如前所述，unique需要一个排序的容器。此外，unique实际上并没有从容器中删除元素。相反，它们被复制到末尾，unique返回一个指向第一个重复元素的迭代器，并且您需要调用erase来实际删除这些元素。

这里有一个模板来帮你做这件事:

template<typename T>
void removeDuplicates(std::vector<T>& vec)
{
    std::sort(vec.begin(), vec.end());
    vec.erase(std::unique(vec.begin(), vec.end()), vec.end());
}

这样称呼它:

removeDuplicates<int>(vectorname);

我不知道你在用这个干什么，所以我不能100%肯定地说，但通常当我想到“排序的，唯一的”容器时，我想到std::set。它可能更适合你的用例:

std::set<Foo> foos(vec.begin(), vec.end()); // both sorted & unique already

否则，在调用unique之前进行排序(正如其他答案所指出的那样)才是正确的方法。

如果你的类很容易转换为int型，并且你有一些内存， Unique可以在没有排序的情况下完成，而且速度快得多:

#include <vector>
#include <stdlib.h>
#include <algorithm>
int main (int argc, char* argv []) {
  //vector init
  std::vector<int> v (1000000, 0);
  std::for_each (v.begin (), v.end (), [] (int& s) {s = rand () %1000;});
  std::vector<int> v1 (v);
  int beg (0), end (0), duration (0);
  beg = clock ();
  {
    std::sort (v.begin (), v.end ());
    auto i (v.begin ());
    i = std::unique (v.begin (), v.end ());
    if (i != v.end ()) v.erase (i, v.end ());
  }
  end = clock ();
  duration = (int) (end - beg);
  std::cout << "\tduration sort + unique == " << duration << std::endl;

  int n (0);
  duration = 0;
  beg = clock ();
  std::for_each (v1.begin (), v1.end (), [&n] (const int& s) {if (s >= n) n = s+1;});
  std::vector<int> tab (n, 0);
  {
    auto i (v1.begin ());
    std::for_each (v1.begin (), v1.end (), [&i, &tab] (const int& s) {
      if (!tab [s]) {
        *i++ = s;
        ++tab [s];
      }
    });
    std::sort (v1.begin (), i);
    v1.erase (i, v1.end ());
  }
  end = clock ();
  duration = (int) (end - beg);
  std::cout << "\tduration unique + sort == " << duration << std::endl;
  if (v == v1) {
    std::cout << "and results are same" << std::endl;
  }
  else {
    std::cout << "but result differs" << std::endl;
  }  
}

典型结果: Duration sort + unique == 38985 持续时间唯一+排序== 2500 结果是一样的

Nate Kohl建议的标准方法，使用vector, sort + unique:

sort( vec.begin(), vec.end() );
vec.erase( unique( vec.begin(), vec.end() ), vec.end() );

对指针向量不起作用。

仔细看看cplusplus.com上的这个例子。

在他们的例子中，移到最后的“所谓的副本”实际上显示为?(未定义的值)，因为这些“所谓的重复”有时是“额外的元素”，有时是原始向量中的“缺失元素”。

当在指向对象的指针向量上使用std::unique()时，会出现一个问题(内存泄漏，从HEAP读取数据的错误，重复释放，这会导致分割错误等)。

这是我对这个问题的解决方案:用ptgi::unique()替换std::unique()。

请看下面的ptgi_unique.hpp文件:

