Accessing private data during finalize/dispose

[Guest]

I understand the basics of finalization and implementing IDisposable and how one is guaranteed access to managed objects only when working through the IDisposable interface.

My question is to what extent can I rely on access to my private data during the finalization process when *not* coming through the IDisposable interface? Are private members still valid or are they managed objects that may already be gone? Are some types accessible (int, char, double) but others not (string)?

Here is my dilemma... class A provides management for the creation and use of objects of class B, including marking these objects as "in use" (and unmarking them) out in the external world. Each B object instantiated keeps a reference to the A object used to create it. Class B implements the IDisposable interface and I want to ensure that the external "in use" flag for a B object is properly cleaned up no matter how the B object is being disposed of. Here is some pseudocode:

class A
{
public B CreateB(int ID)
{
if (MarkInUse(ID))
return new B(this, ID);
else
return null;
}

public bool MarkInUse(int ID) { ... }

public void MarkNotInUse(int ID) { ... }
}

class B : IDisposable
{
private A m_Manager;
private int m_ID;

public B(A manager, int ID)
{
m_Manager = manager;
m_ID = ID;
}

~B() { Dispose(false); }

public Dispose()
{
Dispose(true);
GC.SupressFinalize();
}

protected Dispose(bool disposing)
{
if (disposing)
// Clean up managed resources here

// Now clean up external resources
m_Manager.MarkNotInUse(m_ID); // ??? Is this valid ???
}
}

Can I still refer to 'm_Manager' here? I know I can do so reliably in the "if (disposing)" conditional -- but I need to unmark the object even if the calling code forgets to call Dispose on my object! I could duplicate the code necessary to unmark the object but need to keep two strings and an int to be able to do it. I'm reasonably sure that 'int' private data is accessible during finalization, but how about strings?

Thanks!
-- TB

 

[Stoitcho Goutsev \(100\) [C# MVP]]

Hi TB,

Are some types accessible (int, char, double) but others not (string)?

All data members of value types (int, char, double, enums, structs ) are
accessible, but reference types (strings, arrays, declared as classes) are
not guaranteed.

Can I still refer to 'm_Manager' here? I know I can do so reliably in the

"if (disposing)" conditional -- but I need to unmark the object even if the
calling code forgets to call Dispose on my object! I could duplicate the
code necessary to unmark the object but need to keep two strings and an int
to be able to do it. I'm reasonably sure that 'int' private data is
accessible during finalization, but how about strings?

Since class A doesn't declare *finalizer* that means you won't use any
unmanaged resources from A then "Yes, you can access m_Manager safely".
Why? Because class-B objects have finalizers. When they become garbage they
will be moved to the freachable queue, which is considered as a *root*
all objects referenced from class-B objects are still alive. Finalizers will
be executed by a worker thread and after that class-B objects will be
eligable for GC. How you can see class-A will be collected not before
finalization of the last class-B object it has created.

 

[Guest]

----- Stoitcho Goutsev (100) [C# MVP] wrote: -----

Hi TB,

Are some types accessible (int, char, double) but others not (string)?

All data members of value types (int, char, double, enums, structs ) are
accessible, but reference types (strings, arrays, declared as classes) are
not guaranteed.

Ok, this is what I gathered from reading the documentation. But this seems to contradict what you say next...

Can I still refer to 'm_Manager' here? I know I can do so reliably in the

"if (disposing)" conditional -- but I need to unmark the object even if the
calling code forgets to call Dispose on my object! I could duplicate the
code necessary to unmark the object but need to keep two strings and an int
to be able to do it. I'm reasonably sure that 'int' private data is
accessible during finalization, but how about strings?

Since class A doesn't declare *finalizer* that means you won't use any
unmanaged resources from A then "Yes, you can access m_Manager safely".
Why? Because class-B objects have finalizers. When they become garbage they
will be moved to the freachable queue, which is considered as a *root*
all objects referenced from class-B objects are still alive. Finalizers will
be executed by a worker thread and after that class-B objects will be
eligable for GC. How you can see class-A will be collected not before
finalization of the last class-B object it has created.

