Adapting Object-Oriented CAD Database Concepts for Computer Aided Software Engineering (CASE)


Jeffrey S. Poulin and Martin Hardwick

Department of Computer Science

Rensselaer Polytechnic Institute

Troy, New York 12180


Abstract- There is currently a large research effort underway to develop new techniques and methods for the efficient development of software.  However, much of this effort ignores the vast sum of knowledge that has been acquired through our experiences in the field of engineering CAD, especially in the area of VLSI design.  Much of what has been learned in this area centers on database support for the design process, and in particular, efficient object-oriented modeling techniques for design data.  We believe that the data model for software is the central issue surrounding the development of CASE systems.  After comparing the nature of the design process in the fields of VLSI and Software, a semantic data model for supporting software development is presented. 


Keywords- Software reusability, CAD Databases, Software Engineering, CASE

1         Introduction

Engineering CAD applications and designers have at their disposal a host of tools to assist in the development of their products.  The experience that these designers have with CAD/CAM systems has demonstrated that good application tools contribute greatly to the design process.  With this in mind, recent work has been to develop similar tools for software engineering.  Early efforts in this area indicate that many of the same advantages that were realized by the users of engineering CAD systems can be found using a CASE system for the development of software.  However, since the software design process is less well understood than that of other engineering disciplines, much work remains to be done.


One thing that has been learned is that there are many benefits to using an underlying database to organize the large quantities of data typically managed by a CAD environment.  However, many CAD tools have employed conventional database systems in this role, only to discover that the relational data model on which they are based cannot gracefully represent engineering design data.  It is necessary to view this data as a set of design objects, and new models for design data based on this object-oriented concept were developed.


If an object-oriented database supported design tool is to be effective, it has to represent application data in a form that matches the user’s conceptual view of the data.  This requires that these tools be built around a semantic data model that is appropriate for the application.  Therefore, at the core of these tools is a data model that mirrors the design of the final product.


This paper concentrates on data modeling requirements for software engineering.  Section 2 provides a synopsis of the advantages of applying database technology to engineering design environments.  Section 3 motivates the application of CAD research to the domain of software engineering.  In Section 4, some of the problems that are largely unique to software engineering are discussed.  A review of current CAD modeling techniques such as complex objects and molecular objects is provided in Section 5.  In Section 6, our data model for CASE applications is shown and explained.  Section 7 addresses how this model relates to object-oriented programming and design doctrine.  Finally, Section 8 discusses some of the implementation issues faced in the development of a software engineering environment based on this model.

2         The Argument for Database Support of CAD and CASE

As engineering CAD systems have developed, at some point all have had to address the problem of managing large quantities of design and administrative data.  As the design and size of the project grows, so does the average response time for basic operations.  This dictates the need for efficient access to design data for the purpose of updates as well as for the generation of reports and other documentation.  Developers of CAD systems quickly realized the advantages of capitalizing on readily available database technology, and applied such indigenous database features as file indexing, standard query language capabilities, and controls on access to data for concurrency and security reasons.

3         The Correspondence Between VLSI CAD and CASE

Most of the research and current literature involving engineering CAD systems has centered on VLSI applications.  The goal of this project is to adapt what has been learned in the field of VLSI here at RPI [Har87a, Sam87] and elsewhere [Bat85, Bro84, Buc85, Gut82, Has82, Kat85] to the domain of software development [Pou88].  The motivation is not only the desire to use as much of what is already known as possible, but the realization that there is a series of basic design problems that are the same; namely, version control, the large quantities of data, and the reuse of existing components.  Furthermore, both VLSI and software development generally employ a hierarchical design process, using stepwise refinement as a means of reducing the problem to pieces of manageable size.  Finally, both processes employ various graphical tools to assist in the design.  The tools not only aid the designer by providing a spatial representation of a design object, but may allow the designer to view the problem from different perspectives, such as the logical function of a circuit versus its physical layout. 

4         Special Requirements of CASE

There are several problems that are present in the development of software that do not arise in VLSI design.  A large portion of these is derived from the fact that software is primarily composed of ideas and algorithms, in contrast with a VLSI component that is both conceptually and physically more concrete.  In a sense this makes the design and manufacture of VLSI components “easier,” and therefore extensively studied and well defined.


