr_t)s1->suspect > (intptr_t)s2->suspect ? 1 : -1;
403 if (s1->s_lno == s2->s_lno)
404 return 0;
405 return s1->s_lno > s2->s_lno ? 1 : -1;
406}
407
408void blame_sort_final(struct blame_scoreboard *sb)
409{
410 sb->ent = llist_mergesort(sb->ent, get_next_blame, set_next_blame,
411 compare_blame_final);
412}
413
09002f1b
JS
414static int compare_commits_by_reverse_commit_date(const void *a,
415 const void *b,
416 void *c)
417{
418 return -compare_commits_by_commit_date(a, b, c);
419}
420
b543bb1c
JS
421/*
422 * For debugging -- origin is refcounted, and this asserts that
423 * we do not underflow.
424 */
425static void sanity_check_refcnt(struct blame_scoreboard *sb)
426{
427 int baa = 0;
428 struct blame_entry *ent;
429
430 for (ent = sb->ent; ent; ent = ent->next) {
431 /* Nobody should have zero or negative refcnt */
432 if (ent->suspect->refcnt <= 0) {
433 fprintf(stderr, "%s in %s has negative refcnt %d\n",
434 ent->suspect->path,
435 oid_to_hex(&ent->suspect->commit->object.oid),
436 ent->suspect->refcnt);
437 baa = 1;
438 }
439 }
440 if (baa)
441 sb->on_sanity_fail(sb, baa);
442}
443
444/*
445 * If two blame entries that are next to each other came from
446 * contiguous lines in the same origin (i.e. <commit, path> pair),
447 * merge them together.
448 */
449void blame_coalesce(struct blame_scoreboard *sb)
450{
451 struct blame_entry *ent, *next;
452
453 for (ent = sb->ent; ent && (next = ent->next); ent = next) {
454 if (ent->suspect == next->suspect &&
455 ent->s_lno + ent->num_lines == next->s_lno) {
456 ent->num_lines += next->num_lines;
457 ent->next = next->next;
458 blame_origin_decref(next->suspect);
459 free(next);
460 ent->score = 0;
461 next = ent; /* again */
462 }
463 }
464
465 if (sb->debug) /* sanity */
466 sanity_check_refcnt(sb);
467}
468
469/*
470 * Merge the given sorted list of blames into a preexisting origin.
471 * If there were no previous blames to that commit, it is entered into
472 * the commit priority queue of the score board.
473 */
474
475static void queue_blames(struct blame_scoreboard *sb, struct blame_origin *porigin,
476 struct blame_entry *sorted)
477{
478 if (porigin->suspects)
479 porigin->suspects = blame_merge(porigin->suspects, sorted);
480 else {
481 struct blame_origin *o;
482 for (o = porigin->commit->util; o; o = o->next) {
483 if (o->suspects) {
484 porigin->suspects = sorted;
485 return;
486 }
487 }
488 porigin->suspects = sorted;
489 prio_queue_put(&sb->commits, porigin->commit);
490 }
491}
492
09002f1b
JS
493/*
494 * Fill the blob_sha1 field of an origin if it hasn't, so that later
495 * call to fill_origin_blob() can use it to locate the data. blob_sha1
496 * for an origin is also used to pass the blame for the entire file to
497 * the parent to detect the case where a child's blob is identical to
498 * that of its parent's.
499 *
500 * This also fills origin->mode for corresponding tree path.
501 */
502static int fill_blob_sha1_and_mode(struct blame_origin *origin)
503{
504 if (!is_null_oid(&origin->blob_oid))
505 return 0;
916bc35b 506 if (get_tree_entry(&origin->commit->object.oid, origin->path, &origin->blob_oid, &origin->mode))
09002f1b 507 goto error_out;
0df8e965 508 if (oid_object_info(the_repository, &origin->blob_oid, NULL) != OBJ_BLOB)
09002f1b
JS
509 goto error_out;
510 return 0;
511 error_out:
512 oidclr(&origin->blob_oid);
513 origin->mode = S_IFINVALID;
514 return -1;
515}
516
b543bb1c
JS
517/*
518 * We have an origin -- check if the same path exists in the
519 * parent and return an origin structure to represent it.
520 */
521static struct blame_origin *find_origin(struct commit *parent,
522 struct blame_origin *origin)
523{
524 struct blame_origin *porigin;
525 struct diff_options diff_opts;
526 const char *paths[2];
527
528 /* First check any existing origins */
529 for (porigin = parent->util; porigin; porigin = porigin->next)
530 if (!strcmp(porigin->path, origin->path)) {
531 /*
532 * The same path between origin and its parent
533 * without renaming -- the most common case.
534 */
535 return blame_origin_incref (porigin);
536 }
537
538 /* See if the origin->path is different between parent
539 * and origin first. Most of the time they are the
540 * same and diff-tree is fairly efficient about this.
541 */
542 diff_setup(&diff_opts);
0d1e0e78 543 diff_opts.flags.recursive = 1;
b543bb1c
JS
544 diff_opts.detect_rename = 0;
545 diff_opts.output_format = DIFF_FORMAT_NO_OUTPUT;
546 paths[0] = origin->path;
547 paths[1] = NULL;
548
549 parse_pathspec(&diff_opts.pathspec,
550 PATHSPEC_ALL_MAGIC & ~PATHSPEC_LITERAL,
551 PATHSPEC_LITERAL_PATH, "", paths);
552 diff_setup_done(&diff_opts);
553
554 if (is_null_oid(&origin->commit->object.oid))
583c6a22 555 do_diff_cache(&parent->tree->object.oid, &diff_opts);
b543bb1c 556 else
a6f38c10
JH
557 diff_tree_oid(&parent->tree->object.oid,
558 &origin->commit->tree->object.oid,
559 "", &diff_opts);
b543bb1c
JS
560 diffcore_std(&diff_opts);
561
562 if (!diff_queued_diff.nr) {
563 /* The path is the same as parent */
564 porigin = get_origin(parent, origin->path);
565 oidcpy(&porigin->blob_oid, &origin->blob_oid);
566 porigin->mode = origin->mode;
567 } else {
568 /*
569 * Since origin->path is a pathspec, if the parent
570 * commit had it as a directory, we will see a whole
571 * bunch of deletion of files in the directory that we
572 * do not care about.
