I'm trying to implement a simple system using template struct, the code is very simple and compile fine with MSVC, yet i cannot understand why CLANG gives me this error: "lld-link : error : undefined symbol: public: static struct FMyStruct const TSpec<1>::m_struct"
I compile on a windows 64bitmachine with VisualStudio IDE but CLANG LLVM as compiler. The code works fine with MSVC.
I simplified my problem to the very minimum, i tried to put everything in one single cpp file, with no result. I also tried explicit template instanciation.
I want to be compliant with C++14, no C++17. One thing i tried that worked was declaring the m_struct member as an inline variable, but then i get this warning: "inline variables are a C++17 extension"
struct FMyStruct
{
const int _p0;
const int _p1;
const int _p2;
};
template< int > struct TSpec {
static constexpr FMyStruct m_struct = { 0, 0, 0 };
};
FMyStruct
Function( int i )
{
return TSpec< 1 >::m_struct;
}
int main()
{
return 0;
}
Result:
"lld-link : error : undefined symbol: public: static struct FMyStruct const TSpec<1>::m_struct"
I expect the linker to find the symbol m_struct since it is defined very next to it ...
The weirdest part is that if i try:
int
Function( int i )
{
return TSpec< 1 >::m_struct._p0;
}
the program will compile fine.
Edit: My CLANG version is 9.0.0, prebuilt distributed version for windows from the official website.
clang version 9.0.0 (trunk)
Target: x86_64-pc-windows-msvc
Thread model: posix
InstalledDir: C:\Program Files\LLVM\bin
It indeed seems to be a bug related to the CLANG version, thanks #Sombrero Chicken for pointing this out.
So this is definitely weird but i managed to solve this avoiding the C++17-specific 'inline' declaration of the static member by adding this after the template struct definition:
template< int N > const FMyStruct TSpec< N >::m_struct;
By the way, it does not seem to be related to the template declaration at all.
For summary, it gives this program that will compile fine.
struct FMyStruct
{
const int _p0;
const int _p1;
const int _p2;
};
template< int > struct TSpec {
static constexpr FMyStruct m_struct = { 0, 0, 0 };
};
template< int N > const FMyStruct TSpec< N >::m_struct;
FMyStruct
Function( int i )
{
return TSpec< 1 >::m_struct;
}
int main()
{
return 0;
}
I still do not really understand why this is necessary since the static member is public to the struct, and part of the same unit & file; i guess this is a different matter but i'd like to be enlightened. Thank you.
What is the best way to initialize a private, static data member in C++? I tried this in my header file, but it gives me weird linker errors:
class foo
{
private:
static int i;
};
int foo::i = 0;
I'm guessing this is because I can't initialize a private member from outside the class. So what's the best way to do this?
The class declaration should be in the header file (Or in the source file if not shared).
File: foo.h
class foo
{
private:
static int i;
};
But the initialization should be in source file.
File: foo.cpp
int foo::i = 0;
If the initialization is in the header file then each file that includes the header file will have a definition of the static member. Thus during the link phase you will get linker errors as the code to initialize the variable will be defined in multiple source files.
The initialisation of the static int i must be done outside of any function.
Note: Matt Curtis: points out that C++ allows the simplification of the above if the static member variable is of const integer type (bool, char, char8_t [since C++20], char16_t, char32_t, wchar_t, short, int, long, long long, or any implementation-defined extended integer types, including any signed, unsigned, and cv-qualified variants.). You can then declare and initialize the member variable directly inside the class declaration in the header file:
class foo
{
private:
static int const i = 42;
};
For a variable:
foo.h:
class foo
{
private:
static int i;
};
foo.cpp:
int foo::i = 0;
This is because there can only be one instance of foo::i in your program. It's sort of the equivalent of extern int i in a header file and int i in a source file.
