I am using a framework that passes around function pointers as void*. I want a mock to return a function pointer, and I want to define the function in-place (like a lambda; which does not work as shown below).
A minimal working example is shown below.
#include <gtest/gtest.h>
#include <gmock/gmock.h>
using namespace std;
using namespace testing;
class Original
{
public:
typedef int(*fptr)();
void Func()
{
void* f = Func2();
fptr func = reinterpret_cast<fptr>(f);
if (func) {
int i = func();
if (i == 1) {
//do something
} else if (i == 3) {
//NOTE my unit test should test this decision branch
}
}
}
static int Func3() {return 1;}
virtual void* Func2() {return (void*)&Func3;}
};
class MyMock : public Original
{
public:
MOCK_METHOD0(Func2, void*());
};
My main goal: I want to get rid of this function, and define it inline in the EXPECT_CALL(). See below.
int MockFunc() {cout << "mock func" << endl; return 3;}
The test case:
TEST(MYTEST, Test)
{
MyMock m;
//WORKS: compiles and works as expected,
//but I do not want to use **MockFunc**
EXPECT_CALL(m, Func2()).Times(AtLeast(1))
.WillRepeatedly(Return(&MockFunc));
//DOES NOT WORK: Does not compile, of course
//(compiler message below this code block)
EXPECT_CALL(m, Func2()).Times(AtLeast(1))
.WillRepeatedly(Return((void*)&([](){return 3;})));
m.Func();
}
main.cpp:117:90: error: taking address of temporary [-fpermissive]
For completeness, the main():
int main(int argc, char** argv)
{
::testing::InitGoogleTest(&argc, argv);
return RUN_ALL_TESTS();
}
So the question again: How can I get rid of the MockFunc() function and define its contents in-place within the Return()?
The following statement should work:
EXPECT_CALL(m, Func2()).Times(AtLeast(1)).WillRepeatedly(Return((void*)(+([](){return 3;}))));
It exploits the fact that non-capturing lambdas decay to function pointers.
In other terms, the resulting type of the expression (+([](){return 3;}) is int(*)(). You can then cast it to void* as you did. The error should disappear as well, for you are no longer getting the address of a temporary.
How can I properly pass a const string, for example: "Hello" as a default parameter to a constructor without getting any warnings and errors? In other words how can I preserve memory for them before the function is called?
#include <iostream>
const char *defString = "Hello";
void foo(const char *str = defString) {
std::cout << str;
}
int main() {
foo();
foo("Hello, world!\n");
}
I built an interface taking pointers to functions. Sometimes this calculation depends on state, which I want to encapsulate in a class and pass its method:
#include <iostream>
class Printer {
public:
static void print(int i) { // Want to get rid of the static
std::cout << i << "\n";
}
};
template<typename int_func>
void with_1(int_func func) {
func(1);
}
int main(int argc, char const *argv[]) {
Printer printer;
with_1(printer.print);
return 0;
}
I need non-static methods (and would even prefer overloading operator()). However removing the static results in error: a pointer to a bound function may only be used to call the function.
I could use a dummy like this:
Printer printer;
void dummy(int i) {
printer.print(i);
}
int main(int argc, char const *argv[]) {
with_1(dummy);
return 0;
}
But that does not look elegant to me. Can I write a template that accepts both, function pointers and non-static methods? Or is there even a better design pattern for my problem?
You can not simply pass non-static method like this, because they work on instance. A simply solution is to use lambda:
#include <iostream>
class Printer {
public:
static void print(int i) { // Want to get rid of the static
std::cout << i << "\n";
}
};
template<typename int_func>
void with_1(int_func func) {
func(1);
}
int main(int argc, char const *argv[]) {
Printer printer;
// Can use capture by reference because we are sure printer still
// exist during execution of with_1
with_1([&printer](int i){ printer.print(i); });
return 0;
}
example
Try this:
int main(int argc, char const *argv[]) {
Printer printer;
with_1( std::bind( &Printer::print, printer, std::placeholders::_1 ) );
return 0;
}
(You'll need to #include <functional>.)
I have a C library function that expects a function pointer for callback, and I want to pass in a C++ member function. The C++ function modifies a member variable, so I can't use a static free function (as suggested in several similar posts). My attempt (shown below) fails with a compiler error.
