为结构定义operator <

Question

I sometimes use small structs as keys in maps, and so I have to define an operator< for them. Usually, this ends up looking something like this:

struct MyStruct
{
    A a;
    B b;
    C c;

    bool operator<(const MyStruct& rhs) const
    {
        if (a < rhs.a)
        {
           return true;
        }
        else if (a == rhs.a)
        {
            if (b < rhs.b)
            {
                return true;
            }
            else if (b == rhs.b)
            {
                return c < rhs.c;
            }
        }

        return false;
    }
};

This seems awfully verbose and error-prone. Is there a better way, or some easy way to automate definition of operator< for a struct or class ?

I know some people like to just use something like memcmp(this, &rhs, sizeof(MyStruct)) < 0 , but this may not work correctly if there are padding bytes between the members, or if there are char string arrays that may contain garbage after the null terminators.

Answer 1

This is quite an old question and as a consequence all answers here are obsolete. C++11 allows a more elegant and efficient solution:

bool operator <(const MyStruct& x, const MyStruct& y) {
    return std::tie(x.a, x.b, x.c) < std::tie(y.a, y.b, y.c);
}

Why is this better than using boost::make_tuple ? Because make_tuple will create copies of all the data members, which can be costly. std::tie , by contrast, will just create a thin wrapper of references (which the compiler will probably optimise away entirely).

In fact, the above code should now be considered the idiomatic solution to implementing a lexicographical compare for structures with several data members.

Answer 2

Others have mentioned boost::tuple , which gives you a lexicographical comparison. If you want to keep it as a structure with named elements, you can create temporary tuples for comparison:

bool operator<(const MyStruct& x, const MyStruct& y)
{
    return boost::make_tuple(x.a,x.b,x.c) < boost::make_tuple(y.a,y.b,y.c);
}

In C++0x, this becomes std::make_tuple() .

UPDATE: And now C++11 is here, it becomes std::tie() , to make a tuple of references without copying the objects. See Konrad Rudolph's new answer for details.

Answer 3

I would do this:

#define COMPARE(x) if((x) < (rhs.x)) return true; \
                   if((x) > (rhs.x)) return false;
COMPARE(a)
COMPARE(b)
COMPARE(c)
return false;
#undef COMPARE

Answer 4

在这种情况下，您可以使用boost::tuple<int, int, int> -它的operator <可以按照您想要的方式工作。

Answer 5

I think the easiest way is to stick with the < operator for all comparisons and don't use > or ==. Below is the pattern I follow, and you can follow for all your structs

typedef struct X
{
    int a;
    std::string b;
    int c;
    std::string d;

    bool operator <( const X& rhs ) const
    {
        if (a < rhs.a) { return true; }
        else if ( rhs.a < a ) { return false; }

        // if neither of the above were true then 
        // we are consdidered equal using strict weak ordering
        // so we move on to compare the next item in the struct

        if (b < rhs.b) { return true; }
        if ( rhs.b < b ) { return false; }

        if (c < rhs.c) { return true; }
        if ( rhs.c < c ) { return false; }

        if (d < rhs.d) { return true; }
        if ( rhs.d < d ) { return false; }

        // if both are completely equal (based on strict weak ordering)
        // then just return false since equality doesn't yield less than
        return false;
    }
};

Answer 6

The best way I know is to use a boost tuple . It offers among others a builtin comparison and constructors.

#include <boost/tuple/tuple.hpp>
#include <boost/tuple/tuple_comparison.hpp>

typedef boost::tuple<int,int,int> MyStruct;

MyStruct x0(1,2,3), x1(1,2,2);
if( x0 < x1 )
   ...

I also like Mike Seymors suggestion to use temporary tuples through boost's make_tuple

Answer 7

I usually implement lexicographical ordering this way:

bool operator < (const MyObject& obj)
{
    if( first != obj.first ){
        return first < obj.first;
    }
    if( second != obj.second ){
        return second < obj.second;
    }
    if( third != obj.third ){
        return third < obj.third
    }
    ...
}

Mind you it needs extra consideration for floating point values (G++ warnings), for those something like this would be better:

bool operator < (const MyObject& obj)
{
    if( first < obj.first ){
        return true;
    }
    if( first > obj.first ){
        return false;
    }
    if( second < obj.second ){
        return true;
    }
    if( second > obj.second ){
        return false;
    }
    ...
}

Answer 8

#include <iostream>

#include <boost/fusion/include/adapt_struct.hpp>
#include <boost/fusion/include/less.hpp>

struct MyStruct {
   int a, b, c;
};

BOOST_FUSION_ADAPT_STRUCT( MyStruct,
                           ( int, a )
                           ( int, b )
                           ( int, c )
                          )

bool operator<( const MyStruct &s1, const MyStruct &s2 )
{
   return boost::fusion::less( s1, s2 );
}

int main()
{
   MyStruct s1 = { 0, 4, 8 }, s2 = { 0, 4, 9 };
   std::cout << ( s1 < s2 ? "is less" : "is not less" ) << std::endl;
}

Answer 9

if you can't use boost, you could try something like:

#include <iostream>

using namespace std;

template <typename T>
struct is_gt
{
  is_gt(const T& l, const T&r) : _s(l > r) {}

  template <typename T2>
  inline is_gt<T>& operator()(const T2& l, const T2& r)
  {
    if (!_s)
    {
      _s = l > r;
    }
    return *this;
  }

  inline bool operator!() const { return !_s; }

  bool _s;
};

struct foo
{
  int a;
  int b;
  int c;

  friend bool operator<(const foo& l, const foo& r);
};

bool operator<(const foo& l, const foo& r)
{
  return !is_gt<int>(l.a, r.a)(l.b, r.b)(l.c, r.c);
}

int main(void)
{
  foo s1 = { 1, 4, 8 }, s2 = { 2, 4, 9 };
  cout << "s1 < s2: " << (s1 < s2) << endl;
  return 0;
}

I guess this avoids any macros, and as long as the types in the structure support <, it should work. Of course there is overhead for this approach, constructing is_gt and then superflous branches for each parameter if one of the values is greater...

