Re[2]: Language Design for Refactoring.

On Thu, 1 Jul 2004, Michael Feathers wrote:

> JC>> Given a clean slate, what would be the characteristics of a language
> JC>> designed explicitly with making refactoring of programs written in it easy
> JC>> and reliable?
>
> G> Clarity of side efects and dependencies.
>
> I've thought about that one a little.  If a language had
> compiler-enforced "effect annotations" could be easier for a tool or a
> person to make changes.  Method const in C++ is a simple kind of
> effect annotation.  There could be more involved ones.  On the other
> hand, maintaining them could be hard.  I suspect the better thing is to
> keep them out of the code and let a refactoring tool compute them.

That is the current path of my thoughts. Make it possibly for the compiler 
/ IDE to tell to programmer what the "effect annotation" _and_ type is, 
not the otherway round.

Experience with Ruby tells me dynamic typed is better than static type.

My experience with C++ tells me you can get into vast sea changes just 
because you added a single "const" to a piece of bad legacy code. (One of 
the things I truly truly hated about Rational Rose. Add a const and it can 
be a million mouse clicks later before things finally compile.

I also love the notion of Linear Logic, or "reference counting on the 
fingers of one thumb".

ie. At any one time, there may only be a single reference to every object.