Deterministic destruction across scope and state machines in c#

[Nate]

Excuse me if this has already been answered. I've done many searches
on newsgroups, websites, and books, and have not found a method to do
what I want.

Does there exist an idom or pattern in c# that will destroy, finalise,
or call some specified routine at a specified code point automatically
that will in-turn destroy or finalise all member objects of the class
automatically at the same code point?

By "automatically", I mean that I do not want to have to write code
that calls each member object's destruction method, as that is not a
safe programming idiom (an engineer may add more member variables down
the line and forget to add calls to their finalisation method).

As an example, in c++ with RAII, this is automated by the compiler
generated destructor mechanism.

In particular, the usage I want is similar to a c++ smart-pointer in
that I do not want to have to make an explicit call to the
finalisation method of the containing object, but instead intend to
just "switch out" objects (so a new one replaces the old one -
initiating the destruction).

This is used heavily in many classic designs in c++, particularly the
state pattern. An event to a state machine may cause one state to
transition to another state, which eventually becomes code like:

currentState_ = newState; // destroy old state and start using new

Ref objects only act like smart pointers in the reference counting,
not the deterministic destruction. The "using" keyword appears to
only be a single-scope mechanism, so does not appear suitable for
event-based lifetimes. The IDisposable methodology appears to require
explicit calls to dispose the object, to implement the IDisposable
interface, and recursively must dispose of all members explicitly.

Some popular c# state libraries either do not appear to allow states
to contain objects or do not manage the lifetime of those objects
automatically. I have not found anything that accomplishes this level
of automation yet, but I think I must be overlooking something simple.

 

[Nate]

Excuse me if this has already been answered.  I've done many searches
on newsgroups, websites, and books, and have not found a method to do
what I want.

Does there exist an idom or pattern in c# that will destroy, finalise,
or call some specified routine at a specified code point automatically
that will in-turn destroy or finalise all member objects of the class
automatically at the same code point?

By "automatically", I mean that I do not want to have to write code
that calls each member object's destruction method, as that is not a
safe programming idiom (an engineer may add more member variables down
the line and forget to add calls to their finalisation method).

As an example, in c++ with RAII, this is automated by the compiler
generated destructor mechanism.

In particular, the usage I want is similar to a c++ smart-pointer in
that I do not want to have to make an explicit call to the
finalisation method of the containing object, but instead intend to
just "switch out" objects (so a new one replaces the old one -
initiating the destruction).

This is used heavily in many classic designs in c++, particularly the
state pattern.  An event to a state machine may cause one state to
transition to another state, which eventually becomes code like:

currentState_ = newState;  // destroy old state and start using new

Ref objects only act like smart pointers in the reference counting,
not the deterministic destruction.  The "using" keyword appears to
only be a single-scope mechanism, so does not appear suitable for
event-based lifetimes.  The IDisposable methodology appears to require
explicit calls to dispose the object, to implement the IDisposable
interface, and recursively must dispose of all members explicitly.

Some popular c# state libraries either do not appear to allow states
to contain objects or do not manage the lifetime of those objects
automatically.  I have not found anything that accomplishes this level
of automation yet, but I think I must be overlooking something simple.

And just to be clear about my reasons here, I'm not just looking to
destroy objects to clean up memory or because it is what I'm used to
from c++. I wanted to make that point because I have seen a few
discussions on this topic that mentioned one shouldn't think in those
terms with a language like c# that has garbage collection, and that's
not why I'm asking.

In the state pattern, the whole point is that the state holds not only
methods to handle events while the system is in that state but also
actors in the system who exist while in that state. For instance, one
wants:

class AlertState : BaseState
{
NotificationBox alert_;
// rest of member objects and methods
};

where alert_ may be a notification message added to the model of the
UI while in the alert state. This should be removed when the
AlertState is replaced with another object. Good programming design
should prevent the need to explicitly remove the box on exit (because
of the N state x M object complexity of such code, or because of the
duplicated lifetime specification being error prone and possibly
divergent, etc.).

