|
Hi all, hope you could maybe give me some advice on the following...
Is the following code legal?
MyClass foo()
{
MyClass * p = new MyClass();
// do something to *p here
return *p;
}
void test()
{
foo();
}
void main()
{
test();
}
The above code compiles and works fine. The intention is to create a MyClass instance in foo, and return it without onvoking the constructor on exit. I then want this instance to be available in test. On exit of test i want this instance to be automatically deleted.
It seems to work fine, but when I use the same mechanism in large programs, including classes etc., I am running into some difficulties when the "return *p" executes.
Thanks,
Paul

|
|
|
|
|
In your
void test()
{
foo();
}
You did not handle "p". This is a memory leak. You should delete p;
void test()
{
Myclass* p = foo();
do something with "p" then
delete p;
}
|
|
|
|
|
Thanks for the reply.
I've debugged the presented code step by step and witnessed that the compiler deletes p itself on exit from test() !
However, the presented code is not OOP. When the same mechanism is used in an OOP set-up SOMETIMES I get errors. Perhaps I'm missing something.
--------------------------------------------------------------
Anyway, is there a way to achieve what I want without having to delete p by hand?... I basically want to instantiate an object in foo and then return it to the calling function. I would then like the calling function to auto-delete that instance on exit.
Paul
|
|
|
|
|
paolosh wrote: I've debugged the presented code step by step and witnessed that the compiler deletes p itself on exit from test() !
It does not.
What happens is this:
new MyClass creates a new MyClass instance on the heap
return *p returns a COPY of that object
That copy is a temporary variable, and will be destroyed. However, the instance allocated on the heap will not be destroyed.
You may want to lear a bit more about C++ pointers, copy constructors and temporary values (it is not an easy topic).
When you understand these things, I would suggest my article about smart pointers[^] as an introduction to a common solution to your original question - "returning large objects from a function".
|
|
|
|
|
So what of the overloaded functions?
in standard C++ you can do something like this...
double x, y, z;
z = x + y;
does x.operator +(y) return a temporary double? which is then destroyed after it's contents is assigned to z??
|
|
|
|
|
Yes, exactly.
With a "double", there is no constructor or destructor, so there is not much to see.
But you can try these things with a class like this you can play around and see the effects:
class MyTest
{
public:
MyTest() { cout << "Creating MyTest " << this << endl; }
MyTest(MyTest const & rhs) { cout << "Copy-Construct MyTest " << this << " from " << &rhs << endl; }
MyTest& operator=(MyTest const & rhs) { cout << "Assign MyTest " << this << " from " << &rhs << endl; }
~MyTest() { cout << "Destroying " << this; }
}
Be aware that the effects are not exactly the same on every compiler, since the standard allows some optimizations.
|
|
|
|
|
I have a question for you.
I checked following:
p1 = new myclass();
p2 = foo()
p1 is exactly same as p2.
If foo() return a COPY of a POINTER, why the address of the COPY will be same as the original one?
Thanks
|
|
|
|
|
paolosh's original code is more akin to this:
MyClass * p1 = new MyClass;
MyClass temp = *p1;
|
|
|
|
|
When returning a "POINTER", it return POINTER itself or just a copy of the POINTER?
I debug the code. I found out the address of the "return" one is exactly same as the original one (new Myclass()).
Thanks,
|
|
|
|
|
Returning pointer:
MyClass * Get()
{
MyClass * p = new MyClass;
return p;
}
Returning copy, and leaking heap object:
MyClass Get()
{
MyClass * p = new MyClass;
return *p;
}
|
|
|
|
|
I think you're question is that you want the foo function to create the class and the test function to use the class.
Here is a snippet if this is what you need.
void foo(MyClass*& p)
{
p = new MyClass();
}
void test()
{
MyClass* p
foo(p);
delete p;
}
void main()
{
test();
}
«_Superman_»
I love work. It gives me something to do between weekends.
|
|
|
|
|
Right, I think it's because my copy-constructors have not been defined correctly !
Thanks for the help.
Paul
|
|
|
|
|
No, you're not o the right path.
You code leaks memory. Let me expaly why:
void test()
{ foo(); }
literally means this:
- a space wide enough to store a MyClass is leaved on the stack (That's what
Myclass foo() means to the compiler) - the actual code address is pushed on stack, than
- a jump to the beginning of
foo is executed. at thet point
- a space wide enough to store a pointer is leaved on stack. (That's what
MyClass* p requires)
- new MyClass() is executed, that means:
- a bunch of bytes enouh to store MyClass are taken from the heap
- The MyClass::Myclass() constructor is called
- The address of the allocated memory is copied to p
- now, we are at
return *p that means: - a copy of the value pointed by p (that's what
*p means) is copied into the space previously left on the stack by calling MyClass(const MyClass&). That's what return If you did not declare it that's not important, the compiler generates it implicitly.
Now there are two distinct copy of MyClass: one taken from the heap whose address is stored in p and one that is stored temporarily on the stack containing a copy of it.
After the return,
- the p pointer is destroyed by unrolling the stack. The value it store (the address of the heap containing the first Myclass) is lost forever, but the object is still there (hence the leak)
- a jump is executed to the address stored in the stack, thus causing the execution to return after the call.
Now we are back to test() .
- No actions are required on the return value, hence it is destroyed and the stack unrolled
Now you have destroyed the temporary copy. The original object is still there.
In general every time you call new, you've to guess where and when there is a corresponding delete that will be called.
if you want to prperly debug all this, yu have to explicitly declare all what the compiler will elsewhere do implicitly: that is
class MyClass
{
public:
MyClass() {}
Now set a breakpoint at all of those functions, and try to debug, and you'll see funny things.
