REGISTER  |  SIGN IN
United States Patent6038378
Kita , ; et al.March 14, 2000
Title
Method and apparatus for testing implementations of software specifications
Abstract
A method and apparatus for automatically generating validation tests for implementations of a program specification for an operating system, software application or a machine, where the program specification is expressed at least in part in terms of data structures and relationships. The method is carried out by a computer. The program specification is expressed in an interface specification language which is automatically parsed, and is then transformed into an extended finite state machine (EFSM) or multiple-EFSM architecture internally represented in the computer, the EFSM including objects representing states and transitions between those states representing executable functions, annotated to the states. The annotations may represent predicates, test data, value assignments, branch conditions, etc. The EFSM or architecture is traversed by a path traversal procedure, either exhaustively or in part, thereby producing path files, one for each path taken. Each path file is linked to a program shell, which is automatically generated for the specification, resulting in one independent validation test for each path file. Each validation test includes a call to the implementation of the program specification, and presents that implementation with a test vector representing a given path through the model. Failure and success responses are produced, indicating whether the implementation behaved as expected. Multiple validation tests may be linked or combined in a variety of ways to form a superstructure (architecture) of validation tests for testing of many routines in conjunction with one another, such as for testing all the routines specified for an operating system at the same time.
Inventors:Kita; Ronald Allen (Hollis, NH), Trumpler; Mark Edward  (Lexington, MA), Elkind; Lois Scirocco  (Hollis, NH)
Assignee:Digital Esquipment Corporation (Maynard, MA)
Appl. No.:677774
Filed:July 10, 1996
Current U.S. Class:714/38 
Current International Class:G06F 11/36 (20060101)
Field of Search:395/183.14,183.15,704
U.S. Patent Documents
4696003September 1987Kerr et al.
4991176February 1991Dahbura et al.
5038307August 1991Krishnakumar et al.
5121497June 1992Kerr et al.
5164912November 1992Osborne et al.
5218605June 1993Low et al.
5388258February 1995Larrson et al.
5481717January 1996Gaboury
5513316April 1996Rodrigues et al.
5572727November 1996Larrson et al.
5657438August 1997Wygodny et al.
Other References
Wang, Generating Test Cases For EFSM With Given Fault Models 1993. .
Wang, Axiomatic Test Sequence Generation For Extended Finite State Machines, 1992. .
Chanson, A Unified Approach To Protocol Test Sequence Generation, 1993..~
Primary Examiner: Oberley; Alvin E.
Assistant Examiner: Courtenay, III; St. John
Parent Case Text


This application is a continuation of application Ser. No. 08/545,142, filed Oct. 16, 1995 which is a continuation of application Ser. No. 08/408,090, filed Mar. 21, 1995 which is a continuation of application Ser. No. 08/100,006 filed Jul. 29, 1993, all now abandoned.
Claims

What is claimed is:
1. A method for generating tests for an implementation of a program to insure that the implementation conforms to a natural language and computer readable program specification that specifies the functions the program will have to fulfill, defines the types and uses of variables, defines parameters and their relationships, and other system requirements, the method being carried out in a processor and including the steps of:
(1) accepting as an input the natural language and computer readable program specification, wherein data relationships are expressed in the program specification with predefined syntax and semantics;
(2) parsing the program specification in accordance with said predefined syntax and semantics to generate data structures and relationships between the data structures which represent parameters and functionality expressed in the program specification;
(3) modeling the program specification by transforming the parsed program specification according to said semantics into a finite state machine (FSM), which represents data relationships in the program specification and which includes at least a starting state, an ending state, a plurality of intermediate states, and a set of transitions defining conditions under which the FSM is traversed from a given state to a subsequent state;
(4) generating a test specification comprised of a set of computer readable path files, wherein the path files are generated by traversing paths in the FSM, each path file defining a path beginning at the starting state and ending at the ending state and including a unique traversal of said set of transitions and intermediate states and defining data and variables for one test;
(5) generating executable program code as an implementation of the program specification and derived from the FSM, wherein the program code is in the form of a program shell; and
(6) generating executable test programs for testing the implementation of the program specification, each test program being generated by merging into a copy of the program shell one of said path files.

2. The method of claim 1, wherein the FSM is an extended finite state machine (EFSM), and step 4 includes the step of maintaining context during the traversal of the EFSM.

3. The method of claim 1, wherein the traversal of step 4 is carried out by traversing all paths through the FSM.

4. The method of claim 1, further including, before step 4, a step of generating constraints for confining the traversal of step 4 to a subset of states of the FSM.

5. The method of claim 1, wherein step 3 includes the steps of:
generating at least one value partition for each of at least one variable value to be tested, each said partition corresponding to a value range to be tested for said at least one variable; and
generating a transition for each said partition, said transition including test information for assigning to said at least one variable a value corresponding to said partition.

6. The method of claim 1, wherein at least one said transition includes a predicate correlated with it, and including, as part of step 4, the step of determining whether said predicate is true, and if so, taking the transition.

7. The method of claim 1, wherein at least one said transition includes an action correlated with it, and including, as part of step 4, the step of executing the action.

8. The method of claim 1, wherein at least one said transition includes a test data annotation correlated with it, and including, as part of step 4, the step of generating a test routine for testing a function of the implementation.

9. The method of claim 2, further including the steps of:
combining the EFSM produced in step 3 with a second EFSM for generating a multiple-model architecture; and
carrying out step 4 on the multiple-model architecture for generating path files for use in steps 5-6 to generate test programs for the multiple-model architecture.

10. The method of claim 5, wherein said value partition includes a plurality of subsidiary partitions, and wherein step 3 includes the steps of:
generating a branch transition for each said subsidiary partition, said branch transition including test information for assigning to said at least one variable a value corresponding to said subsidiary partition.

11. The method of claim 6, wherein at least one said transition includes a predicate correlated with it, and including, as part of step 4, the step of generating a first branch transition and a second branch transition for said predicate, said first branch transition representing a test case where said predicate is true, and said second branch transition representing a case where said predicate is false.

12. A method for testing an implementation of a program to insure that the implementation conforms to a natural language and computer readable program specification that specifies the functions the program will have to fulfill, defines the types and uses of variables, defines parameters and their relationships, and other system requirements, the method being carried out in a processor and including the steps of:
(1) accepting as a input the natural language and computer readable program specification, wherein data relationships are expressed in the program specification with predefined syntax and semantics;
(2) parsing the program specification in accordance with said predefined syntax and semantics to generate data structures and relationships between the data structures which represent parameters and functionality expressed in the program specification;
(3) modeling the program specification by transforming the parsed program specification according to said semantics into a finite state machine (FSM), which represents data relationships in the program specification and which includes at least a starting state, an ending state, a plurality of intermediate states, and a set of transitions defining conditions under which the FSM is traversed from a given state to a subsequent state;
(4) generating a test specification comprised of a set of computer readable path files, wherein the path file are generated by traversing paths in the FSM, each path file defining a path beginning at the starting state and ending at the ending state and including a unique traversal of said set of transitions and intermediate states and defining data and variables for one test;
(5) generating executable program code as an implementation of the program specification and derived from the FSM, wherein the program code is in the form of a program shell;
(6) generating executable test programs for testing the implementation of the program specification, each test program being generated by merging into a copy of the program shell one of said path files;
(7) executing each said test program;
(8) determining whether outputs of each said test program are consistent with the program specification; and
(9) for each output which is not consistent with the program specification, generating a diagnostic indicating failure of the implementation.

13. The method of claim 12, wherein the FSM is an extended finite state machine (EFSM), and step 4 includes the step of maintaining context during the traversal of the EFSM.

14. The method of claim 12, wherein the traversal of step 4 is carried out by traversing all paths through the FSM.

15. The method of claim 12, further including, before step 4, a step of generating constraints for confining the traversal of step 4 to a subset of states of the FSM.

16. The method of claim 12, wherein step 3 includes the step of:
generating at least one value partition for each of at least one variable value to be tested, each said partition corresponding to a value range to be tested for said at least one variable; and
generating a transition for each said partition, said transition including test information for assigning to said at least one variable a value corresponding to said partition.

17. The method of claim 12, wherein at least one said transition includes an action correlated with it, and including, as part of step 4, the step of executing the action.

18. The method of claim 12, wherein at least one said transition includes a test data annotation correlated with it, and including, as part of step 4, the step of generating a test routine for testing a function of the implementation.

19. The method of claim 13, further including the steps of:
combining the EFSM produced in step 3 with a second EFSM for generating a multiple-model architecture; and
carrying out step 4 on the multiple-model architecture, for generating path files for use in steps 5-6 to generate test programs for the multiple-model architecture.

20. The method of claim 16, wherein said value partition includes a plurality of subsidiary partitions, and wherein step 3 includes the steps of:
generating a branch transition for each said subsidiary partition, said branch transition including test information for assigning to said at least one variable a value corresponding to said subsidiary partition.

21. The method of claim 16, wherein at least one said transition includes a predicate correlated with it, and including, as part of step 4, the step of determining whether said predicate is true, and if so, taking the transition.

22. The method of claim 21, wherein at least one said transition includes a predicate correlated with it, and including, as part of step 4, the step of generating a first branch transition and a second branch transition for said predicate, said first branch transition representing a test case where said predicate is true, and said second branch transition representing a case where said predicate is false.

23. Apparatus for generating tests for an implementation of a program specification to insure that the implementation conforms to a natural language and computer readable program specification that specifies the functions the program will have to fulfill, defines the types and uses of variables, defines parameters and their relationships, and other system requirements, the apparatus operating in a processor and comprising:
means for accepting as an input the natural language and computer readable program specification, wherein data relationships are expressed in the program specification with predefined syntax and semantics;
a parser which parses the modeled program specification in accordance with said predefined syntax and semantics to generate data structures and relationships between the data structures which represent parameters and functionality expressed in the program specification;
a state machine generator which transforms the parsed program specification according to said semantics to generate a finite state machine (FSM) model of the program specification, which represents data relationships in the program specification and includes at least a starting state, an ending state, a plurality of intermediate states, and a set of transitions defining conditions under which the FSM is traversed from a given state to a subsequent state;
a path file generator which traverses paths in the FSM for generating a set of path files forming a test specification, each path file defining a path beginning at the starting state and ending at the ending state and including a unique traversal of said set of transitions and intermediate states and defining data and variables for one test;
a program shell generator for generating executable program code as an implementation of the program specification, wherein the implementation is in the form of a program shell and is the implementation to be tested; and
a test program generator which generates executable test programs for testing the implementation by merging one of said path files into a copy of the program shell.

