Rule Object: A Pattern Language for Adaptive and Scalable
Business Rule Construction
Part 1: Rule Object
Ali Arsanjani
IBM Enterprise Java Services National Practice, Raleigh, NC, USA
Maharishi University of Management, Fairfield, Iowa, USA
Arsanjan@us.ibm.com
Abstract
Rules are changing everyday in the face of rapidly volatile business requirements. How do we handle this
change while keeping our systems efficiently maintainable, reusable and extensible? How do we model and
handle (represent) rules, for greater reuse, maintainability, performance?
Business rules tend to change more frequently than the rest of the business object with which they are
associated. These rules are typically implemented within the rule methods of a business object. Rules also
refer to other business objects that their encompassing business object associates with; creating a web of
implicit and increasingly unmaintainable dependencies. Thus, changing a business rule can impact the set
of objects dependent upon that rule. Entropy increases even more when the code that is implementing a
rule is scattered across several methods within a class, or across several methods of collaborating classes.
This lack of centralization leads to ripple effects; the impact of changing a rule’s constituent if-else
statements leads to side-effects.
The Rule Object Pattern Language contains eighteen patterns, which, for brevity have been excluded from
our discussion1. We have included only the major one, Rule Object and have mentioned two others,
Assessor and Action, that are intimately related to Rule Object. This pattern language balances the forces
in the above problem domain by providing a sequential unfoldment of a set of patterns that address the
increasing need for handling scalability, adaptability and complexity. The Rule Object pattern language
can best be seen to resolve forces in the greater context of the Service Provider domain pattern
[Arsanjani99a;b].
Introduction: Rule Design
Rules are encountered in most portions of software systems. They are of various types, scopes and scales.
They are often changed, and changed often, to accommodate new requirements. Various authors have
provided taxonomies of rules, particularly business rules [Odell96]. Rules also tend to evolve along with
the rapid pace of business change and evolution.
Let us define a Simple Rule, within the context of a programming language, to be of the form: “if
<condition> then <action> else <action>”. Long chains of nested if-else statements in various parts of code
that implements a class tend to clutter the code and render it virtually un-maintainable as repeated changes
are made over time. In order to avoid ripple effects, based on the principle of variation-oriented design
[GHJV95], [Arsanjani99;b], we encapsulate that which tends to change frequently and in a non-uniform
fashion. We therefore need to extricate these chains of if-statements and case-statements from within the
application layer code (e.g., business logic); to enable rapid alterations to rule structures as dictated by
business needs.
This allows us to insulate the system, to a large degree, against ripple effects that result from changes to
individual objects and sets of collaborating objects, collectively fulfilling a business objective (clusters).
Rules come in various types.
Here is a list of scenarios that provides motivation to define a taxonomy of rule types based on complexity,
scale, layer and function. Specifically, 1-7 address tier-specific types of rules and 8 contains a list of nontier
specific types of rules.
1. Field on a form needs to be validated for non-blank, correct domain of values, etc. Objects check these;
UI sends the app object a message to check this data value.
2. Set of interdependent fields must be validated collectively.
3. Compound data spread across multiple forms needs to be cross-validated; data in one form should not
violate another data value in another, related form. Implemented using a mediator pattern whom
checks to see if all data are consistent without needing the forms to know about one another.
3.1. Invalid combinations need to be identified and reported or disallowed altogether
3.2. Valid combinations tested for validity
4. Checking with policies and verifying business rules in the middle tier constitutes the bulk of
application (or “business” ) logic; which may be accessed by many types of clients: such as PDAs,
thin-clients, fat clients, etc.
5. Checking rules at the database layer in the form of triggers and stored procedures.
6. Middleware communication rules.
7. A variety of security-related rules; authorization, authentication, non-repudiation, etc.
8. Regardless of tier or layer s in which they reside, rules can pertain to computations (functions,
mappings, transformations usually expressed or expressible via mathematical formulae), for example:
When a Coverage is selected (added to a policy) it may have causes of loss associated with it. These are fixed, and
cannot be changed. If a coverage has causes of loss, when the coverage premium is calculated, it asks each cause
of loss to calculate its contribution to the premium and sums them. If there are no causes of loss for a coverage,
then the coverage has a premium of $0.00.
8.1. some restrict and constrain business processes and data that are input to drive them;
8.2. some pertain to the workflow order and “Route”, “Rules”, “Roles”;
8.3. some constitute “terms of agreement” or “conditions”, such as “service level agreements” which
are more legal in nature
8.4. The enforcenment of certain types of rules results in “positive inclusion”: a Coverage is valid if
and only if …. Sometimes, however, we have “negative inclusion”: for example, “If the zip code
of the applicant is within <list of high-fraudulent zip codes> then we will have to do a pre-paid
calling card; other times we have “invalid combinations” the conjunction of a set of conditions
results in an invalid state: “if the loan requester has a second house and the primary house is
within a high-risk earthquake zone and they have no insurance, then we will have to reject the
loan if credit is below <xxx> amount”.
9. The design of rules and their implementation are two separate things: e.g., rules can be designed as
business constraints and implemented in the data layer via referential integrity rules that can be
modeled as specific types of associations/relationships.
Pattern 1: Rule Object
Intent
Make the design and implementation of computerized business processes extensible and adaptible, without
endangering them with intrusive changes, by making the rules governing them pluggable.
Motivation
Example 1
We want to write a program to allow conversion of temperatures from Celsius into Fahrenheit and you
wanted to be able to go both ways : F_ C and C _ F.
This seems deceptively simple at first blush. Actually, there are a number of “rules” that arise when tryoing
to implement this design: “We would like to enter in a value and click OK and have the other textfield
display the temperature in the other scale. We may implement it as follows. We have two text boxes (one
for C and one for F) and one button (to start the computation) . You would enter the number into one text
box and click on the Compute button. The process would check to see whether there was a value in the
other “converted” text box. If there is a value, is it a valid conversion? If not (or if it is blank (no value in
it), reconvert, other wise, give a message saying “Already converted.” “
So you need to do validations and checks before perfroming actions. This is a rather trivial example of a
more important problem in larger software systems: objects usually do not carry out actions (execute
methods) “mindlessly”; rather, some condition is usually checked and if those series of conditions apply
then you can perform the action.
Next we decide to add the Kelvin scale. We perform checks to see if a value was entered in a given
textfield and whether the other two were blank. If they were not to check to see if they were valid
conversions of each other before prompting the user to clear the textfields that do not have focus.