2 bugs found.
> recompile ...
65534 bugs found.
|
|
|
|
|
paolosh wrote: The intention is to create a MyClass instance in foo, and return it without onvoking the constructor on exit.
instead of
return *p;
call
return p;
(of course, you need to declare foo to return a pointer)
and than in the caller do something like:
std::auto_ptr<MyClass> p(foo());
This way the object will be automatically destroyed on exit of your test, and you won't have a memory leak.
|
|
|
|
|
Hi all,
I'm trying to animate something using GDI+. At the moment, to prevent smearing when my picture moves on the screen, I clear the screen before redrawing each picture with drawimage(hdc).
Of course, that causes the screen to flicker.
Is there a way to stop that using GDI+? With GDI, I was able to stop the flickering by BITBLTing the pictures in a buffer HDC before BITBLTing in the window HDC. However, it doesnt look like BITBLT works with GDI+ (unless I'm doing something wrong).
I tried using drawimage to draw to a buffer HDC and then using BITBLT to copy the buffer HDC to the window HDC, but that doesn't work.
Does anyone know an alternative way? Or a way to draw in a buffer hdc and then copying that buffer to the window hdc using GDI+, not GDI.
|
|
|
|
|
The only way I know how (and the way I use) is to use GDI+ to draw on a memory HDC (in xp at least, GDI+ is not accelerated, so you're not losing much), then Blt that using classic GDI to the main screen.
So...
void CMyView::OnDraw (CDC *dc)
{
CMemDC MemDC (dc);
Graphics gfx (MemDC);
....
}
See Flicker Free Drawing In MFC[^] for the article I refer to. I use it in 83% of my drawing code.
Iain.
I have now moved to Sweden for love (awwww).
If you're in Scandinavia and want an MVP on the payroll (or happy with a remote worker), or need cotract work done, give me a job! http://cv.imcsoft.co.uk/[ ^]
|
|
|
|
|
I recommend doing all of your drawing into a memory Bitmap and then use DrawImage. Something like this:
void YourClass::OnPaint()
{
CPaintDC dc(this);
Graphics g(dc.m_hDC);
CRect r;
GetClientRect(r);
Bitmap memBitmap(r.Width(), r.Height());
Graphics memGraphics(&memBitmap);
g.DrawImage(&memBitmap, 0, 0);
}
How To Draw on a Memory Bitmap in GDI+[^]
I also highly recommend using the CachedBitmap Class[^]. Draw your background/pictures a single time and store them in a CachedBitmap object.
Best Wishes,
-David Delaune
|
|
|
|
|
I have a two very simple client server applications that work well. Client can communicate with Server and Server too can communicate with Client.
My goal is to access the applications of servers from client like TeamViewer. For that purpose, I have created a client/server application. Now what next I should do?
|
|
|
|
|
What is 'TeamViewer'? Could you elaborate on your requirements or technical C++ problems?
|
|
|
|
|
It sounds like you are attempting to create a remote desktop viewer which is not a trivial task. If so I suggest that you look at something like TightVNC.
|
|
|
|
|
YES. U are right. I have studied the VNC source code and could not grasp it completely. I would apprecaite if you guide me step by step.
eg , as I have already developed the client/server application. so what I should do next. How to capture the remote machine applications and then how to transmitt using client/server application.
Even I am not sure , either I need the client/server application or not?
I can guees that might be I capture the remote machine desktop by using printscreen function and then transmitt this image to viewer application (client) but then the question arises , how I may perform functions over an image, like mouse click to start/close the application.
THANKS for earlier help.
|
|
|
|
|
Madhu_Rani wrote: I have studied the VNC source code and could not grasp it completely
I would go back and study it a bit more. There is a wealth of information available in open source remote desktop apps and not just in the source code. Review the change logs and the SVN history of files responsible for the functionality that you are particularly interested in. Seeing the problems they have had to deal with can give you an idea of what you might encounter during the development process. You can even go back to very early versions which might be a bit primitive but will be easier to understand.
Madhu_Rani wrote: I would apprecaite if you guide me step by step.
At a very basic level most of what you are trying to accomplish is fairly straight forward. You might approach this by having your remote desktop server capture the desktop as an image, prepare the image data and then transmit it to the client. On the client you would then simply process the incoming image data and display it in a window. Without compression however you will be transmitting a significant amount of data for each screen update.
|
|
|
|
|
THANKS for the detailed reply.
I am trying to capture the desktop as an image and then I would send the captured image to the client and if delay is on higher side, then I would do compression by using some built-in-API..
I am little confused, how a transmitted image ( from server) would responds to mouse clicks. eg If I have a click over START button of transmitted image, how server would open the START menu ... as it is an image, so how clicks would trigger the corresponding events...
I can capture these events on client machine but how these events would be mapped to server sides...
THANKS.
|
|
|
|
|
Madhu_Rani wrote: If I have a click over START button of transmitted image, how server would open the START menu
Well this is pretty straight forward as well. If you click on the image representing the remote desktop you will receive a button message (e.g. WM_LBUTTONDOWN) which includes the mouse coordinates. The message and it's associated parameters (wParam/lParam) are simply transmitted to the remote desktop server which simulates the event. Since the mouse coordinates supplied are relative to the upper left hand corner of the window client area (on the remote client side) you shouldn't have to do any conversion. To simulate mouse and keyboard events on the server side you can make a call to SendInput().
|
|
|
|
|
This was my biggest fear and now I got the idea. Any how today I gained knowledge about DC and how to capture desktop and now I am integrating all these pieces of code with client server application. Then after integration I would do the final stuff that you mentioned in your reply. THANKS
|
|
|
|