ChatGPT解决这个技术问题 Extra ChatGPT

Initializing a static std::map<int, int> in C++

What is the right way of initializing a static map? Do we need a static function that will initialize it?


F
Ferruccio

Using C++11:

#include <map>
using namespace std;

map<int, char> m = {{1, 'a'}, {3, 'b'}, {5, 'c'}, {7, 'd'}};

Using Boost.Assign:

#include <map>
#include "boost/assign.hpp"
using namespace std;
using namespace boost::assign;

map<int, char> m = map_list_of (1, 'a') (3, 'b') (5, 'c') (7, 'd');

Every time I see something like that done with C++, I think of all the horrendous template code that must be behind it. Good example!
The beauty of all the horrendous template code that implements these utilities is that it is neatly encapsulated in a library and the end user rarely needs to deal with the complexity.
@QBziZ: If your company declines using Boost on the grounds of it not being "standard enough", I wonder what C++ library would be "standard enough". Boost is the standard companion for the C++ coder.
My problem with Boost (here, and elsewhere) is that you can often get on without it (in this case with C++11 or before C++11 with a function). Boost adds a significant compile time overhead, had tons of files to park into your repository (and to have to copy around/zip/extract if you are making an archive). That's the reason I try not to use it. I know you can choose what files to include/not include, but you usually don't want to have to worry about Boost's cross dependencies with itself so you just copy the whole thing around.
My problem with Boost is that it often has several new library dependencies, which generally means MORE packages that need to be installed to work correctly. We already need libstdc++. For example, the Boost ASIO library, requires at least 2 new libraries(probably more) that need to be installed. C++11/14 does make it a lot easier to not need Boost.
P
PierreBdR

Best way is to use a function:

#include <map>

using namespace std;

map<int,int> create_map()
{
  map<int,int> m;
  m[1] = 2;
  m[3] = 4;
  m[5] = 6;
  return m;
}

map<int,int> m = create_map();

Why is this the 'best'? Why for example is it better than @Dreamer's answer?
I think it's "best" because it's really simple and doesn't depend on other structures existing (such as the Boost::Assign or a reimplementation of it). And compared to @Dreamer's answer, well, I avoid creating a whole structure only for initializing a map ...
Note there is a danger here. extern variables will not have their correct values in this "before main run-time constructor" if the compiler only saw the extern declaration, but has not run into the actual variable definition yet.
No, the danger is that there is nothing saying in which order the static variables should be initialized (at least across compilation units). But this is not a problem linked to this question. This is a general problem with static variables.
no boost AND no C++11 => +1. Notice that function can be used to initialize a const map<int,int> m = create_map() (and so, initialize const members of a class in the initialization list: struct MyClass {const map<int, int> m; MyClass(); }; MyClass::MyClass() : m(create_map())
V
Vite Falcon

It's not a complicated issue to make something similar to boost. Here's a class with just three functions, including the constructor, to replicate what boost did (almost).

template <typename T, typename U>
class create_map
{
private:
    std::map<T, U> m_map;
public:
    create_map(const T& key, const U& val)
    {
        m_map[key] = val;
    }

    create_map<T, U>& operator()(const T& key, const U& val)
    {
        m_map[key] = val;
        return *this;
    }

    operator std::map<T, U>()
    {
        return m_map;
    }
};

Usage:

std::map mymap = create_map<int, int >(1,2)(3,4)(5,6);

The above code works best for initialization of global variables or static members of a class which needs to be initialized and you have no idea when it gets used first but you want to assure that the values are available in it.

If say, you've got to insert elements into an existing std::map... here's another class for you.

template <typename MapType>
class map_add_values {
private:
    MapType mMap;
public:
    typedef typename MapType::key_type KeyType;
    typedef typename MapType::mapped_type MappedType;

    map_add_values(const KeyType& key, const MappedType& val)
    {
        mMap[key] = val;
    }

    map_add_values& operator()(const KeyType& key, const MappedType& val) {
        mMap[key] = val;
        return *this;
    }

    void to (MapType& map) {
        map.insert(mMap.begin(), mMap.end());
    }
};

