C++Talk.NET

Joseph Harvell · Guest

I am trying to provide a base class to be used in a system where *most* of the memory allocation requires two parameters: a
memory handle and a memory type, with an options for pooled allocation and alignment. Accordingly, I defined the class
MyPlacementAllocator (class definition shown below).

The intended use of this class is that any non-abstract class that wants to use this allocation mechanism simply inherits
publicly from MyPlacementAllocator. This is working fine. But the problem I am having is that any class that has
MyPlacementAllocator as an ancestor cannot use any other signature for new.

I discovered this problem becuase I have a base class that is not abstract (MyExtendableConcreteClass). So I made it inherit
from MyPlacementAllocator so that anyone using an instance of this class can use the new operators I have provided *if they want
to*. Then I tried using the expression 'new MyExtendableConcreteClass()' and it failed because it could not find a matching
operator new.

Am I supposed to add 'inline static void *operator new(std::size_t size){return ::new MyExtendableConcreteClass();}' to the
class definition for MyExtendableConcreteClass? I don't think this is right because the latter new expression should result in
a call to the default constructor. I don't want to construct the object twice. I just want to be able to say 'Use ::operator
new(std::size_t) for any of the standard signature new expressions for MyExtendableConcreteClass and andy of its subclasses.'
Actually, if there is some syntax for this I would put it in MyPlacementAllocator instead: (e.g. 'inline static void *operator
new(std::size_t size){return ::new(size);}).

Another question. I used both 'inline' and 'static' for each operator new. I used static, becaue I think of new as being
static since there is clearly no object yet. However, this makes me wonder if it is overridable if I make it static. Does it
matter whether I say static or not? Also, I used 'inline' because the implementations of these are one line calls to other
functions. I figure that anyone who is newing up an object of a class must be able to see the definition of the class (and all
of its base classes). But I also want to make sure these can be overridden. It seems like the 'inline' would make that
impossible. But I think 'new' and 'delete' have special rules.

Another question: I haven't provided an equivalent signature for operator delete for each version of operator new. I can do
this I guess, but what happens if I don't? There is no syntax for me to actually invoke them. Everything compiles and runs so far.

class MyPlacementAllocator
{
public:
// marker type
struct PooledAllocation {};

// Non-pooled; throws exception
inline static
void *operator new(std::size_t size,
MemHandle memHdl,
int memType,
unsigned alignment=DEFAULT_ALIGNMENT) throw(std::bad_alloc);

// Pooled; throws exception
inline static
void *operator new(std::size_t size,
const PooledAllocation&,
MemHandle memHdl,
int memType,
unsigned alignment=DEFAULT_ALIGNMENT) throw(std::bad_alloc);

// Non-pooled, returns 0
inline static
void *operator new(std::size_t size,
const std::nothrow_t&,
MemHandle memHdl,
int memType,
unsigned alignment=DEFAULT_ALIGNMENT) throw();

// Pooled, returns 0
inline static
void *operator new(std::size_t size,
const PooledAllocation&,
const std::nothrow_t&,
MemHandle memHdl,
int memType,
unsigned alignment=DEFAULT_ALIGNMENT) throw();

inline static
void *operator new(std::size_t size);

inline static
void *operator new(std::size_t size,
const std::nothrow_t&);

// common delete for all
inline
void operator delete(void *mem,std::size_t size);

/*
virtual destructor is not necessary since delete does not use size parameter
*/
};

extern const MyPlacementAllocator::PooledAllocation Pooled;

[ See http://www.gotw.ca/resources/clcm.htm for info about ]
[ comp.lang.c++.moderated. First time posters: Do this! ]

Gianluca Silvestri · Guest

"Joseph Harvell" <jharvell (AT) dogpad (DOT) net> ha scritto nel messaggio
news:btIAe.540$c41.467 (AT) newssvr30 (DOT) news.prodigy.com...

Thomas Richter · Guest

Hi,