Obviously, I don't need the objects dtor to be the method called, nor
any known finalisation routines. The ability to call any method
automatically would be sufficient here if I architect it early enough.

 

[Nate]

On 5/31/11 1:26 PM, Nate wrote: [...]

By "automatically", I mean that I do not want to have to write code
that calls each member object's destruction method, as that is not a
safe programming idiom (an engineer may add more member variables down
the line and forget to add calls to their finalisation method).

Most member fields should not need any treatment whatsoever. The only
cleanup your code should worry about is when dealing with resources that
.NET itself cannot be aware of, either directly or indirectly (by
holding a reference to an object that implements IDisposable).

I tried to make this clearer in my discussion of actors. Making
states IDisposable does not automate the lifetime management of their
members. Lifetime management is important not just for memory, but
system participation in general. Deterministic finalisation should
automate the finalisation of actors at the lifetime of the state. Not
doing so is bad design, as it leaves open the possibility of an
engineer forgetting to do the finalisation action explicitly. This is
the same common design argument for all duplication of tasks - whether
code duplication (where an engineer may updte one piece of code but
not it's duplicate), serialisation (engineers may add a member but
forget to serialise it), or managing type information as switch-on-
enums (updating switch statements may miss a switch for a new enum) -
in general you should reduce the specification an engineer must update
to add functionality and prevent duplication of specification to
prevent divergence. In the examples, the common design solutions are
to use the same code (separate into a function), use introspective
serialisation, and use encapsulation into objects.

It's probably wrong to think of finalisation as needing to occur for
"resources" because most people think only OS resources like files,
windows, graphics surfaces, etc. Any actors that require finalisation
as a part of lifetime management must be deterministically finalised.
Some examples of actors that don't involve resources include:

- regulatory logs of security activity (events of opening and closing
doors on secure devices, for instance - the finalisation event being
the logging of the closing door or leaving the open state)
- notification data while in some state - addition and removal in the
UI model (not to be confused with the UI graphic resources, but the
actual data in the model when using proper MVC / MVP tripartite
presentation designs)
- communication with a host system on state notification (an "exiting
state X" communication found in many protocols)

The whole point of using a state pattern is that these things should
become automated, to reduce the O(N events x M states) complexity of
event handling to just those events that expect to be handled and to
reduce the O(M^2 state-to-state) transition complexity to O(M state)
construction/finalisation management.

Only for stack-allocated objects. In C#, one pretty much never
allocates reference-type objects on the stack.


You can't overload the assignment operator in C#, so that won't work
(note that in C++ the above is not the same as RAII; the "dispose on
assignment" works for a different reason, namely the implementation of
the overloaded assignment operator that goes with your smart-pointer class).

The use of the term RAII in architectural circles is that there should
never exist "zombie" objects, ie. objects which are not alive but not
quite dead. Two phase initialisation, multiphase destruction, etc.
all fall under the use of this term as espoused by Bjarne Stroustrup
(who coined the term). The point is that objects should be usable for
their lifetime (or that accessibility as objects should = their
usability lifetime) to prevent errors where engineers accidentally use
a zombie, and that in a language that allows exceptions, management of
zombies can be combinatorial in complexity.

So, my use here is fairly standard in those circles. I understand it
may not be understood the same way in c# circles, so I hope this
clarifies.

Properly encapsulated, it is trivial to add the necessary call to
IDisposable.Dispose() when updating the current state reference (e.g. by
using a single property or other method to handle the update, and
calling Dispose() there).

Again note that this has nothing to do with C++-style RAII, as the
objects are not allocated on the stack, nor does destruction happen as a
result of exiting a stack frame.