24. The apparatus of claim 23, wherein the path file generator comprises means for traversing all paths through the FSM.

25. The apparatus of claim 23, further comprising a multiple model generator which generates a second FSM and combines it with the FSM to form a combined FSM so that the path file generator traverses paths in the combined FSM to generate path files.

26. The apparatus of claim 24 wherein the path file generator comprises means for generating constraints for confining the traversal of the FSM to a subset of states of the FSM.

27. A computer program product for generating tests for an implementation of a program to insure that the implementation conforms to a natural language and computer readable program specification that specifies the functions the program will have to fulfill, defines the types and uses of variables, defines parameters and their relationships, and other system requirements, the computer program product comprising a computer usable medium having computer readable program code thereon including:
program code for accepting as an input a computer readable and natural language program specification, which expresses data relationships in predefined syntax and semantics;
program code for parsing the program specification in accordance with said predefined syntax and semantics to generate data structures and relationships between the data structures which represent parameters and functionality expressed in the program specification;
program code for modeling the program specification by transforming the parsed program specification according to said semantics into a finite state machine (FSM), which represents data relationships in the program specification and which includes at least a starting state, an ending state, a plurality of intermediate states, and a set of transitions defining conditions under which the FSM is traversed from a given state to a subsequent state;
program code for generating a test specification comprised of a set of computer readable path files, wherein the path files are generated by traversing paths in the FSM, each path file defining a path beginning at the starting state and ending at the ending state and including a unique traversal of said set of transitions and intermediate states and defining data and variables for one test;
program code for generating executable program code as an implementation of the program specification and derived from the FSM, wherein the program code is in the form of a program shell; and
program code which generates executable test programs for testing the implementation of the program specification, each test program being generated by merging one of said path files into a copy of the program shell.

28. The computer program product of claim 27, wherein the path file generator comprises means for traversing all paths through the FSM.

29. The computer program product of claim 27, further comprising a multiple model generator which generates a second FSM and combines it with the FSM to form a combined FSM so that the path file generator traverses paths in the combined FSM to generate path files.

30. The computer program product of claim 28 wherein the path file generator comprises means for generating constraints for confining the traversal of the FSM to a subset of states of the FSM.
Description


FIELD OF THE INVENTION

This invention is directed to an apparatus and method for automatic generation of validation tests for an implementation of a specification expressed as data relationships.

BACKGROUND OF THE INVENTION

In the course of designing a new computer application or operating system, typically one first writes a program specification for the application. Authoring of the program specification includes specifying the functions the application or system will have to fulfill, defining the types and uses of variables, defining parameters and their relationships, and the other requirements which go into a conventional program specification.

Such a program specification may be quite extensive. For instance, for the widespread version of UNIX known as POSIX, the specification for the system application program interface (API) for the C programming language runs over 350 pages in the book entitled ISO/IEC 9945-1 (IEEE Std 1003.1)/Information Technology-Portable Operating System Interface (POSIX)-Part I (First Edition 1990-12-07), which is incorporated herein by reference.

Software application developers, system engineers, end users operating in a UNIX environment and others refer to such a specification whenever they author applications based upon it. Typically, such applications are tested during and after their creation, for internal consistency and to ensure that they perform as intended. Such testing will be referred to herein as validation testing.

In the case of POSIX, for example, the above-mentioned IEEE reference provides the complete English-language specification of POSIX for the C programming language. Whenever a programmer writes an application in C to run on the POSIX operating system, this reference must be consulted to ensure that all of the standards are being complied with.

It is a very difficult and involved matter to ensure that an implementation of a long program specification is consistent with the specification of POSIX, and yet this is crucial for the proper operation of the implementation and anything based upon it. Testing and verifying the implementation is particularly important when the program specification is for a computer language such as C or an operating system such as UNIX. If, for instance, an operating system based upon POSIX includes errors, those errors will create bugs in all programs that run on the erroneous operating system. Thus, before the implementation of POSIX is ever released, it should be fully tested.

Such an implementation will, for the POSIX 1003.1 specification mentioned above, take the form of a program written in the C language. Thus, the validation tests to be run must verify the correctness of this C implementation of the program specification.

The concept of specification testing is represented, for example, by STATEMATE, which provides a system by which a programmer can model a specification and test it. See The STATEMATE Approach to Complex Systems, by i-Logix of Burlington, Mass. (1990); and D. Harel et al., STATEMATE: A Working Environment for the Development of Complex Reactive Systems, IEEE Transactions on Software Engineering, Vol. 16, No. (April 1990). Testing the implementation of a specification (i.e. an application based on the specification), based upon traversal of a graph model based upon the implementation, is discussed in W. H. Jessop et al., Atlas-An Automated Software Testing System, Bell Laboratories, Inc., 13th IEEE Computer Society International Conference, pp. 337-340 (1976) and 2nd International Conference on Software Engineering, pp. 629-635 (1976). Various test selection methods for systems expressed as finite state machines are discussed in S. Fujiwara et al., Test Selection Based on Finite State Models, IEEE Transactions on Software Engineering, Vol. 17, No. 6 (June 1991) and W. H. Jessop et al. Each of the foregoing articles is incorporated herein by reference.

With the STATEMATE system, the user must convert the English-language (as opposed to computer code) program specification into a machine-readable coding. Neither STATEMATE nor other prior systems automatically generate validation tests for the actual implementation of the program specification, i.e. for applications with which they are designed to comply.

To generate validation tests of an application, the systems described in the Jessop and Fujiwara articles rely upon an expression of the program specification as a finite state machine (FSM). Essentially any program specification or software application can theoretically be expressed as an FSM, whether the modeled entity is state-rich or expressed as data relationships. An "extended" finite state machine adds context to the transitions (i.e. a record of the history of traversal of the transitions). See the Introduction below for a formal definition of EFSM.

FSMs and EFSMs are particularly suited to state-rich problems. A program is state-rich if it controls a system where the state of the system at any given time is important, such as a program to control an automatic teller machine. A program specification for an operating system interface or a program routine, on the other hand, is typically data-rich; that is, the specification is described largely in terms of detailed data structures and their relationships.

Although one can, if necessary, generate an FSM to represent a data-rich program specification as states with transitions from one state to another, such a model does not serve well as a human-readable program specification.

In addition, hand-generated graph models are subject to the limitations of the programmer who creates them, which can lead to testing inefficiencies or inaccuracies, particularly in the case of a data-rich setting. Data-rich problems are particularly unsuited to being expressed as graph models such as FSMs, since they are by definition descriptions of data relationships, not of states and transitions. Thus, although a data-rich program specification can be expressed as an FSM, it is very cumbersome to do so, and is not an intuitive way to approach the problem. Moreover, while such a model would be computer-readable, it could not easily be interpreted by humans if complicated data structures are involved. For this reason, FSM-based Systems (such as Atlas) are not, in general, useful for data-rich problems. The cited articles do not teach any mechanism for automatically generating validation tests using the program specification itself as a starting point.

A system is needed which bypasses the necessity for human intervention in the creation of models to generate testing software against a program specification, and which at the same time an produce a set of tests which are accurate and which can be tailored to any degree of comprehensiveness, from testing all possible paths through the model to testing only few. In particular, a system is needed which allows for modeling of a program specification in a manner which is suitable for humans, but which leads directly to an expression of the model which can be used by a computer to generate validation tests for the implementation of the software.

SUMMARY OF THE INVENTION

The present invention comprises a system and method for automatically converting a program specification, particularly one expressed as data relationships, into an EFSM or a multiple-EFSM architecture, as required, and for automatically generating validation tests for implementations of that program specification. The program specification may be for an operating system, a software application or routine, or other systems or apparatus definable in program specifications, especially those expressed as data structures. The invention thus provides a method and apparatus for automatically generating validity tests by traversing multiple-model EFSM architectures.

The validation tests are generated by traversing valid paths through the EFSM (or multiple-EFSM architecture) and coupling each such path with the source code of the implementation in a program shell. This creates a multitude of test programs, each of which is executed to determine whether the implementation behaves as the program specification requires for a particular path through the architecture. Error diagnostics are output for each case where the implementation is inconsistent with the program specification. If execution of all the test programs produces no errors, then the implementation is determined to comply with the program specification.

The program specification is expressed in a computer language, which can be compiled by a processor to generate tests for the program specification, by describing data relationships in the input and output to an application programming interface (API). For this purpose, applicant has created a new language called Interface Specification Language (ISL). The program specification is thus expressed as ISL source code. When this source code is compiled, the output includes:

(1) an extended finite state machine (EFSM) or a multiple-EFSM architecture, which is a model of the program specification;

(2) user documentation for the specified interface; and

(3) validation code, i.e. callable routines allowing compile and run-time validation of parameters to be used before service is provided.

An EFSM or multiple-EFSM architecture generated by compiling the ISL source contains states, transitions and outputs, with context, event and predicate definitions. These concepts are conventional in the definition of an EFSM. See "ESTELLE--A FDT based on an extended state transition model", ISO TC 97/SC21 DP 9074, September 1985. Each transition in the model contains test information describing the result of satisfying a unique data relationship, or providing flow control to allow only valid combinations of data relationships to be selected. Test information is written in the target test language (C, Ada, FORTRAN, etc.) by the ISL compiler.

The EFSMs are, in the preferred embodiment, constructed in applicants Path Flow Language (PFL). The construction and use of EFSMs are discussed in detail in applicant's U.S. patents and co-pending patent application, as follows:

"Method and Apparatus for Testing Structures Expressed as Finite State Machines", U.S. Pat. No. 5,394,347;

"Method and Apparatus for Transforming a Specification into an Extended Finite State Machine" (the "ISL Transformation" application), Ser. No. 08/403,547; and

"Method and Apparatus for Schematic Routing" (STE Routing), U.S. Pat. No. 5,408,597.

Each of the foregoing patent applications was filed with the United States Patent Office on the same date as the present application, and each of the patents and application cited above is incorporated herein by reference.

