Why is initialization of enum class temporaries with arbitrary values allowed? - c++

I came across some code like the following in one the CppCon 2014 talks that confused the heck out of me. The audience accepted it without comment, so I presume that it's legal:
enum class Foo { Bar };
Foo const v1 = Foo(5);
The question is: why does this compile? I would expect compilation to fail and complain that we can't convert an int to a Foo. The slightly modified line below fails with the expected error:
Foo const v1(5);

Scoped enumeration types have an implicit underlying type of int, assuming no other underlying type is specified. All possible values of type int can be represented.
7.2p5:
[...] For a scoped enumeration type, the underlying type is int if it is not explicitly specified. In both of these cases, the underlying type
is said to be fixed. [...]
7.2p8:
For an enumeration whose underlying type is fixed, the values of the enumeration are the values of the underlying type. [...]
And any integral value that can be represented by the enumeration can be explicitly converted to that enumeration type, as #Columbo had pointed out in his now-deleted answer:
5.2.9p10:
A value of integral or enumeration type can be explicitly converted to an enumeration type. The value is unchanged if the original value is within the range of the enumeration values (7.2). [...]
Since there is some confusion in the comments about what that means:
enum class Foo { Bar };
Foo const v1 = Foo(5);
is well-defined. Not undefined, not unspecified, not even implementation-defined. The parts of the standard I quote explain that:
The underlying type of Foo is int, and that the underlying type is fixed.
The values of Foo are the values of int.
Since 5 is in the range of the enumeration values, the value is unchanged by the conversion.

Related

Converting integral type to enum: functional cast vs initialization

Assuming there is an enum like this:
enum foo: int {
first,
second
}
Then I use it as follows:
foo f(1); // error: cannot initialize a variable of type 'foo' with an rvalue of type 'int'
foo f = foo(1); // OK !
I was wondering what is the difference between the two ?
I understand that the second version can be seen as a functional-style cast but why does this make any difference ?
For example, if I do this:
class Bar {};
Bar b = Bar(1); // no matching conversion for functional-style cast from 'int' to 'Bar'
I obviously get an error which makes sense. Therefore, this leads me to believe that in order for the second version of the foo example above to work there must be a conversion from int to enum defined somewhere but if there is such a conversion then why do I get an error in the first version ?
I do apologize if this is a duplicate. I am suspecting it is.
This seems relevant: Is this a cast or a construction?
... but not quit.
Thanks in advance !
Yes, the two forms are quite different, in a subtle way. Let's look at the first one, which results in an error. It's initialization of f, of type foo, from an int. It's described here, emphasis mine:
[dcl.init]/17.8
Otherwise, the initial value of the object being initialized is the
(possibly converted) value of the initializer expression. Standard
conversions will be used, if necessary, to convert the initializer
expression to the cv-unqualified version of the destination type; no
user-defined conversions are considered. If the conversion cannot be
done, the initialization is ill-formed.
The pertinent conversions in this case are integral conversions, mainly the one specified by the following:
[conv.integral]/1
A prvalue of an unscoped enumeration type can be converted to a
prvalue of an integer type.
So a an unscoped enumeration can be converted to an integer implicitly, but the converse is not true. Which is why the initialization is ill-formed. However, that functional-style cast notation is essentially a static cast. And a static cast can perform the inverse of (almost) any valid standard conversion. So the casted 1 is then used to initialize f, but at this point we are copy-initializing from a foo prvalue, which is of course perfectly fine.

Why am I allowed to use an incomplete enum class?

Why does the code below compile without any errors?
enum class Enumeration;
void func()
{
auto enumeration = static_cast<Enumeration>(2);
auto value = static_cast<int>(enumeration);
}
It compiles because the compiler knows at compile time the size of Enumeration (which happens to be empty).
You see it explicitly using the following syntax:
enum class Enumeration : short;
The compiler knows everything there is to know about the Enumeration.
Enumeration is a opaque-enum-declaration which means also that the type is complete i.e. you can use sizeofon it. If needed you can specify the list of enumerators in a later redeclaration (unless the redeclaration comes with a different underlying type, obviously).
Note that since you are using enum class usage of static_cast is mandatory.
Strongly typed enum does not allow implicit conversion to int but you can safely use static_cast on them to retrieve their integral value.
They are still enum afterall.
Quoting cppreference
There are no implicit conversions from the values of a scoped
enumerator to integral types, although static_cast may be used to
obtain the numeric value of the enumerator.
More on this topic here: How to automatically convert strongly typed enum into int?