The whole point of my question is how to avoid the explicit calls to
all member actors in a state that would be placed inside the
Dispose(bool) if I used IDisposable. The IDisposable pattern does not
automate member finalisation - it must be done explicitly, which is
error prone and bad design. I'm looking for other patterns that
people may have in c# that solve this design issue (which is most
certainly about RAII - RAII has nothing to do with stack-v-heap).

There's no reference counting in .NET, except where you've implemented
it explicitly for some reason.

I didn't say anything of the sort. I was saying that management of
reference lifetimes ends when all references are no longer held, like
smart pointers managed by reference counts. The fact that the GC
walks reference graphs instead of using ref counting is immaterial to
the point - they act like smart pointers in one regard, but don't
provide the functionality (deterministic finalisation) that I need.
So I'm looking for something else.

Yet, it is .NET's equivalent to RAII, which is what you said you were
asking about (but of course as I've noted above, you've strayed from
that line of thought, so who really knows what you're asking).

It's a shame that many c# users have such a limited view of RAII,
because that totally misses Bjarne's original point.

I believe you are overlooking the usefulness of the IDisposable pattern.

Unfortunately, you have not actually posted any concrete code example of
a scenario showing what you'd like to do and why IDisposable or some
other approach doesn't help.

-------------- Program.cs ---------------------

namespace TestStateMachine
{
class Program
{
static void Main(string[] args)
{
// Make a state machine (default state 1) and fire event 1
StateMachine.Instance.Event1();

// fire event 2
StateMachine.Instance.Event2();

// fire event 3
StateMachine.Instance.Event3();

System.Threading.Thread.Sleep(10000);
}
}
}


--------- StateMachine.cs -------------

namespace TestStateMachine
{
public sealed class StateMachine
{
private static readonly StateMachine instance_ = new
StateMachine();

private StateBase currentState_ = new State1();

private StateMachine()
{ }

public static StateMachine Instance
{
get
{
return instance_;
}
}

public void Event1()
{
currentState_.Event1();
}

public void Event2()
{
currentState_.Event2();
}

public void Event3()
{
currentState_.Event3();
}

public void Transition(StateBase newState)
{
currentState_.Dispose();

currentState_ = newState;
}
};
}

------------ StateBase.cs ------------

using System;

namespace TestStateMachine
{
public interface StateBase : IDisposable
{
void Event1();
void Event2();
void Event3();
}
}

------------ State1.cs -----------------

using System;

namespace TestStateMachine
{
public class State1 : StateBase
{
TestManageThisClass someObject = new TestManageThisClass();

public State1()
{
Console.WriteLine("Entered State1");
}

~State1()
{
Console.WriteLine("Leaving State1");

Dispose(false);
}

public void Event1()
{
if (disposed_) throw new
ObjectDisposedException("State1");

StateMachine.Instance.Transition(new State2());
}

public void Event2()
{
if (disposed_) throw new
ObjectDisposedException("State1");

StateMachine.Instance.Transition(new State3());
}

public void Event3()
{
if (disposed_) throw new
ObjectDisposedException("State1");

StateMachine.Instance.Transition(new State4());
}

private bool disposed_ = false;

protected virtual void Dispose(bool disposing)
{
Console.WriteLine("State1 in Dispose");

if (!disposed_)
{
if (disposing)
{
// do nothing for managed to illustrate that
members are not deterministic
}

Console.WriteLine("State1 now disposed");

disposed_ = true;
}
}

public void Dispose()
{
Console.WriteLine("State1 in outer Dispose");

Dispose(true);
GC.SuppressFinalize(this);
}
}
}

-------------------- TestManageThisClass.cs -----------------

using System;

namespace TestStateMachine
{
public class TestManageThisClass
{
public TestManageThisClass()
{
Console.WriteLine("TestManageThisClass ctor");
}

~TestManageThisClass()
{
Console.WriteLine("TestManageThisClass dtor");
}
}
}


// then similar for states 2 - 4 (left out for brevity)
--------------------------------------------------

When I run my program, the output is:

TestManageThisClass ctor
Entered State1
Entered State2
State1 in outer Dispose
State1 in Dispose
State1 now disposed
Entered State4
State2 in outer Dispose
State2 in Dispose
State2 now disposed
Entered State3
State4 in outer Dispose
State4 in Dispose
State4 now disposed
....

