Blame uncommitted deletions starting from HEAD

[krobelus]

When starting a blame in the stage view on a deleted line, it seems
appropriate to find the commit that added the line. To this end, instead
of starting the blame view from the working-tree-state of a file, use
the version at HEAD, which still has the deleted line (unless the line
was only just staged). Compute the position of the deleted line to
start the blame view there.

Closes #1008

[koutcher]

Couldn't view->env->file be passed as an argument to git diff-index to avoid parsing all the diffs ? Also, you may return a pointer on a local variable in find_deleted_line_in_head. Maybe it could be changed to bool find_deleted_line_in_head(struct view *view, struct line *line, char *filename, unsigned long *lineno) and stage_request() changes could be reduced to:

+               if (find_deleted_line_in_head(view, line, view->env->file, &view->env->goto_lineno))
+                       string_copy(view->env->ref, "HEAD");
+               else
                        view->env->goto_lineno = diff_get_lineno(view, line, false);
                if (view->env->goto_lineno > 0)
                        view->env->goto_lineno--;

[krobelus]

Couldn't view->env->file be passed as an argument to git diff-index to avoid parsing all the diffs ?

This is to deal with RM files; If file x is renamed to y in the index, then git diff-index HEAD y doesn't know about the rename. I think there is no way around computing the diff of the entire tree to detect renames.

However, dealing with renames here is not super important to me, maybe we should leave that out for now?
Alternatively, I can do the entire diff (to find the renaming) only for those cases that actually need it: where view->env->file exists in the index but not in HEAD.

[koutcher]

The use case is fine, I just didn't realize that providing a pathspec to diff-index was defeating the rename detection.

[krobelus]

Addressed all comments.
I added an ls-tree call that allows to skip rename detection most of the time - when a file with the same name exists in HEAD, this will always use that filename Technically there can be false positives, but rename detection is only a heuristic, too.

[koutcher]

Looks good to me. As Steve used to say, one more thing: so far, @jonas has maintained compatibility with pre-C99 compilers. Can you keep variables declaration at the beginning of blocks ? Also can you check NEWS.adoc, Improvements are listed before Bug fixes.

[krobelus]

Should be done now, thanks!

[koutcher]

This is a great improvement. Thanks.