struct B
{
};
struct A
{
operator A&() const;
operator B&() const;
};
int main()
{
const A a;
B& br = a;
A& ar = a;
}
Why can I create cast operator to B&, but not to A&.
May be it does not have much sense (one can use it to erase const modifier, as in example), but it at least inconsistent!
You can't do this because it's explicitly forbidden. N3290 § 12.3.2 states:
Such functions are called
conversion functions. No return type can be specified. If a conversion function is a member function, the
type of the conversion function (8.3.5) is “function taking no parameter returning conversion-type-id”. A
conversion function is never used to convert a (possibly cv-qualified) object to the (possibly cv-qualified)
same object type (or a reference to it), to a (possibly cv-qualified) base class of that type (or a reference to
it), or to (possibly cv-qualified) void.
(Emphasis mine)
This is discussed further in a note:
These conversions are considered as standard conversions for the purposes of overload resolution (13.3.3.1, 13.3.3.1.4) and
therefore initialization (8.5) and explicit casts (5.2.9).
Which explains this decision - it would interfere with the built-in mechanics too much. (For little gain).
If you really want something non-const from a const object the only smart way to do this is constructing a new instance using the copy constructor.
As a work around you could introduce a lightweight intermediary (like a smart pointer):
struct B {};
struct A {};
namespace {
B b_inst;
A a_inst;
}
struct A_wrapper {
A& inst;
// This is perfectly fine: const alters the reference, not what it refers to
operator A&() const { return inst; }
operator B&() const { return b_inst; }
A_wrapper() : inst(a_inst) {}
};
int main() {
const A_wrapper a;
B& br = a;
A& ar = a;
}
But really, wanting to do this in the first place looks like a code smell.
The proper way to do this would be to use const_cast.
For example,
#include <iostream>
using namespace std;
void f(int* p) {
cout << *p << endl;
}
int main(void) {
const int a = 10;
const int* b = &a;
// Function f() expects int*, not const int*
// f(b);
int* c = const_cast<int*>(b);
f(c);
// Lvalue is const
// *b = 20;
// Undefined behavior
// *c = 30;
int a1 = 40;
const int* b1 = &a1;
int* c1 = const_cast<int*>(b1);
// Integer a1, the object referred to by c1, has
// not been declared const
*c1 = 50;
return 0;
}
Declaring a conversion to a reference to self is not ill-formed. Your problem comes at the time where your reference is initialized. As the type of the reference and the type of the initialization expression are the same, the reference is bound directly and your user defined conversion operator is never considered. Thus normal conversion rules apply and const conversion makes the code ill-formed.
Anyway, what your are doing is basically asking yourself to get shot in the foot. If you don't like constness, don't use it. If you do it consistently, it will never bother you, but it is not going to make you new friends.
Related
Consider this simple example:
template <class Type>
class smartref {
public:
smartref() : data(new Type) { }
operator Type&(){ return *data; }
private:
Type* data;
};
class person {
public:
void think() { std::cout << "I am thinking"; }
};
int main() {
smartref<person> p;
p.think(); // why does not the compiler try substituting Type&?
}
How do conversion operators work in C++? (i.e) when does the compiler try substituting the type defined after the conversion operator?
Some random situations where conversion functions are used and not used follow.
First, note that conversion functions are never used to convert to the same class type or to a base class type.
Conversion during argument passing
Conversion during argument passing will use the rules for copy initialization. These rules just consider any conversion function, disregarding of whether converting to a reference or not.
struct B { };
struct A {
operator B() { return B(); }
};
void f(B);
int main() { f(A()); } // called!
Argument passing is just one context of copy initialization. Another is the "pure" form using the copy initialization syntax
B b = A(); // called!
Conversion to reference
In the conditional operator, conversion to a reference type is possible, if the type converted to is an lvalue.
struct B { };
struct A {
operator B&() { static B b; return b; }
};
int main() { B b; 0 ? b : A(); } // called!
Another conversion to reference is when you bind a reference, directly
struct B { };
struct A {
operator B&() { static B b; return b; }
};
B &b = A(); // called!
Conversion to function pointers
You may have a conversion function to a function pointer or reference, and when a call is made, then it might be used.
typedef void (*fPtr)(int);
void foo(int a);
struct test {
operator fPtr() { return foo; }
};
int main() {
test t; t(10); // called!
}
This thing can actually become quite useful sometimes.