The output models (i.e. EFSMs or multiple-EFSM architectures) allow one of several path generation methods to be used, filling the requirements of exhaustive testing, thorough testing and identification testing as defined in IEEE Std.
1003.3-1991/IEEE Standard for Information Technology-Test Methods for Measuring Conformance to POSIX (1991). See also Chow, T., Testing Software Design Modeled by Finite-State Machines, IEEE Transactions on Software Engineering, Vol. 4, No. 3, May 1978. In ISL parlance, exhaustive testing may be referred to as "allpaths", and thorough testing or identification testing as "edge cover" or "transition cover".

A given path through an EFSM represents a unique combination of data relationships, and traversing the path produces values for parameters and executable code that correspond to those data relationships, in addition to an API call that, when invoked, traverses the combination of data relationships. The path traversal also produces conditions which will appear in the program shell that is used to test the implementation of the program specification.

The process of compiling ISL source into EFSMs is an equivalence transformation--that is, each data relationship or other construct in the original ISL source (and hence in the original specification) is represented by a corresponding structure (one or more objects) in the resultant EFSM or architecture. Thus, the error detection capabilities of the tests realizable by the EFSM definitions apply directly to the ISL program specification. In practice, this means that the transition coverage procedure ensures that each data relationship appears in at least one test path, and that all possible output errors from the original model can be detected. (The Chow paper cited above discusses output errors). For an API call, the impact of every data relationship can accordingly be verified. If the "allpaths" coverage scheme is used, the error detection output includes both output errors and sequence errors. Thus, all combinations of all data relationships can be verified.

Since specifications (such as that for POSIX mentioned above) can be quite complex, typically many EFSMs will be generated, each of which corresponds to some portion of the specification. Thus, an EFSM would be generated for the portion of the specification describing the open (open file) function, another for the close (close file) function, another for the mkdir (create a directory) function, another for chmod (change mode of the file), and so on for each of the scores of POSIX functions. All of these EFSMs are integrated into a single architecture which produces executable test programs for testing implementation of POSIX. The term "architecture" refers to a multiple-EFSM state model; however, in the following description, when building or traversing an architecture is referred to, it should be understood that the term may be taking broadly in context to include a single EFSM (i.e., a single-EFSM architecture).

Since some data structure descriptions represent potentially infinite domains, such as recursively defined data structures, the model created by ISL is annotated to constrain these conditions. This may, at the option of the user, be accomplished by extending the ISL source to allow specification of such execution control (or execution constraint) information, or by directly editing the output model using a suitable model editor. Such an editor has been developed for the present invention by the applicant; it is called the State Transition Editor (STE), and is fully described in the aforementioned "STE Testing" patent.

Test data generated by ISL is translated (via traversals of the output model paths) into a block of code that sets up and executes API calls. In order to achieve a fully executable test, this code block is inserted into an automatically generated program (or "program shell") by integrating it into a larger state model describing relationships between API calls. It is then executed in a suitable test environment. The program and the test environment contain the objects needed for the tests.

ISL source comprises statements that describe data relationships. These statements are unambiguous and, when compiled, can be directly converted to nroff source. "Nroff" is a conventional documentation language used to produce UNIX manual pages. See M. E. Lesk, "Typing Documents on the Unix System: Using the -ms Macros with Troff and Nroff", Bell Labs (Murray Hill, N.J.), incorporated herein by reference.

ISL source can also be compiled to generate a routine in a procedural language (such as C) that takes a specific list of parameters from a real call and returns a positive response if the parameters are consistent (i.e., taken together they represent valid data relationships), and a negative response otherwise.

In this case, the ISL compiler produces a program that executes a set of control statements against the supplied parameters. If any invalid relationships are found, a failure return is made. If the data values are compile-time constants, the routine can execute at compile time to screen out incorrect calls before code is produced. Otherwise, it is called when the actual request for serviced is made. This would occur after a client's request for service and servicing of the request to ensure that the server does not act on bad parameter combinations. (This is independent of bad parameter data for the particular problem). For example, a banking transaction system would always fail a request wherein the customer's bank identification code was of improper syntax; but even if the code is syntactically correct, the request will still be failed if the code does not match the bank's records. Validation code as in the present invention is automatically generated to detect the former (syntax) error, while the latter error must be detected in the server itself, i.e. cannot be detected other than in an actual usage situation for the application.

BRIEF DESCRIPTION OF THE DRAWINGS

The above and further advantages of the invention may be better understood by referring to the following description in conjunction with the accompanying drawings, described below.

FIG. 1 is a block diagram of a computer system implementing the invention.

FIG. 2 is a state diagram of a simple example of an extended finite state machine (EFSM).

FIG. 3 is a flow chart of the overall method of the invention.

FIG. 4 is a functional block diagram of the transformation procedure of the invention, showing the relationships among subsidiary method modules of the transformation procedure.

FIG. 5 is a state diagram of an EFSM generated by the procedure of FIG. 3 for a sample program as specified in Part E.

FIGS. 6A-6B together form a state diagram of an EFSM generated by the procedure of FIG. 3 for the POSIX "chmod" (change mode) function.

FIG. 7 is a state diagram of an EFSM showing a multiple-EFSM architecture (EFSMA), including the POSIX "chmod" and "Chown" routines as two submodel EFSMs.

FIG. 8 is a flow chart illustrating a method for generating a multiple-EFSM architecture.

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENT

The Description of the Preferred Embodiments includes the following sections:

Introduction: Formal Definition of an Extended Finite State Machine

I. The Apparatus of the Invention

II. The Method of the Invention: Generating and Traversing an EFSM

Method Stage 1: Create specification source code.

Method Stage 2: Parse the specification source.

The reference numeral strings appearing in the parsed listings

Method Stage 3: Transform the parsed source.

a. Components of an EFSM produced by the transformation procedure

b. Elements of a transition

c. Events

d. Reading the transformation listings (Parts I and S) for the examples

e. The functions of modules M1-M16

(Description of each of the modules M1-M16)

Method Stage 4: Traverse paths through the EFSM.

Method Stage 5: Post-process the path files.

Method Stage 6: Create program shells

Method Stage 7: Couple test programs with program shells to create validation tests

The routine "sample"

Method Stage 8: Execute

III. The Interface Specification Language (ISL)

A. Property Section(s)

B. Constant Section(s)

C. Type Section(s)

D. Routine Section

1. Syntax

2. Parameters

a. I/O status

b. Optional/Required status

c. Parameter (or value) constraints

3. Values

IV. Output

A. Graphical representation of EFSMs: Path Flow Language

B. Path Files: the validation tests

C. Documentation

V. Testing an Implementation of an Example Routine: "Sample"

(Description of the eight method stages for Sample)

VI. Another example: "chmod".

A. The chmod specification source code

Background on the POSIX (UNIX) "chmod" routine

B. The parsing of the source listing for chmod

C. Transformation of the parsed chmod listing

D. Traversal of the chmod EFSM and production of validation tests

VII. Method for Generating a Multiple-EFSM Architecture

The description of the invention is made in light of the figures and the following attached parts:

______________________________________ Part Contents ______________________________________ A Keywords for ISL B Backus-Naur Format (BNF) for ISL C Transformation Method Description for ISL D ISL User's Guide E ISL specification source listing for "Sample" routine F An implementation of the "Sample" specification G A faulty implementation of the "Sample" specification H Parsed ISL listing for "Sample" I Description of transformation of Part H J Path listing for first traversal of "Sample" EFSM K Path listing for second traversal of "Sample" EFSM L Path listing for third traversal of "Sample" EFSM M First portion of program shell for Parts J, K, L N Last portion of program shell for Parts J, K, L O Procedure for assembling and executing validation tests P POSIX specification, chmod excerpt Q ISL source listing for chmod routine R Parsed ISL listing for chmod routine S Description of transformation of Part R T Test program for chmod routine ______________________________________

Parts A-D specify the syntax and the semantics, including legal and illegal value and structure combinations, for the Interface Specification Language (ISL) that applicant has developed particularly for the system of the present invention. Part D is a user's guide to ISL.

Parts E-O demonstrate the transformation of the specification for a simple function (appearing in Part E) into an EFSM (illustrated in FIG. 5), and the generation of validation tests based upon path files derived from that EFSM.

Parts P-T demonstrate the transformation of the specification for a more complex function, namely "chmod" (change mode) for POSIX, into an architecture for path generation to create tests.

These Parts are discussed in detail below.

Introduction: Formal Definition of an Extended Finite State Machine

A finite state machine (FSM) includes a set of states and a set of transitions connecting those states, where the transitions represent actions, events, predicates, or test information between executed in traversing from one state to the next. An extended finite state machine (EFSM) brings the concept of context to the FSM. An EFSM includes the following elements:

1. A set of states S, each state being of a class, the class being "entry"

(=starting), "exit" (=ending), or "other". An "other" state may be an intermediate state or a submodel, both of which are discussed below.

2. A set of events (inputs, stimuli) E, specified by e.sub.ij, where i represents a from-state and j represents a to-state for a transition t.sub.ij (see below) from state i to state j, and e.sub.ij is an event along the transition t.sub.ij.

3. A set of outputs O.

4. A set of variables V.

5. A set of contexts C, such that

i.e. such that the context c; for a given path up to state i maps a set of variables v (i.e. a variable stack or ordered set of variables) onto an ordered set of variable values n.sub.1, n.sub.2, n.sub.3, etc. Thus, at each state, for a given path traversed through the EFSM up to that state, a context c.sub.i is defined that maps all variables onto their respective values due to the particular path traversed. The context reflects the history of the machine execution up to a given point in that execution, including the set of states through which the machine has traveled, the set of transitions the machine has taken, and the set of events that have occurred during the execution.

6. A set of predicates P, such that

where c.sub.i is a member of the set C, i.e. is a context as defined in 5 above.

7. A set of transitions (mappings) T, such that if and only if

then each transition t.sub.ij consists of:

(a) next output: {s.sub.i, c.sub.i, e.sub.ij } ->o.sub.ij

(b) next state: {s.sub.i, c.sub.i, e.sub.ij } ->s.sub.j

(c) next context: {s.sub.i, c.sub.i, e.sub.ij } ->c.sub.j

These are three mappings of the transition t.sub.ij, which are performed only if the predicate p.sub.ij is true for the given context c.sub.i at the from-state s.sub.i. The transitions describe under what circumstances the machine changes state, and specifies the state to which is changes. Mapping (a) indicates that the from-state s.sub.i, the context c.sub.i (which has been generated over the path taken up to this from-state), and an event e.sub.ij along the transition to the next state (which will be s.sub.i), map onto an output o.sub.ij. The same three elements also map onto the next state s.sub.j, as well as onto the new context c.sub.j, at the new state, which is derived from the previous context c.sub.i, updated according to any functions performed in taking the transition t.sub.ij to the state s.sub.j.