The first issue that is adversely affected by this is the reusability problem [Rei87, Weg84].  At low levels, both VLSI components (adders, shifters, and registers) and software components (stacks, queues, and searching routines) are fairly standardized and are quite well understood.  Therefore, they can be easily classified.  However, where concepts and algorithms predominate at the higher levels of design, the categorization of components becomes much more abstract.  This problem is especially true in the case of software.  Of course, effective reuse depends on the ability to access available components.  What is required is a catalogue of design objects based on some universal criteria or keywords [Onu87], a subject that is currently getting much attention [Bur87, Pri87].  A CASE system would then be able to use these keywords to assist the program designer in locating related and potentially reusable components.


An additional problem caused by the greater abstraction in software design is that there remains the need to manipulate large quantities of text.  In a typical VLSI CAD application, the user rarely deals with chip geometries and layouts directly in file form; rather, he completes his design almost totally in a graphic manner.[1]


In software, the final product is composed of text, and although there is currently a lot of work being done in the area of graphical programming [Bal85, Rae85], there exists a point in the software design process when this fact must be addressed.  In short, the CAD system does a better job at hiding the details of the actual physical design representation, an ability that should be extended to the design of software.


Another problem found in software is the management of an infinite number of possibly very complex data types.  In contrast, data types in VLSI tend to be few and are better defined, (for example, simple bits, or combinations of bits into bytes or words).  Some of these bit combinations may be in a specified format such as integer or real, but the complexity of these formats does not rival the intricate data structures found in structured programming languages.  Any CASE system must be capable of handling this diversity of data.


Finally, wherever a VLSI circuit is needed in the final product, a copy of it is made and burned onto the chip.  This instantiation of the design object does not apply to software, where only one copy of a module may exist at any time.  In VLSI CAD, a reference to an object during the design stage results in a copy of the physical object later.  In software, references to an object during the design stage remain references to a single copy of that object, even in the final product.


5         Existing Database Support for CAD Systems

5.1      For VLSI

Although many VLSI CAD systems exist and have been supported by relational databases, it is now well recognized that the standard relational database is inadequate for modeling and storing design data [Bat84, Has82, Hel87, Sid80].  To meet the need for a powerful and straightforward representation of design objects, complex objects and object-oriented database systems were developed [Kim87, Lor83].  Complex objects are hierarchical groups of tuples consisting of a root tuple that represents a data object, and a set of dependent tuples that define the object.  Complex objects can succinctly represent the recursive, non-disjoint objects that the relational model cannot easily handle [Bat84, Buc85]. 


The molecular object model is an example of a semantic data model for design data that is based on complex objects.

In this model, objects are defined to have two distinct parts, an interface and an implementation [Bat85, Kat85].  As shown in Figure 1, the interface of an object consists of connections to the outside world, and defines how other objects can use or access the design object.  It contains the common attributes of the existing implementations for that object.  The implementation of the object defines how the object does its job, and in VLSI is typically made up of instances of sub circuits and the wires that interconnect them.


Figure 1.  A 4-input AND gate.


 When using the molecular model, a designer may refer to a sub circuit without specifying an implementation for that sub circuit. In this case, the designer references only the interface for the object.  If the designer does not bind an implementation to this interface, the interface is referred to as a socket in the design of the higher-level circuit.  This socket must then be plugged with an implementation for that interface before the design can be considered complete. The plug becomes an instance of the subcircuit in the design.

5.2      For Software Engineering

To date, software engineering database support has been primarily provided by relational systems [Gut82, Rou83].  This is especially true with regards to commercially available environments.  Papers on object-oriented database support for software engineering tend to concentrate on the functionality and requirements of the final product [Ber87, Rom87, Yau87].  So while there are many existing CASE systems, there is not a lot of research focusing on data modeling problems.  However, the data model is a central issue that should motivate the design of the entire CASE system.  For this reason, we have developed an object-oriented data model for software, and are currently implementing a CASE environment based on this modeling concept on DEC VAXstations using the ROSE engineering database system.


ROSE is an experimental database system that provides graphics and user interface abilities for CAD applications.

ROSE is fast, manages data clusters as objects, and provides access to the database through a combination of powerful

control structures based on the ‘C’ programming language and database commands that are an extension of the relational algebra [Har87a, Har87b].  Additionally, data structures in ROSE are defined in an expressive AND/OR tree format [McL83].


