CN106933572B - Measurement model based on LLVM intermediate representation program slice - Google Patents

Abstract

The invention discloses a measurement model based on LLVM intermediate representation program slice, which comprises the following steps: converting the source program into a corresponding LLVM IR language through an LLVM command line; constructing a program system dependency graph, and slicing program LLVM IR statements; obtaining the module size of each module in the program, the number of output variables, the maximum slice set size and the minimum slice set size in the slice sets of all the output variables of each measurement module and the common slice set of all the variables according to the program slicing result and the LLVM IR statement obtained by converting the source program; substituting the value obtained in the previous step into a measurement function to finish LLVM IR measurement of the function, and then calculating Halstead measurement of a program; removing the measurement values with the same function from all measurement results of the program, constructing a Complexity measurement function Complexity, calculating the Complexity of each module of the program and sequencing according to the size of the Complexity, wherein software development or test can reasonably distribute development resources according to the Complexity of each module and carry out error detection. The invention effectively improves the universality and the measurement precision of program slicing.

Description

Measurement model based on LLVM intermediate representation program slice

Technical Field

The invention relates to a method for correlating program slices and software measurement, provides a method for constructing a measurement model based on LLVM intermediate representation program slices, and belongs to the technical field of computer programs.

Background

With the rapid development of the computer industry, a large number of software products are applied to social life. While the number and scale of software systems continues to grow, the problems that arise in software development, testing, and the cost of software maintenance also continue to increase. Many conventional software analysis methods have been unable to meet the requirements of existing software analysis, and new techniques and methods need to be found to help solve the problem. Thus, program analysis becomes a very important component in the software field.

Since the program slicing was proposed in the doctor's paper by m.weiser in 1979, the theory and application development of program slicing have been greatly developed, so that the program slicing plays a very important role in software testing, maintenance, program understanding, optimization and the like. Program slicing is a technique for analyzing and understanding a program by slicing each point of interest in a source program. In short, the program slicing technique is to simplify the program to facilitate the research of the program.

Software metrics are important indicators for ensuring software quality. After measurement indexes such as Halstead, Maccabe, cyclic Complexity, Design Complexity and the like are continuously applied to software analysis. Program slice-based metrics are also applied to software modularization analysis. The slicing-based measurement is to slice the output variable in the module by using a program slicing technology and process the slicing result, so as to obtain a series of quantitative indexes through calculation, and to quantify the strength of functional dependence of elements in the module through indexes, so as to analyze the developed program. The slice-based measurement method plays a great role in measuring the progress of software re-engineering or predicting new software bugs which may occur through the analysis of software error modules.

Although program slice-based metrics have been used in program analysis, since the basic unit of program slice analysis is a statement, the construction of the system dependency graph is also based on the program line number, and the resulting slice set of slice results is a statement line number set. The object of the slice is generally a program written in a specific language such as C, C + +, Python, etc., and the generality is not strong.

LLVM IR (Low Level Virtual Machine Intermediate Representation) is a format between c language and assembly language, has readability of a high-Level language, can reflect the operation and transmission conditions of data of a computer bottom layer more comprehensively, and is refined and efficient. Any one programming language can be converted to LLVMIR, which can also be converted to any one programming language. The basic units of the system dependency graph constructed by LLVM IR and the result set of program slicing are all instructions, which can increase the generality of program slicing and the precision of program analysis.

The invention uses LLVM IR to slice the program, and measures and calculates the slicing result and the instruction set, thereby greatly improving the universality of the program slicing and the measurement precision.

Disclosure of Invention

The invention aims to overcome the defects of the prior art and provides a method for constructing a measurement model based on LLVM intermediate representation program slices, which can effectively improve the universality and measurement precision of the program slices.

In order to solve the problems of the prior art, the invention adopts the following technical scheme.

