|
semantics and documentation arguments aside, did your question get answered?
|
|
|
|
|
The second post in this thread answered the original question. Thanks, Mark.
|
|
|
|
|
oh right, wasn't paying close enough attention. And yes MS semantics are horribly confusing.
In that case, the C# ~Class1() is the finalizer. And by explicitly defining one you ensure that it is called by the garbage collector before the memory for an instance of the class is reclaimed. So although you're printing 'destructed' its actually the garbage collector calling the finalizer before memory is reclaimed. Its not a destructor that runs when the variable goes out of scope as you would get if it was a C++/CLI ref class with ~Class1() defined. And unlike destruction, finalization is nondeterministic and incurs a high performance cost for both allocation and deallocation. So you never want to write one unless your class contains unmanaged resources that won't be cleaned up by the garbage collector. In that case the finalizer will ensure the unmanaged stuff is cleaned up even if someone forgets to call delete or Dispose .
|
|
|
|
|
I don't want to argue, but you're not being specific enough. A C# destructor is more than syntactic sugar. The code within it will be called and then an implicit call is made to Finalize(). I'm not disputing or judging what actually occurs, I'm just pointing out that "C# destructors do not exist" is not a true statement. And I'm certainly not advocating that every C# class define a destructor. I'm glad I made my original mistake in assuming that C++/CLI ctor/dtor behavior is the same as C++ ctor/dtor behavior because it cleared things up after clouding them for awhile.
|
|
|
|
|
That's just it though, it is syntactic sugar. When you write ~Class1() in C# you've got a finalizer. It's not doing something then calling Finalize() I don't believe. Internally the compiler synthesizes that to a call to Dispose(false) which is invoked by the garbage collector right before the memory is reclaimed nondeterministically.
By contrast, a destructor in C++/CLI is synthesized to
Dispose()<br />
{<br />
Dispose(true);<br />
GC::SuppressFinalize();<br />
}
While a call to !Class1() in C++/CLI (the finalizer) results in a call to Dispose(false) .
The ~Class1() is a finalizer, not a destructor. To call it a destructor is a complete misnomer and doesn't convey when or how its called. A destructor would be called automatically when the object goes out of scope (via a using block in C# or via stack semantics in C++/CLI) or when invoked specifically by Dispose() . You must not reference other finalizable objects in that finalizer and other similar rules for finalizers that aren't present in a destructor. In C++/CLI a class destructor is created by the compiler automatically and will clean up managed resources for you. A finalizer (~Class1() in C#) is not generated by the compiler. It only exists if you explicitly write one. Most important is the fact that the ~Class1() in C# is nondeterminitic which by its nature makes it not a destructor (since they're deterministic) and the performance penalty for creating a finalizer for a class (In C++/CLI you would incur this cost with !Class1() but ~Class1() would enable actual RAII destructor semantics like native C++).
In C++/CLI you would write ~Class1() to properly handle the closing of resources in a way that can be done deterministically (when the object goes out of scope or a call to delete is made). This is a destructor. In C# when you write ~Class1() you're providing a way for the garbage collector to handle resource cleanup or unmanaged cleanup at the last second before memory reclamation in the case that the user failed to manually release it with Dispose or a using block. This is a finalizer. There is no destructor. You can't call delete. You have to write a Dispose method and/or use using blocks.
The C++/CLI dtor/ctor behavior is semantically the same as native C++. Although the destructor in C++/CLI doesn't cause the memory to actually be reclaimed (as that's always the garbage collectors job for managed objects) it is invoked deterministically (unlike the C# ~Class1() finalizer) which ensures resources will be closed immediately (unlike the C# finalizer).
Hopefully that wasn't too rambling...
|
|
|
|
|
From the source code perspective, I have code in a destructor that gets called. That's it. I'm not concerned with what goes on underneath. We're debating semantics. When you write ~Class1() in C#, you've got a destructor, both in name and behavior.
In native C++, one would expect a destructor to be called implicitly when an automatic variable goes out of scope. Such is not the case in C++/CLI. What's defined as a destructor in C++/CLI does not get called when an automatic variable goes out of scope. What's defined as a finalizer in C++/CLI gets called when an automatic variable goes out of scope. That, to me, is a big difference.
|
|
|
|
|
You seem to be confused.