Usage:

typedef std::map<int, int> Int2IntMap;
Int2IntMap testMap;
map_add_values<Int2IntMap>(1,2)(3,4)(5,6).to(testMap);

See it in action with GCC 4.7.2 here: http://ideone.com/3uYJiH

############### EVERYTHING BELOW THIS IS OBSOLETE #################

EDIT: The map_add_values class below, which was the original solution I had suggested, would fail when it comes to GCC 4.5+. Please look at the code above for how to add values to existing map.


template<typename T, typename U>
class map_add_values
{
private:
    std::map<T,U>& m_map;
public:
    map_add_values(std::map<T, U>& _map):m_map(_map){}
    map_add_values& operator()(const T& _key, const U& _val)
    {
        m_map[key] = val;
        return *this;
    }
};

Usage:

std::map<int, int> my_map;
// Later somewhere along the code
map_add_values<int,int>(my_map)(1,2)(3,4)(5,6);

NOTE: Previously I used a operator [] for adding the actual values. This is not possible as commented by dalle.

##################### END OF OBSOLETE SECTION #####################


I'm using your first sample as to bind error-numbers (from an enum) with messages - it is working like a charm - thank you.
operator[] only takes a single argument.
@dalle: Good catch! For some reason I thought overloaded [] operators could accept more.
This is a fantastic answer. It's a shame the OP never selected one. You deserve mega props.
the map_add_values doesn't work in gcc, which complains: error: conflicting declaration ‘map_add_values<int, int> my_map’ error: ‘my_map’ has a previous declaration as ‘std::map<int, int> my_map’
B
Brian Neal

Here is another way that uses the 2-element data constructor. No functions are needed to initialize it. There is no 3rd party code (Boost), no static functions or objects, no tricks, just simple C++:

#include <map>
#include <string>

typedef std::map<std::string, int> MyMap;

const MyMap::value_type rawData[] = {
   MyMap::value_type("hello", 42),
   MyMap::value_type("world", 88),
};
const int numElems = sizeof rawData / sizeof rawData[0];
MyMap myMap(rawData, rawData + numElems);

Since I wrote this answer C++11 is out. You can now directly initialize STL containers using the new initializer list feature:

const MyMap myMap = { {"hello", 42}, {"world", 88} };

i
isnullxbh

For example:

const std::map<LogLevel, const char*> g_log_levels_dsc =
{
    { LogLevel::Disabled, "[---]" },
    { LogLevel::Info,     "[inf]" },
    { LogLevel::Warning,  "[wrn]" },
    { LogLevel::Error,    "[err]" },
    { LogLevel::Debug,    "[dbg]" }
};

If map is a data member of a class, you can initialize it directly in header by the following way (since C++17):

// Example

template<>
class StringConverter<CacheMode> final
{
public:
    static auto convert(CacheMode mode) -> const std::string&
    {
        // validate...
        return s_modes.at(mode);
    }

private:
    static inline const std::map<CacheMode, std::string> s_modes =
        {
            { CacheMode::All, "All" },
            { CacheMode::Selective, "Selective" },
            { CacheMode::None, "None" }
            // etc
        };
}; 

In this case in the example, it's better to use std::array
@prehistoricpenguin, why?
For performance reasons, this function may be a hot point, std::array would be faster than a map look up.
@prehistoricpenguin, could you provide an example with std::array?
Maybe for CPU performance . . . but without knowing the integer value of LogLevel you're taking a risk for memory performance. This enum is going to be terrible in an array. enum LogLevel { Disabled=-100, Info, Warning=500, Error, Debug=32768 };
D
Drealmer

I would wrap the map inside a static object, and put the map initialisation code in the constructor of this object, this way you are sure the map is created before the initialisation code is executed.


