A: Almost always [use a vector instead of an array]. Vectors are efficient and flexible. They do require a little more memory than arrays, but this tradeoff is almost always worth the benefits.
That’s an over-simplification. It’s fairly common to use arrays, and can be attractive when:
-
the elements are specified at compile time, e.g.
const char project[] = "Super Server";,const Colours colours[] = { Green, Yellow };- with C++11 it will be equally concise to initialise
std::vectors with values
- with C++11 it will be equally concise to initialise
-
the number of elements is inherently fixed, e.g.
const char* const bool_to_str[] = { "false", "true" };,Piece chess_board[8][8]; -
first-use performance is critical: with arrays of constants the compiler can often write a memory snapshot of the fully pre-initialised objects into the executable image, which is then page-faulted directly into place ready for use, so it’s typically much faster that run-time heap allocation (
new[]) followed by serialised construction of objects-
compiler-generated tables of
constdata can always be safely read by multiple threads, whereas data constructed at run-time must complete construction before other code triggered by constructors for non-function-localstaticvariables attempts to use that data: you end up needing some manner of Singleton (possibly threadsafe which will be even slower) -
In C++03,
vectors created with an initial size would construct one prototypical element object then copy construct each data member. That meant that even for types where construction was deliberately left as a no-operation, there was still a cost to copy the data elements – replicating their whatever-garbage-was-left-in-memory values. Clearly an array of uninitialised elements is faster.
-
-
One of the powerful features of C++ is that often you can write a
class(orstruct) that exactly models the memory layout required by a specific protocol, then aim a class-pointer at the memory you need to work with to conveniently interpret or assign values. For better or worse, many such protocols often embed small fixed sized arrays. -
There’s a decades-old hack for putting an array of 1 element (or even 0 if your compiler allows it as an extension) at the end of a struct/class, aiming a pointer to the struct type at some larger data area, and accessing array elements off the end of the struct based on prior knowledge of the memory availability and content (if reading before writing) – see What’s the need of array with zero elements?
-
classes/structures containing arrays can still be POD types
-
arrays facilitate access in shared memory from multiple processes (by default
vector‘s internal pointers to the actual dynamically allocated data won’t be in shared memory or meaningful across processes, and it was famously difficult to force C++03vectors to use shared memory like this even when specifying a custom allocator template parameter). -
embedding arrays can localise memory access requirement, improving cache hits and therefore performance
That said, if it’s not an active pain to use a vector (in code concision, readability or performance) then you’re better off doing so: they’ve size(), checked random access via at(), iterators, resizing (which often becomes necessary as an application “matures”) etc.. It’s also often easier to change from vector to some other Standard container should there be a need, and safer/easier to apply Standard algorithms (x.end() is better than x + sizeof x / sizeof x[0] any day).
UPDATE: C++11 introduced a std::array<>, which avoids some of the costs of vectors – internally using a fixed-sized array to avoid an extra heap allocation/deallocation – while offering some of the benefits and API features: http://en.cppreference.com/w/cpp/container/array.