Cppcheck is an analysis tool for C/C++ code. It provides unique code analysis to detect bugs and focuses on detecting undefined behaviour and dangerous coding constructs. The goal is to detect only real errors in the code (i.e. have very few false positives).
We don't know which approach (project file or manual configuration) will give you the best results. It is recommended that you try both. It is possible that you will get different results so that to find most bugs you need to use both approaches.
To exclude a file or folder, there are two options. The first option is to only provide the paths and files you want to check.
cppcheck src/a src/b
All files under src/a and src/b are then checked.
The second option is to use -i, with it you specify files/paths to ignore. With this command no files in src/c are checked:
cppcheck -isrc/c src
This option does not currently work with the `--project` option and is only valid when supplying an input directory. To ignore multiple directories supply the -i multiple times. The following command ignores both the src/b and src/c directories.
suggestions about defensive programming to prevent bugs
**style**
stylistic issues related to code cleanup (unused functions, redundant code, constness, and such)
**performance**
Suggestions for making the code faster. These suggestions are only based on common knowledge. It is not certain you'll get any measurable difference in speed by fixing these messages.
**portability**
portability warnings. 64-bit portability. code might work different on different compilers. etc.
**information**
Configuration problems. The recommendation is to only enable these during configuration.
If you use `--project` then Cppcheck will use the preprocessor settings from the imported project. Otherwise you'll probably want to configure the include paths, defines, etc.
### Defines
Here is a file that has 2 preprocessor configurations (with A defined and without A defined):
#ifdef A
x = y;
#else
x = z;
#endif
By default Cppcheck will check all preprocessor configurations (except those that have #error in them). So the above code will by default be analyzed both with `A` defined and without `A` defined.
You can use `-D` to change this. When you use `-D`, cppcheck will by default only check the given configuration and nothing else. This is how compilers work. But you can use `--force` or `--max-configs` to override the number of configurations.
Check all configurations:
cppcheck file.c
Only check the configuration A:
cppcheck -DA file.c
Check all configurations when macro A is defined
cppcheck -DA --force file.c
Another useful flag might be `-U`. It tells Cppcheck that a macro is not defined. Example usage:
cppcheck -UX file.c
That will mean that X is not defined. Cppcheck will not check what happens when X is defined.
### Include paths
To add an include path, use `-I`, followed by the path.
Cppcheck's preprocessor basically handles includes like any other preprocessor. However, while other preprocessors stop working when they encounter a missing header, cppcheck will just print an information message and continues parsing the code.
The purpose of this behaviour is that cppcheck is meant to work without necessarily seeing the entire code. Actually, it is recommended to not give all include paths. While it is useful for cppcheck to see the declaration of a class when checking the implementation of its members, passing standard library headers is highly discouraged because it will result in worse results and longer checking time. For such cases, .cfg files (see below) are the better way to provide information about the implementation of functions and types to cppcheck.
## Suppressions
If you want to filter out certain errors you can suppress these.
Please note that if you see a false positive then we (the Cppcheck team) want that you report it so we can fix it.
### Plain text suppressions
You can suppress certain types of errors. The format for such a suppression is one of:
[error id]:[filename]:[line]
[error id]:[filename2]
[error id]
The `error id` is the id that you want to suppress. The easiest way to get it is to use the --template=gcc command line flag. The id is shown in brackets.
The filename may include the wildcard characters \* or ?, which match any sequence of characters or any single character respectively. It is recommended that you use "/" as path separator on all operating systems.
### Command line suppression
The `--suppress=` command line option is used to specify suppressions on the command line. Example:
cppcheck --suppress=memleak:src/file1.cpp src/
### Suppressions in a file
You can create a suppressions file. Example:
// suppress memleak and exceptNew errors in the file src/file1.cpp
memleak:src/file1.cpp
exceptNew:src/file1.cpp
// suppress all uninitvar errors in all files
uninitvar
Note that you may add empty lines and comments in the suppressions file.