// ptgi::unique()
//
// Fix a problem in std::unique(), such that none of the original elts in the collection are lost or duplicate.
// ptgi::unique() has the same interface as std::unique()
//
// There is the 2 argument version which calls the default operator== to compare elements.
//
// There is the 3 argument version, which you can pass a user defined functor for specialized comparison.
//
// ptgi::unique() is an improved version of std::unique() which doesn't looose any of the original data
// in the collection, nor does it create duplicates.
//
// After ptgi::unique(), every old element in the original collection is still present in the re-ordered collection,
// except that duplicates have been moved to a contiguous range [dupPosition, last) at the end.
//
// Thus on output:
//  [begin, dupPosition) range are unique elements.
//  [dupPosition, last) range are duplicates which can be removed.
// where:
//  [] means inclusive, and
//  () means exclusive.
//
// In the original std::unique() non-duplicates at end are moved downward toward beginning.
// In the improved ptgi:unique(), non-duplicates at end are swapped with duplicates near beginning.
//
// In addition if you have a collection of ptrs to objects, the regular std::unique() will loose memory,
// and can possibly delete the same pointer multiple times (leading to SEGMENTATION VIOLATION on Linux machines)
// but ptgi::unique() won't.  Use valgrind(1) to find such memory leak problems!!!
//
// NOTE: IF you have a vector of pointers, that is, std::vector<Object*>, then upon return from ptgi::unique()
// you would normally do the following to get rid of the duplicate objects in the HEAP:
//
//  // delete objects from HEAP
//  std::vector<Object*> objects;
//  for (iter = dupPosition; iter != objects.end(); ++iter)
//  {
//      delete (*iter);
//  }
//
//  // shrink the vector. But Object * pointers are NOT followed for duplicate deletes, this shrinks the vector.size())
//  objects.erase(dupPosition, objects.end));
//
// NOTE: But if you have a vector of objects, that is: std::vector<Object>, then upon return from ptgi::unique(), it
// suffices to just call vector:erase(, as erase will automatically call delete on each object in the
// [dupPosition, end) range for you:
//
//  std::vector<Object> objects;
//  objects.erase(dupPosition, last);
//
//==========================================================================================================
// Example of differences between std::unique() vs ptgi::unique().
//
//  Given:
//      int data[] = {10, 11, 21};
//
//  Given this functor: ArrayOfIntegersEqualByTen:
//      A functor which compares two integers a[i] and a[j] in an int a[] array, after division by 10:
//  
//  // given an int data[] array, remove consecutive duplicates from it.
//  // functor used for std::unique (BUGGY) or ptgi::unique(IMPROVED)
//
//  // Two numbers equal if, when divided by 10 (integer division), the quotients are the same.
//  // Hence 50..59 are equal, 60..69 are equal, etc.
//  struct ArrayOfIntegersEqualByTen: public std::equal_to<int>
//  {
//      bool operator() (const int& arg1, const int& arg2) const
//      {
//          return ((arg1/10) == (arg2/10));
//      }
//  };
//  
//  Now, if we call (problematic) std::unique( data, data+3, ArrayOfIntegersEqualByTen() );
//  
//  TEST1: BEFORE UNIQ: 10,11,21
//  TEST1: AFTER UNIQ: 10,21,21
//  DUP_INX=2
//  
//      PROBLEM: 11 is lost, and extra 21 has been added.
//  
//  More complicated example:
//  
//  TEST2: BEFORE UNIQ: 10,20,21,22,30,31,23,24,11
//  TEST2: AFTER UNIQ: 10,20,30,23,11,31,23,24,11
//  DUP_INX=5
//  
//      Problem: 21 and 22 are deleted.
//      Problem: 11 and 23 are duplicated.
//  
//  
//  NOW if ptgi::unique is called instead of std::unique, both problems go away:
//  
//  DEBUG: TEST1: NEW_WAY=1
//  TEST1: BEFORE UNIQ: 10,11,21
//  TEST1: AFTER UNIQ: 10,21,11
//  DUP_INX=2
//  
//  DEBUG: TEST2: NEW_WAY=1
//  TEST2: BEFORE UNIQ: 10,20,21,22,30,31,23,24,11
//  TEST2: AFTER UNIQ: 10,20,30,23,11,31,22,24,21
//  DUP_INX=5
//
//  @SEE: look at the "case study" below to understand which the last "AFTER UNIQ" results with that order:
//  TEST2: AFTER UNIQ: 10,20,30,23,11,31,22,24,21
//
//==========================================================================================================
// Case Study: how ptgi::unique() works:
//  Remember we "remove adjacent duplicates".
//  In this example, the input is NOT fully sorted when ptgi:unique() is called.
//
//  I put | separatators, BEFORE UNIQ to illustrate this
//  10  | 20,21,22 |  30,31 |  23,24 | 11
//
//  In example above, 20, 21, 22 are "same" since dividing by 10 gives 2 quotient.
//  And 30,31 are "same", since /10 quotient is 3.
//  And 23, 24 are same, since /10 quotient is 2.
//  And 11 is "group of one" by itself.
//  So there are 5 groups, but the 4th group (23, 24) happens to be equal to group 2 (20, 21, 22)
//  So there are 5 groups, and the 5th group (11) is equal to group 1 (10)
//
//  R = result
//  F = first
//
//  10, 20, 21, 22, 30, 31, 23, 24, 11
//  R    F
//
//  10 is result, and first points to 20, and R != F (10 != 20) so bump R:
//       R
//       F
//
//  Now we hits the "optimized out swap logic".
//  (avoid swap because R == F)
//
//  // now bump F until R != F (integer division by 10)
//  10, 20, 21, 22, 30, 31, 23, 24, 11
//       R   F              // 20 == 21 in 10x
//       R       F              // 20 == 22 in 10x
//       R           F          // 20 != 30, so we do a swap of ++R and F
//  (Now first hits 21, 22, then finally 30, which is different than R, so we swap bump R to 21 and swap with  30)
//  10, 20, 30, 22, 21, 31, 23, 24, 11  // after R & F swap (21 and 30)
//           R       F 
//
//  10, 20, 30, 22, 21, 31, 23, 24, 11
//           R          F           // bump F to 31, but R and F are same (30 vs 31)
//           R               F      // bump F to 23, R != F, so swap ++R with F
//  10, 20, 30, 22, 21, 31, 23, 24, 11
//                  R           F       // bump R to 22
//  10, 20, 30, 23, 21, 31, 22, 24, 11  // after the R & F swap (22 & 23 swap)
//                  R            F      // will swap 22 and 23
//                  R                F      // bump F to 24, but R and F are same in 10x
//                  R                    F  // bump F, R != F, so swap ++R  with F
//                      R                F  // R and F are diff, so swap ++R  with F (21 and 11)
//  10, 20, 30, 23, 11, 31, 22, 24, 21
//                      R                F  // aftter swap of old 21 and 11
//                      R                  F    // F now at last(), so loop terminates
//                          R               F   // bump R by 1 to point to dupPostion (first duplicate in range)
//
//  return R which now points to 31
//==========================================================================================================
// NOTES:
// 1) the #ifdef IMPROVED_STD_UNIQUE_ALGORITHM documents how we have modified the original std::unique().
// 2) I've heavily unit tested this code, including using valgrind(1), and it is *believed* to be 100% defect-free.
//
//==========================================================================================================
// History:
//  130201  dpb dbednar@ptgi.com created
//==========================================================================================================

