Is there a good rule when to use a class template instead of using member templates?
As I understand it, your are forced to use a class template if your class contains member variable templates but otherwise you are free of choice since this:
template<typename T>
class Foo
{
public:
void someMethod(T x);
};
behaves like this:
class Foo
{
public:
template<typename T>
void someMethod(T x);
};
Is my claim wrong? Or is there any rule of thumb when to use what?
The two are not at all the same. With the first:
Foo<int> f;
f.someMethod('a');
the called function is someMethod(int).
With the second:
Foo f;
f.someMethod('a');
the called function is someMethod(char).
You can choose to make your class a template, rather than having member function templates, for several reasons. Say you have a template parameter T.
Like you said, if you have a member variable of type T, your class needs to be a template.
If you have a function returning T and not accepting T, and you don't want to manually specify the type in each invocation, your class needs to be a template.
If you need a virtual function that depends on T, your class needs to be a template.
If you need a different class layout (i.e. member variables) per instantiation of T, your class needs to be a template.
If you need to make sure that your functions all operate on a single type rather than generate different versions of them arbitrarily, your class needs to be a template.
The best rule of thumb would be to use the simplest thing that makes sense. In general, member function templates tend to be more rare—and virtually non-existent for the use case you're talking about. Maybe that's no coincidence.
Related
I have the following code:
#include <iostream>
template <typename T>
struct Base
{
using Type = int;
};
template <typename T>
struct Derived : Base<T>
{
//uncommmenting the below cause compiler error
//using Alias = Type;
};
int main()
{
Derived<void>::Type b = 1;
std::cout << b << std::endl;
return 0;
}
Now the typename Type is available to Derived if its in a deduced context - as shown by the perfectly valid declaration of b. However, if I try to refer to Type inside the declaration of Derived itself, then I get a compiler error telling me that Type does not name a type (for example if the definition of Alias is uncommented).
I guess this is something to do with the compiler not being able to check whether or not Type can be pulled in from the base class when it is parsing the definition of Derived outside the context of a specific instantiation of parameter T. In this case, this is frustrating, as Base always defines Type irrespective of T. So my question is twofold:
1). Why on Earth does this happen? By this I mean why does the compiler bother parsing Derived at all outside of an instantiation context (I guess non-deduced context), when not doing so would avoid these 'bogus' compiler errors? Perhaps there is a good reason for this. What is the rule in the standard that states this must happen?
2). What is a good workaround for precisely this type of problem? I have a real-life case where I need to use base class types in the definition of a derived class, but am prevented from doing so by this problem. I guess I'm looking for some kind of 'hide behind non-deduced context' solution, where I prevent this compiler 'first-pass' by putting required definitions/typedefs behind templated classes or something along those lines.
EDIT: As some answers below point out, I can use using Alias = typename Base<T>::Type. I should have said from the outset, I'm aware this works. However, its not entirely satisfactory for two reasons: 1) It doesn't use the inheritance hierarchy at all (Derived would not have to be derived from Base for this to work), and I'm precisely trying to use types defined in my base class hierarchy and 2) The real-life case actually has several layers of inheritance. If I wanted to pull in something from several layers up this becomes really quite ugly (I either need to refer to a non-direct ancestor, or else repeat the using at every layer until I reach the one I need it at)
Because type is in a "dependent scope" you can access it like this:
typename Base<T>::Type
Your Alias should then be defined like this:
using Alias = typename Base<T>::Type;
Note that the compiler doesn't, at this point, know if Base<T>::type describes a member variable or a nested type, that is why the keyword typename is required.
Layers
You do not need to repeat the definition at every layer, here is an example, link:
template <typename T>
struct intermediate : Base<T>
{
// using Type = typename Base<T>::Type; // Not needed
};
template <typename T>
struct Derived : intermediate<T>
{
using Type = typename intermediate<T>::Type;
};
Update
You could also use the class it self, this relies on using an unknown specializations.
template <typename T>
struct Derived : Base<T>
{
using Type = typename Derived::Type; // <T> not required here.
};
The problem is that Base<T> is a dependent base class, and there may be specializations for it in which Type is not anymore defined. Say for example you have a specialization like
template<>
class Base<int>
{}; // there's no more Type here
The compiler cannot know this in advance (technically it cannot know until the instantiation of the template), especially if the specialization is defined in a different translation unit. So, the language designers chose to take the easy route: whenever you refer to something that's dependent, you need to explicitly specifify this, like in your case
using Alias = typename Base<T>::Type;
I guess this is something to do with the compiler not being able to check whether or not Type can be pulled in from the base class when it is parsing the definition of Derived outside the context of a specific instantiation of parameter T.
Yes.
In this case, this is frustrating, as Base always defines Type irrespective of T.
Yes.