You can specify suppressions in a XML file. Example file:
<?xml version="1.0"?>
<suppressions>
<suppress>
<id>uninitvar</id>
<fileName>src/file1.c</fileName>
<lineNumber>10</lineNumber>
<symbolName>var</symbolName>
</suppress>
</suppressions>
The xml format is extensible and may be extended with further attributes in the future.
You can use the suppressions file like this:
cppcheck --suppress-xml=suppressions.xml src/
### Inline suppressions
Suppressions can also be added directly in the code by adding comments that contain special keywords. Before adding such comments, consider that the code readability is sacrificed a little.
This code will normally generate an error message:
void f() {
char arr[5];
arr[10] = 0;
}
The output is:
cppcheck test.c
[test.c:3]: (error) Array 'arr[5]' index 10 out of bounds
To suppress the error message, a comment can be added:
void f() {
char arr[5];
// cppcheck-suppress arrayIndexOutOfBounds
arr[10] = 0;
}
Now the `--inline-suppr` flag can be used to suppress the warning. No error is reported when invoking cppcheck this way:
cppcheck --inline-suppr test.c
You can specify that the inline suppression only applies to a specific symbol:
samples/arrayIndexOutOfBounds/bad.c,6,error,arrayIndexOutOfBounds,Array 'a[2]' accessed at index 2, which is out of bounds.
### User defined output format (multi line)
Many warnings have multiple locations. Example code:
void f(int *p)
{
*p = 3; // line 3
}
int main()
{
int *p = 0; // line 8
f(p); // line 9
return 0;
}
There is a possible null pointer dereference at line 3. Cppcheck can show how it came to that conclusion by showing extra location information. You need to use both --template and --template-location at the command line.
multiline.c:3: warning: Possible null pointer dereference: p
*p = 3;
^
multiline.c:8: note: Assignment 'p=0', assigned value is 0
int *p = 0;
^
multiline.c:9: note: Calling function 'f', 1st argument 'p' value is 0
f(p);
^
multiline.c:3: note: Null pointer dereference
*p = 3;
^
The first line in the warning is formatted by the --template format.
The other lines in the warning are formatted by the --template-location format.
#### Format specifiers for --template
The available specifiers for --template are:
**{file}**
File name
**{line}**
Line number
**{column}**
Column number
**{callstack}**
Write all locations. Each location is written in [{file}:{line}] format and the locations are separated by ->. For instance it might look like: [multiline.c:8] -> [multiline.c:9] -> [multiline.c:3]
**{inconclusive:text}**
If warning is inconclusive then the given text is written. The given text can be any arbitrary text that does not contain }. Example: {inconclusive:inconclusive,}
Now if you want you can limit the analysis. You probably know what the target compiler is. If `-D` is supplied and you do not specify `--force` then Cppcheck will only check the configuration you give.
$ cppcheck -D __GNUC__ test.c
Checking test.c ...
Checking test.c: __GNUC__=1...
### Unused templates
If you think Cppcheck is slow and you are using templates, then you should try how it works to remove unused templates.
Imagine this code:
template <classT> struct Foo {
T x = 100;
};
template <classT> struct Bar {
T x = 200 / 0;
};
int main() {
Foo<int> foo;
return 0;
}
Cppcheck says:
$ cppcheck test.cpp
Checking test.cpp ...
[test.cpp:7]: (error) Division by zero.
It complains about division by zero in `Bar` even though `Bar` is not instantiated.
You can use the option `--remove-unused-templates` to remove unused templates from Cppcheck analysis.
Example:
$ cppcheck --remove-unused-templates test.cpp
Checking test.cpp ...
This lost message is in theory not critical, since `Bar` is not instantiated the division by zero should not occur in your real program.
Cppcheck has an addon that checks for MISRA C 2012 compliance.
### Requirements
You need:
Python 2.X or 3.X
The MISRA C 2012 PDF. You can buy this from http://www.misra.org.uk (costs 15-20 pounds)
#### MISRA Text file
It is not allowed to publish the MISRA rule texts. Therefore the MISRA rule texts are not available directly in the addon. We can not publish the rule texts. Instead, you must provide the rule texts; the addon can read the rule texts from a text file. If you copy/paste all text in "Appendix A Summary of guidelines" from the MISRA pdf, then you have all the rule texts.
