C / C++ / MFC

Terminating window-less process?

22-Oct-02 13:19

If you're using MFC, then all you need to do is create an instance of your dialog box, and call it's 'DoModal' function.

Christian

Hey, at least Logo had, at it's inception, a mechanical turtle. VB has always lacked even that... - Shog9 04-09-2002

During last 10 years, with invention of VB and similar programming environments, every ill-educated moron became able to develop software. - Alex E. - 12-Sept-2002

Abin22-Oct-02 12:29

Abin

22-Oct-02 12:29

I've searched quite a bit on this forum for the ways to terminate a process, in most case people obtain a HWND by calling ::FindWindow then get the process ID through the HWND, and kill the process.

Now my problem is that my program does not have a window, and suppose I do not know its file name either at this point, I do, however, have its HINSTANCE which was returned by the original ::ShellExecute call(that's how I launched this process). If in some ugly case this application gone frozen, is there a good way to terminate it?

Re: Terminating window-less process?

Hans Ruck22-Oct-02 23:41

Hans Ruck

22-Oct-02 23:41

Bin wrote:
I do, however, have its HINSTANCE

Are you sure? MSDN describes that HINSTANCE as nothing but a return value to be casted for an integer.
Unless you're using a verb, you may try CreateProcess instead of ShellExecute.

rechi

overloaded operator new and STL vector

Nemanja Trifunovic22-Oct-02 12:30

22-Oct-02 12:30

When I try to compile this piece of code

#include <cstddef>
#include <vector>

class klasa
{
char data[100];
public:

void* operator new (size_t)
{
return ::new klasa;
}

void operator delete (void* pt)
{
if (pt == 0) return;

::delete(pt);
}
};

void main()
{
std::vector <klasa> kontejner;
klasa k;
kontejner.push_back(k);
}

I got compiler error:
d:\program files\microsoft visual studio\vc98\include\xmemory(34) : error C2660: 'new' : function does not take 2 parameters
d:\program files\microsoft visual studio\vc98\include\xmemory(66) : see reference to function template instantiation 'void __cdecl std::_Construct(class klasa *,const class klasa &)' being compiled

WTF is going on here?

Beer | [beer]

Christian Graus22-Oct-02 13:34

22-Oct-02 13:34

My *guess* would be that there is an operator new that takes two parameters and that it is involved in the creation of a vector.

BTW, you don't need to do this:

Nemanja Trifunovic wrote:
if (pt == 0) return;

delete (0) is guarenteed to be safe.

Christian

Hey, at least Logo had, at it's inception, a mechanical turtle. VB has always lacked even that... - Shog9 04-09-2002

During last 10 years, with invention of VB and similar programming environments, every ill-educated moron became able to develop software. - Alex E. - 12-Sept-2002

Nemanja Trifunovic22-Oct-02 15:58

22-Oct-02 15:58

Christian Graus wrote:
My *guess* would be that there is an operator new that takes two parameters and that it is involved in the creation of a vector.

True. Operator new with two arguments is required in an allocator function. The question still is what can I do about it? What to do with this second parameter?
Christian Graus wrote:

delete (0) is guarenteed to be safe.

Yeah, I know. My original new and delete are more complicated: not just wrappers for global ones.

Beer | [beer]

Christian Graus22-Oct-02 16:01

22-Oct-02 16:01

I would think your object needs to impliment operator new taking the same two parameters then.

Christian

No offense, but I don't really want to encourage the creation of another VB developer. - Larry Antram 22 Oct 2002

Hey, at least Logo had, at it's inception, a mechanical turtle. VB has always lacked even that... - Shog9 04-09-2002

During last 10 years, with invention of VB and similar programming environments, every ill-educated moron became able to develop software. - Alex E. - 12-Sept-2002

Nemanja Trifunovic22-Oct-02 16:10

22-Oct-02 16:10

Christian Graus wrote:
I would think your object needs to impliment operator new taking the same two parameters then.

Any idea what to do with this "extra" parameter (which is a pointer to the class type)?

Beer | [beer]

Christian Graus22-Oct-02 16:17

22-Oct-02 16:17

Nope - I've never done this stuff. If it were me, I'd write the function that takes the pointer, then I'd run it and check it like crazy for any sort of leak or problem. Or I'd look in the STL source for a similar operator to see what it gets used for.

