JPH0863350A - Program analysis method and device therefor - Google Patents

Abstract

PURPOSE: To easily support an analyzer in his understanding of the program contents by acquiring the concept terms from the names of program identifiers. CONSTITUTION: The concept terms are acquired from the names of program identifiers and a concept term dictionary is prepared, and the similarity is measured among these concept terms and a concept network is prepared as an image that shows the relation among the concept terms. Then a cluster analysis is applied to a routine based on the degree of distribution of concept terms so that an analyzer can fix the module constitution of a program. Furthermore the analyzer can construct an information model of the program and therefore the concept terms can deal with the objects, attributes and links. Thus the concept terms can be acquired from the identifier names reflecting the program contents as the information on the abstract levels that show the program contents.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a program analysis method and apparatus for extracting abstract level information from a program.

[0002]

2. Description of the Related Art Generally, in order to perform a safe maintenance work of a program, a domain analysis for finding a reusable software component from the program, and a modification of the program in a good manner, the purpose and specifications of the program and It is necessary to understand the design and operation.

In order to understand the program as described above, it is general to refer to the document concerning the specifications and design created during the program development. However, it is difficult to understand the program well from the document unless the description of the document is sufficient or the document is not updated in response to the change of the program.

In such a case, the program analyst
You will need to analyze the program directly to understand it, but this is cumbersome and inefficient.

As a program analysis device for solving such a problem, there is a DESIRE system which analyzes a source code and supports the understanding of the program. This DESIRE system holds information about domain models and naming, and uses informal information such as program comments and program naming to establish its own domain model through approximate search and fuzzy search. Mapping with naming. In this way,
Identify the domain of the program and search the program for abstractions.

[0006]

DISCLOSURE OF THE INVENTION Problems to be Solved by the Invention
The system is designed to identify the domain from the program by mapping the domain model it has and the naming, but the effect is not sufficient, and another method is required.

[0007]

According to the first aspect of the present invention,
In a program analysis method that acquires information for understanding the contents of a program that has a hierarchical structure of modules with variables, routines, types, macros, and other identifiers set, acquires a conceptual term from the name of the program identifier. I decided to do it.

According to a second aspect of the present invention, in the first aspect of the invention, a conceptual term dictionary in which abbreviated expressions frequently used for the names of program identifiers are preset together with the original expressions is provided, and the conceptual term dictionary is provided. , The abbreviation is detected from the name of the program identifier, and the detected abbreviation is restored to the original expression to be used as a conceptual term.

According to a third aspect of the present invention, in the first aspect of the present invention, the sequence of lexical phrases extracted from the name of the variable is a noun concept term, and the lexical sequence extracted from the name of the routine is a noun conceptual concept. I tried to use a set of terms and verb-like conceptual terms.

According to a fourth aspect of the invention, in the invention according to the third aspect, the connectives and prepositions of the lexical sequences of the noun-like conceptual terms are distinguished to recognize the dependency relations, and the lexical relations are recognized in accordance with the dependency relations. The order is changed to the normal expression of noun phrases.

According to a fifth aspect of the invention, in the third aspect of the invention, a concept term dictionary in which a verb frequently used for the name of the program identifier is preset is provided, and the lexical sequence obtained from the routine name is arranged. When a verb set in the conceptual term dictionary is detected from this, this lexical phrase is made to be a verb-like conceptual term, and other lexical phrases are made to be noun-like conceptual terms.

According to a sixth aspect of the present invention, in the invention according to the first aspect, the similarity of the concept terms is measured based on the phenomenon of simultaneous appearance of the concept terms in an expression of a program or an assignment statement. An image showing the relationship between a plurality of conceptual terms is generated by the dimension scaling method.

According to a seventh aspect of the present invention, in the first or sixth aspect of the invention, the similarity of the conceptual terms is measured based on the phenomenon of the relation between the type and the typed one in the program. An image showing the relationship between a plurality of conceptual terms was generated by the multidimensional scaling method.

The invention according to claim 8 is the invention according to claim 1, 6 or 7, wherein the similarity of the conceptual terms is measured based on the phenomenon of simultaneous appearance of the conceptual terms in the file structure of the program. In this way, an image showing the relationship between a plurality of conceptual terms is generated by the multidimensional scaling method.

According to a ninth aspect of the present invention, in the first aspect of the invention, a measure of the effect of hiding the conceptual term acquired from the name of the external reference of the routine is defined as the similarity of the routine, and the cluster is defined by the similarity. The analysis was made to build a hierarchical cluster of routines based on the concealment of conceptual terms.

According to a tenth aspect of the invention, in the invention of the ninth aspect, the degree of distribution of the conceptual terms in the routine is quantified as an entropy in the routine space, and as a reduction amount of the entropy when the two routines are clustered. I tried to measure the similarity between two routines.

According to an eleventh aspect of the present invention, in the first aspect of the invention, conceptual terms are made to correspond to an object, an attribute, and a link that construct an information model of a program.

According to the twelfth aspect of the invention, in the eleventh aspect of the invention, as a rule for identifying a conceptual term in an object, the conceptual term acquired from a higher-order module is set as an object candidate.

According to a thirteenth aspect of the present invention, in the invention according to the eleventh aspect or the twelfth aspect, as a rule for identifying a conceptual term in an object, the conceptual term acquired from the name of the composite type is set as a candidate for the object.

The invention according to claim 14 is the invention according to claims 11 and 1.
In the invention described in 2 or 13, as a rule for identifying a conceptual term in an object, the conceptual term acquired from the names of global variables applied to a plurality of files is set as a candidate for the object.

The invention according to claim 15 is the invention according to claims 11 and 1.
In the invention described in 2, 13, or 14, as a rule for identifying a concept term as an attribute, when a concept term acquired from a composite type name is an object, the concept term acquired from a composite type subfield name is an attribute. I made it a candidate.

The invention according to claim 16 is the invention according to claims 11 and 1.
In the invention described in 2, 13, 14 or 15, as a rule for identifying a conceptual term in a link, a verb-like conceptual term acquired from the name of a routine is used as a link candidate, and the direction of this link is performed within the routine. The decision is made based on the information about the operation of the specified argument and the verb case control information preset in the conceptual term dictionary.

According to a seventeenth aspect of the present invention, there is provided a program analysis device for acquiring information for understanding the contents of a program formed by a hierarchical structure of modules in which identifiers such as variables, routines, types and macros are set. , The term generation means for acquiring the concept term from the name of the program identifier is provided.

[0024]

According to the first aspect of the invention, the concept term is acquired from the name of the identifier of the program, thereby acquiring the concept term as the information of the abstract level which can support the understanding of the content of the program.