Let me see if I have this right...

Because my class B declares a finalizer (the ~B( ) "destructor"), it will be placed on the finalizer queue when garbage collected and kept "live". Because it is still "live" in this queue, any reference the class B instance holds is still valid -- even if these are to reference types (strings, arrays, classes). [doesn't this contradict the statement above??]

However, if class A also declared a finalizer (a ~A( ) "destructor"), then I would be SOL because both A and B instances would be placed in the finalizer queue and, although both are "live" while in the queue, their order of finalization is indeterminate and I can't guarantee that the reference to the A object inside of B is still valid when B's finalizer is eventually called.

So..., in this type of pattern, where a subordinate class object maintains a reference to a supervising object and needs access to it during finalization it is critical that the supervising class *not* implement a finalizer!!

Have I got it right??
Thanks!
-- TB

 

[Stoitcho Goutsev \(100\) [C# MVP]]

--
B\rgds
100

----- Stoitcho Goutsev (100) [C# MVP] wrote: -----

Hi TB,

Are some types accessible (int, char, double) but others not

(string)?
All data members of value types (int, char, double, enums, structs ) are
accessible, but reference types (strings, arrays, declared as classes) are
not guaranteed.

Ok, this is what I gathered from reading the documentation. But this

seems to contradict what you say next...

Yes, you are right. Sorry. That was my last post before the end of the day
My statement above is not correct. You have access to all members of the
class.

However, if class A also declared a finalizer (a ~A( ) "destructor"), then

I would be SOL because both A and B instances would be placed in the
finalizer queue and, although both are "live" while in the queue, their
order of finalization is indeterminate and I can't guarantee that the
reference to the A object inside of B is still valid when B's finalizer is
eventually called.

So..., in this type of pattern, where a subordinate class object maintains

a reference to a supervising object and needs access to it during
finalization it is critical that the supervising class *not* implement a
finalizer!!

Yes, you got it. However, If class A declares finalizer that doesn't mean
class-A object is not alive. It is still alive, but it might be already
finalized. which means that all unmanaged resources might be already
released. So you don't have to use anything which might be affected by
finalization. My point is that finalizer doesn't destroy the object in the
managed heap. That's why is somehow misleading to call it *destructor*. Bare
in mind that it is good practice to use finalizers to release unmanaged
resources only. I believe it is not good idea nulling references for
example. You don't gain anything with that.
In your case: Yes it is safe to unregister class-B objects when the objects
are disposed by the GC.

B\rgds
100

 

[Guest]

----- Stoitcho Goutsev (100) [C# MVP] wrote: ----
--
B\rgd
10

----- Stoitcho Goutsev (100) [C# MVP] wrote: ----
Hi TB
Are some types accessible (int, char, double) but others no

(string)
All data members of value types (int, char, double, enums, structs ar
accessible, but reference types (strings, arrays, declared a classes) ar
not guaranteed

Ok, this is what I gathered from reading the documentation. But thi

seems to contradict what you say next..

Yes, you are right. Sorry. That was my last post before the end of the da
My statement above is not correct. You have access to all members of th
class

No problem, thanks for clearing that up

However, if class A also declared a finalizer (a ~A( ) "destructor"), the

I would be SOL because both A and B instances would be placed in th
finalizer queue and, although both are "live" while in the queue, thei
order of finalization is indeterminate and I can't guarantee that th
reference to the A object inside of B is still valid when B's finalizer i
eventually calleda reference to a supervising object and needs access to it durin
finalization it is critical that the supervising class *not* implement
finalizer!

