Abstract
In the article, we will compare three mechanisms of code analysis from the viewpoint of detecting 64-bit errors: the Visual C++ 2010 compiler, the Code Analysis for C/C++ component included into Visual Studio 2010 and Viva64 analyzer included into PVS-Studio 3.60. I will show both the capabilities of detecting defects in 64-bit projects and preliminary diagnosis of 64-bit errors in the 32-bit code of projects.Introduction
Our company OOO "Program Verification Systems" develops a specialized static code analyzer Viva64 intended to detect 64-bit errors in Windows-applications. The Viva64 analyzer is included into the PVS-Studio package integrating into the Visual Studio 2005/2008/2010 environment.Our potential users who think about purchasing PVS-Studio often ask us what advantages our tool has over diagnostic capabilities of the Visual C++ compiler and Code Analysis for C/C++ component available in extended editions of Visual Studio (for instance, in Visual Studio 2010 Premium/Ultimate).
Our users are also interested in the capability of preliminary detection of 64-bit errors at the stage when there is no 64-bit project yet.
In this article, we will compare various tools by 31 patterns of 64-bit errors and show how efficient they are when checking 32-bit and 64-bit projects. In the third section, there are links for you to learn about each error pattern in detail and comments on the comparison tables. The test project that we used as a basis for comparison and that contains all the error patterns can be downloaded here: http://www.viva64.com/external-pictures/ErrorExamples-vs2010-project.7z.
1. Comparison of tools when analyzing 64-bit projects
Comparison of tools and percentage of defects found when analyzing 64-bit projects are presented in Table 1. Note that the column referring to Code Analysis for C/C++ is empty. The reason is that Code Analysis for C/C++ does not work with 64-bit projects.Also note that we enabled all the warnings of the Visual C++ compiler with the /Wall switch, i.e. all its diagnostic capabilities are in use. The /Wp64 switch is disabled since it is ignored (has no sense) when compiling 64-bit projects.
The coloring of the table cells (the legend):
- Grey - cannot be diagnosed.
- Blue background - can be diagnosed partly (see explanations in the third section).
- Green background - can be diagnosed.
Table 1 - Comparison of capabilities of the Visual C++ 2010 compiler, Code Analysis for C/C++ (Visual Studio 2010 Premium) and Viva64 (PVS-Studio 3.60) in detecting 64-bit errors in a 64-bit project
Conclusion
The diagnostic capabilities of the Viva64 static analyzer exceed those of Visual C++ 2010 several times when searching for 64-bit errors in 64-bit projects. The Code Analysis for C/C++ is useless in searching for this type of errors since it cannot work with the code of 64-bit projects at the moment.
2. Comparison of tools when analyzing 32-bit projects
People are often interested in the possibility of detecting 64-bit errors already at the stage of working with the 32-bit project already. This interest results from the following two tasks:- To estimate the cost of porting a 32-bit application to a 64-bit system.
- To eliminate as many 64-bit errors as possible before porting the application.
The /Wall and /Wp64 switches are enabled for the Visual C++ compiler to use its diagnostic capabilities to the full extent. For the Code Analysis for C/C++ unit, we have also enabled all the possible warnings.
The coloring of the table cells (the legend):
- Grey - cannot be diagnosed.
- Blue background - can be diagnosed partly (see explanations in the third section).
- Green background - can be diagnosed.
Table 2 - Comparison of capabilities of the Visual C++ 2010 compiler, Code Analysis for C/C++ (Visual Studio 2010 Premium) and Viva64 (PVS-Studio 3.60) in detecting 64-bit errors in a 32-bit project
Conclusion
The diagnostic capabilities of the Viva64 analyzer exceed those of Visual C++ 2010 several times when searching for 64-bit errors in 32-bit projects.
The diagnostic capabilities of Visual C++ 2010 turn out to be less efficient when analyzing 32-bit projects than in case of 64-bit projects. This is explained by the fact that the compiler uses a different data model (ILP32) when compiling 32-bit projects.
The Code Analysis for C/C++ component is a general-purpose static analyzer and does not help in detecting the type of 64-bit errors we consider here.
The Viva64 analyzer performed equally full analysis both for 32-bit and 64-bit projects. In practice, the Viva64 analyzer still might miss up to 5% of errors and show fewer warnings. To learn more about it, please see - Lesson 28. Estimating the cost of 64-bit migration of C/C++ applications.
3. Description of comparison parameters
A detailed description of each error pattern will take too much space in the article. So let me just give you links to various sources where you may thoroughly study each of the patterns and see various examples. I will also give some comments explaining why some types of errors can be diagnosed only partly.3.1. Implicit conversion of a 32-bit type to a memsize-type
Description:- PVS-Studio documentation. V101. Implicit assignment type conversion to memsize type.
- Lessons on development of 64-bit C/C++ applications. Lesson 11. Pattern 3. Shift operations.
