On 21/08/2011 13:45, Wolfgang Laun wrote:


On 20 August 2011 21:05, Michael Anstis <michael.anstis@gmail.com> wrote:
I like Mario's suggestion.

Use of a switch statement could also be an option (to muddy the waters further):-

Any good reason for calling a "switch" what is usually an if/elsif? Also, would the branches have an implied "break"? Or should any number of branches fire for a single match of the CEs preceding the "switch"? If so, what if "a1" (in the example) retracts tha fact matching "D()", and then "a2" needs to be executed?

Anybody making proposals is cordially invited not just to provide a syntax silhouette.

Thank you
-W
 

rule R1
    when
        D()
        switch
            A()
              then
                  a1
            B()
              then
                  b1
            C()
              then
                  c1
        switch
    then
        d1
end

I definitely don't like the ">" "{..}" notation.

Cheers,

Mike

On 20 August 2011 00:34, Mark Proctor <mproctor@codehaus.org> wrote:
For further info, this is what ilog do. They only allow an implicit "else" on the last "evaluate" expression. Which while simple is quite restrictive.:
http://publib.boulder.ibm.com/infocenter/brjrules/v7r1/index.jsp?topic=/com.ibm.websphere.ilog.jrules.doc/Content/Business_Rules/Documentation/_pubskel/JRules/ps_JRules_Global1895.html
rule ruleName {
   when 
      {conditioni evaluate (expression)}
   then 
       {[action1 ... actionm]}
   else 
       {[action1 ... actionp]}
}; 

OPSJ is the only engine that i know of that uses labels:

x.name == obj,
x.weight == "light",
x.location != g.location);
mky: monkey;
[4] (mky.holds == obj);
} do {
makegoal("walkto", loc);
} else( 4 ) {
makegoal("holds", obj);
}
Although I would encourage people to think beyond simple "if/else", the proposal I put forward would allow for tree like data flows for signal processing - which will map very nicely to GUI tooling.
I proposed something like this, based around the OPSJ labels approach. It helps if you think of a "then" block as a function literal and the {func_lit_name} as an inline code fragment that calls the function literal.
So this:
$o : Object() from stream
( A() | {a} from $o or
  B() | {b} from $o or
  C() | {c} from $o )

becomes:
case( Object() from stream; // notice the semi colon to separate the object that specifies the default 'from'
          A() | {a},
          B() | {b},
          C() | {c} )
then
...
then.a
....
then.b
....
then.c
end.

The above would currently work like 'or'. All matching branches would activate. I thought about a short cut or. I think the answer there is to build a prolog like 'cut' operator, which we need for queries anyway. However adding 'cut' is far from trivial as it'll involve substantial changes to the current Rete algorithm, but assuming we could eventually make those changes, maybe something like:
case( Object() from stream; // notice the semi colon to separate the object that specifies the default 'from'
          A() | {a} | cut,
          B() | {b}| cut,
          C() | {c} | cut)

When thinking about the | symbol, think of unix pipes. The pipe indicates to filter the output of the pattern, the pipe allows chaining of these "filters", much like unix, such that the propagation then goe sfrom the labelled "functional literal" to the cut CE. I plan to allow the | symbol to be used to integrate user pluggable filters, such as "distinct". The following will filter all A() matches using a distinct filter, making sure only distinct matches are execute by the "a" functional literal. The cut symbpol ensures that only the first match is found, all other matches (join attempts) will be stopped - including those in the various 'or' branches if they exist.
A() | distinct | {a} | cut

For consistency we will allow time windows to be used as filters. Over the last 30 seconds find the distinct A and activate on it, but do not find any more matches after that.
A() | time.window(30s) | distinct | {a} | cut.

Hopefully that shows why I want to introduce a general purpose "|" pipe symbol for connecting "filters" and use that in a very generic way. Rather than introducing various different keywords.

This syntax is not aimed at business users, who will be using the guided editor anyway and the verbalisations that come with that, even more so true once we start to add natural language. So I'm not overly keen to restrict or limit things for a group of people won't use the low level syntax anyway.


Mark

Mark



On 19/08/2011 13:56, Toni Rikkola wrote:
I got the same feeling that Geoffrey had about readability.

