/*
* src typically is on-stack; we want to copy the information in it to
* a malloced blame_entry that gets added to the given queue. The
* origin of dst loses a refcnt.
*/
static void dup_entry(struct blame_entry ***queue,
struct blame_entry *dst, struct blame_entry *src)
{
blame_origin_incref(src->suspect);
blame_origin_decref(dst->suspect);
memcpy(dst, src, sizeof(*src));
dst->next = **queue;
**queue = dst;
*queue = &dst->next;
}
/*
* It is known that lines between tlno to same came from parent, and e
* has an overlap with that range. it also is known that parent's
* line plno corresponds to e's line tlno.
*
* <---- e ----->
* <------>
* <------------>
* <------------>
* <------------------>
*
* Split e into potentially three parts; before this chunk, the chunk
* to be blamed for the parent, and after that portion.
*/
static void split_overlap(struct blame_entry *split,
struct blame_entry *e,
int tlno, int plno, int same,
struct blame_origin *parent)
{
int chunk_end_lno;
memset(split, 0, sizeof(struct blame_entry [3]));
if (e->s_lno < tlno) {
/* there is a pre-chunk part not blamed on parent */
split[0].suspect = blame_origin_incref(e->suspect);
split[0].lno = e->lno;
split[0].s_lno = e->s_lno;
split[0].num_lines = tlno - e->s_lno;
split[1].lno = e->lno + tlno - e->s_lno;
split[1].s_lno = plno;
}
else {
split[1].lno = e->lno;
split[1].s_lno = plno + (e->s_lno - tlno);
}
if (same < e->s_lno + e->num_lines) {
/* there is a post-chunk part not blamed on parent */
split[2].suspect = blame_origin_incref(e->suspect);
split[2].lno = e->lno + (same - e->s_lno);
split[2].s_lno = e->s_lno + (same - e->s_lno);
split[2].num_lines = e->s_lno + e->num_lines - same;
chunk_end_lno = split[2].lno;
}
else
chunk_end_lno = e->lno + e->num_lines;
split[1].num_lines = chunk_end_lno - split[1].lno;
/*
* if it turns out there is nothing to blame the parent for,
* forget about the splitting. !split[1].suspect signals this.
*/
if (split[1].num_lines < 1)
return;
split[1].suspect = blame_origin_incref(parent);
}
/*
* Append a new blame entry to a given output queue.
*/
static void add_blame_entry(struct blame_entry ***queue,
const struct blame_entry *src)
{
struct blame_entry *e = xmalloc(sizeof(*e));
memcpy(e, src, sizeof(*e));
blame_origin_incref(e->suspect);
e->next = **queue;
**queue = e;
*queue = &e->next;
}
/*
* Locate an existing origin or create a new one.
* This moves the origin to front position in the commit util list.
*/
static struct blame_origin *get_origin(struct commit *commit, const char *path)
{
struct blame_origin *o, *l;
for (o = commit->util, l = NULL; o; l = o, o = o->next) {
if (!strcmp(o->path, path)) {
/* bump to front */
if (l) {
l->next = o->next;
o->next = commit->util;
commit->util = o;
}
return blame_origin_incref(o);
}
}
return make_origin(commit, path);
}
/*
* The blobs of origin and porigin exactly match, so everything
* origin is suspected for can be blamed on the parent.
*/
static void pass_whole_blame(struct blame_scoreboard *sb,
struct blame_origin *origin, struct blame_origin *porigin)
{
struct blame_entry *e, *suspects;
if (!porigin->file.ptr && origin->file.ptr) {
/* Steal its file */
porigin->file = origin->file;
origin->file.ptr = NULL;
}
suspects = origin->suspects;
origin->suspects = NULL;
for (e = suspects; e; e = e->next) {
blame_origin_incref(porigin);
blame_origin_decref(e->suspect);
e->suspect = porigin;
}
queue_blames(sb, porigin, suspects);
}
/*
* best_so_far[] and potential[] are both a split of an existing blame_entry
* that passes blame to the parent. Maintain best_so_far the best split so
* far, by comparing potential and best_so_far and copying potential into
* bst_so_far as needed.