Imagine what would happen if instead of temperature, the fields were currencies; the problem would be the
similar and the solution would be analogous.
Example 2
You are designing a Property Insurance Application. The Insurance company has a number of client s(the
“insured”) who have purchased a number of policies. Each policy has a number of Coverages associated
with it by which the Insured will protect their buildings against some Cause of Loss (COL). Buildings are
located on a given premise within a geographic region. Coverage is specialized in four types: Building
Coverage, Personal Property Coverage, Debris Removal and Special Provision. Each of these subclasses
have a set number of COLs associated with them.
At the heart of the requirements, are the business rules. You will find countless rules governing a business
application such as this. These are often scattered across tiers: GUIs, middle-tier business logic supporting
business processes and database tiers may each have their own set of rules.
Let’s consider a few business tier rules that do not require a lengthy explanation of the domain:
_ There can be no duplicate coverage.
_ In order to add a Personal Property Coverage to a Policy, there must be an existing Building Coverage.
_ Or take InflationGuard; this is an additional coverage that can be used if a Building Coverage is part of
the Policy and the Policy does not already contain a Debris Removal Coverage.
_ Policies should not have Coverage with overlapping dates.
Policies, Coverage and rules associated with individual Coverage are often changing. The business needs to
change rules to ensure market share and profit; not to mention survival by prudently changing business
processes to accommodate rapidly changing business needs and competition. In order to rapidly change the
design of business requirements, analysis invariants (collectively, business rules) , we need to locate and
modify rules without suffering side-effects. We may typically want to restore the previous conditions of the
rule at a later time, for example when a promotional offering expires. Thus the requirement for variation
and change, entails the requirement for pluggability and adaptability. But the rules must be well-organized
in order to do this. If rules are scattered throughout the application, then changing them and anticipating
side-effects can be costly, cumbersome and in some cases infeasible. Business process adaptability is thus
compromised.
Instead of using a hard-coded set of if-statements to enforce the rule, that is hard to maintain and change,
we setup a tiny framework which will allow us to do rule checking by plugging-in conditions and actions as
the business sees fit.
Rules, however, do not change in their entirety; new conditions need to be applied to augment old
conditions, new or modified actions need to be replaced or recombined in new combinations. Therefore, we
the requirement to componentize rules for pluggability emerges. We can start by encapsulating their
constituent elements: conditions, actions, properties passed to them (context), result objects along with the
rule itself as an adaptable component.
This furnishes us with the ability to rapidly change, extend, recombine and reuse rules and rule constituents
and components. For example, it would be convenient to be able to plug in some conditions to be checked
for a new type of Property Insurance, say Personal Property. If the new conditions apply, we would like to
perform some action, even something as simple as displaying a warning message to inform the user that
they have entered overlapping Coverages that cannot be inserted into the same policy.
Typical Business Rule Implementation
Rules are usually implemented within the body of various methods that an object implements. These
typically scattered and often-nested If-else statements tend to clutter code and render it unitelligible; thus
making it difficult to maintain.
Rules as Methods (Rule Method)
To be properly implemented, Business Rules that are first identified will be listed, categorized and then
implemented by a single method. This typcially returns a Boolean value to acknowledge whether the rule
applied or not. If the rule is complex enough, this rule method may call other rule methods and try to
enforce the rule through collaboration with the rules of other business objects. So, for example, in our
property insurance case study we may have something called a Debris Removal Coverage. This is an
additional coverage that can be used if:
} building coverage is part of the policy, and
} inflation guard is not part of the policy, and
} a SpecialProvision for business service has not been applied, and DebrisRemoval’s limit has been
increased.
A Brief Look at Rule Object’s Peer Patterns
Peer Patterns2
Rule Object acts as a Mediator to Assessors and Actions; Properties (Context) and Results. Rule Objects
determine which Assessors should be used in the assessment of Properties, with a possible recording of
Results in an Assessment result. If the assessment was successful, the appropriate Actions are invoked,
possibly updating or changing the State of
Assessor
Assessors encapsulate and evaluate conditions based on a set of input Properties and record the results of
the evaluation within their AssessmentResults. Assessor is similar to Command. A Command executes an
operation whereas an Assessor assesses a set of conditions based on input State or Properties. The
difference is that it has an evaluate() method rather than an execute(); and collaborates with a Property List and a ErrorLog or ResultSet
ACTION
An Action executes by updating State or invoking behavior in other collaborating objects. It records the
result of its actions in an ActionResult which may contain an ErrorResult or ErrorLog. Actions can be
implemented as Strategies, Visitors, Interpreters, Commands. Here is the strcuture of its cluster:
Actions are related to Assessors in the following way:
Sequential Unfoldment of Rule Object: Simple Rule Object
How can you design something that you can use in a simple case and yet start scaling your design?
Rule Object has a structure that is best understood through a sequential unfoldment of complexity that
attempts to balance forces that require these additional layers of complexity and functionality.
Make the rules governing an object’s behavior extensible and adaptable through transparently attaching
rule objects, each one representing a rule the object has to enforce in its domain. The object manages its
rule object(s) dynamically. By representing rules as individual objects, different business process flows are
kept separate from the rules governing them and their interactions. Thus the process of changing them is
simplified.
Rule Object starts out with a simple Validator. This uses methods for conditions and actions. The Validator
has a method which accesses its conditions methods and if they are all evaluated to true, calls one or more
action methods.
The next step is a Simple Rule Object. Now you find yourself having to define many methods or complex
methods for assessing conditions; and these conditions may start changing; so you would like to encasulate
the methods so you can change them easily. The same goes for actions. So you decide to reify the
conditions as an Assessor(s) and the actions as an Action(s). The Rule Object acts as a Mediator ([Gof])
between the Assessors and the Actions. It may use a Factory (Builder or Abstract Factory or plain Factory
Method) to create appropriate Assessors and Actions from Serializer Objects (see [Riehle98]). Other
patterns in the Rule Cluster pattern language provide solutions to “hash and cache” the Rule Object or its
Constituent Assessors and Actions so they may be created or plugged into the Rule Object on-demand.
Now you have the following design. Please note that SimpleRuleObject is an abstract class, Assessor and
Action can be interfaces or abstract classes with some default behavior.
Cluster
Telecommunications Provisioning where a nested set of Rule Objects were needed along with their nested
set of appropriate Assessors and Actions.