For a constant you can put the value straight in the class declaration:
class foo
{
private:
static int i;
const static int a = 42;
};
Since C++17, static members may be defined in the header with the inline keyword.
http://en.cppreference.com/w/cpp/language/static
"A static data member may be declared inline. An inline static data member can be defined in the class definition and may specify a default member initializer. It does not need an out-of-class definition:"
struct X
{
inline static int n = 1;
};
For future viewers of this question, I want to point out that you should avoid what monkey0506 is suggesting.
Header files are for declarations.
Header files get compiled once for every .cpp file that directly or indirectly #includes them, and code outside of any function is run at program initialization, before main().
By putting: foo::i = VALUE; into the header, foo:i will be assigned the value VALUE (whatever that is) for every .cpp file, and these assignments will happen in an indeterminate order (determined by the linker) before main() is run.
What if we #define VALUE to be a different number in one of our .cpp files? It will compile fine and we will have no way of knowing which one wins until we run the program.
Never put executed code into a header for the same reason that you never #include a .cpp file.
Include guards (which I agree you should always use) protect you from something different: the same header being indirectly #included multiple times while compiling a single .cpp file.
With a Microsoft compiler[1], static variables that are not int-like can also be defined in a header file, but outside of the class declaration, using the Microsoft specific __declspec(selectany).
class A
{
static B b;
}
__declspec(selectany) A::b;
Note that I'm not saying this is good, I just say it can be done.
[1] These days, more compilers than MSC support __declspec(selectany) - at least gcc and clang. Maybe even more.
int foo::i = 0;
Is the correct syntax for initializing the variable, but it must go in the source file (.cpp) rather than in the header.
Because it is a static variable the compiler needs to create only one copy of it. You have to have a line "int foo:i" some where in your code to tell the compiler where to put it otherwise you get a link error. If that is in a header you will get a copy in every file that includes the header, so get multiply defined symbol errors from the linker.
If you want to initialize some compound type (f.e. string) you can do something like that:
class SomeClass {
static std::list<string> _list;
public:
static const std::list<string>& getList() {
struct Initializer {
Initializer() {
// Here you may want to put mutex
_list.push_back("FIRST");
_list.push_back("SECOND");
....
}
}
static Initializer ListInitializationGuard;
return _list;
}
};
As the ListInitializationGuard is a static variable inside SomeClass::getList() method it will be constructed only once, which means that constructor is called once. This will initialize _list variable to value you need. Any subsequent call to getList will simply return already initialized _list object.
Of course you have to access _list object always by calling getList() method.
C++11 static constructor pattern that works for multiple objects
One idiom was proposed at: https://stackoverflow.com/a/27088552/895245 but here goes a cleaner version that does not require creating a new method per member.
main.cpp
#include <cassert>
#include <vector>
// Normally on the .hpp file.
class MyClass {
public:
static std::vector<int> v, v2;
static struct StaticConstructor {
StaticConstructor() {
v.push_back(1);
v.push_back(2);
v2.push_back(3);
v2.push_back(4);
}
} _staticConstructor;
};
// Normally on the .cpp file.
std::vector<int> MyClass::v;
std::vector<int> MyClass::v2;
// Must come after every static member.
MyClass::StaticConstructor MyClass::_staticConstructor;
int main() {
assert(MyClass::v[0] == 1);
assert(MyClass::v[1] == 2);
assert(MyClass::v2[0] == 3);
assert(MyClass::v2[1] == 4);
}
GitHub upstream.
Compile and run:
g++ -ggdb3 -O0 -std=c++11 -Wall -Wextra -pedantic -o main.out main.cpp
./main.out
See also: static constructors in C++? I need to initialize private static objects
Tested on Ubuntu 19.04.
C++17 inline variable
Mentioned at: https://stackoverflow.com/a/45062055/895245 but here is a multifile runnable example to make it even clearer: How do inline variables work?