573 */
574 int i;
575 struct diff_filepair *p = NULL;
576 for (i = 0; i < diff_queued_diff.nr; i++) {
577 const char *name;
578 p = diff_queued_diff.queue[i];
579 name = p->one->path ? p->one->path : p->two->path;
580 if (!strcmp(name, origin->path))
581 break;
582 }
583 if (!p)
584 die("internal error in blame::find_origin");
585 switch (p->status) {
586 default:
587 die("internal error in blame::find_origin (%c)",
588 p->status);
589 case 'M':
590 porigin = get_origin(parent, origin->path);
591 oidcpy(&porigin->blob_oid, &p->one->oid);
592 porigin->mode = p->one->mode;
593 break;
594 case 'A':
595 case 'T':
596 /* Did not exist in parent, or type changed */
597 break;
598 }
599 }
600 diff_flush(&diff_opts);
601 clear_pathspec(&diff_opts.pathspec);
602 return porigin;
603}
604
605/*
606 * We have an origin -- find the path that corresponds to it in its
607 * parent and return an origin structure to represent it.
608 */
609static struct blame_origin *find_rename(struct commit *parent,
610 struct blame_origin *origin)
611{
612 struct blame_origin *porigin = NULL;
613 struct diff_options diff_opts;
614 int i;
615
616 diff_setup(&diff_opts);
0d1e0e78 617 diff_opts.flags.recursive = 1;
b543bb1c
JS
618 diff_opts.detect_rename = DIFF_DETECT_RENAME;
619 diff_opts.output_format = DIFF_FORMAT_NO_OUTPUT;
620 diff_opts.single_follow = origin->path;
621 diff_setup_done(&diff_opts);
622
623 if (is_null_oid(&origin->commit->object.oid))
583c6a22 624 do_diff_cache(&parent->tree->object.oid, &diff_opts);
b543bb1c 625 else
a6f38c10
JH
626 diff_tree_oid(&parent->tree->object.oid,
627 &origin->commit->tree->object.oid,
628 "", &diff_opts);
b543bb1c
JS
629 diffcore_std(&diff_opts);
630
631 for (i = 0; i < diff_queued_diff.nr; i++) {
632 struct diff_filepair *p = diff_queued_diff.queue[i];
633 if ((p->status == 'R' || p->status == 'C') &&
634 !strcmp(p->two->path, origin->path)) {
635 porigin = get_origin(parent, p->one->path);
636 oidcpy(&porigin->blob_oid, &p->one->oid);
637 porigin->mode = p->one->mode;
638 break;
639 }
640 }
641 diff_flush(&diff_opts);
642 clear_pathspec(&diff_opts.pathspec);
643 return porigin;
644}
645
646/*
647 * Append a new blame entry to a given output queue.
648 */
649static void add_blame_entry(struct blame_entry ***queue,
650 const struct blame_entry *src)
651{
652 struct blame_entry *e = xmalloc(sizeof(*e));
653 memcpy(e, src, sizeof(*e));
654 blame_origin_incref(e->suspect);
655
656 e->next = **queue;
657 **queue = e;
658 *queue = &e->next;
659}
660
661/*
662 * src typically is on-stack; we want to copy the information in it to
663 * a malloced blame_entry that gets added to the given queue. The
664 * origin of dst loses a refcnt.
665 */
666static void dup_entry(struct blame_entry ***queue,
667 struct blame_entry *dst, struct blame_entry *src)
668{
669 blame_origin_incref(src->suspect);
670 blame_origin_decref(dst->suspect);
671 memcpy(dst, src, sizeof(*src));
672 dst->next = **queue;
673 **queue = dst;
674 *queue = &dst->next;
675}
676
677const char *blame_nth_line(struct blame_scoreboard *sb, long lno)
678{
679 return sb->final_buf + sb->lineno[lno];
680}
681
682/*
683 * It is known that lines between tlno to same came from parent, and e
684 * has an overlap with that range. it also is known that parent's
685 * line plno corresponds to e's line tlno.
686 *
687 * <---- e ----->
688 * <------>
689 * <------------>
690 * <------------>
691 * <------------------>
692 *
693 * Split e into potentially three parts; before this chunk, the chunk
694 * to be blamed for the parent, and after that portion.
695 */
696static void split_overlap(struct blame_entry *split,
697 struct blame_entry *e,
698 int tlno, int plno, int same,
699 struct blame_origin *parent)
700{
701 int chunk_end_lno;
702 memset(split, 0, sizeof(struct blame_entry [3]));
703
704 if (e->s_lno < tlno) {
705 /* there is a pre-chunk part not blamed on parent */
706 split[0].suspect = blame_origin_incref(e->suspect);
707 split[0].lno = e->lno;
708 split[0].s_lno = e->s_lno;
709 split[0].num_lines = tlno - e->s_lno;
710 split[1].lno = e->lno + tlno - e->s_lno;
711 split[1].s_lno = plno;
712 }
713 else {
714 split[1].lno = e->lno;
715 split[1].s_lno = plno + (e->s_lno - tlno);
716 }
717
718 if (same < e->s_lno + e->num_lines) {
719 /* there is a post-chunk part not blamed on parent */
720 split[2].suspect = blame_origin_incref(e->suspect);
721 split[2].lno = e->lno + (same - e->s_lno);
722 split[2].s_lno = e->s_lno + (same - e->s_lno);
723 split[2].num_lines = e->s_lno + e->num_lines - same;
724 chunk_end_lno = split[2].lno;
725 }
726 else
727 chunk_end_lno = e->lno + e->num_lines;
728 split[1].num_lines = chunk_end_lno - split[1].lno;
729
730 /*
731 * if it turns out there is nothing to blame the parent for,
732 * forget about the splitting. !split[1].suspect signals this.
733 */
734 if (split[1].num_lines < 1)
735 return;
736 split[1].suspect = blame_origin_incref(parent);
737}
738
739/*
740 * split_overlap() divided an existing blame e into up to three parts
741 * in split. Any assigned blame is moved to queue to
742 * reflect the split.
743 */
744static void split_blame(struct blame_entry ***blamed,
745 struct blame_entry ***unblamed,
746 struct blame_entry *split,
747 struct blame_entry *e)
748{
749 if (split[0].suspect && split[2].suspect) {
750 /* The first part (reuse storage for the existing entry e) */
751 dup_entry(unblamed, e, &split[0]);
752
753 /* The last part -- me */
754 add_blame_entry(unblamed, &split[2]);
755
756 /* ... and the middle part -- parent */
757 add_blame_entry(blamed, &split[1]);
758 }
759 else if (!split[0].suspect && !split[2].suspect)
760 /*
761 * The parent covers the entire area; reuse storage for
762 * e and replace it with the parent.
763 */
764 dup_entry(blamed, e, &split[1]);
765 else if (split[0].suspect) {
766 /* me and then parent */
767 dup_entry(unblamed, e, &split[0]);
768 add_blame_entry(blamed, &split[1]);
769 }
770 else {
771 /* parent and then me */
772 dup_entry(blamed, e, &split[1]);
773 add_blame_entry(unblamed, &split[2]);
774 }
775}
776
777/*
778 * After splitting the blame, the origins used by the
779 * on-stack blame_entry should lose one refcnt each.