Type of an enumerator in the declaration of its enum

In C++, specifically in C++14 n4296, there are two paragraps talking about type of an enumerator, which seem to be contradictory to me. See 7.2/5 (which is 10.2/5 in n4659):
Each enumeration defines a type that is different from all other types. Each enumeration also has an underlying type. The underlying type can be explicitly specified using an enum-base. For a scoped enumeration type, the underlying type is int if it is not explicitly specified. In both of these cases, the underlying type is said to be fixed. Following the closing brace of an enum-specifier, each enumerator has the type of its enumeration. If the underlying type is fixed, the type of each enumerator prior to the closing brace is the underlying type and the constant-expression in the enumerator-definition shall be a converted constant expression of the underlying type [...]
And 5.1.1/11 (which is 8.1.4.2/4 in n4659) writes:
A nested-name-specifier that denotes an enumeration (7.2), followed by the name of an enumerator of that enumeration, is a qualified-id that refers to the enumerator. The result is the enumerator. The type of the result is the type of the enumeration. The result is a prvalue.
Then, what happens when we refer to an enumerator through nested-name-specifier prior to closing brace of the declaration? Take for example the following snippet:
template < typename T1, typename T2 >
struct fail_if_not_same {
static_assert(std::is_same<T1, T2>::value, "Fail!");
static constexpr int value = 0;
};
enum class E : short {
A,
B = A + 1,
C = fail_if_not_same<decltype(B), short>::value,
D = fail_if_not_same<decltype(E::B), short>::value
};
What is the type of the expression E::B above? Is this a contradiction in standard? Both gcc and clang follows the 7.2/5.
I think the standard contradicts itself here as you have in 5.1.1/11
The result is the enumerator. (1)
and
The type of the result is the type of the enumeration. (2)
If (1) is true, then the result type shall be the type of the enumerator, which, according to 7.2/5, is either the underlying type of the enumeration or the type defined by the enumeration depending if it is before or after the closing brace.
Meaning that your code sample should compile fine because E::B is B and the type of B is short.
Now if you take (2) into account, it does not change anything after the closing brace. But if (2) is true before the closing brace, it means that the type of E::B is E and at the same time the type of B is short, so you end up with E::B != B which contradicts (1).

Converting enum to enum via int in C++

Is it valid to convert an enum to a different enum via int conversion, like illustrated below ?
It looks like gcc for x64 has no problem with it, but is it something to expect with other compilers and platforms as well ?
What happens when a equals A_third and has no equivalent in enum_B ?
enum enum_A {
A_first = 0,
A_second,
A_third
};
enum enum_B {
B_first = 0,
B_second
};
enum_A a = A_first;
enum_B b;
b = enum_B(int(a));
You have to be careful when doing this, due to some edge cases:
From the C++11 standard (§7.2,6):
For an enumeration whose underlying type is not fixed, the underlying
type is an integral type that can represent all the enumerator values
defined in the enumeration. If no integral type can represent all the
enumerator values, the enumeration is ill-formed. It is
implementation-defined which integral type is used as the underlying
type except that the underlying type shall not be larger than int
unless the value of an enumerator cannot fit in an int or unsigned
int.
This means that it is possible that an enum is a larger type than an int so the conversion from enum to int could fail with undefined results.
Subject to the above, it is possible to convert an int to an enum that results in an enumerator value that the enum does not specify explicitly. Informally, you can think of an enum as being an integral type with a few values explicitly defined with labels.

Why can't C++11 strongly-typed enum be cast to underlying type via pointer?