But in general it would be entirely infeasible to detect whether this were true, and extremely confusing if the semantics of the language changed when it were true.
Perhaps there is a good reason for this.
C++ is a general-purpose programming language, not an optimised-for-the-program-Smeeheey-is-working-on-today programming language. :)
What is the rule in the standard that states this must happen?
It's this:
[C++14: 14.6.2/3]: In the definition of a class or class template, if a base class depends on a template-parameter, the base class scope is not examined during unqualified name lookup either at the point of definition of the class template or member or during an instantiation of the class template or member. [..]
What is a good workaround for precisely this type of problem?
You already know — qualification:
using Alias = typename Base<T>::Type;
When defining a template, sometimes things need to be a bit more explicit:
using Alias = typename Base<T>::Type;
Rougly speaking: a template is a blueprint of sorts. Nothing actually exists until the template get instantiated.
When doing the same thing, but in a non-template context, a C++ compiler will try to figure out what Type is. It'll try to find it in the base classes, and go from there. Because everything is already declared, and things are pretty much cut-and-dry.
Here, a base class does not really exist until the template gets instantiated. If you already know about template specialization, that you should realize that the base class may not actually turn out to have a Type member, when the template gets instantiated if there's a specialization for the base class defined later down the road, that's going to override the whole thing, and turn it inside and out.
As such, when encountering just a plain, old, Type in this context, the compiler can't make a lot of assumptions. It can't assume that it can look in any defined template base classes because those base classes may not actually look anything like the compiler thinks they will look, when things start to solidify; so you have spell everything out, explicitly, for the compiler, and tell the compiler exactly what your are trying to do, here.
You cannot use your base class types in a non-deduced context. C++ refuses to assume unbound names refer to things in your base class.
template <typename T>
struct Base {
using Type = int;
};
template<>
struct Base<int> {};
using Type=std::string;
template <typename T>
struct Derived : Base<T> {
using Alias = Type;
};
Now let us look at what is going on here. Type is visible in Derived -- a global one. Should Derived use that or not?
Under the rule of "parse nothing until instantiated", we use the global Type if and only if T is int, due to the Base specialization that removes Type from it.
Following that rule we run into the problem that we can diagnose basically no errors in the template prior to it being instantiated, because the base class could replace the meaning of almost anything! Call abs? Could be a member of the parent! Mention a type? Could be from the parent!
That would force templates to basically be macros; no meaningful parsing could be done prior to instantiating them. Minor typos could result in massively different behavior. Almost any incorrectness in a template would be impossible to diagnose without creating test instances.
Having templates that can be checked for correctness means that when you want to use your parent class types and members, you have to say you are doing so.
As a partial response about the point
[T]his is frustrating, as Base always defines Type irrespective of T.
I'd say: no it does not.
Please consider the following example differing from yours only by the one line definition of Base<void> and of the definition of Alias:
#include <iostream>
template <typename T>
struct Base
{
using Type = int;
};
template <typename T>
struct Derived : Base<T>
{
using Alias = typename Base<T>::Type; // error: no type named 'Type' in 'struct Base<void>'
};
template<> struct Base<void> {};
int main()
{
Derived<void>::Type b = 1;
std::cout << b << std::endl;
return 0;
}
In the context of template <typename T> struct Derived : Base<T>, there is no guaranty that Type exists. You must explicitly tell your compiler than Base<T>::Type is a type (with typename) and if you ever fail this contract, you'll end up with a compilation error.
I'm wondering how template member functions work. In particular, when there is an instantiation of the template member function, is the whole class redefined? My confusion comes from the fact that (if I'm right) template classes are not classes in the proper sense. i.e., when instantiated, the compiler creates the definition for a completely new class. The same for template functions. However, classes with a template function seem to be actual classes, so I'm not sure how they could possibly work. Thus, I'm wondering, after instantiating a template member function, what happens with the class definition? Moreover, if I pass a class with a template member function to a template class, can I use the template member function? Could that cause a problem? I tried it once but got an error saying that several functions where defined more that once, although I'm not sure if that was the reason or if there could be an other reason for my error. Is there any further caveat when using static template member functions?
The class definition remains as it is; all the template function does is generate a family of member functions for that class. As an example:
class A {
public:
template<typename T> foo (T &t);
}
Is not conceptually different from you writing:
class A {
public:
foo (bool &t);
foo (int &t);
foo (double &t);
}
...just more convenient. And in the last example, you wouldn't expect a new class to be created for each function would you?
Perhaps the confusion comes from the notion that functions are somehow part of the memory layout of a class; that each function is itself contained in the class, and will be instantiated somewhere in memory whenever an object of the class is created. This notion is incorrect. Functions (templated, global, member, lambda, or otherwise) are never created on the fly or copied around in memory; they are a static and unchanging part of the executable image. The memory layout of the class is not changed by the presence of an extra set of functions, even if those happen to be generated by a template member.