Conversion to non class types
The implicit conversions that happen always and everywhere can use user defined conversions too. You may define a conversion function that returns a boolean value
struct test {
operator bool() { return true; }
};
int main() {
test t;
if(t) { ... }
}
(The conversion to bool in this case can be made safer by the safe-bool idiom, to forbid conversions to other integer types.) The conversions are triggered anywhere where a built-in operator expects a certain type. Conversions may get into the way, though.
struct test {
void operator[](unsigned int) { }
operator char *() { static char c; return &c; }
};
int main() {
test t; t[0]; // ambiguous
}
// (t).operator[] (unsigned int) : member
// operator[](T *, std::ptrdiff_t) : built-in
The call can be ambiguous, because for the member, the second parameter needs a conversion, and for the built-in operator, the first needs a user defined conversion. The other two parameters match perfectly respectively. The call can be non-ambiguous in some cases (ptrdiff_t needs be different from int then).
Conversion function template
Templates allow some nice things, but better be very cautious about them. The following makes a type convertible to any pointer type (member pointers aren't seen as "pointer types").
struct test {
template<typename T>
operator T*() { return 0; }
};
void *pv = test();
bool *pb = test();
The "." operator is not overloadable in C++. And whenever you say x.y, no conversion will automatically be be performed on x.
Conversions aren't magic. Just because A has a conversion to B and B has a foo method doesn't mean that a.foo() will call B::foo().
The compiler tries to use a conversion in four situations
You explicitly cast a variable to another type
You pass the variable as an argument to a function that expects a different type in that position (operators count as functions here)
You assign the variable to a variable of a different type
You use the variable copy-construct or initialize a variable of a different type
There are three types of conversions, other than those involved with inheritance
Built-in conversions (e.g. int-to-double)
Implicit construction, where class B defines a constructor taking a single argument of type A, and does not mark it with the "explicit" keyword
User-defined conversion operators, where class A defines an operator B (as in your example)
How the compiler decides which type of conversion to use and when (especially when there are multiple choices) is pretty involved, and I'd do a bad job of trying to condense it into an answer on SO. Section 12.3 of the C++ standard discusses implicit construction and user-defined conversion operators.
(There may be some conversion situations or methods that I haven't thought of, so please comment or edit them if you see something missing)
Implicit conversion (whether by conversion operators or non-explicit constructors) occurs when passing parameters to functions (including overloaded and default operators for classes). In addition to this, there are some implicit conversions performed on arithmetic types (so adding a char and a long results in the addition of two longs, with a long result).
Implicit conversion does not apply to the object on which a member function call is made: for the purposes of implicit conversion, "this" is not a function parameter.
The compiler will attempt one(!) user-defined cast (implicit ctor or cast operator) if you try to use an object (reference) of type T where U is required.
The . operator, however, will always try to access a member of the object (reference) on its left side. That's just the way it's defined. If you want something more fancy, that's what operator->() can be overloaded for.
You should do
((person)p).think();
The compiler doesn't have the information for automatically casting to person, so you need explicit casting.
If you would use something like
person pers = p;
Then the compiler has information for implicit casting to person.
You can have "casting" through constructors:
class A
{
public:
A( int );
};
A a = 10; // Looks like a cast from int to A
These are some brief examples. Casting (implicit, explicit, etc) needs more to explain. You can find details in serious C++ books (see the questions about C++ books on stack overflow for good titles, like this one).
//Virtual table Fuction(VFT)
#include <iostream>
using namespace std;
class smartref {
public:
virtual char think() { }//for Late bindig make virtual function if not make virtual function of char think() {} then become early binding and pointer call this class function
smartref() : data(new char) { }
operator char(){ return *data; }
private:
char* data;
};
class person:public smartref
{
public:
char think() { std::cout << "I am thinking"; }
};
int main() {
smartref *p;//make pointer of class
person o1;//make object of class
p=&o1;//store object address in pointer
p->think(); // Late Binding in class person
return 0;
}
Consider this simple example:
template <class Type>
class smartref {
public:
smartref() : data(new Type) { }
operator Type&(){ return *data; }
private:
Type* data;
};
class person {
public:
void think() { std::cout << "I am thinking"; }
};
int main() {
smartref<person> p;
p.think(); // why does not the compiler try substituting Type&?
}
How do conversion operators work in C++? (i.e) when does the compiler try substituting the type defined after the conversion operator?
Some random situations where conversion functions are used and not used follow.
First, note that conversion functions are never used to convert to the same class type or to a base class type.