Figure 2.  The Four Types of AND/OR Trees.


Each node in an AND/OR tree defines a domain for an object or one of its sub-objects.  AND nodes represent aggregation abstractions, OR nodes represent generalization abstractions [Smi77], and an asterisk under either type of node represents an association abstraction [Bro84]. Figure 2 contains examples of these abstraction types.

6         A Data Model For CASE

The primary concern in developing a data model for CASE is that it must address the issues of design data capture, classification, and retrievability discussed above, and it must be supportable by an underlying database system.  It must also give consideration to the practical requirements of implementation, and any possible restrictions on access time and memory management.  In our model, the software module is considered the basic design object.  The model is strongly influenced by the molecular object model of Batory because of a natural correspondence between the molecular view of VLSI circuits and object-oriented view of software modules.  First, there is the similarity between the VLSI interface, which is composed of a list of pins, and the software interface, which is composed of a list of parameters.  Second, there is the correspondence between the VLSI implementation, which is composed of gates, sub circuits, and wires, and the control statements, calls to subprograms, and flow of control found in the implementation of a software module.  Our data model extends the two-part molecular model in order to fully support the unique requirements of the software design process.


The following modifications to the molecular model are required for use in a CASE system.  First, the concept of instantiation needs to be adapted to mean a reference, and not a copy.  Second, in the molecular model the interface of the object is used in two distinct roles, one in the definition of the object and one in the definition of a socket, which is a call to an object.  We argue for dividing the interface portion of the data model into two parts in order to accommodate this conflict. 


Modification of the concept of instantiation is required in order to clarify the notion that in the final software product, just as in the software design, only one copy of a design object exists.  Calls, or instances of the design object, are references to a software module; during execution of the program a call to a module results in the transfer of program control to the physical location of that routine.  In VLSI design, an instance of a design object is a reference to a single copy of the design specifications for that circuit.  However, in the final hardware product, every instance of a sub circuit in the design results in a copy of that circuit being created and placed on the chip.  This view of instantiation, which implies the creation of multiple copies of the design object, is not appropriate for software. 


The second change to the VLSI model involves the dual role of the molecular interface.  The problem with the interface in the molecular model is that the interface found in the declaration and the interface found in the module call are not the same, and, in fact, have completely different functions.  The difference is that the interface found in the declaration represents the object.  Since it may represent many alternatives and versions of that object, there are strict limitations on how the user may modify the interface.  The interface in the module call, on the other hand, is a request for

service.  It differs from the declaration interface because it evolves with the design, and may not even represent a particular design object, especially early in the design process.  Since the details of the request may change as the design develops, the second kind of interface should, in a sense, be more flexible, and provided with operations that help guide the designer towards satisfying the need for service.  An example of the two ways that a module interface

is used is shown in Figure 3.


The software interface in the module declaration:

Procedure Sort(Var Variable1: Array[1..100] of integer;

Variable2: Boolean);


The software interface in the module call:

Sort(Integer__array, Error__flag);


Figure 3. The Two Roles of the Software Interface.


What typically happens is that when a designer needs a subprogram to perform some function, he first starts with a partially defined requirements statement representing what he needs to do.   With this need in mind, he searches the software database for available routines.  If he finds an appropriate routine, he incorporates the interface for it into his design as a subprogram call (socket).  If not, he must design his own interface, thereby creating a new interface object.    However, this is difficult to do given that requirements are often ill defined and likely to continue to evolve over time.  However, since the molecular model treats definition interfaces and request-for-service interfaces the same, the designer must fully detail the required interface at this time.  This is required because by treating the defining interface as common to many implementations, the molecular model necessarily places a host of restrictions on how they can be modified later. 


What is needed is a meta-interface that provides a place for the designer to sketch the requirements for a subcomponent without the need to know exactly what he wants.  A new object type, the <f 2>call,  is introduced to meet this need. The call not only has a much more flexible set of operations to allow the interface to evolve, but it also has an associated set of descriptive keywords and performance constraints that the designer may use to assist in electing available components to fill the socket. 


The resulting three-part model for software, termed the Interactive Development Model (IDM), is shown in Figure 4. 

