[rules-users] Abstraction between rules and data model?
Mark Proctor
mproctor at codehaus.org
Mon Jun 2 10:12:41 EDT 2014
On 2 Jun 2014, at 13:59, Horváth Péter Gergely <h.peter at mailbox.hu> wrote:
> Hi Davide,
>
> Thanks, I just wanted to get a grasp of the time frame we could expect (whether it is a month or half a year etc.) :)
We haven’t had much feedback on traits, and the system is still in flux. So we cannot give any time frame. At some point there will be a momentum in the number of users, that will allow us to have a degree of confidence in it, but at the moment we’ve had close to zero feedback from users on this functionality.
But there is nothing to stop you using this now. The only difference is that we may or may not change it up until the point we call it “non-experimtal”. It won’t make any difference to the execution stability and bugs.
>
> A bit more about our use case: we have a massive old system, with basically all of the logic embedded into Java. We are in the process of refactoring and moving up-to-date Java technologies. We would like to introduce a rule engine so as to extract logic from the core of the application. Since we need to re-structure the whole back-end and persistence layer, we face a challenge with rules: as the object model changes, the rules have to change also, which simply would not be manageable. (Imagine something really crazy/complex here: 6-8-... levels of object hierarchy, grown most of the time organically during the last 15 years.) Deferring the introduction of the rule engine until the data model classes reach the final, ideal state is not something we can now afford, the rule engine and data model refactoring projects should be done in parallel, without the one paralyzing the other one...
>
> Take the following as an example (from the manual). Say, you have Applicant class, with age and name fields:
>
> public class Applicant {
>
> private String name;
> private int age;
>
> // getters/setters
> }
>
> Then, you could have a simple rule, requiring, that name must be supplied:
>
> rule "Supplied a proper name"
> when
> Applicant( name == null)
> $a : Application()
> then
> output.println("### RULE MATCH: Invalid application: No name specified!");
>
> $a.setValid( false );
>
> end
>
> But there is an issue: every single time you refactor / rename any field in your domain model, the rules are no longer valid. That is why we have a problem: we should start developing rules against a relatively fluid object model. Say Applicant.name is renamed to Applicant.fullName, then the existing rules are no longer valid. That is why I would like to have some sort of abstraction, since the rules do not really need to know the exact details of the underlying class.
>
> As a demonstration of my idea, I managed to hack together a custom parameterizable operator, that calls the method where the annotation contains the same parameter as in the rule. With this approach, changes to the field names would no longer affect the rules. Again, I have to highlight, I don't think this should necessarily be implemented in the core engine. What I am looking for is some kind of extension point allowing us to hook into the expression evaluation part of Drools. (Instead of a custom operator, something, where we can access the object and the expression being resolved so that we can implement our custom logic for returning the value of the expression)
>
> The rule now looks like this - note the custom satisfiesRule operation. The "Name of this Applicant" expression now abstracts away the name of the actual field (the second parameter is the operator for the test, please ignore it for now):
>
> rule "Supplied a proper name"
> agenda-group "evaluate-application"
> auto-focus true
> when
> Applicant( this satisfiesRule[ "Name of this Applicant" , "is equal to" ] null )
> $a : Application()
> then
> output.println("### RULE MATCH: Invalid application: No name specified!");
>
> $a.setValid( false );
>
> end
>
> The name is simply mapped by a custom annotation:
>
> public class Applicant {
>
> private String name;
> private int age;
>
> @MyBusinessExpression("Name of this Applicant")
> public String getName() {
> return name;
> }
> // getters/setters
> }
>
>
> I think having a level of abstraction between the rules and the data model would not only be useful for de-coupling, but also can make rule authoring easier for the business users. The custom operator (please see attachment) is a kind of hacky workaround, but demonstrates how something similar could be achieved at run time (without using DSL or any other rule transformation).