I'm with you on this one. It's also a tad faster :)
Tad faster than what? A global static with an initializer? No, it's not (remember about RVO).
Nice answer. I would be happy if I see the actual example code
u
user3826594

Just wanted to share a pure C++ 98 work around:

#include <map>

std::map<std::string, std::string> aka;

struct akaInit
{
    akaInit()
    {
        aka[ "George" ] = "John";
        aka[ "Joe" ] = "Al";
        aka[ "Phil" ] = "Sue";
        aka[ "Smitty" ] = "Yando";
    }
} AkaInit;

this doesn't work for object without default constructor, insert method should be preferred IMHO
i
isnullxbh

You can try:

std::map <int, int> mymap = 
{
        std::pair <int, int> (1, 1),
        std::pair <int, int> (2, 2),
        std::pair <int, int> (2, 2)
};

You cannot use initializer lists with non-aggregate types before C++11, in which case you may as well use the shorter syntax {1, 2} instead of std::pair<int, int>(1, 2).
E
Emanuele Benedetti

If you are stuck with C++98 and don't want to use boost, here there is the solution I use when I need to initialize a static map:

typedef std::pair< int, char > elemPair_t;
elemPair_t elemPairs[] = 
{
    elemPair_t( 1, 'a'), 
    elemPair_t( 3, 'b' ), 
    elemPair_t( 5, 'c' ), 
    elemPair_t( 7, 'd' )
};

const std::map< int, char > myMap( &elemPairs[ 0 ], &elemPairs[ sizeof( elemPairs ) / sizeof( elemPairs[ 0 ] ) ] );

e
eduffy

This is similar to PierreBdR, without copying the map.

#include <map>

using namespace std;

bool create_map(map<int,int> &m)
{
  m[1] = 2;
  m[3] = 4;
  m[5] = 6;
  return true;
}

static map<int,int> m;
static bool _dummy = create_map (m);

It probably wouldn't have been copied anyway.
but this way map couldn't be static const, could it?
H
Hans Olsson

In addition to the good top answer of using

const std::map<int, int> m = {{1,1},{4,2},{9,3},{16,4},{32,9}}

there's an additional possibility by directly calling a lambda that can be useful in a few cases:

const std::map<int, int> m = []()->auto {
  std::map<int, int> m;
  m[1]=1;
  m[4]=2;
  m[9]=3;
  m[16]=4;
  m[32]=9;
  return m;
}();

Clearly a simple initializer list is better when writing this from scratch with literal values, but it does open up additional possibilities:

const std::map<int, int> m = []()->auto {
  std::map<int, int> m;
  for(int i=1;i<5;++i) m[i*i]=i;
  m[32]=9;
  return m;
}();

(Obviously it should be a normal function if you want to re-use it; and this does require recent C++.)


u
user2185945

You have some very good answers here, but I'm to me, it looks like a case of "when all you know is a hammer"...

The simplest answer of to why there is no standard way to initialise a static map, is there is no good reason to ever use a static map...

A map is a structure designed for fast lookup, of an unknown set of elements. If you know the elements before hand, simply use a C-array. Enter the values in a sorted manner, or run sort on them, if you can't do this. You can then get log(n) performance by using the stl::functions to loop-up entries, lower_bound/upper_bound. When I have tested this previously they normally perform at least 4 times faster than a map.

The advantages are many fold... - faster performance (*4, I've measured on many CPU's types, it's always around 4) - simpler debugging. It's just easier to see what's going on with a linear layout. - Trivial implementations of copy operations, should that become necessary. - It allocates no memory at run time, so will never throw an exception. - It's a standard interface, and so is very easy to share across, DLL's, or languages, etc.

I could go on, but if you want more, why not look at Stroustrup's many blogs on the subject.


Performance is not the only reason for using a map. For example, there are many cases, where you want to link values together (for example, an error code with an error message), and a map makes the use and access relatively simple. But a link to these blog entries may be interesting, maybe I'm doing something wrong.
An array is much easier and has higher performance if you can use it. But if the indices (keys) are not contiguous, and widely spaced, you need a map.
A map is also a useful form for representing a partial function (function in the mathematical sense; but also, kind of, in the programming sense). An array does not do that. You can't, say, lookup data from an array using a string.
Your answer does not attempt to answer the valid question, and instead speculates on the limitations of the language, proposes solutions to different problems, hence downvote. A real scenario - mapping (continuous or not) library error codes to text strings. With array, search time is O(n), which can be improved by static maping to O(log(n)).
If indeed "there is no good reason to ever use a static map..." then it is very strange that syntax (initializer lists) that make them easy to use was added in C++11.