The interface portion of the IDM is used exclusively to describe software modules, and plays the declaration role of the molecular interface, as shown in Figure 3.  The alternative portion of the model is also used exclusively for describing the software module; however, the alternative defines the code and other implementation details of the module.  The final portion of the model is the call.  The purpose of the call is to describe a request for service in a sufficiently abstract way, in order to assist the designer to bridge the gap between requirements and any objects available to meet those requirements.  It is important to remember that the call represents what the designer thinks he needs, where the interface and alternative objects serve to define the object.  For this reason, components of the call, such as the parameter list, are tentative and modifiable, where in the interface and alternative they define a module and may not, in general, be changed.  Other parts of the model and their function are described below. 

Figure 4.  The Data Model for Software


The “Header” fields contain the object name and other administrative data about the object, such as the name of the

responsible designer and the date the object was created.  The “Description” in the interface and the call portion of the object consist of general comments and a set of keyword identifiers.  These keyword identifiers are used in the call to outline the call’s required service, and serve as search parameters into the library of interfaces, where the keywords tell the user the attributes of each interface.  The “Performance__Constraints” in the module call correspond to the “Performance__Attributes” of each alternative implementation.  As in searching the software library for interfaces based on the keyword identifiers, suitable alternatives of an interface are located in the library based on the call’s stated constaints on performance and on each alternative’s ability to meet those constraints. 


The addition of the call object to the molecular data model has an important consequence for the reuse of design objects.  Because there are criteria available to link requests for service with routines potentially able to meet them, the filling of module calls (sockets) should and can be automated. In many cases this may even be left until the design is compiled or validated.  At that time, the program that compiles the design will have three sets of criteria on which to base the selection of an appropriate module to fill the call.  First, the initial description of the call provides keywords that partially identify an interface.  Second, the parameter list in the call is compared with the parameter lists in the potential interfaces in both number and type of parameters.  At this point, if there are candidate routine(s), an interface for the call is identified, or at a minimum narrowed to a few choices that the designer may easily browse.  Finally, the performance constraints specified in the call are matched against the performance attributes of the alternatives for the identified interface in order to provide a final selection of implementation. 


The “Bound” fields in the call object allow the designer to save the name of an interface or alternative once he has located one that will fill the call.  Note, however, that with the IDM it is possible for objects filling a call to be referenced explicitly as a specific reference, or implicitly according to some default criteria.  The designer has the option of which type of assignment to make.  A temporary, or implicit, assignment would be used if he always wanted to fill the call with the latest version of a given alternative.  An explicit, or permanent, assignment would be warranted if, for example, he specifically wants version V1.2 of “Binary__sort” to fill a given call. 


This matching of constraints with performance criteria and the use of dynamic binding of calls with components makes the IDM particularly effective for use for an evolutionary design process.  The advantages of dynamically binding design objects to sockets have also recently been recognized by Dittrich and Lorie [Dit88].  While their solution defines and stores “environment” characteristics that will be used at run time to qualify candidate objects for sockets, we feel that for conceptual clarity it is better to incorporate this and all other design-related information directly into the design data. 


Figure 5. Object Hierarchy.


Control of versions and alternative implementations is accomplished in the data model through the use of association abstractions (multiple reference pointers) from one part of the model to the others.  The relationships between these objects are defined in the Entity-Relationship diagram of Figure 5.  A design object, which in our case is a software module, has exactly one interface.  The interface is stored as an Interface object as defined in Figure 4, and may have any number of alternative implementations in the design.  While the differentiation between alternatives of an interface is left to the user, they are typically distinguished by a fundamental difference in algorithm.  These alternatives are related to the interface by means of forward pointers in the association abstraction called “Alternative__list.”


Each alternative implementation is represented in the database by the Alternative object of Figure 4.  An alternative may be composed of any number of versions, which are snapshots of the alternative, usually as it evolved over time in response to bugs or minor changes.  The version of the alternative contains the pseudo-code representation of the module’s function, and is accessed through the association abstraction “Version__list.”  It is worthwhile to note that other related concepts, such as configurations (implementations for some specified purpose such as site requirements), are easily incorporated into the model as alternatives. 

