Tight class cohesion metric, all or just public methods?

Question

I am researching cohesion topic and found out that some claim TCC metric should only include public methods, some other sources claim all methods. Would it be wrong to use either approach? Why private methods should be excluded, as some suggest?

Answer 1

I would think you need to consider both the public methods and the private methods as follows:

The metric is based on evaluating the cohesion of among the public methods — the interface to the class as seen by external clients, and doesn't report the cohesion of the private internals as part the interface that a client consumes (as clearly they are not part of that).

However, during the computation of the cohesion among the public methods, you must consider cohesion among the instance variables that are accessed not just directly from the public methods, but also by private methods invoked by the public methods.

Imagine we have a class whose public method simply calls a private method:

class Foo {
    private bool _state;
    private void _setState ( bool choice ) { _state = choice; }
    public void Set() { _setState ( true ); }
    public void Reset () { _setState ( false ); }
}

Now, TCC concerns itself with the cohesion among Set & Reset, but not _setState.

However, in considering Set and Reset we need to take into account that they call _setState. If we don't consider the private methods in the static call graph for the public methods, then they will appear as failing to have cohesion with the each other, when in fact, they are not lacking cohesion: they interact over the _state instance variable.