US 5,313,636 AGrant
Mosaic Objects and Method for Optimizing Object Representation Performance in an Object-Oriented Representation System
Issue Date:1994-05-17
•2 Claims
•3 Drawing Sheets
Abstract
Object-oriented software-based systems may be optimized by providing a data representation which is initially permitted to be completely dynamic such that at compile time representation need not be specified and thereafter the data structure and representation is progressively optimized. The invention is based on the recognition that the important data structures in object-oriented systems are not the objects but rather the slots, i.e., the attributes of objects.
Metadata
Assignee
- Intellicorp, Inc.
Inventors
- Robert E. Noble
- Gregory D. Clemenson
Application Information
Application Number:US 1037234
Filing Date:1993-08-05
Priority Date:1990-09-27
Art Unit:236
Classifications
IPC:
G06F 1200G06F 1500
Field of Search:
395364700DIG. 1
Patent Drawings (3 sheets)
Description
COPYRIGHT NOTICE
A portion of the disclosure of this patent document contains material which is subject to copyright protection. The copyright owner has no objection to the facsimile reproduction by anyone of the patent document or the patent disclosure as it appears in the Patent and Trademark Office patent file or records, but otherwise reserves all copyright rights whatsoever.
BACKGROUND OF THE INVENTION
This invention relates to the field generally known as artificial intelligence and more particularly to programming in frame-based systems or more generally to object-oriented systems. Specifically, this invention relates to techniques for software development of an object-oriented system.
In the past, optimization of software-based systems and specifically of object-oriented systems has been dependent on the developer. While no solution for general software-based systems is known, characteristics of object-oriented systems have features which, according to the invention disclosed hereinafter, permit developer-independent optimization.
Many known computer languages support static data representation. Fewer languages support dynamic data representation. In static data representation, the decision on the internal layout of the data structure and representation is made at compile time of the computer program. All static data representation systems require prior specification of individual language, of each individual field and of each data type, such as integer, floating point, aggregate or the like. This greatly reduces flexibility of programming and data handling and extends the development time, although it does provide the best possible performance, since data access characteristics ar all determined at compile time. An example of an object-oriented language with static data representation capability is the C++ language.
In a language capable of dynamic data representation, the decision on internal layout of data structure is deferred until run time of the program, which allows efficient applications development. The dynamic representation capability is an important distinction over static data representation. The proto-typical example of an object-oriented language capable of dynamic data representation is the SmallTalk language of Xerox Corporation. Because of frequent and extensive lookup at program run time, the typical system with a dynamic data structure yields significantly slower run-time performance than a comparable system having a static data structure. Although it allows for easy applications development, SmallTalk does not run many practical applications in a generally acceptable amount of time. As a consequence, programmers may use SmallTalk or other similar languages with dynamic data structures during system development but thereafter recode a production system in another language which supports static data representation.
It is therefore desireable to provide a programming environment which is easy and flexible for development of a software-based system and which does not require extensive reengineering in a different language for production of a software-based system.
The problem area herein addressed is not to be confused with the area of database management, where database management systems employ dynamic look-up capabilities as part of an indexing scheme on a revisable database. The problem addressed herein is how to minimize recoding of source code in a generalized object-oriented programming environment.
What is needed is a programming environment and methodology that allows delay until compile time of any specification about how static the structure is to be, and what is needed is a range of options between the formation of a dynamic data structure and the formation of a fully static data with commensurate improvements over a program with fully dynamic data representation.
Definitions
In order to more fully understand the invention, an understanding of certain terms is helpful. These definitions are not necessarily all-inclusive, and it should be understood that even within certain audiences of the relevant arts, common concepts employ different terminology and common terms have various meanings.
An "object" is a specific term of art, sometimes referred to as a "unit" in the literature, which is a data structure. It is analogous to a "row" in a relational database. Typically the term "object" refers to data structures which carry both attributes ("slots") and behaviors or methods.
An "attribute" is a specific term of art, sometimes referred to as a "slot" in the literature, which is a "property" of an "object," such as a method associated with an "object." A parallel concept is the "fields" of a data structure. The artificial intelligence community refers to the concept by both terms.
A "frame" is an "object" with some additional capabilities, such as secondary structure, or attributes of attributes, active capabilities called for example "monitors," "active values," or "triggers", and self-descriptive capabilities, such as an understanding what attributes it has, what its parents are or what its parents or children are.
A "frame-based system" is a system of frames, or in other words, a system which has a self-descriptive capability.
An "object system" is a "frame-based system" which in many cases does not have the ability to describe itself.
A "value" is a term of art defining what is "in" an attribute ("slot") at any particular moment.
