Re: FP/IP and performance (in general) and Patterns... (Re: [Caml-list] Avoiding shared data)

[skaller <skaller@users.sourceforge.net>]

On Wed, 2005-10-05 at 14:10 +0200, Oliver Bandel wrote:

> For example, I'm not really a fan of OO-programming, but when
> programming GUI-software I think it would be the best choice,
> and FP-programming lacks flexibility here 

I contend that concurrent programming is best here.
However, I don't know: I have a tool (Felix) to explore
this, but haven't got around to actually trying the
one-thread/one-widget paradigm. 

The OO solution is reactive and requires you to maintain
the whole widget state without use of the stack, thereby
discarding everything that is known about block structured
and functional programming, and going back to the bad old
days of the early 70s with a flat state space, except that
the state space is partitioned into objects.

The thread based solution control inverts this paradigm,
giving you a stack and control flow for every widget,
this is *active* or algorithmic programming. So I content
it must be better, because it not only partitions the state
space as OO does, it *also* partitions the control space,
and allows you to use all the advantages of a functional
or block structured style.

Actually, in practice, I expect a mix of the solutions
will be best. The reason is: we know a LOT about
state machines. So subparts of problems which correspond
to them can easily be programmed and reasoned about,
simply because we have lots of mature theory.

-- 
John Skaller <skaller at users dot sf dot net>
Felix, successor to C++: http://felix.sf.net