*/
static void copy_split_if_better(struct blame_scoreboard *sb,
struct blame_entry *best_so_far,
struct blame_entry *potential)
{
int i;
if (!potential[1].suspect)
return;
if (best_so_far[1].suspect) {
if (blame_entry_score(sb, &potential[1]) <
blame_entry_score(sb, &best_so_far[1]))
return;
}
for (i = 0; i < 3; i++)
blame_origin_incref(potential[i].suspect);
decref_split(best_so_far);
memcpy(best_so_far, potential, sizeof(struct blame_entry[3]));
}
/*
* Process one hunk from the patch between the current suspect for
* blame_entry e and its parent. This first blames any unfinished
* entries before the chunk (which is where target and parent start
* differing) on the parent, and then splits blame entries at the
* start and at the end of the difference region. Since use of -M and
* -C options may lead to overlapping/duplicate source line number
* ranges, all we can rely on from sorting/merging is the order of the
* first suspect line number.
*/
static void blame_chunk(struct blame_entry ***dstq, struct blame_entry ***srcq,
int tlno, int offset, int same,
struct blame_origin *parent)
{
struct blame_entry *e = **srcq;
struct blame_entry *samep = NULL, *diffp = NULL;
while (e && e->s_lno < tlno) {
struct blame_entry *next = e->next;
/*
* current record starts before differing portion. If
* it reaches into it, we need to split it up and
* examine the second part separately.
*/
if (e->s_lno + e->num_lines > tlno) {
/* Move second half to a new record */
int len = tlno - e->s_lno;
struct blame_entry *n = xcalloc(1, sizeof (struct blame_entry));
n->suspect = e->suspect;
n->lno = e->lno + len;
n->s_lno = e->s_lno + len;
n->num_lines = e->num_lines - len;
e->num_lines = len;
e->score = 0;
/* Push new record to diffp */
n->next = diffp;
diffp = n;
} else
blame_origin_decref(e->suspect);
/* Pass blame for everything before the differing
* chunk to the parent */
e->suspect = blame_origin_incref(parent);
e->s_lno += offset;
e->next = samep;
samep = e;
e = next;
}
/*
* As we don't know how much of a common stretch after this
* diff will occur, the currently blamed parts are all that we
* can assign to the parent for now.
*/
if (samep) {
**dstq = reverse_blame(samep, **dstq);
*dstq = &samep->next;
}
/*
* Prepend the split off portions: everything after e starts
* after the blameable portion.
*/
e = reverse_blame(diffp, e);
/*
* Now retain records on the target while parts are different
* from the parent.
*/
samep = NULL;
diffp = NULL;
while (e && e->s_lno < same) {
struct blame_entry *next = e->next;
/*
* If current record extends into sameness, need to split.
*/
if (e->s_lno + e->num_lines > same) {
/*
* Move second half to a new record to be
* processed by later chunks
*/
int len = same - e->s_lno;
struct blame_entry *n = xcalloc(1, sizeof (struct blame_entry));
n->suspect = blame_origin_incref(e->suspect);
n->lno = e->lno + len;
n->s_lno = e->s_lno + len;
n->num_lines = e->num_lines - len;
e->num_lines = len;
e->score = 0;
/* Push new record to samep */
n->next = samep;
samep = n;
}
e->next = diffp;
diffp = e;
e = next;
}
**srcq = reverse_blame(diffp, reverse_blame(samep, e));
/* Move across elements that are in the unblamable portion */
if (diffp)
*srcq = &diffp->next;
}
/*
* We have an origin -- check if the same path exists in the
* parent and return an origin structure to represent it.
*/
static struct blame_origin *find_origin(struct commit *parent,
struct blame_origin *origin)
{
struct blame_origin *porigin;
struct diff_options diff_opts;
const char *paths[2];
/* First check any existing origins */
for (porigin = parent->util; porigin; porigin = porigin->next)
if (!strcmp(porigin->path, origin->path)) {
/*
* The same path between origin and its parent
* without renaming -- the most common case.
*/
return blame_origin_incref (porigin);
}
/* See if the origin->path is different between parent
* and origin first. Most of the time they are the
* same and diff-tree is fairly efficient about this.