Notice that the contained object TestManageThisClass of State1 is not
automatically and deterministically finalised. Instead, I would have
to explicitly write in the Dispose(bool) method where I've left a
comment an explicit call to finalisation, likely by deriving the
TestManageThisClass type from IDisposable and callng Dispose(). As
I've tried to point out, this is widely acknowledged to be a bad
design and error prone, because it requires that an engineer duplicate
the specification of member actors in the state in the member list and
in the Dispose(bool) method, introducing the possibility of
divergence.

I mentioned the patterns I am aware of (using, IDisposable) for
deterministic finalisation. None of these solve the problem for
automating state based lifetime management. I was hoping someone
might offer another means of solving this that is widely used, as I
hope there are some c# developers out there that use good
architectural practices and have solved this. Possibly using T4? Or
reflection?

There are 4 lifetimes that are commonly used in architecting programs:

temporary - commonly anonymous objects that live for a statement or
between sequence points
scope - imperative objects used to perform procedural calculations
within a function or more explicit scope
state - actors that operate for the lifetime of a specific program
state - ie. between asynchronous events
application - objects that exist for the lifetime of the application

It is clear that Microsoft eventually realised the necessity for
deterministic finalisation of scope objects, because they created a
number of statements in c# to manage these (lock, using, ...).
Temporary objects have a number of useful idioms in other languages
that take advantage of deterministic finalisation (like expression
templates and expression proxies), but I don't need those for my
project right now, so I'm not asking about those yet (they are also
more advanced techniques, so I think I'd be less likely to get
answers). Application objects already have proper lifetime
management.

It is the scope lifetimes that I would hope I could fnd some well
known idioms to use, as they should have been developed with any
project that uses good architectural practices.

Is this any clearer?

 

[Nate]

It's clear that my question is being taken as an attack against c#. I
understand that there must be sensitivities in play here, but I am not
attacking anything. I am asking a serious question that I think is
not being understood because of these sensitivities. I will try again
to be clearer.

[...]
The whole point of my question is how to avoid the explicit calls to
all member actors in a state that would be placed inside the
Dispose(bool) if I used IDisposable.  The IDisposable pattern does not
automate member finalisation - it must be done explicitly, which is
error prone and bad design.

Baloney.  It's no more error-prone than C++ objects allocated on the
heap.  In fact, because C# offers a language-supported way to control
the effective lifetime of an IDisposable object in a code block-scoped
way, it's less error-prone than the equivalent approach in C++.

Calling it "error prone" and "bad design" simply reveals your own
anti-C# bias.  Whether that's because you've got an axe to grind, or if
you've simply not had enough experience with C# to appreciate the
different idioms in the language that work just as well as those found
elsewhere, I don't know.  But the bias is clear.

There is absolutely no bias. In fact, I am asking for an idiom to
solve a design issue. Let me show the c++ side of what I would do:

class State1 : public StateBase
{
SomeActorType actor1_;
SomeActorType actor2_;

public:
// Event handlers interface, etc.
};

When I enter State1, actor1_ and actor2_ would be created and their
ctor called. When I leave State1, actor1_ and actor2_ would have
their dtor called. All of this is automated by the language.

In c#, if I were to use the IDisposable pattern, I require

public class State1 : StateBase
{
private SomeActorType actor1_;
private SomeActorType actor2_;

protected virtual void Dispose(bool disposing)
{
// if ! disposed and disposing (just abbreviating)
actor1_.Dispose();
actor2_.Dispose();

// rest of pattern
}
};

My point is that with the IDisposable pattern, you have to repeat the
actors that you've already specified in the member list also in the
Dispose(bool) function. That repitition is error-prone, because
someone may add a member and forget to put it in the Dispose(bool).