Conversion during argument passing
Conversion during argument passing will use the rules for copy initialization. These rules just consider any conversion function, disregarding of whether converting to a reference or not.
struct B { };
struct A {
operator B() { return B(); }
};
void f(B);
int main() { f(A()); } // called!
Argument passing is just one context of copy initialization. Another is the "pure" form using the copy initialization syntax
B b = A(); // called!
Conversion to reference
In the conditional operator, conversion to a reference type is possible, if the type converted to is an lvalue.
struct B { };
struct A {
operator B&() { static B b; return b; }
};
int main() { B b; 0 ? b : A(); } // called!
Another conversion to reference is when you bind a reference, directly
struct B { };
struct A {
operator B&() { static B b; return b; }
};
B &b = A(); // called!
Conversion to function pointers
You may have a conversion function to a function pointer or reference, and when a call is made, then it might be used.
typedef void (*fPtr)(int);
void foo(int a);
struct test {
operator fPtr() { return foo; }
};
int main() {
test t; t(10); // called!
}
This thing can actually become quite useful sometimes.
Conversion to non class types
The implicit conversions that happen always and everywhere can use user defined conversions too. You may define a conversion function that returns a boolean value
struct test {
operator bool() { return true; }
};
int main() {
test t;
if(t) { ... }
}
(The conversion to bool in this case can be made safer by the safe-bool idiom, to forbid conversions to other integer types.) The conversions are triggered anywhere where a built-in operator expects a certain type. Conversions may get into the way, though.
struct test {
void operator[](unsigned int) { }
operator char *() { static char c; return &c; }
};
int main() {
test t; t[0]; // ambiguous
}
// (t).operator[] (unsigned int) : member
// operator[](T *, std::ptrdiff_t) : built-in
The call can be ambiguous, because for the member, the second parameter needs a conversion, and for the built-in operator, the first needs a user defined conversion. The other two parameters match perfectly respectively. The call can be non-ambiguous in some cases (ptrdiff_t needs be different from int then).
Conversion function template
Templates allow some nice things, but better be very cautious about them. The following makes a type convertible to any pointer type (member pointers aren't seen as "pointer types").
struct test {
template<typename T>
operator T*() { return 0; }
};
void *pv = test();
bool *pb = test();
The "." operator is not overloadable in C++. And whenever you say x.y, no conversion will automatically be be performed on x.
Conversions aren't magic. Just because A has a conversion to B and B has a foo method doesn't mean that a.foo() will call B::foo().
The compiler tries to use a conversion in four situations
You explicitly cast a variable to another type
You pass the variable as an argument to a function that expects a different type in that position (operators count as functions here)
You assign the variable to a variable of a different type
You use the variable copy-construct or initialize a variable of a different type
There are three types of conversions, other than those involved with inheritance
Built-in conversions (e.g. int-to-double)
Implicit construction, where class B defines a constructor taking a single argument of type A, and does not mark it with the "explicit" keyword
User-defined conversion operators, where class A defines an operator B (as in your example)
How the compiler decides which type of conversion to use and when (especially when there are multiple choices) is pretty involved, and I'd do a bad job of trying to condense it into an answer on SO. Section 12.3 of the C++ standard discusses implicit construction and user-defined conversion operators.
(There may be some conversion situations or methods that I haven't thought of, so please comment or edit them if you see something missing)
Implicit conversion (whether by conversion operators or non-explicit constructors) occurs when passing parameters to functions (including overloaded and default operators for classes). In addition to this, there are some implicit conversions performed on arithmetic types (so adding a char and a long results in the addition of two longs, with a long result).
Implicit conversion does not apply to the object on which a member function call is made: for the purposes of implicit conversion, "this" is not a function parameter.
The compiler will attempt one(!) user-defined cast (implicit ctor or cast operator) if you try to use an object (reference) of type T where U is required.
The . operator, however, will always try to access a member of the object (reference) on its left side. That's just the way it's defined. If you want something more fancy, that's what operator->() can be overloaded for.
You should do
((person)p).think();
The compiler doesn't have the information for automatically casting to person, so you need explicit casting.
If you would use something like
person pers = p;
Then the compiler has information for implicit casting to person.
You can have "casting" through constructors:
class A
{
public:
A( int );
};
A a = 10; // Looks like a cast from int to A
These are some brief examples. Casting (implicit, explicit, etc) needs more to explain. You can find details in serious C++ books (see the questions about C++ books on stack overflow for good titles, like this one).
