Merge branch 'ab/sha1dc'
[gitweb.git] / blame.c
diff --git a/blame.c b/blame.c
index 798e61b8c7620f50b17125bb32e1ec139e4e572b..91e26e93e8a23d0af63e1bc8a74e9f7690276ab4 100644 (file)
--- a/blame.c
+++ b/blame.c
@@ -4,6 +4,7 @@
 #include "mergesort.h"
 #include "diff.h"
 #include "diffcore.h"
+#include "tag.h"
 #include "blame.h"
 
 void blame_origin_decref(struct blame_origin *o)
@@ -49,7 +50,7 @@ static struct blame_origin *make_origin(struct commit *commit, const char *path)
  * Locate an existing origin or create a new one.
  * This moves the origin to front position in the commit util list.
  */
-struct blame_origin *get_origin(struct commit *commit, const char *path)
+static struct blame_origin *get_origin(struct commit *commit, const char *path)
 {
        struct blame_origin *o, *l;
 
@@ -98,7 +99,7 @@ static struct commit_list **append_parent(struct commit_list **tail, const struc
 {
        struct commit *parent;
 
-       parent = lookup_commit_reference(oid->hash);
+       parent = lookup_commit_reference(oid);
        if (!parent)
                die("no such commit %s", oid_to_hex(oid));
        return &commit_list_insert(parent, tail)->next;
@@ -142,9 +143,9 @@ static void set_commit_buffer_from_strbuf(struct commit *c, struct strbuf *sb)
  * Prepare a dummy commit that represents the work tree (or staged) item.
  * Note that annotating work tree item never works in the reverse.
  */
-struct commit *fake_working_tree_commit(struct diff_options *opt,
-                                       const char *path,
-                                       const char *contents_from)
+static struct commit *fake_working_tree_commit(struct diff_options *opt,
+                                              const char *path,
+                                              const char *contents_from)
 {
        struct commit *commit;
        struct blame_origin *origin;
@@ -228,7 +229,7 @@ struct commit *fake_working_tree_commit(struct diff_options *opt,
                if (strbuf_read(&buf, 0, 0) < 0)
                        die_errno("failed to read from stdin");
        }
-       convert_to_git(path, buf.buf, buf.len, &buf, 0);
+       convert_to_git(&the_index, path, buf.buf, buf.len, &buf, 0);
        origin->file.ptr = buf.buf;
        origin->file.size = buf.len;
        pretend_sha1_file(buf.buf, buf.len, OBJ_BLOB, origin->blob_oid.hash);
@@ -313,8 +314,7 @@ static void fill_origin_blob(struct diff_options *opt,
 static void drop_origin_blob(struct blame_origin *o)
 {
        if (o->file.ptr) {
-               free(o->file.ptr);
-               o->file.ptr = NULL;
+               FREE_AND_NULL(o->file.ptr);
        }
 }
 
@@ -410,6 +410,13 @@ void blame_sort_final(struct blame_scoreboard *sb)
                                  compare_blame_final);
 }
 
+static int compare_commits_by_reverse_commit_date(const void *a,
+                                                 const void *b,
+                                                 void *c)
+{
+       return -compare_commits_by_commit_date(a, b, c);
+}
+
 /*
  * For debugging -- origin is refcounted, and this asserts that
  * we do not underflow.
@@ -482,6 +489,32 @@ static void queue_blames(struct blame_scoreboard *sb, struct blame_origin *porig
        }
 }
 
+/*
+ * Fill the blob_sha1 field of an origin if it hasn't, so that later
+ * call to fill_origin_blob() can use it to locate the data.  blob_sha1
+ * for an origin is also used to pass the blame for the entire file to
+ * the parent to detect the case where a child's blob is identical to
+ * that of its parent's.
+ *
+ * This also fills origin->mode for corresponding tree path.
+ */
+static int fill_blob_sha1_and_mode(struct blame_origin *origin)
+{
+       if (!is_null_oid(&origin->blob_oid))
+               return 0;
+       if (get_tree_entry(origin->commit->object.oid.hash,
+                          origin->path,
+                          origin->blob_oid.hash, &origin->mode))
+               goto error_out;
+       if (sha1_object_info(origin->blob_oid.hash, NULL) != OBJ_BLOB)
+               goto error_out;
+       return 0;
+ error_out:
+       oidclr(&origin->blob_oid);
+       origin->mode = S_IFINVALID;
+       return -1;
+}
+
 /*
  * We have an origin -- check if the same path exists in the
  * parent and return an origin structure to represent it.
@@ -520,11 +553,11 @@ static struct blame_origin *find_origin(struct commit *parent,
        diff_setup_done(&diff_opts);
 