The invention discloses a construction method of a measurement model based on LLVM intermediate representation program slices, which is characterized by comprising the following steps:

step (1) converting a source program into a corresponding LLVM IR language through a command line of the LLVM;

constructing a system dependency graph by using LLVM IR to obtain a corresponding program slice;

and (3) obtaining the information of the instruction and the instruction set, the program operator and the operand corresponding to the output variable of the source program, namely: obtaining the module size of each module in the program and the number of output variables according to the program slicing result and LLVM IR statements obtained by converting the source program, and obtaining the maximum slice set size and the minimum slice set size in all slice sets of output variables of each measurement module and a common slice set of all variables;

and (4) calculating various measurement values of the function module according to the slicing result of the LLVM IR and various measurement formulas, namely: bringing the value obtained in the step (3) into a measurement function to complete LLVM IR measurement of the function, and then calculating Halstead measurement of a program;

and (5) calculating and evaluating according to the measurement values, namely: removing the measurement values with the same function from all measurement results of the program, and then sequencing the rest measurement values according to the influence conditions on the reliability and the stability of the program to obtain the module check sequence before the program is wrongly issued or the program is wrongly issued.

The specific process of the step (1) is as follows:

step 1.1) get binary bc file using command clean-emit-llvm-c hello.c-o hello.bc:

step 1.2) using a command llvm-dis hello.bc to obtain a corresponding readable hello.ll file, and opening the file through a text editor.

The specific process of the step (2) is as follows:

step 2.1) constructing a control flow graph CFG of the program;

step 2.2) adding corresponding entry nodes, format-in nodes, format-out nodes, and actual-in nodes corresponding to the format-in nodes and actual-out nodes corresponding to the format-out nodes in each process;

step 2.3) adding a control dependent edge;

step 2.4) adding a data dependent edge;

step 2.5) obtaining the dependence of the actual-out point on the actual-in point in the program calling process according to the corresponding relation between the formal-in/-out and the actual-in/-out nodes, and adding a summary edge;

and 2.6) based on the slicing criterion, performing backward traversal on the system dependency graph by using a two-stage graph reachability algorithm of the HRB.

The specific process of the step (3) is as follows:

step 3.1) finding a return value RV of each subprogram in the source program, outputting an output variable PV in a statement, a changed global variable GV in the program, and putting a function application parameter RFV into an output variable set OVS, wherein the function application parameter RFV is used as an output variable;

step 3.2) finding an instruction slice set corresponding to each subprogram OVS from the slice set;

step 3.3) calculating the # OVS, # Mod, { modi }, maxMod, minMod and comMod of the subprogram through the instruction set of the OVS;

the number # OVS refers to the number of OVS, the number # Mod refers to the number of instructions of the subprogram, { modi } refers to the number of instructions of each element in the OVS, maxMod refers to the number of instructions of the largest instruction slice set among the elements in the OVS in each module, minMod refers to the number of instructions of the smallest instruction slice set among the elements in the OVS in each module, and comMod refers to the number of identical instructions in the slice sets corresponding to all the elements in the OVS;

step 3.4) obtaining the unique operand quantity η of the program according to the LLVM IR program corresponding to the program₁number of unique operators η₂All operands N present in the program₁And programAll operators N present in₂。

The specific process of the step (4) is as follows:

step 4.1) obtaining Total IRs of the program according to the basic attribute and the slice basic value of the function;

step 4.2) according to

Obtaining the maximum coverage condition of the OVS of the program;

step 4.3) according to

Obtaining the minimum coverage condition of the OVS of the program;

step 4.4) according to

Obtaining the tightness condition of the OVS of the program;

step 4.5) according to

Obtaining the overlapping condition of OVS of the program;

step 4.6) according to

Obtaining the OVS coverage condition of the program;

step 4.7) obtaining the coupling condition CouplingIR (f, g) among all the subprograms in the program, wherein f and g are different subprograms in the program;

step 4.8) according to N ═ N₁+N₂N × log (η), yielding program length N in Halstead, and program capacity V;

and 4.9) obtaining the cyclic compatibility as e-n +2p according to the dependency graph of the program, wherein the e, n and p are the number of edges, the number of nodes and the number of connected components in the control flow graph respectively.