Not all aspects of the use of context are presented in the present application, but the concept is fully discussed in applicant's aforementioned PFL patent. The present invention is, in any case, applicable to EFSMs as defined above.

Apparatus of the Invention

The invention is preferably implemented in a computer system such as system 10 depicted in FIG. 1. The system 10 includes a central processing unit (CPU) 15 including a microprocessor 20 coupled to main memory 25 and program memory (such as ROM or RAM) 30. Computer input and output may be displayed on a monitor 35 or other output device, such as a printer (not shown). User input is provided via a keyboard 40, a mouse or trackball 45, and other standard input devices.

The system 10 is conventional, and may be a microprocessor system, a workstation or networked terminal system, or any of a variety of other apparatus providing computing and input/output capabilities.

All of the method steps described below, except for the original creation of the program specification whose implementations are to be tested, are carried out in the microprocessor 20. The implementations themselves are, of course, applications designed by programmers. Thus, the apparatus of FIG. 1 carries out all the steps of the invention automatically. Given the following detailed description of the method of the invention, it will be a routine matter for a programmer to produce code to carry out all of the described steps.

II. The Method of the Invention: Generating and Traversing an EFSM

As illustrated in the flow chart 50 of FIG. 3, there are eight basic stages involved in testing a given implementation of a program specification. These are discussed in general terms below, followed by detailed descriptions of the application of the method to two examples (in Sections III and IV below).

Stage 1 (box 55). Create the program specification source code, using the BNF and the Transformation Method Description for syntax and semantics. This program specification source code is an ISL file compilable by a parser (see stage 2). It may be derived from a plain-English program specification or may be coded as an initial matter in ISL.

Stage 2 (box 60). Parse the program specification source code, using a parser created by a conventional automatic compiler as applied to ISL.

Stage 3 (box 65). Transform the parsed program specification, using a program meeting the description of the Transformation Method (Part C), thus generating an EFSM representing the original program specification.

Stage 4 (box 70). Traverse paths through the EFSM (or architecture) thus generating individual validation tests of the original program specification. The EFSM may be comprehensively traversed in an "exhaustive testing" (allpaths) mode, or selected paths may be traversed in either a "thorough testing" or an "identification testing" (edge cover) mode, by using execution constraints. Validation tests are flagged. The output is a path file which is in the language upon which the specification was based (such as C).

Stage 5 (box 75). Post-process the path files to place validation tests at the ends of the respective files.

Stage 6 (box 80). Create a program shell for the implementation of the program specification.

Stage 7 (box 85). Add the program shell to each path file, thus generating a test program for each path file.

Stage 8 (box 90). Execute these test programs, and generate error outputs for those conditions under which the implementation is inconsistent with the program specification.

Method Stage 1: Create specification source code.

The program specification source code created in stage 1 is important to the automatic generation of an EFSM representing the program specification, and hence of tests for the implementation thereof. Thus, the ISL (discussed in Section III below) is an important element of the invention; however, alternative languages may be developed and used for this purpose, as long as they can express a program specification and can be parsed as in stage 2 to produce data structures and relationships which can then be transformed into an EFSM according to stage 3.

The specific requirements for an ISL source listing for a program specification are given in Section III below.

Given Part B (the BNF for ISL), an ordinary English program specification or a list of specification requirements, and a knowledge of a language such as C, a programmer will be able to generate the source code for the program specification needed to practice the present invention. Examples of ISL specifications appear in Parts E (for a routine called "Sample") and Q (for the chmod operation under POSIX, shown in Part P).

Method Stage 2: Parse the specification source.

The ISL source listing of is parsed in a conventional manner by a parser generated by a standard compiler, such as YACC. See Stephen Johnson, "YACC: Yet Another Compiler Compiler", Bell Labs (Murray Hill, N.J.), which is incorporated herein by reference. (YACC is also discussed in virtually any UNIX documentation.) The compiler doing the parsing uses the BNF listing for the language in which the program specification was written (here, applicant's ISL), along with annotations and other definitions that the compiler needs, including definitions of objects which are to be referenced, definitions of symbols which are used, input/output functions, and so on. Automatic parsing procedures are well known.

The results of parsing the source listings of Parts E and Q appear in Part H (for "Sample") and R (for chmod), which also include applicants added numerical references relating the parsed results to the general transformation method set forth in Part C and to the comments contained in the respective ISL source listings.

The parsing procedure generates a symbol table reflecting the information expressed in the original program specification. The symbol table organizes the information by its role in the API, such as: properties and constants (including header file or comment information); routine name declarations; syntax declarations; parameters; and parameter constraint information. Parameter constraints are discussed below. (Execution constraints, mentioned above, are different: they are restrictions on the traversal of the EFSM, not on values of parameters. See discussion of Method Stage 4, below.)

The parsing of the program specification breaks down all expressions and declarations into their components, and stores them in a family of structures that captures both the raw data and the context in which it is used. This allows reconstruction into the original expressions for display purposes. When parsing is complete, the symbol table is filled in only in skeletal form.

Part C, the Transform Method for ISL, defines (beginning on page 1) the data structures which are used (sections A through I). Each data structure includes an enumerated list of characteristics, such as items 1-4 under "Routine", lists which lead to the numerical string references accompanying the parsed listings of Parts H and R.

The reference numeral strings appearing in the parsed listings

It should be noted that these numerical references are used only to facilitate the analysis of the method of the invention, and do not form a part of the method per se. Once it is understood how these numerical references are generated, they are useful in tracing the creation of the parsed source listings and the transformation of the parsed listings into EFSMs or architectures.

It will be seen that some of the defined structures A-I in Part C refer to other data structures. Thus a ROUTINE has a name and a type; and in order to determine how the type may be defined, one must refer to B (the TYPE listing). The type may then include a partition (item 8), for which we must refer to I: PARTITION (on page 3 of Part C). A partition may itself have a type denoted by a type partition (item 6), or a property denoted by a property partition (item 7), or may in fact include a list of partitions (item 8). For item 6, we refer back to the TYPE structure; for item 7, to G: PROPERTY; and for item 8, we refer recursively to I: PARTITION. A given data structure having this mixture of characteristics would be defined by a concatenated numerical string identifying the path followed through the defined data structures A-I.

Each entry in the parsed specifications appearing in Parts H and R is identifiable by such a concatenation of numerals. For instance, the first two lines in Part H state:

The routine specification describes a routine that

1: is named "sample"

Since the program specification defines a routine, we refer to A: ROUTINE in Part C. "Name" appears under item 1, so "is named `sample`" is accompanied by a reference to "1".

The next lines are:

2) has a return type that

2.1) is not named

2.2) is classified as an integer type

2.3) occupies four (4) units of storage

2.4) is signed

These lines indicate that "sample" returns an integer, as in fact it does (see under SYNTAX in Part E). The reference to the type of the routine correlates to item 2 under "Routine" in Part C, namely "the type (q.v.) describing the data type of the routine's output".

Referring now to TYPE (data structure B in Part C), item 1 refers to "name"; so the line above stating that the type is "not named" has the reference 2.1. Similarly, the type classification as an integer has reference 2.2 (since type classification is time 2 under "Type"), the size (4 bytes) has reference 2.3, and the sign characteristic has reference 2.4.

Similarly, "sample" includes a field group ("a") which is defined in the parsed listing. The corresponding numerical references begin with a "3" because the field groups are under item 3 in the ROUTINE section of Part C. "Sample" contains only one field group in its "list" of field groups, identified at line 3.1 (the "1" referring to the first field group). The single field group also contains only one field, leading to the designation 3.1.1. (In general, when a list is encountered, the number of the item in the list will be identified by an additional numeral. The second field in the group, if there were one, would have the reference number 3.1.2.)

Using this approach, any entry in Part H or Part R can be analyzed. For instance, numerical entry 3.1.1.2.8.8.5.7 in Part R. It defines a data structure as follows:

______________________________________ Digit Corresponding entry in Part C ______________________________________ 3 Routine: list of field groups 1 first group in the list of field groups (go to C: FIELD GROUP) 1 first (and only) field in this group (item C(1); go to D: FIELD) 2 type (item D(2); go to B: TYPE) 8 has a partition (item B(8); go to I: PARTITION) 8 the partition is composed of a list of partitions (item I(8); recursively go to I: PARTITION again) 5 the fifth partition in the list of partitions 7 has a property (item(7)), namely the name "file.sub.-- does.sub.-- not.sub.-- exist". ______________________________________

(The meaning of this structure will be made clear in Section V below.)

These numerical cross-references in the listings of Parts E/H and Q/R, manually created by applicant, show in the above manner how they relate to one another and to Part C. As noted above, the parsed result represented by Part H is itself generated automatically by the parser created by YACC; that is, the processor stores an internal representation of the relationships and hierarchies expressed in Part H. In normal use, one would not print out this parsed code, nor insert the numerical reference strings; this is one here for explanatory purposes. By inspecting each numerical reference string in Parts E/H and Q/R in light of the transformation method data structures in Part C, it can be shown that Parts H and R represent valid parsings of the coded specifications of Parts E and Q, respectively.

Method Stage 3: Transform the parsed source.

The output of the transformation in stage 3 is an EFSM representing the entire input program specification. (The EFSM may, however, include submodels which themselves include submodels, and so on.) The EFSM is an internal representation of a connected graph corresponding to the specification, and can be visually represented, as in FIGS. 5 and 6A-6B. It is expressed in applicant's Path Flow Language (PFL) (see above-mentioned "PFL" patent) or another language suitable for representing EFSMs. See the general discussion of EFSM graphical representations in Section III.E below.

The EFSM is generated by traversing the data structures created during the parsing procedure, creating states and transitions as needed for each data structure. Some data is transformed into testing information that annotates the transitions of the connected graph, while some is transformed into control information (or parameter constraints) on the graph.

Part C includes a description of the method of transforming the parsed listing (Part H of Part R) into an EFSM. Part C is a plain-English representation of the transformation method of the invention. The method is largely self-explanatory from this part, but will be discussed in some detail for the examples.

The overall method includes sixteen subsidiary method modules M1 through M16, which are listed at page 8 of Part C and are described in detail on the succeeding pages. (To maintain the distinction between the overall transformation method and its subsidiary methods M1-M16, the latter will generally be referred to simply as "modules".) In implementation, the transformation method is coded to provide automatic transformation of the parsed listing of the program specification in question. Given Part C, coding the method is a routine matter for a programmer.