Christian

No offense, but I don't really want to encourage the creation of another VB developer. - Larry Antram 22 Oct 2002

Hey, at least Logo had, at it's inception, a mechanical turtle. VB has always lacked even that... - Shog9 04-09-2002

During last 10 years, with invention of VB and similar programming environments, every ill-educated moron became able to develop software. - Alex E. - 12-Sept-2002

Nemanja Trifunovic22-Oct-02 18:46

22-Oct-02 18:46

Well, I just looked at "The Book" (St Bjarne Stroustrup), and found out that this new with two arguments is so-called "placement operator" - it usually just returns the pointer that comes as second argument. Strange...

Beer | [beer]

Christian Graus22-Oct-02 18:48

22-Oct-02 18:48

Unfortunately, and for reasons that I do not understand, we do not have that book at work, hence my inability to check on such details. Sounds like a winner though.

Christian

No offense, but I don't really want to encourage the creation of another VB developer. - Larry Antram 22 Oct 2002

Hey, at least Logo had, at it's inception, a mechanical turtle. VB has always lacked even that... - Shog9 04-09-2002

During last 10 years, with invention of VB and similar programming environments, every ill-educated moron became able to develop software. - Alex E. - 12-Sept-2002

Joaquín M López Muñoz22-Oct-02 21:09

Joaquín M López Muñoz

22-Oct-02 21:09

Uff, there's a couple of misunderstandings here.

void* operator new (size_t)
{
return ::new klasa;
}

This is the first problem in your program. An overloaded new operator is meant to be taking care of memory allocation only: the call to the class contructor is generated automatically by the compiler. Please note that the new you provide is calling implicitly the ctor through the use of golbal ::new (altough in this case the error will go unnoticed cause kasa has a trivial ctor). Instead, write something like:

void* operator new (size_t size)
{
return ::new char[size];
}

void* operator new (size_t size)
{
return malloc(size);
}

Similarly, your overloaded delete operator should be only concerned with deallocating memory without calling (implicitly) klasa::~klasa.
Now for the second problem. Once you provide an overload for new, the compiler won't use (for the relevant class) any of the predefined global new oeprators available. In particular, it won't use the so-called placement new, whose signature is:

void* operator new (size_t size,void* pos);

Placement new does not really do anything: the caller already provides the space (in pos) where the object is to be constructed. An example of an explicit call to placement new:

void *pos; // pointing to some suitable block of memory
whatever *p=new(pos)whatever;

Now, placement new is extensively used by allocators in STL, and this is what the compiler is complaining about: since you overloaded new you also have to provide an overload for the placement variant:

class klasa{
...
  void* operator new (size_t size,void* pos)
  {
    return pos;
  }
};

Hope I made myself clear enough Smile | :)

Joaquín M López Muñoz
Telefónica, Investigación y Desarrollo

Nemanja Trifunovic23-Oct-02 5:10

23-Oct-02 5:10

Hello, Joaquín. Thanks for your answer.

Joaquín M López Muñoz wrote:
This is the first problem in your program. An overloaded new operator is meant to be taking care of memory allocation only: the call to the class contructor is generated automatically by the compiler. Please note that the new you provide is calling implicitly the ctor through the use of golbal ::new (altough in this case the error will go unnoticed cause kasa has a trivial ctor).

As I already wrote to Christian, my original new and delete look quite different. I replaced my code with calls to global new and delete, because I wanted to make my examlple short and clear. I am aware of the issues you are talking about.

Joaquín M López Muñoz wrote:
In particular, it won't use the so-called placement new, whose signature is:void* operator new (size_t size,void* pos);
Placement new does not really do anything: the caller already provides the space (in pos) where the object is to be constructed.

Exactly. And that's what confuses me a lot. If you take a look at the implementation of placement new and delete in <new> you will see

inline void *__cdecl operator new(size_t, void *_P)
	{return (_P); }
#if     _MSC_VER >= 1200
inline void __cdecl operator delete(void *, void *)
	{return; }

What is this good for? Confused | :confused:

Joaquín M López Muñoz wrote:
Now, placement new is extensively used by allocators in STL, and this is what the compiler is complaining about: since you overloaded new you also have to provide an overload for the placement variant

You are 100% right. The default allocator uses placement new in function construct, and when I put this code in my class

void* operator new (size_t size,void* pos)  {    return pos;  }
void operator delete (void*, void*) {}

everything seems to work fine.

However, I still don't understand why. What is such implementation of placement new and delete good for? I know that in some cases you can provide i.e. your custom memory pool as a second argument, but here it is not the case. Why do they use this placement new in default allocator?

Beer | [beer]

Nemanja Trifunovic23-Oct-02 5:36