The specific process of the step (5) is as follows:

step 5.1) carrying out principal component analysis on the multiple measurement indexes, and removing the measurement indexes having the same influence on the program according to a principal component analysis method;

step 5.2) the measurement indexes obtained after the processing of the step 5.1) are sorted according to the respective contribution values of the measurement indexes;

step 5.3) construction of a complexity metric function according to step 5.2)

α + β + χ is 1, and the weighting factors α, β and χ have values of α being 0.5, β being 0.4 and χ being 0.1, respectively;

step 5.4) according to the analysis of each measurement index to the program, according to different conditions of the influence of each measurement index, obtaining a sequence of each subprogram in the aspect of complexity; the complexity function takes into account various aspects including: stability, portability, and error-prone conditions; compared with the prior art, the invention has the following advantages and beneficial effects:

(1) the invention uses LLVM IR as the basic unit of program slice, not a specific programming language, and takes out the slice result corresponding to the line number, and uses the program slice result and LLVM converted from the source program to perform measurement calculation, which improves the universality of program slice and the accuracy of software measurement based on program slice.

(2) The analysis granularity of the program in the metric model provided by the invention is finer, and the statement level is changed into the LLVM IR instruction level, so that the obtained metric result is more accurate.

(3) The invention combines the cohesion, the coupling degree and Halstead based on the slice measurement for use, and provides the Complexity measurement Complexity based on the slice measurement, which can help developers to reasonably distribute development resources.

Drawings

FIG. 1 is a flow chart of a construction method of an embodiment of the invention.

Detailed Description

The present invention will be described in further detail with reference to the accompanying drawings.

FIG. 1 is a flow chart of a method for constructing a metrology model based on LLVM intermediate representation program slicing according to an embodiment of the present invention. As shown in fig. 1, the method of this embodiment includes the steps of:

step (1) converting a source program into a corresponding LLVM IR language through a command line of the LLVM;

constructing a system dependency graph by using LLVM IR to obtain a corresponding program slice;

and (3) obtaining the information of the instruction and the instruction set, the program operator and the operand corresponding to the output variable of the source program, namely: obtaining the module size of each module in the program and the number of output variables according to the program slicing result and LLVM IR statements obtained by converting the source program, and obtaining the maximum slice set size and the minimum slice set size in all slice sets of output variables of each measurement module and a common slice set of all variables;

and (4) calculating various measurement values of the function module according to the slicing result of the LLVM IR and various measurement formulas, namely: bringing the value obtained in the step (3) into a measurement function to complete LLVM IR measurement of the function, and then calculating Halstead measurement and circle complexity measurement of a program;

and (5) calculating and evaluating according to the measurement values, namely: removing the measurement values with the same function from all measurement results of the program, and then sequencing the rest measurement values according to the influence conditions on the reliability and the stability of the program to obtain the module check sequence before the program is wrongly issued or the program is wrongly issued.

The specific process of the step (1) is as follows:

step 1.1) get binary bc file using command clean-emit-llvm-c hello.c-o hello.bc:

step 1.2) using a command llvm-dis hello.bc to obtain a corresponding readable hello.ll file, and opening the file through a text editor.

The specific process of the step (2) is as follows:

step 2.1) constructing a control flow graph CFG of the program;

step 2.2) adding corresponding entry nodes, format-in nodes, format-out nodes, and actual-in nodes corresponding to the format-in nodes and actual-out nodes corresponding to the format-out nodes in each process;

step 2.3) adding a control dependent edge;

step 2.4) adding a data dependent edge;

step 2.5) obtaining the dependence of the actual-out point on the actual-in point in the program calling process according to the corresponding relation between the formal-in/-out and the actual-in/-out nodes, and adding a summary edge;

and 2.6) based on the slicing criterion, performing backward traversal on the system dependency graph by using a two-stage graph reachability algorithm of the HRB.

The specific process of the step (3) is as follows:

step 3.1) finding a return value RV of each subprogram in the source program, outputting an output variable PV in a statement, a changed global variable GV in the program, and putting a function application parameter RFV into an output variable set OVS, wherein the function application parameter RFV is used as an output variable;

step 3.2) finding an instruction slice set corresponding to each subprogram OVS from the slice set;

step 3.3) calculating the # OVS, # Mod, { modi }, maxMod, minMod and comMod of the subprogram through the instruction set of the OVS;

the number # OVS refers to the number of OVS, the number # Mod refers to the number of instructions of the subprogram, { modi } refers to the number of instructions of each element in the OVS, maxMod refers to the number of instructions of the largest instruction slice set among the elements in the OVS in each module, minMod refers to the number of instructions of the smallest instruction slice set among the elements in the OVS in each module, and comMod refers to the number of identical instructions in the slice sets corresponding to all the elements in the OVS;

step 3.4) obtaining the unique operand quantity η of the program according to the LLVM IR program corresponding to the program₁number of unique operators η₂All operands N present in the program₁And all operators N present in the program₂。

