SOFTWARE—PRACTICE AND EXPERIENCE, VOL.

25(10), 1155–1181 (OCTOBER 1995)

Software Maintenance: An Approach to Impact Analysis

of Objects Change

SAMUEL AJILA∗
CRIN-CRNS, Bâtiment LORIA, Campus scientifique, BP 239, 54506 Vandouvre-lès-nancy, France
(e-mail: [email protected])

SUMMARY
The maintenance of a software system requires a tool for impact analysis and the propagation of change.
This paper presents a knowledge-based model for both. This model is generic because it is not based on
any language or design method. Therefore, it can be applied on the basis of a ‘problem’ to be solved. It
also captures four life-cycle phases: requirement, specification, design and programming. We also provide a
domain-specific view that allows the dependency analysis of fine-grain objects. Two kinds of dependencies
are identified: inter-phase dependencies, these are dependency relations between the objects of one phase
and another; and intra-phase dependencies, these are dependency relations between the objects of the same
phase. In order to validate this model, we also present a prototype based on two life-cycle phases: design
and programming.

KEY WORDS: software maintenance; impact analysis; objects change; dependency relations; software life cycle; knowledge
based model

1. INTRODUCTION
Software maintenance is generally considered11,14,17,18,24,27,30 to be expensive, difficult and
time consuming. One of the reasons for this is that it is difficult for software maintenance
engineers to fully understand a software system that they did not design or implement. Even
when they develop their own system, it is easy to forget the details of the design or code;
even more so if there is little or no documentation. Another reason is that any change in
part of the software must take into consideration the inter-relationships between the change
and the rest of the system. There are complex inter-dependent relationships between the
different components of the software.12,19
There are various reasons for changes in a software system:3,26,28,29

1. the performance of the software must be maintained (perfective maintenance);

2. the errors in specification, design and implementation must be corrected (corrective
maintenance);
3. the final product must operate in diverse environments (adaptive maintenance);
4. the final product must also evolve (evolution maintenance).

The problem of objects change in a software can be viewed as:

∗ Current address: Department of Computer Science, University of Ibadan, Nigeria.

CCC 0038–0644/95/101155–27 Received 19 October 1994

Ó 1995 by John Wiley & Sons, Ltd. Revised 6 April 1995
1156 S. AJILA

1. identifying the victims of change;

2. identifying the types of changes that may operate on each object of the software;
3. evaluating the effects of a change;
4. acquiring knowledge on how to perform a change.
Impact analysis can be defined as the activity of identifying which object to modify to
accomplish a change, that is, estimating the potential consequences of carrying out a change.
Impact analysis of a change is necessary for the following reasons:

1. To estimate the cost of a change. If a proposed change has the possibility of affecting
large disjoint sections of the software, then the change might need to be re-examined
and if possible abandoned in order to design another safer change.
2. To understand the meaning and the relationship between the item of change and the
structure of the software. The impact analysis tool can use the notion of conductivity
because there are some dependency relations that are better conductors of changes.
These types of relation allow the identification of the parts of the software that must
be modified and that parts may be modified for a given change-type.
3. To record the history of change related information and to evaluate the quality of
a particular change. Examining the feedback information from users of the changed
product helps in maintaining the quality of the software.
4. To determine the parts of the software that may need to be regression tested after a
change is carried out and to point out the vulnerability of some critical sections of the
software in order to determine whether the functionality of those parts of the software
is susceptible to changes made to the software.

We propose a generic knowledge-based model for the impact analysis of changes in

software. The model software view is based on four life-cycle phases and a domain-specific
view.9
As a part of this research, we have developed a prototype impact analysis environment
that provides dependency analysis to indicate the possible effects of software modifications.
The rest of this paper is structured as follows: the next section presents the impact analysis
tool model. Section 3 gives the details of our prototype and we give a conclusion and report
future research in Section 4.

2. THE MODEL WHAT-IF

The model WHAT-IF is a generic knowledge-based model. This model is generic because
it is not based on any type of programming or specification language nor on any method of
design. Therefore, it can be applied on the basis of the problem to be solved. The objective
of this model is to evaluate a priori the impact of objects change in a software system.

2.1. Conceptual framework

Let S be a software system represented by a set of objects OS = {o1 , o2 , . . . , on }. Let
Ts = {t1 , t2 , . . . , tn } be the set of change types that can be carried out on the objects of S
such that for a given change {ti , oj } we can define

fimpact {ti , oj } −→ {o1 , . . . , oi , ok , . . .}

in such a way that some system properties in S remain invariant if the change is finally
 SOFTWARE MAINTENANCE 1157

carried out and the affected objects {o1 , . . . , oi , ok , . . .} are also modified accordingly and
where:
—the objects in OS are linked by explicit dependencies (a set of ordered pairs);
—Ts depend on the software system view;
—fimpact is the impact analysis function; and
—{o1 , . . . , oi , ok , . . .} are the ‘direct victims’ of the change, i.e. those objects that have
direct relationships with oj . The ability to estimate the impact of a change (in terms of
objects that may be affected by a change) and perform change propagation depends on the
following factors:
1. software system views;
2. change types;
3. set of objects to be manipulated;
4. dependency relationships between objects of S ; and
5. fimpact and its properties.
Therefore, there is the need for us to explicitly define these factors which are in the
sub-sections that follow.