Anyone who has worked in a large project knows that specifying the
intent more than once is a major source of errors. In fact, c# uses
the GC to handle the common error in c of not matching all "malloc"s
with "free". c# uses the "lock" statement to handle the common error
in c of not matching "mutex_lock" calls with "mutex_unlock". And c#
has the "using" statement to match many of the common scope-based
duplication like file open/close, socket open/close, logging begin/
end, ... (although Microsoft has pointed out that since the Dispose
method may throw and hide exceptions in the using block, it is often
best not to use "using" but revert to the try/catch/finally triad).

The trend in architecture, since the start of the generative
programming days, is to reduce the specification of intent to best
match the efficiencies found in the domain language. One of the
simplest reductions is to remove duplicated specification. That is
all I am trying to do. All I am asking is if there is a known idiom
or pattern that people use to do this for _state_ lifetimes. You
don't employ "using" for state lifetimes, because the objects don't
live inside a scope. The objects live inside a state object because
there are asynchronous events that determine the start and end of the
lifetime.

I am fairy positive that this can be done with reflection (I am almost
finished with a solution that uses this and will be testing this
afternoon), and as with all generative programming, I am certain it
could also be done with code generation like T4. I know there are
ways to solve this. I am simply asking if there is a "standard" or
"well-known" pattern that has been used regularly. It is always best
to use common solutions, because they are regularly less complex
because they have had more eyes on them and more time to be refined.

I expected this was something that any good architects moving to c#
would have faced many times in the past and come up with something. I
didn't expect to touch on silly sensitivities and misunderstandings of
the intent of RAII and start some newsgroup war, but I guess those
kinds of things are hard to avoid on the newsgroups...

 

[Nate]

It's clear that my question is being taken as an attack against c#.  I
understand that there must be sensitivities in play here, but I am not
attacking anything.  I am asking a serious question that I think is
not being understood because of these sensitivities.  I will try again
to be clearer.

[...]
The whole point of my question is how to avoid the explicit calls to
all member actors in a state that would be placed inside the
Dispose(bool) if I used IDisposable.  The IDisposable pattern does not
automate member finalisation - it must be done explicitly, which is
error prone and bad design.

Baloney.  It's no more error-prone than C++ objects allocated on the
heap.  In fact, because C# offers a language-supported way to control
the effective lifetime of an IDisposable object in a code block-scoped
way, it's less error-prone than the equivalent approach in C++.

Calling it "error prone" and "bad design" simply reveals your own
anti-C# bias.  Whether that's because you've got an axe to grind, or if
you've simply not had enough experience with C# to appreciate the
different idioms in the language that work just as well as those found
elsewhere, I don't know.  But the bias is clear.

There is absolutely no bias.  In fact, I am asking for an idiom to
solve a design issue.  Let me show the c++ side of what I would do:

class State1 : public StateBase
{
  SomeActorType actor1_;
  SomeActorType actor2_;

public:
  // Event handlers interface, etc.

};

When I enter State1, actor1_ and actor2_ would be created and their
ctor called.  When I leave State1, actor1_ and actor2_ would have
their dtor called.  All of this is automated by the language.

In c#, if I were to use the IDisposable pattern, I require

public class State1 : StateBase
{
  private SomeActorType actor1_;
  private SomeActorType actor2_;

  protected virtual void Dispose(bool disposing)
  {
    // if ! disposed and disposing (just abbreviating)
    actor1_.Dispose();
    actor2_.Dispose();

    // rest of pattern
  }

};

My point is that with the IDisposable pattern, you have to repeat the
actors that you've already specified in the member list also in the
Dispose(bool) function.  That repitition is error-prone, because
someone may add a member and forget to put it in the Dispose(bool).