The specific process of the step (4) is as follows:

step 4.1) obtaining Total IRs of the program according to the basic attribute and the slice basic value of the function;

step 4.2) according to

Obtaining the maximum coverage condition of the OVS of the program;

step 4.3) according to

Obtaining the minimum coverage condition of the OVS of the program;

step 4.4) according to

Obtaining the tightness condition of the OVS of the program;

step 4.5) according to

Obtaining the overlapping condition of OVS of the program;

step 4.6) according to

Obtaining the OVS coverage condition of the program;

step 4.7) obtaining the coupling condition CouplingIR (f, g) among all the subprograms in the program, wherein f and g are different subprograms in the program;

step 4.8) according to N ═ N₁+N₂N × log (η), yielding program length N in Halstead, and program capacity V;

and 4.9) obtaining the cyclic compatibility as e-n +2p according to the dependency graph of the program, wherein the e, n and p are the number of edges, the number of nodes and the number of connected components in the control flow graph respectively.

The specific process of the step (5) is as follows:

step 5.1) carrying out principal component analysis on the multiple measurement indexes, and removing the measurement indexes having the same influence on the program according to a principal component analysis method;

step 5.2) the measurement indexes obtained after the processing of the step 5.1) are sorted according to the respective contribution values of the measurement indexes;

step 5.3) construction of a complexity metric function according to step 5.2)

α + β + χ is 1, and the weighting factors α, β and χ have values of α being 0.5, β being 0.4 and χ being 0.1, respectively;

step 5.4) according to the analysis of each measurement index to the program, according to different conditions of the influence of each measurement index, obtaining a sequence of each subprogram in the aspect of complexity; the complexity function takes into account various aspects including: stability, portability, and error-prone conditions;

and 5.5) if the software or the program has problems, checking and modifying according to the sequence in which the software is easy to make mistakes.

In summary, the present invention provides a metric model based on LLVM intermediate representation program slices. Firstly, converting a source program into a corresponding LLVM IR language through a command line of an LLVM; and constructing a program system dependency graph, and slicing LLVMIR statements of the program. And then obtaining the module size of each module in the program, the number of variables, the maximum size of the slice set and the minimum size of the slice set in the slice sets of all the variables, and a common slice set of all the variables according to the program slicing result and the LLVM IR statement obtained by converting the source program. The slice base values from the above steps are then substituted into the LLVM IR metric function and the commonly used Halstead, cyclic Complexity, etc. metric functions involving re-use in LLVM IR and Halstead and circle Complexity metrics for the inter-function cohesion and coupling related metric functions proposed and improved by m.weiser and Longworth et al. Finally, removing the measurement values with the same function from all measurement results of the program, and then sequencing the rest measurement values according to the influence conditions on the reliability and the stability of the program to obtain the module check sequence before the program is wrongly issued or the program is wrongly issued.

Claims (6)

1. A method for constructing a metric model based on LLVM intermediate representation program slices is characterized by comprising the following steps:

step (1) converting a source program into a corresponding LLVM IR language through a command line of the LLVM;

constructing a system dependency graph by using LLVM IR to obtain a corresponding program slice;

and (3) obtaining the information of the instruction and the instruction set, the program operator and the operand corresponding to the output variable of the source program, namely: obtaining the module size of each module in the program and the number of output variables according to the program slicing result and LLVM IR statements obtained by converting the source program, and obtaining the maximum slice set size and the minimum slice set size in all slice sets of output variables of each measurement module and a common slice set of all variables;

and (4) calculating various measurement values of the function module according to the slicing result of the LLVM IR and various measurement formulas, namely: bringing the value obtained in the step (3) into a measurement function to complete LLVM IR measurement of the function, and then calculating Halstead measurement of a program;

and (5) calculating and evaluating according to the measurement values, namely: all measurement results of the program are firstly removed from the measurement values with the same action, then the rest measurement values are sequenced according to the influence conditions on the reliability and the stability of the program, and the development resources can be reasonably arranged according to module sequencing to check the modules in the software testing stage.