This post comes closest to what I need:
Using a C++ class member function as a C callback function
How can I do this without static functions? Thanks!
test.h
#ifndef TEST_H_
#define TEST_H_
#ifdef __cplusplus
extern "C" {
#endif
typedef void (*handler_t)(int foo, void *bar);
void set_handler(handler_t h);
#ifdef __cplusplus
}
#endif
#endif
test.c
#include "test.h"
#include <stdlib.h>
static handler_t handler_ = NULL;
void set_handler(handler_t h) {
handler_ = h;
}
void handle_event(int foo, void *bar) {
if (handler_ != NULL) handler_(foo, bar);
}
test.cpp
#include "test.h"
#include <iostream>
using namespace std;
class Foo {
public:
Foo() : ctr_(0) {};
// handler needs to access non-static variable, so it can't be static
void handler(int foo, void *bar) { ++ctr_; }
private:
int ctr_;
};
int main(int argc, char **argv) {
// error: can't convert to "void (*)(int, void*)"
set_handler(&Foo::handler);
cout << "done" << endl;
return 0;
}
GCC barf
$ gcc test.cpp test.c
test.cpp: In function ‘int main(int, char**)’:
test.cpp:18: error: cannot convert ‘void (Foo::*)(int, void*)’ to ‘void (*)(int, void*)’ for argument ‘1’ to ‘void set_handler(void (*)(int, void*))’
It is not possible, at least with that handler_t signature.
While you can create a free function on your .cpp to wrap the member call, you need a pointer to the Foo instance:
void my_wrap(int foo, void* bar) {
Foo* some_foo_instance = ...;
some_foo_instance->handler(foo, bar);
}
int main(int argc, char **argv) {
set_handler(&my_wrap);
}
You need some void* to pass the Foo instance as a handler attribute:
// Header
typedef void (*handler_t)(int foo, void *bar, void* arg1);
void set_handler(handler_t h, void* arg1);
// Impl.
void set_handler(handler_t h, void* arg1) {
handler_ = h;
handler_arg1_ = arg1;
}
// cpp
void my_wrap(int foo, void* bar, void* arg1) {
Foo* some_foo_instance = static_cast<Foo*>(arg1);
some_foo_instance->handler(foo, bar);
}
// main
int main(int argc, char **argv) {
Foo some_concrete_instance;
set_handler(&my_wrap, static_cast<void*>(&some_concrete_instance));
}
The big question is how many times you need to call set_handler multiple times to call methods on different objects. If this answer is one, you can do something like this:
#include <boost/function.hpp>
class HandlerContext
{
static boost::function<void (int, void*)> s_func
static void forward(int foo, void* bar)
{
s_func(foo, bar);
}
public:
static void set(boost::function<int, void*> const& f)
{
s_func = f;
set_handler(&HandlerContext::forward);
}
};
If the answer is "more than once", you can have multiple forwarding functions that get their function objects out of an array. You will need to preassign slots in this case, because the function in use will indicate which callback to make.
This sentence:
I have a C library function
This means you can NOT pass it any C++ object.
If the library you are using is a C library it does not know about C++ so it can not using anything that is C++ it can only use C stuff.
You MUST make it call a free function in you code.
Now your free function can then call a method on an object (that is why C callbacks have a void* parameter (so you can pass context to the callback)).
Suppose you create a mapping function:
Foo *inst = // some instance of Foo you're keeping around...
void wrapper(int foo, void *bar){
inst->handler(foo, bar);
}
Then use wrapper as the callback. Instance semantics in a callback are kind of strange, so I'm not sure how you're going to be sure you bind to the correct instance -- if this is a singleton maybe that doesn't matter.
Here is an ugly hack I invented awhile ago to solve this problem:
#include <boost/function.hpp>
#include <boost/bind.hpp>
using ::boost::function;
using ::boost::bind;
typedef int (*callback_t)(const char *, int);
typedef function<int(const char *, int)> MyFTWFunction;
template <MyFTWFunction *callback>
class callback_binder {
public:
static int callbackThunk(const char *s, int i) {
return (*callback)(s, i);
}
};
extern void register_callback(callback_t f);
int random_func(const char *s, int i)
{
if (s && *s) {
return i;
} else {
return -1;
}
}
MyFTWFunction myfunc;
class FooClass {
public:
virtual int callme(const char *s, int x) { return 0; };
};
int main(int argc, const char *argv[])
{
FooClass foo;
myfunc = bind(&FooClass::callme, &foo, _1, _2);
register_callback(&callback_binder<&myfunc>::callbackThunk);
return 0;
}
This could probably be fixed to use stuff from TR1 and remove the dependency on Boost.
And also, of course, myfunc is a global variable. It has to be a global variable. You must have one global variable per different possible object you'd want to call back into. OTOH, you can have as many of these globals as you want.
The main issue here is that it is absolutely impossible to do what you want within the given constraints. The pointer to the object you want to call back into has to come from somewhere. In some languages (like Python for example) you can create a function on-the-fly that has it's own copy of the object pointer. This cannot be done in C++. All functions must exist completely at compile time. You cannot create new function instances at run time.
With C++0x, you can sort of create functions at runtime with lambda functions. But these functions have an unspecified type and there is absolutely no way you could ever then pass them to a C function and have it work. Lambda expressions are meant to be supplied as template parameters and it's pretty hard to use them for anything else because their address can't be taken, and even if it could you ccouldn't actually know what type the pointer is pointing to.