This awesome C++17 feature allow us to:
conveniently use just a single memory address for each constant
store it as a constexpr: How to declare constexpr extern?
do it in a single line from one header
main.cpp
#include <cassert>
#include "notmain.hpp"
int main() {
// Both files see the same memory address.
assert(¬main_i == notmain_func());
assert(notmain_i == 42);
}
notmain.hpp
#ifndef NOTMAIN_HPP
#define NOTMAIN_HPP
inline constexpr int notmain_i = 42;
const int* notmain_func();
#endif
notmain.cpp
#include "notmain.hpp"
const int* notmain_func() {
return ¬main_i;
}
Compile and run:
g++ -c -o notmain.o -std=c++17 -Wall -Wextra -pedantic notmain.cpp
g++ -c -o main.o -std=c++17 -Wall -Wextra -pedantic main.cpp
g++ -o main -std=c++17 -Wall -Wextra -pedantic main.o notmain.o
./main
GitHub upstream.
I don't have enough rep here to add this as a comment, but IMO it's good style to write your headers with #include guards anyway, which as noted by Paranaix a few hours ago would prevent a multiple-definition error. Unless you're already using a separate CPP file, it's not necessary to use one just to initialize static non-integral members.
#ifndef FOO_H
#define FOO_H
#include "bar.h"
class foo
{
private:
static bar i;
};
bar foo::i = VALUE;
#endif
I see no need to use a separate CPP file for this. Sure, you can, but there's no technical reason why you should have to.
You can also include the assignment in the header file if you use header guards. I have used this technique for a C++ library I have created. Another way to achieve the same result is to use static methods. For example...
class Foo
{
public:
int GetMyStatic() const
{
return *MyStatic();
}
private:
static int* MyStatic()
{
static int mStatic = 0;
return &mStatic;
}
}
The above code has the "bonus" of not requiring a CPP/source file. Again, a method I use for my C++ libraries.
The linker problem you encountered is probably caused by:
Providing both class and static member definition in header file,
Including this header in two or more source files.
This is a common problem for those who starts with C++. Static class member must be initialized in single translation unit i.e. in single source file.
Unfortunately, the static class member must be initialized outside of the class body. This complicates writing header-only code, and, therefore, I am using quite different approach. You can provide your static object through static or non-static class function for example:
class Foo
{
// int& getObjectInstance() const {
static int& getObjectInstance() {
static int object;
return object;
}
void func() {
int &object = getValueInstance();
object += 5;
}
};
I follow the idea from Karl. I like it and now I use it as well.
I've changed a little bit the notation and add some functionality
#include <stdio.h>
class Foo
{
public:
int GetMyStaticValue () const { return MyStatic(); }
int & GetMyStaticVar () { return MyStatic(); }
static bool isMyStatic (int & num) { return & num == & MyStatic(); }
private:
static int & MyStatic ()
{
static int mStatic = 7;
return mStatic;
}
};
int main (int, char **)
{
Foo obj;
printf ("mystatic value %d\n", obj.GetMyStaticValue());
obj.GetMyStaticVar () = 3;
printf ("mystatic value %d\n", obj.GetMyStaticValue());
int valMyS = obj.GetMyStaticVar ();
int & iPtr1 = obj.GetMyStaticVar ();
int & iPtr2 = valMyS;
printf ("is my static %d %d\n", Foo::isMyStatic(iPtr1), Foo::isMyStatic(iPtr2));
}
this outputs
mystatic value 7
mystatic value 3
is my static 1 0
Also working in privateStatic.cpp file :
#include <iostream>
using namespace std;
class A
{
private:
static int v;
};
int A::v = 10; // possible initializing
int main()
{
A a;
//cout << A::v << endl; // no access because of private scope
return 0;
}
// g++ privateStatic.cpp -o privateStatic && ./privateStatic
What about a set_default() method?
class foo
{
public:
static void set_default(int);
private:
static int i;
};
void foo::set_default(int x) {
i = x;
}
We would only have to use the set_default(int x) method and our static variable would be initialized.
This would not be in disagreement with the rest of the comments, actually it follows the same principle of initializing the variable in a global scope, but by using this method we make it explicit (and easy to see-understand) instead of having the definition of the variable hanging there.
One "old-school" way to define constants is to replace them by a enum:
class foo
{
private:
enum {i = 0}; // default type = int
enum: int64_t {HUGE = 1000000000000}; // may specify another type
};
This way doesn't require providing a definition, and avoids making the constant lvalue, which can save you some headaches, e.g. when you accidentally ODR-use it.
