[rules-dev] Backwards chaining: the difference between input and output variables
Mark Proctor
mproctor at codehaus.org
Thu Apr 21 14:04:57 EDT 2011
The aim of the positional syntax is to appeal to those that work this
way or prefer positional, such as prolog users.
I would stress that the positional stuff is purely optional. Those that
wish can completely ignore it as though it never existed without any
loss of functionality.
You can call queries and have unification, without ever using
positional, and thus you'll never see the ; separator. The ; is used
only when you use positional syntax, to differentiate it from a boolean
eval.
Ceylon has adopted this and also proposes ; as a separator between named
and positional constructors on java objects.
The other option we discussed was wrapping positional arguments in an
array symbol, to make it more obviously different.
predName( [pos1, pos2, pos3], name == xx )
Edson preferred the above, I preferred ; as too many parenthesis and
brackets is confusing. It also adds an undue readability burdon on those
wishing to work purely in positional, meaning they would probably be
discinclined to adopt drools which would defeat the goal of the proposal
- which is to encourage positional users. ; is not over bearing enough
to discourage people who favour positional, and named argument people
can continue as though it never existed.
So that explains the motive and the decisions on ;
There is the other issue that geoffrey has brought up which is related
to our decision to support a unification syntax in a manner that
Jess/Clips and Prolog do now. The following two unification approaches
are now possible:
Cheese( $cn : name )
Person( $name : name, $cn : likes) // $cn constraint is a unification.
---
Cheese( $cn ) // assuming cheese name was position 1
Person( $name, $cn)
Geoffrey wante a way to differentiate unification, whether an argument
is an input or an output. Any symbol added to the positional unification
argument would need to also be applicable to the named unfication
argument for consistency. <- and -> would just be over kill. We briefly
considered using : on positional arguments, but realised it created an
inconsitency with named which would still retain the issue of what is <-
and what is ->
i.e. while this would work for positional
Cheese( $cn : ) // shows that $cn is an output unification variable
Person( $name, $cn ) // $cn is an input variable and would error if
$cn does not exist.
But this wouldn't help named, which would need a further symbole to
differentiate. As mentioned earlier == is not applicable as it's not a
constraint filter, it is a unification variable and has different wider
reaching behaviour.
Cheese( $cn : name )
Person( $name : name, $cn : likes)
Over all I feel what we have proposed is a reasonable compromise on
readability and regularity.
Mark
On 21/04/2011 18:07, Wolfgang Laun wrote:
> On 21 April 2011 18:23, Mark Proctor<mproctor at codehaus.org> wrote:
>> Wolfgang,
>>
>> Jess/clips have never differentiated between binding and constraint. has
>> that been a major problem for Jess?
>> (cheese name ?cn )
>> (person likes ?cn )
> It was, apparently, not applauded throughout, and I can understand
> that. Not a "major" problem, but also small inconveniences irk. The
> 1st occurrence defines, then it's usage. The parser distinguishes
> easily, and users get used to it.
>
> BUT, overall, the syntax here is not at all convenient and that's why
> the much more plausible form using "{likes == ?cn}", etc., has been
> introduced where you would use Java-style relations for constraints,
> leaving "{name ?cn}" for binding.
>
>
>> In previous DRL we did not allow unification like above. So you were forced
>> to separate binding and constraint:
>> Cheese( cn : name )
>> Person( likes == cn )
>>
> Perhaps Jess followed Drools here :) - it's clearer, and preferable.
>
>> In most common cases I think the separation is preferred, users do not need
>> to be concerned with the concepts of unification.
> Yes, yes, and yes.
>
>> What I have done is to
>> allow unification in the manner than Jess and Clips does, this is important
>> for prolog itself because the second pattern is not a filter. So == does not
>> make sense, it is a unification.
>
> You are not trying to tell me here that tje creation of an activation
> by selection of a fact by equality of one or more of its fields is
> not a "unification"? Or that the distinction "identity" vs. "equality"
> suddenly does not matter? It's not really "unification" in the sense
> that object must be IDENTICAL - it's still based on equality,
> whichever way it is defined. I see no reason not to use "==". But use
> another operator (e.g., -:- ) if you think that this is an issue.
>
> In Prolog-style terms you don't need an operator because you don't use
> anything except equality. BTW: What would happen if you permit the
> usual relational operators in those BC-terms?
>
> But mainly it's the positional-ness I don't like.