>
> Please let me know what you think. :)
>
> Cheers,
> Peter
>
>
>
>
>
>
>
>
>
>
> 2014-06-02 11:54 GMT+02:00 Davide Sottara <dsotty at gmail.com>:
> I can't guarantee a public date.. as a community member, I work on a "best effort" basis...
> I'll try to do it before the end of the month, though.
> For now, as a workaround, I would create derived getter/setter pairs that expose the desired
> computations. A concrete example of what you are trying to do exactly would also be helpful,
> feel free to contact me privately if you can't share your code here
> Best
> Davide
>
>
> On 05/28/2014 10:08 AM, Péter Gergely, Horváth wrote:
>> Thanks for the explanation, I was a bit confused because of the terminology; "virtual" is not mentioned in the docs. ;)
>>
>> Is there any plan for the public release of Trait property binding to a nested path? We would definitely need something like that in our environment. Or do you see any way we could hook into the property look-up mechanism? Based on what I know, I don't see any official extension point for that.
>>
>> My only idea would be using some Java proxy voodoo-magic to wrap objects before they are inserted to the session, but my gut feeling is that it would be a way to debug hell...
>>
>> What do you think?
>>
>> Cheers,
>> Peter
>>
>>
>>
>> 2014-05-27 19:42 GMT+02:00 Davide Sottara <dsotty at gmail.com>:
>> Consider that a trait is an interface applied to some class. In the context of the pair:
>> A "hard" field is a property (get/set) exposed by the interface AND the underlying class
>> A "soft" (or "virtual") field is a property exposed by the interface BUT NOT by the underlying class
>> A "hidden" field is a field of the underlying class NOT exposed by the interface
>>
>> Hard and Soft fields can be accessed using the interface, hidden fields are accessible using the map-like
>> construct fields[ "fieldName" ].
>>
>> This said,
>> the mapping is by default done using the property name and (then) the property type.
>> However, this mapping can be decoupled using the annotation @Alias() on either the class OR the trait.
>> E.g.
>> declare Core
>> name : String @Alias( "any-Id-or-even-an-IRI-here" )
>> end
>>
>> declare trait SomeTrait
>> label : String @Alias( "..." ) // if two "aliases" match, this will be considered a hard field
>> end
>>
>> The "accessor", i.e. the ability to bind a trait property to a (possibly deeply) nested path is what I'm working
>> on these days, I have the same requirement from another urgent use case
>>
>> For the time being, you can probably create a "shortcut" accessor pair in your implementation class,
>> to execute the complex expression, and @Alias it to the trait field.
>>
>> Please let me know if you find any issue/bugs and any feature request you may have!
>> Best
>> Davide
>>
>>
>>
>> On 05/27/2014 07:57 AM, Horváth Péter Gergely wrote:
>>> Hi Davide,
>>>
>>> Drools trait functionality is one of the powerful concepts which makes Drools a good candidate for the project. So keep up the good work! :) However I'm not sure if its current level of flexibility would be sufficient for our use case. I've checked the documentation, but haven't really found the term virtual field -- could you please elaborate on this?
>>>
>>> Do you think we could somehow hook into the evaluation of the aliases or the "fields" Map? Sometimes you would need slightly more than merely aliasing fields to something else; e.g. calculating values for the purpose of rule processing or extracting a value from a more complex object tree etc. Citing the example -- GoldenCustomer( fields[ "age" ] > 18 ) -- being able to get a reference to the target object and the field map expression "age" would be quite close to what I imagined. Our custom code could then perform the appropriate translation and return the requested value, hiding the fact whether "age" is an actual field in the Customer object itself/retrieved from an encapsulated complex object e.g. replacing expression "customer.personalInformation.birthData.age"/calculated on the flight.
>>>
>>> What do you think?