Anyone who has worked in a large project knows that specifying the
intent more than once is a major source of errors.  In fact, c# uses
the GC to handle the common error in c of not matching all "malloc"s
with "free".  c# uses the "lock" statement to handle the common error
in c of not matching "mutex_lock" calls with "mutex_unlock".  And c#
has the "using" statement to match many of the common scope-based
duplication like file open/close, socket open/close, logging begin/
end, ... (although Microsoft has pointed out that since the Dispose
method may throw and hide exceptions in the using block, it is often
best not to use "using" but revert to the try/catch/finally triad).

The trend in architecture, since the start of the generative
programming days, is to reduce the specification of intent to best
match the efficiencies found in the domain language.  One of the
simplest reductions is to remove duplicated specification.  That is
all I am trying to do.  All I am asking is if there is a known idiom
or pattern that people use to do this for _state_ lifetimes.  You
don't employ "using" for state lifetimes, because the objects don't
live inside a scope.  The objects live inside a state object because
there are asynchronous events that determine the start and end of the
lifetime.

I am fairy positive that this can be done with reflection (I am almost
finished with a solution that uses this and will be testing this
afternoon), and as with all generative programming, I am certain it
could also be done with code generation like T4.  I know there are
ways to solve this.  I am simply asking if there is a "standard" or
"well-known" pattern that has been used regularly.  It is always best
to use common solutions, because they are regularly less complex
because they have had more eyes on them and more time to be refined.

I expected this was something that any good architects moving to c#
would have faced many times in the past and come up with something.  I
didn't expect to touch on silly sensitivities and misunderstandings of
the intent of RAII and start some newsgroup war, but I guess those
kinds of things are hard to avoid on the newsgroups...

I finished up a preliminary implementation of a reflection solution to
demonstrate that that method worked. It's basic procedure for an
external IDisposable looks like:


// use reflection to get the objects in holder
FieldInfo[] mInfos = holder.GetType().GetFields();

// loop through members
foreach (FieldInfo mInfo in mInfos)
{
// first print what info I can get
Console.WriteLine("Found member: " + mInfo.Name + " of type " +
mInfo.FieldType);

.// now try to call Dispose on each member
if (typeof(IDisposable).IsAssignableFrom(mInfo.FieldType))
{
Console.WriteLine("This is derived from IDisposable!");

// get the value object
object value = mInfo.GetValue(holder);
Console.WriteLine("Got value object of type " +
value.GetType().Name);

// so invoke
mInfo.FieldType.GetMethod("Dispose", new
Type[0]).Invoke(value, new object[0]);
}
}


Now, instead of just using this at a single level working on an
external object (the holder type above), I've taken a recursive
approach and created a type RecursiveDisposable derived from
IDisposable which has a method RecursivelyDispose that implements the
same technique on the self type. This allows deterministic
finalisation of all RecursiveDisposable types. Then, my state machine
calls RecursivelyDispose on the state that is exiting and all actors
will be removed from the system properly and without duplication of
specification during the state transition.

My biggest concern is the time involved in invoking methods by
string. So I will pursue a T4 solution next to weigh options here.
Are there any other approaches foreseeable?

 

[Nate]

Of course there is.  You've stated several times that C# is deficient in
this area, when in fact it's not.  It simply has a different idiom than
you're used to.

Your false criticisms of the language can occur only either because
you've started with a false premise that you have some compulsion to
prove, or simply out of ignorance.  Either way, it's a bias.

Dude, I really have no idea what your issues are, why you like to
attack people online instead of answer their questions, who hurt you,
whatever. I don't really want to know. This will be the last time I
respond to you because I have actual work to do and family to be with.

