Setting the last vertex to 0 was the source of those 3D polygon bugs, is something similar going to come up here?

I don't think so. When the code returned is agg::path_cmd_stop, the x and y values are actually meaningless and should be ignored by the caller. This fix is only to silence a compiler warning that it is theoretically possible to get a vertex where the x y values were never set (if it were a 0-length path), since the compiler doesn't understand the semantics of the vertex codes.