//Virtual table Fuction(VFT)
#include <iostream>
using namespace std;
class smartref {
public:
virtual char think() { }//for Late bindig make virtual function if not make virtual function of char think() {} then become early binding and pointer call this class function
smartref() : data(new char) { }
operator char(){ return *data; }
private:
char* data;
};
class person:public smartref
{
public:
char think() { std::cout << "I am thinking"; }
};
int main() {
smartref *p;//make pointer of class
person o1;//make object of class
p=&o1;//store object address in pointer
p->think(); // Late Binding in class person
return 0;
}
I have a class that 'remembers' a reference to some object (e.g. an integer variable). I can't have it reference a value that's destructed immediately, and I'm looking for a way to protect the users of my class from doing so by accident.
Is an rvalue-reference overload a good way to prevent a temporary to be passed in?
struct HasRef {
int& a;
HasRef(int& a):a(a){}
void foo(){ a=1; }
};
int main(){
int x=5;
HasRef r1(x);
r1.foo(); // works like intended.
HasRef r2(x+4);
r2.foo(); // dereferences the temporary created by x+4
}
Would a private rvalue overload do?
struct HasRef {
int& a;
HasRef( int& a ):a(a){}
void foo(){ a=1; }
private:
HasRef( int&& a );
};
... HasRef r2(x+1); // doesn't compile => problem solved?
Are there any pitfalls I didn't see?
If you have to store a const reference to some instance of type B into your class A, then surely you want to be ensured, that lifetime of A instance will be exceeded by the lifetime of B instance:
B b{};
A a1{b}; // allowed
A a2{B{}}; // should be denied
B const f() { return B{}; } // const result type may make sense for user-defined types
A a3{f()}; // should also be denied!
To make it possible you should explicitly to = delete; all the constructor overloadings, which can accept rvalues (both const && and &&). For this to achieve you should just to = delete; only const && version of constructor.
struct B {};
struct A
{
B const & b;
A(B const & bb) : b(bb) { ; } // accepts only `B const &` and `B &`
A(B const &&) = delete; // prohibits both `B &&` and `B const &&`
};
This approach allows you to prohibit passing to the constructor all kinds of rvalues.
This also works for built-in scalars. For example, double const f() { return 0.01; }, though it cause a warning like:
warning: 'const' type qualifier on return type has no effect [-Wignored-qualifiers]
it still can has effect if you just = delete; only && version of constructor:
struct A
{
double const & eps;
A(double const & e) : eps(e) {} // binds to `double const &`, `double &` AND ! `double const &&`
A(double &&) = delete; // prohibit to binding only to `double &&`, but not to `double const &&`
};
double const get_eps() { return 0.01; }
A a{0.01}; // hard error
A a{get_eps()}; // no hard error, but it is wrong!
For non-conversion constructors (i.e. non-unary) there is an issue: you may have to provide = delete;-d versions for all the combinatorically possible versions of constructors as follows:
struct A
{
A(B const &, C const &) {}
A(B const &&, C const &&) = delete;
// and also!
A(B const &, C const &&) = delete;
A(B const &&, C const &) = delete;
};
to prohibit mixed-cases like:
B b{};
A a{b, C{}};
Ignoring the fact the code isn't valid and just answering the question about the private overload...
In C++11 I would prefer a deleted function to a private function. It's a bit more explicit that you really can't call it (not even if you're a member or friend of the class.)
N.B. if the deleted constructor is HasRef(int&&)=delete it will not be chosen here:
int i;
HasRef hr(std::forward<const int>(i));
With an argument of type const int&& the HasRef(const int&) constructor would be used, not the HasRef(int&&) one. In this case it would be OK, because i really is an lvalue, but in general that might not be the case, so this might be one of the very rare times when a const rvalue reference is useful:
HasRef(const int&&) = delete;
That shouldn't compile. A good C++ compiler (or really almost any C++ compiler that I've ever seen) will stop that from happening.
I'm guessing you're compiling in MSVS. In that case, turn off language extensions and you should get an error.
Otherwise, not that even marking the reference const extends the lifetime of the temporary until the constructor finishes. After that, you'll refer to an invalid object.
Consider a type bar which has user-defined conversion operators to references of type bar:
struct bar
{
operator bar & ();
operator const bar & () const;
};
When would these conversions be applied? Moreover, what does it imply if these operators were deleted? Is there any interesting use of either feature?