FIG. 4 is a functional block diagram of the transformation method, showing the interrelationships among subsidiary modules M1-M16 (represented by boxes 101-116 and described fully in Part C). Each arrow from one module box to another represents a call from one module to the other. For instance, arrows 101.2, 101.3, 101.4 and 101.5 indicate that the main transformation module M1 may directly call any of the modules Ms, M3, M4 or M5. An arrow looping around from a box to itself represents a module recursively calling itself, such as arrow 102.2 coming from, and ending at, module M2 (box 102). (Note that the reference numeral for a given arrow is the same as that of the module from which it emanates, plus a decimal indicating the module to which it is directed.)

Following is a description of the function of each of the sixteen transformation method modules M1-M16. Which of these modules is used in the transformation of a given parsed specification, and the order in which the modules are called, depends on which elements are present in that-specification. However, every transformation procedure must begin with the main transformation module M1, and must follow steps M1:1 through M1:14 of that module.

a. Components of an EFSM produced by the transformation procedure

As noted above, EFSMs are known and used in the art of validation testing. It is useful to note the elements of an EFSM:

(1) a set of states, including one labeled as the starting state and one labeled as the ending state. There will generally be many intermediate states for any non-trivial EFSM.

(2) a set of transitions (see below) that describe under what circumstances the EFSM changes state, and to what state it changes.

(3) a context describing the path history of the current EFSM execution at any intermediate point in the execution, including the following:

(3a) the set of states through which the machine (EFSM) has traveled;

(3b) the set of transitions the machine has taken; and

(3c) the set of events (see below) that have occurred during the execution.

b. Elements of a transition

A transition is a mapping from one EFSM to another, and includes the following:

(1) input conditions for the transition, including:

(1a) a state in which the machine must be for this transition to be taken;

(1b) an input event (optional);

(1c) a parameter constraint (optional) which the current machine context must satisfy for this transition to be taken; and

(1d) a predicate (optional), comprising textual references to components of the machine context, combined into a single text string using logical operators (AND, OR, NOT).

(2) output conditions for the transition, including:

(2a) the state into which the machine changes if the transition is traversed;

(2b) a new context (optional) or change in the context, which becomes the new context if the transition is traversed;

(2c) an event (optional), which is added to the new context if this transition is traversed; and

(2d) a string of test data.

c. Events

An event is a named object that may be attached to a transition, and that may have data associated with it which can be retrieved by name during the execution of the EFSM.

d. Reading the transformation listings Parts I and S) for the examples

The transformation procedure 100 of FIG. 4, including the sixteen modules discussed below, converts the parsed ISL code into an EFSM stored in the memory 30 of the CPU 20. The EFSM 130 is merely a graphical representation of the internally stored model of the specification.

The transformation method always begins with module M1 (see the listing in Part C). Step 1 of method module M1 (which may be referred to as step M1:1) indicates that a starting state must be created. Part I, which represents the application of the transformation method (Part C) to the parsed "sample" listing (Part H), indicates execution of step M1:1 under "State SO", which states that "an entry state is created, SO". This state SO is graphically represented as vertex 200 in FIG. 5.

In general, reference herein to a method (or module) step such as M1:1, M6.2.2.1 or the like refers to either the general statement of the transformation method found in Part C or the specific application of that method to one of the examples as set forth in Part I (for "sample") or S (for chmod), depending on context.

From FIG. 4 (and from the listings of the modules in Part C), it can be seen that all of the modules except for modules M1, M5, M12 and M16 may call themselves recursively. This is indicated in Parts I and R by parentheses. Thus, in Part I under "State S3", the first call of module M6 starts with M6:1-i.e., step 1 of module M6. When it calls itself a few lines later, however, the entry is M6(2):1-step 1 of module 6, second "nested" call (first recursive call).

In this way, Parts I and S can be interpreted. They will be discussed in detail in Sections III and IV, respectively.

e. The functions of modules M1-M16

The transformation procedure of Part C involves two basic procedures: (1) the creation and annotation of EFSM components (see above); and (2) the building of a sequence of these EFSM components, along with annotations, based upon the content of the program routine specification.

The transformation modules perform the following general functions in generating an EFSM from the data structures represented by the parsed specification:

(1) States are created, including a starting and ending state.

(2) Transitions are created between states, going in each case from the present input state (1(a) under Elements of a Transition above) to the present output state (a(b) under Elements of a Transition).

(3) One or more events may optionally be attached to any transition, defined by the name of the event and the data associated with the event.

(4) Tests data, characterized by a data string, may optionally be attached to any transition

(5) One or more predicates, characterized by a string containing the predicate expression, may optionally be attached to any transition.

Module M1: Main Transformation

This is the main module, through which every transformation must proceed. It specifies the fundamental steps required to convert a parsed program specification into an EFSM. The steps are summarized below, and are set forth in more complete detail in Part C.

Step M1:1. The starting state is created. An example is state SO (vertex 200) in FIG. 5.

Step M1:2. A declaration (in the form of a data string) for the return value of the routine is generated, using module M2.

Step M1:3. Another state is created.

Step M1:4. A transition is created between the starting state and the second state.

Step M1:5. A program header string is added as test data annotation to the transition.

Step M1:6. The declaration string from step M1:2 is added to the transition.

Step M1:7. The second state (created in step M1:3) is remembered as the current state.

Step M1:8. This step declares the type of each field in each of the field groups specified in the declaration of the routine. For each such field, a new state is created, and a transition is created from the current state to the new state.

A data string representing the type declaration is added to the transition. The new state is then remembered as the current state. This step calls module M2 in order to declare the field type. The step is reiterated until each field in each field group is represented by a type declaration annotated to a transition to a new state.

Step M1:9. This step also processes each field of each field group in the routine declaration. Sub-steps M1:9.2* (i.e. the block of steps M1:9.s through M1:9.2.6.2.6) form the core of this step, including constructing (by a call to module M3 at step M1:9.2.4) a section of the EFSM for each field and assigning values to the field as identified by a field partition.

In sub-steps M1:9.2.6.1*, including a call at step M1:9.2.6.1.1 to module M4, a predicate expression is created for each value constraint, and it is annotated to a new transition going to a new state.

In sub-steps M1:9.2.6.2*, module M4 is again called for any fields that are identified as required if or required iff, and a predicate string is created reflecting the requirement. A transition is created to the current state and a predicate annotated to it reflecting the negation of the predicate string.

Sub-steps M1:9.3.1* create a new state and transition for any field indicated to be optional. Substep M1:9.3.2 adds a string indicating assignment of no value to the transition for each field that is not indicated to be required.

Step M1:10. If the routine declaration contains a value statement (which is optional, but in practice will usually be present), this step calls module M5 to build a portion of the EFSM corresponding to the value statement.

Steps M1:11-12. Here an exit state is created, and a transition to the exit state.

Step M1:13. A data string is added to the final transition, indicating an invocation of the specified routine. The parameter values are the names of the fields, and the output value is "result".

Step M1:14. Another data string is added to the final transition, representing a program closing.

Module M2: Declare Type

This module, called in step M1:8.1.1 of module M1, receives as input a string and a type, usually the name and type of a filed, and constructs and returns a text string corresponding to the input. Different data types are treated differently: pointer (sub-steps 2.1*); array (2.2*); void (2.3*); integer (2.4*); opaque or indirect (2.5*); and floating point (2.6*). Recursive calls to M2 are made until all depths of the input type have been processed.

Module M3: Transform Type

This module is called by step M1:9.2.4, and constructs a section of the EFSM including value assignments to the field identified by the input string. (The value assignments are test data on the transitions.) The module has as input a string identifying the name of the field, a partition identifying the values of the type, a starting state, and an ending state. In addition to this information, the global state and transition locations are referenced, well as the global partition level indicating the invocation level in calls of module M6 (Transform Partition).

Different field types are treated differently: pointer (sub-steps 1.1*); opaque, integer, floating or void (1.2*); struct or union (1.3*); and indirect (1.4*).

Module M4: Transform Expression

Module M4, as shown in FIG. 4, is a module central to the transform procedure, including calls to most of the other modules. Module M4 is itself called by module M1, steps M1:9.1.6.1.1. and M1:9.2.6.2.1, for, respectively, fields that have value constraints and fields that are indicated as being "required if" or "required iff". It receives as input an expression identifying an expression, and converts this into a predicate string representing references to all transitions corresponding to value references made in the expression. If there are Boolean operators in the expression, these will be reflected in the predicate string. The predicate string is returned to the calling routine. States and transitions may be created, and the predicate string(s) annotated onto the appropriate transition(s).

Different operator types (see Section III.D below) are treated differently, namely mask, choice, or array (sub-steps M4:1.1*) or range (sub-steps M4:1.2*). In the latter case, module M7 (Particle to String) is called.

The expression is classified (see Part C, Section F: Expression) depending upon the number of subsidiary expressions, as follows:

Expression Class 0: a basic object

Expression Class 1: having a single subsidiary expression

Expression Class 2: having two subsidiary expressions

Expression Class n: having more than two subsidiary expressions (or a non-fixed number of subsidiary expressions)

Part C, Section F gives further detail as to expression operators, types, etc. The classification of the expression operators corresponds to the classification of the expression itself. Operators of the four classifications are described in detail in Part C, under "Expression Operators".

Steps M4.2* treat expressions of classification 2, with the treatment depending upon the operator of the expression:

for OR: sub-steps 2.1* are invoked, including recursive calls to module M4;

for AND: sub-steps 2.2*, also including recursive calls to M4;

for "has property": sub-steps 2.3*, including calls to module M11 (Transform Property Partition);

for "mask contains": sub-steps 2.4*, including a call to module M8 (Transform Mask Partition) if the expression is NULL;

for "type"; sub-steps 2.5*, including a call to module M10 (Transform Type Partition);

for "pointsto" or "dot": sub-steps 2.6*, including recursive calls to M4;

for "property": sub-steps 2.7*; and

for all other types of operators: sub-steps 2.8*, including a call to M16 (Relational Operators).