>>>
>>> Cheers,
>>> Peter
>>>
>>>
>>>
>>> 2014-05-26 17:58 GMT+02:00 Davide Sottara <dsotty at gmail.com>:
>>> We are working on the trait framework for cases like this. Essentially,
>>> it allows to use
>>> interfaces when writing rules AND to inject the interfaces dynamically
>>> at runtime,
>>> at the instance level. It relies on transparent proxies which wrap the
>>> data classes
>>> and implement the required interfaces. A simple field aliasing mechanism
>>> is provided
>>> (work in progress). For more complex transformations, "virtual" fields
>>> can be added.
>>> See section 7.7.8 of the manual for more details and let me know if it
>>> can help
>>> with your use case.
>>> Best,
>>> Davide
>>>
>>> On 05/26/2014 09:55 AM, Wolfgang Laun wrote:
>>> > Even a relatively sophisticated transformation would be easier to implement
>>> > and most certainly safer from changes in the unstable Drools API than some
>>> > hook-and-intercept mechanism built into Drools.
>>> >
>>> > Notice that violent structural departure of the model the BUs see from what
>>> > you call "persistence model" might make it impossible for the BUs to come
>>> > up with rules that can be transformed to match the other model at all;
>>> > if it is possible, rules might still incur a heavy performance penalty.
>>> >
>>> > It is (IMHO) a myth that "Rules" is a foolproof way of establishing
>>> > business logic
>>> > independent from the data model and application environment with which
>>> > this logic should be able to cooperate. As long as everything is kept in the
>>> > abstract (i.e., formulated in terms of mathematics) it will look good, but
>>> > any implementation may throw a spanner in the works, or worse.
>>> >
>>> > -W
>>> >
>>> > On 26/05/2014, Péter Gergely, Horváth <h.peter at mailbox.hu> wrote:
>>> >> Hi Wolfgang,
>>> >>
>>> >> Thank you for your input. You are right that some of the cases could simply
>>> >> be covered by regexp-replace, but I'm afraid, not all of them. Interfaces
>>> >> could also help, but we have a requirement that the business rules should
>>> >> not be tightly coupled to the underlying persistence model. (I understand
>>> >> that some might say this is not ideal, but that is our current situation)
>>> >>
>>> >> I am wondering whether it is possible to hook into Drools engine and
>>> >> intercept field value reference expression evaluations in run time (e.g if
>>> >> "foo.bars" is used in an expression, we could return "foo.barList") ? By
>>> >> injecting some custom code, we could make the necessary decisions and
>>> >> extract the proper value from an object. Unfortunately these parts of
>>> >> Drools are pretty much undocumented.
>>> >>
>>> >> Regards,
>>> >> Peter
>>> >>
>>> >>
>>> >>
>>> >> 2014-05-26 13:57 GMT+02:00 Wolfgang Laun <wolfgang.laun at gmail.com>:
>>> >>
>>> >>> What you describe can be done with /bin/sed.
>>> >>>
>>> >>> Notice that the DSL processor doesn't require you to translate entire
>>> >>> patterns; there is a mechanism for translating "keywords", which is
>>> >>> just arbitrary tokens to whatever replacement text.
>>> >>>
>>> >>> If a "bar" must be translated to a "barList" in the context of a class
>>> >>> "Foo" but not in any other context, a more sophisticated translation
>>> >>> is required in any case (with /bin/sed still being sufficient if
>>> >>> patterns aren't split across lines).
>>> >>>
>>> >>> Some say that good design makes use of Interfaces, which leaves room
>>> >>> for actual implementations being changed as long as the interfaces are
>>> >>> implemented. Here, note that rules can be written against interface
>>> >>> types.
>>> >>>
>>> >>> -W
>>> >>>
>>> >>>
>>> >>> On 26/05/2014, Péter Gergely, Horváth <h.peter at mailbox.hu> wrote:
>>> >>>> Hi All,
>>> >>>>
>>> >>>> We are evaluating Drools 6 for our use case, and face challenges where
>>> >>>> we
>>> >>>> would need some ideas from more experienced users of Drools.