The following program does not appear to apply either conversion:
#include <iostream>
struct bar
{
operator bar & ()
{
std::cout << "operator bar &()" << std::endl;
return *this;
}
operator const bar & () const
{
std::cout << "operator const bar &() const" << std::endl;
return *this;
}
};
void foo(bar x)
{
}
int main()
{
bar x;
bar y = x; // copy, no conversion
y = x; // assignment, no conversion
foo(x); // copy, no conversion
y = (bar&)x; // no output
y = (const bar&)x; // no output
return 0;
}
C++11 §12.3.2
A conversion function is never used to convert a (possibly cv-qualified) object to the (possibly cv-qualified) same object type (or a reference to it), to a (possibly cv-qualified) base class of that type (or a reference to it), or to (possibly cv-qualified) void
The feature that you are allowed to define a conversion function from a type to itself, but that the conversion function is never used, can be a useful feature in template programming, where two type parameters may or may not refer to the same type, depending on the instantiation. Some code of mine relies on this feature. It saves having to provide specializations for cases where two or more of the type parameters end up referring to the same type.
I can't see any reason why it would ever be called. The conversion functions are called to... convert. If you already have the right type, there is absolutely no reason to add a conversion operation before the copy.
Just for the record. I managed to create this kind of structure in a bigger software project, blindly trusted the compiler warning and removed the "never used method". Well, I guess I found a scenario where it is actually called. The compiler seems to miss the base class.
#include <iostream>
struct D;
struct B
{
virtual operator D& ()
{
throw "foo";
}
};
struct D : public B
{
virtual operator D& ()
{
std::cout << "bar" << std::endl;
return *this;
}
};
int main()
{
B* b = new D();
D& d = *b;
return 0;
}
The following pattern has arisen in a program I'm writing. I hope it's not too contrived, but it manages to mutate a Foo object in the const method Foo::Questionable() const, without use of any const_cast or similar. Basically, Foo stores a reference to FooOwner and vice versa, and in Questionable(), Foo manages to modify itself in a const method by calling mutate_foo() on its owner. Questions follow the code.
#include "stdafx.h"
#include <iostream>
using namespace std;
class FooOwner;
class Foo {
FooOwner& owner;
int data;
public:
Foo(FooOwner& owner_, int data_)
: owner(owner_),
data(data_)
{
}
void SetData(int data_)
{
data = data_;
}
int Questionable() const; // defined after FooOwner
};
class FooOwner {
Foo* pFoo;
public:
FooOwner()
: pFoo(NULL)
{}
void own(Foo& foo)
{
pFoo = &foo;
}
void mutate_foo()
{
if (pFoo != NULL)
pFoo->SetData(0);
}
};
int Foo::Questionable() const
{
owner.mutate_foo(); // point of interest
return data;
}
int main()
{
FooOwner foo_owner;
Foo foo(foo_owner, 0); // foo keeps reference to foo_owner
foo_owner.own(foo); // foo_owner keeps pointer to foo
cout << foo.Questionable() << endl; // correct?
return 0;
}
Is this defined behavior? Should Foo::data be declared mutable? Or is this a sign I'm doing things fatally wrong? I'm trying to implement a kind of lazy-initialised 'data' which is only set when requested, and the following code compiles fine with no warnings, so I'm a little nervous I'm in UB land.
Edit: the const on Questionable() only makes immediate members const, and not the objects pointed to or referenced by the object. Does this make the code legal? I'm confused between the fact that in Questionable(), this has the type const Foo*, and further down the call stack, FooOwner legitimately has a non-const pointer it uses to modify Foo. Does this mean the Foo object can be modified or not?
Edit 2: perhaps an even simpler example:
class X {
X* nonconst_this; // Only turns in to X* const in a const method!
int data;
public:
X()
: nonconst_this(this),
data(0)
{
}
int GetData() const
{
nonconst_this->data = 5; // legal??
return data;
}
};
Consider the following:
int i = 3;
i is an object, and it has the type int. It is not cv-qualified (is not const or volatile, or both.)
Now we add:
const int& j = i;
const int* k = &i;
j is a reference which refers to i, and k is a pointer which points to i. (From now on, we simply combine "refer to" and "points to" to just "points to".)