780 */
781static void decref_split(struct blame_entry *split)
782{
783 int i;
784
785 for (i = 0; i < 3; i++)
786 blame_origin_decref(split[i].suspect);
787}
788
789/*
790 * reverse_blame reverses the list given in head, appending tail.
791 * That allows us to build lists in reverse order, then reverse them
792 * afterwards. This can be faster than building the list in proper
793 * order right away. The reason is that building in proper order
794 * requires writing a link in the _previous_ element, while building
795 * in reverse order just requires placing the list head into the
796 * _current_ element.
797 */
798
799static struct blame_entry *reverse_blame(struct blame_entry *head,
800 struct blame_entry *tail)
801{
802 while (head) {
803 struct blame_entry *next = head->next;
804 head->next = tail;
805 tail = head;
806 head = next;
807 }
808 return tail;
809}
810
811/*
812 * Process one hunk from the patch between the current suspect for
813 * blame_entry e and its parent. This first blames any unfinished
814 * entries before the chunk (which is where target and parent start
815 * differing) on the parent, and then splits blame entries at the
816 * start and at the end of the difference region. Since use of -M and
817 * -C options may lead to overlapping/duplicate source line number
818 * ranges, all we can rely on from sorting/merging is the order of the
819 * first suspect line number.
820 */
821static void blame_chunk(struct blame_entry ***dstq, struct blame_entry ***srcq,
822 int tlno, int offset, int same,
823 struct blame_origin *parent)
824{
825 struct blame_entry *e = **srcq;
826 struct blame_entry *samep = NULL, *diffp = NULL;
827
828 while (e && e->s_lno < tlno) {
829 struct blame_entry *next = e->next;
830 /*
831 * current record starts before differing portion. If
832 * it reaches into it, we need to split it up and
833 * examine the second part separately.
834 */
835 if (e->s_lno + e->num_lines > tlno) {
836 /* Move second half to a new record */
837 int len = tlno - e->s_lno;
838 struct blame_entry *n = xcalloc(1, sizeof (struct blame_entry));
839 n->suspect = e->suspect;
840 n->lno = e->lno + len;
841 n->s_lno = e->s_lno + len;
842 n->num_lines = e->num_lines - len;
843 e->num_lines = len;
844 e->score = 0;
845 /* Push new record to diffp */
846 n->next = diffp;
847 diffp = n;
848 } else
849 blame_origin_decref(e->suspect);
850 /* Pass blame for everything before the differing
851 * chunk to the parent */
852 e->suspect = blame_origin_incref(parent);
853 e->s_lno += offset;
854 e->next = samep;
855 samep = e;
856 e = next;
857 }
858 /*
859 * As we don't know how much of a common stretch after this
860 * diff will occur, the currently blamed parts are all that we
861 * can assign to the parent for now.
862 */
863
864 if (samep) {
865 **dstq = reverse_blame(samep, **dstq);
866 *dstq = &samep->next;
867 }
868 /*
869 * Prepend the split off portions: everything after e starts
870 * after the blameable portion.
871 */
872 e = reverse_blame(diffp, e);
873
874 /*
875 * Now retain records on the target while parts are different
876 * from the parent.
877 */
878 samep = NULL;
879 diffp = NULL;
880 while (e && e->s_lno < same) {
881 struct blame_entry *next = e->next;
882
883 /*
884 * If current record extends into sameness, need to split.
885 */
886 if (e->s_lno + e->num_lines > same) {
887 /*
888 * Move second half to a new record to be
889 * processed by later chunks
890 */
891 int len = same - e->s_lno;
892 struct blame_entry *n = xcalloc(1, sizeof (struct blame_entry));
893 n->suspect = blame_origin_incref(e->suspect);
894 n->lno = e->lno + len;
895 n->s_lno = e->s_lno + len;
896 n->num_lines = e->num_lines - len;
897 e->num_lines = len;
898 e->score = 0;
899 /* Push new record to samep */
900 n->next = samep;
901 samep = n;
902 }
903 e->next = diffp;
904 diffp = e;
905 e = next;
906 }
907 **srcq = reverse_blame(diffp, reverse_blame(samep, e));
908 /* Move across elements that are in the unblamable portion */
909 if (diffp)
910 *srcq = &diffp->next;
911}
912
913struct blame_chunk_cb_data {
914 struct blame_origin *parent;
915 long offset;
916 struct blame_entry **dstq;
917 struct blame_entry **srcq;
918};
919
920/* diff chunks are from parent to target */
921static int blame_chunk_cb(long start_a, long count_a,
922 long start_b, long count_b, void *data)
923{
924 struct blame_chunk_cb_data *d = data;
925 if (start_a - start_b != d->offset)
926 die("internal error in blame::blame_chunk_cb");
927 blame_chunk(&d->dstq, &d->srcq, start_b, start_a - start_b,
928 start_b + count_b, d->parent);
929 d->offset = start_a + count_a - (start_b + count_b);
930 return 0;
931}
932
933/*
934 * We are looking at the origin 'target' and aiming to pass blame
935 * for the lines it is suspected to its parent. Run diff to find
936 * which lines came from parent and pass blame for them.
937 */
938static void pass_blame_to_parent(struct blame_scoreboard *sb,
939 struct blame_origin *target,
940 struct blame_origin *parent)
941{
942 mmfile_t file_p, file_o;
943 struct blame_chunk_cb_data d;
944 struct blame_entry *newdest = NULL;
945
946 if (!target->suspects)
947 return; /* nothing remains for this target */
948
949 d.parent = parent;
950 d.offset = 0;
951 d.dstq = &newdest; d.srcq = &target->suspects;
952
953 fill_origin_blob(&sb->revs->diffopt, parent, &file_p, &sb->num_read_blob);
954 fill_origin_blob(&sb->revs->diffopt, target, &file_o, &sb->num_read_blob);
955 sb->num_get_patch++;
956
957 if (diff_hunks(&file_p, &file_o, blame_chunk_cb, &d, sb->xdl_opts))
958 die("unable to generate diff (%s -> %s)",
959 oid_to_hex(&parent->commit->object.oid),
960 oid_to_hex(&target->commit->object.oid));
961 /* The rest are the same as the parent */
962 blame_chunk(&d.dstq, &d.srcq, INT_MAX, d.offset, INT_MAX, parent);
963 *d.dstq = NULL;
964 queue_blames(sb, parent, newdest);
965
966 return;
967}
968
969/*
970 * The lines in blame_entry after splitting blames many times can become
971 * very small and trivial, and at some point it becomes pointless to
972 * blame the parents. E.g. "\t\t}\n\t}\n\n" appears everywhere in any
973 * ordinary C program, and it is not worth to say it was copied from
974 * totally unrelated file in the parent.
