[go-nuts] Question about interface composition

I had been building what I'll call *main* interfaces that had embedded *sub*
interfaces. I would then make structs that fulfilled the *sub* interface
and attach them to an implementation of the *main* interface. But then I
got the idea that I should make the *main* interface just be functions, and
then build structs that were made up of other structs that implemented all
of the needed function calls. This way a function could move from one
struct to another and the over all duck typing would be maintained.

Code here:

http://play.golang.org/p/nk1CGrCZEH

Is the sane? A bad practice? Do I not understand interface composition?

Thanks in advance.