questions about how to implement a gist index

I would like to use different data structures to represent the internal and leaf entries. I think I have found several examples in the postgresql source code where a variable length array is used to pack different data consistently with the value of a "type" attribute. Is this the suggested way to proceed, or would it be also possible to use two actually different data structures? I at first considered this latter approach, but some erroneous behavior suggested that maybe I am not fully clear about which data structure should be produced or expected by some parts of the implementation.