2.2. Software system views

A software system view Sview is a set of abstractions that gives possible characteristics of
the software system. When making a change, it is necessary to define the ‘objects change
life cycle’ and this definition must be based on the characteristics of the software system.
Software system views have been categorized9 as: programming language and structural
view, architectural view, software life-cycle view (in this case one can have a waterfall
life-cycle view, spiral life-cycle view, etc) and domain view.
Each of these views has advantages and disadvantages. For example, the programming
view has the advantage that it is easy to implement and in most cases it is based on
the structure of the programming language. Its disadvantage is that it focuses only on the
programming language dependency relationships and requires that the maintainer creates a
sort of ‘mental knowledge’ of the relationships between the design or the specification and
the programming level.
In this work, we are interested in the impact of change across software life-cycle phases.
Therefore, our system of views is based on what we choose to call the ‘mixed-view ap-
proach’: waterfall and domain views.
In this approach, we have four life-cycle phases. Each phase represents a view. Therefore,
we have ‘requirement view’, ‘specification view’, ‘design view’, and ‘programming view’.13
For each view there is a ‘domain view’19 which makes explicit the fine-grained dependencies
between system parts. This is illustrated in Figure 1. The advantage of this approach is that
we can explain the impact of a change in terms of a domain-specific view and a life-
cycle view. It also allows us to specify the impact of the change in terms of the following
combination of views: requirement-specification view (r1); specification-design view (r2);
design-programming view (r3); requirement-programming view (r4); requirement-design
view (r5); and specification-programming view (r6).

2.3. Change types and set of objects in WHAT-IF

For each view, we need to define change types that will operate on the objects of the
1158 S. AJILA

Figure 1. WHAT-IF model system of views

view. In some cases, such changes may be trivial such as type change (variable, function,
etc) in the programming view, merging and deletion in the design-programming view or
modification and elimination in the requirement-specification view. It may be complex in
some cases and this may require the intervention of specialists and users of the system
before the change type can be specified.
The most basic object-oriented concept is the ‘object’ itself. An object may be described
as an entity which can be observed. This observation creates and distinguishes the object
from its environment. Our notion of object in this work is based on the notion of object in
a conventional programming language. An object in a conventional programming language
can be defined as a collection of routines, types, and/or data items.28 Routines implement
the methods associated with the object; types structure the data it processes; and data items
represent the instance of the object class.
The objects manipulated by WHAT-IF are described by the following hierarchy:

1. requirement objects, e.g. chapters, sections, sub-sections, etc, in the requirement doc-
ument;
 SOFTWARE MAINTENANCE 1159

2. specification objects, e.g. specification functions and operations;

3. design objects, e.g. objects that encode design decisions such as active and passive
objects and operations in HOOD (Hierarchical Object Oriented Design);
4. implementation objects, e.g. modules, procedures, functions, blocks, packages, tasks
and variables.

2.4. Dependency relationships between objects

Software objects are related to each other by complex dependencies and constraints.1,19
Therefore, an understanding of system dependencies is fundamental for efficient software
change.
In this work, we identify the following kinds of dependencies:

1. inter-phase dependency relationships: these are relationships between objects of dif-

ferent views, and
2. intra-phase dependency relationships: these dependency relationships depend on a
particular system of view. For example: in the programming view we may have the
following: Si -calls-Sj , if both Si and Sj are sub-programs; Si -creates-Sj , e.g. tasks
in Ada; Si -is-declared-in-Sj , if Si is declared inside Sj ; Si --is-a-parameter-of-Sj ,
if Si is a parameter and Sj is a function; etc.

2.5. fimpact and its properties

Defining fimpact {ti ,oj } −→ {o1 , . . . , oi , ok , . . .} is not enough for impact analysis and
change propagation. What we actually need is to define:
fimpact {ti , oj } −→ {o1 (o11 , o12 , . . . , o1n ), . . . , oi (oi1 , oi2 , . . . , oin ), ok (ok1 , ok2 , . . . , okn ), . . .}

where
{(o11 (o12 , . . . , o1n ), . . . , (ok1 , ok2 , . . . , okn ), . . .}
are the objects that are indirectly concerned by the change.
In this case we can compute both the direct and indirect ‘victims’ of change. In order to
do this we have a choice between two methods of representation:32 the ‘shallow method’
and the ‘deep method’. Generally speaking, the shallower a representation (i.e. the less in-
formation it represents explicitly), the easier it is to implement. However, the less explicitly
information is represented, the harder it is for a tool to manipulate it. For example, it is
impossible for a tool that represents requirements merely as textual documents to detect
logical inconsistencies. Therefore, if a tool is to provide comprehensive support for a com-
plex software task that requires the deep understanding of a software artifact, then it must
represent the artifact in such a way that all of the relevant information is available to it. An
important benefit of deeper representations is that they allow a deeper level of integration
between tools.32
Therefore to represent fimpact , do impact analysis and change propagation, we adopt the
following:

1. A graph is used to represent the structure of each view. The nodes of the graph
correspond to the objects of the view while the arcs correspond to the dependencies
between the objects. Therefore, for the entire system, we have a ‘multi-graph’10,15,40
1160 S. AJILA

Figure 2. WHAT-IF global architecture

consisting of a graph of the requirement view, a graph of the specification view, a

graph of the design view, and a graph of programming view. Graph representation is
chosen here because it can be used as a powerful software engineering technique.36
2. The multi-graph is then represented by the set of its simple graph components and by
their dependencies.
3. Composite dependency relations, constraints and all other information such as pre-
and post-conditions for the application of a particular type of change are organized as
rules into the knowledge base of the model (Figure2).
4. The graph components and their dependencies are also organized as a set of facts
(Figure 2).
5. Information specific to each software component is stored in a database (Figure 2).
In what follows, we present a prototype of this model. The prototype is based on the
design-programming view (r3 in Figure 1). We use the HOOD design method to capture
the design view and the Ada programming language for the programming view.