We added "from" its really easy to get, why not add "else".

when
    Person( name == "darth" )  else  [darthIsMissing] 
    A()
then
   ....
then.darthIsMissing
  log("Darth was never found");
end

or

when
    Person( name == "darth" )  else  { log("Darth was never found"); }
    A()
then
   ....
end

"Inline then" could be done with inner rules. Similar to what Mario suggested.

rule "Handle Login"
  when
    $loginRequest :LoginRequest()
    AuthorizedUsers( list contains $loginRequest.user ) else [unsuccessfulLoginAttempt]

      inner rule "Check if Admin"
         $p :AdminRights( user == $loginRequest.user )
      then 
         showAdminMenu(); 
       end

  then
    logInUser( $loginRequest.user );

  then.unsuccessfulLoginAttempt
    log( "There was and unsuccessful login attempt with the user name " + $loginRequest.user.name );    
end

Toni

On Aug 19, 2011, at 2:59 PM, Geoffrey De Smet wrote:

I like Mario's proposal because I can actually read it.
Those special chars | < are gibberish to me.

The only reason we're not debating to use a new readable, intuitive keyword, is because of the back-wards compatibility issues involved.
But using unreadable, unintuitive special char just for that, is probably not a good idea.
I wonder if we reserve new keywords by prefix them with reserved special char like "@"?
Then we can introduce as many keywords as we want without breaking backwards compatibility.

Who's our target users for DRL authors?
A) Supersmart computer science guys
B) Blue collar Java programmers
C) Domain experts (= not programmers)

I 'd classify "{notA} < A()" as (given some time to learn it) readable for A, but not for B and C.

Op 18-08-11 23:35, Mario Fusco schreef:
Hi Mark,

Since you're gathering 2 cents here and there I decided to add also mine even if I am pretty sure that I am still missing the whole picture and anyway at the moment I cannot see all the consequences of what I am going to propose.

To tell you the truth I find the label syntax not very intuitive and I was wondering if we could avoid it in some way. In the end what the 90% of the users are asking for is just something like:

rule R
    when
        A()
    then
        do something
    else
        do something else
end       

while we are going to give them something that is not exactly the same:

rule R
    when
        {notA} < A()
    then
        do something
    then.notA
        do something else
end       

In particular I was thinking if we could keep the when ... then ... else syntax that should be familiar to the biggest part of the users and at the same time obtain a flexibility similar to the one provided by the labels syntax. Probably we could do it with a kind of nested rules so, for instance, the rule:

rule R1
    when
        {af} < A() > {at}
        B()
    then
        DO
    then.af
        DO.af
    then.at
        DO.at
end

could be rewritten as it follows:

rule R1
    when
        B()
    then
        DO
        rule R1A
            when
                A()
            then
                DO.at
            else
                DO.af
        end       
end           

Of course the nested rule couldn't be used by the Drools engine as it is, but we could implement a kind of "linearization" process at compile time that translates it more or less as:

rule R1_1
    when
        A()
        B()
    then
        DO
        DO.at
end

rule R1_2
    when
        not A()
        B()
    then
        DO
        DO.af
end

In the same way the "or" example:

rule R1
when
    (     A() > {a1} or
        B() > {b1} or
        C() > {c1} )
    D()
then
    DO
then.a1
    DO.a1
then.b1
    DO.b1
then.c1
    DO.c1
end

could be written as:

rule R1
    when
        D()
    then
        DO
        rule R1A
            when
                A()
            then
                DO.a1
        end       
        rule R1B
            when
                B()
            then
                DO.b1
        end       
        rule R1C
            when
                C()
            then
                DO.c1
        end
end       

and then linearized at compile time in a similar way as I wrote before.

Once again I still haven't evaluated all the implications of my suggestion neither I know if we can cover with it all the cases proposed by Mark. I am pretty sure I am missing something important to be honest, but since we are in a "brainstorming phase" I thought it could worth to consider it at least.

My 2 cents,
Mario


_______________________________________________
rules-dev mailing list
rules-dev@lists.jboss.org
https://lists.jboss.org/mailman/listinfo/rules-dev


-- 
With kind regards,
Geoffrey De Smet
_______________________________________________
rules-dev mailing list
rules-dev@lists.jboss.org
https://lists.jboss.org/mailman/listinfo/rules-dev



_______________________________________________
rules-dev mailing list
rules-dev@lists.jboss.org
https://lists.jboss.org/mailman/listinfo/rules-dev



_______________________________________________
rules-dev mailing list
rules-dev@lists.jboss.org
https://lists.jboss.org/mailman/listinfo/rules-dev



_______________________________________________
rules-dev mailing list
rules-dev@lists.jboss.org
https://lists.jboss.org/mailman/listinfo/rules-dev




_______________________________________________
rules-dev mailing list
rules-dev@lists.jboss.org
https://lists.jboss.org/mailman/listinfo/rules-dev