Forces
_ Changes to business requirements entails changes to the design and implementation of rules.
Business rules are expected to change more frequently than the rest of the business object. At the
analysis, architectural, design and implementation level, Business Rules may need to be redefined and
updated to reflect the changes in policy and business. Business Rules tend to evolve over time as a
result of new business requirements. Changes to existing rules must be such as to leave the integrity of
the system of rules intact and in a consistent state. Consistency implies that modifications to a set of
conditions not adversely impact the rest of the system and produce unwanted side-effects.
_ Rules may be time-sensitive. This is when the rules in a domain are time-constrained; for example,
they may pertain to Service Offerings that are offered for only a limited duration for a promotion. Such
domains have a rapidly expanding and changing set of rules that may frequently change; sometimes on
a day-to-day basis. Although necessary, changing rules in a program are typically costly, as these
changes are usually intrusive and will thus have side-effects that will require further debugging and
testing. Making intrusive changes to production code is unsafe and costly; the potential of sideeffects
leads to the need for extensive regression testing and re-certification of components.
_ Rules should be centralized, making them easy to locate and change. Frequently, Rules are found
scattered throughout the design and implementation; typically in nested if-then-else clauses with
many dependencies. Tracing rule requirements to rule design and implementation can be a very errorprone
and resource-intensive project.
_ Rules should be de-centralized and assigned to their relevant classes and clusters. Objects have
manners. Manners govern the ways in which their methods should be invoked to guarantee a consistent
and valid set of states. Thus, rules can be assigned to classes. Each such class (aka, business object)
has a set of methods. Rules govern their valid state, and valid sequences of message-sends to their
collaborators and self methods. Rules usually check the state of an object or set of collaborating
objects. They operate on data that has been submitted to them via a context. We need to track the status
of the application of rules to this set of data (properties). Rules should know about relevant
properties (data); but the properties should be oblivious to the rules that check their collective
integrity from a business perspective.
_ Rules should be scalable. The same rule structure designed for a small application should scale up to
the needs and non-functional requirements of a larger application. They start out in a deceptively small
context and pretty soon need to be scaled up to handle larger scale transactions. Thus, rules should be
simple yet designed to be scaleable.
_ Non-uniform treatment of rule types. Treating different types of rules differently leads to a variety
of unwanted consequences. We need to strive to have uniformity with respect to the fact that every
business object should carry its own manners. There are different types of rules from different
perspectives that apply to different layers in the user-interface, application logic tier and persistence
layer. Rules need to have a holistic view of all their types whether they are a set of data elements;
knowing (report back) their valid and invalid combinations and values or complex business logic at the
middle-tier of an application server.
_ Code clutter: Nested If-then-else statements tend to clutter code and make it difficult to maintain.
Rules are frequently implemented as [nested] if-then-else structures. Rules are usually grown through
piecemeal growth [Foote96] rather than designed top-down.
_ Architectural Layers and Rules: Different rules may apply at different layers of the architecture.
Rules may apply differently at each layer: GUI, Webserver, Application, Middleware and Database.
Rules are present at different layers and require different mechanisms for their implementation; simple
validations in the GUI, complex cross-rule checking at the application layer, stored procedures and
triggers at the database layer.Yet their essential structure needs to be similar to be easily traceable
and uniformly applicable.
_ Business domains contain business processes that are governed by a set of business rules. These
rules are captured as part of company policy and workflow; operations and procedures. They need to
be standard procedure, but also need to be updated frequently. They are frequently modeled in
information systems whose design needs ot change along with the needs of the business.
_ Rules need to be created by and visible to management; rules and rule changes need to be visible
to programmers . Programmers implement business rules in code and must change them when
executives change them. Management must know which rules have been implemented.
_ Rules are the valid ways in which objects within a domain are allowed to interact and change state.
They should thus be considered as first-class constructs of the object paradigm and be identified
very early in the analysis process; not as an afterthought. In this new paradigm, in addition to having
identity, state and behavior, a class has manners (i.e., rules that govern the behavior or methods).
Manners are rules plus methods (object behavior) plus the meta-data needed to apply the rules and
govern the interaction and collaboration protocols of an object.
Applicability
Use Rule Object
_ to add/modify conditions, actions and rules to business objects dynamically and transparently, that is,
without affecting other objects or rules.
_ when complexity and scalability makes the use of Rule Method obsolete. Rules can be implemented as
if-statements or as methods (e.g., isCompatible()) returning boolean. When they grow beyond a
proportion of simple checks, it is time to reify them into Rule Objects. This is especially tru eif
they tend to change frequently.
when maintaining a system by implementing changes of requirements that can be captured as rules,
invariants, business requirements (predicates and conditions), policies, terms of agreements, etc. and
we would like to make non-intrusive changes; to reuse existing conditions and actions or create a new
rule that is “slightly” different from an existing one, but which will only be in effect for a “short
diration” (highly volatile requirements).
_ Business objects should know their own manners: how to use their methods in concert with other
collaborating business objects; what are valid states and what are invalid combinations of states among
business objects. Thus, the laws governing the use of methods, the meta-data that may be required to
store this information and the reification of the conditions and actions that embody these “laws” are
collectively called a business objects’ “manners”.[Arsanjani 99;a] Instead of rules being solely
accessible through their expression as logic within each business object, rules are collated together into
their own objects to facilitate non-intrusive changeability and pluggability. Business objects would
then contain a set of business rules.
_ When features that a set of business objects need to portray need to be adapted and customized to meet
different scalability and complexity constraints. Rule Object can be reduced to a simple Validator for a
GUI data entry field or it can scale to a Composite Rule Object with Composite Assessors (Conditions)
and Actions; with Properties, and Results. Here is the full spectrum.
Structure
Business rules are expected to change more frequently than the rest of the business object. The impact of
these changes will be minimized if the rules are encapsulated in their own classes and held separately;
ready to be plugged in and reused. Therefore, reify rules and their conditions and actions; making them
interchangeable and pluggable.
Maintain rules by making non-intrusive changes to Rule Objects. Add or change existing conditions and
actions; add properties (aka, a context) to be inspected for valid or invalid combinations (State) by using
Rule Objects. A Rule Object Cluster is a Composite of Condition (Assessor) and Action Composites along
with their helper classes. These helper classes are Properties that provide a name value set of “fields” that
the Condition(s) (Assessor) much check. The result of this Assessment may be recorded in an Assessment
Result. If the Conditions all apply, then Actions are performed to change the state of the object or other
objects. The Result of the application of these Actions can be recorded in Action Results for reporting or
analysis purposes.