At this point, we have two cv-qualified variables, j and k, that point to a non-cv-qualified object. This is mentioned in §7.1.5.1/3:
A pointer or reference to a cv-qualified type need not actually point or refer to a cv-qualified object, but it is treated as if it does; a const-qualified access path cannot be used to modify an object even if the object referenced is a non-const object and can be modified through some other access path. [Note: cv-qualifiers are supported by the type system so that they cannot be subverted without casting (5.2.11). ]
What this means is that a compiler must respect that j and k are cv-qualified, even though they point to a non-cv-qualified object. (So j = 5 and *k = 5 are illegal, even though i = 5 is legal.)
We now consider removing the const from those:
const_cast<int&>(j) = 5;
*const_cast<int*>(k) = 5;
This is legal (§refer to 5.2.11), but is it undefined behavior? No. See §7.1.5.1/4:
Except that any class member declared mutable (7.1.1) can be modified, any attempt to modify a const object during its lifetime (3.8) results in undefined behavior.
Emphasis mine.
Remember that i is not const and that j and k both point to i. All we've done is tell the type system to remove the const-qualifier from the type so we can modify the pointed to object, and then modified i through those variables.
This is exactly the same as doing:
int& j = i; // removed const with const_cast...
int* k = &i; // ..trivially legal code
j = 5;
*k = 5;
And this is trivially legal. We now consider that i was this instead:
const int i = 3;
What of our code now?
const_cast<int&>(j) = 5;
*const_cast<int*>(k) = 5;
It now leads to undefined behavior, because i is a const-qualified object. We told the type system to remove const so we can modify the pointed to object, and then modified a const-qualified object. This is undefined, as quoted above.
Again, more apparent as:
int& j = i; // removed const with const_cast...
int* k = &i; // ...but this is not legal!
j = 5;
*k = 5;
Note that simply doing this:
const_cast<int&>(j);
*const_cast<int*>(k);
Is perfectly legal and defined, as no const-qualified objects are being modified; we're just messing with the type-system.
Now consider:
struct foo
{
foo() :
me(this), self(*this), i(3)
{}
void bar() const
{
me->i = 5;
self.i = 5;
}
foo* me;
foo& self;
int i;
};
What does const on bar do to the members? It makes access to them go through something called a cv-qualified access path. (It does this by changing the type of this from T* const to cv T const*, where cv is the cv-qualifiers on the function.)
So what are the members types during the execution of bar? They are:
// const-pointer-to-non-const, where the pointer points cannot be changed
foo* const me;
// foo& const is ill-formed, cv-qualifiers do nothing to reference types
foo& self;
// same as const int
int const i;
Of course, the types are irrelevant, as the important thing is the const-qualification of the pointed to objects, not the pointers. (Had k above been const int* const, the latter const is irrelevant.) We now consider:
int main()
{
foo f;
f.bar(); // UB?
}
Within bar, both me and self point to a non-const foo, so just like with int i above we have well-defined behavior. Had we had:
const foo f;
f.bar(); // UB!
We would have had UB, just like with const int, because we would be modifying a const-qualified object.
In your question, you have no const-qualified objects, so you have no undefined behavior.
And just to add an appeal to authority, consider the const_cast trick by Scott Meyers, used to recycle a const-qualified function in a non-const function:
struct foo
{
const int& bar() const
{
int* result = /* complicated process to get the resulting int */
return *result;
}
int& bar()
{
// we wouldn't like to copy-paste a complicated process, what can we do?
}
};
He suggests:
int& bar(void)
{
const foo& self = *this; // add const
const int& result = self.bar(); // call const version
return const_cast<int&>(result); // take off const
}
Or how it's usually written:
int& bar(void)
{
return const_cast<int&>( // (3) remove const from result
static_cast<const foo&>(*this) // (1) add const to this
.bar() // (2) call const version
);
}
Note this is, again, perfectly legal and well-defined. Specifically, because this function must be called on a non-const-qualified foo, we are perfectly safe in stripping the const-qualification from the return type of int& boo() const.
(Unless someone shoots themselves with a const_cast + call in the first place.)