According to the second aspect of the invention, when an abbreviated expression is present in the name of the program identifier, the abbreviated expression is restored to the original expression to make it a conceptual term, so that the conceptual term can be easily understood. And

According to the third aspect of the present invention, the sequence of lexical terms extracted from the name of the variable is used as a noun-like conceptual term, and the lexical sequence extracted from the name of the routine is defined as a noun-like conceptual term and a verb-like conceptual term. With the set of, the form of the conceptual term is easily predicted depending on the type of the identifier.

The invention according to claim 4 arranges the lexical sequences of the noun-like conceptual terms into the normal expressions of the noun phrases in accordance with the dependency relation, thereby making the conceptual terms easy to understand.

According to the fifth aspect of the present invention, the verbs in the lexical sequence obtained from the name of the routine are verb-like conceptual terms, and the other lexical terms are noun-like conceptual terms. And a noun-like conceptual term are easily distinguished.

According to the sixth aspect of the present invention, the relationship between a plurality of conceptual terms is shown by a multidimensional scaling method based on the similarity of the conceptual terms measured based on the phenomenon of simultaneous appearance of the conceptual terms in a program expression or an assignment statement. By creating an image, we will make it easier to understand the interrelationship of conceptual terms.

The invention according to claim 7 is an image showing a relationship between a plurality of concept terms by a multidimensional scaling method based on the similarity of the concept terms measured based on the phenomenon of the relationship between the type and the typed one in the program. By creating the, the relationship between the conceptual terms is made easy to understand.

According to an eighth aspect of the present invention, the similarity of the concept terms is measured based on the phenomenon that the concept terms appear at the same time in the file structure of the program. By generating the image shown, the relationship between the conceptual terms is made easy to understand.

According to the ninth aspect of the present invention, the hierarchical analysis of the routine based on the hiding of the conceptual term in the cluster analysis is performed by the similarity of the routine defined by the measure of the effect of hiding the conceptual term obtained from the name of the external reference of the routine. By constructing a cluster, we obtain a hierarchical cluster that reflects the module structure of the program as abstract level information that can support understanding of the program content.

According to a tenth aspect of the present invention, the degree of distribution of conceptual terms in a routine is quantified as entropy in the routine space, and the similarity between the two routines is measured as the amount of entropy reduction when the two routines are clustered. By doing so, the module structure of the program is reflected well in the hierarchical cluster acquired as the information of the abstract level that can support the understanding of the content of the program.

According to the eleventh aspect of the present invention, the information for constructing the information model is acquired from the conceptual terms by associating the conceptual terms with the objects, attributes and links that construct the information model of the program.

According to the twelfth aspect of the present invention, the conceptual terms acquired from the higher module are used as object candidates, so that the objects for constructing the information model are easily acquired.

According to the thirteenth aspect of the present invention, the conceptual term acquired from the composite type name is used as an object candidate to easily acquire the object for constructing the information model.

According to the fourteenth aspect of the present invention, the concept terms acquired from the names of the global variables applied to the plurality of files are candidates for the object, so that the object for constructing the information model is easily acquired.

According to the fifteenth aspect of the present invention, when the concept term acquired from the name of the composite type is used as an object, the concept term acquired from the name of the subfield of the composite type is used as an attribute candidate, whereby the information model is obtained. Easily get the attributes of the object that builds the.

According to the sixteenth aspect of the present invention, a verb-like conceptual term acquired from the name of the routine is used as a link candidate, and the direction of this link is information regarding the operation of the argument performed inside the routine and the conceptual term dictionary. The link between the objects that construct the information model is easily obtained by making a decision based on the verb case control information set in advance.

According to the seventeenth aspect of the present invention, the term generation means acquires the concept term from the name of the program identifier, thereby acquiring the concept term as information at an abstract level capable of supporting the understanding of the content of the program.

[0041]

An embodiment of the present invention will be described below with reference to the drawings. First, the program analysis device (not shown) of the present embodiment acquires abstraction level information for understanding the contents of the program. Therefore, the program analysis method of the present embodiment uses the concept term from the name of the program identifier. Is to be acquired. The term “conceptual term” is a term that represents a concept useful for understanding a program, such as a tangible object, a role, an event, an interaction, a specification, and the like.

Therefore, a program for which such a conceptual term is acquired will be described. In a general program, identifiers such as variables, routines, types, and macros are set at important points, and they are formed in a hierarchical structure of modules. The identifier is also called a variable or routine name, and there are restrictions on the length of the character string and the type of characters that can be used. For example, in C language, as illustrated below, 1. Variable name (eg NumOfPartTimeEmployee) 2. Function name (eg CalcPaymentOfEmployee) 3. Type name of structure or union (eg struct person) 4. Member name of structure or union (eg p-> age, p-> name, p-> sex) 5. Type name given by typedef (eg person_t) 6. Macro (and its argument) (eg MAX_EMPLOYEES_NUM) 7. There are seven types of identifiers, such as the destination label of a GOTO sentence (eg goto error :).

The identifier name of such a program generally reflects the contents of the problem area to which the identifier is added. Therefore, the program analysis method and apparatus of the present embodiment targets a routine identifier, a variable identifier of its argument, a global variable identifier, and a program-defined type identifier, and creates conceptual terms from the names of these identifiers. To earn.

In actual programs, it is not uncommon for an identifier name to be inappropriate or to be omitted. For this reason, it is difficult to completely automate the process of acquiring the concept term from the name of the identifier, and the justification when automated is not guaranteed. Therefore, when the program analysis device is actually manufactured, it is desirable to form the program analysis device so that the processing proceeds in an interactive manner with the analyst.

The program analysis device of this embodiment is a RAM (Random Access Memory) that stores various types of information in an updatable manner.
A device such as a device has a conceptual term dictionary, and a CPU (Centr) that executes various processes by referring to the conceptual term dictionary.
al Processing Unit) and other devices are also provided.

Since basic conceptual terms covering the concepts of programming and computer environment are preset in the conceptual term dictionary, verbs, prepositions, and conjunctions often used in the names of program identifiers are These terms are also set with similar words, abbreviations, grammatical information such as case dominance and part-of-speech, and the like. Therefore, the abbreviated expressions that are often used in the names of program identifiers are also set in advance along with their original expressions, and the verbs that are often used in the names of program identifiers are also set in advance along with their case control information. . In addition, this conceptual term dictionary is
The content is updated as the program analysis work progresses.

A program analysis method by the program analysis apparatus of this embodiment will be schematically described with reference to FIG.
First, a concept term is acquired from the name of a program identifier, a concept term dictionary is generated, the similarity of concept terms is measured, and a concept network is generated as an image showing the relationship between concept terms. Next, in order to let the analyst identify the module structure of the program, the routine is cluster analyzed by the degree of distribution of the conceptual terms, and further, in order to let the analyst construct the information model of the program, the conceptual terms are linked with the objects and attributes. And correspond to.