7         Relationship to “Object-Oriented” Program Design

One of the most important aspects of the object-oriented design paradigm is the encapsulation of certain parts of the design in order to hide implementation details from the user [Boo84, Pyl81, Wir85].  In this context we define “object-oriented” not in the Smalltalk sense but rather in the Ada sense, as data and related operations encapsulated or packaged so as to be invisible to and protected from the user.  However, care must be taken in object-oriented design so that during the design process certain key data remains accessible.  Much of the information that the designer requires in order to (i) retrieve appropriate design objects from the database, and (ii) select the best implementation for his needs, is specific to the implementation and is not available by looking solely at the interface for that object. 


The data model above reflects the belief that, although one part of the object’s interface is, indeed, composed of the externally visible attributes common to all its implementations, the complete interface must provide access to all the externally visible attributes of a module, even those that are implementation-specific.  The primary examples of such attributes are performance characteristics that may influence a designer’s decision to use that object, such as those required in order to conform to a set of design constraints.  Other such attributes include memory requirements, object code size, and the time complexity of the algorithm.  In the VLSI domain, such attributes include surface area, power consumption, and delay time. 

8         Implementation in ROSE

The complete ROSE implementation of a CASE system based on this data model includes several graphical editors for design data input, a combination of pull-down menus and text-entry boxes, and mouse-driven icons.  These interface considerations are techniques designed to assist the designer as he interacts with the system and uses the reusability features that are built into the model.  A sample session with the ROSE CASE Tool is shown in Figure 6.


For support of reusability, a common set of descriptors for the software library and the performance attributes of the library components has been defined.  These descriptors are based largely on the work of [Pri87] and [Bur87] and are used as keys during searches of the software library. 



Screen shot not available.



Figure 6. Sample Session with the CASE Tool.


First, in order to reduce the amount of text entry and to distance the user from the details of the data model, the CASE system provides real-time graphic tools to support the design process.  Several major software design methods, including data flow, data structure, and structured design, were considered for this purpose [Ber85].  Mappings from these methodologies to program structure are well defined [Jac75, Mye78, You75], and are used to dynamically capture graphical edits made by the designer.  These edits are stored in a pseudocode-type format.  This is possible because the ROSE AND/OR tree description of the data model mirrors the Backus-Naur descriptions used by many of the major structured programming languages.  This AND/OR tree description is also used to describe the multitude of data structures available to the software designer.



Screen shot not available.



Figure 7. The CASE Tool Dynamic Structure Editor Options.


As a example of how the designer uses the tools to develop an abstract request-for-service into a specific module instance, consider the case where the software designer has a need for some kind of sorting routine.  Using the edit options provided in Figure 7, he adds a generic call to the design, graphically portrayed in the Dynamic Structure Editor by a Call icon. 

Through the use of a set of text-entry tools, the designer has tagged this call icon as some kind of sort.  Using the same text-entry tools, he now consults the public software library for possible routines that might be available for his use.  After querying the software library he discovers an existing interface for an integer array sort that appears as if it will serve the function he requires.  Content with this possibility, he binds the interface to the call and proceeds to investigate possible alternative implementations for this interface.  The graphical editor responds by replacing the call icon with an interface icon.

Again using the text-entry and searching tools, the designer finally uncovers in the library a quicksort version of the interface that he prefers to use, based on his particular time constraints, memory use, and other considerations.  He binds this alternative to the call and thus fulfills his original software requirement.  He could just as easily copied the existing software into his workspace and mode modifications to the code if it were necessary.  As shown below, the graphical editor now replaces the interface icon with an alternative icon, thereby representing the fact that the call has been plugged with defined components.

9         Conclusion

In light of the numerous advantages of database support for the engineering design process, much research has gone into applying this technology to CAD, especially with respect to VLSI design.  Little, however, has been done to use this knowledge in the design of software.  We have identified several of the important similarities as well as some significant differences in the two domains.  These include the question of reuse of program modules, version control requirements, and textual versus graphical representations.  We have demonstrated a data model specifically for use in

CASE that addresses these issues, and identified valid operations on the model.


The three-part Interactive Development Model separates the interface portion of the molecular object model into two parts.  These two parts reflect the two different roles that an interface has in the semantic representation of design data. First, the interface defines and represents the object.  Second, the call is used to a request service from the object.  Finally, the third part of the data model is the alternative implementation of the object.  Operations on this model are identified in the appendix and are adapted to the roles each part of the model plays in representing the software module.  Also, while this data model was specifically developed for use in a software engineering environment, much of this concept applies equally well to engineering CAD/CAM applications. 