If you have installed xpdf, such text file can be generated on the command line (using pdftotext that is included in xpdf):
pdftotext misra-c-2012.pdf output.txt
The output might not be 100% perfect so you might need to make minor tweaks manually.
Other pdf-to-text utilities might work also.
To create the text file manually, copy paste Appendix A "Summary of guidelines" from the MISRA PDF. Format:
Appendix A Summary of guidelines
Rule 1.1
Rule text
Rule 1.2
Rule text
...
Rules that you want to disable does not need to have a rule text. Rules that don't have rule text will be suppressed by the addon.
## Library configuration
When external libraries are used, such as WinAPI, POSIX, gtk, Qt, etc, Cppcheck doesn't know how the external functions behave. Cppcheck then fails to detect various problems such as leaks, buffer overflows, possible null pointer dereferences, etc. But this can be fixed with configuration files.
Cppcheck already contains configurations for several libraries. They can be loaded as described below. Note that the configuration for the standard libraries of C and C++, std.cfg, is always loaded by cppcheck. If you create or update a configuration file for a popular library, we would appreciate if you upload it to us.
You can create and use your own .cfg files for your projects. Use `--check-library` and `--enable=information` to get hints about what you should configure.
It is recommended that you use the `Library Editor` in the `Cppcheck GUI` to edit configuration files. It is available in the `View` menu. Not all settings are documented in this manual yet.
If you have a question about the .cfg file format it is recommended that you ask in the forum (<http://sourceforge.net/p/cppcheck/discussion/>).
The command line cppcheck will try to load custom .cfg files from the working path - execute cppcheck from the path where the .cfg files are.
The cppcheck GUI will try to load custom .cfg files from the project file path. The custom .cfg files should be shown in the Edit Project File dialog that you open from the `File` menu.
Cppcheck has configurable checking for leaks, e.g. you can specify which functions allocate and free memory or resources and which functions do not affect the allocation at all.
The code example above has a resource leak - CreatePen() is a WinAPI function that creates a pen. However, Cppcheck doesn't assume that return values from functions must be freed. There is no error message:
$ cppcheck pen1.c
Checking pen1.c...
If you provide a configuration file then Cppcheck detects the bug:
The allocation and deallocation functions are organized in groups. Each group is defined in a `<resource>` or `<memory>` tag and is identified by its `<dealloc>` functions. This means, groups with overlapping `<dealloc>` tags are merged.
To specify the behaviour of functions and how they should be used, `<function>` tags can be used. Functions are identified by their name, specified in the name attribute and their number of arguments. The name is a comma-separated list of function names. For functions in namespaces or classes, just provide their fully qualified name. For example: `<function name="memcpy,std::memcpy">`. If you have template functions then provide their instantiated names `<function name="dostuff<int>">`.
The arguments a function takes can be specified by `<arg>` tags. Each of them takes the number of the argument (starting from 1) in the nr attribute, `nr="any"` for arbitrary arguments, or `nr="variadic"` for variadic arguments. Optional arguments can be specified by providing a default value: `default="value"`. The specifications for individual arguments override this setting.
##### Not bool
Here is an example program with misplaced comparison:
void test()
{
if (MemCmp(buffer1, buffer2, 1024==0)) {}
}
Cppcheck assumes that it is fine to pass boolean values to functions:
$ cppcheck notbool.c
Checking notbool.c...
If you provide a configuration file then Cppcheck detects the bug:
$ cppcheck --library=notbool.cfg notbool.c
Checking notbool.c...
[notbool.c:5]: (error) Invalid MemCmp() argument nr 3. A non-boolean value is required.