Therefore, the program analysis method of such a program analysis device will be described step by step below.

First, when acquiring a conceptual term from the name of a program identifier, the lexical detection means extracts a lexical representing a concept such as a tangible object, a role, an event, an interaction, and a specification from the name of the program identifier. Then, when the abbreviation detecting means detects the abbreviation from the extracted token,
The original expression restoring means restores the abbreviated expression to the original expression by referring to the conceptual term dictionary.

If the lexical identifier is a variable, type or member as described above, the term identifying means
The lexical sequence extracted from the name is a noun-like concept term. On the other hand, when the identifier is a routine, the sequence of lexical words extracted from the name is a set of noun-like conceptual terms and verb-like conceptual terms.

Further, when a noun-like conceptual term is obtained as described above, the syntactic analysis unit discriminates a conjunction or a preposition in a lexical sequence, recognizes a dependency relation, and recognizes the dependency relation according to this dependency relation. The expression restoring means rearranges the sequence of to the regular expression of the noun phrase.

As described above, the conceptual terms can be obtained in a form that is easy to understand from the names of identifiers such as variables and routines. The conceptual terms thus obtained are set in the conceptual term dictionary.

A specific example of acquiring a concept term from the name of the program identifier will be described below. First, if the variable identifier name is “max points of team”,
As illustrated below: Extract lexical name from identifier name → (max points of of
team) 2. Restore the lexical abbreviation to the original expression → (maximum point
s of team) 3. Recognize conceptual terms from lexical words → Acquire conceptual terms in the step of concept: (maximum team point).

A concrete example of acquiring a conceptual term from a program by such steps is as follows: program → concept term int CalcMonthlytotal (int NewPurchases) → VC: caluclate NC: (monthly total) {int Monthlytotal; → NC: (monthly total) Monthly total = NewPurchases + SalesTax (NewPurchases); → NC: (monthly total), NC: (new purchases), NC: (sales tax) retern (MonthlyTotal)}.

Therefore, the contents of the above-mentioned steps for obtaining the conceptual term from the name of the identifier will be sequentially described in detail below.

First, as step 1 of the process of acquiring the conceptual term from the name of the identifier, the case of extracting a token from the name of the identifier will be described.

Generally, the name of the program identifier is
It consists of several lexical sequences. For example, when the program is written in the C language, breaks in the lexical name of the identifier can be easily recognized as a delimiter by a hyphen, a change from a lowercase letter to an uppercase letter, or the like. Therefore, since the name of the identifier is decomposed into individual tokens by using the characteristics of the programming language as described above, specifically, Example 1 win_hght_siz → win hght siz) Example 2 CalcMonthlyPymntOfEmple → calc monthly pymnt of emple And so on.

Next, as step 2 of the process of acquiring the conceptual term from the name of the identifier, the process operation of restoring the abbreviation of the lexical phrase extracted from the name of the identifier to the original expression will be described.

Generally, abbreviated expressions are often used in the names of program identifiers, and such abbreviated expressions include abbreviations. The abbreviation is formed, for example, by removing vowels from the character string of the original expression or by only the first few characters of the original expression. Here, abbreviations that are often used in identifier names but are abbreviated expressions that are inappropriate as conceptual terms are win (= window), scr (= screen), del (= delete), prev
It is like (= previous).

Therefore, if an abbreviation such as an abbreviation exists in the lexical name extracted from the identifier name, the abbreviation is restored to the original expression by referring to the conceptual term dictionary. First, the lexical phrase extracted from the name of the identifier is spell checked, and an abbreviation that matches the detected lexical phrase is searched from the conceptual term dictionary. Therefore, when a matching abbreviation is searched for, abbreviations such as Example 1 win hght siz → window height size Example 2 calc monthly pymnt of emple → calculate monthly payment of employee are obtained as shown below. The abbreviation is restored to the original expression set in the concept term dictionary.

If an abbreviation that matches a lexical phrase is not retrieved from the conceptual term dictionary, this lexical item is displayed on a display together with a message, and the lexical phrase is restored by an analyst's keyboard input. In addition, it is preferable to let the analyst restore the expression even when the matching abbreviations conflict due to the variety of abbreviations (eg, calculate → cal, calc, calcu, clc ...). Further, as temporary variables and loop control variables, meaningless variables such as i, x, p, etc. may be used, but these variables are discarded without being targeted for restoration.

Next, as a step 3 of the process of acquiring the conceptual term from the name of the identifier, the processing operation of recognizing the conceptual term by rearranging the lexical words will be described below.

First, in general, a name given to an identifier by a programmer reflects a concept belonging to the problem area and a concept relating to design and implementation. These are noun-like conceptual terms and verb-like conceptual terms. Will be expressed as Here, noun-like conceptual terms indicate things, states, times, etc. that accompany adnominal modification (eg, top level window), and verb-like conceptual terms include physical positions, shapes, quantities, etc., or states. Indicates generation, change, disappearance, etc.

The name given to the identifier by such a concept can be predicted whether the concept term is noun or verb, depending on the kind of the identifier. That is, since variables and types are generally given noun names, it can be expected that the conceptual terms obtained from the names are nouns. In addition, since the routine is generally given a name that expresses the action or action of the process, it can be expected to be a set of noun-like conceptual reading and verb-like conceptual term.

Therefore, in the program analysis device of the present embodiment, when the identifier discriminating means discriminates that the identifier is a variable, a type or a member, the term recognizing means takes the noun concept of the arrangement of the tokens extracted from the name. Recognize as a term. Further, when the identifier discriminating means discriminates that the identifier is a routine, the term recognition means recognizes the lexical sequence extracted from the name as a set of a noun-like conceptual term and a verb-like conceptual term.

Although the program is generally written in English, since the name of the identifier has a limited number of characters, a natural language syntax rule cannot be established. However, since connectives and prepositions can be identified relatively easily, it is possible to rearrange lexical words of noun conceptual terms into ordinary expressions by using this to recognize lexical dependency relationships. .

Therefore, when the noun-like conceptual term is acquired from the name of the identifier as described above, the syntactic analysis unit recognizes the dependency relation by discriminating the connective or preposition of the lexical sequence of the noun-like conceptual term. Then, the expression restoration means identifies the adnominal modifier and the noun from the dependency relationship and rearranges the lexical sequence into the normal expression of the noun phrase.