>>> >>>>
>>> >>>> We have an application with a massive code base and a large number of
>>> >>> model
>>> >>>> (entity) classes. We are in the process of moving away from inherited
>>> >>>> legacy technologies and refactoring the old code base. As a part of
>>> >>>> this
>>> >>>> work we would like extract some of the hard-coded business logic to
>>> >>>> external rules, that is why we are looking at Drools as a potential
>>> >>>> solution.
>>> >>>>
>>> >>>> What we would like to have is some kind of abstraction or mapping
>>> >>>> between
>>> >>>> actual entities and rules the business users can define so that they do
>>> >>> not
>>> >>>> have to know the _exact_ details of the data model (field names,
>>> >>>> precise
>>> >>>> relations etc). This would be important for us so that we can refactor
>>> >>> the
>>> >>>> old model classes without affecting business rules; also it would make
>>> >>> life
>>> >>>> easier for the business users. While IDE support might make refactoring
>>> >>>> easier, we definitely want to have a separation between rules and
>>> >>> entities.
>>> >>>> Given our situation, writing and maintaining "stable" wrapper/adapter
>>> >>>> classes for the sole purpose of rule processing is out of question. I
>>> >>> have
>>> >>>> checked the documentation of Drools DSL support and for me it seems to
>>> >>>> be
>>> >>>> overkill for our use case: we do not really need a custom language, but
>>> >>>> simply an abstraction between rules and the data model classes.
>>> >>>>
>>> >>>> What I could imagine is a piece of code, (a custom property resolver? -
>>> >>> no
>>> >>>> sure how it is called) which maps property expressions to actual
>>> >>> properties
>>> >>>> based on a custom annotation on the entity class or something like
>>> >>>> that,
>>> >>> so
>>> >>>> that a rule containing "Foo.bars" expression does not have to change
>>> >>>> even
>>> >>>> if we decide to rename "Foo.bars" to "Foo.barList" in the model
>>> >>>> classes.
>>> >>>> (This was just a simple example of a potential use cases)
>>> >>>>
>>> >>>> Could you please share your thoughts on this topic and point me into
>>> >>>> the
>>> >>>> right direction?
>>> >>>>
>>> >>>> Thanks,
>>> >>>> Peter
>>> >>>>
>>> >>> _______________________________________________
>>> >>> rules-users mailing list
>>> >>> rules-users at lists.jboss.org
>>> >>> https://lists.jboss.org/mailman/listinfo/rules-users
>>> >>>
>>> > _______________________________________________
>>> > rules-users mailing list
>>> > rules-users at lists.jboss.org
>>> > https://lists.jboss.org/mailman/listinfo/rules-users
>>> >
>>>
>>> _______________________________________________
>>> rules-users mailing list
>>> rules-users at lists.jboss.org
>>> https://lists.jboss.org/mailman/listinfo/rules-users
>>>
>>>
>>>
>>> _______________________________________________
>>> rules-users mailing list
>>> rules-users at lists.jboss.org
>>> https://lists.jboss.org/mailman/listinfo/rules-users
>>
>>
>> _______________________________________________
>> rules-users mailing list
>> rules-users at lists.jboss.org
>> https://lists.jboss.org/mailman/listinfo/rules-users
>>
>>
>>
>> _______________________________________________
>> rules-users mailing list
>> rules-users at lists.jboss.org
>> https://lists.jboss.org/mailman/listinfo/rules-users
>
>
> _______________________________________________
> rules-users mailing list
> rules-users at lists.jboss.org
> https://lists.jboss.org/mailman/listinfo/rules-users
>
> <BusinessFieldEvaluatorDefinition.java><MyBusinessExpression.java>_______________________________________________
> rules-users mailing list
> rules-users at lists.jboss.org
> https://lists.jboss.org/mailman/listinfo/rules-users
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.jboss.org/pipermail/rules-users/attachments/20140602/c9fa8136/attachment-0001.html
More information about the rules-users
mailing list