std::vector of std::vectors contiguity

问题:

I know that std::vector<T> internally stores it's data contiguously (unless it is std::vector<bool>) both in the old C++03 standard and the new C++11.

Nice stackoverflow questions that deal with this and quote the standard: answer, answer.

What about the data inside nested vectors std::vector <std::vector <T> >? How is that stored?

If every internal vector needs to store it's data contiguously, how can it be true that &v[n] == &v[0] + n for all 0 <= n < v.size().

To phrase this slightly different, is it possible to access all the elements stored in such nested structure "simply" and sequentially (via a pointer or similar) the same way it can be done for a 1-D vector?

回答1:

No. The elements of a vector are stored in a dynamically allocated block of memory; otherwise, the capacity of the vector could not increase. The vector object just holds a pointer to that block.

The requirement that the elements be stored sequentially applies only to the elements themselves, and not to any dynamically allocated members of those elements.

回答2:

To answer your final question: No. The elements of a vector of vectors are not stored contiguously.

Consider the following code:

std::vector<std::vector<int> > vv;
.... fill in v[0], v[1], v[2], etc
std::vector <int> & v = vv[1];
v.push_back (23);

If they were all stored contiguously, then this would cause each element in vv[2], vv[3], etc to move. How would that possibly work, since you're just affecting a single vector 'v'?

回答3:

std::vector< std::vector<T> > is a vector of objects, that are stored in contiguous block of memory. The fact that these objects are vectors too is irrelevant though.

Although elements of the vector are stored in the contiguous block of memory, the memory where elements reside is not the part of the vector object itself.

"is it possible to access all the elements stored in such nested structure "simply" and sequentially (via a pointer or similar) the same way it can be done for a 1-D vector?"
For accessing elements of std::vector, it's better to use operator[] or at() method than retrieving the address of first element and using pointer arithmetic. For multidimensional arrays represented as vector of vectors, I suggest you to stay with operator[], which is easy to use and easy to read as well: myVector[i][j]. Worth to see vector::at vs. vector::operator[] as well :)

回答4:

is it possible to access all the elements stored in such nested structure "simply" and sequentially (via a pointer or similar) the same way it can be done for a 1-D vector?

Yes, if:

• you only ever need to add stuff to the end of your vector of vectors, and

• you're willing to replace the vector of vectors construct with a custom data structure

What you can do then is to concatenate all those sub vectors into a single contiguous buffer, with another index buffer used for accessing this by top level entry index.

See my article here for more discussion about this, and an example 'collapsed vector vector' class implementation..