The mesh often has more than one <polygons> section, and the Collada spec allows the vertex input to reference different position streams for each of these.

If the mesh is skinned, then these vertices represent the mesh at a particular frame. So I use the BIND_SHAPE_POSITIONS. This is a single list of positions however.

So my question is this: For a skinned mesh, will each <polygons> section always reference the same vertex input, so that I can just substitute the BIND_SHAPE_POSITIONS in place of those vertices and the correct mesh will be generated?