>
> Wolfgang
>
>> Mark
>>
>>
>> On 21/04/2011 07:27, Wolfgang Laun wrote:
>>
>> Designing syntax well is not easy. With extensions, one should strive for as
>> much
>> conformity with the existing language, while trying to follow general
>> principles.
>>
>> One might have discussed (for instance) the use of field names for
>> referencing
>> the query relations, taken from their parameter definition. And then one
>> could write,
>> as usual:
>>
>> ?editableThings(food: thing, location == loc )
>>
>> or
>>
>> ?editableThings(food: thing, loc: location )
>>
>> And the in/out is clear to all who know a little legacy DRL.
>>
>> And the ugly semicolon evaporates.
>>
>> And the maintainability/readability disadvantage of "positional" is gone.
>>
>> Cheers
>> -W
>>
>>
>> On 20 April 2011 22:52, Michael Anstis<michael.anstis at gmail.com> wrote:
>>> Simple yes, but consistent too should be a factor.
>>>
>>> Most questions we have to the user mailing list involve people writing DRL
>>> not using tooling.
>>>
>>> So DRL, IMO, has to be seen as the "tool" to author rules. Drop the
>>> proposed colon altogether or make it's use consistent.
>>>
>>> On 20 April 2011 17:42, Mark Proctor<mproctor at codehaus.org> wrote:
>>>> My personally opinion is to keep the language simple and instead have the
>>>> tooling inject what ever is necessary as a visulation. Be it different
>>>> colouring, hover over or graphic symbol. It keeps the language simple and
>>>> actually achieve the desired result better.
>>>>
>>>> Mark
>>>> On 20/04/2011 14:00, Leonardo Gomes wrote:
>>>>
>>>> +1 for Michael's suggestion.
>>>>
>>>> It's a bit more verbose, but makes things clear.
>>>>
>>>> The semicolon here:
>>>> ?editableThings(food : ?, loc;)
>>>>
>>>> Is a typo, right? You actually meant:
>>>>
>>>> ?editableThings(food : ?, loc);
>>>>
>>>> - Leo.
>>>>
>>>>
>>>>
>>>> On Wed, Apr 20, 2011 at 11:59 AM, Michael Anstis
>>>> <michael.anstis at gmail.com> wrote:
>>>>> Hmmmmm....
>>>>>
>>>>> Personally, I don't like the use of ":" i isolation as it's what we
>>>>> currently use to bind variables and I feel "cheese:" as an output definition
>>>>> could just make people question whether they've missed something. Perhaps
>>>>> "cheese : ?" would be a viable alternative. This would be in keeping with
>>>>> (a) current variable declaration, (b) the use of "?" to identify a call to a
>>>>> query. Geoffrey's examples would then become:-
>>>>>
>>>>> rule outputinput
>>>>> when
>>>>> Here( loc : location)
>>>>> ?editableThings(food : ?, loc;)
>>>>> then
>>>>> System.out.println("Food " + food + " at location " + loc);
>>>>> // Output:
>>>>> // Food crackers at location kitchen
>>>>> // Food apple at location kitchen
>>>>> end
>>>>>
>>>>> rule outputOutput
>>>>> when
>>>>> ?editableThings(food : ?, loc : ?;)
>>>>> then
>>>>> System.out.println("Food " + food + " at location " + loc);
>>>>> // Output:
>>>>> // Food crackers at location kitchen
>>>>> // Food apple at location kitchen
>>>>> // Food chocolate at location living room
>>>>> // Food chips at location living room
>>>>> end
>>>>>
>>>>> rule typo
>>>>> when
>>>>> Here( looc : location)
>>>>> ?editableThings(food : ?, loc : ?;)
>>>>> then
>>>>> System.out.println("Food " + food + " at location " + loc);
>>>>> // Output:
>>>>> // Food crackers at location kitchen
>>>>> // Food apple at location kitchen
>>>>> // Food chocolate at location living room
>>>>> // Food chips at location living room
>>>>> // looc is just an unused bound variable
>>>>> end
>>>>>
>>>>> On 20 April 2011 10:16, Geoffrey De Smet<ge0ffrey.spam at gmail.com>
>>>>> wrote:
>>>>>> Mark and I were discussing backwards chaining
>>>>>>
>>>>>> http://blog.athico.com/2011/04/backward-chaining-emerges-in-drools.html
>>>>>> on IRC and we 'd like your opinion on a design issue.