A "selector" is a reference to an "attribute." It is a name given to an "attribute" and the way by which access to the "attribute" is specified.
Other definitions have been or will be introduced as is helpful for an understanding of the invention.
SUMMARY OF THE INVENTION
According to the invention, an object-oriented software-based system is optimized by providing a data representation which is initially permitted to be completely dynamic, that is, a decision on layout of internal data structure is deferred such that at compile time during development data need not be specified and thereafter the data structure and representation is progressively specified and optimized. The invention is based on the recognition that the important data structures in object-oriented systems as viewed from an internal perspective are not objects but rather slots, i.e., the attributes of objects. An object of the type according to the invention begins as a pointer to a memory location. Specifically, optimization is based on the understanding that it is attributes that carry the dynamic or static characteristics and that an object is only the aggregate or mosaic of the static and dynamic characteristics of the entity which it represents.
The invention allows flexibility in terms of algorithm selection. There is the recognition that specific declarations about specific attributes can be changed to affect flexibility and thereby change both the internal representation of the attributes as well as, through means of dispatch tables, the operations that apply to them. It may be necessary to have an applicable "get value" operation, but the actual code to be executed can be customized according to the declaration made of the subject attribute. In order to effect such a modification, at compile time the system chooses a specific form of the "get value" operation suited to the level of development or optimization desired. In a specific embodiment, a minimum version is an index from a reference point or pointer in computer memory; the system will know the relative location of the specified value and then access that value. In the more flexible cases, the system may actually search using a reference data structure itself to find the right access location of the value. The attribute is specified by a reference name, which provides abstraction; the programmer need not know specific access information, such as memory offsets and indices. The name is useful as an indirect index to the data structure.
The invention will be better understood by reference to the following detailed description in connection with the following drawings.
BRIEF DESCRIPTION OF THE DRAWINGS
FIG. 1 is a graphical representation of an object-oriented system illustrating a single object according to the invention.
FIG. 2 is a graphical representation of an attribute access descriptor in a portion of computer memory according to the invention.
FIG. 3 is a graphical r presentation of an attribute value array in a further portion of computer memory according to the invention.
FIG. 4 is a graphical representation of an object-oriented system employing mosaic objects according to the invention.
FIG. 5 is a flow chart of a general procedure for optimizing access to data representation in accordance with the invention.
DETAILED DESCRIPTION OF SPECIFIC EMBODIMENTS
In order to understand the invention, it is helpful to first outline the structure of objects and then show how to access the attributes in a fully dynamic mode. Thereafter, it is helpful to show how to access the attributes in a progressively more efficient and more static mode. With that one can appreciate the corresponding significant performance enhancement of the invention, which herein is referred to as a "mosaic object"-based system. It will be seen that a mosaic object is implemented as a pointer to a memory location in a software system on a computer.
It is useful to separate the way in which attribute values are accessed from the values themselves. Referring to FIG. 1, there is shown a graphical representation of a portion of memory 10 in an object-oriented system according to the invention. An object is represented in related storage locations of a memory 10 by an attribute access table 12 and an attribute value array 14. The attribute access table 12 need not be a sequential order of memory locations. The attribute value array 14 is as a rule unique to the object and as a consequence an attribute value array 14 is a set of sequential storage locations containing structure provided for each object as represented by the elements 32, 34, 36, 38, 40, 42, 44, 46.
The attribute access table 12 is a hash table which contains structure or attribute access descriptors 31, 33, 35, 37, 39, 41, 43, 45 which are pointers to locations of the elements 32, 34, 36, 38, 40, 42, 44, 46 in order to interpret the attribute value array 14. Referring to FIG. 2, there is shown a table representing the attribute access descriptors structure 31 of one element 16 of an attribute access table 12. The first word or field in the structure 31 is the attribute selector (AS) 48. Another significant field is the attribute offset (AO) 50. Other fields are Inheritance Information (IO) 52, Attribute Type (AT) 54, and flags (F1) 56, (F2) 58, (F3) 60. The attribute offset 50, for example, is a value indicating an offset from a base address in an attribute value array 14 where the desired value is to be found.
Referring to FIG. 3, there is shown a representation of two elements 32, 34 of the attribute value array 14. The attribute value array 14 is divided into types. Two significant types for this discussion are the locked type 62 and the unlocked type 64, as identified by Field 54 (FIG. 2). These types determine ancillary functionality for accessing the desired value or ways in which a representation of the structure which contains the value might be optimized or performance improvements can be obtained.
In a specific embodiment of the attribute value array, elements 62 and 64 have regions of memory of alternating attribute value 66 and associated secondary attributes 68 of the attribute. Thus attributes and secondary attributes may be associated closely.