2. The method for constructing a metric model based on LLVM intermediate representation program slice as claimed in claim 1, wherein the specific process of step (1) is:

step 1.1) get binary bc file using command clean-emit-llvm-c hello.c-o hello.bc:

step 1.2) using a command llvm-dis hello.bc to obtain a corresponding readable hello.ll file, and opening the file through a text editor.

3. The method for constructing a metric model based on LLVM intermediate representation program slice as claimed in claim 1, wherein the specific process of step (2) is:

step 2.1) constructing a control flow graph CFG of the program;

step 2.2) adding corresponding entry nodes, format-in nodes, format-out nodes, and actual-in nodes corresponding to the format-in nodes and actual-out nodes corresponding to the format-out nodes in each process;

step 2.3) adding a control dependent edge;

step 2.4) adding a data dependent edge;

step 2.5) obtaining the dependence of the actual-out point on the actual-in point in the program calling process according to the corresponding relation between the formal-in/-out and the actual-in/-out nodes, and adding a summary edge;

and 2.6) based on the slicing criterion, performing backward traversal on the system dependency graph by using a two-stage graph reachability algorithm of the HRB.

4. The method for constructing a metric model based on LLVM intermediate representation program slice as claimed in claim 1, wherein the specific process of step (3) is:

step 3.1) finding a return value RV of each subprogram in the source program, outputting an output variable PV in a statement, a changed global variable GV in the program, and putting a function application parameter RFV into an output variable set OVS, wherein the function application parameter RFV is used as an output variable;

step 3.2) finding an instruction slice set corresponding to each subprogram OVS from the slice set;

step 3.3) calculating the # OVS, # Mod, { modi }, maxMod, minMod and comMod of the subprogram through the instruction set of the OVS;

the number # OVS refers to the number of OVS, the number # Mod refers to the number of instructions of the subprogram, { modi } refers to the number of instructions of each element in the OVS, maxMod refers to the number of instructions of the largest instruction slice set among the elements in the OVS in each module, minMod refers to the number of instructions of the smallest instruction slice set among the elements in the OVS in each module, and comMod refers to the number of identical instructions in the slice sets corresponding to all the elements in the OVS;

step 3.4) obtaining the unique operand quantity η of the program according to the LLVM IR program corresponding to the program₁number of unique operators η₂All operands N present in the program₁And all operators N present in the program₂。

5. The method for constructing a metric model based on LLVM intermediate representation program slice as claimed in claim 1, wherein the specific process of step (4) is:

step 4.1) obtaining Total IRs of the program according to the basic attribute and the slice basic value of the function;

step 4.2) according to

Obtaining the maximum coverage condition of the OVS of the program;

step 4.3) according to

Obtaining the minimum coverage condition of the OVS of the program;

step 4.4) according to

Obtaining the tightness condition of the OVS of the program;

step 4.5) according to

Obtaining the overlapping condition of OVS of the program;

step 4.6) according to

Obtaining the OVS coverage condition of the program;

step 4.7) obtaining the coupling condition CouplingIR (f, g) among all the subprograms in the program, wherein f and g are different subprograms in the program;

step 4.8) according to N ═ N₁+N₂N × log (η), yielding program length N in Halstead, and program capacity V;

and 4.9) obtaining the cyclic compatibility as e-n +2p according to the dependency graph of the program, wherein the e, n and p are the number of edges, the number of nodes and the number of connected components in the control flow graph respectively.

6. The method for constructing a metric model based on LLVM intermediate representation program slice as claimed in claim 1, wherein the specific process of step (5) is:

step 5.1) carrying out principal component analysis on the multiple measurement indexes, and removing the measurement indexes having the same influence on the program according to a principal component analysis method;

step 5.2) the measurement indexes obtained after the processing of the step 5.1) are sorted according to the respective contribution values of the measurement indexes;

step 5.3) construction of a complexity metric function according to step 5.2)

the weighting factors alpha, β and chi respectively take the values of alpha, β and chi as 0.5, β as 0.4 and chi as 0.1;

step 5.4) according to the analysis of each measurement index to the program, according to different conditions of the influence of each measurement index, obtaining a sequence of each subprogram in the aspect of complexity; the complexity function takes into account various aspects including: stability, portability, and error-prone conditions;

and 5.5) if the software or the program has problems, checking and modifying according to the sequence in which the software is easy to make mistakes.

Images