Yes, you got it. However, If class A declares finalizer that doesn't mea
class-A object is not alive. It is still alive, but it might be alread
finalized. which means that all unmanaged resources might be alread
released. So you don't have to use anything which might be affected b
finalization. My point is that finalizer doesn't destroy the object in th
managed heap. That's why is somehow misleading to call it *destructor*. Bar
in mind that it is good practice to use finalizers to release unmanage
resources only. I believe it is not good idea nulling references fo
example. You don't gain anything with that
In your case: Yes it is safe to unregister class-B objects when the object
are disposed by the GC

Excellent, thanks! Not only do I think that I understand this much more now, but I can do what I need to do in my application safely!!! :-

-- T

 

[Andreas Huber]

I understand the basics of finalization and implementing IDisposable and how one is guaranteed access to managed objects only when working through the IDisposable interface.

My question is to what extent can I rely on access to my private data during the finalization process when *not* coming through the IDisposable interface? Are private members still valid or are they managed objects that may already be gone? Are some types accessible (int, char, double) but others not (string)?
Here is my dilemma... class A provides management for the creation

and use of objects of class B, including marking these objects as "in
use" (and unmarking them) out in the external world. Each B object
instantiated keeps a reference to the A object used to create it.
Class B implements the IDisposable interface and I want to ensure that
the external "in use" flag for a B object is properly cleaned up no
matter how the B object is being disposed of. Here is some
pseudocode:

class A
{
public B CreateB(int ID)
{
if (MarkInUse(ID))
return new B(this, ID);
else
return null;
}

public bool MarkInUse(int ID) { ... }

public void MarkNotInUse(int ID) { ... }
}

class B : IDisposable
{
private A m_Manager;
private int m_ID;

public B(A manager, int ID)
{
m_Manager = manager;
m_ID = ID;
}

~B() { Dispose(false); }

public Dispose()
{
Dispose(true);
GC.SupressFinalize();
}

protected Dispose(bool disposing)
{
if (disposing)
// Clean up managed resources here

// Now clean up external resources
m_Manager.MarkNotInUse(m_ID); // ??? Is this valid ???
}
}

Can I still refer to 'm_Manager' here? I know I can do so reliably

in the "if (disposing)" conditional -- but I need to unmark the object
even if the calling code forgets to call Dispose on my object! I
could duplicate the code necessary to unmark the object but need to
keep two strings and an int to be able to do it. I'm reasonably sure
that 'int' private data is accessible during finalization, but how
about strings?

You can safely access other objects from a finalizer, as long they are
in no way affected by another finalizer. This is because no longer
accessible objects are finalized in completely arbitrary order (which
is in stark contrast to C++ where the order of destruction is
defined).

Therefore, your example works as long as class A does not have a
finalizer which modifies members that are directly or indirectly
accessed from ~B() (through MarkNotInUse). Moreover, none of A's
fields (or any of their fields) must have a finalizer. Since primitive
types (int, float, ... ) and strings do not define finalizers,
everything should work.

It wouldn't work if A contained e.g. a FileStream field that is
accessed from MarkNotInUse(), because FileStream defines a finalizer.

HTH,

Andreas

 

[Guest]

----- Andreas Huber wrote: ----

[snip]

You can safely access other objects from a finalizer, as long they ar
in no way affected by another finalizer. This is because no longe
accessible objects are finalized in completely arbitrary order (whic
is in stark contrast to C++ where the order of destruction i
defined)

Therefore, your example works as long as class A does not have
finalizer which modifies members that are directly or indirectl
accessed from ~B() (through MarkNotInUse). Moreover, none of A'
fields (or any of their fields) must have a finalizer. Since primitiv
types (int, float, ... ) and strings do not define finalizers
everything should work

It wouldn't work if A contained e.g. a FileStream field that i
accessed from MarkNotInUse(), because FileStream defines a finalizer

HTH

Andrea

Yes, thanks Andreas (and Stoitcho)... It is intuitively satisfying, and most importantly..., it allows me do do what I need to do!! :-

-- T