Implementation of the prototype CASE environment based on these modeling ideas is well underway.  Several graphical editors are provided, including a Structured Diagram editor, a program structure editor, and a variety of formatted text entry tools.  A comprehensive query facility is provided and activated through a series of pull-down menus and mouse-driven icons.  Special emphasis has been placed on providing flexible access to the reusability characteristics built into the model. 


Future work will focus on completing the prototype CASE system.  As this work continues, we will be addressing various questions involving user interface considerations for flexible, efficient access to the data model, as well as special application area topics such as might be required in an operating systems environment.  Special support for animation of control flows and data flows are also being explored.  Such a capability will allow the designer to receive a visual demonstration of how the design will perform or react to particular modifications.

10   Acknowledgements

We gratefully acknowledge the many contributions made by David L. Spooner in the preparation of this paper.  We would also like to acknowledge the help of Margarita Rovira in providing diagrams for the text.  This work was partially supported by a grant from the International Business Machines Corporation, Meyers Corners, New York.


Appendix: Operations 


On Interfaces-      



On Calls-



On Alternatives-


11   References

[Bal85]  Balzer, Robert. “A 15 Year Perspective on Automatic Programming,” in IEEE Transactions on Software Engineering, Vol. SE-11, No. 11, November 1985, pp. 1257-1268.


[Bat84] Batory, D.S. and Alejandro Buchmann, “Molecular Objects, Abstract Data Types, and Data Models: A Framework,” in Proceedings of the 10th International Conference on Very Large DataBases, Signapore, August, 1984, pp. 172-184.


[Bat85] Batory, D.S. and Won Kim, “Modeling Concepts for VLSI CAD Objects,” in ACM Transactions on Database Systems, Vol. 10, No. 3, September 1985, pp. 322-346.


[Ber85] Bergland, G.D. “A Guided Tour of Program Design Methodologies,” in IEEE Tutorial on Software Quality Assurance, ed. Tsun S. Chow, IEEE Computer Society Press, Silver Spring, Maryland, 1985, pp. 219-243.


[Ber87] Bernstein, Philip A., “Database System Support for Software Engineering: An Extended Abstract,” in Proceedings of the 9th International Conference on Software Engineering, Monterey, California, April, 1987, pp. 166-178.


[Boo84] Booch, Grady. Software Engineering with Ada, 2nd. ed. The Benjamin/Cummings Publishing Company, Inc, Menlo Park, California, 1984.


[Bro84] Brodie, M., B. Blaustein, U. Dayal, F. Manola, and A. Rosenthal, “CAD/CAM Database Management,” Database Engineering,  Volume 7, Number 2, 1984.


[Buc85] Buchmann, Alejandro P. and Concepcion Perez de Celis, “An Architecture and Data Model for CAD Databases,” in Proceedings of the 11th International Conference on Very Large Data Bases, Stockholm, 1985, pp. 105-114.


[Bur87] Burton, Bruce A., Rhonda Wienk Aragon, Stephen A. Bailey, Kenneth D. Koehler, and Lauren A. Mayes, “The Reusable Software Library,” in IEEE Software, July, 1987, pp. 25-33.


[Dit88] Dittrich, Klaus R. and Raymond A. Lorie, “Version Support for Engineering Database Systems,” in IEEE Transactions on Software Engineering, Vol. 14, No. 4, April 1988.


[Gut82] Guttman, Antonin and Michael Stonebraker, “Using a Relational Database Management System for Computer Aided Design Data,” in Database Engineering, Vol. 5, No. 2, June 1982, pp. 21-28.


[Har87a] Hardwick, Martin, George Samaras and David Spooner. “Evaluating Recursive Queries in CAD Using an Extended Projection Function,” in Proceedings of the 3rd International Conference on Data Engineering, Los Angeles, California, February, 1987, pp. 138-148.


[Har87b] Hardwick, Martin. “Why ROSE is Fast: Five Optimizations in the Design of an Experimental Database

System for CAD/CAM Applications,” in Proceedings of ACM SIGMOD, San Francisco, California, May, 1987, pp. 292-298.


[Has82] Haskin, Roger and Raymond Lorie, “Using a Relational Database System for Circuit Design,” in Database Engineering, Vol. 5, No. 2, June 1982, pp. 10-14.