I highly recommend not using it. The little void * most callback interfaces allow you to specify that gets handed back to you along with the data is meant to hold an object pointer of some kind. If possible, you should be doing that instead.
If you have control over how handler is defined, I recommend using Boost function objects instead of function pointers.
If you HAVE to use function pointers, define handler_t with an extra void* whose value is passed along with the handler, watch out for the gotchas Martin York linked in a comment. Then you have something like this:
typedef void (*handler_t)(int foo, void *bar, void *data);
static handler_t handler_ = NULL;
static void* handler_data_ = NULL;
void set_handler(handler_t h, void *d = NULL) {
handler_ = h;
handler_data = d;
}
void handle_event(int foo, void *bar) {
if (handler_ != NULL) handler_(foo, bar, handler_data_);
}
void foo_handler(int foo, void *bar, void *data) {
Foo *fooObj = static_cast<Foo*>(data);
fooObj->handler(foo, bar);
}
// in main
set_handler(foo_handler, &some_foo_object);
Out of curiosity, I thought I'd try and write a basic C++ class that mimics C#'s multiple delegate pattern. The code below mostly does the job, with the nasty sacrifice of losing almost all type-safety, but having to use the initial dummy parameter to set up the va_list really seems a bit off. Is there a way to use va_list without this?
I do realize there are ways to do this with (for example) boost, but I was aiming for something dead simple that used just the standard library.
#include <vector>
#include <iostream>
#include <string>
#include <stdarg.h>
#include <algorithm>
using namespace std;
class CDelegate
{
public:
virtual bool operator()(va_list params) = 0;
};
class CMultipleDelegateCaller
{
public:
typedef vector<CDelegate*> CDelegateVector;
CMultipleDelegateCaller& operator+=(CDelegate &rDelegate)
{
m_apDelegates.push_back(&rDelegate);
return (*this);
}
CMultipleDelegateCaller& operator-=(CDelegate &rDelegate)
{
CDelegateVector::iterator iter =
find(m_apDelegates.begin(), m_apDelegates.end(), &rDelegate);
if (m_apDelegates.end() != iter) m_apDelegates.erase(iter);
return (*this);
}
bool Call(int iDummy, ...)
{
va_list params;
CDelegate* pDelegate;
CDelegateVector::iterator iter;
for (iter = m_apDelegates.begin(); iter != m_apDelegates.end(); ++iter)
{
pDelegate = *iter;
va_start(params, iDummy);
if (!(*pDelegate)(params)) return false;
va_end(params);
}
return true;
}
private:
CDelegateVector m_apDelegates;
};
class CTestDelegate:
public CDelegate
{
public:
CTestDelegate():m_iId(++s_iCount) {}
virtual bool operator()(va_list params)
{
int iIntParam = va_arg(params, int);
char* szCharPtrParam = va_arg(params, char*);
string* psStringParam = va_arg(params, string*);
cout<<m_iId<<"{"
<<iIntParam<<", "
<<szCharPtrParam<<", "
<<*psStringParam<<"}"<<endl;
return true;
}
int m_iId;
static int s_iCount;
};
int CTestDelegate::s_iCount = 0;
int main(int argc, char* argv[])
{
CMultipleDelegateCaller cDelegateCaller;
CTestDelegate cTestDelegate1;
CTestDelegate cTestDelegate2;
cout<<"--------------------"<<endl;
cDelegateCaller += cTestDelegate1;
cDelegateCaller += cTestDelegate2;
string sString("World");
cDelegateCaller.Call(1, 2, "Hello", &sString);
cout<<"--------------------"<<endl;
cDelegateCaller -= cTestDelegate1;
cDelegateCaller.Call(1, 2, "Hello", &sString);
cout<<"--------------------"<<endl;
cDelegateCaller -= cTestDelegate2;
cDelegateCaller.Call(1, 2, "Hello", &sString);
cout<<"--------------------"<<endl;
cin>>sString;
return 0;
}
Functions with ellipsis in C++ is only for compatibility with C. Using C++ I'd return temporary helper object in Call function and add template operator% to pass variable number of arguments. To use it in the following way:
cDelegateCaller.Call() % 2 % "Hello" % sString; // dummy argument isn't required
As to your question, Standard requires to invoke va_start before any access to the unnamed arguments. And va_start requires second argument which is the identifier of the rightmost parameter in the variable parameter list in the function definition.
Out of Kirill's answer you can conclude that it's possible to create a type-safe delegate, using a template argument-combining function. This function also needs a dummy starting point, but has the benefit of type-safety.
The FastFormat library uses this, boost uses this, and I once provided another example in an answer to another question.