Participants
_ Abstract Rule Object
defines the interface for objects that can have rules added to them dynamically.
_ Concrete Rule Object
defines an object to which additional rules can be attached.
_ Rule Object
Mediate between conditions and actions; apply rules
_ Assessor
check conditions and store results
_ Action
perform actions based on success or failure of corresponding Assessors; record results and change state
of collaborating objects
_ Properties (aka Context)
pass in information and State for evaluation (Assessor) or update of State (Action)
_ Result
Provide a super class for AssessorResult, ActionResult and Errors so other participants can record
results and report back to client.
_ AssessorResult
The specialization of Result
_ ActionResult
A specialization of Result used to record results of actions
_ MediationStrategy
Determine how we will execute the Assessors and Actions or even Rule Objects in the case of
Compound Rule Objects, Assessors and Actions.
Simple Rule Object: Static View
Compound Rule Object: Static View
The Rule Cluster holds a cluster of Composite Rules; each with their its potential Composite Conditions
and Actions. A Simple Rule is a leaf node of the composite and can exist by itself to handle (for example)
GUI field editing and validation.
Figure 10: Static View of Compound Rule Object and its Composite Hierarchy
Here is an alternative view of the above diagram. The concentration is on the dynamically pluggable nature
of the Rule Object and its constituent parts rather than on its Composite nature as depicted above.
Figure 11: Alternative View of Compound Rule Object
Collaborations
Here is a set of sample collaborations:
1. Set up rules (e.g., cache in a
hashtable: “cache and hash”)
2. Submit candidate state (pass in
property or hashtable or just parameters
for state you want to check consistency
of)
3. Check conditions on submitted state
using Assessor(s) . An Assessor will
usually go through the list or hashtable
and check each condition, or have a
Strategy that will check each condition
using an algorithm that in the simplest
default case is round robin, but you can
choose or define your own optimized
Assessor algorithm by creating, extending a given Assessment Strategy.
4. For each set of assessors, you have a corresponding set of Action(s) that must be performed or
ResultState or ErrorLog that must be written to, created, reported back. This may be a Null Object by
default for non-important or unimplemented ErrorResults or ResultLogs.
The following sequence diagram depicts the Set up Rules step above:
Figure 12: Setting Up Rules
Here are some variations in default collaborations:
One Strategy for Rule Object: (many to many)
If (assessorList.assess(inState))
actionList.perform(inState, outState)
else errorLog.reportOutcome(outState);
Simple Strategy: (one for one)
if (assessor.evaluate(inState))
action.performAction(inState, outState); else
errorLog.reportErrors(outState);
ValidCombinations:
A Combination is a State which is coupled with a
Strategy and a Rule Object. An initially InvalidState
State is transitioned into a ValidState if the rule
applies. Otherwise, errorLog records the fact and
reasons for the invalid State.
Invalid Combinations:
If this inState is an InvalidState then
errorLog.logThisAsInvalidState();
Go through all Rule Objects (Using specified or
default [optimizing] Strategy). Determine if this
inState matches an InvalidState Combination, if so,
reportInvalidState using the
outState = assessor.assess(inState);
if (outState.isINvalidState())
errorLog.reportErrors(outState) else continue; //
with next check for next possible invalid state
Consequences
1. Rules become more easily changeable and reusable; simpler to maintain. Non-intrusive changes can be
made to maintain rules and their Assessors and Actions. The individual Assessors and Actions can be
potentially reused in multiple different scenarios. Systems built this way tend to adhere more to the
15
open/closed principle [Meyer84]. Changes to Business Rules have much less of a ripple effect; they
are encapsulated within a Rule Object. New Rules can be added by adding more Rule Objects, by
creating a Compound Rule within a Rule Object context or by changing the Strategy for a Validator in
the case of a Simple Rule. [Client has Abstract Rule Object].
2. Rules become scalable. As Composites, their rules, assessors and actions can be stored in a database or
cached to account for increased volume, demand and availability, in proporation to the growth of
individual rules.
3. Uniform treatment of Rule Types. There is no “best “ way to design and implement business rules.
Actually, rules come in various categories: business rules, validation rules, usage rules, collaboration
rules. Validation Rules usually apply to a GUI; some fields are mandatory, some have to be within a
given range, some fields are related in an interdependent way: enter values in some, others are enabled
or disabled (state-based). Business Rules check valid combinations of inputs; check against invalid
combinations . Rules at each layer are treated in a similar manner; though provisions for the
uniqueness of each layer is made through customizing a rule. For example, instead of a composite rule,
a simple rule (Validator) can be used to validate a textbox in a GUI. Alternatively, a Composite Rule
with Composite Assessors and Composite Actions can be used to implement requirements for relating
customer care and telecommunications provisioning.
4. Rules become more easily testable. Following a rigorous requirements approach for business rules also
means that each business rule captured must have a means of being tested.
5. New subsystems will be needed to hold Rule Objects and to allow the rules to be changed by
privileged users.
6. Company Policy Repository. Rule Object prepares the infrastructure for having a central repository of
rules within a software development organization or within a corporation. Although the Rules (and
Company Policies) are scattered /distributed within the structure of the organization, they can be
centrally managed and browsed, defined and changed from a central location, allowing all interested
parties who have registered interest in the Rule or Rule Type to receive notification of its change. This
is done using Observer or Publish-Subscribe.
7. Rule Object Repository allows corporate executives to be able to define and manipulate rules as
policies from a GUI-based Rule Browser. This can then be propagated within an entire organizational
structure so that the programmers who will ultimately responsible for implementing the rules in
business objects will have a common basis or reference point of traceability.
Implementation
Consider the following implementation issues:
1. Avoid putting simple logic in Rule Objects unless they are apt to change on a frequent basis; Rule
Methods are a simpler way of handling rules if they are not needed to be pluggable, adaptable or
extensible.
2. Reuse of existing conditions and actions is a good sign of opting to use a Rule Object.
3. Setting up Rule Objects may take a bit of effort but once the framework is set up (see example code
below in the Sample Code section) then defining new rules, actions and conditions becomes simple.