975 *
976 * Compute how trivial the lines in the blame_entry are.
977 */
978unsigned blame_entry_score(struct blame_scoreboard *sb, struct blame_entry *e)
979{
980 unsigned score;
981 const char *cp, *ep;
982
983 if (e->score)
984 return e->score;
985
986 score = 1;
987 cp = blame_nth_line(sb, e->lno);
988 ep = blame_nth_line(sb, e->lno + e->num_lines);
989 while (cp < ep) {
990 unsigned ch = *((unsigned char *)cp);
991 if (isalnum(ch))
992 score++;
993 cp++;
994 }
995 e->score = score;
996 return score;
997}
998
999/*
abeacb25
BW
1000 * best_so_far[] and potential[] are both a split of an existing blame_entry
1001 * that passes blame to the parent. Maintain best_so_far the best split so
1002 * far, by comparing potential and best_so_far and copying potential into
b543bb1c
JS
1003 * bst_so_far as needed.
1004 */
1005static void copy_split_if_better(struct blame_scoreboard *sb,
1006 struct blame_entry *best_so_far,
abeacb25 1007 struct blame_entry *potential)
b543bb1c
JS
1008{
1009 int i;
1010
abeacb25 1011 if (!potential[1].suspect)
b543bb1c
JS
1012 return;
1013 if (best_so_far[1].suspect) {
abeacb25
BW
1014 if (blame_entry_score(sb, &potential[1]) <
1015 blame_entry_score(sb, &best_so_far[1]))
b543bb1c
JS
1016 return;
1017 }
1018
1019 for (i = 0; i < 3; i++)
abeacb25 1020 blame_origin_incref(potential[i].suspect);
b543bb1c 1021 decref_split(best_so_far);
abeacb25 1022 memcpy(best_so_far, potential, sizeof(struct blame_entry[3]));
b543bb1c
JS
1023}
1024
1025/*
1026 * We are looking at a part of the final image represented by
1027 * ent (tlno and same are offset by ent->s_lno).
1028 * tlno is where we are looking at in the final image.
1029 * up to (but not including) same match preimage.
1030 * plno is where we are looking at in the preimage.
1031 *
1032 * <-------------- final image ---------------------->
1033 * <------ent------>
1034 * ^tlno ^same
1035 * <---------preimage----->
1036 * ^plno
1037 *
1038 * All line numbers are 0-based.
1039 */
1040static void handle_split(struct blame_scoreboard *sb,
1041 struct blame_entry *ent,
1042 int tlno, int plno, int same,
1043 struct blame_origin *parent,
1044 struct blame_entry *split)
1045{
1046 if (ent->num_lines <= tlno)
1047 return;
1048 if (tlno < same) {
abeacb25 1049 struct blame_entry potential[3];
b543bb1c
JS
1050 tlno += ent->s_lno;
1051 same += ent->s_lno;
abeacb25
BW
1052 split_overlap(potential, ent, tlno, plno, same, parent);
1053 copy_split_if_better(sb, split, potential);
1054 decref_split(potential);
b543bb1c
JS
1055 }
1056}
1057
1058struct handle_split_cb_data {
1059 struct blame_scoreboard *sb;
1060 struct blame_entry *ent;
1061 struct blame_origin *parent;
1062 struct blame_entry *split;
1063 long plno;
1064 long tlno;
1065};
1066
1067static int handle_split_cb(long start_a, long count_a,
1068 long start_b, long count_b, void *data)
1069{
1070 struct handle_split_cb_data *d = data;
1071 handle_split(d->sb, d->ent, d->tlno, d->plno, start_b, d->parent,
1072 d->split);
1073 d->plno = start_a + count_a;
1074 d->tlno = start_b + count_b;
1075 return 0;
1076}
1077
1078/*
1079 * Find the lines from parent that are the same as ent so that
1080 * we can pass blames to it. file_p has the blob contents for
1081 * the parent.
1082 */
1083static void find_copy_in_blob(struct blame_scoreboard *sb,
1084 struct blame_entry *ent,
1085 struct blame_origin *parent,
1086 struct blame_entry *split,
1087 mmfile_t *file_p)
1088{
1089 const char *cp;
1090 mmfile_t file_o;
1091 struct handle_split_cb_data d;
1092
1093 memset(&d, 0, sizeof(d));
1094 d.sb = sb; d.ent = ent; d.parent = parent; d.split = split;
1095 /*
1096 * Prepare mmfile that contains only the lines in ent.
1097 */
1098 cp = blame_nth_line(sb, ent->lno);
1099 file_o.ptr = (char *) cp;
1100 file_o.size = blame_nth_line(sb, ent->lno + ent->num_lines) - cp;
1101
1102 /*
1103 * file_o is a part of final image we are annotating.
1104 * file_p partially may match that image.
1105 */
1106 memset(split, 0, sizeof(struct blame_entry [3]));
1107 if (diff_hunks(file_p, &file_o, handle_split_cb, &d, sb->xdl_opts))
1108 die("unable to generate diff (%s)",
1109 oid_to_hex(&parent->commit->object.oid));
1110 /* remainder, if any, all match the preimage */
1111 handle_split(sb, ent, d.tlno, d.plno, ent->num_lines, parent, split);
1112}
1113
1114/* Move all blame entries from list *source that have a score smaller
1115 * than score_min to the front of list *small.
1116 * Returns a pointer to the link pointing to the old head of the small list.
1117 */
1118
1119static struct blame_entry **filter_small(struct blame_scoreboard *sb,
1120 struct blame_entry **small,
1121 struct blame_entry **source,
1122 unsigned score_min)
1123{
1124 struct blame_entry *p = *source;
1125 struct blame_entry *oldsmall = *small;
1126 while (p) {
1127 if (blame_entry_score(sb, p) <= score_min) {
1128 *small = p;
1129 small = &p->next;
1130 p = *small;
1131 } else {
1132 *source = p;
1133 source = &p->next;
1134 p = *source;
1135 }
1136 }
1137 *small = oldsmall;
1138 *source = NULL;
1139 return small;
1140}
1141
1142/*
1143 * See if lines currently target is suspected for can be attributed to
1144 * parent.
1145 */
1146static void find_move_in_parent(struct blame_scoreboard *sb,
1147 struct blame_entry ***blamed,
1148 struct blame_entry **toosmall,
1149 struct blame_origin *target,
1150 struct blame_origin *parent)
1151{
1152 struct blame_entry *e, split[3];
1153 struct blame_entry *unblamed = target->suspects;
1154 struct blame_entry *leftover = NULL;
1155 mmfile_t file_p;
1156
1157 if (!unblamed)
1158 return; /* nothing remains for this target */
1159
1160 fill_origin_blob(&sb->revs->diffopt, parent, &file_p, &sb->num_read_blob);
1161 if (!file_p.ptr)
1162 return;
1163
1164 /* At each iteration, unblamed has a NULL-terminated list of
1165 * entries that have not yet been tested for blame. leftover
1166 * contains the reversed list of entries that have been tested
1167 * without being assignable to the parent.