But it is strange how you've convinced yourself that I believed there
was no answer to my question when 1) I mentioned reflection and T4
early in the discussion, and 2) I ended up answering my own question
because you were too busy masturbating about how great the idiom that
didn't answer my question really was. If I had wanted to start an
argument that simply bashes a language, I would have gone over to the c
++ groups and pointed out how even the newest updates to the language
fail to allow serialisation without duplication (but c# reflection
solves it fine). Or I would have come over here and pointed out that
without RAII, c# makes you have to write finalisation code everywhere
a class requiring disposal is used instead of how in c++, the
exception mechanism calls dtors and automates cleanup, making c++ a
much safer language in the face of exceptions. Or I'd go to the OCaml
groups and point out that since they allow imperative state, they
aren't a true functional language. Every language has something true
about it that hurts people and mentioning it makes it users sore.

But this _obviously_ wasn't one, since I mentioned possible solutions
early on and eventually solved it myself. So how you've convinced
yourself I was here to pick a fight, when I was only mentioning
truthful statements about how the standard idioms failed to resolve my
particular issue, is completely unknown.

Only because you have embedded the types within the container type.  If
you were actually using true reference types throughout, you'd have the
same issue as in C#.

It is a really strange complaint that I am using the state pattern as
one is supposed to. Sure, if I do things wrong, I'll get bad
results. Why does this line of thought even come to your mind?

If you follow the ComponentModel pattern found in .NET, then disposable
objects are added to a list of components.  They still need to be added,
but the Dispose() method requires no changes if disposable objects are
added.

Sure, I can change compiletime contracts to runtime contracts and lose
even greater compiler error detection. That's the wrong direction to
go based on nearly every book on safe architecture I've ever read,
where the big goal in domain language mapping is to enforce more
compiler contracts through compiletime type information. But sure, if
I like to do things poorly, I totally could.

More to the point, you have lots of opportunities besides this to
"forget" to do something important.  Fact is, disposable objects are
much less common than plain vanilla ones that don't require special
handling, and for skilled programmers, it's not a problem at all to
remember to dispose one's disposable objects.

Adding a new field to a class involves a laundry list of things that
have to be done in order to incorporate that object into the class.
Unless you're a particularly careless person, you are much more likely
to screw up the main implementation details than to forget to add the
object to the IDisposable implementation (especially since most of the
time, it's just going to be automatic as soon as you add the field to
also add the dispose call).

Of course. Which is why assembly is so popular. I mean, we don't
need procedural names since any skilled programmer can choose the
right relocatable address to jump to as needed, or the proper data
layout that they intend to manipulate.

Seriously, you are arguing that my concern is unwarranted when
Microsoft took the effort and money to add a GC and "using" and "lock"
statements to the language and suggest that using them is preferred
practice, when all design in c++ has held that as best practice, when
there is a mathematical theorem in computational grammar correctness
theory that proves such a reduction reduces the code points where
contradicting intent may occur...

You can code in unsafe mode c# all you want, remember all the silly
things that c# could be taking care of for you, and prove to yourself
what a manly programmer you are. I've coded in Fortran and c earlier
in my career, so I know I can do it. I don't need to prove to myself
anything on my company's time. Instead, I want to provide a design to
my engineers that makes it so they don't have unnecessary
possibilities of error. They can screw up the intent through logical
errors and there is nothing I can do to prevent that - that's always
the case. I see no need in adding places where they can mess up by
failing to duplicate data.

[...] (although Microsoft has pointed out that since the Dispose
method may throw and hide exceptions in the using block, it is often
best not to use "using" but revert to the try/catch/finally triad).

More misunderstanding on your part.  I have no idea what you read that
gave you that idea, but whatever it was, you didn't understand it correctly.

A Dispose() method that catches exceptions will do so whether you use a
"using" statement or a "try/finally".  The only reason to use
"try/catch/finally" instead of "using" is if you have other stuff that
needs to go into the "finally" anyway.

I love how you, who misunderstood my question in a number of hilarious
(and probably embarrassing) posts want desperately to claim that I am
the one misunderstanding things.