[Hel87] Helier, Sandra, Umeshwar Dayal, Jack Orenstein, and Susan Radke-Sproull, “An Object-Oriented Approach to Data Management: Why Design Databases Need it,” in Proceedings of the 24th Design Automation Conference, Los Vegas, Nevada, 1987, pp. 335-340.


[Jac75] Jackson, Michael. Principles of Program Design. Academic Press, 1975.


[Kat85] Katz, R. H.  Information Management for Engineering Design.  Springer-Verlag, Berlin, FRG, 1985.


[Kim87] Kim, Won, Hong-Tai Chou and Jay Banerjee, “Operations and Implementation of Complex Objects,” in Proceedings of the3rd International Conference on Data Engineering, Los Angeles, California, 1987, pp. 626-633.


[Lor83] Lorie, Raymond and Wilfred Plouffe, “Complex Objects and Their Use in Design Transactions,” in Proceedings of Annual Meeting of Engineering Design Applications,  San Jose, California, May 1983, pp. 115-121.


[Mcl83] Mcleod, D, et al., “An Approach to Information Management for CAD/VLSI Applications,” in Proceedings of ACM Database Week, SIGMOD Conference, San Jose, California, May 1983.


[Mye78] Myers, Glenford J. Composite/Stuctured Design. Van Nostrand Reinhold Company, New York, 1978.


[Onu87] Onuegbe, Emmanuael O., “Database Management System Requirements for Software Engineering Environments,” in Proceedings of the 3rd International Conference on Data Engineering, Los Angeles, California, 1987, pp. 501-509.


[Pou88] Poulin, Jeffrey S., “Object-Oriented Database Support for Computer Aided Software Engineering,” Research Summary, Department of Computer Science, Rensselaer Polytechnic Institute, February, 1988.


[Pri87] Prieto-Diaz, Ruben, and Peter Freeman. “Classifying Software for Reusability,” in IEEE Software, Los Alomitos, California, January 1987, pp. 6-16.


[Pyl81] Pyle, T.C. The Ada Programming Language.  Prentice-Hall, Englewood Cliffs, New Jersey, 1981.


[Rae85]  Raeder, Georg. “A Survey of Current Graphical Programming Techniques,” in IEEE Computer, Vol. 18, Number 8, August 1985, pp. 11-24.


[Rei87] Reilly, Angela. “Roots of Reuse,” in IEEE Software, Los Alomitos, California, January 1987, pp. 4-5.


[Rom87] Roman, Gruia-Catalin, “Data Engineering in Software Development Environments,” in Proceedings of the 3rd International Conference on Data Engineering, Los Angeles, California, 1987, pp. 85-86.


[Rou83] Roussopoulos, Nick and Stephen Kelly, “A Relational Database to Support Graphical Design and Documentation,” in Proceedings of annual meeting of Engineering Design Applications, San Jose, California, May 1983, pp. 135-149.


[Sam87]  Samaras, George and Martin Hardwick, “User Manual for a VLSI CAD System Developed on ROSE,” Department of Computer Science Technical Report, Rensselaer Polytechnic Institute, Troy, New York, pp. 1-24.


[Sid80] Sidle, Thomas W., “Weaknesses of Commercial Data Base Management Systems in Engineering Applications,” in Proceedings of the 17th Design Automation Conference, New York, 1980, pp. 57-61.


[Smi77] Smith, J. and D. Smith, “Data Abstractions: Aggregation and Generalization,” ACM Transactions on Database Systems, Volume 3, Number 3, 1977, pp. 105-133.


[Weg84] Wegner, Peter, “Capital Intensive Software Technology,” IEEE Software, Vol. 1, Number 3, July 1984, pp. 7-45.


[Wir85] Wirth, Nicklaus.  Programming in Modula-2.  Springer-Verlag, New York, 1985.


 [Yau87] Yau, Stephen S. “Relationship Between Data Engineering and Software Engineering,” in Proceedings, 3rd IEEE International Conference on Data Engineering, Los Angeles, California, 1987, pp. 84.


[You75] Yourdon, Edward.  Techniques of Program Structure and Design.  Prentice-Hall, Inc., Englewood Cliffs, New Jersey, 1975.


[1] It is possible, for example in VHDL, to completely specify a VLSI component in text form.  However, CAD systems generally employ graphical tools for data input.