1168 */
1169 do {
1170 struct blame_entry **unblamedtail = &unblamed;
1171 struct blame_entry *next;
1172 for (e = unblamed; e; e = next) {
1173 next = e->next;
1174 find_copy_in_blob(sb, e, parent, split, &file_p);
1175 if (split[1].suspect &&
1176 sb->move_score < blame_entry_score(sb, &split[1])) {
1177 split_blame(blamed, &unblamedtail, split, e);
1178 } else {
1179 e->next = leftover;
1180 leftover = e;
1181 }
1182 decref_split(split);
1183 }
1184 *unblamedtail = NULL;
1185 toosmall = filter_small(sb, toosmall, &unblamed, sb->move_score);
1186 } while (unblamed);
1187 target->suspects = reverse_blame(leftover, NULL);
1188}
1189
1190struct blame_list {
1191 struct blame_entry *ent;
1192 struct blame_entry split[3];
1193};
1194
1195/*
1196 * Count the number of entries the target is suspected for,
1197 * and prepare a list of entry and the best split.
1198 */
1199static struct blame_list *setup_blame_list(struct blame_entry *unblamed,
1200 int *num_ents_p)
1201{
1202 struct blame_entry *e;
1203 int num_ents, i;
1204 struct blame_list *blame_list = NULL;
1205
1206 for (e = unblamed, num_ents = 0; e; e = e->next)
1207 num_ents++;
1208 if (num_ents) {
1209 blame_list = xcalloc(num_ents, sizeof(struct blame_list));
1210 for (e = unblamed, i = 0; e; e = e->next)
1211 blame_list[i++].ent = e;
1212 }
1213 *num_ents_p = num_ents;
1214 return blame_list;
1215}
1216
1217/*
1218 * For lines target is suspected for, see if we can find code movement
1219 * across file boundary from the parent commit. porigin is the path
1220 * in the parent we already tried.
1221 */
1222static void find_copy_in_parent(struct blame_scoreboard *sb,
1223 struct blame_entry ***blamed,
1224 struct blame_entry **toosmall,
1225 struct blame_origin *target,
1226 struct commit *parent,
1227 struct blame_origin *porigin,
1228 int opt)
1229{
1230 struct diff_options diff_opts;
1231 int i, j;
1232 struct blame_list *blame_list;
1233 int num_ents;
1234 struct blame_entry *unblamed = target->suspects;
1235 struct blame_entry *leftover = NULL;
1236
1237 if (!unblamed)
1238 return; /* nothing remains for this target */
1239
1240 diff_setup(&diff_opts);
0d1e0e78 1241 diff_opts.flags.recursive = 1;
b543bb1c
JS
1242 diff_opts.output_format = DIFF_FORMAT_NO_OUTPUT;
1243
1244 diff_setup_done(&diff_opts);
1245
1246 /* Try "find copies harder" on new path if requested;
1247 * we do not want to use diffcore_rename() actually to
1248 * match things up; find_copies_harder is set only to
a6f38c10 1249 * force diff_tree_oid() to feed all filepairs to diff_queue,
b543bb1c
JS
1250 * and this code needs to be after diff_setup_done(), which
1251 * usually makes find-copies-harder imply copy detection.
1252 */
1253 if ((opt & PICKAXE_BLAME_COPY_HARDEST)
1254 || ((opt & PICKAXE_BLAME_COPY_HARDER)
1255 && (!porigin || strcmp(target->path, porigin->path))))
0d1e0e78 1256 diff_opts.flags.find_copies_harder = 1;
b543bb1c
JS
1257
1258 if (is_null_oid(&target->commit->object.oid))
583c6a22 1259 do_diff_cache(&parent->tree->object.oid, &diff_opts);
b543bb1c 1260 else
a6f38c10
JH
1261 diff_tree_oid(&parent->tree->object.oid,
1262 &target->commit->tree->object.oid,
1263 "", &diff_opts);
b543bb1c 1264
0d1e0e78 1265 if (!diff_opts.flags.find_copies_harder)
b543bb1c
JS
1266 diffcore_std(&diff_opts);
1267
1268 do {
1269 struct blame_entry **unblamedtail = &unblamed;
1270 blame_list = setup_blame_list(unblamed, &num_ents);
1271
1272 for (i = 0; i < diff_queued_diff.nr; i++) {
1273 struct diff_filepair *p = diff_queued_diff.queue[i];
1274 struct blame_origin *norigin;
1275 mmfile_t file_p;
abeacb25 1276 struct blame_entry potential[3];
b543bb1c
JS
1277
1278 if (!DIFF_FILE_VALID(p->one))
1279 continue; /* does not exist in parent */
1280 if (S_ISGITLINK(p->one->mode))
1281 continue; /* ignore git links */
1282 if (porigin && !strcmp(p->one->path, porigin->path))
1283 /* find_move already dealt with this path */
1284 continue;
1285
1286 norigin = get_origin(parent, p->one->path);
1287 oidcpy(&norigin->blob_oid, &p->one->oid);
1288 norigin->mode = p->one->mode;
1289 fill_origin_blob(&sb->revs->diffopt, norigin, &file_p, &sb->num_read_blob);
1290 if (!file_p.ptr)
1291 continue;
1292
1293 for (j = 0; j < num_ents; j++) {
1294 find_copy_in_blob(sb, blame_list[j].ent,
abeacb25 1295 norigin, potential, &file_p);
b543bb1c 1296 copy_split_if_better(sb, blame_list[j].split,
abeacb25
BW
1297 potential);
1298 decref_split(potential);
b543bb1c
JS
1299 }
1300 blame_origin_decref(norigin);
1301 }
1302
1303 for (j = 0; j < num_ents; j++) {
1304 struct blame_entry *split = blame_list[j].split;
1305 if (split[1].suspect &&
1306 sb->copy_score < blame_entry_score(sb, &split[1])) {
1307 split_blame(blamed, &unblamedtail, split,
1308 blame_list[j].ent);
1309 } else {
1310 blame_list[j].ent->next = leftover;
1311 leftover = blame_list[j].ent;
1312 }
1313 decref_split(split);
1314 }
1315 free(blame_list);
1316 *unblamedtail = NULL;
1317 toosmall = filter_small(sb, toosmall, &unblamed, sb->copy_score);
1318 } while (unblamed);
1319 target->suspects = reverse_blame(leftover, NULL);
1320 diff_flush(&diff_opts);
1321 clear_pathspec(&diff_opts.pathspec);
1322}
1323
1324/*
1325 * The blobs of origin and porigin exactly match, so everything
1326 * origin is suspected for can be blamed on the parent.