#ifndef PTGI_UNIQUE_HPP
#define PTGI_UNIQUE_HPP

// Created to solve memory leak problems when calling std::unique() on a vector<Route*>.
// Memory leaks discovered with valgrind and unitTesting.


#include <algorithm>        // std::swap

// instead of std::myUnique, call this instead, where arg3 is a function ptr
//
// like std::unique, it puts the dups at the end, but it uses swapping to preserve original
// vector contents, to avoid memory leaks and duplicate pointers in vector<Object*>.

#ifdef IMPROVED_STD_UNIQUE_ALGORITHM
#error the #ifdef for IMPROVED_STD_UNIQUE_ALGORITHM was defined previously.. Something is wrong.
#endif

#undef IMPROVED_STD_UNIQUE_ALGORITHM
#define IMPROVED_STD_UNIQUE_ALGORITHM

// similar to std::unique, except that this version swaps elements, to avoid
// memory leaks, when vector contains pointers.
//
// Normally the input is sorted.
// Normal std::unique:
// 10 20 20 20 30   30 20 20 10
// a  b  c  d  e    f  g  h  i
//
// 10 20 30 20 10 | 30 20 20 10
// a  b  e  g  i    f  g  h  i
//
// Now GONE: c, d.
// Now DUPS: g, i.
// This causes memory leaks and segmenation faults due to duplicate deletes of same pointer!