Steps M4:3* treat expressions of classification 1, with the treatment again depending upon the operator of the expression:

for "has property": sub-steps 3.1*;

for "property": sub-steps 3.2*;

for "present": sub-steps 3.3*, including a recursive call to module M4;

for "verify": sub-steps 3.5;

for "pointer": sub-steps 3.6*, also including a recursive call to module M4; and

for "negate": sub-steps 3.7*, including a call to module M3 (Transform Expression).

Steps M4.3* treat expressions of classification 0, generating a string corresponding to the name of the operator (for "number" or "string"), or the name of the described type, field or property, as the case may be.

Module M5: Transform Value Statement

This module receives a statement as input, which may be of the type "block", "simple", "conditional", or be empty. It constructs a section of the EFSM corresponding to the input statement. A block statement, treated in sub-steps M4.2*, results in a section of the EFSM with a sequence of independent substatements. An empty statement (substep M5.4.3) results in no new addition to the EFSM. A simple statement (sub-steps M5:4.4*) leads to a call of module M4 (Transform Expression). This call is represented in FIG. 4 by arrow 105.4, leading from module M5 (box 105) to module M4 (box 104).

A conditional branch statement (sub-steps M5:4.5*) results in the creation of new states and transitions representing the alternatives (see steps M5:4.5.2-M5.4.5.6), and M4 (Transform Expression) is called (step M5:4.5.7). The transitions are annotated with strings representing the alternatives of the conditional statement.

Module M6: Transform Partitions

This module has as inputs a string identifying a location where values are to be stored, a partition identifying those values, a starting state and an ending state. The partition relates to possible values for a variable, such as an assignment, where alternative values must ultimately be tested. For instance, an assignment of a =2 will lead to a partition where values of a<2, 1=2 and a>2 are all tested in the validity tests of the implementation. A partition may be divided, in which case it includes a set of subsidiary partitions covering the set of values indicated.

This module results in a section of the EFSM with a single new transition representing an assignment statement (for a non-divided partition), or with new transitions corresponding to alternative value assignments as determined by the partitions. Different types of partitions are treated in different portions of the module:

Divided partitions are treated in sub-steps M6:2*, including a recursive call to module

M6 for each subsidiary partition of the divided partition.

Mask partitions are treated in sub-steps M6.3*, including calls to modules M14 (Expression to Partition) and M15 (Expression to String).

Type partitions are treated in sub-steps M6.4*, including a call to module M3 (Transform Type).

Discrete partitions are treated in sub-steps M6:5*.

Range partitions are treated in sub-steps M6:6*.

Expression partitions are treated in sub-steps M6.7*, including a call to module M15 (Expression to String).

Property partitions are treated in sub-steps M6:8*.

Module M7: Partition to String

This module receives two partitions as input, the second of which is a range partition, and the first of which may be a range partition, a discrete partition or a divided partition. If the first partition is a range partition and its value falls within the second partitions's range, a string referring to the first partition's transition is returned; otherwise, a null string is returned (see steps M7:1*). If the first partition is a range partitions (see steps M7:2*), a similar procedure is executed.

If the first partition is a divided partition (see steps M7:3*), module M7 is recursively called for each of the subsidiary partitions of the divided partition. If the first partition is any other kind of partition, a null string is returned.

Modules M8, M9, M10 and M11

These modules (Transform Mask Partition, Transform Expression Partition, Transform Type Partition and Transform Property Partition) perform essentially identical operations. Each receives as input a partition and an object which may be an expression (for a mask partition (M8) or an expression partition (M9)), a type (for a type partition (M10)), or a property (for a property partition (M11)). Each module returns a predicate expression corresponding to a transition at which the value corresponding to the expression is assigned. Step 2 calls module M8 and step 3 calls module M9.

Module M12: Type to String

The input to this module is a type. The output is a string containing a C expression identifying the type. For instance, for an integer occupying four bytes of storage, this module would return a string ("LONG INT") reflecting this.

Module M13: Locate Partition

This module receives as input two partitions, the second of which is a discrete partition whose range falls within that of the first partition. A string is returned containing reference to a transition corresponding to the subsidiary partition of the first partition that matches the value of the second partition--unless the value match is found in a mask partition.

If the first partition is divided, sub-steps M13:1.1* perform a recursive call to module M13 for each subsidiary partition of the first partition. Sub-steps M13:2* show the treatment of a discrete partition. If the first partition is a mask partition (sub-steps M13:3*), module M13 is recursively called in a more complicated treatment (including also sub-steps M13:3*). For other types of partitions, an empty string is returned (steps M13:4*).

Module M14: Expression to Partition

This module is called by module M6 (Transform Partitions), which is itself called during parameter or value statement transformations by modules M3 (Transform Type), M15 (Expression to String) or M16 (Transform Relational Operators), M15 and M16
themselves being called by M4 (Transform Expression). See FIG. 4.

Module M14 receives as input an expression and a partition. The expression is a subsidiary expression of an expression. If the partition is null (sub-steps M14:2*), a null is returned. The treatment of the expression depends on whether it is or classification 1 (steps M14:3*), 2(steps M14:4*), 0(M14:5*) or n(M14:6*).

This module is part of the process of producing text strings for use as predicates or test information.

Module M15: Expression to String

Module M15 is called by module M4, as shown in FIG. 4, and receives an expression as input, such as an expression representing a data location. The output is a string containing a C expression corresponding to the input expression. The output string produced depends upon the kind of operator of the input expression. For certain kinds of operators (dot, pointsto and length), M15 is called recursively; see steps M15:2*, M15:3* and M15:4*, respectively. If the expression operator is "constraint", "number" or "string", the text string contains the constant's name (M15:5*), the formatted umber (M15:6), or the expressions' string, respectively (M15:7*).

Module M16: Transform Relational Operators

This module is called by M4 (Transform Expression), as shown in FIG. 4. The input to the module is an expression which is a subsidiary expression to the expression in the calling modules, and which is a binary expression including a relational operator. It returns a predicate string corresponding to both halves of the binary expression and of the operator.

If the input expression has a partition associated with it, then a predicate expression corresponding to the edge in that partition is returned. This situation is treated in steps M16:1.2*.

If the input expression does not have a partition, this means that a more complicated predicate must be constructed. The situation is covered in steps M16:1.1*. This arises under certain circumstances, such as when (1) the operator of the predicate expression is == (EQUAL) or != (NOT EQUAL), and (2) operator of the second subsidiary expression is choice, then the predicate expression that is formed is the result of processing the second subsidiary expression, and all the subsidiary expressions of that choice expression will have partitions associated with them, and those partitions will have associated edges, all of which must be Or-ed together. If the operator of the input expression is NOT EQUAL, then these Or-ed together edges are negated in the output predicate expression (see M16:1.2.1.2*.)

Treatment of "eq" and "neq" (for "equals" or "does not equal") operators are in steps M16:1*, and of other non-null operators in M16:2*. The second subsidiary expression of the input expression is processed (step M16:3) by a call to module M14
(Expression to Partition). The partition of the second subsidiary expression is processed by a call step (step M16:4) to module M13 (Locate Partition). If the operator is "neq", it is further treated (steps M16:1.5*) by the creation of a text string which is the NOT of the result of the call to module M13.

Operators of the types "ge" or "gt" (for "is greater than or equal to" or "is greater than") (steps M16:2*) and "le" or "it" (for "is less than or equal to" or "is less than") (M16:3*) result in a partition of the type of the first subsidiary expression of the binary expression, and a text string associated with the partition's transition.

The result of the transformation is, as noted, an EFSM representing the program specification. Graphical representations of EFSMs are shown in FIGS. 2, 5 and 6A-6B. See Section III.E below for further discussion of these representations.

Method Stage 4: Traverse paths through the EFSM.

In stage 4, the EFSM is traversed, which may be accomplished by a conventional path generation method. For instance, the method described in the Jessop article on Atlas may be used. Each test will consist of a single path through the EFSM, each such path effectively constituting a set of test data (such a set being known as a test vector) which is input to the implementation. If there are, for instance, fourteen paths generated through the EFSM, then there are fourteen text vectors on which the actual program must be run to determine whether, in each case, it complies with the specification.

Applicants aforementioned PFL patent includes a thorough description of a method of traversing the EFSM.

As noted, stage 4 may be carried out exhaustively or selectively. For small EFSMs, i.e. those with relatively few paths (from several to several thousand), the exhaustive approach is preferable, since the program is comprehensively tested for all possible combinations of variable values, etc.

However, a complicated specification may have very large numbers of possible combinations of variable values or other parameters, and hence very large numbers of possible paths through he resultant EFSM. For instance, if there are fourteen variables, each with ten possible values, there are at least 10.sup.14, or 100 trillion, possible paths through the EFSM. This is an unwieldy number of paths to generate, and would consume an enormous amount of computer time, both to generate the paths and to run the resultant tests. Even greater numbers are possible where loops are created with large numbers of iterations. Where a routine is recursive and there are no internal limitation on the number of times it may call itself, the number of paths may in fact be infinite.

To avoid such situations, execution constraints may be placed on the process of generating paths. Thus, if a loop exists with a large possible number of iterations, the user may specify a constraint for that loop of, say, three iterations, so that values for variables generated in the loop may be tested, without testing every possible iteration of the loop.

Similarly, means are provided for flagging unconstrained recursions. During execution of the path generation program, any path with an unlimited recursive call would be identified, and either no paths would be generated including that recursion, or a limited number of recursions (perhaps one or two) may be executed, so that at least some paths are generated which include the recursive call.

Execution constraints are more important in the design of state-rich models, and are discussed in detail in applicant's PFL patent.

As the transformation according to stage 4 proceeds, any tests which are encountered (like Boolean tests) are flagged by the addendum of a "% verify" prefix. The purpose of this is explained in the discussion of Method Stage 5, following. An example of a "% verify" statement in the specification for "sample" appears in Part E, which is discussed in detail in Section V below.

The result of the path traversal stage is a set of path files. For allpaths (exhaustive testing), all paths through the EFSM will be traversed, and one path file will be generated for each of these paths. For edge cover (thorough testing and identification testing), one path file will be generated for each path actually traversed, which will be fewer than the total number of possible paths.

Method Stage 5: Post-process the path files.

The path files form the core of the validation tests. Before they can be used as tests, however, they must be post-processed.

The test statements are flagged with "% verify" because the validation tests based upon the path files must be executed; and the tests cannot be executed unless calls upon which they rely have already been made. Since, in a specification, the tests may be stated anywhere (such as before the related call is defined), this would result in errors, namely the failure of a test because it does not have the required parameter. Thus, in order to ensure that the calls in the test programs occur before the "verify" tests are executed, these tests are flagged during traversal of the EFSM, while the path files are created; and then post-processing is carried out, which involves moving all of the "verify" statements to the end of each validation test program.