>>>>>>
>>>>>> The example
>>>>>> ========
>>>>>>
>>>>>> Let's say you have this data:
>>>>>> Location("crackers", "kitchen")
>>>>>> Location("apple", "kitchen")
>>>>>> Location("chocolate", "living room")
>>>>>> Location("chips", "living room")
>>>>>>
>>>>>> Let's say you have this code:
>>>>>>
>>>>>> query editableThings( String thing, String location )
>>>>>> Location(thing, location)
>>>>>> end
>>>>>> And then these 3 rules:
>>>>>>
>>>>>> rule outputinput
>>>>>> when
>>>>>> Here( loc : location)
>>>>>> ?editableThings(food, loc;)
>>>>>> then
>>>>>> System.out.println("Food " + f + " at location " + loc);
>>>>>> // Output:
>>>>>> // Food crackers at location kitchen
>>>>>> // Food apple at location kitchen
>>>>>> end
>>>>>>
>>>>>> rule outputOutput
>>>>>> when
>>>>>> ?editableThings(food, loc;)
>>>>>> then
>>>>>> System.out.println("Food " + f + " at location " + loc);
>>>>>> // Output:
>>>>>> // Food crackers at location kitchen
>>>>>> // Food apple at location kitchen
>>>>>> // Food chocolate at location living room
>>>>>> // Food chips at location living room
>>>>>> end
>>>>>>
>>>>>> rule typo
>>>>>> when
>>>>>> Here( looc : location)
>>>>>> ?editableThings(food, loc;)
>>>>>> then
>>>>>> System.out.println("Food " + f + " at location " + loc);
>>>>>> // Output:
>>>>>> // Food crackers at location kitchen
>>>>>> // Food apple at location kitchen
>>>>>> // Food chocolate at location living room
>>>>>> // Food chips at location living room
>>>>>> end
>>>>>>
>>>>>> The discussion
>>>>>> =========
>>>>>>
>>>>>> Both rules have the same statement:
>>>>>> ?editableThings(food, loc;)
>>>>>>
>>>>>> In the outputInput rule, "loc" is an input variable.
>>>>>> In the outputOutput rule, "loc" is an output variable.
>>>>>>
>>>>>> I am wondering if we don't need a visual demarcation that a variable is
>>>>>> an output variable,
>>>>>> to make it stand out of an input variable?
>>>>>>
>>>>>> Proposition 1: Suffix output variables with ":"
>>>>>>
>>>>>> rule outputinput
>>>>>> when
>>>>>> Here( loc : location)
>>>>>> ?editableThings(food:, loc;)
>>>>>> then ... end
>>>>>>
>>>>>> rule outputOutput
>>>>>> when
>>>>>> ?editableThings(food:, loc:;)
>>>>>> then ... end
>>>>>> rule typo
>>>>>> when
>>>>>> Here( looc : location)
>>>>>> ?editableThings(food:, loc;) // compiler error because input
>>>>>> variable loc is not declared
>>>>>> then ... end
>>>>>>
>>>>>> --
>>>>>> With kind regards,
>>>>>> Geoffrey De Smet
>>>>>> _______________________________________________
>>>>>> rules-dev mailing list
>>>>>> rules-dev at lists.jboss.org
>>>>>> https://lists.jboss.org/mailman/listinfo/rules-dev
>>>>>>
>>>>>
>>>>> _______________________________________________
>>>>> rules-dev mailing list
>>>>> rules-dev at lists.jboss.org
>>>>> https://lists.jboss.org/mailman/listinfo/rules-dev
>>>>>
>>>>
>>>> _______________________________________________
>>>> rules-dev mailing list
>>>> rules-dev at lists.jboss.org
>>>> https://lists.jboss.org/mailman/listinfo/rules-dev
>>>>
>>>>
>>>> _______________________________________________
>>>> rules-dev mailing list
>>>> rules-dev at lists.jboss.org
>>>> https://lists.jboss.org/mailman/listinfo/rules-dev
>>>>
>>>
>>> _______________________________________________
>>> rules-dev mailing list
>>> rules-dev at lists.jboss.org
>>> https://lists.jboss.org/mailman/listinfo/rules-dev
>>>
>>
>> _______________________________________________
>> rules-dev mailing list
>> rules-dev at lists.jboss.org
>> https://lists.jboss.org/mailman/listinfo/rules-dev
>>
>>
>> _______________________________________________
>> rules-dev mailing list
>> rules-dev at lists.jboss.org
>> https://lists.jboss.org/mailman/listinfo/rules-dev
>>
>>
> _______________________________________________
> rules-dev mailing list
> rules-dev at lists.jboss.org
> https://lists.jboss.org/mailman/listinfo/rules-dev
>
>
More information about the rules-dev
mailing list