4. Strategy can be used to check rules because there may be a family of rules that are related, and need to
be applied depending on the state of a given object. Command may also be used to check state, or to
execute an action after a condition is checked. Assessor reifies a set of conditions that need to be
checked. For example, an cluster’s state consists of a number of states of its objects. Each needs to be
checked for valid conditions, or they may alternatively (an entirely different kind of effort) be checked
for the presence of invalid conditions, so permutations of valid and invalid combinations need to be
checked.
5. Results of rule application. The application of rules, leaves a trace: results of the application of rules;
their conditions and actions may need to be logged and recorded. Efficiency considerations disallow
the logging of every single condition and action set.
16
6. Rule Complexity: Rules can be simple or complex (compound). They can be seen as Composites; or
as simple Validators. They may be implemented as a Composite, a Strategy or a Command.
One of the less thought-about factors in business rule design is the fact that error handling is closely
coupled to it. It is as important, perhaps, to know why a rule failed as it is to know that it succeeded.
Therefore error handling will be considered as part of the implementation consideration section.
Guidelines on Implementation
Each Business Rule is encapsulated within its own class. This can be Simple Rule Object or a Compound
Rule Object. A Composite Rule object consists of a Composite Structure containing Assessors and Actions.
The RuleObject contains a Composite Assessor, A Composite Action and an ErrorResult. The ErrorResult
is there in order to specify what the error was that disallowed the rule to fire the action after evaluating the
Assessor. The Assesor is similar to a Command. It has a method “assess()” which returns boolean. If it is a
Composite, the Assessor must successfully assess all of its constituent elements before returning true. If
there is a problem, the Assessor logs this as an error message/condition in the ErrorResult.
If all is well and the Assessor has evaluated the conditions to be true, then the Action is invoked. The
Action is a Command, possibly Composite that will either change the state of the current Client object
which is using, containing the Rule Object, or more likely, will collaborate with other objects to create a
valid state for the system, based on the evaluation by the Assessor. If at any time there are error conditions
arising, these will be logged within the ErrorResult object to which Action has a reference.
A Rule Object usually forms a cluster. This Cluster consists of {RuleObject, Properties, Assessor, Action,
Result}. The attributes or state that is passed into the RuleObject is assessed or evaluated by the Assessor.
If this evaluation is successful, RuleObject (acting as a Mediator between Property, Assessor, Action and
Result) sets the results of the Assessment in the result Object then asks Action to execute on the Properties,
potentially changing their State.
The results of the execution are set in the Result object. This may provide meta-data as to the fact that, for
example, a
a single object’s attribute values. Therefore, not only are we interested in evaluating a set of attributes and
performing actions based on the result of the evaluation, we frequently would like to know of the
intermediate results of the evaluation and the execution of each action as intermediate steps. This
information is recorded in the Result object. The State or set of attributes and values which are used as the
basis for evaluation by the Assessor are passed into the Rule Object or the Rule Object aggregates and
creates its own Properties object(s) and passes them to its Assessors and Actions on a demand-driven basis.
Sample Code
Here is an example of a Simple Rule Object that can be applied at the GUI layer. Although this
demonstrates Simple Rule Object for a GUI layer, Rule Object is applicable in all layers: application,
protocol and persistence layers.
class BusinessRule implements ActionListener{
private JDialog theDialog;
private Frame theFrame;
public void setTheDialogJDialog aDialog){
theDialog = aDialog;
}
public BusinessRule(){
}
public boolean assess(int number){
if (MIN_SERVICES <=number && number<= MAX_SERVICES)
return true;
return false;
}
public void actionPerformed(ActionEvent event){
tion 2
Enumeration e = tempVector.elements();
// for each assessor in the enumeration, call the evaluate
// method and get the return value
while(e.hasMoreElements())
{
tempAssessor = (Assessor)(e.nextElement());
assessorReturns = assessorReturns &&
tempAssessor.evaluateAssessor(getRuleProperties());
// determine if there are any actions that need to be executed for this assessor
Vector tempActionsVector = getActionsForAssessor(tempAssessor);
if(tempActionsVector != null)
{17
if (event.getActionCommand() == "SubmitButton"){
theDialog.dispose();
}
}
}
This is the simplest case where the Rule Object is merely one Class and acts as the Listener. Each time a
key is typed in a field in a Dailog, assess() is called to assess whether a valid value has been entered or not.
In this case it is a range that is being checked.
Below, is an instance of use.
/**
* Register the field for entering the number of people
* @param field The field used to enter the number people
*/
public void registerNumberField(final JTextComponent field) {
numberOfServices = field;
DocumentAdapter documentAdapter = new DocumentAdapter() {
protected void parseDocument() {
int count = 0;
try {
count = Integer.parseInt(field.getText());
} catch (NumberFormatException e) {
}
if (rangeRule.assess(count) )
serviceCount = count;
else
serviceCount = 0;
} // parseDocument()
};
field.getDocument().addDocumentListener(documentAdapter);
} // registerserviceCountField(JTextComponent)
Register the Rule Object as the button’s listener:
public void registerOKButton(final JButton btn){
submitButton = btn;
submitButton.addActionListener(rangeRule);
submitButton.setActionCommand("SubmitButton");
rangeRule.setTheDialog(myParentDialog);
}
Sample Code : The Rule Object Framework
Define a Rule Object that contains its Assessors and Action; Properties and an ActionAssessor Map to help
determine which Actions relate to which Assessors.
public abstract class RuleObject
{
private Vector assessorVector;
private Vector actionVector;
private RuleProperties ruleProperties;
private ActionAssessorMap actionAssessorMap;
public RuleObject()
{
actionVector = new Vector();
assessorVector = new Vector();
actionAssessorMap = new ActionAssessorMap();
ruleProperties = new RuleProperties();
}
public boolean applyRule(){
boolean assessorReturns = true;
Assessor tempAssessor;
Action tempAction;
// get an enumeration of the assessor vector
Vector tempVector = getAssessors();
Enumeration tempEnum = tempActionsVector.elements();
while(tempEnum.hasMoreElements())
{
tempAction = (Action)(tempEnum.nextElement());
tempAction.performAction(getRuleProperties(),
new boolean(assessorReturns));
}
}
}
return assessorReturns;
}
/**
}
Known Uses
The Rule Object has been used on several projects by various teams that the author has been involved with.
Domains include Telecommunications (Customer Care and Billing), Healthcare, Insurance, Automotive,
Higher Education industries, Sales, e-brokerage.