Further, in the program analysis device of this embodiment, since the verbs that are frequently used as the names of the program identifiers are preset in the conceptual term dictionary, the verb detection means arranges the lexical sequences acquired from the routine names. When the verb set in the conceptual term dictionary is detected from the vocabulary, the term determining means recognizes the lexical term detected as the verb as a verb-like conceptual term and recognizes other lexical terms as a noun-like conceptual term.

Here, the formats of the noun-like conceptual term and the verb-like conceptual term are as follows: Noun-like conceptual term format: NC: (adnominal modifier 1, adnominal modifier 2, ... noun) Verb-like concept Term format: VC: Verb etc.

Further, an actual example of recognizing a sequence of tokens acquired from the name of the identifier as a conceptual term is shown below. In addition, the example shown here is a case in which the lexical sequence of the concept terms obtained by restoring the abbreviations of the concept terms to the original expressions in step 2 is further recognized as the noun concept terms and the verb concept terms. Yes, Example 1 (window height size) → NC: (window height si)
ze) Example 2 (calculate only payment of em)
playee) → VC: calculate, NC: (employee payment), etc.

As described above, since the concept term can be obtained from the program identifier, the concept term thus obtained is set in the concept term dictionary, and as shown in FIG. A conceptual network is generated as an image showing the relationship between conceptual terms by measuring the degree of similarity.

The concept network is a network in which each of a plurality of concept terms obtained from a program is used as a node and a link is formed based on the semantic relation of these concept terms. Then, in order to construct such a concept network, the program analysis device of the present embodiment measures the similarity of the concept terms by the similarity measuring means and generates an image by projecting the similarity of the concept terms on a plane. By means of a multidimensional scaling method. In addition, this multidimensional scaling method is
It is a statistical method for finding a placement in which the degree of similarity between objects is expressed as the closeness of the positions of points representing the objects.

Therefore, the method for generating the concept network from the concept terms in this way will be described below in order. First, since it is necessary to measure the similarity of the conceptual terms, the program analysis device of this embodiment measures the similarity of the conceptual terms by the following three types of methods.

First, in the first method of measuring the similarity,
The similarity of conceptual terms is measured based on the phenomenon of simultaneous appearance of conceptual terms in program expressions and assignment statements.

That is, since a plurality of variables described together in a program expression or assignment statement generally have a dependency relationship, the conceptual terms obtained from these variables also have a semantic relevance. It is expected that. The same applies when a function is described in addition to a variable.

For example, "MonthlyTotal = NewParchases +"
From the variables and functions of the assignment statement (assignment expression in C language) such as “SalesTax (NewParchases)”, NC: (monthly tota
Three conceptual terms, l), NC: (new parchases) and NC: (sales tax), are acquired, and it is expected that these conceptual terms are semantically related.

Therefore, the program analysis apparatus of the present embodiment investigates the phenomenon of the simultaneous appearance of the conceptual terms in the expressions and assignment statements as described above by the appearance detection means in the entire program, and totalizes and calculates them. By processing, the semantic similarity of conceptual terms is measured. That is, the degree of similarity between the conceptual terms T1 and T2 that appear simultaneously in an expression or an assignment statement is an expression in which the conceptual terms T1 and T2 are the same for all expressions (calculations and logical expressions) and assignment statements included in the program. And the number of occurrences in assignment statements. Therefore, as illustrated below, this equation becomes a similarity matrix of n rows and n columns: | pij | = ΣOccurrence1 (sk, ci, cj) {k: 1 ... m}. In this mathematical expression, n is the total number of conceptual terms obtained from the program, m is the total number of expressions and assignment statements of the program, pij is the element at the i-th row and j-th column of the similarity matrix, sk is the k-th expression or assignment statement of the program. ci: i-th conceptual term cj: j-th conceptual term Occurrence1 (sk, ci, cj): "c" in the expression or assignment statement
When the concept terms "i, cj" appear at the same time, it becomes "1",
Other than this, the function is "0".

Next, in the second method of measuring the degree of similarity, the degree of similarity of the conceptual terms is measured based on the phenomenon of the relation between the type and the typed one in the program. In other words, the variables of the program have an attribute called type, which determines the meaning of the in-memory value that represents the variable, so it is expected that the meanings of each of these variables and types will be related. It Similarly, since the return value of a function is also designated with a type, it is expected that the meanings of each of these functions and types are related. Especially in the case of complex types, types are often associated with a given target, and from a name perspective, it is expected that there will be a correlation between the name of the type and the name of the typed one. .

For example, from the types illustrated below and typed variables, an example of variable declaration: struct xy_coodes PreviousMousePoin
The concept term NC: (previous mouse point) of type t; and the concept term NC: (xy codinate) of variable are acquired, but it is expected that these concept terms are semantically related.

Therefore, the program analysis apparatus of the present embodiment investigates the phenomenon of the relationship between the above type and the typed one with respect to the entire program by the appearance detection means, and totalizes this to perform arithmetic processing. By doing so, the semantic similarity of the conceptual terms is measured. That is, the similarity between the concept terms T1 and T2 based on the phenomenon of type and typed thing is that the concept terms T1 and T2 are the same for all expressions (calculation expressions and logical expressions) and assignment statements included in the program. It is measured by the number of occurrences in the expression and assignment statement. As exemplified below, this mathematical expression is a similarity matrix of n rows and n columns: | pij | = ΣOccurrence2 (dk, ci, cj) {k: 1 ... m}. In this mathematical expression, n: the total number of conceptual terms obtained from the program, m: the total number of expressions and assignment statements of the program, pij: the element in the i-th row and j-th column of the similarity matrix, dk: the k-th variable or function of the program Total number of declarations ci: i-th conceptual term cj: j-th conceptual term Occurrence2 (dk, ci, cj): When the conceptual terms of "ci, cj" appear at the same time in the declaration of variables or functions of "dk", "1"", And other than this, it is a function that becomes" 0 ".

When the program is based on object-oriented programming, class objects correspond to types and instances correspond to types.

In the third method of measuring the similarity, the similarity of the conceptual terms is measured based on the phenomenon that the conceptual terms simultaneously appear in the file structure of the program. That is, according to the file structure of the program, it is expected that the concept terms that exist at the same time in the same function, the same file, the same directory, and the like are related. In general, the routines in a file are often set as modules that realize the same function, and this also applies to files and directories.