3. THE PROTOTYPE WHAT-IF

The prototype WHAT-IF is based on two sub-systems of WHAT-IF mixed views. These
are the design view and the programming view. We have also chosen the design method
HOOD for the design view and the programming language and Ada for the coding view.
This choice is based on the fact that Ada as a programming language has some constructs
that are rich in semantics and can be defined in two steps, first the declaration and then the
body. Theses constructs are sub-programs, tasks, generic units, private types, incomplete
types and deferred constants.35 HOOD as a method of design is based on the concept of
the object and the history of this method is linked with Ada. The translation from HOOD
to Ada is possible via the ODS (Object Description Skeleton).23,25 In addition to this there
are some software environments that allow the automatic translation of HOOD to the Ada
specification part.
 SOFTWARE MAINTENANCE 1161

Figure 3. The architecture of the prototype WHAT-IF

3.1. Architecture of the prototype

The prototype consists of three major sub-systems: the extraction sub-system, the knowledge-
base, and the abstractor (cf Figure 3).
The extraction sub-system accepts as inputs the HOOD ODS and the Ada code. It pro-
duces as output direct relationships which are later converted to PROLOG facts and rules.
It also produces tables showing the characteristics of each object in the system. The tables
are stored in the database and the facts and rules are stored respectively in the facts base
and the rules base. The abstractor receives the queries, executes the abstraction process on
the relations and produces replies to the user. In what follows, we are going to give the
details of each of the three major sub-systems.

3.2. The extraction sub-system

The extraction sub-system consists of three parts: the analysers, the processor and the
transformator (cf Figure 3).
1162 S. AJILA

Figure 4. The architecture of the analysers

3.2.1. The analysers

The two analysers are based on the same architecture (cf Figure 4). They are implemented
by using LEX and YACC, two standard UNIXTM tools. The Ada code analyser is based on
the Ada 83 syntax and that of HOOD ODS on the syntax enumerated in Reference 23.

3.2.2. The processor

The processor processes the decorated abstract trees and produces data structures represent-
ing the direct relationships. The tables containing the various objects and their characteristics
are stored in the database (Figures 5 and 6).

Figure 5. Ada identifier table

 SOFTWARE MAINTENANCE 1163

Figure 6. HOOD identifier table

Figure 7 shows a simplified version of the data structure produced by the processor for
the Ada software system. The Ada object table contains the names of objects in an Ada
software system and for each object it contains the following lists: files-list, sub-programs-
list, packages-list, tasks-list, variables-list, import-list, use-list, generic-list, renaming-list
and sub-program-call-list.
The Ada object table summarizes several direct relations. To specify these relationships,
we define the following entities:

1. variable(V): signifies that V is a variable of the type integer, real, or enumerated† .

2. composite-variable(C): signifies that C is a composite variable of the type array or
record.
3. type(T): signifies that T is a data type.
4. sub-program(S): signifies that S is a sub-program and that S must be an Ada function
or a procedure.
5. ada-object(X): signifies that X is an Ada abject and X can be of type package, task,
function, or procedure.
6. integer(N): signifies that N is an integer.

Given the previous entities and the Ada object table, we specify the following direct
relations for the Ada software system:
(R1) is-parameter-of :- {S, V, N} The relation is-parameter-of(obj, var, i), obj ∈ S ,
var ∈ V , i ∈ N exists if and only if the variable var is declared in the module obj
as a formal parameter at position i.
(R2) calls :- {X, S} The relation calls(obj1 , obj2 ), obj1 ∈ X, obj2 ∈ S exists if and only
if obj1 directly calls the sub-program obj2 .
(R3) type-return-value :- {S, T} The relation type-return-value(obj, t), obj ∈ S , t ∈ T
exists if and only if the function obj returns t as its value after its invocation.
(R4) obj-is-declared-in :- {X, X} The relation obj-is-declared-in(obj1 , obj2 ), obj1 obj2 ∈
X exists if and only if the module obj2 is declared inside obj1 .
(R5) var-is-declared-in :- {X, Vt }, Vt ∈(V, C) The relation var-is-declared-in(obj, var),
obj ∈ X , var ∈ Vt exists if and only if the variable var is declared inside module
obj .
† Ada enumeration type.
 1164 S. AJILA

Figure 7. Ada object table

(R6) record-contains-var :- {C, Vt }, Vt ∈ (V,C) The relation record-contains-var(var1 ,

var2 ), var1 ∈ C var2 ∈ Vt exists if and only if var2 is defined inside var1 and that
var1 is of the type record.
(R7) obj-rename :- {X, X} The relation obj-rename(obj1 , obj2 ), obj1 obj2 ∈ X exists if
and only if obj1 is renamed as obj2 by using the renaming mechanism in Ada.
(R8) var-rename :-{Vt ,VT },Vt ∈ (V, C) The var-rename(var1 , var2 ), var1 , var2 ∈ Vt ex-
ists if and only if var1 is renamed var2 by using the renaming mechanism in Ada.
(R9) import :- {X, X} The relation import(obj1 , obj2 ), obj1 obj2 ∈ X exists if and only if
the module obj1 imports obj2 by using the Ada clause use.
(R10) is-specified-by :- {X, X} The relation is-specified-by(obj1 , obj2 ), obj1 obj2 ∈ X exists
if and only if module obj1 is the declaration part of obj2 . Then obj2 is the body part.
(R11) instantiates :- {X, X} The relation instantiates(obj1 , obj2 ), obj1 obj2 ∈ X exists if
and only if obj2 is defined as a generic unit of obj1 by using the Ada clause new.
(R12) mapping :- {X, S, V, N} The relation mapping(obj1 , obj2 , var, i), obj1 ∈ X , obj2 ∈
S, var ∈ V , and i ∈ N , exists if and only if obj1 calls obj2 and var is a parameter of
obj1 at position i, i.e. calls(obj1 , obj2 ), and is-parameter-of(obj1 , var, i).
In the same way, Figure 8 shows a simplified version of the data structure produced by
the processor for the HOOD design system. The HOOD object table contains the names
of objects in a HOOD design system and for each object contains the following lists:
 SOFTWARE MAINTENANCE 1165

Figure 8. HOOD object table

operations-list, used-objects-list, is-implemented-by-list and internal-objects-list.