1327 */
1328static void pass_whole_blame(struct blame_scoreboard *sb,
1329 struct blame_origin *origin, struct blame_origin *porigin)
1330{
1331 struct blame_entry *e, *suspects;
1332
1333 if (!porigin->file.ptr && origin->file.ptr) {
1334 /* Steal its file */
1335 porigin->file = origin->file;
1336 origin->file.ptr = NULL;
1337 }
1338 suspects = origin->suspects;
1339 origin->suspects = NULL;
1340 for (e = suspects; e; e = e->next) {
1341 blame_origin_incref(porigin);
1342 blame_origin_decref(e->suspect);
1343 e->suspect = porigin;
1344 }
1345 queue_blames(sb, porigin, suspects);
1346}
1347
1348/*
1349 * We pass blame from the current commit to its parents. We keep saying
1350 * "parent" (and "porigin"), but what we mean is to find scapegoat to
1351 * exonerate ourselves.
1352 */
1353static struct commit_list *first_scapegoat(struct rev_info *revs, struct commit *commit,
1354 int reverse)
1355{
1356 if (!reverse) {
1357 if (revs->first_parent_only &&
1358 commit->parents &&
1359 commit->parents->next) {
1360 free_commit_list(commit->parents->next);
1361 commit->parents->next = NULL;
1362 }
1363 return commit->parents;
1364 }
1365 return lookup_decoration(&revs->children, &commit->object);
1366}
1367
1368static int num_scapegoats(struct rev_info *revs, struct commit *commit, int reverse)
1369{
1370 struct commit_list *l = first_scapegoat(revs, commit, reverse);
1371 return commit_list_count(l);
1372}
1373
1374/* Distribute collected unsorted blames to the respected sorted lists
1375 * in the various origins.
1376 */
1377static void distribute_blame(struct blame_scoreboard *sb, struct blame_entry *blamed)
1378{
1379 blamed = llist_mergesort(blamed, get_next_blame, set_next_blame,
1380 compare_blame_suspect);
1381 while (blamed)
1382 {
1383 struct blame_origin *porigin = blamed->suspect;
1384 struct blame_entry *suspects = NULL;
1385 do {
1386 struct blame_entry *next = blamed->next;
1387 blamed->next = suspects;
1388 suspects = blamed;
1389 blamed = next;
1390 } while (blamed && blamed->suspect == porigin);
1391 suspects = reverse_blame(suspects, NULL);
1392 queue_blames(sb, porigin, suspects);
1393 }
1394}
1395
1396#define MAXSG 16
1397
1398static void pass_blame(struct blame_scoreboard *sb, struct blame_origin *origin, int opt)
1399{
1400 struct rev_info *revs = sb->revs;
1401 int i, pass, num_sg;
1402 struct commit *commit = origin->commit;
1403 struct commit_list *sg;
1404 struct blame_origin *sg_buf[MAXSG];
1405 struct blame_origin *porigin, **sg_origin = sg_buf;
1406 struct blame_entry *toosmall = NULL;
1407 struct blame_entry *blames, **blametail = &blames;
1408
1409 num_sg = num_scapegoats(revs, commit, sb->reverse);
1410 if (!num_sg)
1411 goto finish;
1412 else if (num_sg < ARRAY_SIZE(sg_buf))
1413 memset(sg_buf, 0, sizeof(sg_buf));
1414 else
1415 sg_origin = xcalloc(num_sg, sizeof(*sg_origin));
1416
1417 /*
1418 * The first pass looks for unrenamed path to optimize for
1419 * common cases, then we look for renames in the second pass.
1420 */
1421 for (pass = 0; pass < 2 - sb->no_whole_file_rename; pass++) {
1422 struct blame_origin *(*find)(struct commit *, struct blame_origin *);
1423 find = pass ? find_rename : find_origin;
1424
1425 for (i = 0, sg = first_scapegoat(revs, commit, sb->reverse);
1426 i < num_sg && sg;
1427 sg = sg->next, i++) {
1428 struct commit *p = sg->item;
1429 int j, same;
1430
1431 if (sg_origin[i])
1432 continue;
1433 if (parse_commit(p))
1434 continue;
1435 porigin = find(p, origin);
1436 if (!porigin)
1437 continue;
1438 if (!oidcmp(&porigin->blob_oid, &origin->blob_oid)) {
1439 pass_whole_blame(sb, origin, porigin);
1440 blame_origin_decref(porigin);
1441 goto finish;
1442 }
1443 for (j = same = 0; j < i; j++)
1444 if (sg_origin[j] &&
1445 !oidcmp(&sg_origin[j]->blob_oid, &porigin->blob_oid)) {
1446 same = 1;
1447 break;
1448 }
1449 if (!same)
1450 sg_origin[i] = porigin;
1451 else
1452 blame_origin_decref(porigin);
1453 }
1454 }
1455
1456 sb->num_commits++;
1457 for (i = 0, sg = first_scapegoat(revs, commit, sb->reverse);
1458 i < num_sg && sg;
1459 sg = sg->next, i++) {
1460 struct blame_origin *porigin = sg_origin[i];
1461 if (!porigin)
1462 continue;
1463 if (!origin->previous) {
1464 blame_origin_incref(porigin);
1465 origin->previous = porigin;
1466 }
1467 pass_blame_to_parent(sb, origin, porigin);
1468 if (!origin->suspects)
1469 goto finish;
1470 }
1471
1472 /*
1473 * Optionally find moves in parents' files.
1474 */
1475 if (opt & PICKAXE_BLAME_MOVE) {
1476 filter_small(sb, &toosmall, &origin->suspects, sb->move_score);
1477 if (origin->suspects) {
1478 for (i = 0, sg = first_scapegoat(revs, commit, sb->reverse);
1479 i < num_sg && sg;
1480 sg = sg->next, i++) {
1481 struct blame_origin *porigin = sg_origin[i];
1482 if (!porigin)
1483 continue;
1484 find_move_in_parent(sb, &blametail, &toosmall, origin, porigin);
1485 if (!origin->suspects)
1486 break;
1487 }
1488 }
1489 }
1490
1491 /*
1492 * Optionally find copies from parents' files.