Therefore, the program analysis apparatus of the present embodiment investigates the phenomenon of simultaneous appearance of the conceptual terms in the file structure of the program as described above by the appearance detection means for the whole program, and totalizes and calculates this. By processing, the semantic similarity of conceptual terms is measured. That is,
The similarities of the conceptual terms T1 and T2 based on the phenomenon that the conceptual terms in the file structure of the program appear at the same time have high values in the order of simultaneous appearance in the same function, simultaneous appearance in the same file, and simultaneous appearance in the same directory. West,
Furthermore, by summing up all such simultaneous occurrences of conceptual terms, an n-by-n similarity matrix: | pij | = ΣOccurrence3 (fk, fl, ci, cj) {k: 1 ... m} {l The similarity of conceptual terms is measured as: 1 ... m}. In this mathematical expression, n is the total number of conceptual terms obtained from the program, m is the total number of functions defined in the program, pij is the element at the i-th row and j-th column of the similarity matrix, fk is the k-th element defined in the program. Function fl: l-th function defined in the program ci: i-th conceptual term cj: j-th conceptual term Occurrence3 (fk, fl, ci, cj): “ci, cj” conceptual term is “fk, Appear simultaneously in the function of "fl" and "fk, fl"
Are "3" if they are the same function, "2" if "fk, fl" are included in the same source file, and "1" if "fk, fl" are included in the source files on the same directory. ", Otherwise it is a function that becomes" 0 ".

As described above, since the similarities of the conceptual terms are measured as three types of matrices, the program analysis apparatus of the present embodiment uses the image forming means with each of the three types of similarity matrices as an input. By performing a multidimensional scaling method on
As illustrated in FIG. 3, an image in which the conceptual terms are plotted on a plane is generated, and this image is output by displaying on a display (not shown) or printing on a printer (not shown). Also,
A similar image can be generated from a matrix obtained by linearly combining three matrices having similarities.

The image thus generated corresponds to the above-mentioned concept network and two-dimensionally expresses the relationship between the concept terms of the program, so that it is possible to assist the analyst in understanding the content of the program. .

Further, in the program analysis device of this embodiment, the configuration analysis means performs the cluster analysis of the routine on the degree of distribution of the conceptual terms, and provides the result to the analyst as abstract level information for supporting the understanding of the contents of the program. To do.
Therefore, this analyst will identify the module structure of the program according to the acquired information.

Here, the program targeted by the program analysis apparatus of the present embodiment is formed in a hierarchical structure of modules as illustrated in FIG. This module is a set of routines that realize a predetermined function. In order for the analyst to understand the content of the program, it is important to identify the module and recognize the hierarchical structure, and therefore it is necessary to identify the module configuration of the program.

First, in designing a module in a program, information hiding is generally considered important, and it is desirable to hide detailed information in the module. The module manufactured in this way is
It can be expected to include unique conceptual terms. For example, external references to routines include global variables, externally defined types, routine invocations, etc., but the set of routines that form a module defines the terminology acquired from external references to those routines. It can be assumed to be a set of routines that conceal well.

Therefore, in the program analysis device of this embodiment, the measure of the effect of concealing the conceptual term acquired from the external reference name of the routine is defined as the similarity of the routine by the similarity measuring means. Then, the analysis executing means executes the cluster analysis based on the similarity, thereby constructing a hierarchical cluster of routines based on the concealment of the conceptual term. This cluster hierarchy serves as information that suggests the module structure and the like existing in the program.

The cluster analysis is often used for the problem of organizing the target data into a meaningful structure and developing the classification. In cluster analysis, target data are combined based on similarity and distance to form a cluster, and finally a hierarchical tree diagram of the cluster is generated. In the process, all two target data sets are combined. The degree of similarity is quantified. This definition of similarity affects the meaning of the result of the cluster analysis, so it is necessary to give a definition that reflects the purpose.

Therefore, in the program analysis device of this embodiment, the degree sum distribution means quantifies the degree of distribution of conceptual terms in the routine as entropy in the routine space, and the similarity measurement means clusters the two routines. The similarity between the two routines is measured as the amount of entropy reduction in the case. Thus, the process of cluster analysis corresponds to the process of optimally reducing the clutter of the distribution of concept terms in the routine space, so that the concept terms obtained from the names of routine external references are well localized in clusters. It is possible to construct a hierarchical cluster of routines rooted in information hiding in the design of programs.

Here, in the program analysis apparatus of this embodiment, an entropy measuring method by the degree totalizing means,
The routine similarity measurement method by the similarity measurement means will be described in detail below. First, for an event in which a conceptual term appears in a routine that forms a module, the occurrence probability P (atr
_insi, ej) is

[0093]

[Equation 1]

It is calculated as In this formula, atr_insi: i-th conceptual term ej: j-th routine NumEty: total number of routines in the program NumAtrIns: total number of conceptual terms obtained from external reference in the program NumOfOccur (atr_insi, ej): atr_insi in ej Is the number of appearances of.

Then, the entropy H (atr_insi, [e ₁ , e ₂ , ..., e _NumEty ]) of the conceptual term in all the routines is

[0096]

[Equation 2]

It is calculated as

Next, the distance Di between the two routines ep and eq
stance (ep, eq) is the difference between the entropy of the conceptual term before clustering the two routines ep and eq, and the entropy of the conceptual term after clustering,

[0099]

(Equation 3)

It is calculated as follows.

Then, in this way, the two routines ep
Since the degree of similarity between the two routines ep and eq is calculated as the reciprocal of the distance between eq and eq, such degree of similarity is calculated for all combinations of routines.

In this way, the similarity of the routines that form the modules of the program is measured, and the similarity of this routine is used to determine the hierarchical cluster of routines based on the concealment of conceptual terms in the cluster analysis. Generate as. The process of the cluster analysis includes three steps of generating an initial cluster set, selecting a clustering candidate, and updating the cluster set.

Therefore, in the first step of cluster analysis, the initial cluster set is formed by all the routines to be clustered. Next, in the second step, two elements (initially, routines) included in the cluster set are selected as a set and registered as a candidate set for clustering. Then, in the third step, a cluster including the same element is selected from the registered candidate set, this cluster set is held as data, and the contents of the candidate set are reflected in the cluster set and updated. Therefore, by repeating the second step and the third step until the candidate set becomes empty and aggregating the cluster set held as data, as shown in FIG. 4, the hierarchical cluster of the routine has a hierarchical tree shape. Generated as a figure.

As described above, the program analysis apparatus of this embodiment generates the hierarchical tree diagram of the routine, and outputs it by displaying on the display or printing by the printer.
The image thus output reflects the hierarchical structure of the modules of the program, so that it is possible to assist the analyst in understanding the contents of the program.

Next, in the program analysis apparatus of this embodiment, the model generation means associates the conceptual terms with the objects, the attributes, and the links, and provides them to the analyst as abstract level information indicating the contents of the program. Therefore, this analyst will construct an information model of the program by using the acquired objects, attributes and links.

Here, the information model of the program will be described. First, this information model is an S. Schreier and S. J. It is proposed by Meller and expresses the problem area targeted by the program mainly in data. For example, it is called an object model in an object-oriented program.

