Overhauling Parameter Passing Conventions for Optimization
In a recent talk, Herb Sutter suggested that passing std::vector and std::string objects by const & is becoming obsolete. This claim has sparked discussions among programmers.
Sutter's Rationale
Sutter argues that in cases where the function returns an rvalue (i.e., a temporary object), passing the input parameter by value allows for more efficient handling using move semantics. This avoids unnecessary copies or memory allocations.
Consider the following code snippet:
std::string do_something ( std::string inval ) { std::string return_val; // ... do stuff ... return return_val; }
Passing inval by value allows the function to directly modify and return the value, eliminating the need for a copy constructor.
Concerns about Efficiency
Critics argue that passing large objects like std::string by value can still be inefficient due to the overhead of moving the object. A std::string instance consists of several components, including a heap pointer and a member char[] for short string optimization.
Arguments for Reference Parameters
Despite the potential inefficiencies, proponents of reference parameters argue that they offer several benefits:
Conclusion
The choice of whether to pass objects by const & or by value depends on the specific use case. In situations where efficiency is critical and the function operates on rvalues, passing by value with move semantics can provide performance benefits. However, in cases where encapsulation, simplicity, or ownership management is important, reference parameters remain valuable.
The above is the detailed content of Should We Still Pass `std::string` and `std::vector` by `const &`?. For more information, please follow other related articles on the PHP Chinese website!