It is now possible to describe a developmental system according to the invention and contrast it with a run-time system which has been fully optimized. It is significant that the access function or call is identical in all cases. Only the attribute declaration statement is changed. None of the functional source code is changed as among the different systems.
Referring to FIG. 4, there is shown an object 11 according to the invention labeled "JOHN" with an attribute access table 12 and an attribute value array 14 having three attribute values 63, 65, 67, and there is an object 13 labeled "PERSON" with an attribute access table 112 and an attribute value array 114 having three corresponding attribute values 163, 165, 167. One of the attribute values, herein element 65 in the first array 14 and element 165 in the second array 114, is "EYECOLOR". The generic call in the code used in a specific embodiment of the invention for objects 11 and 13 is:
The desired result is the value of John's eye color. In the fully dynamic developmental case, access to the value is by means of the attribute access table 12, which directs to an attribute access descriptor 31 (FIG. 2) with the access selector 48 of interest, which in this example is "EYECOLOR". Upon reference to the attribute access descriptor 31, the access offset 50 and the attribute type 54 are located. (The attribute type 54 is either locked or unlocked.) The start of the particular attribute types 62 or 64 of interest is found, and then it is added to the access offset (all of which are merely memory locations) which identifies the particular memory cell. In this case it is the element 65 in a portion of the attribute value array 14. All of these actions are considered necessary to perform a fully dynamic lookup of John's eye color, and it is considered a fairly significant computation.
By employing an attribute declaration statement according to the invention to redefine the getvalue statement, a more efficient static process is carried out without having to replace the source code. According to this action, the instruction fetches the mosaic object 11 labeled John, goes to the attribute value array 14, uses the compile time offset value (AO 50 in FIG. 2) and examines one word at element 65. This consumes significantly less computational overhead than the prior technique.
The organization of the system is such that a series of declarations may be provided to partition the possibilities between these two extremes. What has been recited are the most dynamic case and the most static case. There are also a number of incremental steps between those two extremes providing flexibility which are useful in the course of development and in certain application programs. When an application program is in development, the program is designed to follow the full path of access to the attribute value array to get at a memory cell. However, upon compilation of a fully developed program, the declarations may be set such that resultant translation specifies a minimal access path in order to go directly to the specific memory cell of the attribute value array.
The attribute declaration 52, 54, 56, 58, 60 are stored in the attribute access descriptor data structure 16 (FIG. 2 , and they can be modified at any time, from the beginning of development to the point of desired optimization.
A system according to the invention typically works as described in connection with FIG. 5. A typical declaration in the language of interest (C or a dialect) is as follows:
This source code statement creates a class in the repository "attributes" with no parents and three attributes, namely, "eyecolor", "weight", and "height", with an attribute declaration for "eyecolor" of "locked" attribute type, to which is added raw attribute data.
A further relationship is defined by the sentence.
Thus the compiler is advised "John" is an "instance" of the "class" "person". (#pragma is C construct.)
To fully understand the concept of declarations it is now necessary to define a further term, namely, "class". Referring to FIG. 4, a "class" is a specification for a type of object. In the case of a person, it specifies attributes which are appropriate to a person. Some of the secondary attributes of a person might also relate a person to other classes. A further term of art is an "instance". An "instance", rather than specifying a general concept such as person, specifies a specific concept, or a specific realization, for example, John. The "object" 11 "John" is a particular "instance" of the "class" "person", which in turn is represented by the "object" 13. There might be many instances of a person but there's only one John, for example.
Declarations are made at the class level. All such declarations are inherited in an object-representation system to the instance level. Referring to FIG. 3, in the unlocked type element 64 of the attribute value array, there is contained non-optimized attributes, and the locked type element 62 contains the optimized attributes. For example, by adding a declaration at the class level (object 13, FIG. 4), the instance would inherit whether or not a particular attribute (eye color) would be located in the unlocked or unoptimizable location or the locked or optimizable location. The attribute caller cannot tell the location of the particular attribute.
Significantly under "person", it is possible to add another level of indirection. It could have been stated that there is a subclass of person called "computer programmer" and then that "John" was a computer programmer and still advise the compiler that "John" is a "person" even though it is indirect. Because a particular attribute of interest is eye color, which is defined at the "person" level, not the "computer programmer" level, recoding is not required.