You have to be concerned with what goes on underneath or you won't understand how this works. When you write ~Class1() in C# you have a finalizer in behavior. The syntax is like a C++ destructor and this is confusing on Microsoft's part. Assume you have a C# class with ~Class1() defined. Now, whenever you instantiate an object of Class1 you incur a performance penalty simply because the finalizer exists. The Class1 instance must register in a Finalizable queue and an freachable queue when the GC does its business. ~Class1() (really a call to Dispose(false) ) will be invoked by the garbage collector every time an object of Class1 is about to be reclaimed. The simple existence of ~Class1() and the subsequent placement in the freachable queue can cause objects of Class1 to be promoted to another generation and not be collected by the GC when it really could be. Most importantly, you have no idea when ~Class1() will be called by the garbage collector. Its completely nondeterministic. This by its very nature means its not a destructor. The finalizer is there to let you code last minute cleanup of unmanaged resources (which the GC cannot work with) in case the user of your class did not properly call Dispose() on an instance of Class1 or use a using block. And like I said before, there's a huge number of restrictions for what you can do in a finalizer vs a destructor.
Now assume we have Class1 in C++/CLI with ~Class1() defined. This will be called when an object of Class1 goes out of scope IF you declared that object with stack semantics (i.e., Class1 myClass1Object ). If you dynamically allocate memory with gcnew then you must use delete to invoke the destructor deterministically. This is the same as what happens in native C++. The difference in C++/CLI is that the memory itself won't be immediately reclaimed immediately (as that's the GC's job and it works nondeterministically) but that's not an issue. It will be reclaimed when more memory is required. The important part is that the destructor (~Class1() ) will always be invoked implicitly when the automatic variable goes out of scope. A C# class with ~Class1() defined does not function this way because in C# ~Class1() is a finalizer not a destructor. So in C++/CLI you put resource-release code in your destructor ~Class1() and then an automatic variable (such as a FileStream object) will have the destructor implicitly called and close the stream for you so there is no resource leak. In C# your finalizer ( ~Class1() ) will not be called implicitly when the object goes out of scope. It will be called sometime later by the GC nondeterministically. This can lead to errors when you think a resource has been released but it hasn't because the GC hasn't invoked the finalizer yet. Hence the need for using blocks in C# when C++/CLI can simply use stack semantics because of deterministic destruction via destructors (~Class1() ).
In C++/CLI a finalizer is declared with !Class1() as opposed to C# using ~Class1(). The same rules for finalizers is true in C++/CLI as C# (as outlined in my second paragrah).
|
|
|
|
|
I may be reading your comments wrong, but did you state a C++ ref class destructor is a finalizer?
That's not the case. The destructor maps to Dispose() and a finalizer must be explicitly added.
Destructors do not get called if you allocate an object on the managed heap and never explicitly
delete it. The finalizer, however will be called when the handle no longer has any references to it
and the GC gets around to cleaning it up. That's why native objects should be freed in a finalizer,
not a destructor (although the destructor should call the finalizer - the compiler takes care of
SuppressFinalize).
Having to explicitly delete objects would be counterintuitive, as Michael originally stated.
Being able to (deterministically) using the familiar C++ syntax, however, is a nice, powerful
feature.
Note I'm only referring to C++ - I'll make no comments about C# except for the related Dispose/Finalize
pattern of CLR.
Cheers,
Mark
Mark Salsbery
Microsoft MVP - Visual C++
|
|
|
|
|
I'm pretty sure we've said the same thing. However...
You say destructors are not called on an object allocated on the managed heap which isn't strictly true. If you allocate it on the managed heap with gcnew then yes the destructor is never called if you don't manually call delete . However, if you instantiate the ref class object with stack semantics then the destructor is called implicitly when the object goes out of scope (even though underneath its still allocated on the managed heap like all ref class objects). This is the critical addition to C++/CLI (vs MC++) that allowed for RAII style coding like native C++. Michael was claiming this is not the case (as it is in native C++) and that the finalizer is what is called, which is not correct. The finalizer will be called eventually either way if you define one, but the way things work is obviously much different than what a destructor provides.