The purpose of the information model is to extract from the program an entity group that constitutes the program, that is, an object group such as tangible objects, roles, events, interactions, and specifications. An object is an abstraction of a set of things in the real world, and all the things that belong to this set have the same property, and all instances follow the same rules and rules. Is pleased.

Furthermore, the information model includes expressions such as attributes and links in addition to objects.
It is a characteristic of an object, and is an abstraction of one property owned by all entities that are abstracted as one object. A link is an abstraction of a set of relationships that systematically holds between different kinds of things in the real world, and is often expressed as a verb in a problem description in a program (example: work), implemented as a pointer from object to object (eg company → employee).

Therefore, the method by which the program analysis device of this embodiment constructs the information model by the model generation means will be sequentially described below.

At this stage, all the conceptual terms extracted from the names of the program identifiers are set in the conceptual term dictionary, but in order to identify the conceptual terms in the object in the information model, this conceptual term is used. The term is a conceptual term related to the problem area of the program, and it is necessary to satisfy the condition as an object.

However, since all the conceptual terms set in the conceptual term dictionary are directly extracted from the program as described above, the conceptual terms related to the implementation of the program design and the conceptual terms that are too detailed are included. Therefore, as described above, the conceptual terms related to the implementation and the conceptual terms that are too detailed are excluded, and only the conceptual terms related to the problem area targeted by the program are extracted.

Here, an example of the conceptual terms relating to the implementation that needs to be excluded and an example of the conceptual terms relating to the problem area that needs to be taken out are respectively illustrated below. The conceptual terms relating to the implementation are (binary search) ( tree structure) (quick sort), the conceptual terms for the problem area are (employee work time) (student credit) (airplane
altitude).

That is, in order to identify such a conceptual term as an object in the information model, it is necessary to determine whether the conceptual term is a conceptual term related to the problem area of the program and whether the condition as the object is satisfied. Need to judge.

Therefore, the program analysis apparatus according to the present embodiment uses, as a rule for identifying a conceptual term as an object,
The first to third HRs (Heuristic Rul)
e) is set in rule storage means such as RAM.

Since these HR are not rules for automatically generating the final conclusion but rules for supporting the analyst's judgment, the program analysis device performs processing in an interactive manner. It is formed to proceed. In this case, the program analysis device selects a conceptual term that is a candidate for an object by the HR and displays it on the display. Therefore, the analyst determines whether or not to identify the conceptual term as an object and designates it by input operation on the keyboard. Will be done.

First, as the first HR, the conceptual terms acquired from the upper module are used as object candidates.

That is, since a program composed of hierarchically structured modules is generally designed top down, the upper module realizes the upper design concept and the lower module realizes the lower design concept. ing. For this reason, the lower-level module often implements a part that depends on the hardware and a more detailed design such as the implementation of an abstract level data structure. Often related to problem areas.

Therefore, it is expected that the conceptual term acquired from the higher-level module is a conceptual term related to the problem area of the program, and therefore this is provided to the analyst as a candidate for the object.

Next, as the second HR, the conceptual term acquired from the name of the composite type is a candidate for an object.

That is, the complex type defined in the program generally represents the information of the target in the real world. Therefore, the conceptual term acquired from the name of the composite type can be a candidate for an object, and therefore this is provided to the analyst as a candidate for the object. Here, exemplifying the complex type defined in the program and the conceptual term acquired from this complex type, the complex type defined in the program is struct person {int age; int name; int sex ;} ;, the concept term acquired from this complex type is concept: (person).

Further, as the third HR, the conceptual term acquired from the names of global variables applied to a plurality of files is a candidate for an object.

That is, it is generally said that it is not desirable to use a global variable as data to be exchanged between modules in order to strengthen the coupling between the modules. However, global variables may be used for data whose existence is broadly defined in a program or for data hidden in a module. Therefore, the global variables applied to multiple source files can be regarded as an object of a rather large unit.
This is provided to the analyst as a candidate for the object.

With the first to third HRs as described above, the program analysis apparatus of this embodiment provides the analyst with the conceptual terms that are object candidates. You will be identified as an object.

If the attributes of the objects thus obtained and the links between the objects can be obtained, the information model of the program can be constructed. Therefore, in the program analysis device according to the present embodiment, a rule that associates a concept term with an attribute and a rule that associates a concept term with a link are also set in a rule storage unit such as a RAM. Therefore, a process of associating a predetermined conceptual term with an attribute and a link according to such a rule will be sequentially described below.

First, as a rule for associating a concept term with an attribute, when a concept term acquired from a composite type name is an object, a concept term acquired from a composite type subfield name is used as an attribute candidate. .

That is, when the composite type includes a subfield, this subfield is a composite type "a-part-of".
Therefore, if the concept term acquired from the name of the composite type is an object, the concept term acquired from the name of the subfield of the composite type can be an attribute of the object.

Here, exemplifying the complex type defined in the program and the conceptual term acquired from the name of the subfield of this complex type, the definition of the complex type is struct person {int age; int name; int sex;} ;, the conceptual terms are age, name, sex. Therefore, the above-mentioned conceptual terms (age, nam
e, sex) is a candidate for the attribute of the object (person).

As described above, since the concept term can be provided to the analyst as a candidate for the attribute, it can be identified as the attribute by the judgment of the analyst. Then, when the object and the attribute can be acquired in this way, the information model can be constructed after that if the link can be acquired.

Therefore, as a rule for associating a concept term with a link, a verb-like concept term acquired from the name of a routine is used as a link candidate. At this time, the direction of the link is determined by the link generation means based on the information about the operation of the argument performed inside the routine and the verb case control information preset in the concept term dictionary.

That is, it is assumed that the verb-like expression set as the name of the routine of the program indicates a link between objects. Therefore, this will be sequentially described below. The following is an example of the routine name and arguments and the conceptual terms obtained from the routine name.If the routine name and arguments are add_new_customer (table, customer_name, customer_id), the routine name is The acquired conceptual term is (VERB: add CONCEPT: (new customer)).

Here, the concept term dictionary used for identifying such verb-like concept terms will be described. As described above, since there are not many types of verbs that are frequently used in program names, it is possible to predict this and set it in the concept term dictionary. Further, in this conceptual term dictionary, information on meaning and syntax based on grammar is also set as case control information for each set verb. This case control information is the type of deep case that a verb can have,
The range and the essentiality that can be taken, and from the viewpoint of data manipulation on the program, is it simply referred to or is it subject to change such as value substitution?
It consists of such information. Here, "add" set as a verb-like concept term in the concept term dictionary is illustrated below together with its case dominance information: (add (abbreviation ad) (similar word increment) (syntactic part-of-speech verb) (case dominance information) ((Deep case subject) (Noun meaning marker object ...) (No necessity) (Refer to data manipulation) (Link source)) ((Deep case object) (Noun meaning marker object ...) (Required) (Data manipulation change) (link destination)) ((deep case receiver) (noun meaning marker object ...) (required) (data manipulation reference) (link source)).