1493 */
1494 if (opt & PICKAXE_BLAME_COPY) {
1495 if (sb->copy_score > sb->move_score)
1496 filter_small(sb, &toosmall, &origin->suspects, sb->copy_score);
1497 else if (sb->copy_score < sb->move_score) {
1498 origin->suspects = blame_merge(origin->suspects, toosmall);
1499 toosmall = NULL;
1500 filter_small(sb, &toosmall, &origin->suspects, sb->copy_score);
1501 }
1502 if (!origin->suspects)
1503 goto finish;
1504
1505 for (i = 0, sg = first_scapegoat(revs, commit, sb->reverse);
1506 i < num_sg && sg;
1507 sg = sg->next, i++) {
1508 struct blame_origin *porigin = sg_origin[i];
1509 find_copy_in_parent(sb, &blametail, &toosmall,
1510 origin, sg->item, porigin, opt);
1511 if (!origin->suspects)
1512 goto finish;
1513 }
1514 }
1515
1516finish:
1517 *blametail = NULL;
1518 distribute_blame(sb, blames);
1519 /*
1520 * prepend toosmall to origin->suspects
1521 *
1522 * There is no point in sorting: this ends up on a big
1523 * unsorted list in the caller anyway.
1524 */
1525 if (toosmall) {
1526 struct blame_entry **tail = &toosmall;
1527 while (*tail)
1528 tail = &(*tail)->next;
1529 *tail = origin->suspects;
1530 origin->suspects = toosmall;
1531 }
1532 for (i = 0; i < num_sg; i++) {
1533 if (sg_origin[i]) {
1534 drop_origin_blob(sg_origin[i]);
1535 blame_origin_decref(sg_origin[i]);
1536 }
1537 }
1538 drop_origin_blob(origin);
1539 if (sg_buf != sg_origin)
1540 free(sg_origin);
1541}
1542
1543/*
1544 * The main loop -- while we have blobs with lines whose true origin
1545 * is still unknown, pick one blob, and allow its lines to pass blames
1546 * to its parents. */
1547void assign_blame(struct blame_scoreboard *sb, int opt)
1548{
1549 struct rev_info *revs = sb->revs;
1550 struct commit *commit = prio_queue_get(&sb->commits);
1551
1552 while (commit) {
1553 struct blame_entry *ent;
1554 struct blame_origin *suspect = commit->util;
1555
1556 /* find one suspect to break down */
1557 while (suspect && !suspect->suspects)
1558 suspect = suspect->next;
1559
1560 if (!suspect) {
1561 commit = prio_queue_get(&sb->commits);
1562 continue;
1563 }
1564
1565 assert(commit == suspect->commit);
1566
1567 /*
1568 * We will use this suspect later in the loop,
1569 * so hold onto it in the meantime.
1570 */
1571 blame_origin_incref(suspect);
1572 parse_commit(commit);
1573 if (sb->reverse ||
1574 (!(commit->object.flags & UNINTERESTING) &&
1575 !(revs->max_age != -1 && commit->date < revs->max_age)))
1576 pass_blame(sb, suspect, opt);
1577 else {
1578 commit->object.flags |= UNINTERESTING;
1579 if (commit->object.parsed)
1580 mark_parents_uninteresting(commit);
1581 }
1582 /* treat root commit as boundary */
1583 if (!commit->parents && !sb->show_root)
1584 commit->object.flags |= UNINTERESTING;
1585
1586 /* Take responsibility for the remaining entries */
1587 ent = suspect->suspects;
1588 if (ent) {
1589 suspect->guilty = 1;
1590 for (;;) {
1591 struct blame_entry *next = ent->next;
1592 if (sb->found_guilty_entry)
1593 sb->found_guilty_entry(ent, sb->found_guilty_entry_data);
1594 if (next) {
1595 ent = next;
1596 continue;
1597 }
1598 ent->next = sb->ent;
1599 sb->ent = suspect->suspects;
1600 suspect->suspects = NULL;
1601 break;
1602 }
1603 }
1604 blame_origin_decref(suspect);
1605
1606 if (sb->debug) /* sanity */
1607 sanity_check_refcnt(sb);
1608 }
1609}
09002f1b
JS
1610
1611static const char *get_next_line(const char *start, const char *end)
1612{
1613 const char *nl = memchr(start, '\n', end - start);
1614 return nl ? nl + 1 : end;
1615}
1616
1617/*
1618 * To allow quick access to the contents of nth line in the
1619 * final image, prepare an index in the scoreboard.
1620 */
1621static int prepare_lines(struct blame_scoreboard *sb)
1622{
1623 const char *buf = sb->final_buf;
1624 unsigned long len = sb->final_buf_size;
1625 const char *end = buf + len;
1626 const char *p;
1627 int *lineno;
1628 int num = 0;
1629
1630 for (p = buf; p < end; p = get_next_line(p, end))
1631 num++;
1632
1633 ALLOC_ARRAY(sb->lineno, num + 1);
1634 lineno = sb->lineno;
1635
1636 for (p = buf; p < end; p = get_next_line(p, end))
1637 *lineno++ = p - buf;
1638
1639 *lineno = len;
1640
1641 sb->num_lines = num;
1642 return sb->num_lines;
1643}
1644
1645static struct commit *find_single_final(struct rev_info *revs,
1646 const char **name_p)
1647{
1648 int i;
1649 struct commit *found = NULL;
1650 const char *name = NULL;
1651
1652 for (i = 0; i < revs->pending.nr; i++) {
1653 struct object *obj = revs->pending.objects[i].item;
1654 if (obj->flags & UNINTERESTING)
1655 continue;
1656 obj = deref_tag(obj, NULL, 0);
1657 if (obj->type != OBJ_COMMIT)
1658 die("Non commit %s?", revs->pending.objects[i].name);
1659 if (found)
1660 die("More than one commit to dig from %s and %s?",
1661 revs->pending.objects[i].name, name);
1662 found = (struct commit *)obj;
1663 name = revs->pending.objects[i].name;
1664 }
1665 if (name_p)
9e7d8a9b 1666 *name_p = xstrdup_or_null(name);
09002f1b
JS
1667 return found;
1668}
1669
1670static struct commit *dwim_reverse_initial(struct rev_info *revs,
1671 const char **name_p)
1672{
1673 /*
1674 * DWIM "git blame --reverse ONE -- PATH" as
1675 * "git blame --reverse ONE..HEAD -- PATH" but only do so
1676 * when it makes sense.