Here are all possibilities and errors in one simple example ...
#ifndef Foo_h
#define Foo_h
class Foo
{
static const int a = 42; // OK
static const int b {7}; // OK
//static int x = 42; // ISO C++ forbids in-class initialization of non-const static member 'Foo::x'
//static int y {7}; // ISO C++ forbids in-class initialization of non-const static member 'Foo::x'
static int x;
static int y;
int m = 42;
int n {7};
};
// Foo::x = 42; // error: 'int Foo::x' is private
int Foo::x = 42; // OK in Foo.h if included in only one *.cpp -> *.o file!
int Foo::y {7}; // OK
// int Foo::y {7}; // error: redefinition of 'int Foo::y'
// ONLY if the compiler can see both declarations at the same time it,
// OTHERWISE you get a linker error
#endif // Foo_h
But better place this in Foo.cpp. This way you can separately compile each file and link them later, otherwise Foo:x will be present in multiple object files and cause a linker error. ...
// Foo::x = 42; // error: 'int Foo::x' is private, bad if Foo::X is public!
int Foo::x = 42; // OK in Foo.h if included in only one *.cpp -> *.o file!
int Foo::y {7}; // OK
Does this serves your purpose?
//header file
struct MyStruct {
public:
const std::unordered_map<std::string, uint32_t> str_to_int{
{ "a", 1 },
{ "b", 2 },
...
{ "z", 26 }
};
const std::unordered_map<int , std::string> int_to_str{
{ 1, "a" },
{ 2, "b" },
...
{ 26, "z" }
};
std::string some_string = "justanotherstring";
uint32_t some_int = 42;
static MyStruct & Singleton() {
static MyStruct instance;
return instance;
}
private:
MyStruct() {};
};
//Usage in cpp file
int main(){
std::cout<<MyStruct::Singleton().some_string<<std::endl;
std::cout<<MyStruct::Singleton().some_int<<std::endl;
return 0;
}
I just wanted to mention something a little strange to me when I first encountered this.
I needed to initialize a private static data member in a template class.
in the .h or .hpp, it looks something like this to initialize a static data member of a template class:
template<typename T>
Type ClassName<T>::dataMemberName = initialValue;
We use the Intel C++ compiler and detected that it miscompiles (?) the following, reduced from a use of boost::function<Ponies()> f(unnamedNamespacedFunctor).
a1.cc:
template<typename T>
int f(T) { static int x = T::x; return x; }
namespace { struct A { static const int x = 1; }; }
int f1() {
return f(A());
}
a2.cc:
template<typename T>
int f(T) { static int x = T::x; return x; }
namespace { struct A { static const int x = 0; }; }
int f2() {
return f(A());
}
main.cc:
#include <cstdio>
int f1();
int f2();
int main() {
std::printf("%d != %d\n", f1(), f2());
}
Command line:
# icpc a1.cc a2.cc main.cc -o main
# ./main
0 != 0
My question is: Is this compliant? Does using static locals in such instantiations produce undefined behavior? When inspecting the produced symbols, I noted that while f is has local linkage, as I suspected, the x static variable receives weak linkage and therefore the two x'es are merged and it becomes lottery which is chosen
# icpc a2.cc a1.cc main.cc -o main
# ./main
1 != 1
I would be grateful for help. Perhaps this is actually a compiler-bug after all and it has already been reported?
This looks like a bug to me. Letting A1 be one of the instantiations of A and A2 be the other:
I presume that the static x has weak linkage so that the linker can merge copies of the static between multiple copies of the same instantiation. (If you managed to instantiate f<A1> in two different translation units, for example.)
Either f<A1> and f<A2> should have different name mangling, which would cause the two versions of x to have different name mangling (I think some compilers actually generate a random value to make names inside anonymous namespaces unique), or else x should not have internal linkage (because a local type was used to instantiate f, which should make it impossible to replicate in another translation unit).
