C / C++ / MFC

4-Feb-04 1:42

Hi, I have a logging class which contains a <code>static std::map</code> of logfilename to std::ofstream, so that I can ensure every logging object uses the correct stream.

When I first attempt to access the map in any way, by calling <code>insert()</code> or anything else, I get an Access Violation from within STL. Has anybody come across this before, or am I missing something obvious?

My classes are outlined as follows:

<code>class LogThis
{
public:
LogThis(const std::string& logName);
~LogThis();
...
private:
...
static std::map<std::string, LogFileStream> mLogFileStreams;
};</code>

Where LogFileStream is just a simple class with a pointer to an std::ofstream and a reference counter of how many logging objects are using the stream, so I know when I can close the stream and delete the pointer.

The logging class is contained within a static lib and I am using Visual Studio.Net.

Any help would be much appreciated.

Thanks,
Christopher

The bomb lives only as it is falling

Diddy4-Feb-04 2:03

4-Feb-04 2:03

How have you then "declared" this static variable (mLogFileStreams? By that I mean in the CPP for the class, you will have something to initialise the map, like this:

std::map<std::string, LogFileStream> LogThis::mLogFileStreams;

What have you got there?

Try a simple test, change the map to map "simple" types, ie map int-to-int, that way you can pinpoint the problem a bit more to either the LogFileStream class causing the map problems, or the map itself.

inter8ection4-Feb-04 3:34

4-Feb-04 3:34

Yes, I initialise it at the top of my cpp file, otherwise it wouldn't compile.

std::map<std::string, LogFileStream> LogThis::mLogFileStreams = std::map<std::string, LogFileStream>();

I also tried just this:

std::map<std::string, LogFileStream> LogThis::mLogFileStreams;

But it didn't seem to make any difference. Anyway, I shall experiment with some primitive types and report back.

The bomb lives only as it is falling

Diddy4-Feb-04 3:44

4-Feb-04 3:44

simcoe23 wrote:
Yes, I initialise it at the top of my cpp file, otherwise it wouldn't compile.

Note is says "how have you" not "have you" Smile | :)

Anyway, std::map definatly do work as static memebers, so it's probably something with your other class. For instance this works without problem

typedef map<int, int> MYMAP;

class A
{
public:

static MYMAP aMap;
};

MYMAP A::aMap;

int main(int argc, char* argv[])
{
A aa;
aa.aMap.insert(MYMAP::value_type(3, 3));
return 0;
}

inter8ection4-Feb-04 4:01

4-Feb-04 4:01

Hi there, sorry I didn't notice the "how have you" Smile | :)

.

I just tried with a std::map<int,int> and exactly the same thing happens.

The map.insert code is being called in the constructor of my main class, might that be a problem? Basically you create this logging object by passing a filename into the constructor, it checks if the file stream is already open, if not it opens a new stream and stores it in this static map using the filename as a key.

The other thing to mention is that the logging object which is being constructed is a global - I declare it using 'extern' at the bottom of my header file, which may or may not be the right thing to do.

The bomb lives only as it is falling

Diddy4-Feb-04 4:07

4-Feb-04 4:07

Yes

- Maybe

The important bit is "logging object which is being constructed is a global " does this mean that the class that is using the map, which is static, is also being declared as global? Or does it mean the logging class which uses the map is declared global, then you have another "main" class that uses the logging one, which isn't global?

When your app starts up, the C runtime libray will init all your static and global members and doesn't do these in any perticualr order - when your constructor calls insert on the map, it's likley that the map hasn't been created yet.

Comment out the line that insert's into the map into the ctor, put a break point in your classes ctor, when it gets hit, check the call stack and go back to the _inititem functions - this is the CRT calling the ctors for all your global variables. Step back out of your ctor to this loop, and you should see that the loop then go into the ctor for the map - IE the map is being created after your ctor is being called, and you get a AV.

You could have a Init() method on your logging object or something, there are a few soultions to this problem really.

inter8ection4-Feb-04 4:40

4-Feb-04 4:40

Cool, thanks. Changed it so that the stream is only opened the first time you write to it, and that seems to work. Now of course I am having the same problem when the object destructs, but at least i know what's causing it.

Thanks for your time.

The bomb lives only as it is falling

Diddy4-Feb-04 4:47

Problem with Kernel Object Security

4-Feb-04 4:47

No problem Smile | :)

Nowt to do today at work anyway Big Grin | :-D

particle2k4-Feb-04 1:15

particle2k

4-Feb-04 1:15

hello,

i need to adjust the access rights to the local cd drives installed in the computer. for this i'm retrieving kernel object security and adjust the dacl by adding a grant access with the necessary rights for the users group.
this worked perfectly in the past, the settings are applied and the security changes affected the system as expected.
but now the settings are removed after restarting the computer. all functions return a successful condition and the settings are changed as long as the computer is not restarted.
does this have to do with w2k sp3 ? we were using sp1 previously and it was working there.
does anyone has a clue what i can check or where i can search ?

many thanks in advance
enrico

Re: Problem with Kernel Object Security

Alexander ^M.,4-Feb-04 3:09

Alexander ^M.,

4-Feb-04 3:09

first: sp4 is the latest!
2nd: why do u need to adjust the access rights for your cd drive?

Don't try it, just do it! Wink | ;-)

Re: Problem with Kernel Object Security

particle2k4-Feb-04 3:31

particle2k

4-Feb-04 3:31

i need to adjust the security settings to be able to record cds with the nero software even if i do not have admin rights. nero is providing a tool to adjust the burn rights, but i tried to recreate the functionality to include it in another application, so i do not need the nero burn rights tool and can automate the process of changing the security settings.
and this was working perfectly, but in the meantime the behaviour changed. i cannot say, if it is related to the service pack, but i have no idea what i can check. adjusting the security with the nero utility is still working by the way.