        if (is_null_oid(&origin->commit->object.oid))
-               do_diff_cache(parent->tree->object.oid.hash, &diff_opts);
+               do_diff_cache(&parent->tree->object.oid, &diff_opts);
        else
-               diff_tree_sha1(parent->tree->object.oid.hash,
-                              origin->commit->tree->object.oid.hash,
-                              "", &diff_opts);
+               diff_tree_oid(&parent->tree->object.oid,
+                             &origin->commit->tree->object.oid,
+                             "", &diff_opts);
        diffcore_std(&diff_opts);
 
        if (!diff_queued_diff.nr) {
@@ -589,11 +622,11 @@ static struct blame_origin *find_rename(struct commit *parent,
        diff_setup_done(&diff_opts);
 
        if (is_null_oid(&origin->commit->object.oid))
-               do_diff_cache(parent->tree->object.oid.hash, &diff_opts);
+               do_diff_cache(&parent->tree->object.oid, &diff_opts);
        else
-               diff_tree_sha1(parent->tree->object.oid.hash,
-                              origin->commit->tree->object.oid.hash,
-                              "", &diff_opts);
+               diff_tree_oid(&parent->tree->object.oid,
+                             &origin->commit->tree->object.oid,
+                             "", &diff_opts);
        diffcore_std(&diff_opts);
 
        for (i = 0; i < diff_queued_diff.nr; i++) {
@@ -1213,7 +1246,7 @@ static void find_copy_in_parent(struct blame_scoreboard *sb,
        /* Try "find copies harder" on new path if requested;
         * we do not want to use diffcore_rename() actually to
         * match things up; find_copies_harder is set only to
-        * force diff_tree_sha1() to feed all filepairs to diff_queue,
+        * force diff_tree_oid() to feed all filepairs to diff_queue,
         * and this code needs to be after diff_setup_done(), which
         * usually makes find-copies-harder imply copy detection.
         */
@@ -1223,11 +1256,11 @@ static void find_copy_in_parent(struct blame_scoreboard *sb,
                DIFF_OPT_SET(&diff_opts, FIND_COPIES_HARDER);
 
        if (is_null_oid(&target->commit->object.oid))
-               do_diff_cache(parent->tree->object.oid.hash, &diff_opts);
+               do_diff_cache(&parent->tree->object.oid, &diff_opts);
        else
-               diff_tree_sha1(parent->tree->object.oid.hash,
-                              target->commit->tree->object.oid.hash,
-                              "", &diff_opts);
+               diff_tree_oid(&parent->tree->object.oid,
+                             &target->commit->tree->object.oid,
+                             "", &diff_opts);
 