Rule Object has been used in the implementation of the Java Business Frameworks [Arsanjani99b]
IBM San Francisco uses Policy Common Business Object and Pattern which use a similar concept.
Rule Object motivated and was used in the “If-Then-Else” Framework, by Paul Corazza [Corazza].
IBM WebSphere Application Server Enterprise Edition, Component Broker’s Managed Object Framework
implements Rule Object.
David Taylor mentions a similar scheme in his Object Magazine column on Business Rules.
Related Patterns
Peer Patterns3
Rule Object acts as a Mediator to Assessors and Actions; Properties (Context) and Results. Rule Objects
determine which Assessors should be used in the assessment of Properties, with a possible recording of
Results in an Assessment result. If the assessment was successful, the appropriate Actions are invoked,
possibly updating or changing the State of
Other Patterns
Rule Object uses several more fundamental design patterns. It can therefore be considered to be a
Compound Pattern (or a Composite Pattern [Riehle98]). But not all Compound Patterns are Patterns
themselves; Compound Patterns are merely a namespace that labels a set of patterns that are repeatedly
found to work in concert in many different occasions. To implement the Condition participant, we suggest
the use of an Assessor [Arsanjani98]. To implement the actions (Simple or Compound) use a Command.
The Compound Rule is itself a Composite. SimpleRule uses a Strategy to implement its Validator
participant. The Rule Cluster has a Builder which uses an Abstract Factory to produce individual
instances of Rules, Conditions and Actions.
The Assessor is really a special case of a Command that is found to recur in multiple contexts. Instead of
executing a command, the Assessor has an assess() method which returns either a boolean (in the case of a
simple assessor) or a composite (in the case of a Compound Assessor). The Assessor may further be
implemented as an Interpreter if it needs to determine the validity of a “rule string”; i.e., a string containing
sentences of a “rule language”.
Rule Object is connected with Visitor in that it shares the following applicability to a large degree: “many
distinct and unrelated operations need to be performed on objects in an object structure, and you want to
void "polluting" their classes with these operations. Visitor lets you keep related operations together by
defining them in one class. When the object structure is shared by many applications, use Visitor to put
operations in just those applications that need them. ”
Alternatively, Assessor may be used in the context of Grammar-Oriented Programming [Arsanjani89]
where domain analysis determines a domain language. The domain language is then described in terms of a
domain grammar. The interaction between the domain objects is fully described by the domain grammar.
Use-cases that trigger collaborations, trigger the domain grammar and the message is passed as an input
stream into the parser that is interpreting or parsing the grammar. Object’s “manners” are described in
terms of the meta-model that is represented as a grammar.
References
[Arsanjani99;a] Ali Arsanjani. "Service Provider: A Domain Pattern and Its Business Framework
Implementation," presented to PloP '99.
http://st-www.cs.uiuc.edu/~plop/plop99/proceedings/Arsanjani/provider3.pdf
[Arsanjani99;b] Ali Arsanjani. "Analysis, Design, and Implementation of Distributed Java Business
Frameworks Using Domain Patterns" in Proceedings of Technology of Object-oriented Languages and
Systems 30, IEEE Computer Society Press 1999, pp. 490-500.
[Arsanjani89] Concepts of Grammar-Oriented Programming,
[Corazza] Paul Corazza. Using the if-then-else framework, Part 1: Code maintainable branching logic with
the if-then-else framework. Available at: http://www.javaworld.com/javaworld/jw-03-2000/jw-0324-
ifthenelse.html
[Fow96] Martin Fowler. Analysis
Patterns.
[GHJV95] E. Gamma, R. Helm, R. Johnson and J. Vlissides. Design Patterns: Elements of Reusable
Object-oriented
Software.
[MRB98] Robert Martin, Dirk Riehle, and Frank Buschmann (eds.). Pattern Languages of Program
Design
3.
[Odell96] James Odell and James Martin, Object-oriented Methods: Pragmatic Considerations. Prentice-
Hall, 1996.
[Riehle98] Dirk Bäumer, Dirk Riehle, Wolf Siberski, and Martina Wulf. "Role Object." In Proceedings of
the 1997 Conference on Pattern Languages of Programs (PLoP '97). Technical Report WUCS-97-34.
[VCK96] John M. Vlissides, James O. Coplien, and Norman L. Kerth (eds.). Pattern Languages of
Program Design
2.
Appendix A: Rule Object: A Pattern Language for Adaptive and
Scalable Rule Design and Construction (Management)
The following table summarizes the patterns in this pattern language and provides an initial definition and
context for each one. The next section in this appendix outlines a map of the pattern language showing how
the patterns relate to one another; supplying transition criteria (for going from one pattern to another) and
the forces that will be encountered before the transition and once the transition is made.
ROPL Patterns
1. Rule Object –
Provide extensibility and adaptability to business processes, without endangering them with intrusive
changes, by making the rules governing them pluggable.
2. Assessor – Assess a set of conditions based on an input set of Properties; record results of evaluations
3. Action – Perform actions in continuation of the results of the Assessment of conditions, record results
and update Properties and State of pertinent objects in collaboration.
4. Rule Cluster – Components the Composite definition and application of Rule Objects; optimize rule
application through the definition and selection of a rule application policy.
5. Rules have State –
6. Rules are Tracked – track history, changes, condition/action set pairs
7. Document Rules as Patterns—capture rules as patterns to track and report reasons for solution of issues
and consequences
8. Rule Object Repository: Centralize Rules in Corporate Repository
9. Rule Access Rights – managers should be able to create rules; give access rights to control unwanted
or accidental corruption of rules
10. Rules Change Process – New rules impact old processes
11. Clusters Have Manners – clusters of collaborating objects have laws governing their behavior and
meta-data about these rules (laws).
12. Rules as First-class Constructs -- conducting analysis and design based on object “manners”
13. Rules as Production Rules of the Application Domain Grammar – Grammar-oriented object design;
define a domain language and grammar for a domain; implement it using a parser accepting input from
an application running in that domain
14. Persistent Rules—Handle proliferation of subclasses and objects as “data”
15. Hash and Cache—provide efficient and quick access to subclasses and objects as the numbers
increase
16. Remedy Rule Proliferation—Handle object proliferation syndrome
17. Rules Evolve – Rule evolution for business survival
18. Rule Change Impacts Architecture – information system architecture, functional and non-functional
requirements are impacted by changes in rules.
