c ++将序列上限传递给可变函数模板

Question

我在模板的帮助下以一种特定的方式调用 APIs，我在传递常量参数时遇到了一个问题。

我尝试 int bound:

    template <typename F, typename ...Ts> 
          static int f3(int bound, CString file, int line, CString Caller,
                        CString f_name, F f, Ts&& ...ts) {
                 int err = fn(bound, file, line, Caller, f_name,
                 f, std::tuple<Ts...>(ts...), seq3<bound>{},  // error C2975 
                 seq1<sizeof...(Ts)>{});                  
                 return err;
    }

主线：

int const bound; 
bound = 4; 

err = fn(bound, api(GetModuleFileName), rval, nullptr, path, MAX_PATH, L"EXE-path");

compiler error C2975: 'N': invalid template argument for 'seq3', expected compile-time constant expression

如何解决这个问题？

我现在的解决方法：

err = f3(api(GetModuleFileName), rval, nullptr, path, MAX_PATH, L"EXE-path");

f3 是具有 3 个参数的 API 的特化，因为我到现在为止无法传递上限 - 在这种情况下为 4 - 用于生成序列：<1,2 ,3>。调用带有 3 个参数的 API 需要此序列，其中 tupel 从 f3() 中的参数 rval 开始。

背景：

api 是一个#define

f3 调用 API。

f3 在 sequence/tupel 的 0 位置处理 API 的 return 值。

f3 使用所有参数调用另一个用于记录调试信息的可变参数函数。

两个函数调用的一个元组和两个序列。

问题：

我想传递一个参数来控制不是由 tupel-size 而是由 API 函数签名给出的序列的上限。

我只想要一个 fn() 用于所有 APIs 而不是 f0()、f1()、f2()、f3() ..... 用于 APIs 0、1、2、3 ... 参数。

我想要这样的东西：

err = fn(seq3<4>, api(GetModuleFileName), rval, nullptr, path, MAX_PATH, L"EXE-path")

这是我的工作代码：

#include <windows.h>
#include <atlstr.h>
#include <tuple>
#include <utility>

template <int ... Ns> struct seq_3 {};                                  
template <int ... Ns> struct seq3_n {};  

template <int I, int ... Ns> struct seq3_n<I, Ns...>{
   using type = typename seq3_n<I - 1, I - 1, Ns...>::type;};

template <int ... Ns> struct seq3_n<1, Ns...>{
// skip first argument : rval, because it doesn't fit to API,
// but needed for calling other function    
   using type = seq_3<Ns...>;                };

template <int N>
   using seq3 = typename seq3_n<N>::type;   

template <int ... Ms> struct seq_1 {};                          
template <int ... Ms> struct seq1_n {};    
template <int J, int ... Ms> struct seq1_n<J, Ms...>{
   using type = typename seq1_n<J - 1, J - 1, Ms...>::type; };    
template <int ... Ms> struct seq1_n<0, Ms...> {
   using type = seq_1<Ms...>;                };
template <int M>
   using seq1 = typename seq1_n<M>::type;       

template <typename F, typename TUP, int ... INDICES3, int ... INDICES1>                        
   static int fn(CString file,  int line, CString Caller, CString f_name,
              F f, TUP tup, seq_3<INDICES3...>, seq_1<INDICES1...>) {   
                int err = 0;
                // handling of rval = first element of tuple 
                std::get<0>(tup) = f(std::get<INDICES3>(tup) ...);  // calling API  
                err = GetLastError();   
                /* calling next function (variadic too) with same tupel, but other sequence 
                 myOpenDebugOutputString(project, file, line, Caller, f_name, std::get<INDICES1>(tup) ..., "stop");
                */ 
                return err; }

template <typename F, typename ...Ts> 
   static int f3(CString file, int line, CString Caller, CString f_name,
              F f, Ts&& ...ts)  {
                int err = fn(file, line, Caller, f_name,
                f, std::tuple<Ts...>(ts...), seq3<4>{},  // sequence fixed by f3 
                seq1<sizeof...(Ts)>{});                  // 3 arguments api  + skip 1 rval = 4 
                return err;                              // given by signature of API 
}


int main() {    
    // for calling simple API GetModulFileName with 3 arguments     
    //                                      returns len(path)   
    wchar_t     path[MAX_PATH];     
    DWORD           rval = 0;   
    int         err = 0;
    rval = GetModuleFileName( nullptr, path, MAX_PATH);     
    err  = GetLastError(); 