HOOD provides an object-based notation to describe systems. Therefore, the basic notion
present in HOOD designs is the object. HOOD objects provide an interface to the rest of
the system, encapsulate its data and internal structure and may have a particular dynamic
behaviour. A design step in this method consists of decomposing an object (called the parent
object) into sub-objects (child object).
The HOOD object table summarizes several direct relations. We define the following
entities for the purpose of specifying these relations:
1. HOOD-object(H): signifies that H is a HOOD object. H can be active or passive and
it can also be a parent or child object.
2. HOOD-operation(O): signifies that O is a HOOD operation.

Given the entities above and the HOOD object table, we specify the following direct
relations for HOOD design system:
(R13) defines :- {H, O} The relation defines(obj, op), obj ∈ H , op ∈ O exists if and only
if the operation op is defined inside the object obj. obj can be active or passive.
(R14) includes :- {H, H} The relation includes(obj1 , obj2 ), obj1 , obj2 ∈ H exists if and only
if obj1 is the father of obj2 . Therefore, obj2 is the child of obj1 .
(R15) implemented-by :- {O, O} The relation implemented-by(op1 , op2 ), op1 , op2 ∈ O
exists if and only if there are two objects obj1 , obj2 ∈ H such that defines(obj1 , op1 ),
defines(obj2 , op2 ) and includes(obj1 , obj2 ).
(R16) uses :- {H, H} The relation uses(obj1 , obj2 ), obj1 , obj2 ∈ H exists if and only if there
are two operations op1 , op2 ∈ O such that implemented-by(op1 , op2 ) and [(either obj1
and obj2 are active objects) or (obj1 is active and obj2 is passive) or (obj1 and obj2
are passive objects)].
We have also created a table of correspondences between the Ada objects and HOOD
objects (cf Figure 9). Given the entities defined for the Ada software system, the HOOD
design system and the table of correspondences, we specify the following relation:
(R17) corresponds-to :- {H, X} The relation corresponds-to(obj1 , obj2 ), obj1 ∈ H , obj2 ∈ X
exists if and only if obj2 is an Ada object (package, task, procedure, or function)
1166 S. AJILA

Figure 9. Table of correspondences

and obj1 is a HOOD object (class, object, operation, opcs, obcs) and that the corre-
spondence between obj1 and obj2 is based on the rule of correspondence defined in
Figure 7. The inverse of this relation (i.e. corresponds-to−1 (obj2 , obj1 )) is also true.

3.2.3. The transformator

The transformator transforms the direct relations into PROLOG facts and rules.
We give below the definitions of the usage of some terms in the context of PROLOG.
1. facts base: we call facts base the set of facts relative to the problem that a user
wants to formalize in PROLOG. Formally, in first-order logic, facts are positive Horn
clauses.
2. rules base: we call rules base the set of rules relative to the problem a user will want
to formalize in PROLOG. Rules in this context are strict Horn clauses.
3. query or question: A query (or question) in PROLOG is a negative Horn clause or
goal clause.
4. PROLOG inference engine: the PROLOG inference engine is the part of the system
which performs the logic inference on the facts and the questions asked by the user.
It is a backward chaining non monotone engine.5,21
The direct relationships are transformed into facts and rules as follows:
1. The entities defined for the Ada software system and HOOD design system are trans-
formed to PROLOG facts.
2. The relations R1–R14 and R17 are transformed to PROLOG facts.
3. The relations R12, R15 and R16 are transformed to PROLOG rules.
Table II shows part of a set of facts generated from the reduced Ada program example
shown in Table I.

3.3. The abstractor

The abstraction sub-system is not a finished product. It is designed in such a way as
to allow the users to formulate relations by adding rules to the system and by relating the
SOFTWARE MAINTENANCE 1167

Table I.
1168 S. AJILA
Table II.

existing ones in such a way as to prove some assertions or to create personal queries. At
present, it is implemented to give real-time answers to some pre-defined queries.
The abstractor consists of three parts: the user interface, query manager and the inference
engine (cf Figure 10).

Figure 10. The abstraction sub-system

The query manager is implemented in PROLOG. The inference engine is essentially the
PROLOG inference engine and it is completed by a function that allows the interrogation
of the engine from the interface and the activation of the engine based on the type of query
(cf see sections 3.3.1–3.3.3).
 SOFTWARE MAINTENANCE 1169

3.3.1. The query manager

The query manager as its name indicates manages the queries by defining a sort of agenda
for each query. In this system, a query is normally based on a meta-rule and a meta-rule is
a set of rules/ Formally, we define a query as follows:
< query >→< meta-rule >;
< meta-rule >→< [rules]+ >;
< rules >→< rule, [rule]∗ >;
< rule >→< PROLOG rule >.
In this prototype, we have implemented three type of query: WHAT-IS, WHAT-IF, and
WHAT-LIST.