        if (!DIFF_OPT_TST(&diff_opts, FIND_COPIES_HARDER))
                diffcore_std(&diff_opts);
@@ -1574,3 +1607,256 @@ void assign_blame(struct blame_scoreboard *sb, int opt)
                        sanity_check_refcnt(sb);
        }
 }
+
+static const char *get_next_line(const char *start, const char *end)
+{
+       const char *nl = memchr(start, '\n', end - start);
+       return nl ? nl + 1 : end;
+}
+
+/*
+ * To allow quick access to the contents of nth line in the
+ * final image, prepare an index in the scoreboard.
+ */
+static int prepare_lines(struct blame_scoreboard *sb)
+{
+       const char *buf = sb->final_buf;
+       unsigned long len = sb->final_buf_size;
+       const char *end = buf + len;
+       const char *p;
+       int *lineno;
+       int num = 0;
+
+       for (p = buf; p < end; p = get_next_line(p, end))
+               num++;
+
+       ALLOC_ARRAY(sb->lineno, num + 1);
+       lineno = sb->lineno;
+
+       for (p = buf; p < end; p = get_next_line(p, end))
+               *lineno++ = p - buf;
+
+       *lineno = len;
+
+       sb->num_lines = num;
+       return sb->num_lines;
+}
+
+static struct commit *find_single_final(struct rev_info *revs,
+                                       const char **name_p)
+{
+       int i;
+       struct commit *found = NULL;
+       const char *name = NULL;
+
+       for (i = 0; i < revs->pending.nr; i++) {
+               struct object *obj = revs->pending.objects[i].item;
+               if (obj->flags & UNINTERESTING)
+                       continue;
+               obj = deref_tag(obj, NULL, 0);
+               if (obj->type != OBJ_COMMIT)
+                       die("Non commit %s?", revs->pending.objects[i].name);
+               if (found)
+                       die("More than one commit to dig from %s and %s?",
+                           revs->pending.objects[i].name, name);
+               found = (struct commit *)obj;
+               name = revs->pending.objects[i].name;
+       }
+       if (name_p)
+               *name_p = name;
+       return found;
+}
+
+static struct commit *dwim_reverse_initial(struct rev_info *revs,
+                                          const char **name_p)
+{
+       /*
+        * DWIM "git blame --reverse ONE -- PATH" as
+        * "git blame --reverse ONE..HEAD -- PATH" but only do so
+        * when it makes sense.
+        */
+       struct object *obj;
+       struct commit *head_commit;
+       struct object_id head_oid;
+
+       if (revs->pending.nr != 1)
+               return NULL;
+
+       /* Is that sole rev a committish? */
+       obj = revs->pending.objects[0].item;
+       obj = deref_tag(obj, NULL, 0);
+       if (obj->type != OBJ_COMMIT)
+               return NULL;
+
+       /* Do we have HEAD? */
+       if (!resolve_ref_unsafe("HEAD", RESOLVE_REF_READING, head_oid.hash, NULL))
+               return NULL;
+       head_commit = lookup_commit_reference_gently(&head_oid, 1);
+       if (!head_commit)
+               return NULL;
+
+       /* Turn "ONE" into "ONE..HEAD" then */
+       obj->flags |= UNINTERESTING;
+       add_pending_object(revs, &head_commit->object, "HEAD");
+
+       if (name_p)
+               *name_p = revs->pending.objects[0].name;
+       return (struct commit *)obj;
+}
+
+static struct commit *find_single_initial(struct rev_info *revs,
+                                         const char **name_p)
+{
+       int i;
+       struct commit *found = NULL;
+       const char *name = NULL;
+
+       /*
+        * There must be one and only one negative commit, and it must be
+        * the boundary.
+        */
+       for (i = 0; i < revs->pending.nr; i++) {
+               struct object *obj = revs->pending.objects[i].item;
+               if (!(obj->flags & UNINTERESTING))
+                       continue;
+               obj = deref_tag(obj, NULL, 0);
+               if (obj->type != OBJ_COMMIT)
+                       die("Non commit %s?", revs->pending.objects[i].name);
+               if (found)
+                       die("More than one commit to dig up from, %s and %s?",
+                           revs->pending.objects[i].name, name);
+               found = (struct commit *) obj;
+               name = revs->pending.objects[i].name;
+       }
+
+       if (!name)
+               found = dwim_reverse_initial(revs, &name);
+       if (!name)
+               die("No commit to dig up from?");
+
+       if (name_p)
+               *name_p = name;
+       return found;
+}
+
+void init_scoreboard(struct blame_scoreboard *sb)
+{
+       memset(sb, 0, sizeof(struct blame_scoreboard));