Therefore, a method of finding an object associated with a link by using the case dominance information set in the concept term dictionary together with the verb will be described below.

First, the variables of the input and output of the routine are
By parsing and clarifying with the variable acquisition means,
Get information about argument manipulations done inside the routine. The input variable is a dummy argument or a global variable of the routine, and indicates a variable simply referred to in the routine. The output variable is a dummy argument or a global variable of the routine, and indicates a variable whose contents are changed in the routine.

Next, the case governing information of the verb-like concept term acquired from the name of the routine is read from the concept term dictionary, and the contents of the information regarding the operation of the argument are referred to by referring to the necessity and data operation. judge. Therefore, if there is no essentiality and the data operation is set as reference, it is judged as the subject of the verb, and if it is essential and the data operation is set as change, it is judged as the target of the verb and there is essentiality. If the data operation is set to reference, it is determined to be the recipient of the verb.

Here, an example of the name and argument of the routine and three conceptual terms acquired from the argument of the routine will be illustrated below. When the example of the routine and argument is add_new_customer (table, customer_name, customer_id) , The conceptual term obtained from the routine argument is CONCEPT: (table) CONCEPT: (customer name) CONCEPT: (customer id).

Therefore, the routine "add_new_customer"
When parsing, the data operation of the argument “table” is changed, and the data operation of “customer name” and “customer id” is a reference. Therefore, based on this information and the case governing information of the verb "add", the direction of the link is: link source object: (table) link destination object: (customer name), (customer i)
d) can be obtained as

As described above, the verb-like conceptual term “add” acquired from the name of the routine is converted into the object “table”.
Since it can be provided to the analyst as a candidate for a link from "customer name" to "customer id",
This can be identified as a link at the discretion of the analyst.

As described above, the program analysis device of the present embodiment can associate the concept term with the object, the attribute, and the link, and analyze this as abstract level information that can support understanding of the content of the program. The analyst can build an information model of the program based on the acquired information.

[0139]

According to the invention described in claim 1, in the program analysis method for acquiring information for understanding the content from the program, the concept term is acquired from the name of the identifier of the program. Since the conceptual term can be acquired from the name of the identifier that reflects the content of the program as the information of the abstraction level, it is possible to easily assist the analyst in understanding the content of the program.

According to the second aspect of the invention, a conceptual term dictionary in which abbreviated expressions frequently used for the names of program identifiers are set in advance together with the original expressions is provided, and the conceptual term dictionary is referred to to identify the program identifiers. By detecting the abbreviation from the name and restoring the detected abbreviation to the original expression to make it a conceptual term, it is possible to make the conceptual term easier to understand. It has the effect of being able to favorably support content understanding.

According to the third aspect of the present invention, the sequence of words and phrases extracted from the name of the variable is used as a noun concept term, and the sequence of words and phrases extracted from the name of the routine is defined as a noun concept term and a verb concept term. By making it a set of
Since the form of the conceptual term can be easily discriminated from the type of the identifier, the subsequent processing for the conceptual term can be simplified and the efficiency of generating abstract level information indicating the content of the program can be improved. Have an effect.

The invention according to claim 4 recognizes a dependency relation by discriminating a conjunction or a preposition in a lexical sequence of a noun-like conceptual term, and recognizes the lexical sequence into a normal expression of a noun phrase according to the dependency relation. The rearrangement allows the conceptual terms to be in a form that is easy to understand, which has the effect of being able to favorably support the analyst's understanding of the program contents.

According to the invention described in claim 5, a concept term dictionary in which a verb frequently used for a name of a program identifier is preset is provided, and the concept term dictionary is set from a sequence of lexical words acquired from a routine name. When a verb is detected, this lexical phrase is used as a verb-like conceptual term, and other lexical phrases are also used as noun-like conceptual terms, so that a verb-like conceptual term and a noun-like conceptual term can be easily distinguished. As a result, the subsequent processing for the conceptual term can be simplified, and there is an effect that the generation efficiency of the information at the abstract level indicating the content of the program can be improved.

According to the sixth aspect of the present invention, the similarity of the concept terms is measured based on the phenomenon of simultaneous appearance of the concept terms in the expression of the program or the assignment statement, and by this similarity, a plurality of concept terms are calculated by the multidimensional scaling method. By generating the image showing the relationship between the terms, it is possible to make the mutual relationship of the conceptual terms into a form that is easy to understand, and thus it is possible to favorably assist the analyst in understanding the contents of the program.

The invention according to claim 7 measures the similarity of the concept terms based on the phenomenon of the relationship between the type and the typed one in the program, and by this similarity, a plurality of concept terms of the concept terms are calculated by the multidimensional scaling method. By generating the image showing the relationship, it is possible to make the mutual relationship of the conceptual terms into a form that is easy to understand, and thus it is possible to favorably support the analyst to understand the contents of the program.

According to the eighth aspect of the present invention, the similarity of the concept terms is measured based on the phenomenon that the concept terms simultaneously appear in the file structure of the program, and the relation between a plurality of concept terms is calculated by the multidimensional scaling method based on the similarity. By generating the image showing "," it is possible to make the mutual relationship of the conceptual terms into a form that is easy to understand, and thus it is possible to favorably support the analyst to understand the contents of the program.

According to the ninth aspect of the present invention, the measure of the effect of hiding the concept term acquired from the name of the external reference of the routine is defined as the degree of similarity of the routine, and based on the degree of similarity, the cluster analysis is based on the hiding of the term. By constructing a hierarchical cluster of routines, it is possible to obtain a hierarchical cluster that reflects the module structure of the program as abstract level information that can help understand the content of the program. It has the effect of being able to support understanding well.

According to the tenth aspect of the present invention, the degree of distribution of the conceptual terms in the routine is quantified as entropy in the routine space, and the similarity between the two routines is measured as the entropy reduction amount when the two routines are clustered. By doing so, the module structure of the program can be reflected well in the hierarchical cluster acquired as information at the abstract level that can support the understanding of the program content, so that the analyst can favorably understand the program content. Has the effect of being able to.

According to the eleventh aspect of the present invention, the concept terms are made to correspond to the object, the attribute and the link for constructing the information model of the program. Since you can get attributes and links,
Based on this, the analyst can construct an information model of the program, and has the effect of being able to favorably support the analyst's understanding of the program content.