Pattern Language Map
The Rule Object Pattern Language can be briefly described in the following narrative. We will then take a
look at the relationships between the individual patterns and how one can potentially follow another or is
somehow related to another pattern.
Note that some patterns are tiny pattern languages in disguise: Rule Object consists of Validator, Simple
Rule Object, Assessor, Action, ErrorResult, Properties, Compound Rule Object and MediationStrategy
(showing what algorithm to apply in sequence when applying rules, conditions or actions. These can be
simple round-robin style or have a more complicated algorithm such as the Rete Algorithm to apply rules.
The default is the simple round robin; just go through the vector of rules and apply each one sequentially.
You may want to have a weighted vector or a Hashtable which gives a priority to the application of each
rule of assessor or action, should they be a collection or should they be Composite).
You can start at various parts of the ROPL language and work your way around by applying patterns to
resolve forces in the problem domain. So for example, there are many use-cases for starting the journey and
resolving forces that arise in the problem space trhough the application of the pattern language’s patterns.
Use-case 1: You want to build an insurance application and need to implement existing business rules that
are given as part of the requirements specification.
Use-case 2: The organization wants to organize its business rules. You are on a business rules hunt and
Document Rules as Patterns, Create a Corporate Rule Repository and store the rules.
Use-case 3: You have a Corporate Rules Repository and want to configure it for general use. You assign
Rule Access Rights to managers and developers who have access to the Rules in the Repository. You Log
Changes to Rules as management changes the business to meet market and operational demands. Having
this repository affects how business is conducted so Rules Changes Business Processes. As changes are
made to rules, the architecture is impacted; maybe rules are concentrated in a middle tier rather than being
scattered and duplicated in various middle tiers, database triggers, GUIs, etc.
Use-case 4: During development with Rules Objects, Rules Evolve and changes to rules are handled by
creating new Assessors and Actions, or reusing existing ones in new combinations to reflect new rules and
processes. Properties are created as raw inputs for Assessment and Actions may update the Properties,
yielding a Results or ErrorLog which is reported to the user and/or logged in a persistent store.
Use-case 5: As Rules start proliferating, we Handle Rule Proliferation by encapsulating related rules in
Rule Clusters and compose Compound Rules from Simple Rules. We start reusing Assessors and Actions.
We may Hash and Cache them in memory or in a middle tier for optimal performance. As they proliferate,
we may decide to persist our Assessors, Actions and Rules , Properties and ErrorResults and thus Create
Persistent Rule Objects.
Use-case 6: During the course of
using Rule Objects we may find that we need to
maintaining it as a Momento or some other mechanism, maybe even a simple static AssessorResult would
do the trick.
Use-case 7: As we conduct more requirements analysis and create more Rule Objects, we find that there are
Clusters of collaborating classes tat work with each other to achieve a business goal. These Business
Objects form Clusters. These Clusters tend to have rules governing their interactions within the cluster, we
call this their “manners”. Thus, Clusters have manners and Business Objects have Manners.
23
Keeping these use-cases in mind, the following pattern language map will aid you in choosing paths
through the language:
Number Transition Criteria/ Forces
1. Complexity, adaptability, composability, organization
2. Frequent changes to rules rapid turnaround needed; reuse of designs, rules, conditions and
actions; maintenance issues as complexity increases; scalability
3. Uses; interchangeable conditions needed (again zip code is checked against the service type
and features)
4. Uses ; interchangeable and reusable, recurring actions (e.g., error messages, updates to db,
etc.)
5. Subclasses and objects proliferate
6. Greater proliferation (object proliferation syndrome); need to have faster access to objects;
need to manage and maintain object assets
7. Even greater proliferation; faster access; treat objects as “data”; databases handle this well
8. Need to track changes and report
9. Many people have access, access needs to be controlled; legal and security purposes;
stability and control
10. Common portal for corporate rules; organized and browseable; development teams need
frame of reference from which to be updated on new needs for rules changes; dynamic,
cross-domain reuse
11. Multi-domain reuse; product line architectures; create a domain grammar; employ grammaroriented
object design [Arsanjani90]
12. Architecture and business process driven by rules
13. Rules and business change; need to stay in business, maintain market share and profit;
promotions, new services, new offerings; new legislation; deregulation, etc.
14. Change of rules impact architecture in terms of functionality and non-functional
requirements (availability, security, performance, persistence, scalability, etc.)
15. Corporate awareness of rationale behind rules, issues the rules are trying to solve need to be
documented, what are the reasons and issues; how does this rule solve them; what is the
resulting consequence (sound familiar?!)
16. <same as above>
17. Implement rules starting with the most appropriate design or implementation mechanism
Appendix B: Rule Pattern Language Map – Business Process Emphasis
The Rule Object Pattern Language can be briefly described in the following narrative. We will then take a
look at the relationships between the individual patterns and how one can potentially follow another or is
somehow related to another pattern.
Note that some patterns are tiny pattern languages in disguise: Rule Object consists of Validator, Simple
Rule Object, Assessor, Action, ErrorResult, Properties, Compound Rule Object and MediationStrategy
(showing what algorithm to apply in sequence when applying rules, conditions or actions. These can be
simple round-robin style or have a more complicated algorithm such as the Rete Algorithm to apply rules.
The default is the simple round robin; just go through the vector of rules and apply each one sequentially.
You may want to have a weighted vector or a Hashtable which gives a priority to the application of each
rule of assessor or action, should they be a collection or should they be Composite).
You can start at various parts of the ROPL language and work your way around by applying patterns to
resolve forces in the problem domain. So for example, there are many use-cases for starting the journey and
resolving forces that arise in the problem space trhough the application of the pattern language’s patterns.
Use-case 1: You want to build an insurance application and need to implement existing business rules that
are given as part of the requirements specification.
Use-case 2: The organization wants to organize its business rules. You are on a business rules hunt and
Document Rules as Patterns, Create a Corporate Rule Repository and store the rules.
Use-case 3: You have a Corporate Rules Repository and want to configure it for general use. You assign
Rule Access Rights to managers and developers who have access to the Rules in the Repository. You Log
Changes to Rules as management changes the business to meet market and operational demands. Having
this repository affects how business is conducted so Rules Changes Business Processes. As changes are
made to rules, the architecture is impacted; maybe rules are concentrated in a middle tier rather than being
scattered and duplicated in various middle tiers, database triggers, GUIs, etc.