The confusion is that in C# when you declare ~Class1() this is actually mapped to the finalizer (Dispose(false) ) in the dispose pattern. C# has no concept of destructors like C++ or C++/CLI even though the semantics there look like you're declaring a destructor. In C++/CLI you accomplish the same thing by declaring !Class1() which more properly distinguishes the finalizer from the destructor. Then declaring ~Class1() in C++/CLI does a very different thing than ~Class1() in C#. That's why in C# you need to explicitly call Dispose() in a finally block or use a using block with things like FileWriter objects to ensure there is no resource leak. C++/CLI's destructor semantics provide the handy native C++ style, FileWriter fw; which will automatically be disposed of properly when it goes out of scope (deterministically).
|
|
|
|
|
iddqd515 wrote: in C# when you declare ~Class1() this is actually mapped to the finalizer (Dispose(false)) in the dispose pattern.
Dispose() isn't the finalizer...Finalize() is, right?
Regardless, C++ does the same - the destructor becomes Dispose() (quite literally,
as Michael's original warning message stated).
In C++ the Destructor is explicitly called when a stack-based syntax managed object
goes out of scope, the destructor is called directly, a managed heap object
is explicitly deleted with delete, or another language calls Dispose() on an object.
If the class implements a finalizer then that needs to be (at least SHOULD be) called from
the destructor. The compiler takes care of calling SuppressFinalize so the finalizer will not
be called again by the GC.
Isn't that the same as the C# (and probably VB) Dispose/Finalize pattern?
Mark
Mark Salsbery
Microsoft MVP - Visual C++
|
|
|
|
|
Are you being purposefully daft? I understand what's going on; I'm arguing that "C# does not have destructors" is an untrue statement. I'm not repudiating the mechanics.
Dispute this: The C# language specification allows for the definition of destructors for class types.
|
|
|
|
|
From: http://en.wikipedia.org/wiki/Destructor_%28computer_science%29
In object-oriented programming, a destructor (sometimes shortened to dtor) is a method which is automatically invoked when the object is destroyed. Its main purpose is to clean up and to free the resources which were acquired by the object along its life cycle and unlink it from other objects or resources invalidating any references in the process. The use of destructors is key to the concept of RAII.
How in C# do you declare such a thing? A finalizer is not automatically invoked when the object is destroyed deterministically. It is nondeterministically called by the GC at some point in the future. Most important, in no way does a finalizer enable the RAII pattern.
Take a FileStream class which encapsulates a resource in C#. Declare a finalizer for it. When an instance of a FileStream class goes out of the scope the finalizer is not called then. You do not know when it is called. Furthermore, even if the GC does a collect and there are no roots in the code to the object, the simple existence of the finalizer counts as a root and can cause the FileStream object to be promoted into the next generation and not collected by the GC until the next collect cycle (who knows when that will be). So while your FileStream object may have gone out of scope ages ago, by not calling Dispose and assuming the finalizer will close the FileStream , you have potentially caused a resource leak. The next function call that assumes the file you previously had used with the FileStream object is closed may throw an exception because you were relying on the garbage collector to nondeterministically release a critical resource.
On the other hand, a destructor in C++/CLI will be called the moment the FileStream goes out of scope (assuming you declared the object with stack semantics and not gcnew ). Then in the destructor you make sure there is a call to Dispose and you ensure that the resource is released immediately. You don't need a try/finally and you don't need to explicitly call Dispose and nor does anyone who uses your class in the future. The destructor incurs no performance penalty during allocation or deallocation like a finalizer does either. There's no way to declare such a function in C#.
To say 'I'm not repudiating the mechanics' therefore C# has destructors is completely wrong. The mechanics are exactly what separate a finalizer and a destructor. You can't just say 'oh it has a ~Class() declaration therefore its a destructor even if it works completely differently.' They work in fundamentally different ways. Start declaring finalizers for all your C# classes as if it was a C++ destructor and watch the programs performance and reliability plummet.
|
|
|
|
|
iddqd515 wrote: How in C# do you declare such a thing?
~Class()
|
|
|
|
|
Well like I said then, make sure to declare ~Class() in all your C# classes and don't bother explicitly invoking Dispose . I'm sure you'll find it works fantastically. Also make sure to take advantage of finalizers to declare reference classes in C# with stack semantics.
-- modified at 15:30 Monday 13th August, 2007
|
|
|
|
|
No need to get pissy. As I said earlier, I'm not advocating the use of what the language itself calls (or, apparently, used to call) a destructor; I'm just claiming that such a beast exists.