#define api(a)  __FILE__, __LINE__, __func__, L#a, a   
// L#a becomes L"GetModuleFileName" 

    err = f3(api(GetModuleFileName), rval, nullptr, path, MAX_PATH, L"EXE-path");   

    return 0; }

提前致谢。

P.S。我正在使用 Microsoft Visual Studio 2015

更新：

我尝试按照 Richard Hodges 解决方案中的模板 api_call 进行操作。

std::tuple<GivenArgs...> tup(args...);   

// OK, but only for an api with 3 arguments 
callsite.function(std::get<0>(tup), std::get<1>(tup), std::get<2>(tup));

// compiler error too many arguments  
callsite.function(std::forward<GivenArgs>(args)..., seq1<callsite.nofArgs()>{}); 

// compiler error too few arguments
callsite.function(tup, seq1<callsite.nofArgs()>{});

备注：

seq1<3> = seq_1<0,1,2>

callsite.nofArg() = 3

How to get the correct number of arguments?

Answer 1

还不完全清楚您想如何处理错误等。我假设返回错误代码和值的元组。

这是一个通用模式，我认为它可以满足您的需求。您需要小心 emit_log 的特化和重载，尤其是对于可能不是 null 终止或包含非打印字符的字节数组。

为了方便起见，我使用了窄字符，但这个想法只需进行一些编辑就可以用于宽字符。

注意：在 linux gcc 上编辑，所以我模拟了 windows API。

#include <cstdint>
#include <utility>
#include <iostream>
#include <variant>

#define WINAPI
#define _In_opt_
#define _Out_
#define _In_

struct _hmodule {};
using HMODULE = _hmodule*;
using LPTSTR = char*;
using LPCTSTR = const char*;
using DWORD = std::uint32_t;

extern DWORD WINAPI GetModuleFileName(
  _In_opt_ HMODULE hModule,
  _Out_    LPTSTR  lpFilename,
  _In_     DWORD   nSize
);

extern WINAPI DWORD GetLastError();

template<class Ret, class...Args>
struct api_call_site
{
    const char* file;
    int line;
    const char* current_function;
    const char* called_function;
    Ret (* function)(Args...);
};

template<class Ret, class...Args>
auto make_api_call_site(const char* file, int line, const char* callername, const char* calleename, Ret (* WINAPI callee)(Args...))
{
    return api_call_site<Ret, Args...>
    {
        file, 
        line,
        callername,
        calleename,
        callee
    };
}

template<class T>
void emit_log(LPCTSTR& sep, std::ostream& os, T&& x)
{
    os << sep << x;
    sep = ",";
}

template<class Ret>
struct error_with_value
{
    DWORD error;
    Ret value;

    bool has_error() const { return error != 0; }
    friend std::ostream& operator<<(std::ostream& os, const error_with_value& ewv)
    {
        os << "{ error: " << ewv.error << ", value: ";
        LPCTSTR sep = "";
        emit_log(sep, os, ewv.value);
        os << " }";
        return os;
    }
};


#define api(a) make_api_call_site(__FILE__, __LINE__, __func__, #a, a)


// this will need some specialisations...
void emit_log(LPCTSTR& sep, std::ostream& os, std::nullptr_t)
{
    os << sep << "nullptr";
    sep = ",";
}

template<class Ret, class...Args, class...GivenArgs>
auto api_call(api_call_site<Ret, Args...> const& callsite, GivenArgs&&...args) -> error_with_value<Ret>
{
    // log call here
    std::clog << callsite.file << ":" << callsite.line << "@" << callsite.current_function << " - ";
    std::clog << "calling " << callsite.called_function << "(";
    // appropriate code to print arguments in a safe way here...
    LPCTSTR sep = "";
    using expand = int[];
    void(expand{0,
        (emit_log(sep, std::clog, args),0)...
    });
    std::clog << ")";
    error_with_value<Ret> result
    {
        0,
        callsite.function(std::forward<GivenArgs>(args)...)
    };
    result.error = GetLastError();

    std::clog << " -> returns: " << result;
    return result;
}

int main()
{
    char buffer[255];
    DWORD bufsize = 255;

    auto result = api_call(api(GetModuleFileName), nullptr, buffer, bufsize);
    if (! result.has_error())
    {
        //
    }