To summarize:
struct foo
{
foo(void) :
i(),
self(*this), me(this),
self_2(*this), me_2(this)
{}
const int& bar() const
{
return i; // always well-formed, always defined
}
int& bar() const
{
// always well-formed, always well-defined
return const_cast<int&>(
static_cast<const foo&>(*this).
bar()
);
}
void baz() const
{
// always ill-formed, i is a const int in baz
i = 5;
// always ill-formed, me is a foo* const in baz
me = 0;
// always ill-formed, me_2 is a const foo* const in baz
me_2 = 0;
// always well-formed, defined if the foo pointed to is non-const
self.i = 5;
me->i = 5;
// always ill-formed, type points to a const (though the object it
// points to may or may not necessarily be const-qualified)
self_2.i = 5;
me_2->i = 5;
// always well-formed, always defined, nothing being modified
// (note: if the result/member was not an int and was a user-defined
// type, if it had its copy-constructor and/or operator= parameter
// as T& instead of const T&, like auto_ptr for example, this would
// be defined if the foo self_2/me_2 points to was non-const
int r = const_cast<foo&>(self_2).i;
r = const_cast<foo* const>(me_2)->i;
// always well-formed, always defined, nothing being modified.
// (same idea behind the non-const bar, only const qualifications
// are being changed, not any objects.)
const_cast<foo&>(self_2);
const_cast<foo* const>(me_2);
// always well-formed, defined if the foo pointed to is non-const
// (note, equivalent to using self and me)
const_cast<foo&>(self_2).i = 5;
const_cast<foo* const>(me_2)->i = 5;
// always well-formed, defined if the foo pointed to is non-const
const_cast<foo&>(*this).i = 5;
const_cast<foo* const>(this)->i = 5;
}
int i;
foo& self;
foo* me;
const foo& self_2;
const foo* me_2;
};
int main()
{
int i = 0;
{
// always well-formed, always defined
int& x = i;
int* y = &i;
const int& z = i;
const int* w = &i;
// always well-formed, always defined
// (note, same as using x and y)
const_cast<int&>(z) = 5;
const_cast<int*>(w) = 5;
}
const int j = 0;
{
// never well-formed, strips cv-qualifications without a cast
int& x = j;
int* y = &j;
// always well-formed, always defined
const int& z = i;
const int* w = &i;
// always well-formed, never defined
// (note, same as using x and y, but those were ill-formed)
const_cast<int&>(z) = 5;
const_cast<int*>(w) = 5;
}
foo x;
x.bar(); // calls non-const, well-formed, always defined
x.bar() = 5; // calls non-const, which calls const, removes const from
// result, and modifies which is defined because the object
// pointed to by the returned reference is non-const,
// because x is non-const.
x.baz(); // well-formed, always defined
const foo y;
y.bar(); // calls const, well-formed, always defined
const_cast<foo&>(y).bar(); // calls non-const, well-formed,
// always defined (nothing being modified)
const_cast<foo&>(y).bar() = 5; // calls non-const, which calls const,
// removes const from result, and
// modifies which is undefined because
// the object pointed to by the returned
// reference is const, because y is const.
y.baz(); // well-formed, always undefined
}
I refer to the ISO C++03 standard.
IMO, you are not doing anything technically wrong. May-be it would be simpler to understand if the member was a pointer.
class X
{
Y* m_ptr;
void foo() const {
m_ptr = NULL; //illegal
*m_ptr = 42; //legal
}
};
const makes the pointer const, not the pointee.
Consider the difference between:
const X* ptr;
X* const ptr; //this is what happens in const member functions
As to references, since they can't be reseated anyway, the const keyword on the method has no effect whatsoever on reference members.
In your example, I don't see any const objects, so you are not doing anything bad, just exploiting a strange loophole in the way const correctness works in C++.
Without actually getting to whether it is/should/could be allowed, I would greatly advice against it. There are mechanisms in the language for what you want to achieve that don't require writing obscure constructs that will most probably confuse other developers.
Look into the mutable keyword. That keyword can be used to declare members that can be modified within const member methods as they do not affect the perceivable state of the class. Consider class that gets initialized with a set of parameters and performs a complex expensive calculation that may not be needed always:
class ComplexProcessor
{
public:
void setInputs( int a, int b );
int getValue() const;
private:
int complexCalculation( int a, int b );
int result;
};
A possible implementation is adding the result value as a member and calculating it for each set:
void ComplexProcessor::setInputs( int a, int b ) {
result = complexCalculation( a, b );
}
But this means that the value is calculated in all sets, whether it is needed or not. If you think on the object as a black box, the interface just defines a method to set the parameters and a method to retrieve the calculated value. The instant when the calculation is performed does not really affect the perceived state of the object --as far as the value returned by the getter is correct. So we can modify the class to store the inputs (instead of the outputs) and calculate the result only when needed:
class ComplexProcessor2 {
public:
void setInputs( int a, int b ) {
a_ = a; b_ = b;
}
int getValue() const {
return complexCalculation( a_, b_ );
}
private:
int complexCalculation( int a, int b );
int a_,b_;
};
Semantically the second class and the first class are equivalent, but now we have avoided to perform the complex calculation if the value is not needed, so it is an advantage if the value is only requested in some cases. But at the same time it is a disadvantage if the value is requested many times for the same object: each time the complex calculation will be performed even if the inputs have not changed.