namespace ptgi {

// Return the position of the first in range of duplicates moved to end of vector.
//
// uses operator==  of class for comparison
//
// @param [first, last) is a range to find duplicates within.
//
// @return the dupPosition position, such that [dupPosition, end) are contiguous
// duplicate elements.
// IF all items are unique, then it would return last.
//
template <class ForwardIterator>
ForwardIterator unique( ForwardIterator first, ForwardIterator last)
{
    // compare iterators, not values
    if (first == last)
        return last;

    // remember the current item that we are looking at for uniqueness
    ForwardIterator result = first;

    // result is slow ptr where to store next unique item
    // first is  fast ptr which is looking at all elts

    // the first iterator moves over all elements [begin+1, end).
    // while the current item (result) is the same as all elts
    // to the right, (first) keeps going, until you find a different
    // element pointed to by *first.  At that time, we swap them.

    while (++first != last)
    {
        if (!(*result == *first))
        {
#ifdef IMPROVED_STD_UNIQUE_ALGORITHM
            // inc result, then swap *result and *first

//          THIS IS WHAT WE WANT TO DO.
//          BUT THIS COULD SWAP AN ELEMENT WITH ITSELF, UNCECESSARILY!!!
//          std::swap( *first, *(++result));

            // BUT avoid swapping with itself when both iterators are the same
            ++result;
            if (result != first)
                std::swap( *first, *result);
#else
            // original code found in std::unique()
            // copies unique down
            *(++result) = *first;
#endif
        }
    }

    return ++result;
}

template <class ForwardIterator, class BinaryPredicate>
ForwardIterator unique( ForwardIterator first, ForwardIterator last, BinaryPredicate pred)
{
    if (first == last)
        return last;

    // remember the current item that we are looking at for uniqueness
    ForwardIterator result = first;

    while (++first != last)
    {
        if (!pred(*result,*first))
        {
#ifdef IMPROVED_STD_UNIQUE_ALGORITHM
            // inc result, then swap *result and *first

//          THIS COULD SWAP WITH ITSELF UNCECESSARILY
//          std::swap( *first, *(++result));
//
            // BUT avoid swapping with itself when both iterators are the same
            ++result;
            if (result != first)
                std::swap( *first, *result);

#else
            // original code found in std::unique()
            // copies unique down
            // causes memory leaks, and duplicate ptrs
            // and uncessarily moves in place!
            *(++result) = *first;
#endif
        }
    }

    return ++result;
}

// from now on, the #define is no longer needed, so get rid of it
#undef IMPROVED_STD_UNIQUE_ALGORITHM

} // end ptgi:: namespace

#endif

下面是我用来测试它的UNIT Test程序:

// QUESTION: in test2, I had trouble getting one line to compile,which was caused  by the declaration of operator()
// in the equal_to Predicate.  I'm not sure how to correctly resolve that issue.
// Look for //OUT lines
//
// Make sure that NOTES in ptgi_unique.hpp are correct, in how we should "cleanup" duplicates
// from both a vector<Integer> (test1()) and vector<Integer*> (test2).
// Run this with valgrind(1).
//
// In test2(), IF we use the call to std::unique(), we get this problem:
//
//  [dbednar@ipeng8 TestSortRoutes]$ ./Main7
//  TEST2: ORIG nums before UNIQUE: 10, 20, 21, 22, 30, 31, 23, 24, 11
//  TEST2: modified nums AFTER UNIQUE: 10, 20, 30, 23, 11, 31, 23, 24, 11
//  INFO: dupInx=5
//  TEST2: uniq = 10
//  TEST2: uniq = 20
//  TEST2: uniq = 30
//  TEST2: uniq = 33427744
//  TEST2: uniq = 33427808
//  Segmentation fault (core dumped)
//
// And if we run valgrind we seen various error about "read errors", "mismatched free", "definitely lost", etc.
//
//  valgrind --leak-check=full ./Main7
//  ==359== Memcheck, a memory error detector
//  ==359== Command: ./Main7
//  ==359== Invalid read of size 4
//  ==359== Invalid free() / delete / delete[]
//  ==359== HEAP SUMMARY:
//  ==359==     in use at exit: 8 bytes in 2 blocks
//  ==359== LEAK SUMMARY:
//  ==359==    definitely lost: 8 bytes in 2 blocks
// But once we replace the call in test2() to use ptgi::unique(), all valgrind() error messages disappear.
//
// 130212   dpb dbednar@ptgi.com created
// =========================================================================================================

