Question

This is a purely theoretical question, I know that the standard containers interface are not likely to change now…

I read recently one of Herb Sutter “guru of the week” where he complained about the fact that empty() was implemented as a member function in std::string. I did not agree with all the arguments, because e.g. std::list would require a different implementation of the same function, as size() is O(n) and empty() is obviously O(1).

However, is there a reason why the standard specify that the empty() member function of (for instance) std::string is implemented as "size () == 0" instead of "begin () == end ()" ?

It seems to me that the latter allow the same O(1) implementation of the empty() function for all containers, and I can’t think of any drawback. Is it less efficient?

Thanks,

N.G.

Answer 1

The standard doesn’t specify exactly how anything in the library is implemented. It merely specifies the meaning of something. So if empty() is true, then size() must also be 0. That doesn’t mean that empty() must actually call std::basic_string::size() and compare it to 0. The spec is simply saying that if empty() returns true, then calling size() immediately after will return 0, and if empty() returned false, then calling size() immediately after will not return 0.

The spec could have said that its begin() would equal its end(), and it would force no implementations to change.

The inconsistency is most likely a consequence of the std::basic_string class having come from a different place from the rest of the STL containers during the development of C++98. That’s why it has so many member functions for doing things that STL containers would typically do with algorithms.