- Lessons on development of 64-bit C/C++ applications. Lesson 17. Pattern 9. Mixed arithmetic.
3.2. Dangerous address arithmetic
Description:- PVS-Studio documentation. V102. Usage of non memsize type for pointer arithmetic.
- Lessons on development of 64-bit C/C++ applications. Lesson 13. Pattern 5. Address arithmetic.
3.3. Implicit conversion of a memsize-type to a 32-bit type
Description:- PVS-Studio documentation. V103. Implicit type conversion from memsize type to 32-bit type.
- Lessons on development of 64-bit C/C++ applications. Lesson 17. Pattern 9. Mixed arithmetic.
- Andrey Karpov. OOO "Program Verification Systems" company blog. Problems of 64-bit code in real programs: FreeBSD.
3.4. Implicit conversion of a 32-bit type to a memsize-type in a comparison operation
Description:- PVS-Studio documentation. V104. Implicit type conversion to memsize type in an arithmetic expression.
- Lessons on development of 64-bit C/C++ applications. Lesson 17. Pattern 9. Mixed arithmetic.
- Andrey Karpov. OOO "Program Verification Systems" company blog. Why A + B != A - (-B).
3.5. Implicit conversion of a 32-bit type to a memsize-type in a ternary operation
Description:- PVS-Studio documentation. V105. N operand of '?:' operation: implicit type conversion to memsize type.
- Lessons on development of 64-bit C/C++ applications. Lesson 17. Pattern 9. Mixed arithmetic.
3.6. Implicit conversion of a 32-bit type to a memsize-type when calling a function
Description:- PVS-Studio documentation. V106. Implicit type conversion N argument of function 'foo' to memsize type.
- Lessons on development of 64-bit C/C++ applications. Lesson 17. Pattern 9. Mixed arithmetic.
3.7. Implicit conversion of a memsize-type to a 32-bit type when calling a function
Description:- PVS-Studio documentation. V107. Implicit type conversion N argument of function 'foo' to 32-bit type.
- Lessons on development of 64-bit C/C++ applications. Lesson 17. Pattern 9. Mixed arithmetic.
3.8. A non-memsize type is used as an index
Description:- PVS-Studio documentation. V108. Incorrect index type: 'foo[not a memsize-type]'. Use memsize type instead.
- Lessons on development of 64-bit C/C++ applications. Lesson 13. Pattern 5. Address arithmetic.
3.9. Implicit conversion of a 32-bit type to a memsize-type inside the return operator
Description:- PVS-Studio documentation. V109. Implicit type conversion of return value to memsize type.
- Lessons on development of 64-bit C/C++ applications. Lesson 17. Pattern 9. Mixed arithmetic.
3.10. Implicit conversion of a memsize-type to a 32-bit type inside the return operator
Description:- PVS-Studio documentation. V110. Implicit type conversion of return value from memsize type to 32-bit type.
- Lessons on development of 64-bit C/C++ applications. Lesson 17. Pattern 9. Mixed arithmetic.
3.11. Functions with a variable number of arguments, a value of a memsize-type is passed as the parameter
Description:- PVS-Studio documentation. V111. Call function 'foo' with variable number of arguments. N argument has memsize type.
- Lessons on development of 64-bit C/C++ applications. Lesson 10. Pattern 2. Functions with variable number of arguments.
3.12. Dangerous magic number
Description:- PVS-Studio documentation. V112. Dangerous magic number N used.
- Lessons on development of 64-bit C/C++ applications. Lesson 9. Pattern 1. Magic numbers.
3.13. Attempt to store a value of a memsize-type in a variable of the double type
Description:- PVS-Studio documentation. V113. Implicit type conversion from memsize to double type or vice versa.
- PVS-Studio documentation. V203. Explicit type conversion from memsize to double type or vice versa.
- Lessons on development of 64-bit C/C++ applications. Lesson 18. Pattern 10. Storage of integer values in double.
When building a 32-bit project, the Visual C++ compiler warns only about the conversion of the double type to size_t and does not warn about the opposite conversion.
3.14. Incorrect change of the pointer's type
Description:- PVS-Studio documentation. V114. Dangerous explicit type pointer conversion.
- Lessons on development of 64-bit C/C++ applications. Lesson 14. Pattern 6. Changing an array's type.
- Andrey Karpov. OOO "Program Verification Systems" company blog. 64-bit code issues in real programs: pointer type change.
3.15. Using memsize types when handling exceptions
Description:- PVS-Studio documentation. V115. Memsize type is used for throw.
- PVS-Studio documentation. V116. Memsize type is used for catch.
- Lessons on development of 64-bit C/C++ applications. Lesson 20. Pattern 12. Exceptions.
3.16. Memsize-types in unions
Description:- PVS-Studio documentation. V117. Memsize type is used in the union.
