Re: [q-lang-users] Special values or failing functions

[Alexander N. <AN...@sp...>]

Hello Tim,

Thursday, May 17, 2007, 12:19:18 PM, you wrote:


TH> Might () be appropriate?

Great idea. While [] could be a 'proper' answer sometimes, empty tuple
is never a value.

TH> As a bonus idea, can you seek to work with lists more, thereby allowing a
TH> filter (<>()) expression as a tidy-up phase (before or after getmax-ing)?

In fact, I already changed the dist. Now it returns list of distances,
the whole path from A to B and [] is a valid failure designator - it means
'no path found'

AG> but IMHO it just doesn't
AG> seem to fit the rewriting model very well, where each equation is
AG> considered a separate entity.

Well, for me it's just a shortcut and it is always clear where it
belongs to. The only problem is existing code which may use global
variables with the same name.

AG> I was talking about an operation which
AG> bails out of the current reduction, not a general-purpose sentinel value
AG> to be returned by a function.

AG> Alexander, I think that this is what you wanted, right?

Well, maybe. Here's how I see it :

length []    = 0;
length [H|T] = _FAIL_ if (islist H);
             = 1+length T otherwise;

but also look at :

len []    = 0;
len [[H|Hs]|T] = _FAIL_;
len [H|T]      = 1+len T;

so semantics of _FAIL_ would be : rewrite the current match with
itself and return.
So, 'fail' applies to the left side, but _FAIL_ applies to the right side.
no matter how you call it - reject, normal, agree, refuse - which
strangely means the same.

Each time I am learning a new language I am thinking about how I would
implement it. It helps understanding and in fact it is the only time
when I can think about it. When I am already used to a language I just
use it.

For example
(X+Y)*Z         = X*Z+Y*Z;
X*(Y+Z)         = X*Y+X*Z;
is good, but
X+Y = Y+X; // commutative rule
will not work.


-- 
Best regards,
 Alexander                            mailto:AN...@sp...