*/
diff_setup(&diff_opts);
DIFF_OPT_SET(&diff_opts, RECURSIVE);
diff_opts.detect_rename = 0;
diff_opts.output_format = DIFF_FORMAT_NO_OUTPUT;
paths[0] = origin->path;
paths[1] = NULL;
parse_pathspec(&diff_opts.pathspec,
PATHSPEC_ALL_MAGIC & ~PATHSPEC_LITERAL,
PATHSPEC_LITERAL_PATH, "", paths);
diff_setup_done(&diff_opts);
if (is_null_oid(&origin->commit->object.oid))
do_diff_cache(&parent->tree->object.oid, &diff_opts);
else
diff_tree_oid(&parent->tree->object.oid,
&origin->commit->tree->object.oid,
"", &diff_opts);
diffcore_std(&diff_opts);
if (!diff_queued_diff.nr) {
/* The path is the same as parent */
porigin = get_origin(parent, origin->path);
oidcpy(&porigin->blob_oid, &origin->blob_oid);
porigin->mode = origin->mode;
} else {
/*
* Since origin->path is a pathspec, if the parent
* commit had it as a directory, we will see a whole
* bunch of deletion of files in the directory that we
* do not care about.
*/
int i;
struct diff_filepair *p = NULL;
for (i = 0; i < diff_queued_diff.nr; i++) {
const char *name;
p = diff_queued_diff.queue[i];
name = p->one->path ? p->one->path : p->two->path;
if (!strcmp(name, origin->path))
break;
}
if (!p)
die("internal error in blame::find_origin");
switch (p->status) {
default:
die("internal error in blame::find_origin (%c)",
p->status);
case 'M':
porigin = get_origin(parent, origin->path);
oidcpy(&porigin->blob_oid, &p->one->oid);
porigin->mode = p->one->mode;
break;
case 'A':
case 'T':
/* Did not exist in parent, or type changed */
break;
}
}
diff_flush(&diff_opts);
clear_pathspec(&diff_opts.pathspec);
return porigin;
}
/*
* The main loop -- while we have blobs with lines whose true origin
* is still unknown, pick one blob, and allow its lines to pass blames
* to its parents. */
void assign_blame(struct blame_scoreboard *sb, int opt)
{
struct rev_info *revs = sb->revs;
struct commit *commit = prio_queue_get(&sb->commits);
while (commit) {
struct blame_entry *ent;
struct blame_origin *suspect = commit->util;
/* find one suspect to break down */
while (suspect && !suspect->suspects)
suspect = suspect->next;
if (!suspect) {
commit = prio_queue_get(&sb->commits);
continue;
}
assert(commit == suspect->commit);
/*
* We will use this suspect later in the loop,
* so hold onto it in the meantime.
*/
blame_origin_incref(suspect);
parse_commit(commit);
if (sb->reverse ||
(!(commit->object.flags & UNINTERESTING) &&
!(revs->max_age != -1 && commit->date < revs->max_age)))
pass_blame(sb, suspect, opt);
else {
commit->object.flags |= UNINTERESTING;
if (commit->object.parsed)
mark_parents_uninteresting(commit);
}
/* treat root commit as boundary */
if (!commit->parents && !sb->show_root)
commit->object.flags |= UNINTERESTING;
/* Take responsibility for the remaining entries */
ent = suspect->suspects;
if (ent) {
suspect->guilty = 1;
for (;;) {
struct blame_entry *next = ent->next;
if (sb->found_guilty_entry)
sb->found_guilty_entry(ent, sb->found_guilty_entry_data);
if (next) {
ent = next;
continue;
}
ent->next = sb->ent;
sb->ent = suspect->suspects;
suspect->suspects = NULL;
break;
}
}
blame_origin_decref(suspect);
if (sb->debug) /* sanity */
sanity_check_refcnt(sb);
}
}
static void pass_blame(struct blame_scoreboard *sb, struct blame_origin *origin, int opt)
{
struct rev_info *revs = sb->revs;
int i, pass, num_sg;
struct commit *commit = origin->commit;
struct commit_list *sg;
struct blame_origin *sg_buf[MAXSG];
struct blame_origin *porigin, **sg_origin = sg_buf;
struct blame_entry *toosmall = NULL;
struct blame_entry *blames, **blametail = &blames;
num_sg = num_scapegoats(revs, commit, sb->reverse);
if (!num_sg)
goto finish;
else if (num_sg < ARRAY_SIZE(sg_buf))
memset(sg_buf, 0, sizeof(sg_buf));
else
sg_origin = xcalloc(num_sg, sizeof(*sg_origin));
/*
* The first pass looks for unrenamed path to optimize for
* common cases, then we look for renames in the second pass.