1677 */
1678 struct object *obj;
1679 struct commit *head_commit;
583c6a22 1680 struct object_id head_oid;
09002f1b
JS
1681
1682 if (revs->pending.nr != 1)
1683 return NULL;
1684
1685 /* Is that sole rev a committish? */
1686 obj = revs->pending.objects[0].item;
1687 obj = deref_tag(obj, NULL, 0);
1688 if (obj->type != OBJ_COMMIT)
1689 return NULL;
1690
1691 /* Do we have HEAD? */
49e61479 1692 if (!resolve_ref_unsafe("HEAD", RESOLVE_REF_READING, &head_oid, NULL))
09002f1b 1693 return NULL;
583c6a22 1694 head_commit = lookup_commit_reference_gently(&head_oid, 1);
09002f1b
JS
1695 if (!head_commit)
1696 return NULL;
1697
1698 /* Turn "ONE" into "ONE..HEAD" then */
1699 obj->flags |= UNINTERESTING;
1700 add_pending_object(revs, &head_commit->object, "HEAD");
1701
1702 if (name_p)
1703 *name_p = revs->pending.objects[0].name;
1704 return (struct commit *)obj;
1705}
1706
1707static struct commit *find_single_initial(struct rev_info *revs,
1708 const char **name_p)
1709{
1710 int i;
1711 struct commit *found = NULL;
1712 const char *name = NULL;
1713
1714 /*
1715 * There must be one and only one negative commit, and it must be
1716 * the boundary.
1717 */
1718 for (i = 0; i < revs->pending.nr; i++) {
1719 struct object *obj = revs->pending.objects[i].item;
1720 if (!(obj->flags & UNINTERESTING))
1721 continue;
1722 obj = deref_tag(obj, NULL, 0);
1723 if (obj->type != OBJ_COMMIT)
1724 die("Non commit %s?", revs->pending.objects[i].name);
1725 if (found)
1726 die("More than one commit to dig up from, %s and %s?",
1727 revs->pending.objects[i].name, name);
1728 found = (struct commit *) obj;
1729 name = revs->pending.objects[i].name;
1730 }
1731
1732 if (!name)
1733 found = dwim_reverse_initial(revs, &name);
1734 if (!name)
1735 die("No commit to dig up from?");
1736
1737 if (name_p)
9e7d8a9b 1738 *name_p = xstrdup(name);
09002f1b
JS
1739 return found;
1740}
1741
1742void init_scoreboard(struct blame_scoreboard *sb)
1743{
1744 memset(sb, 0, sizeof(struct blame_scoreboard));
1745 sb->move_score = BLAME_DEFAULT_MOVE_SCORE;
1746 sb->copy_score = BLAME_DEFAULT_COPY_SCORE;
1747}
1748
1749void setup_scoreboard(struct blame_scoreboard *sb, const char *path, struct blame_origin **orig)
1750{
1751 const char *final_commit_name = NULL;
1752 struct blame_origin *o;
1753 struct commit *final_commit = NULL;
1754 enum object_type type;
1755
1756 if (sb->reverse && sb->contents_from)
1757 die(_("--contents and --reverse do not blend well."));
1758
1759 if (!sb->reverse) {
1760 sb->final = find_single_final(sb->revs, &final_commit_name);
1761 sb->commits.compare = compare_commits_by_commit_date;
1762 } else {
1763 sb->final = find_single_initial(sb->revs, &final_commit_name);
1764 sb->commits.compare = compare_commits_by_reverse_commit_date;
1765 }
1766
1767 if (sb->final && sb->contents_from)
1768 die(_("cannot use --contents with final commit object name"));
1769
1770 if (sb->reverse && sb->revs->first_parent_only)
1771 sb->revs->children.name = NULL;
1772
1773 if (!sb->final) {
1774 /*
1775 * "--not A B -- path" without anything positive;
1776 * do not default to HEAD, but use the working tree
1777 * or "--contents".
1778 */
1779 setup_work_tree();
1780 sb->final = fake_working_tree_commit(&sb->revs->diffopt,
1781 path, sb->contents_from);
1782 add_pending_object(sb->revs, &(sb->final->object), ":");
1783 }
1784
1785 if (sb->reverse && sb->revs->first_parent_only) {
1786 final_commit = find_single_final(sb->revs, NULL);
1787 if (!final_commit)
1788 die(_("--reverse and --first-parent together require specified latest commit"));
1789 }
1790
1791 /*
1792 * If we have bottom, this will mark the ancestors of the
1793 * bottom commits we would reach while traversing as
1794 * uninteresting.
1795 */
1796 if (prepare_revision_walk(sb->revs))
1797 die(_("revision walk setup failed"));
1798
1799 if (sb->reverse && sb->revs->first_parent_only) {
1800 struct commit *c = final_commit;
1801
1802 sb->revs->children.name = "children";
1803 while (c->parents &&
1804 oidcmp(&c->object.oid, &sb->final->object.oid)) {
1805 struct commit_list *l = xcalloc(1, sizeof(*l));
1806
1807 l->item = c;
1808 if (add_decoration(&sb->revs->children,
1809 &c->parents->item->object, l))
1810 die("BUG: not unique item in first-parent chain");
1811 c = c->parents->item;
1812 }
1813
1814 if (oidcmp(&c->object.oid, &sb->final->object.oid))
1815 die(_("--reverse --first-parent together require range along first-parent chain"));
1816 }
1817
1818 if (is_null_oid(&sb->final->object.oid)) {
1819 o = sb->final->util;
1820 sb->final_buf = xmemdupz(o->file.ptr, o->file.size);
1821 sb->final_buf_size = o->file.size;
1822 }
1823 else {
1824 o = get_origin(sb->final, path);
1825 if (fill_blob_sha1_and_mode(o))
1826 die(_("no such path %s in %s"), path, final_commit_name);
1827
0d1e0e78 1828 if (sb->revs->diffopt.flags.allow_textconv &&
09002f1b
JS
1829 textconv_object(path, o->mode, &o->blob_oid, 1, (char **) &sb->final_buf,
1830 &sb->final_buf_size))
1831 ;
1832 else
b4f5aca4 1833 sb->final_buf = read_object_file(&o->blob_oid, &type,
1834 &sb->final_buf_size);
09002f1b
JS
1835
1836 if (!sb->final_buf)
1837 die(_("cannot read blob %s for path %s"),
1838 oid_to_hex(&o->blob_oid),
1839 path);
1840 }
1841 sb->num_read_blob++;
1842 prepare_lines(sb);
1843
1844 if (orig)
1845 *orig = o;
9e7d8a9b
SG
1846
1847 free((char *)final_commit_name);
09002f1b 1848}
bd481de7
JS
1849
1850
1851
1852struct blame_entry *blame_entry_prepend(struct blame_entry *head,
1853 long start, long end,
1854 struct blame_origin *o)
1855{
1856 struct blame_entry *new_head = xcalloc(1, sizeof(struct blame_entry));
1857 new_head->lno = start;
1858 new_head->num_lines = end - start;
1859 new_head->suspect = o;
1860 new_head->s_lno = start;
1861 new_head->next = head;
1862 blame_origin_incref(o);
1863 return new_head;
1864}