To do this, the flagged "verify" statements are collected together in a list. The very last step of the transformation procedure (method step M1:14) includes a statement (".backslash.nverify;") which is an instruction to dump (list) all of the "verify" tests which have been placed in the VERIFY list. The word "verify" is stripped from each of the "verify" tests, so that they now constitute implementations of the tests in the specification. The sets of "verify" tests (a unique set for each generated path) thus appear at the ends of the path files. When validation test programs are built using these path files (see Method Stages 6-8, following), the "verify" tests appear as conditional statements in the programs.

An alternative method is to append the work "verify" to each test in the specification as it is encountered, and build a list as the specification is transformed (per the transformation method of Part C). Each "verify" line in this list is then correlated to the edge to which it pertains. Then, at the end of the transformation, a series of new states is created, one for each of the "verify" tests. This series of states is added to the EFSM for each of the tests generated by the path generator. Each state is connected to the previous state by two edges, with opposite predicates. One predicate will be something in the form of "P: edge.sub.-- X.sub.-- was.sub.-- taken", and the other will be its opposite, as "P:ledge.sub.-- X.sub.-- was.sub.-- taken". The first of these edges includes the test which pertains to a path including edge X, and will be taken only if the edge X was taken. Thus, only those generated paths which include edge X will run the test attached to the edge to which "P: edge.sub.-- X.sub.-- was.sub.-- taken" is also attached. This ensures that a given test cannot be made if its prerequisite call has not been made.

Note that this alternative method results in the concatenation of the same set of edges and states at the end of all the EFSMs that are generated, and that these additional edges include all of the "verify" tests present in the specification. If a specification includes twenty such tests, then twenty states connected in series via forty edges (two to each state) will be added at the end of each test.

Typically, the majority of the predicates will fail for a given test; i.e., each generated path will be likely to include only a small subset of the edges resulting from the "verify" tests.

Method Stage 6: Create program shells

The path files generated in stage 4 and post-processed as necessary in stage 5 are in the form of source code representing the paths taken through the EFSM. This source code is in C or whatever other language is specified. In stage 6, each path file is inserted into a program shell that also includes the actual implementation to be tested.

An example of such a program shell for the routine "sample" is illustrated in Parts M (the beginning of the shell) and N (the ending of the shell). Thus, each test program constitutes an actual C language program consisting of the program shell fragments coupled with one path file. Each such validation test program includes an executable call to the implementation to test its consistency with the specification for the set of data relationships represented by that program.

The program shells are automatically generated by the method of the invention, and have certain required sections. Since the point of the validation tests will be to verify the correct execution of an application, there must be some feedback to the user as to whether each test failed or not.

The program shell includes:

the beginning of the shell (such as Part M); and

the ending of the shell (such as Part N).

Each path file is inserted between one pair of these shell portions to generate the validation tests, that is, each validation test includes the shell beginning, followed by a single path file, followed by the shell ending.

The program shell is responsible for initializing the variables to be used by the implementation, which is called by each path file. The portions of the program shell are:

1. a definition for a routine (here called "verify.sub.-- failure") that provides an output indicating when an implementation has failed, i.e. does not comply with the program specification;

2. A definition for a routine (here called "pick.sub.-- range") which selects values to variables in ranges determined by partitions reflecting value assignments for the respective variables. For each parameter with a range partition defined in the parsed specification, there will be an edge in the resultant EFSM corresponding to that range. For each such edge, a single "pick.sub.-- range" statement is generated in each path file containing that edge. The "pick.sub.-- range" routine may be exactly as appears in Part M, or may be some other routine for selecting values in the correct ranges.

3. a random number generator routine, preferably, to provide a seed for the "pick.sub.-- range" routine, so that the values selected within the different ranges are otherwise random. Such a routine for the C language running in a UNIX environment appears in Part M.

4. a "main" portion. This portion begins in the first portion of the program shell, includes a path file, and concludes in the second (ending) portion of the program shell.

The beginning of the "main" portion of the program shell initializes a variable called "test.sub.-- result", which is a test failure flag appearing in each path file (as in Parts J, K and L). It also includes a type declaration for a variable having a name of the form "<name.of.routine>.sub.-- result", which also appears in each path file. Thus, for the routine "sample", this becomes "sample-result". The use of these variables is discussed below. Note that these variables could be declared before the "main" section, in which case they would be global instead of local.

The type declared for "<name.of.routine>.sub.-- result" will be the same as the return type of the tested routine. Since the specification for "sample" (Part M) indicates an integer return type, "sample-result" will be declared as an integer in the program shell. It will be used as global variable containing the result of making a call to the implementation from each validation test, and its value returned to the test program. It may then be used for other purposes, for instance as input to one or more tests later on, which might rely upon the outcome of this particular test.

5. an ending portion (see Part N) containing an if-then-else section; if test.sub.-- result is 0 (or whatever value was selected in the specification), then the program is to print out a "success" message, and otherwise it will print out a "failure" message.

The requirements of the program shell are clarified in the discussion of the example routine "sample" under Method Stage 7, below.

Method Stage 7: Couple test programs with program shells to create validation tests

This stage involves creating a copy of each of the first and second portions of the program shell for each path file, and joining them together with that path file to form a complete validation test program. There will thus be one such program for each path file generated in the traversal of the EFSM. An example of a complete test program would be: Part M, followed by one path file (Parts J, K or L), followed by Part N.

Method Stage 8: Execute

Stage 8 is the actual execution of these test programs. This is accomplished simply by making a call to the name of each validation test. The output of the tests will be messages indicating failure or success of each test. The tests that failed are thus flagged, and by inspection of those tests one can immediately determine which paths included erroneous program segments. This leads to efficient debugging, all accomplished automatically by using the program specification as the only hand-generated input.

Part O shows the steps one would go through to manually build and execute validation tests for the routine "sample" defined in the specification of Part E, given: the three path files (Parts J, K and L); the two program shell portions (Parts M and N); and two example implementations of the specification of Part E. These two implementations (Parts F and G) are named "sample.sub.-- good.o" and "sample.sub.-- bad.o". The first of these is designed to implement the program specification correctly, by returning a 0 if and only if a=2. The second is designed (for exemplary purposes) to function incorrectly, by returning a 0 even if a>2. The "Sample" routine is discussed in Section V below in greater detail.

The steps to building the validation tests include:

(1) compile the implementation(s) to create implementation object files;

(2) create test source code files by concatenating the program shell portions with each of the path files;

(3) compile these test source code files to create test object files;

(4) link each test object file with each implementation object file to create test images; and

(5) execute the test images.

These steps can be carried out by keying in the commands if there is a limited number of tests. However, for general purposes it is an easy matter to carry them out automatically.

III. The Interface Specification Language (ISL)

ISL is applicant's preferred language for expressing the specification. Other languages may be developed and used, as long as they fulfill the functions necessary to support the parsing and transformation procedures described below. When reference is made herein to "ISL", it may be taken to mean any language meeting the criteria discussed below that enable the procedures of parsing, transformation and automatic generation of validation tests. Data segments in ISL are generally named according to C's "Ivalue" syntax, allowing for pointer, array and structure-member references.

Part A lists ISL keywords, and Part B presents the ISL Backus-Naur Format (BNF). The primary constructs of ISL are those describing data types and API entry points (or routines).

Part D, the user's guide for ISL, gives detail sufficient (in combination with Parts A and B) to generate source listings suitable for automatic parsing, transformation and ultimately generation of validation tests. Referring to Part D, Section
1.4, and ISL source file (such as Part Q) includes the following sections:

A. Property Section(s) (optional), including a comma-delimited list of symbols that will later be used in the parameters section(s) as constraints or in the values section(s) to declare a characteristic that parameters may have.

B. Constant Section(s) (optional), essentially the same as conventional constant sections in a standard header file.

C. Type Section(s) (optional). This syntactic unit contains the following:

1. the name and syntactic form of the type. This is a C declaration for the named types, which are user-specified. It may be a numeric enumerated, structure or union type, or some other type.

2. specifications for fields for structure and union types. This section is similar to the parameter-definition section for routines (see D.2 below). It is used to describe the characteristics and relationships of the field members.

3. a value-relationship section, similar to the value-relationship section of a routine (see D.3 below).

D. Routine Section (required), the heart of the ISL listing. It includes the following subsections:

1. Syntax (required), including information found in conventional header files, including the name of the routine, a description of the syntax (format) of the routine and declaration of the data type of the routine and the data types of its parameters. This is (in the preferred embodiment) a C declaration of the routine.

2. Parameters (optional), describing the characteristics of the parameters and groups of parameters declared in the Syntax section.

3. Values (optional), describing the relationships among the parameter, and between parameters and the routine.

The Parameters subsection (D.2) of the Routine Section defines characteristics falling into three different categories (see Part D, Section 1.4.4.2):

D.2.a I/O status, specifying the relationships between parameters and the call format, including whether a parameter is for input, output or both, and whether a value must be supplied for a particular parameter. The I/O status may be one of three types:

(1) INPUT, indicating that the parameter is used to pass data into the routine;

(2) OUTPUT, indicating that the parameter is used to return data from the routine; or

(3) INOUT, indicating that the parameter is used to pass a value into the routine, where the value may be updated and passed back out.

D.2.b Optional/Required status, including:

(1) OPTIONAL, indicating that the parameter is always optional;

(2) REQUIRED indicating that the parameter is always required;

(3) REQUIRED.sub.-- IF, indicating that the parameter is required if a specified circumstance is true. The parameter may be present if that circumstance is not true.

(4) REQUIRED.sub.-- IFF, indicating that the parameter is required if and only if a specified circumstance is true.