Multithreading in C++

TV4-Feb-04 0:11

4-Feb-04 0:11

Hello,

I am trying to do a object oriented multithreaded program with MFC.

In order to do that, I have a class CToto whose contructor creates 2 threads with AfxBeginThread. As an argument to AfxBeginThread I pass the this pointer.

CToto::CToto()<br />
{<br />
   /...other inits here.../<br />
    fastThread = AfxBeginThread((AFX_THREADPROC)ObjectFastThreadRoutine, this);<br />
    slowThread = AfxBeginThread((AFX_THREADPROC)ObjectSlowThreadRoutine, this);<br />
}

(The CToto instance is created on the heap with the new operator in the OnInitDialog of a CDialog class.)

The thread proc are as follow:

UINT ObjectFastThreadRoutine(CToto* lpParam) <br />
{ <br />
    lpParam->FastThread();<br />
    return 0;<br />
}

UINT ObjectSlowThreadRoutine(CToto* lpParam) <br />
{ <br />
    lpParam->SlowThread();<br />
    return 0;<br />
}

I have therefore two thread running in two members functions of the CToto class. The goal of this approach is to acces to the variables of this class with both threads.

To be thread safe, I read and write in the member variables that I want to share with functions which are protected with CCriticalSection objects, here is one of those functions used to write to the positionMeasured member variable of CToto:

void CToto::SetPosition(double *pos)<br />
{<br />
    CSingleLock singleLock(critSectionPosition); //critSectionPosition is also a member variable of the class<br />
    singleLock.Lock();<br />
<br />
    for(int i=0; i<4; i++)<br />
    {<br />
        positionMeasured[i] = pos[i];<br />
    }<br />
<br />
    singleLock.Unlock();<br />
}

I have three differents member variables such as positionMeasured that are protected like that. All those three variables have Get and Set functions to read and write.

My problem occurs when the SlowThread tries to access to the SetPosition function. I have a debug assertion failed at the CSingleLock singleLock(critSectionPosition); line. The assertion is the following:

CSingleLock::CSingleLock(CSyncObject* pObject, BOOL bInitialLock)<br />
{<br />
     ASSERT(pObject != NULL);<br />
---->     ASSERT(pObject->IsKindOf(RUNTIME_CLASS(CSyncObject)));<br />
<br />
     m_pObject = pObject;<br />
     m_hObject = pObject->m_hObject;<br />
     m_bAcquired = FALSE;<br />
<br />
     if (bInitialLock)<br />
          Lock();<br />
}<br />
BOOL CObject::IsKindOf(const CRuntimeClass* pClass) const<br />
{<br />
     ASSERT(this != NULL);<br />
     // it better be in valid memory, at least for CObject size<br />
------>     ASSERT(AfxIsValidAddress(this, sizeof(CObject)));<br />
<br />
     // simple SI case<br />
     CRuntimeClass* pClassThis = GetRuntimeClass();<br />
     return pClassThis->IsDerivedFrom(pClass);<br />
}

I really don't know what to do. I am not an expert in multithreading and maybe (sure) I am doing something false with memory access and memory zone of threads. So if anybody can help me on how to solve this problem or can propose me another program structure for one class with shared ressources and two threads accesing them concurently (those threads must be able to access the private members of the class if possible)

Thank a lot for your answers.

Regards

TV

Re: Multithreading in C++

Diddy4-Feb-04 2:11

Re: Multithreading in C++

4-Feb-04 2:11

First of all, you are not _really_ using CSingleLock correctly.

The point of CSingleLock is to basically aid in cleaning up your critical section if your code throws an exception, or returns take this:

fun()
{
CCriticalSection section;
section.lock()
.
.
.
if (ladeda)
return false;

section.unlock();
}

here, if the "if (ladeda)" condition was true and you returned, you would be left with a locked CS. - using it this way you have to remeber to unlcok the CS before you return, it makes it harder to use exceptions or to have multiple return points in a function. CSingleLock overcomes that by locking in the constructor and unlocking in the destuctor, so by declaring one on the stack, your CS will automatically be unlocked when that stack unwinds - where ever you may be.

You use CSingleLock to overcome the problem like this:

fun()
{
CCriticalSection section;
CSingleLock lock(§ion, TRUE).
.
.
if (ladeda)
return false;

// no need to unlock section, done by ~CSingleLock
}

In other words, remove the explicate calls to singleLock.Lock(); and singleLock.Unlock(); in your code, and change the decliaration to be CSingleLock singleLock(critSectionPosition, TRUE); - this causes the constructor to call Lock for you.

As for your actual problem, well, it depends on what type your critSectionPosition data member is... You haven't included that bit Blush | :O

) You have to pass a pointer to a CS (or any other sync object) to CSingleObject, the assert is telling you your what you are passing is not of type CSyncObject - ie your not passing a CCritcalSection* which dervices from CSyncObject.

If you put up the critSectionPosition data type, I should be able to tell you.

TV4-Feb-04 3:58

4-Feb-04 3:58

I defined critSectionPosition as a pointer and create the object with new in the CToto constructor. I also tried to create it in the stack and I get the same problem.

I define my CCriticalSection objects as members of CToto because I have a GetPosition and a SetPosition fonction so I need the same CCriticalSection object to lock the position variable which is accessed by those two function independentely.

I juste tried now replacing my CCriticalSection by CMutex and it is working...
I know that CMutex are not well suited for my case so I would really like to manage to do that with CCriticalSection.

Thanks for your time

Re: Multithreading in C++

Diddy4-Feb-04 4:30