I'm doing something like this
Class.hpp:
class Class {
private:
static const unsigned int arraySize;
int ar[arraySize+2];
};
Class.cpp:
#include <Class.hpp>
const unsigned int arraySize = 384;
The compiler (q++, a c++ compiler for the QNX OS based on g++) gives me error: array bound is not an integer constant while compiling a unit including Class.hpp (not while compiling Class.cpp).
Why isn't that working? I know that a static const member can be used as an array bound, guaranteed by the C++ standard (see this anwser). But why doesn't the compiler see the result of static const + const as a constant?
This is good code which should have been accepted by the compiler:
class Class {
const static int arraySize = 384;
int ar[arraySize+2];
};
and if it isn't, your compiler is broken.
However, if you move actual constant out of the header file to selected translation unit, that invalidates the code.
// Class.h
class Class {
const static int arraySize;
int ar[arraySize+2]; // ERROR
};
// Class.cpp
const int Class::arraySize = 384;
This is because the size of your Class object cannot be determined at compile time from the data available in the header alone. This is not exactly right reason, but reasoning along these lines helps to understand compilation errors such as this.
To avoid making such mistakes, you can replace static const int with an enum, e.g.
class Class {
enum { arraySize = 384 };
int ar[arraySize+2];
};
I'm surprised this actually compiles on gcc, as a comment says. Since the 384 isn't in the header file, the size of the Class is not known to other compilation units. It might not matter in some compilation units depending on how/if they are using Class, but I can't imagine this compiling:
// this is a source file called, say, blah.cpp
#include <Class.hpp>
void someFunc()
{
void *mem = malloc(sizeof(Class)); // size is not known, so this can't compile
// do something with mem
}
You need to have in your .hpp:
class Class {
private:
static const unsigned int arraySize = 384;
int ar[arraySize+2];
};
.. as it is in the OP that you link to here.
please consider these files:
p.h:
#ifndef _p_h_
#define _p_h_
class p{
public:
static void set_func(int(*)());
private:
static int (*sf)();
};
#endif
p.cpp:
#include "p.h"
#include <cstdio>
int (p::*sf)() = NULL; //defining the function pointer
void p::set_func(int(*f)()){
sf = f;
}
main.cpp:
#include "p.h"
#include <iostream>
int function_x(){
std::cout << "I'm function_x()" << std::endl;
return 1234;
}
int main(){
p::set_func(function_x);
}
when compiling, I get this:
$ g++ -o pp main.cpp p.cpp
/tmp/ccIs0M7r.o:p.cpp:(.text+0x7): undefined reference to `p::sf'
collect2: ld returned 1 exit status
but:
$ g++ -c -o pp p.cpp
compiles right.
What's wrong with the code? I just can't find where the problem is, please your help will be more than appreciated.
Thanks.
Your attempt at defining p::sf is incorrect – yours is a definition of a global variable named sf that is of type int (p::*)(), i.e. a pointer to a member function. Consequently p::sf remains undefined, hence the linker error.
Try this instead:
int (*p::sf)() = 0;
// or,
typedef int (*p_sf_t)();
p_sf_t p::sf = 0;
The difference is because error only occurs when you actually link the program. The problem is in your declaration of the static function pointer. The correct syntax is:
int (*p::sf)() = NULL; //defining the function pointer
You define a member function pointer and not a function pointer. I'm not sure what the correct syntax is, but I would have tried something like this:
int (*p::sf)() = NULL;
I will not give another answer (ildjarn answer is correct) but i will suggest you another way of achieving the same without static initialization (and the burdens it implies)
class p{
public:
typedef int (*func_t)();
static void set_func(func_t v) {
func_t& f = getFuncRef();
f = v;
}
static void call_func() {
func_t& f = getFuncRef();
assert( f != 0);
f();
}
private:
static func_t& getFuncRef() {
static func_t sf = 0;
return sf;
}
};
in this way you delegate the static initialization to a static function variable, which doesn't have the initialization order problems that affect static data variables, and is lazy-initialised