D.2.c. Parameter (or value) constraints, expressed as Boolean expressions or using ISL operators (see table below), defining restrictions that may be placed on parameters. Constraints may be placed on relationships between parameters, such as how the structures and values of some parameters affect the structures and values of other parameters, and how specific parameter/data combinations map to specific output results and actions. Thus, constraints may restrict a given parameter to having values within a defined range, or one of a list of values, a mask of values, a specific type or property, or a function of another parameter value. Operators for imposing parameter constraints include length, choice, range, mask, present, mask.sub.-- contains and has.sub.-- prop, defined as follows (see Part D, Sections 1.4.4.2 and 1.5):

ISL OPERATORS

______________________________________ Keyword Syntax Operator Definition ______________________________________ has.sub.-- prop has.sub.-- prop(A,B) A has the properties of B type type(A,B) A has the type of B range range(A,x,y) Range of A is x to y present present(A) A is required to be present choice A=choice(x,y,...) A may be one of x, y, etc. mask A=mask(x,y,...) A is mask of values x, y, etc. length A=length(B) A is length of B mask.sub.-- contains mask.sub.-- contains(A,B) A bitwise AND B is nonzero ______________________________________

These operators are used in ISL with a % symbol, as will be seen from the ISL source listings in Parts E and Q, and have functions as follows:

The has.sub.-- prop operator assigns a named property to a data segment (parameter). The semantics of the named property are defined externally to ISL, typically in the context of a test execution environment, such as C.

The type operator ascribes to a parameter the semantics of a type other than the one given in the syntactic description.

The range operator restricts the value of a parameter to a specified numeric range.

The present operator restricts the value of a parameter to a specified numeric range.

The mask operator restricts the value of a parameter to a subset of specified bitmasks, connected together with the OR operator, and hence performs a bitwise OR on the parameter.

The length operator requires the specified parameter to have the length of the parameter appearing in parentheses.

The choice operator restricts the value of a data segment to one of a set of specific values.

The mask contains operator requires that a bitwise AND of the first subsidiary expression and the second subsidiary expression be nonzero, hence that former includes the latter.

These and additional operators are defined in detail in Part C.

As specified in Part D, the ISL User's Guide, other types of operators used to define parameter constraints include:

relational operators (used in conventional C fashion):

Boolean operators:

The Values subsection (D.3) of the Routines section utilizes a list of operators similar to those used by the Parameters subsection (D.2), and in addition use keywords unique to the Values subsection, namely IF, ELSE and VERIFY.

IV. Output

ISL implements procedures that generate EFSMs, validation tests, and documentation. The generation of EFSMs and validation tests is discussed above (see Sections II/Method Stages 3 and 4-8 respectively). As mentioned, the documentation is produced in a conventional manner using an application such as nroff.

A. Graphical representation of EFSMs: Path Flow Language

An EFSM is represented internally as a set of objects representing states and transitions between states, with the transitions including predicates, parameter constraints, events, actions and tests (e.g. Boolean tests). The EFSMs are no longer expressed in ISL, though ISL was the starting point when the original program specification was generated. Rather, the EFSMs are expressed in applicants Path Flow Language (discussed above) or another suitable language.

While these EFSMs may be displayed graphically, that is not necessary for the generation of tests of the implementation. The graphical interface for creation and modification of EFSMs is described in the STE Users Guide attached as Part B to applicant's aforementioned PFL patent.

An example of a graphical representation of a simple EFSM 92 is shown in FIG. 2, with start state SO (box 94) and exit state (box 96) connected by a transition 98. State SO may also be referred to as a "vertex", as may the exit state; generally, the term "state" will be used herein to refer to the abstract state of the EFSM, i.e. the values of variables, stacks, etc., while the term "vertex" will be used to refer to the symbol in the figures. However, they may be used essentially interchangeably.

Vertex 94 may represent a state where a variable x is as yet unassigned. Transition (or "edge") 98 may represent an assignment (which is identified as test data, denoted by the "T") of x=2. The exit vertex 96 then represents a state where x has a value of 2. Note that the arrow on transition 98 proceeds from vertex 94 to vertex 96, indicating that the flow is in that direction only.

This example, while very simple, represents the pattern of the graphical representations of EFSMs discussed below. In those representations, a start state has the appearance of vertex 94 (or vertex 200 in FIG. 5), and an exit state has the appearance of vertex 96 (or vertex 208 in FIG. 5). Intermediate states have a symbol with a two-way arrow in a box, such as vertices 201, 202, etc. Transitions in the EFSMs of FIGS. 5 and 6A-6B are represented as arcs or lines, with an arrow indicating the direction of flow and the decimal indicating the vertex to which the arrow points; thus, arc 200.1 extends from state SO (vertex 200) to state S1 (vertex 201) in FIG. 5.

If there are two or more alternative paths which may be taken between a given pair of states, they will be represented as consecutively placed arrows, as with transitions 202.2.1, 202.3.2 and 202.3.3 in FIG. 5 (the second decimal indicating which of the arcs is referred to). This occurs where different paths are taken depending upon whether a condition is or is not satisfied; or where a variable is assigned one of two or more values; and in other situations.

Along the transitions may appear any of the abbreviations A (for an action), C (for a constraint), E (for an event), P (for a predicate) and T (for test). In the examples herein, however, illustrations only of tests and predicates are shown. The other types of annotations (see the Introduction above) are more appropriate to representation of a state-rich problem, as in applicants PFL patent.

B. Path files: the validation tests

When the EFSM representing the specification is traversed as described in Section II/Method Stage 4, a "% verity" statement is created for each test encountered, e.g. each "if" statement annotated to an edge in the EFSM model. These statement are concatenated at the ends of the path files, as described for Method Stage 5. The validation tests are then built up, as described for Method Stages 6-8, using the program shells. These tests are in the C language in the present examples, but may be in any language desired by the user; the only requirement is that the annotations (test data) be expressed in the target language (which is controlled by the transformation).

Validation tests include the following elements:

1. a type declaration for "result", the type being determined by the specification.

2. type declarations for any parameters (if any) of the routine being tested, the types being determined by the specification. These may be declared before or in the "main" portion (see below).

3. assignments of values to the parameters (if any) of the routine being called, including the "pick.sub.-- range" statements (see Section II/Method Stage 6, above).

4. a statement to make the call to the routine being tested and assigned the result to a variable "result", the statement basically of the form: routine.name.sub.-- result=result=<routine.name(parameters)> For "sample", this becomes:

sample.sub.-- result=result=sample(a)

The variable "result" is used locally in the validation test, while "sample.sub.-- result" may be used globally. (This could, in this case, be replaced by the single statement

sample.sub.-- result=sample(a),

but there may be reasons for wanting to keep the variables "result" and "sample.sub.-- result" separate.)

5. For each "verify" statement encountered in the traversal of a path (in the creation of path file), an "if NOT" statement is ultimately created to generate failure reports. An example may be seen in Part J; the routine if (! (result==0))

test.sub.-- result=1

verify.sub.-- failure("result==0");

is used to generate a value for test.sub.-- result of 1 if the result of calling the implementation under test fails, i.e. returns a value for result of which is not 0. (It also utilizes the verify.sub.-- failure routine, which is defined in the first part of the program shell, to print out a failure message in the case of an incorrect implementation.)

C. Documentation

ISL provides a facility for passing ASCII text in the source directly through to the documentation. This documentation can completely replace the handwritten nroff source now in use.

Since the ISL source comments are in English (or some other natural language) and are by design unambiguous, documentation can be produced automatically for any specification expressed in ISL. This eliminates the burden of manually creating documentation and updating it when the specification changes. In applicant's current embodiment, documentation throughput is hundreds of manual pages compiled per second into nroff source.

V. Testing an implementation of an example routine: "Sample"

Part E is a specification, expressed in ISL, for the "sample" referred to above, which shall be used to illustrate the method of the invention. Its function is simply to accept an input for the variable "a", and to return a result of 1 if a=2, but otherwise to return a result of 0.

"Sample": Stage 1--Create specification source code.

The specification of Part E represents the result of Stage 1 of the method of the invention (see Section II above). It has been created in ISL, so it is ready to be automatically parsed and transformed. If it had been written in ordinary English, it would have to be coded into ISL.

Part E includes the ISL components set forth in Section III.D above namely a Syntax section, a Parameters section, and a Values section. This particular specification does not require the Property, Constant and Type sections (see Sections III.A, B and C above).

The Syntax section includes a declaration of the type of result to be returned by the routine, in the statement:

int sample (int a)

This indicates that the "sample" routine takes an integer "a" as input, and returns an integer result.

The Parameters sections states that the input "a" is required. Finally, the Values section states that if the input value of "a" is 2, then the routine should return a result of 1; otherwise it should return a result of 0.

The % verify statements at the end of the specification of Part E flag strings that will be used as test statements in the validation test programs created at the end of the overall procedure, as explained in Section II/Method Stage 5, above.

An implementation of the specification of Part E must: (1) accept the integer input "a"; (2) test whether it is equal to 2; (3) if it is equal to 2, return 1; and (4) if not, return 0. An example of such a program, written in C and entitled "sample.sub.-- good.c", appears in Part F. This program can be tested for consistency with the requirements of Part E in the manner discussed below.

Another example implementation of Part E, entitled "sample.sub.-- bad.c", appears in Part G. It is immediately apparent that the implementation in Part G does not meet the specification requirements of Part E:A this implementation returns a value of 1 if "a" is equal to or greater than 2, whereas the specification requires a result of 1 only if "a" is exactly 2. Thus, "sample.sub.-- bad.c" should produce an error result when tested by the invention. That this occurs is demonstrated below.

"Sample": Stage 2--Parse the specification source.

The entire specification of Part E consists of a Routine definition. To parse this, we refer first to section A of Part C. Item 1 identifies the name of the routine; thus, line 1 in the parsed specification of "sample" (see Part H) reads "is named sample".

Line 2 of A: Routine in Part E identifies the datatype of the routine's output, and thus all lines beginning with "2" in Part H related to the datatype. Type is covered in section B of Part C. Line 1 under Type identifies the name of the type; thus, line 2.1 in Part H identifies the name of the output for "sample" (in this case, it is not assigned a name). The return is classified as an integer (line 2.2), as required in the specification (Part E) in the "int sample" declaration.

Lines 2.3-2.4 indicated that the integer is signed and occupies four bytes. This is standard for a C integer. Items 5-7 under Type in Part C (associated types, structure or union, constraints) do not apply here, so there are no corresponding lines 2.5-2.7 in Part H.

Item 8 of the Type description refers to partitioning; since the variable "a" will have a partition associated with it, due to the testing of its value (a=2 or not), we proceed to Section I of Part C: Partition. All lines in Part H beginning with "2.8" will thus relate to this partition. Partitions for the testing of different values for variables will be familiar to those knowledgeable in the area of software testing. Basically, enough partitions are created to test all ranges of values for the variable; here, since a=2 is tested, we will tes