3.3.2. The query WHAT-IS

Every object in a system must be identified. The visibility of each object in a software
system is very important for the purpose of impact analysis of objects that are changed in
the system. Therefore, this query allows such identification and gives an insight into the
visibility of the object.
The syntax of WHAT-IS is as follows:
WHAT-IS(object-name, level)
object-name refers to the object and
level refers to the object view (in this case: HOOD or Ada).
This query interacts directly with the database. We have already give an example of the
invocation of this query. This example is based on the Ada program example in Table II.
In this example we can see that the object sort a procedure-body in the Ada program is
defined in two places (lines 63–88 and lines 204–223) in the program.

3.3.3. The query WHAT-IF

The most important query in the system is the impact analysis query. The global syntax
of the query is as follows:
WHAT-IF(object-name, level, change-type) where
object-name and level are defined in the same way as in WHAT-IS, and
change-type depends on the level and it can be empty in some cases.
This query contains four versions:

1. WHAT-IF-Ada(object-name, change-type),
2. WHAT-IF-HOOD(object-name, change-type),
3. WHAT-IF-Local(object-name, level), and
4. WHAT-IF-Global(object-name, level).

The query WHAT-IF-Ada acts on the Ada software system objects while the query
WHAT-IF-HOOD acts on the HOOD design system objects. We have defined these two
versions of WHAT-IF query in order to take care of the differences between the types of
change in Ada and HOOD.
The query WHAT-IF-Local acts on all the objects at the two levels (Ada and HOOD).
This query takes care of all the objects that are not constrained by any type of change and
the objects that do not have any correspondence. The query WHAT-IF-Global also acts
on all the objects at the two levels. In fact this is the propagation query. It analyses the impact
1170 S. AJILA
Table III.

of a change at a given level and propagates the impact to the other level (cf Example 4 in
Table VI).

3.3.4. The query WHAT-IF-Ada

We have defined three change types for the query WHAT-IF-Ada:
—PArameter change Type (PAT);
—Type Definition change Type (TDT); and
—Assignment Statements change Type (AST).
In addition to the relations (R1–R12) defined above for the Ada software system, we
have also specified the following relations to take care of the visibility of objects beyond
 SOFTWARE MAINTENANCE 1171
Table IV.

the limit of direct relationships.

(R18) object-global-declaration :- {X, X}; this relation is defined as a transitive closure of

the relation obj-is-declared-in, (R4). The relation object-global-declaration(obj1 , obj2 ),
obj1 , obj2 ∈ X exists if and only if obj1 declare directly obj2 or obj2 is declared in a
module obji such that object-global-declaration(obj1 , obji ) exists.
(R19) var-parameter-of :- {X,Vt },Vt ∈ (V , C ); this relation is the composition of the rela-
tions is-parameter-of (R1) and var-is-declared-in (R5). The relation var-parameter-
of(obj, var), obj ∈ X , var ∈ Vt exists if:
(a) var-is-declared-in(obj, var) exists, or
(b) ∃i ∈ N , such that is-parameter-of(obj, var, i) exists.
(R20) An object x declared in obj1 is visible in obj2 if and only if
(a) (obj2 = obj1 ) or
1172 S. AJILA

Figure 11. HOOD design diagram of Ada example in Table II

(b) var-parameter-of(obj, x), such that ∀obji

(a) object-global-declaration(obj1 , obji ) and
(b) object-global-declaration(obji , obj2 ) exist.
These relations (R18–R20) are also transformed to PROLOG rules.
We give an example of the application of the query WHAT-IF-Ada in Table IV. This
example is also based on the Ada program shown in Table II.
This example shows the importance of relations (R18–R20). We can see that the object
project records is visible in data base and sort while data base is visible in project and sort
is visible in list data, etc.

3.3.5. The query WHAT-IF-HOOD

We have defined two change types for the query WHAT-IF-HOOD:
1. Object Mode change Type (OMT) and
2. Operation Constraint change Type (OCT).
The OMT change type is defined because HOOD defines several kinds of objects: passive,
active and classes. Classes are generic objects. Active objects include the dynamic behaviour
of their interface, while passive objects have no special semantics attached to their interface.
The OCT change type is defined to take care of the differences in the interface behaviour
of active and passive objects.
The execution of the query WHAT-IF-HOOD is based on the activation of the rules
specified by the relations R15, R16 and the two change types already given. We give in
SOFTWARE MAINTENANCE 1173

Table V.
1174 S. AJILA

Table V an example of the activation of this query. This example is based on the HOOD
design diagram in Figure 9. This diagram is the HOOD design for the Ada example shown
in Table II.
In this example, we note that we have two types of conclusion: micro-conclusion and
global-conclusion. The reason for this is that the conclusions given by the impact analysis
function may be ‘contradicting’ because each conclusion (micro-conclusion) is based on the
facts, the rules and the objects at hand (i.e. what the impact analysis function is working with
at the time of giving the conclusion). This working set may change for the next analysis; for
example, the facts and the rules may remain the same while the objects may be changed.
In this case the next conclusion may not be the same as the previous one. To resolve this
seeming contradiction, we have introduced the global conclusion which actually gives a
condition in the form of a conclusion.

3.3.6. The query WHAT-IF-Local and the query WHAT-IF-Global

The difference between the query WHAT-IF-Local and the other two queries, WHAT-
IF-Ada and WHAT-IF-HOOD, is that WHAT-IF-Local is not constrained by any change
type and because of this it can be applied to all the objects of a given level (Ada or HOOD).
This query, unlike the other two, does not give any conclusion but the impacts. The query
WHAT-IF-Global has the same characteristics with the exception that it also propagates
the impacts from one level to another. Because of space, we will only give an example
of the query WHAT-IF-Global (Table VI). In this example we can see that we have two
levels of impact. The impact was started with HOOD: Query :- WHAT-IF-Global(h project,
HOOD) and this was propagated to the Ada level by the global mapping:
–>GLOBAL MAPPING ===>h project :- IS MAPPED TO ===>project.