If you look at

http://msdn.microsoft.com/en-us/library/aa355056.aspx

you will see how they discuss the fact that when a Dispose may throw,
like in their client sample for WCF (or many of the other explicit
resource IDisposables in .NET and many user-defined classes requiring
finalisation that adds something to an internal collection like a log,
among other examples), then the using construct may throw upon
closing. This means that exceptions thrown inside the using statement
block may be hidden by the outer finally-originated throw. When the
finaliser may throw, there is the need for Abort-like constructs to
ensure resources are still proper cleaned and no leaks occur.

This is pretty well known by most who have any WCF or related
experience. But I can understand that if you are not informed about
such things, it might sound like it is other people who are
misinformed. It's called the Dunning-Kruger effect, and it's quite
common.

[...]   I
didn't expect to touch on silly sensitivities and misunderstandings of
the intent of RAII and start some newsgroup war, but I guess those
kinds of things are hard to avoid on the newsgroups...

If you don't want to touch on "silly sensititivites" (ad hominem much?),
then you shouldn't go around make ignorant claims about deficiencies of
the language you are asking about.

Since I assumed there was a solution, eventually solved it myself
while you were still stumbling with the question, and have had, in the
process, to bear a series of posts from you labeling me ignorant
(where I never once until this post reciprocated the accusation -
despite you being the one who showed on numerous responses that you
were the one lacking common knowledge of terms commonly used in the
architectural community), I think "silly sensitivities" is a
reasonable description of your behavior. Project much?

Oh, and for your own education, down the road and all, if you ever get
into another these discussions, you might want to learn how to google
search before claiming terms used are made up or not common in front
of all other newsgroup visitors that might read your silliness.

For instance,

http://www.google.com/search?client=safari&rls=en&q=zombie+object&ie=UTF-8&oe=UTF-8

Peace and love, Pete. I hope you get some help. As I mentioned
above, this is my last response to you.

 

[ozbear]

Dude, I really have no idea what your issues are, why you like to
attack people online instead of answer their questions, who hurt you,
whatever. I don't really want to know. This will be the last time I
respond to you because I have actual work to do and family to be with.

<snip>
Nate, your reaction to Pete is understandable. Although occaionally
knowledgeable on some aspects of C# and .net, he uses his knowledge as
a club to belittle and berate people who ask questions. You have come
to the same conclusion as many of us that it's just not worth getting
an answer from such an individual.

Oz

 

[James A. Fortune]

Nate, your reaction to Pete is understandable.  Although occaionally
knowledgeable on some aspects of C# and .net, he uses his knowledge as
a club to belittle and berate people who ask questions.  You have come
to the same conclusion as many of us that it's just not worth getting
an answer from such an individual.

As with all posters, our words eventually stand or fail on their own
merits. I will play your game to this extent: Peter Duniho has not
come across to me the way that you are suggesting in providing any
answers to me. I am determined enough that I will eventually find the
answers on my own even without Peter Duniho being available, so he is
not an exclusive repository of knowledge; but his insights save me
considerable time in coming to certain conclusions and philosophies
about C#. I would eventually like to reach or surpass his expertise
in C#. If so, I also hope to match or exceed his level of
professionalism when answering questions about C#, and to keep in mind
the amount of help I received to get there. Those of us who post
answers based on our experiences hope that those experiences will make
everyone better. I try to be as clear as I can when asking or
answering questions. I try to share what I can with humility and
goodwill, not minding at all if what I share helps someone else
surpass me. Such helpful people should be forgiven any small conceits
that initially help spur us on. Besides, he's had a wit like you to
deal with. That accounts for half of his irritability . You are
actually very clever, but you're also very annoying. As far as
sympathy for being made to feel inferior by Peter, you can recruit
bleeding hearts elsewhere. IMO, you couldn't be made to feel inferior
if your life depended upon it, so don't act so wounded. You've done
that many times. Even if Peter actually made someone feel bad, it
wouldn't be as bad as your sorry act. You've actually helped my
understanding of C# by asking the questions you have - but, O, the
price!

James A. Fortune
(e-mail address removed)