}

示例输出：

main.cpp:120@main - calling GetModuleFileName(nullptr,,255) -> returns: { error: 0, value: 14 }

http://coliru.stacked-crooked.com/a/e5da55af212d5500

How do I get the number of arguments in the API call?

template<class Ret, class...Args>
struct api_call_site
{
    const char* file;
    int line;
    const char* current_function;
    const char* called_function;
    Ret (* function)(Args...);

    // like this
    static constexpr std::size_t nofArgs()
    {
        return sizeof...(Args);
    } 
};

Answer 2

感谢 Richard Hodges 我可以解决我的问题：

现在每个 api 都有一个 api_call，api_call 背后的模板决定了被调用 api 的签名。比我要求的要好得多。

#include <windows.h>
#include <atlstr.h>
#include <tuple>

template <int ... Ns> struct seq_3 {};
template <int ... Ns> struct seq3_n {};
template <int I, int ... Ns> struct seq3_n<I, Ns...> {
    using type = typename seq3_n<I - 1, I - 1, Ns...>::type;
};
template <int ... Ns> struct seq3_n<1, Ns...> {
    // this sequence is more complicated in my real code, because
    // there are more variables for logging, but not for api calling     
    using type = seq_3<Ns...>;
};
template <int N>
using seq3 = typename seq3_n<N>::type;

template <int ... Ms> struct seq_1 {};
template <int ... Ms> struct seq1_n {};
template <int J, int ... Ms> struct seq1_n<J, Ms...> {
    using type = typename seq1_n<J - 1, J - 1, Ms...>::type;
};
template <int ... Ms> struct seq1_n<0, Ms...> {
    using type = seq_1<Ms...>;
};
template <int M>
using seq1 = typename seq1_n<M>::type;

// according to the solution from Richard Hodges
// *********************************************
template<typename Ret, typename...Args>   
struct api_call_site
{
    const CString file;
    int line;
    const CString Caller;
    const CString f_name;
    Ret(WINAPI* function)(Args...);     

    static constexpr std::size_t nofArgs() {
        return sizeof...(Args);
    }
};

template<typename Ret, typename...Args>
auto make_api_call_site(const CString file, int line, const CString Caller, const CString f_name, Ret(WINAPI* callee)(Args...))
// WINAPI see also here  
{
    return api_call_site<Ret, Args...>
    {
        file,
            line,
            Caller,
            f_name,
            callee
    };
}

template <typename Ret, typename...Args, typename TUP, int...INDICES3, int...INDICES1>
int fn(api_call_site<Ret, Args...> const& callsite, TUP tup, seq_3<INDICES3...>, seq_1<INDICES1...>) {
    int err = 0;
    // handling of return value from api call goes always in position 0 from tuple 
    std::get<0>(tup) = callsite.function(std::get<INDICES3>(tup) ...);
    err = GetLastError();
    /* calling next function (variadic too) with same tupel, but other sequence
    myOpenDebugOutputString(project, file, line, Caller, f_name, std::get<INDICES1>(tup) ..., "stop");
    */
    return err;
}

template<typename Ret, typename...Args, typename...GivenArgs>
int api_call(api_call_site<Ret, Args...> const& callsite, GivenArgs&&...args) 
{
    int err;
    err = fn(callsite, std::tuple<GivenArgs...>(args...), seq3 <callsite.nofArgs()+1> {}, seq1 <sizeof...(GivenArgs)> {});
    return err;
}

int main() {
    DWORD       size_path = 20;   // make it small and get error 122 
    wchar_t     path[MAX_PATH];   // ERROR_INSUFFICIENT_BUFFER
    DWORD       rval = 0;
    int         err = 0;
    CString     tolog1(L"EXE-Path determined"); 
    int         tolog2 = 25; 
    // old way without logging information 
    rval = GetModuleFileName(nullptr, path, MAX_PATH);
    err = GetLastError();

    // new way with logging any variables ... behind the must variables for the api  
    // ****************************************************************************
#define api(a) make_api_call_site(__FILE__, __LINE__, __func__, L#a, a)
    err = api_call(api(GetModuleFileName), rval, nullptr, path, size_path, tolog1, tolog2); 

    return 0;
}

c ++将序列上限传递给可变函数模板

c++ passing sequence upper range to variadic function template

c++

templates

sequences

variadic

更新：