3.3.7. The query WHAT-LIST

The query WHAT-LIST gives a list of objects at any given level. This query interacts
directly with the database. The syntax is: WHAT-LIST(level) where level indicates Ada-view
or HOOD-view. An example is given below:

3.4. The user interface

The user interface is implemented by using XF. XF is an integrated programming envi-
ronment that supports the development of graphical user interfaces. XF takes advantage of
Tk, a MotifTM -like widget set that is accessible through Tcl, an efficient interpreted pro-
gramming language. This interface is based on the notion of windows and menu. Therefore,
it is menu driven (cf Figures 12 and 13).

4. RELATED WORK AND DISCUSSION

Researchers have addressed the impact analysis of object change in software from different
points of view:31
1. programming-view: from a programming view by the use of incremental data flow
techniques.33 Similar principles are applied in retesting strategies for modified systems
by combining data flow analysis and logic coverage techniques.20 A good example
SOFTWARE MAINTENANCE 1175

Table VI.
1176 S. AJILA
Table VII.

in this category is the ‘Maintenance Assistant Project’ at the Software Engineering

Research Centre.37,38,39 In this project, program dependency classes are based on data
flow, definition, calling and functional dependencies.

2. document-view: in the course of construction of a software system, software documents

are produced which describe the system from different points of view. We give below
two examples in this category:
(a) Software Document Support environment (SODOS): this environment was de-
veloped at the University of Southern California to support the development and
maintenance of software documentation.22 SODOS uses an object-oriented model
and hypermedia graph of relationships to manage software life-cycle documents.
(b) USC System Factory Project:34 part of the prototype of this system is a hypertext-
based documentation integration facility (DIF). DIF provides a mechanism for
developing and maintaining software documentation with its associated relation-
ships.
3. mixed-views: in this category we have impact analysis tools that cover more than one
phase of the software life-cycle. We would like to point out that our work falls into
this category. It is true that in the document view some of the tools cover more than
one life-cycle phase but the documents in question are of the same form (text) and
presentation. In the case of mixed views, we do not always have the same structure.
We give below a brief survey of two systems in this category.
(a) Maintenance Assistance Capabilities for Software (MACS): In MACS a Proto-
type Software Evolution Expert System (SEES)7,8,9 was implemented on top of
Tool Builder’s Kit (TBK). TBK is a productive environment for the new gener-
ation of integrated CASE tools. The approach is based on modelling the main
objects in any view of the software system in a frame-base representation and
on making explicit the related dependencies in meta-knowledge bases.
 SOFTWARE MAINTENANCE 1177

Figure 12. The menu driven window of the interface

The SEES Knowledge Base Model is specified by outlining the distinction of

the concepts of Views, Model of Views and the Software System Model. The
SEES design and analysis proposes a classification of the types of change that
may trigger the evolution of a software system. This classification is used as
a basis for defining requirements on representations and operations to support
software evolution.
(b) Unified Model (UM): UM is designed to express dependencies between anal-
ysis and design products, respectively DCFD (Data Control Flow Diagrams)
and HOOD (Hierarchical Object Oriented Design) diagrams.16 The approach
presented in UM is to investigate the traceability between analysis and design
software artifacts and to propose a mechanism for recording those traces under
a PCTE (Portable Common Tool Environment).
The Unified Semantic Model (USM) of UM is designed to implement vertical
traceability between analysis and design artifacts. The model (USM) is a ‘bridge’
connecting two different conceptual levels, so that changes in either level can
be propagated through it.
Our work can be compared with the MACS project and the Unified Model (UM) system.
In contrast to MACS, the WHAT-IF model is a generic knowledge-based model. It can
be applied on the basis of the problem to be solved. This gives some flexibility in terms
of application. The WHAT-IF prototype, as well as the Unified Semantic Model (USM)
of UM, address two life-cycle phases. USM addresses the analysis and design products
while WHAT-IF addresses design and coding phases of the life-cycle. UM uses PCTE as a
platform between analysis (DCFD) and design (HOOD). In WHAT-IF, the unifying factor
1178 S. AJILA

Figure 13. The result of the query WHAT-IF-Ada (sensor, PAT)

is the dependence graph.

The possible advantage of our work over other related work is that the abstraction sub-
system of the WHAT-IF prototype is designed in such a way as to allow the users to
formulate relations by adding rules to the system and by relating the existing ones in
such a way as to prove some assertions or to create personal queries. In addition to this,
the implementation of the abstraction sub-system connects real-time answers to a set of
pre-defined queries (cf Subsection 3.3.1). The pre-defined queries are easy to use. Another
advantage is that it is relatively easy to set up the prototype. This is so because the graphical
interface can be separated from the rest of the system. In a situation without graphical
support the prototype can be set up in such a way that the result will be written into a file
and this can then be printed. The queries can also be used on-line without the graphical
supports.
The maintenance process can be viewed from both a product and process perspective.
In either way, the approach presented in this paper is useful for the purpose of an impact
analysis of a change. The WHAT-IF Impact Analysis Tool can be used to analyse a priori
the impact of a change. This will enable the management to determine (or have an idea)
of the cost of carrying out a change. Therefore, as changes are requested, the impact can
be assessed in order to have ‘first hand’ information which is useful for the estimation of
the cost. The result produced by the WHAT-IF impact tool is also useful for the purpose
of regression testing after the change and test planning. We know that regression testing
is vital to guarantee the quality of the ‘new’ product, but it tends to be expensive and is
 SOFTWARE MAINTENANCE 1179

often not performed. We believe that, with the WHAT-IF impact tool, the cost of regression
testing can be reduced.
At present, the WHAT-IF Impact Analysis Tool has not been integrated into any software
development environment neither has it been tested with a ‘big’ software project (an indus-
trial type of project) but the results of the laboratory tests (examples of which are given in
Section 3) are very encouraging.