From the C# Language Specification:
[Note: In the previous version of this standard, what is now referred to as a "finalizer" was called a "destructor". Experience has shown that the term "destructor" caused confusion and often resulted to incorrect expectations, especially to programmers knowing C++. In C++, a destructor is called in a determinate manner, whereas, in C#, a finalizer is not. To get determinate behavior from C#, one should use Dispose. end note]
|
|
|
|
|
So that right there says 'sorry we incorrectly used to refer to this as a destructor. We realized it was confusing because its really not a destructor at all. To get destructor like behavior in C# use the Dispose method.' Therefore, C# has no destructors. And even in the past, it only had them in name, not function. But its been awhile since anyone referred to them as destructors since its blatantly confusing, especially with the introduction of C++/CLI in '05.
|
|
|
|
|
iddqd515 wrote: On the other hand, a destructor in C++/CLI will be called the moment the FileStream goes
out of scope (assuming you declared the object with stack semantics and not gcnew).
Then in the destructor you make sure there is a call to Dispose
That may be where my confusion with your statements is coming from.
That sounds like managed extensions. In VS 2005, you can't call Dispose() - the compiler
won't let you. This is now done implicitly with the destructos symantics.
Mark
Mark Salsbery
Microsoft MVP - Visual C++
|
|
|
|
|
Sorry, I was being ambiguous. All this Dispose talk and switching between C# and C++/CLI gets me mixed up occasionally. You're correct, I meant in the destructor a call to Dispose(true) is made. In any case, the important part (C++/CLI destructors are deterministic) is the same, while a C# finalizer is completely different due to its nondeterministic nature. Luckily I've never had to use MC++. *shudder*
-- modified at 15:49 Monday 13th August, 2007
|
|
|
|
|
Cool. I'm just making sure for my own clarification.
I recently ported code from VS2003 and while I was ready for many of the changes, the
Dispose stuff was a surprise. I want to make sure I've got it clear in my head since
I only used the documentation as a learnin' resource
Cheers,
Mark
Mark Salsbery
Microsoft MVP - Visual C++
|
|
|
|
|
Yeah its taken me over a month to really grasp the whole pattern and how it meshes with traditional C++ semantics. Its not helpful when you try to find stuff about the Dispose pattern and end up with C# examples mixing terms to create more confusion. On the plus side, now C++/CLI basically works just like you would expect native C++ to in most cases and you only need to worry about finalizers for classes with unmanaged members. Certainly an improvement over MC++ (as most things are).
|
|
|
|
|
Yes indeed, there is a destructor-like syntax, but it is actually called a finalizer;
it is a bit confusing if you ask me.
And as you said it has undeterministic behavior: It gets executed by the gc
when it decides to collect the object.
Luc Pattyn [Forum Guidelines] [My Articles]
this weeks tips:
- make Visual display line numbers: Tools/Options/TextEditor/...
- show exceptions with ToString() to see all information
- before you ask a question here, search CodeProject, then Google
|
|
|
|
|
Michael Chapman wrote: That just seems totally counterintuitive to using managed code.
Yes, but so is mixing native types and having destructors.
The garbage collector will only cleanup managed objects if you don't specifically call a destructor.
Since you don't need deterministic cleanup of your native object, the smart pointer is a good solution,
since it provides the destructor/finalizer for you.
Only if you needed deterministic cleanup of your native object would you need to implement a destructor
and possibly a finalizer as shown in the article (although you do it implicitely with the smart pointer).
Mark
Mark Salsbery
Microsoft MVP - Visual C++
|
|
|
|
|
|
Interesting! Thanks for the link George!
Mark
Mark Salsbery
Microsoft MVP - Visual C++
|
|
|
|
|
hmm...my Sunday morning reply made no sense - I'll take another crack at it.
What I meant to say was...
If you have unmanaged resources in a managed class then you'll need a way to
free them, as always.
You could just cleanup your native resources in the destructor BUT the destructor
won't be called unless you explicitly call it (using delete or calling it directly).
Having to do THAT would be counterintuitive
To ensure your native object created with new gets deleted, you should implement
a destructor AND a finalizer in its managed owner class - that's what Nish's smart pointer
class does for you.
That way you don't have to explicitly destruct your managed object. If it is left
for the GC to cleanup, the GC will call the finalizer (NOT the destructor!)...
ref class MyRefClass
{
int *NativeIntArray;
public:
MyRefClass() {NativeIntArray = new int[400];}
~MyRefClass() {this->!MyRefClass();}
!MyRefClass() {delete[] NativeIntArray;}
} Cheers,
Mark
Mark Salsbery
Microsoft MVP - Visual C++
|
|
|
|