According to the twelfth aspect of the present invention, as a rule for identifying a conceptual term in an object, the conceptual term acquired from a higher-order module is set as an object candidate. Since it can be easily acquired, it has the effect of being able to favorably support the analyst's understanding of the program contents.

According to the thirteenth aspect of the present invention, as a rule for identifying a conceptual term to an object, the conceptual term acquired from the name of the composite type is set as an object candidate, thereby constructing the information model of the program. Since it can be easily obtained, it has the effect of being able to favorably support the analyst's understanding of the contents of the program.

According to the fourteenth aspect of the present invention, as a rule for identifying a conceptual term in an object, the conceptual term acquired from the names of global variables applied to a plurality of files is set as a candidate for the object. Since the object for constructing the information model can be easily acquired, it has the effect of being able to favorably support the analyst's understanding of the contents of the program.

According to the fifteenth aspect of the present invention, as a rule for identifying a concept term as an attribute, when the concept term acquired from the composite type name is used as an object, the concept term acquired from the composite type subfield name is used. By making the candidates of the attributes, the attributes of the objects that construct the information model of the program can be easily obtained, and therefore, there is an effect that the analyst can favorably understand the contents of the program.

According to the sixteenth aspect of the present invention, as a rule for identifying a concept term in a link, a verb-like concept term acquired from the name of a routine is used as a link candidate, and the direction of this link is performed inside the routine. It is possible to easily obtain links between objects that construct the information model of the program by making a decision based on the information about the operation of the argument and the verb case control information preset in the conceptual term dictionary. As a result, it is possible to favorably support the analyst's understanding of the contents of the program.

According to the seventeenth aspect of the present invention, in the program analysis apparatus for acquiring information for understanding the content from the program, the term generating means for acquiring the conceptual term from the name of the program identifier is provided, so that the content of the program is Since the conceptual term can be obtained from the name of the identifier that reflects the content of the program as the information of the abstract level indicating the, there is an effect that the analyst can easily support the understanding of the content of the program.

[Brief description of drawings]

FIG. 1 is a schematic diagram illustrating an example of a program analysis method by a program analysis device of the present invention.

FIG. 2 is a plan view exemplifying an image showing a relationship of conceptual terms generated as one of pieces of abstraction level information that can be assisted by the program analysis method to understand the contents of a program by the program analysis method.

FIG. 3 is a schematic diagram exemplifying a hierarchical structure of modules of a program targeted by a program analysis method by a program analysis device.

FIG. 4 is a schematic diagram exemplifying a hierarchical tree diagram that is a hierarchical cluster generated as one of pieces of abstraction level information that can support understanding of the content of a program.

Claims (17)

[Claims] 1. In a program analysis method for obtaining information for understanding the contents of a program formed by a hierarchical structure of modules in which identifiers of variables, routines, types, macros, etc. are set, the name of the program identifier is used. A program analysis method characterized in that a conceptual term is acquired. 2. A conceptual term dictionary in which abbreviated expressions frequently used for the names of program identifiers are set in advance together with the original expressions is provided, and the abbreviated expressions are detected from the names of program identifiers by referring to this conceptual term dictionary. The program analysis method according to claim 1, wherein the detected abbreviated expression is restored to the original expression to be a conceptual term. 3. A lexical sequence extracted from a variable name is used as a noun-like conceptual term, and a lexical sequence extracted from a routine name is used as a set of noun-like and verb-like conceptual terms. The program analysis method according to claim 1, wherein the program analysis method is performed. 4. A method of recognizing a dependency relation by discriminating a conjunction or a preposition in a lexical sequence of a noun-like conceptual term, and rearranging the lexical sequence into a normal expression of a noun phrase according to the dependency relation. 4. The program analysis method according to claim 3, wherein 5. A concept term dictionary in which a verb frequently used for a name of a program identifier is preset is provided, and when the verb set in the concept term dictionary is detected from a lexical sequence acquired from a routine name, 4. The program analysis method according to claim 3, wherein the token is used as a verb-like conceptual term and the other tokens are used as noun-like conceptual terms. 6. The similarity of concept terms is measured based on the phenomenon of simultaneous appearance of concept terms in an expression of a program or an assignment statement, and an image showing a relationship between a plurality of concept terms is obtained by multidimensional scaling by this similarity. The program analysis method according to claim 1, wherein the program analysis method is generated. 7. The similarity of concept terms is measured based on the phenomenon of the relationship between types and types in a program, and the similarity is used to generate an image showing the relationship between a plurality of concept terms by multidimensional scaling. 7. The program analysis method according to claim 1, wherein the program analysis method is performed. 8. The similarity of conceptual terms is measured based on the phenomenon of simultaneous appearance of conceptual terms in the file structure of a program, and an image showing the relationship between a plurality of conceptual terms is generated by multidimensional scaling by this similarity. The program analysis method according to claim 1, 6 or 7, characterized in that. 9. A measure of the effect of concealing a conceptual term obtained from the name of an external reference of a routine is defined as a similarity of the routine, and the similarity analyzes a hierarchical cluster of routines based on the concealment of the conceptual term by cluster analysis. The program analysis method according to claim 1, wherein the program analysis method is constructed. 10. The degree of distribution of conceptual terms in a routine is quantified as an entropy in a routine space,
10. The program analysis method according to claim 9, wherein the similarity between the two routines is measured as the amount of entropy reduction when the two routines are clustered. 11. The program analysis method according to claim 1, wherein conceptual terms are made to correspond to an object, an attribute, and a link that construct an information model of the program. 12. The program analysis method according to claim 11, wherein, as a rule for identifying the conceptual term as an object, the conceptual term acquired from a higher-order module is set as an object candidate. 13. The program analysis method according to claim 11, wherein the rule for identifying the conceptual term to the object is that the conceptual term acquired from the name of the composite type is set as the candidate of the object. 14. A rule for identifying a concept term to an object, wherein a concept term acquired from the names of global variables applied to a plurality of files is set as a candidate for the object. Alternatively, the program analysis method according to item 13. 15. As a rule for identifying a concept term as an attribute, when a concept term acquired from a composite type name is used as an object, a concept term acquired from a composite type subfield name is set as an attribute candidate. 15. The program analysis method according to claim 11, 12, 13 or 14. 16. As a rule for identifying a concept term to a link, a verb-like concept term acquired from the name of a routine is used as a link candidate, and the directionality of this link is information regarding an operation of an argument performed inside the routine. And the case control information of the verb preset in the concept term dictionary.
The program analysis method according to 3, 14 or 15. 17. The name of a program identifier in a program analysis device for acquiring information for understanding the contents of a program formed by a hierarchical structure of modules in which identifiers such as variables, routines, types and macros are set. A program analysis device comprising a term generation means for acquiring a conceptual term from the.