5. CONCLUSION AND FURTHER WORK

The maintenance of software requires a tool for impact analysis and propagation of a change.
This paper has presented an approach for the design and implementation of such a tool.
The WHAT-IF model captures four life-cycle phases of the software.
Even though the WHAT-IF model prototype is based on the HOOD and Ada views, the
global architecture itself (Figure 1) is general and could therefore be used for other types
of view. The prototype also allows the users to participate by building their own relations.
The queries are sufficiently simple to use through the user interface.
This experimentation has shown that a logic-based programming language can be a ‘time
saving’ prototyping language.4 We do not recommend PROLOG as a standalone language
for practising maintenance engineers, but a PROLOG-like language combined with other
programming languages can be a good start for an experiment of this kind. Finally, we
are aware of the problem of the performance with the rule-based systems, most especially
if such systems are implemented in PROLOG (as in our case). However, with the recent
scientific development and research into the integration of relational databases and logic-
bases systems and the availability of inter-process communication with a graphical user
interface environment, we believe that the performance can be improved.
As part of further work to be done, we intend to use the framework for a comparison
proposed by Arnold et al.6 to evaluate our work. The framework outlines three parts:
impact analysis application, impact analysis parts and impact analysis effectiveness. Impact
analysis application examines how an approach is used to accomplish impact analysis.
Impact analysis parts examines the nature of the internal parts and methods used to actually
perform the impact analysis. Impact analysis effectiveness examines the properties of the
the resulting search for impacts, especially how well they accomplish the goals of impact
analysis.
The most interesting part of the framework for our work is the impact analysis effective-
ness. Arnold et al. proposes four concepts that can be used to measure the result of impact
analysis. In order to apply these concepts we need to integrate the ‘Impact Analysis Tool’
(IAT) of WHAT-IF into a software engineering environment having good graphical support
for HOOD tools and a powerful text editor for the code. This will enable us to compare
the input to IAT and the result produced by the tool. It will also enable us to measure the
result with respect to the change proposed by the user and relate the measure to the actual
project cost.
Another area of further work is to incorporate the requirements and the specification
views into the prototype. We intend using DCFD for the requirements view16 and a formal
approach for the specification. In addition to that we also intend extending the mixed-views
proposed in WHAT-IF model to cover the testing view.
Finally, we believe that our approach is practical and it will benefit people engaged in
software design and maintenance.
1180 S. AJILA

ACKNOWLEDGEMENTS

The author would like to thank Professor Jean Claude DERNIAME of CRIN-CNRS, Nancy,
France, Dr Henri Basson at the University of Littoral, Calais, France and Dr Amos David
of CRIN-CNRS, Nancy, France for their helpful discussions. The author would also like to
thank the reviewers for their comments and suggestions on the paper.