The template class definition is instantiated when you instantiate a class. Each member function of it is instantiated when used. This actually allows you to have member functions that would not work if called when the class is instantiated with some types and not others. However, you must ensure that the signature of the function is either syntactically viable or fails with SFINAE. It will be looked up during the first phase of parsing. The body, if the function isn't itself a template, will be checked for name lookup...so dependent names have to be labeled as such via typename.
Where do I have to specify default template parameters of classes member functions (assuming that the declaration is (of course) in the "class body", and the function definition is outside the class body) for each case in C++2011 :
"normal" functions
static functions
friend functions
In the definition, in the declaration or both ?
Well,
From my experiences creating template classes and methods, you specify a template function as such:
template<typename T>
T MyFunc(T &aArg1, T &aArg2)
{
//...Definition Goes Here
}
The typename T is the template argument type for the template function and you need to pass that data type consistently to each argument labeled as "T". This means that aArg2 has to be whatever data type aArg1 is. Now, when you call this function, you call it like so:
MyFunc</*datatype*/int>(iArg1, iArg2); the two arguments have to be data type "int" or you'll get a warning or an error.
Now, this also applies to class methods (I think that is what you meant by "classes member functions") which are the functions supplied by the class (i.e. MyClass::MyFunc()) so when you declare a class method that is a template method, you do it in the same manner. Here is an example class:
class MyClass
{
MyClass();
~MyClass();
template<typename T>
static T MyStaticFunc(T aArg) { return aArg; }
template<typename T>
T MyFunc(T aArg) { return aArg; }
}
As you can see, not to difficult. Now, static functions are the same way you just have to be sure t define then in the same module that the class is built in, otherwise, you'll get an error.
Unfortunately, I never really use "friend" methods, so I don't know how to tackle that. I would suspect you would do it in the same way as the other two. I hoped that whole essay of an answer helped.
Trying these out in Clang suggests the following:
For non-static and static functions, specifying the default in either the definition or
the declaration is acceptable - but not both and certainly not if
they contradict one another;
For friend functions, specifying a
default inside the class definition results in an error.
We have following class definition
template<typename T>
class Klass {...}
and we also have below two instantiations
Klass<int> i;
Klass<double> d;
how many copies of Klass' methods are generated by the C++ compiler?
Can somebody explain it? Thanks!
Klass isn't a type, so it doesn't make sense to talk of Klass's methods. Kalss<int> is a type with it's own methods, and so is Klass<double>. In your example there would be one set of methods for each type.
Edit in real life, it isn't as simple as that. The question of the actual existence of the methods also depends on other factors, see #KerrekSB's answer to this question.
Each template instance is an entirely separate, distinct and independent type of its own. However, the code for class template member functions is only generated if the member function is actually used for a given template instantiation (unless you instantiate the template explicitly for some set of parameters). Among other things, this means that if some class template member function's body doesn't actually make sense for a given template parameter, then you can still use the overall template as long as you don't invoke that member function, since the code for the member function will never get compiled.
Also bear in mind that templates can be specialized:
template <typename T> struct Foo {
int bar;
void chi();
};
template <> struct Foo<int> {
double bar(bool, char) const;
typedef std::vector<bool> chi;
bool y;
};
As you can see, there's a lot you cannot just tell from a template itself until you see which actual instantiations you'll be talking about.
Is this answer considered "good" code or is it just an ugly hack?
And I would like to know how this is forward-declared (both classes).
When I just forward-declare the class with 2 template-parameters, it just always takes this one, no matter what value flag has.
I would like to do this because I have 2 special member functions which should behave differently on flag being true and I don't feel like reimplementing the whole class. Also, it should have the same name. According to this example, this seems to be possible.
And I have to forward-declare it because I'm creating a library in which we forward-declare everything.
Any idea?
It has the drawback that it doesn't really work. The base member function is not overridden, but it is just hidden by the derived class' function when you try to call it from outside. Which means if you call doSomething out of the base class (where presumably all your other functions live) it will call the base class doSomething which is not what is wanted.
The accepted answer on that question shows multiple ways for how you can solve your problem.
In order to use specialisation its definition always has to be visible to the caller. If, for example, you have template <class Type, bool flag> struct something defined in one header and template <class Type> struct something<Type, true> : public something<Type, false> defined in the second one, to use the latter you have to include the second header. Without that you will always get the first, more generic type.
EDIT: the bit about forward-declaring got me thinking. If you want to use only type declaration, as in pointer variable, do the following:
Header
template <class Type, bool flag>
struct something;
struct Test
{
something<int, true>* ptr; // definition not needed
Test();
}
Source
#include "something.h" // header with template
#include "something_spec.h" // header with specialisation
Test::Test()
{
ptr = new something<int, true>(); // specialisation used
}