In C++11 we can cast a strongly-typed enum (enum class) to its underlying type. But it seems we cannot cast a pointer to the same:
enum class MyEnum : int {};
int main()
{
MyEnum me;
int iv = static_cast<int>(me); // works
int* ip = static_cast<int*>(&me); // "invalid static_cast"
}
I'm trying to understand why this should be: is there something about the enum mechanism that makes it hard or nonsensical to support this? Is it a simple oversight in the standard? Something else?
It seems to me that if an enum type is truly built on top of an integral type as above, we should be able to cast not only the values but also the pointers. We can still use reinterpret_cast<int*> or a C-style cast but that's a bigger hammer than I thought we'd need.
TL;DR: The designers of C++ don't like type punning.
Others have pointed out why it's not allowed by the standard; I will try to address why the writers of the standard might have made it that way. According to this proposal, the primary motivation for strongly-typed enums was type safety. Unfortunately, type safety means many things to many people. It's fair to assume consistency was another goal of the standards committee, so let's examine type safety in other relevant contexts of C++.
C++ type safety
In C++ in general, types are unrelated unless explicitly specified to be related (through inheritance). Consider this example:
class A
{
double x;
int y;
};
class B
{
double x;
int y;
};
void foo(A* a)
{
B* b = static_cast<B*>(a); //error
}
Even though A and B have the exact same representation (the standard would even call them "standard-layout types"), you cannot convert between them without a reinterpret_cast. Similarly, this is also an error:
class C
{
public:
int x;
};
void foo(C* c)
{
int* intPtr = static_cast<int*>(c); //error
}
Even though we know the only thing in C is an int and you can freely access it, the static_cast fails. Why? It's not explicitly specified that these types are related. C++ was designed to support object-oriented programming, which provides a distinction between composition and inheritance. You can convert between types related by inheritance, but not those related by composition.
Based on the behavior you've seen, it's clear strongly-typed enums are related by composition to their underlying types. Why might this have been the model the standard committee chose?
Composition vs Inheritance
There are many articles on this issue better written than anything I could fit here, but I'll attempt to summarize. When to use composition vs. when to use inheritance is certainly a grey area, but there are many points in favor of composition in this case.
Strongly-typed enums are not intended to be used as integral values. Thus the 'is-a' relationship indicated by inheritance does not fit.
On the highest level, enums are meant to represent a set of discrete values. The fact that this is implemented through assigning an id number to each value is generally not important (unfortunately C exposes and thus enforces this relationship).
Looking back at the proposal, the listed reason for allowing a specified underlying type is to specify the size and signedness of the enum. This is much more of an implementation detail than an essential part of the enum, again favoring composition.
You could argue for days about whether or not inheritance or composition is better in this case, but ultimately a decision had to be made and the behavior was modeled on composition.
Instead, look at it in a slightly different way. You can't static_cast a long* to int* even if int and long have identical underlying representations. For same same reason an enum based on int is yet treated as a unique, unrelated type to int and as such requires the reinterpret_cast.
An enumeration is a distinct type (3.9.2) with named constants. [...] Each enumeration defines a type that is different from all other types. [...] Two enumeration types are layout-compatible if they have the same underlying type.
[dcl.enum] (§7.2)
The underlying type specifies the layout of the enum in memory, not its relation to other types in the type system (as the standard says, it's a distinct type, a type of its own). A pointer to an enum : int {} can never implicitly convert to an int*, the same way that a pointer to a struct { int i; }; cannot, even though they all look the same in memory.
So why does the implicit conversion to int work in the first place?
For an enumeration whose underlying type is fixed, the values of the
enumeration are the values of the underlying type. [...] The value of
an enumerator or an object of an unscoped enumeration type is
converted to an integer by integral promotion (4.5).
[dcl.enum] (§7.2)
So we can assign values of an enum to an int because they are of type int. An object of enum type can be assigned to an int because of the rules of integer promotion. By the way, the standard here specifically points out that this is only true for C-style (unscoped) enums. This means that you still need the static_cast<int> in the first line of your example, but as soon as you turn the enum class : int into an enum : int it will work without the explicit cast. Still no luck with the pointer type though.