The solution is caching the result. For that we can the result to the class. When the result is requested, if we have already calculated it, we only need to retrieve it, while if we do not have the value we must calculate it. When the inputs change we invalidate the cache. This is when the mutable keyword comes in handy. It tells the compiler that the member is not part of the perceivable state and as such it can be modified within a constant method:
class ComplexProcessor3 {
public:
ComplexProcessor3() : cached_(false) {}
void setInputs( int a, int b ) {
a_ = a; b_ = b;
cached_ = false;
}
int getValue() const {
if ( !cached_ ) {
result_ = complexCalculation( a_, b_ );
cached_ = true;
}
return result_;
}
private:
int complexCalculation( int a, int b );
int a_,b_;
// This are not part of the perceivable state:
mutable int result_;
mutable bool cached_;
};
The third implementation is semantically equivalent to the two previous versions, but avoid having to recalculate the value if the result is already known --and cached.
The mutable keyword is needed in other places, like in multithreaded applications the mutex in classes are often marked as mutable. Locking and unlocking a mutex are mutating operations for the mutex: its state is clearly changing. Now, a getter method in an object that is shared among different threads does not modify the perceived state but must acquire and release the lock if the operation has to be thread safe:
template <typename T>
class SharedValue {
public:
void set( T v ) {
scoped_lock lock(mutex_);
value = v;
}
T get() const {
scoped_lock lock(mutex_);
return value;
}
private:
T value;
mutable mutex mutex_;
};
The getter operation is semantically constant, even if it needs to modify the mutex to ensure single threaded access to the value member.
The const keyword is only considered during compile time checks. C++ provides no facilities to protect your class against any memory access, which is what you are doing with your pointer/reference. Neither the compiler nor the runtime can know if your pointer points to an instance that you declared const somewhere.
EDIT:
Short example (might not compile):
// lets say foo has a member const int Foo::datalength() const {...}
// and a read only acces method const char data(int idx) const {...}
for (int i; i < foo.datalength(); ++i)
{
foo.questionable(); // this will most likely mess up foo.datalength !!
std::cout << foo.data(i); // HERE BE DRAGONS
}
In this case, the compiler might decide, ey, foo.datalength is const,
and the code inside the loop promised not to change foo, so I have to evaluate
datalength only once when I enter the loop. Yippie!
And if you try to debug this error, which will most likely only turn up if you compile with optimizations (not in the debug builds) you will drive yourself crazy.
Keep the promises! Or use mutable with your braincells on high alert!
You have reached circular dependencies. See FAQ 39.11 And yes, modifying const data is UB even if you have circumvented the compiler. Also, you are severely impairing the compiler's capacity to optimize if you don't keep your promises (read: violate const).
Why is Questionable const if you know that you will modify it via a call to its owner? Why does the owned object need to know about the owner? If you really really need to do that then mutable is the way to go. That is what it is there for -- logical constness (as opposed to strict bit level constness).
From my copy of the draft n3090:
9.3.2 The this pointer [class.this]
1 In the body of a non-static (9.3) member function, the keyword this is an rvalue a prvalue expression whose
value is the address of the object for which the function is called. The type of this in a member function
of a class X is X*. If the member function is declared const, the type of this is const X*, if the member
function is declared volatile, the type of this is volatile X*, and if the member function is declared
const volatile, the type of this is const volatile X*.
2 In a const member function, the object for which the function is called is accessed through a const access
path; therefore, a const member function shall not modify the object and its non-static data members.
[Note emphasis mine].
On UB:
7.1.6.1 The cv-qualifiers
3 A pointer or reference to a cv-qualified type need not actually
point or refer to a cv-qualified
object, but it is treated as if it
does; a const-qualified access path
cannot be used to modify an object
even if the object referenced is a
non-const object and can be modified
through some other access path. [
Note: cv-qualifiers are supported by
the type system so that they cannot be
subverted without casting (5.2.11).
—end note ]
4 Except that any class
member declared mutable (7.1.1) can be
modified, any attempt to modify a
const object during its lifetime (3.8)
results in undefined behavior.