*/
for (pass = 0; pass < 2 - sb->no_whole_file_rename; pass++) {
struct blame_origin *(*find)(struct commit *, struct blame_origin *);
find = pass ? find_rename : find_origin;
for (i = 0, sg = first_scapegoat(revs, commit, sb->reverse);
i < num_sg && sg;
sg = sg->next, i++) {
struct commit *p = sg->item;
int j, same;
if (sg_origin[i])
continue;
if (parse_commit(p))
continue;
porigin = find(p, origin);
if (!porigin)
continue;
if (!oidcmp(&porigin->blob_oid, &origin->blob_oid)) {
pass_whole_blame(sb, origin, porigin);
blame_origin_decref(porigin);
goto finish;
}
for (j = same = 0; j < i; j++)
if (sg_origin[j] &&
!oidcmp(&sg_origin[j]->blob_oid, &porigin->blob_oid)) {
same = 1;
break;
}
if (!same)
sg_origin[i] = porigin;
else
blame_origin_decref(porigin);
}
}
sb->num_commits++;
for (i = 0, sg = first_scapegoat(revs, commit, sb->reverse);
i < num_sg && sg;
sg = sg->next, i++) {
struct blame_origin *porigin = sg_origin[i];
if (!porigin)
continue;
if (!origin->previous) {
blame_origin_incref(porigin);
origin->previous = porigin;
}
pass_blame_to_parent(sb, origin, porigin);
if (!origin->suspects)
goto finish;
}
/*
* Optionally find moves in parents' files.
*/
if (opt & PICKAXE_BLAME_MOVE) {
filter_small(sb, &toosmall, &origin->suspects, sb->move_score);
if (origin->suspects) {
for (i = 0, sg = first_scapegoat(revs, commit, sb->reverse);
i < num_sg && sg;
sg = sg->next, i++) {
struct blame_origin *porigin = sg_origin[i];
if (!porigin)
continue;
find_move_in_parent(sb, &blametail, &toosmall, origin, porigin);
if (!origin->suspects)
break;
}
}
}
/*
* Optionally find copies from parents' files.
*/
if (opt & PICKAXE_BLAME_COPY) {
if (sb->copy_score > sb->move_score)
filter_small(sb, &toosmall, &origin->suspects, sb->copy_score);
else if (sb->copy_score < sb->move_score) {
origin->suspects = blame_merge(origin->suspects, toosmall);
toosmall = NULL;
filter_small(sb, &toosmall, &origin->suspects, sb->copy_score);
}
if (!origin->suspects)
goto finish;
for (i = 0, sg = first_scapegoat(revs, commit, sb->reverse);
i < num_sg && sg;
sg = sg->next, i++) {
struct blame_origin *porigin = sg_origin[i];
find_copy_in_parent(sb, &blametail, &toosmall,
origin, sg->item, porigin, opt);
if (!origin->suspects)
goto finish;
}
}
finish:
*blametail = NULL;
distribute_blame(sb, blames);
/*
* prepend toosmall to origin->suspects
*
* There is no point in sorting: this ends up on a big
* unsorted list in the caller anyway.
*/
if (toosmall) {
struct blame_entry **tail = &toosmall;
while (*tail)
tail = &(*tail)->next;
*tail = origin->suspects;
origin->suspects = toosmall;
}
for (i = 0; i < num_sg; i++) {
if (sg_origin[i]) {
drop_origin_blob(sg_origin[i]);
blame_origin_decref(sg_origin[i]);
}
}
drop_origin_blob(origin);
if (sg_buf != sg_origin)
free(sg_origin);
}