#include <iostream> // std::cout, std::cerr
#include <string>
#include <vector>   // std::vector
#include <sstream>  // std::ostringstream
#include <algorithm>    // std::unique()
#include <functional>   // std::equal_to(), std::binary_function()
#include <cassert>  // assert() MACRO

#include "ptgi_unique.hpp"  // ptgi::unique()



// Integer is small "wrapper class" around a primitive int.
// There is no SETTER, so Integer's are IMMUTABLE, just like in JAVA.

class Integer
{
private:
    int num;
public:

    // default CTOR: "Integer zero;"
    // COMPRENSIVE CTOR:  "Integer five(5);"
    Integer( int num = 0 ) :
        num(num)
    {
    }

    // COPY CTOR
    Integer( const Integer& rhs) :
        num(rhs.num)
    {
    }

    // assignment, operator=, needs nothing special... since all data members are primitives

    // GETTER for 'num' data member
    // GETTER' are *always* const
    int getNum() const
    {
        return num;
    }   

    // NO SETTER, because IMMUTABLE (similar to Java's Integer class)

    // @return "num"
    // NB: toString() should *always* be a const method
    //
    // NOTE: it is probably more efficient to call getNum() intead
    // of toString() when printing a number:
    //
    // BETTER to do this:
    //  Integer five(5);
    //  std::cout << five.getNum() << "\n"
    // than this:
    //  std::cout << five.toString() << "\n"

    std::string toString() const
    {
        std::ostringstream oss;
        oss << num;
        return oss.str();
    }
};

// convenience typedef's for iterating over std::vector<Integer>
typedef std::vector<Integer>::iterator      IntegerVectorIterator;
typedef std::vector<Integer>::const_iterator    ConstIntegerVectorIterator;

// convenience typedef's for iterating over std::vector<Integer*>
typedef std::vector<Integer*>::iterator     IntegerStarVectorIterator;
typedef std::vector<Integer*>::const_iterator   ConstIntegerStarVectorIterator;

// functor used for std::unique or ptgi::unique() on a std::vector<Integer>
// Two numbers equal if, when divided by 10 (integer division), the quotients are the same.
// Hence 50..59 are equal, 60..69 are equal, etc.
struct IntegerEqualByTen: public std::equal_to<Integer>
{
    bool operator() (const Integer& arg1, const Integer& arg2) const
    {
        return ((arg1.getNum()/10) == (arg2.getNum()/10));
    }
};

// functor used for std::unique or ptgi::unique on a std::vector<Integer*>
// Two numbers equal if, when divided by 10 (integer division), the quotients are the same.
// Hence 50..59 are equal, 60..69 are equal, etc.
struct IntegerEqualByTenPointer: public std::equal_to<Integer*>
{
    // NB: the Integer*& looks funny to me!
    // TECHNICAL PROBLEM ELSEWHERE so had to remove the & from *&
//OUT   bool operator() (const Integer*& arg1, const Integer*& arg2) const
//
    bool operator() (const Integer* arg1, const Integer* arg2) const
    {
        return ((arg1->getNum()/10) == (arg2->getNum()/10));
    }
};

void test1();
void test2();
void printIntegerStarVector( const std::string& msg, const std::vector<Integer*>& nums );

int main()
{
    test1();
    test2();
    return 0;
}

// test1() uses a vector<Object> (namely vector<Integer>), so there is no problem with memory loss
void test1()
{
    int data[] = { 10, 20, 21, 22, 30, 31, 23, 24, 11};

    // turn C array into C++ vector
    std::vector<Integer> nums(data, data+9);

    // arg3 is a functor
    IntegerVectorIterator dupPosition = ptgi::unique( nums.begin(), nums.end(), IntegerEqualByTen() );

    nums.erase(dupPosition, nums.end());

    nums.erase(nums.begin(), dupPosition);
}

//==================================================================================
// test2() uses a vector<Integer*>, so after ptgi:unique(), we have to be careful in
// how we eliminate the duplicate Integer objects stored in the heap.
//==================================================================================
void test2()
{
    int data[] = { 10, 20, 21, 22, 30, 31, 23, 24, 11};

