constVariable False Positive using >> operator

[Steve Albright]

The example below does not compile but just pretend it does. I have several of these issues so it is going to make upgrading to 2.60 more painful than it should be.

Also note, I have run into the situation before where << was used instead of append with Qt containers and I've typically just converted to append but I'd rather not have to fix that code but I didn't come up with a good example to submit but I've seen it multiple times.

#include <QByteArray>

void constVariableFalsePositiveExample(const QByteArray& buffer)
{
   unsigned char writeDataHereSoItCannotBeConst[2];

   for(int i = 0; i < 2; i++)
   {
      buffer >> writeDataHereSoItCannotBeConst[i]; // ignore compile error because operator int is private
   }

   //(void)writeDataHereSoItCannotBeConst; - odd that no warning with this line
}

[Daniel Marjamäki]

I would like to have a compilable example. If buffer is const it seems strange that you can perform a stream read from the object. I guess the compiler tries to compile your code as:

((int)buffer) >> writeDataHereSoItCannotBeConst[i]

and in this case the Cppcheck warning is correct.

[Steve Albright]

Can we not focus on buffer here and what type of class it is?

CPPCheck is saying to make the unsigned char a const but we are trying to populate it and therefore it can't be const.

Assume we have an override (friend function or whatever) for >> that reads from buffer (pretend buffer isn't const if that helps) and puts into the unsigned char array.

[Daniel Marjamäki]

I think it's important that it is valid code and compiles in this case. If we always assume that ">>" in various expressions is a stream read and bailout then there will be false negatives. We need to be strict.

Here is a example code I think seems more proper;

void foo(std::istream& buffer)
{
   unsigned char writeDataHereSoItCannotBeConst[2];

   for(int i = 0; i < 2; i++)
   {
      buffer >> writeDataHereSoItCannotBeConst[i];
   }
}

[Daniel Marjamäki]

I have created a ticket: https://trac.cppcheck.net/ticket/10547

[Steve Albright]

I was working on another repo and found this one where we are writing to *os so it can't be const.

style: constParameter - Parameter 'os' can be declared with const

void PrintTo(QString const& foo, ::std::ostream os)
{
os << '"' << foo.toStdString() << '"';
}

[CHR]

I think the problem is that the overloaded operators << are not part of either std.cfg or qt.cfg, so cppcheck assumes they behave like the regular ones for integers.

[Paul Fultz]

Cppcheck knows its a stream operator. The isLikelyStreamRead returns true. The problem is that checkConstPointer doesn't use isVariableChanged(which would recognize the variable being modified).