Edit:

Modified based on comments, this version should now short-circuit as well, now uses two bools to keep state (not sure there's a way to do this with a single bool).

template <typename T>
struct is_lt
{
  is_lt(const T& l, const T&r) : _s(l < r), _e(l == r) {}

  template <typename T2>
  inline bool operator()(const T2& l, const T2& r)
  {
    if (!_s && _e)
    {
      _s = l < r;
      _e = l == r;
    }
    return _s;
  }

  inline operator bool() const { return _s; }

  bool _s;
  bool _e;
};

and

bool operator<(const foo& l, const foo& r)
{
  is_lt<int> test(l.a, r.a);
  return test || test(l.b, r.b) || test(l.c, r.c);
}

just build up a collection of such functors for various comparisons..

Answer 10

I just learned the boost::tuple trick, thanks, @Mike Seymour!

If you can't afford Boost, my favorite idiom is:

bool operator<(const MyStruct& rhs) const
{
    if (a < rhs.a)  return true;
    if (a > rhs.a)  return false;

    if (b < rhs.b)  return true;
    if (b > rhs.b)  return false;

    return (c < rhs.c);
}

which I like because it sets everything in parallel structure that makes errors and omissions easier to spot.

But, of course, you are unit testing this anyway, right?

Answer 11

I wrote a perl script to help me. For example given:

class A
{
int a;
int b;
int c;

It would emit:

bool operator<(const A& left, const A& right)
{
    bool result(false);

    if(left.a != right.a)
    {
        result = left.a < right.a;
    }
    else if(left.b != right.b)
    {
        result = left.b < right.b;
    }
    else
    {
        result = left.c < right.c;
    }

    return result;
}

Code (it's a bit long):

#!/usr/bin/perl

use strict;

main:

my $line = <>;
chomp $line;
$line =~ s/^ *//;

my ($temp, $line, $temp) = split / /, $line;

print "bool operator<(const $line& left, const $line& right)\n{\n";
print "    bool result(false);\n\n";

my $ifText = "if";

$line = <>;

while($line)
{
    if($line =~ /{/)
    {
        $line = <>;
        next;
    }
    if($line =~ /}/)
    {
        last;
    }

    chomp $line;
    $line =~ s/^ *//;

    my ($type, $name) = split / /, $line;
    $name =~ s/; *$//;

    $line = <>;
    if($line && !($line =~ /}/))
    {
        print "    $ifText(left.$name != right.$name)\n";
        print "    {\n";
        print "        result = left.$name < right.$name;\n";
        print "    }\n";

        $ifText = "else if";
    }
    else
    {
        print "    else\n";
        print "    {\n";
        print "        result = left.$name < right.$name;\n";
        print "    }\n";

        last;
    }
}

print "\n    return result;\n}\n";

Answer 12

bool operator <(const A& l, const A& r)
{

    int[] offsets = { offsetof(A, a), offsetof(A, b), offsetof(A, c) };
    for(int i = 0; i < sizeof(offsets)/sizeof(int); i++)
    {
        int ta = *(int*)(((const char*)&l)+offsets[i]);
        int tb = *(int*)(((const char*)&r)+offsets[i]);

        if (ta < tb)
             return true;
        else if (ta > tb)
             break;

    }
    return false;
}

Answer 13

When you can produce iterators over the elements defining the lexicographic order you can use std::lexicographic_compare , from <algorithm> .

Otherwise I suggest basing comparisons on old three-value compare functions, eg as follows:

#include <iostream>

int compared( int a, int b )
{
    return (a < b? -1 : a == b? 0 : +1);
}

struct MyStruct
{
    friend int compared( MyStruct const&, MyStruct const& );
    int a;
    int b;
    int c;

    bool operator<( MyStruct const& rhs ) const
    {
        return (compared( *this, rhs ) < 0);
    }
};

int compared( MyStruct const& lhs, MyStruct const& rhs )
{
    if( int x = compared( lhs.a, rhs.a ) ) { return x; }
    if( int x = compared( lhs.b, rhs.b ) ) { return x; }
    if( int x = compared( lhs.c, rhs.c ) ) { return x; }
    return 0;
}

int main()
{
    MyStruct const  s1 = { 0, 4, 8 };
    MyStruct const  s2 = { 0, 4, 9 };
    std::cout << ( s1 < s2 ? "is less" : "is not less" ) << std::endl;
}

I included the last if and return in the compare function just for generality. I imagine it can help maintenance to very rigidly adhere to a single system. Otherwise you could just do a return compared( lhs.c, rhs.c ) there (and perhaps you prefer that).

Cheers & hth.,

− Alf

Answer 14

If three-way comparisons are more expensive than two-way, and if the more-significant portions of the structures will often be equal, it may be helpful to define field comparison functions with a 'bias' parameter, such that if 'bias' is false, they will return true when a>b, and when bias is true, they will return true if a>=b. Then one can find out if a>b by doing something like:

return compare1(a.f1,b.f1, compare2(a.f2,b.f2, compare3(a.f3,b.f3,false)));

Note that all comparisons will be performed, even if a.f1<>b.f1, but comparisons will be two-way instead of three-way.