- Lessons on development of 64-bit C/C++ applications. Lesson 16. Pattern 8. Memsize-types in unions.
3.17. Dangerous expression as an argument of the malloc() function
Description:- PVS-Studio documentation. V118. malloc() function accepts a dangerous expression in the capacity of an argument.
- Lessons on development of 64-bit C/C++ applications. Lesson 9. Pattern 1. Magic numbers.
3.18. Incorrect calculation of object sizes using several sizeof() operators
Description:- PVS-Studio documentation. V119. More than one sizeof() operators are used in one expression.
- Lessons on development of 64-bit C/C++ applications. Lesson 21. Pattern 13. Data alignment.
3.19. The new operator accepts an expression of a 32-bit type as an argument
Description:- PVS-Studio documentation. V121. Implicit conversion of the type of 'new' operator's argument to size_t type.
- Lessons on development of 64-bit C/C++ applications. Lesson 17. Pattern 9. Mixed arithmetic.
3.20. Explicit conversion of a 32-bit type to a memsize-type
Description:- PVS-Studio documentation. V201. Explicit type conversion. Type casting to memsize.
3.21. Explicit conversion of a memsize-type to a 32-bit type
Description:- PVS-Studio documentation. V202. Explicit type conversion. Type casting from memsize to 32-bit.
- Lessons on development of 64-bit C/C++ applications. Lesson 15. Pattern 7. Pointer packing.
- Andrey Karpov. OOO "Program Verification Systems" company blog. Search of explicit type conversion errors in 64-bit programs.
The Visual C++ compiler diagnoses only the explicit conversion of pointers to 32-bit data types but not all the memsize-types.
3.22. Incorrectly defined virtual functions
Description:- PVS-Studio documentation. V301. Unexpected function overloading behavior. See N argument of function 'foo' in derived class 'derived' and base class 'base'.
- Lessons on development of 64-bit C/C++ applications. Lesson 12. Pattern 4. Virtual functions.
- Andrey Karpov. OOO "Program Verification Systems" company blog. Issues of 64-bit code in real programs: virtual functions.
The Visual C++ compiler diagnoses all the cases when the function prototype in the descendant class differs from the prototype of the function defined as a virtual function in the base class. As a result, a lot of warnings are generated which do not refer to 64-bit defects and it complicates the use of this diagnosis type. Moreover, the compiler does not detect this type of errors at all when compiling a 32-bit project.
3.23. Dangerous [] operator
Description:- PVS-Studio documentation. V302. Member operator[] of 'foo' class has a 32-bit type argument. Use memsize-type here.
- Andrey Karpov. OOO "Program Verification Systems" company blog. Search of 64-bit errors in array implementation.
3.24. Using deprecated functions
Description:- PVS-Studio documentation. V303. The function is deprecated in the Win64 system. It is safer to use the 'foo' function.
3.25. Buffer overflow or underflow error
Description:- PVS-Studio documentation. V320. A call of the 'foo' function will lead to a buffer overflow or underflow in a 64-bit system.
The task of searching for the buffer overflow is difficult and often cannot be solved by means of static analysis at all. That is why we specified in the table that the Viva64 analyzer detects only some of the defects of this kind.
3.26. Searching for structures whose sizes can be decreased without performance loss
Description:- PVS-Studio documentation. V401. The structure's size can be decreased via changing the fields' order. The size can be reduced from N to K bytes.
- Lessons on development of 64-bit C/C++ applications. Lesson 23. Pattern 15. Growth of structures' sizes.
- Andrey Karpov. OOO "Program Verification Systems" company blog. Viva64 for optimizing data structures.
The Visual C++ compiler warns about all empty spaces between fields in structures that appear because of data alignment. This information may be used to search for non-optimal structures but it is difficult to do in practice.
3.27. Using a function without preliminarily defining it (in the C language)
Description:- PVS-Studio documentation. V102. Usage of non memsize type for pointer arithmetic.
- Andrey Karpov. OOO "Program Verification Systems" company blog. A nice 64-bit error in C.
The analyzer diagnoses this type of errors indirectly by generating a warning about conversion of the int type to the pointer.
3.28. Incorrect #ifdef..#else
Description:- Andrey Karpov. A Collection of Examples of 64-bit Errors in Real Programs. Example 3.
3.29. Serialization errors (changes of type sizes, byte order changes)
Description:- Lessons on development of 64-bit C/C++ applications. Lesson 19. Pattern 11. Serialization and data interchange.
3.30. Redirection errors (referring to WoW64)
Description:- Andrey Karpov. A Collection of Examples of 64-bit Errors in Real Programs. Example 30.
3.31. Changes of program behavior when using overloaded functions
Description:- Lessons on development of 64-bit C/C++ applications. Lesson 22. Pattern 14. Overloaded functions.
Комментариев нет:
Отправить комментарий