REFERENCES

1. S. Ajila, H. Basson and N. Boudjlida, ‘Software process assistance: a case study of the impact of object
modification during software development’, 1st African Conf. on Research in Computer Science, Yaoundé,
Cameroun, 14–20 October, 1992, Vol 1, pp. 73–84.
2. S. Ajila, ‘Software process assistance: management of objects modifications in an integrated software
engineering environment’, Nigerian Journal of Science, 28, (1993).
3. S. Ajila, H. Basson and J. C. Derniame, ‘WHAT-IF: a function to estimate the impacts of potential changes
in a software’, Second Int. Conf. on Software Quality Management (SQM 94), Vol. 2, 26–28 July 1994,
Computational Mechanics Publications, Southampton Boston, UK, pp. 379–394
4. S. Ajila, J. C. Derniame and H. Basson, ‘A logic-based approach to impact analysis of objects change’,
Third Int. Conf. on the Practical Application of Prolog, 3–7 April, 1995, Paris, France.
5. J.-M. Alliot and T. Schiex, Intelligence Artificielle et Informatique Théorique, Collection Intelligence Ar-
tificielle, Cépadués Editions, Paris, France, 1994.
6. Robert S. Arnold and Shawn A. Bohner, ‘Impact analysis – towards a framework for comparison’, Proc.
Conf. Software Maintenance, 27–30 September, 1993, Montreal, Quebec, Canada.
7. G. Avellis, ‘Software evolution expert system – requirement analysis’, Delivery Report,
WD/RE/SEES/151089-3, 15 December, 1989.
8. G. Avellis, ‘Specifications of SEES knowledge base model’, Delivery Report, WD/RE/SEES/150591, 15
May 1991.
9. G. Avellis, ‘CASE support for software evolution: a dependency approach to control the change process’,
5th Int. Workshop on CASE, 6-10 July, 1992, Montréal, Québec, Canada, pp. 62–73.
10. Henry Basson, M. C. Haton and J. C. Derniame, ‘Characteristics graph of software quality’, Int. Symp. on
Computer and Information Sciences VII, November, 1992, pp. 455–461.
11. Lionel C. Briand, Sandro Morasca and Victor R. Basili, ‘Measuring and assesing maintainability at the end
of high level design’, Proc. Conf. on Software Maintenance, 27–30 September, 1993, Montréal, Quebec,
Canada.
12. S. D. Cooper and M. Munro, ‘Software change information for maintenance management’, Proc. IEEE
Conf. on Software Maintenance (CSM89), Miami FL, 1989, pp. 279–289.
13. C. G. Davis and P. J. Layzell, ‘Rules to govern change in JSD-based cystems’, Proc. IEEE Conf. on
Software Maintenance (CSM88), Phoenix AZ, 1988, pp. 34–40.
14. D. Sandra Fay and G. Denise Holme, ‘Help! I have to update an undocumented program’, Proc. Conf. on
Software Maintenance, IEEE Computer Society Press, 1985, pp. 194–202.
15. J. Ferrante, Karl J. Ottenstein and Joe D. Warren, ‘The program dependence graph and its use in optimiza-
tion’, ACM Trans. Programming Languages and Systems, 9(3), 319–349 (1987).
16. P. Fillon, N. Mitsuda, A. Sawada, T. Ajisaka and Y. Matsumoto, ‘A facility to trace dependencies between
PCTE objects for software maintenance, Proc. PCTE ’94 Conf., California, USA, December, 1994.
17. L. Daniel Freedman and M. Gerald Weinberg, Maintenance Reviews, Techniques of Program and System
Maintenance, Winthrop Publishers, Cambridge, MA, 1982.
18. R. Gopal and S. R. Schach, ‘Using automatic program decomposition techniques in software maintenance’,
Proc. IEEE Conf. on Software Maintenance (CSM89), Miami FL, 1989, pp. 132–141.
19. M. T. Harandi and J. Q. Ning, ‘Knowledge-based program analysis’, IEEE Software, January 1990.
20. J. Hartman and D. Robson, ‘Techniques for selective revalidation’, IEEE Software, January 1990.
21. Jean Paul Haton, Les systèmes à base de connaissances, Intelligence Naturelle et Intelligence Artificielle,
Symposium de l’association de Psychologie Scientfique de Langue Française, Rome, 1991, Presse Univer-
sitaires de France, pp. 19–42
22. E. Horowitz and R. Williamson, ‘SODOS: a software document support environment - its definition’, IEEE
Trans. Software Engineering, SE–12(8), 1986.
23. M. Lai, Conception Orienté Objet: Pratique de la Méthode HOOD, DUNOD, Paris, 1991.
 SOFTWARE MAINTENANCE 1181

24. Janusz Laski and Wojciech Szermer, ‘Identification of program modifications and its application in software
maintenance’, Proc. Conf. on Software Maintenance, 9–12 November, 1992, Orlando, Florida.
25. Cécile Legout, ‘Electre, un langage cible pour la methode HOOD dans le cadre de l’atelier de génie logiciel
CONCERTO’, 5th Int. Conf. on Software Engineering & Its Applications, Toulouse 1992, 7–11 December,
1992.
26. M. M. Lehman and L. A. Belady, ‘Program evolution: processes of software change’, APIC Studies in
Data Processing No. 27, Academic Press, New York, 1985.
27. K. J. Lieberherr and I. M. Holland, ‘Tools for preventive software maintenance’, Proc. IEEE Conf. on
Software Maintenance (CSM89), Miami FL, 1989, pp. 2–13.
28. Panos E. Livadas and Pradal K. Roy, ‘Using dependence analysis’, Proc. Conf. on Software Maintenance,
9–12 November, 1992, Orlando, Florida.
29. Joseph P. Loyall and Susan A. Mathisen, ‘Using dependence analysis to support the software maintenance
process’, Proc. Conf. on Software Maintenance, 27–30 September, 1993, Montréal, Quebec, Canada.
30. Girish Parikh, ‘Restructuring your Cobol programs’, Computerworld Focus, 20(7A), 39–42 (1986).
31. S. L. Pfleeger and S. A. Bohner, ‘A framework for software maintenance metrics’, Proc. IEEE Conf. on
Software Maintenance (CSM90), 1990, pp 320–327.
32. Charles Rich and Richard C. Waters, ‘Knowledge intensive software engineering tools’, IEEE Trans.
Knowledge and Data Enginering, 4(5), (1992).
33. N. Schneidewind, ‘Software maintenace: the need for standardization’, Proc. IEEE, 77(4), 618–624 (1989).
34. A. Taha and S. Thebaut, ‘Program change analysis using incremental data flow techniques’, SERC-TR-14-F,
CIS Department, University of Florida, Gainesville, FL 32611, USA, 1987.
35. Mats Weber, ‘Elaboration order issues in Ada 9X’, ACM Ada Letters, XIII(1), (1993).
36. B. Westfechtel, ‘A graph-based approach to the construction of tools for the life cycle integration between
software documents’, 5th Int. Workshop on CASE, Montréal, Québec, Canada, 6–10 July, 1992.
37. N. Wilde and Allen Chapman, ‘Describing object oriented software: what maintainers need to know’,
SERC Purdue University and University of Florida, Technical Report No. SERC-TR-54-F, 1992.
38. N. Wilde and R. Huitt, ‘A data-base program representation for software maintenance tools’, SERC-TR-
25-F, CIS Department, University of Florida, Gainesville, FL 32611, USA, 1988.
39. N. Wilde, R. Huitt and S. Huitt, ‘Dependency analysis tools: reusable components for software mainte-
nance’, Proc. IEEE Conf. on Software Maintenance (CSM89), Miami FL, 1989, pp. 126–131.
40. S. S, Yau and P. C. Grabow, ‘A model for representing programs using hierarchical graphs’, IEEE Trans.
Software Eng., SE7, 556–574 (1981).