I have two questions:
1) Is it possible to implement an allocator that uses alloca to allocate memory on the stack and is otherwise C++ STL compliant?
If there is code out there, you can make me happy by simply pointing me to the URL. :-) If there is no code out there, perhaps you can sketch the functions allocate and deallocate?
2) If the answer to the above question is 'yes', I'd like to understand how it is possible to allocate memory on the stack for class members. As an example, consider an
std::vector<int, AllocaAllocator<int> >
and suppose that a call of the member function 'resize' of this vector calls first 'deallocate' and then 'allocate' of the allocator.
The scope from which allocate is called is that of the member function resize. Doesn't this mean that the allocated memory is removed from the stack at the end of that function call?
Kind regards, Bjoern
No, this kind of thing isn't possible. For a start, the STL expects to allocate more memory, then free the old memory. How are you going to do that on the stack?
The only thing even remotely equivalent to this is a conservative garbage collector.
Bjoern, it looks like you fundamentally misunderstand how stack and alloca work. Read about them.
What you are asking is impossible because the memory allocated by alloca is "freed" when you return from the function that allocated it (and unlike Patrick said, inlining cannot change its behavior). I write "freed" because it's not actually freed, it just goes out of scope as any other stack variable. So using it afterwards causes undefined behavior.
Suppose you allocate a chunk of memory in
YourAllocator::allocate
which is called fromd.push_back()
:The memory allocated by alloca may be overwritten by the stack-frames of
push_back
andprintf
, so the output may not be 42, it may crash, or any other thing.