+       sb->move_score = BLAME_DEFAULT_MOVE_SCORE;
+       sb->copy_score = BLAME_DEFAULT_COPY_SCORE;
+}
+
+void setup_scoreboard(struct blame_scoreboard *sb, const char *path, struct blame_origin **orig)
+{
+       const char *final_commit_name = NULL;
+       struct blame_origin *o;
+       struct commit *final_commit = NULL;
+       enum object_type type;
+
+       if (sb->reverse && sb->contents_from)
+               die(_("--contents and --reverse do not blend well."));
+
+       if (!sb->reverse) {
+               sb->final = find_single_final(sb->revs, &final_commit_name);
+               sb->commits.compare = compare_commits_by_commit_date;
+       } else {
+               sb->final = find_single_initial(sb->revs, &final_commit_name);
+               sb->commits.compare = compare_commits_by_reverse_commit_date;
+       }
+
+       if (sb->final && sb->contents_from)
+               die(_("cannot use --contents with final commit object name"));
+
+       if (sb->reverse && sb->revs->first_parent_only)
+               sb->revs->children.name = NULL;
+
+       if (!sb->final) {
+               /*
+                * "--not A B -- path" without anything positive;
+                * do not default to HEAD, but use the working tree
+                * or "--contents".
+                */
+               setup_work_tree();
+               sb->final = fake_working_tree_commit(&sb->revs->diffopt,
+                                                    path, sb->contents_from);
+               add_pending_object(sb->revs, &(sb->final->object), ":");
+       }
+
+       if (sb->reverse && sb->revs->first_parent_only) {
+               final_commit = find_single_final(sb->revs, NULL);
+               if (!final_commit)
+                       die(_("--reverse and --first-parent together require specified latest commit"));
+       }
+
+       /*
+        * If we have bottom, this will mark the ancestors of the
+        * bottom commits we would reach while traversing as
+        * uninteresting.
+        */
+       if (prepare_revision_walk(sb->revs))
+               die(_("revision walk setup failed"));
+
+       if (sb->reverse && sb->revs->first_parent_only) {
+               struct commit *c = final_commit;
+
+               sb->revs->children.name = "children";
+               while (c->parents &&
+                      oidcmp(&c->object.oid, &sb->final->object.oid)) {
+                       struct commit_list *l = xcalloc(1, sizeof(*l));
+
+                       l->item = c;
+                       if (add_decoration(&sb->revs->children,
+                                          &c->parents->item->object, l))
+                               die("BUG: not unique item in first-parent chain");
+                       c = c->parents->item;
+               }
+
+               if (oidcmp(&c->object.oid, &sb->final->object.oid))
+                       die(_("--reverse --first-parent together require range along first-parent chain"));
+       }
+
+       if (is_null_oid(&sb->final->object.oid)) {
+               o = sb->final->util;
+               sb->final_buf = xmemdupz(o->file.ptr, o->file.size);
+               sb->final_buf_size = o->file.size;
+       }
+       else {
+               o = get_origin(sb->final, path);
+               if (fill_blob_sha1_and_mode(o))
+                       die(_("no such path %s in %s"), path, final_commit_name);
+
+               if (DIFF_OPT_TST(&sb->revs->diffopt, ALLOW_TEXTCONV) &&
+                   textconv_object(path, o->mode, &o->blob_oid, 1, (char **) &sb->final_buf,
+                                   &sb->final_buf_size))
+                       ;
+               else
+                       sb->final_buf = read_sha1_file(o->blob_oid.hash, &type,
+                                                      &sb->final_buf_size);
+
+               if (!sb->final_buf)
+                       die(_("cannot read blob %s for path %s"),
+                           oid_to_hex(&o->blob_oid),
+                           path);
+       }
+       sb->num_read_blob++;
+       prepare_lines(sb);
+
+       if (orig)
+               *orig = o;
+}
+
+
+
+struct blame_entry *blame_entry_prepend(struct blame_entry *head,
+                                       long start, long end,
+                                       struct blame_origin *o)
+{
+       struct blame_entry *new_head = xcalloc(1, sizeof(struct blame_entry));
+       new_head->lno = start;
+       new_head->num_lines = end - start;
+       new_head->suspect = o;
+       new_head->s_lno = start;
+       new_head->next = head;
+       blame_origin_incref(o);
+       return new_head;
+}