Analysis failed (Cppcheck 1.82)

[Set]

Cppcheck 1.82 failed running analysis of the following simplified code:

#define EVTDELEG(classname,arg1type) template<typename T>\
class C##classname {\
public:\
    typedef void (T::*fnptr)(##arg1type);\
    C##classname( T* object, fnptr fn ) : m_object(object), m_fn(fn) { }\
    void Invoke( ##arg1type value ) {\
        (m_object->*m_fn)( value );\
    }\
private:\
    T*      m_object;\
    fnptr   m_fn;\
};

class EvtH {
    // For each of the events a delegate needs to be declared.
    EVTDELEG(CDG, bool)
};

void main() {
    return 0;
}

Online demo (http://cppcheck.sourceforge.net/demo/) reports:

Cppcheck 1.82

[test.cpp:15]: (error) Analysis failed. If the code is valid then please report this failure.

Done!

I tried to create a ticket in Trac, but I was not able to create an account (tried https://webchat.freenode.net/?channels=#cppcheck and then also this forum: https://sourceforge.net/p/cppcheck/discussion/development/thread/34ddb3ae/#5be5).

[versat]

How do you compile this code?
I tried g++ 6.4.0 with "g++ --std=c++17 analysis_failed.cpp" and got the following output:

analysis_failed.cpp:4:29: error: pasting "(" and "bool" does not give a valid preprocessing token
     typedef void (T::*fnptr)(##arg1type);\
                             ^
analysis_failed.cpp:16:5: note: in expansion of macro ‘EVTDELEG’
     EVTDELEG(CDG, bool)
     ^~~~~~~~
analysis_failed.cpp:6:16: error: pasting "(" and "bool" does not give a valid preprocessing token
     void Invoke( ##arg1type value ) {\
                ^
analysis_failed.cpp:16:5: note: in expansion of macro ‘EVTDELEG’
     EVTDELEG(CDG, bool)
     ^~~~~~~~
analysis_failed.cpp:19:11: error: ‘::main’ must return ‘int’
 void main() {
           ^

[Set]

Thank you for the response. I am sorry for the delay (I had a vacation).
You are correct. With simplification of the original code I left return 0; while main() is void. Just replace void with int and this code compiles. I am using Visual Studio 2015 (C++ CLR). Nevertheless, CPPCheck should not fail such way despite the code is not compilable.

[versat]

Ah ok, with Visual Studio i also do not get the preprocessor errors.
GCC and Visual Studio handle the token pasting operator differently:
According to https://complete-concrete-concise.com/programming/c/preprocessor-the-token-pasting-operator both compilers are working correctly, the standard does not define everything.
So i think Cppcheck should handle this code correctly too and should not issue an analysis failed warning.
I will create a ticket for that.

[versat]

I created the ticket: https://trac.cppcheck.net/ticket/8427

[Set]

Thank you. Nevertheless, amai marked this as a duplicate ticket and closed it. I hope that the developers will finally take care about it.

[versat]

It seems to be really a duplicate of his older ticket and there the result was that it seems to be invalid code.
If you have any information/links that this is really valid code that should be accepted by static analyzers that may help.

[Daniel Marjamäki]

Please tell me why you have the ## in the code :

typedef void (T::*fnptr)(##arg1type);

Is it by mistake?

This usage of ## seems pointless to me and it makes the code invalid/undefined.

[Set]

I agree, you are right. ## is useless in this case. It concatenates an empty token with a token arg1type. Anyway, it is compilable by Visual Studio (at least 2017, without any error or warning).
This is a simplified fragment of a legacy code in our huge stream which is going to be replaced soon (I hope). Nevertheless, the point I wrote about this is that I considered Cppcheck encountered a serious internal error (because it "failed") which is good to report. Failure of a program represents a backlog from my point of view - therefore I just wanted to inform about such an issue to help to improve the behaviour of future versions of Cppcheck. It could inform the user in a better way than with a failure in my opinion.

[Daniel Marjamäki]

For information, I want simplecpp to be compatible with gcc/msvc.. but this feels like a bug in the msvc preprocessor so I will not try to implement this.

It could inform the user in a better way than with a failure in my opinion.

Yes I totally agree. I created this issue: https://github.com/danmar/simplecpp/issues/121

[Set]

Ok. I have something else now. The following code gives: "[test.cpp:13]: (error) Array 'pdispparams->rgvarg[2]' accessed at index 2, which is out of bounds." This error message is suppressed if another assert() is added (just uncomment the last assert and you will see). I do not undersand the behaviour of CPPCheck here.

The following fragment of code is compilable in VS2017.

#include <atlbase.h>
#include <assert.h>

class CBBox {
public:
    CBBox() {}
    STDMETHOD(Invoke)(DISPPARAMS* pdispparams, VARIANT*) {
        assert(pdispparams->cArgs == 3);
        assert(pdispparams->rgvarg[0].vt == VT_R8);
        assert(pdispparams->rgvarg[1].vt == VT_I4);
        assert(pdispparams->rgvarg[2].vt == VT_BSTR);
        //assert(pdispparams->rgvarg[3].vt == VT_BSTR); // Uncommenting this line suppresses the error message of the line below
        CComBSTR name = pdispparams->rgvarg[2].bstrVal;
        return S_OK;
    }
};

int main() {
    return 0;
}

[versat]

I can reproduce this with the current development version of Cppcheck:

$ ./cppcheck --debug-warnings array_outofbounds.cpp
Checking array_outofbounds.cpp ...
[array_outofbounds.cpp:13]: (error) Array 'pdispparams->rgvarg[2]' accessed at index 2, which is out of bounds.
[array_outofbounds.cpp:8]: (debug) Unknown type 'pdispparams.'.
[array_outofbounds.cpp:11]: (debug) Unknown type 'pdispparams.'.

IMHO the error is a false positive and i have no idea how Cppcheck knows (or think it knows) the size of the array pdispparams->rgvarg[]. I am not sure but guessing such things out of asserts is not yet implemented and even if it would be i can not see how this error is justified then.

But i am no insider so maybe one of the core developers has an idea what is going on here.