    // turn C array into C++ vector of Integer* pointers
    std::vector<Integer*> nums;

    // put data[] integers into equivalent Integer* objects in HEAP
    for (int inx = 0; inx < 9; ++inx)
    {
        nums.push_back( new Integer(data[inx]) );
    }

    // print the vector<Integer*> to stdout
    printIntegerStarVector( "TEST2: ORIG nums before UNIQUE", nums );

    // arg3 is a functor
#if 1
    // corrected version which fixes SEGMENTATION FAULT and all memory leaks reported by valgrind(1)
    // I THINK we want to use new C++11 cbegin() and cend(),since the equal_to predicate is passed "Integer *&"

//  DID NOT COMPILE
//OUT   IntegerStarVectorIterator dupPosition = ptgi::unique( const_cast<ConstIntegerStarVectorIterator>(nums.begin()), const_cast<ConstIntegerStarVectorIterator>(nums.end()), IntegerEqualByTenPointer() );

    // DID NOT COMPILE when equal_to predicate declared "Integer*& arg1, Integer*&  arg2"
//OUT   IntegerStarVectorIterator dupPosition = ptgi::unique( const_cast<nums::const_iterator>(nums.begin()), const_cast<nums::const_iterator>(nums.end()), IntegerEqualByTenPointer() );


    // okay when equal_to predicate declared "Integer* arg1, Integer*  arg2"
    IntegerStarVectorIterator dupPosition = ptgi::unique(nums.begin(), nums.end(), IntegerEqualByTenPointer() );
#else
    // BUGGY version that causes SEGMENTATION FAULT and valgrind(1) errors
    IntegerStarVectorIterator dupPosition = std::unique( nums.begin(), nums.end(), IntegerEqualByTenPointer() );
#endif

    printIntegerStarVector( "TEST2: modified nums AFTER UNIQUE", nums );
    int dupInx = dupPosition - nums.begin();
    std::cout << "INFO: dupInx=" << dupInx <<"\n";

    // delete the dup Integer* objects in the [dupPosition, end] range
    for (IntegerStarVectorIterator iter = dupPosition; iter != nums.end(); ++iter)
    {
        delete (*iter);
    }

    // shrink the vector
    // NB: the Integer* ptrs are NOT followed by vector::erase()
    nums.erase(dupPosition, nums.end());


    // print the uniques, by following the iter to the Integer* pointer
    for (IntegerStarVectorIterator iter = nums.begin(); iter != nums.end();  ++iter)
    {
        std::cout << "TEST2: uniq = " << (*iter)->getNum() << "\n";
    }

    // remove the unique objects from heap
    for (IntegerStarVectorIterator iter = nums.begin(); iter != nums.end();  ++iter)
    {
        delete (*iter);
    }

    // shrink the vector
    nums.erase(nums.begin(), nums.end());

    // the vector should now be completely empty
    assert( nums.size() == 0);
}

//@ print to stdout the string: "info_msg: num1, num2, .... numN\n"
void printIntegerStarVector( const std::string& msg, const std::vector<Integer*>& nums )
{
    std::cout << msg << ": ";
    int inx = 0;
    ConstIntegerStarVectorIterator  iter;

    // use const iterator and const range!
    // NB: cbegin() and cend() not supported until LATER (c++11)
    for (iter = nums.begin(), inx = 0; iter != nums.end(); ++iter, ++inx)
    {
        // output a comma seperator *AFTER* first
        if (inx > 0)
            std::cout << ", ";

        // call Integer::toString()
        std::cout << (*iter)->getNum();     // send int to stdout
//      std::cout << (*iter)->toString();   // also works, but is probably slower

    }

    // in conclusion, add newline
    std::cout << "\n";
}

如前所述，unique需要一个排序的容器。此外，unique实际上并没有从容器中删除元素。相反，它们被复制到末尾，unique返回一个指向第一个重复元素的迭代器，并且您需要调用erase来实际删除这些元素。