Integral promotions are defined in the standard at [conv.prom] (§4.5). I'll spare you the details of quoting the full section, but the important detail here is that all rules in there apply to prvalues of non-pointer types, so none of this applies to our little problem.
The final piece of the puzzle can be found in [expr.static.cast] (§5.2.9), which describes how static_cast works.
A value of a scoped enumeration type (7.2) can be explicitly converted
to an integral type.
That explains why your cast from enum class to int works.
But note that all of the static_casts allowed on pointer types (again, I won't quote the rather lengthy section) require some relationship between the types. If you remember the beginning of the answer, each enum is a distinct type, so there is no relationship to their underlying type or other enums of the same underlying type.
This ties in with #MarkB's answer: Static-casting a pointer enum to a pointer to int is analogous to casting a pointer from one integral type to another - even if both have the same memory layout underneath and values of one will implicitly convert to the other by the rules integral promotions, they are still unrelated types, so static_cast will not work here.
I think the error of thinking is that
enum class MyEnum : int {};
is not really inheritance. Of course you can say MyEnum is an int. However, it is different from classic inheritance, inasmuch as not all operations that are available on ints are available for MyEnum also.
Let's compare this to the following: A circle is an ellipse. However, it would almost always be wrong to implement a CirlceShape as inheriting from EllipseShape since not all operations that are possible on ellipses are also possible for circle. A simple example would be scaling the shape in x direction.
Hence, to think of enum classes as inheriting from an integer type leads to the confusion in your case. You cannot increment an instance of an enum class, but you can increment integers. Since it's not really inheritance, it makes sense to prohibit casting pointers to these types statically. The following line is not safe:
++*reinterpret_cast<int*>(&me);
This might be the reason why the committee prohibited static_cast in this case. In general reinterpret_cast is considered to be evil while static_cast is considered to be ok.
The answers to your questions can be found in the section 5.2.9 Static cast in the draft standard.
Support for allowing
int iv = static_cast<int>(me);
can be obtained from:
5.2.9/9 A value of a scoped enumeration type (7.2) can be explicitly converted to an integral type. The value is unchanged if the original value can be represented by the specified type. Otherwise, the resulting value is unspecified.
Support for allowing
me = static_cast<MyEnum>(100);
can be obtained from:
5.2.9/10 A value of integral or enumeration type can be explicitly converted to an enumeration type. The value is unchanged if the original value is within the range of the enumeration values (7.2). Otherwise, the resulting value is unspecified (and might not be in that range).
Support for not allowing
int* ip = static_cast<int*>(&me);
can be obtained from:
5.2.9/11 A prvalue of type “pointer to cv1 B,” where B is a class type, can be converted to a prvalue of type “pointer to cv2 D,” where D is a class derived (Clause 10) from B, if a valid standard conversion from “pointer to D” to “pointer to B” exists (4.10), cv2 is the same cv-qualification as, or greater cv-qualification than, cv1, and B is neither a virtual base class of D nor a base class of a virtual base class of D. The null pointer value (4.10)
is converted to the null pointer value of the destination type. If the prvalue of type “pointer to cv1 B” points to a B that is actually a subobject of an object of type D, the resulting pointer points to the enclosing object of type D. Otherwise, the result of the cast is undefined.
static_cast cannot be used to cast &me to an int* since MyEnum and int are not related by inheritance.
I think the reason for first static_cast is being able to work with functions and libraries that expect old style enum or even used a bunch of defined values for enumerations and directly expect an integral type. But there is no other logical relation between type enum and an integral type, so you should use reinterpret_cast if you want that cast. but if you have problems with reinterpret_cast you can use your own helper:
template< class EnumT >
typename std::enable_if<
std::is_enum<EnumT>::value,
typename std::underlying_type<EnumT>::type*
>::type enum_as_pointer(EnumT& e)
{
return reinterpret_cast<typename std::underlying_type<EnumT>::type*>(&e);
}
or
template< class IntT, class EnumT >
IntT* static_enum_cast(EnumT* e,
typename std::enable_if<
std::is_enum<EnumT>::value &&
std::is_convertible<
typename std::underlying_type<EnumT>::type*,
IntT*
>::value
>::type** = nullptr)
{
return reinterpret_cast<typename std::underlying_type<EnumT>::type*>(&e);
}
While this answer may not satisfy you about the reason of prohibiting static_cast of enum pointers, it give you a safe way to use reinterpret_cast with them.