Jtdcftul

can anybody explaine the memory layout of

std::vector<std::array<int, 5>> vec(2)

does it provide contiguous memory block of a 2D array
with 2 rows of 5 elements?

To my understanding, the vector of vectors

std::vector<std::vector<int>> vec(2, std::vector<int>(5))

provide the memory layout of two contiguous arrays of length 5 elements in different locations in memory.

Will it be the same for the vector of arrays?

Arrays do not have any indirection, but just store their data "directly". That is, a std::array<int, 5> literally contains five ints in a row, flat. And, like vectors, they do not put padding between their elements, so they're "internally contiguous".

However, the std::array object itself may be larger than the set of its elements! It is permitted to have trailing "stuff" like padding. So, although likely, it is not necessarily true that your data will all be contiguous in the first case.

An int

+----+

|    |

+----+



A vector of 2 x int

+----+----+----+-----+        +----+----+

| housekeeping | ptr |        | 1  |  2 |

+----+----+----+-----+        +----+----+

                   |          ^

                   -----------



An std::array<int, 5>

+----+----+----+----+----+----------->

| 1  |  2 |  3 |  4 |  5 | possible cruft/padding....

+----+----+----+----+----+----------->



A vector of 2 x std::array<int, 5>

+----+----+----+-----+        +----+----+----+----+----+----------------------------+----+----+----+----+----+----------->

| housekeeping | ptr |        | 1  |  2 |  3 |  4 |  5 | possible cruft/padding.... | 1  |  2 |  3 |  4 |  5 | possible cruft/padding....

+----+----+----+-----+        +----+----+----+----+----+----------------------------+----+----+----+----+----+----------->

                   |          ^

                   -----------

And, even if it were, due to aliasing rules, whether you'd be able to use a single int* to navigate all 10 numbers would potentially be a different matter!

All in all, a vector of ten ints would be clearer, completely packed, and possibly safer to use.

In the case of a vector of vectors, a vector is really just a pointer plus some housekeeping, hence the indirection (as you say).

The big difference between std::vector and std::array is that std::vector contains a pointer to the memory it wraps, while std::array contains the actual array in itself.

That means a vector of vectors is like a jagged array.

For a vector of arrays, the std::array objects will be placed contiguously but separate from the vector object. Note that the std::array object themselves may be larger than the array they contain, and if so then the data will not be contiguous.

The last bit also means that an array (plain C-style or std::array) of std::array may also not keep the data contiguously. The std::array objects in the array will be contiguous, but not the data.

The only way to guarantee contiguous data for a "multi-dimensional" array is nested plain C-style arrays.

The C++ standard does not guarantee that std::array doesn't contain any payload at the end of the array, so alas you cannot assume that the first element of a subsequent array is just after the last element of a prior array.

Even if that were the case, the behaviour on attempting to reach any element in a array by pointer arithmetic on a pointer to an element in a different array is undefined. This is because pointer arithmetic is only valid within arrays.

The above also applies to a std::array<std::array>.