Use-case 4: During development with Rules Objects, Rules Evolve and changes to rules are handled by
creating new Assessors and Actions, or reusing existing ones in new combinations to reflect new rules and
processes. Properties are created as raw inputs for Assessment and Actions may update the Properties,
yielding a Results or ErrorLog which is reported to the user and/or logged in a persistent store.
Use-case 5: As Rules start proliferating, we Handle Rule Proliferation by encapsulating related rules in
Rule Clusters and compose Compound Rules from Simple Rules. We start reusing Assessors and Actions.
We may Hash and Cache them in memory or in a middle tier for optimal performance. As they proliferate,
we may decide to persist our Assessors, Actions and Rules , Properties and ErrorResults and thus Create
Persistent Rule Objects.
Use-case 6: During the course of using Rule Objects we may
find that we need to
maintaining it as a Momento or some other mechanism, maybe even a simple static AssessorResult would
do the trick.
Use-case 7: As we conduct more requirements analysis and create more Rule Objects, we find that there are
Clusters of collaborating classes tat work with each other to achieve a business goal. These Business
Objects form Clusters. These Clusters tend to have rules governing their interactions within the cluster, we
call this their “manners”. Thus, Clusters have manners and Business Objects have Manners.
© 1999-2001, Ali Arsanjani, IBM – Permission granted to make copies for the PLoP 2001 Conference 29
Figure 17: Rule Object Pattern Language Map
The Following Table discusses the above figure 17 and specifies the transition criteria for each edge of the
graph.
Number Transition Criteria/ Forces
1. Complexity, adaptability, composability, organization
2. Frequent changes to rules rapid turnaround needed; reuse of designs, rules, conditions and
actions; maintenance issues as complexity increases; scalability
3. Uses; interchangeable conditions needed (again zip code is checked against the service type
and features)
4. Uses ; interchangeable and reusable, recurring actions (e.g., error messages, updates to db,
etc.)
5. Subclasses and objects proliferate
6. Greater proliferation (object proliferation syndrome); need to have faster access to objects;
need to manage and maintain object assets
7. Even greater proliferation; faster access; treat objects as “data”; databases handle this well
8. Need to track changes and report
9. Many people have access, access needs to be controlled; legal and security purposes;
stability and control
10. Common portal for corporate rules; organized and browseable; development teams need
frame of reference from which to be updated on new needs for rules changes; dynamic,
cross-domain reuse
11. Multi-domain reuse; product line architectures; create a domain grammar; employ grammaroriented
object design [Arsanjani90]
12. Architecture and business process driven by rules
13. Rules and business change; need to stay in business, maintain market share and profit;
promotions, new services, new offerings; new legislation; deregulation, etc.
Appendix C: The Peer Patterns
A Brief Look at Rule Object’s Peer Patterns
Peer Patterns5
Rule Object acts as a Mediator to Assessors and Actions; Properties (Context) and Results. Rule Objects
determine which Assessors should be used in the assessment of Properties, with a possible recording of
Results in an Assessment result. If the assessment was successful, the appropriate Actions are invoked,
possibly updating or changing the State of
Assessor
Assessors encapsulate and evaluate conditions based on a set of input Properties and record the results of
the evaluation within their AssessmentResults. Assessor is similar to Command. A Command executes an
operation whereas an Assessor assesses a set of conditions based on input State or Properties. The
difference is that it has an evaluate() method rather than an execute(); and collaborates with a Property List
and a ErrorLog or ResultSet:
Figure 13: The Assessor (Condition Evaluator)
Action
An Action executes by updating State or invoking behavior in other collaborating objects. It records the
result of its actions in an ActionResult which may contain an ErrorResult or ErrorLog. Actions can be
implemented as Strategies, Visitors, Interpreters, Commands. Here is the structure of its cluster:
The Action Pattern
Appendix D: Types of Rules
Rules come in various types. Here is a list of scenarios that provides motivation to define a taxonomy of
rule types based on complexity, scale, layer and function. Specifically, 1-7 address tier-specific types of
rules and 8 contains a list of non-tier specific types of rules.
1. Field on a form needs to be validated for non-blank, correct domain of values, etc. Objects check these;
UI sends the app object a message to check this data value.
2. Set of interdependent fields must be validated collectively.
5 Peer Patterns are patterns in the same pattern language that work together to resolve and balance forces.
3. Compound data spread across multiple forms needs to be cross-validated; data in one form should not
violate another data value in another, related form. Implemented using a mediator pattern whom
checks to see if all data are consistent without needing the forms to know about one another.
3.1. Invalid combinations need to be identified and reported or disallowed altogether
3.2. Valid combinations tested for validity
4. Checking with policies and verifying business rules in the middle tier constitutes the bulk of
application (or “business” ) logic; which may be accessed by many types of clients: such as PDAs,
thin-clients, fat clients, etc.
5. Checking rules at the database layer in the form of triggers and stored procedures.
6. Middleware communication rules.
7. A variety of security-related rules; authorization, authentication, non-repudiation, etc.
8. Regardless of tier or layer s in which they reside, rules can pertain to computations (functions,
mappings, transformations usually expressed or expressible via mathematical formulae), for example:
When a Coverage is selected (added to a policy) it may have causes of loss associated with it. These are fixed, and
cannot be changed. If a coverage has causes of loss, when the coverage premium is calculated, it asks each cause
of loss to calculate its contribution to the premium and sums them. If there are no causes of loss for a coverage,
then the coverage has a premium of $0.00.
8.1. Some restrict and constrain business processes and data that are input to drive them;
8.2. Some pertain to the workflow order and “Route”, “Rules”, “Roles”;
8.3. Some constitute “terms of agreement” or “conditions”, such as “service level agreements” which
are more legal in nature
8.4. The enforcement of certain types of rules results in “positive inclusion”: a Coverage is valid if
and only if …. Sometimes, however, we have “negative inclusion”: for example, “If the zip code
of the applicant is within <list of high-fraudulent zip codes> then we will have to do a pre-paid
calling card; other times we have “invalid combinations” the conjunction of a set of conditions
results in an invalid state: “if the loan requester has a second house and the primary house is
within a high-risk earthquake zone and they have no insurance, then we will have to reject the
loan if credit is below <xxx> amount”.
9. The design of rules and their implementation are two separate things: e.g., rules can be designed as
business constraints and implemented in the data layer via referential integrity rules that can be
modeled as specific types of associations/relationships.