WIP: Use call_once for dynamic loading ObjectFactoryBase::Initialize()

[N-Dekker]

Removed ObjectFactoryBasePrivate::m_Initialized. Moved RegisterInternal()
call from ObjectFactoryBase::Initialize() to the default-constructor of
ObjectFactoryBasePrivate.

---

Also cherry-picked the commit from pull request #4225, STYLE: Remove private ObjectFactoryBase member InitializeFactoryList(). (This PR is not meant to supersede PR #4225)

[N-Dekker]

Please note: this is really still in an experimental state. I still need to figure out exactly how things work! (The other PR, #4225 is ready though, independent of the discussion on atomic vs mutex vs call_once)

[issakomi]

A little off topic. Interesting what OpenGL factory is mentioned here :)

/**
 * Register any factories that are always present in ITK like
 * the OpenGL factory, currently this is not done.
 */
void
ObjectFactoryBase::RegisterInternal()

It looks like the class was a long time ago copied from VTK, s. comment in vtkObjectFactory.cxx.

[N-Dekker]

A little off topic. Interesting what OpenGL factory is mentioned here :)

/**
 * Register any factories that are always present in ITK like
 * the OpenGL factory, currently this is not done.
 */
void
ObjectFactoryBase::RegisterInternal()

Thanks for your input, @issakomi I'm now proposing to just remove RegisterInternal(), including the apparently VTK specific comment about "the OpenGL factory". Please check:

• STYLE: Remove private ObjectFactoryBase member RegisterInternal() #4228

[N-Dekker]

I'm afraid I was mistaken when I was thinking that call_once might be useful for ObjectFactoryBase::Initialize(). The member function claims to be a "one time initialization method":

ITK/Modules/Core/Common/src/itkObjectFactoryBase.cxx

Lines 223 to 227 in ecd7bda

	/**
	* A one time initialization method.
	*/
	void
	ObjectFactoryBase::Initialize()

But in fact, it may be called more than once, on the very same ObjectFactoryBase instance, by calling ReHash:

ITK/Modules/Core/Common/src/itkObjectFactoryBase.cxx

Lines 457 to 462 in ecd7bda

	void
	ObjectFactoryBase::ReHash()
	{
	ObjectFactoryBase::UnRegisterAllFactories();
	ObjectFactoryBase::Initialize();
	}

So I guess we may need to reconsider using a mutex (PR #4226?)... at least if we think we have serious thread-safety problem here.

[N-Dekker]

Abandoned: this approach (using call_once) does not work, because of ReHash as I mentioned before at #4227 (comment)