Here is the minimal notbool.cfg
<?xml version="1.0"?>
<def>
<functionname="MemCmp">
<argnr="1"/>
<argnr="2"/>
<argnr="3">
<not-bool/>
</arg>
</function>
</def>
##### Uninitialized memory
Here is an example program:
void test()
{
char buffer1[1024];
char buffer2[1024];
CopyMemory(buffer1, buffer2, 1024);
}
The bug here is that buffer2 is uninitialized. The second argument for CopyMemory needs to be initialized. However, Cppcheck assumes that it is fine to pass uninitialized variables to functions:
$ cppcheck uninit.c
Checking uninit.c...
If you provide a configuration file then Cppcheck detects the bug:
Note that this implies for pointers that the memory they point at has to be initialized, too.
Here is the minimal windows.cfg:
<?xml version="1.0"?>
<def>
<functionname="CopyMemory">
<argnr="1"/>
<argnr="2">
<not-uninit/>
</arg>
<argnr="3"/>
</function>
</def>
##### Null pointers
Cppcheck assumes it's ok to pass NULL pointers to functions. Here is an example program:
void test()
{
CopyMemory(NULL, NULL, 1024);
}
The MSDN documentation is not clear if that is ok or not. But let's assume it's bad. Cppcheck assumes that it's ok to pass NULL to functions so no error is reported:
$ cppcheck null.c
Checking null.c...
If you provide a configuration file then Cppcheck detects the bug:
$ cppcheck --library=windows.cfg null.c
Checking null.c...
[null.c:3]: (error) Null pointer dereference
Note that this implies `<not-uninit>` as far as values are concerned. Uninitialized memory might still be passed to the function.
Here is a minimal windows.cfg file:
<?xml version="1.0"?>
<def>
<functionname="CopyMemory">
<argnr="1">
<not-null/>
</arg>
<argnr="2"/>
<argnr="3"/>
</function>
</def>
##### Format string
You can define that a function takes a format string. Example:
void test()
{
do_something("%i %i\n", 1024);
}
No error is reported for that:
$ cppcheck formatstring.c
Checking formatstring.c...
A configuration file can be created that says that the string is a format string. For instance:
<?xml version="1.0"?>
<def>
<functionname="do_something">
<formatstrtype="printf"/>
<argnr="1">
<formatstr/>
</arg>
</function>
</def>
Now Cppcheck will report an error:
$ cppcheck --library=test.cfg formatstring.c
Checking formatstring.c...
[formatstring.c:3]: (error) do_something format string requires 2 parameters but only 1 is given.
The type attribute can be either:
printf - format string follows the printf rules
scanf - format string follows the scanf rules
##### Value range
The valid values can be defined. Imagine:
void test()
{
do_something(1024);
}
No error is reported for that:
$ cppcheck valuerange.c
Checking valuerange.c...
A configuration file can be created that says that 1024 is out of bounds. For instance:
<?xml version="1.0"?>
<def>
<functionname="do_something">
<argnr="1">
<valid>0:1023</valid>
</arg>
</function>
</def>
Now Cppcheck will report an error:
$ cppcheck --library=test.cfg range.c
Checking range.c...
[range.c:3]: (error) Invalid do_something() argument nr 1. The value is 1024 but the valid values are '0-1023'.
Some example expressions you can use in the valid element:
0,3,5 => only values 0, 3 and 5 are valid
-10:20 => all values between -10 and 20 are valid
:0 => all values that are less or equal to 0 are valid
0: => all values that are greater or equal to 0 are valid
0,2:32 => the value 0 and all values between 2 and 32 are valid
-1.5:5.6 => all values between -1.5 and 5.6 are valid
A configuration file can for instance be created that says that the size of the buffer in argument 1 must be larger than the strlen of argument 2. For instance:
[minsize.c:4]: (error) Buffer is accessed out of bounds: str
There are different types of minsizes:
strlen
buffer size must be larger than other arguments string length. Example: see strcpy configuration in std.cfg
argvalue
buffer size must be larger than value in other argument. Example: see memset configuration in std.cfg
sizeof
buffer size must be larger than other argument buffer size. Example: see memcpy configuration in posix.cfg
mul
buffer size must be larger than multiplication result when multiplying values given in two other arguments. Typically one argument defines the element size and another element defines the number of elements. Example: see fread configuration in std.cfg
strz
With this you can say that an argument must be a zero-terminated string.