In summary, and referring to FIG. 5, the method for implementing objects of the software system comprises generating a data representation of objects of the object-oriented system wherein decision on layout of internal data structure is deferred such that data structure need not be specified at the time source code of the object-oriented software system is compiled into an executable form (Step A), and thereafter utilizing attribute declarations of the attributes to modify data representation of the object-oriented software system to specify whether access to individual attributes of the objects is static or dynamic (Step B). Specifically, the generating step comprises creating an attribute access table for selected objects (Step A1), the attribute access table including pointers to an attribute value array. The attribute value array contains the attribute declarations. The utilizing step then comprises replacing the attribute declarations in the attribute value array with new (or second) attribute declarations (Step B1). The new attribute declarations specify the layout of the internal data structure and can be either static or dynamic. To this end the attribute declarations are prestored in the attribute access descriptor data structure such that the attribute declarations can be modified without affecting source code. Specifically, the software system is operated with mixed dynamic and static representations (Step C); thereafter the representations can be changed to alter the mixture (Step D) and the software systems is operated again (Step C), producing desired results with appropriately altered performance characteristics.
The invention herein enables the construction of an object-oriented system having substantial flexibility. Specifically, the invention enables construction of an object-oriented system from objects of a different character than heretofore known. The object-oriented software system on a computer has objects which are specified by attributes having attribute declarations, wherein the objects comprise data representations in the form of attributes, and attribute declarations defining the attributes and modifying said data representations. More specifically, there may be within a single object first attribute declarations which define static attributes and second attribute declarations define dynamic attributes. The combination of static attributes and dynamic attributes in a single object is a unique and powerful aspect of the present invention.
The following provides guidance to permit a programmer to adapt a conventional compiler of an object-oriented language to employ mosiac objects. Specifically, the following example illustrates programmatic manipulation of an object which contains both static and dynamic attributes. The access of the "Proceeds" attribute is transformed into a static fetch, whereas the access of "Expense" remains a dynamic fetch. OBJECT--VALUE--ARRAY(job) extracts the value array from the object job, and [0] accesses the specific location in the value array where the actual data is stored.
I. EXAMPLE
Give the following source code:
In this example, JOB is a variable containing an object descended from the class "Job". "Job" is a class containing two attributes: "Proceeds" and "Expense". "Job" might be structured as follows:
This source code would be transformed using a simple preprocessor into:
A parser/preprocessor mechanism in a compiler is in the generic case as follows:
The following outlines a description of a Table containing Accessor vs. Attributes, where Table assumes the attribute "a" value is stored in value--array location "i":
The foregoing enables a programmer of a compiler to employ mosiac objects. The compiler acts upon the source code to change from a dynamic data description to a static data description.
In summary, advantages of this invention include time savings of about one to two orders of magnitude between the development version and a corresponding run-time version with identical source code, as well as flexibility in the construction of object-oriented and related systems.
Appendix A is a source code listing incorporating one embodiment of the invention. The source code listing is an illustrative example of the invention hereinabove described in the C programming language. Other embodiments are contemplated, such as using the LISP programming language. Certain aspects of a complete system in the C programming language have been omitted as they may relate to a specific implementation of an object-oriented system. The omitted details are a matter of engineering choice within the skill of the ordinary programmer in this art.
The invention has now been explained with reference to specific embodiments. Other embodiments will be apparent to those of ordinary skill in the art. It is therefore not intended that this invention be limited, except as indicated by the appended claims. ##SPC1##
Claims
What is claimed is:
1. In an object-oriented software system on a computer having a central processing unit and a memory means, said object-oriented system having data structures comprising objects, said objects representing real-world entities, said objects being an aggregation of attributes, wherein form of aggregation of said attributes is specified by attribute declarations, a method for generating a data representation of said objects so that source code is independent of value access and for accessing said data representation for use in implementing said objects of the software system, said method comprising the steps of: using said central processing unit to create an attribute value array for each one of said objects for storing in said memory means specific values of said attributes; using said central processing unit to store in said memory means said attribute declarations in an attribute access descriptor data structure having an attribute selector and an attribute offset, said attribute selector being the name of said attribute, said attribute offset being an index into said attribute value array; using said central processing unit to create an attribute access table for each said object and to store all attribute access tables in said memory means; thereafter using said central processing unit to place in said attribute access tables pointers to said attribute access descriptor data structures for each one of said objects, thereby to provide the data representation of said objects; and thereafter using said central processing unit to access said specific values of attributes in said attribute value array by specifying said attribute selector, using said pointers to find the attribute access descriptor data structure having said attribute selector, and applying said attribute offset as an index into said attribute value array, thereby to access the data representation of said objects.
2. The method of claim 1, wherein said attribute declarations include an element specifying access type, and wherein said access type element has the value "dynamic access" so as to permit generic access to said attribute values, further comprising the step of; thereafter using the central processing unit to replace said "dynamic access" value of said access type element of said attribute declarations in said attribute access descriptor data structure with a value "static access" in order to permit direct reference to said specific values in said attribute value array, said attribute declarations specifying layout of internal data structure.