STL向量中隐藏的副本？(hidden copy in STL vector?)

 使用带类的向量时，我遇到意外的性能下降。 
 
 我有一个csv_File类，它读入一个csv文件并将内容存储在一个2D向量中。 有一个允许访问的成员函数，例如 
 
    csv_File file("file.csv");
    file.access(2,2);
 
 访问2,2元素。 
 
 然后，我有另一个类csv_Array，它在向量中存储多个csv_File对象，例如私有成员向量有一个允许访问的成员函数，即它返回一个csv_File对象，例如： 
 
    csv_Array file_array(5); //store 5 csv_File objects
    file_array.grab(0).access(2,2); 
 
 在第二行中，grab返回一个csv_File对象（在本例中是第一个），access是csv_File对象的成员函数。 
 
 但是，我注意到调用csv_Array.grab（0）.access（2,2）; 比它应该慢得多（它应该只是3个vector :: at calls）。 
 
 是否有某种隐藏的副本正在使这个非常慢？ 
 
 编辑：以下是一些相关的函数原型： 
 
    //Access in csv_File
    std::string access(int row, int column);

    //grab in csv_Array and csv_Analysis (mentioned below)
    csv_File grab(int index);
 
 附加详细信息（如有必要）：此代码的应用是将一堆csv文件加载到内存中，然后将其传递给将对数据进行一些计算的朋友类。 原理上，我有以下内容： 
 
 1）csv_Array具有私有成员向量存储; 2）csv_Analysis是一个csv_Array的朋友的类3）csv_Analysis访问矢量存储，它位于csv_Array中4）这种访问是通过将csv_Analysis传递给csv_Array中的存储（因此没有希望复制....）来完成的，例如public：csv_Analysis（csv_Array＆csv_block）; 
 
 因此，上面给出的调用[file_array.grab（0）.access（2,2）; ]实际上还有一个额外的“级别”，更像是 
 
    csv_Analysis analysis_Object(file_array); 
    analysis_Object.grab(0).access(2,2); 
 
 其中grab以相同的方式起作用，并且也被定义为csv_Analysis类的成员函数。 

I am experiencing unexpected performance degradation when using vectors with classes.
 
I have a csv_File class that reads in a csv file and stores the contents in a 2D vector. There's a member function that allows access, e.g.
 
    csv_File file("file.csv");
    file.access(2,2);
 
To access the 2,2 element.
 
Then, I have another class csv_Array that stores multiple csv_File objects in a vector, e.g private member vector There's a member function that allows access, i.e. it returns a csv_File object, for example:
 
    csv_Array file_array(5); //store 5 csv_File objects
    file_array.grab(0).access(2,2); 
 
In the second line, grab returns a csv_File object (in this case, the first one) and access is a member function of the csv_File object.
 
However, I have noticed that the call csv_Array.grab(0).access(2,2); is much slower than it should be (it should be just 3 vector::at calls).
 
Is there some sort of hidden copy going on that is making this very slow?
 
EDIT: Here are some of the relevant function prototypes:
 
    //Access in csv_File
    std::string access(int row, int column);

    //grab in csv_Array and csv_Analysis (mentioned below)
    csv_File grab(int index);
 
ADDITIONAL DETAILS (if necessary): The application of this code is to load a bunch of csv files into memory before passing it to a friend class that will do some calculations with the data. Schematically, I have the following:
 
1) csv_Array has private member vector storage; 2) csv_Analysis is a class that is a friend of csv_Array 3) csv_Analysis accesses vector storage, which is in csv_Array 4) This access is done by passing csv_Analysis a reference to storage in csv_Array (so no copy hopefully....), e.g public: csv_Analysis(csv_Array &csv_block);
 
Thus, the call given above[file_array.grab(0).access(2,2); ] actually has one additional class "level" in between and is more like
 
    csv_Analysis analysis_Object(file_array); 
    analysis_Object.grab(0).access(2,2); 
 
where grab acts in the same way, and is also defined as a member function of csv_Analysis class.

原文：https://stackoverflow.com/questions/6514438