These correspond to the GCC function attributes `<pure>` and `<const>`.
A pure function has no effects except to return a value, and its return value depends only on the parameters and global variables.
A const function has no effects except to return a value, and its return value depends only on the parameters.
Here is an example code:
void f(int x)
{
if (calculate(x) == 213) {
} else if (calculate(x) == 213) {
// unreachable code
}
}
If calculate() is a const function then the result of calculate(x) will be the same in both conditions, since the same parameter value is used.
Cppcheck normally assumes that the result might be different, and reports no warning for the code:
$ cppcheck const.c
Checking const.c...
If a proper const.cfg is provided, the unreachable code is detected:
$ cppcheck --enable=style --library=const const.c
Checking const.c...
[const.c:7]: (style) Expression is always false because 'else if' condition matches previous condition at line 5.
Here is a minimal const.cfg file:
<?xml version="1.0"?>
<def>
<functionname="calculate">
<const/>
<argnr="1"/>
</function>
</def>
##### Example configuration for strcpy()
The proper configuration for the standard strcpy() function would be:
<functionname="strcpy">
<leak-ignore/>
<noreturn>false</noreturn>
<argnr="1">
<not-null/>
</arg>
<argnr="2">
<not-null/>
<not-uninit/>
<strz/>
</arg>
</function>
The `<leak-ignore/>` tells Cppcheck to ignore this function call in the leaks checking. Passing allocated memory to this function won't mean it will be deallocated.
The `<noreturn>` tells Cppcheck if this function returns or not.
The first argument that the function takes is a pointer. It must not be a null pointer, therefore `<not-null>` is used.
The second argument the function takes is a pointer. It must not be null. And it must point at initialized data. Using `<not-null>` and `<not-uninit>` is correct. Moreover it must point at a zero-terminated string so `<strz>` is also used.
Use this for integer/float/bool/pointer types. Not for structs/unions.
Lots of code relies on typedefs providing platform independent types. "podtype"-tags can be used to provide necessary information to cppcheck to support them. Without further information, cppcheck does not understand the type "uint16_t" in the following example:
void test() {
uint16_t a;
}
No message about variable 'a' being unused is printed:
$ cppcheck --enable=style unusedvar.cpp
Checking unusedvar.cpp...
If uint16_t is defined in a library as follows, the result improves:
<?xml version="1.0"?>
<def>
<podtypename="uint16_t"sign="u"size="2"/>
</def>
The size of the type is specified in bytes. Possible values for the "sign" attribute are "s" (signed) and "u" (unsigned). Both attributes are optional. Using this library, cppcheck prints:
A lot of C++ libraries, among those the STL itself, provide containers with very similar functionality. Libraries can be used to tell cppcheck about their behaviour. Each container needs a unique ID. It can optionally have a startPattern, which must be a valid Token::Match pattern and an endPattern that is compared to the linked token of the first token with such a link. The optional attribute "inherits" takes an ID from a previously defined container.
Inside the `<container>` tag, functions can be defined inside of the tags `<size>`, `<access>` and `<other>` (on your choice). Each of them can specify an action like "resize" and/or the result it yields, for example "end-iterator".
The following example provides a definition for std::vector, based on the definition of "stdContainer" (not shown):
You can convert the XML output from cppcheck into a HTML report. You'll need Python and the pygments module (http://pygments.org/) for this to work. In the Cppcheck source tree there is a folder htmlreport that contains a script that transforms a Cppcheck XML file into HTML output.
This command generates the help screen:
htmlreport/cppcheck-htmlreport -h
The output screen says:
Usage: cppcheck-htmlreport [options]
Options:
-h, --help show this help message and exit
--file=FILE The cppcheck xml output file to read defects from.
Default is reading from stdin.
--report-dir=REPORT_DIR
The directory where the html report content is written.
--source-dir=SOURCE_DIR
Base directory where source code files can be found.
