Re: ref counting - Scott Meyers more effective C++
On 27.03.14 07.26, Christopher Pisz wrote:
I got handed a project that uses ref counting from Scott Meyers "More
Effective C++." It's old. I can't use the latest standard or boost,
currently.
I wonder if there is anyone around that is intimate with those chapters
of his book. There is some confusion about where the real data goes and
if it is or is not a pointer.
The example of String and StringValue is confusing, because char * is
the data. So when trying to use the example, one wonders if the
mentality is StringValue is the data with ref counting as part of the
class or if the char * it contains is the data.
I don't know the book but I have implemented ref counting with char*
(and others) as data type.
Lets say I had a class Animal for example.
Do I make:
Animal - with its data and methods
AnimalRCO : public RefCountedObject - containing a pointer to an animal
AnimalRCP - Containing the template RCO instance as a member
or
AnimalRCO : public RefCountedObject - containing all the data an Animal
would
AnimalRCP - having all the methods an Animal would
I am thinking the latter, but some colleagues are thinking the former.
Both implementations are possible.
The first one is the way boost:shared_ptr works like.
The second one is concept behind boost::intrusive_ptr.
If you look about performance the additional of the first approach might
be counterproductive. Especially if you have many small Animal objects.
On the other side the intrusive concept requires changes to the memory
layout of Animal. This might be not an option if you cannot change the
type or the factory for your Animals. And it might be overhead if not
all Animal objects need their own reference counter. Think of Arrays.
or perhaps even
AnimalRCO : public RefCountedObject - containing all the data _and_
methods an Animal would
AnimalRCP - having methods that call methods on the RCO for the Animal
interface
This is indeed a third option. It makes AnimalRCP behave more like an
object rather than a pointer. But I would not recommend this for public
types AnimalRCO since it might prevent you to distinguish between
reference equality and value equality and causes other confusion.
However, it might be an option if you want to hide the reference
counting entirely from the class user. E.g. a reference counted string
class that provides value semantics from outside and copy on write
internally.
-------------------
Also, the book doesn't seem to cover assignment operator as thoroughly
as I would like
What if an Animal or AnimalRCO has a constructor that takes an int
(perhaps for number of legs)
When I say AnimalRFP = 0; what am I really saying?
This should be a syntax error unless you have gone the 3rd way.
In the other cases it should involve operator new.
Marcel