Help with threads

David Crow

pasztorpisti wrote:

The less hassle free solution is sending a message from the worker thread to the ui thread by using SendMessage()...

What happens when the primary (UI) thread is blocked and the worker thread sends it a message via SendMessage()?

"One man's wage rise is another man's price increase." - Harold Wilson

"Fireproof doesn't mean the fire will never come. It means when the fire comes that you will be able to withstand it." - Michael Simmons

"Show me a community that obeys the Ten Commandments and I'll show you a less crowded prison system." - Anonymous

pasztorpisti

Then the worker thread is waiting until the message is processed on the gui thread. Exploiting this you can for example pop up a messagebox from the worker thread by sending a WM_APP+X message to the gui thread that processes WM_APP+X and "blocks" until the message box is closed. The worker thread continues running only after the message box is closed and the processing of WM_APP+X is finished on the gui thread. I quoted "blocks" because a MessageBox call doesn't really block the gui thread, it just blocks the processing of a single message an it runs an inner messageloop of its own so processing other messages from the queue of the gui thread goes on. For example your windows are still drawn (by WM_PAINT messages) while a messagebox is active. PS: The ui thread should never be blocked, that causes unresponsive UI. Often thats the reason for multithreading and not gaining performance. If your UI thread is blocked then your program is either buggy or poorly designed.

David Crow

pasztorpisti wrote:

Then the worker thread is waiting until the message is processed on the gui thread.

I'm not talking about the primary thread being busy for a few nanoseconds. I'm talking about deadlock. For example, the secondary thread sends a "add item to the control" message (e.g., LB_ADDSTRING) to the primary thread, which is blocked waiting on the secondary thread to complete. Now the secondary thread cannot complete because it is no longer running (waiting on SendMessage() to return). This is the primary reason why SendMessage() should rarely, if ever, be used to communicate between primary and secondary threads.

"One man's wage rise is another man's price increase." - Harold Wilson

"Fireproof doesn't mean the fire will never come. It means when the fire comes that you will be able to withstand it." - Michael Simmons

"Show me a community that obeys the Ten Commandments and I'll show you a less crowded prison system." - Anonymous

pasztorpisti

The gui thread should never block, as a consequence it should never wait for the worker thread to complete. When the worker thread completes it can send/post a message to the gui thread to signal. Why would you want a blocking-wait on the gui thread???

Chuck OToole

DavidCrow wrote:

For example, the secondary thread sends a "add item to the control" message (e.g., LB_ADDSTRING) to the primary thread

This is the scenario that's confused me from time to time. It is really legal for a thread to do a SendMessage() to a control (the target of the LB_ADDSTRING) that is owned by another thread? I thought that Windows checked "thread ownership" of the control and returned an error on the SendMessage(). If the "primary thread" is, as is implied in your reply, the UI thread, then the "secondary thread" should not be allowed access to the control. I've always used "PostMessage()" and used "user defined messages" to have secondary threads pass messages / commands to the primary thread for action. Since "PostMessage()" is just a queueing action, there is no deadlock (although you might not get the immediate feedback of a screen update). Does anyone have the definitive answer on "SendMessage and Control Owner Thread" question?

AndrewG1231

Ok, instead of modifying my original question I will continue the thread since it has been really helpful. So, my program uses the following code to create a new thread.

m_Thread = AfxBeginThread(Acq_Data,this,THREAD_PRIORITY_HIGHEST);

This thread is told to look at the current class and the controlling function is Acq_Data. In Acq_Data there is the following code.

while(dlg->m_Continue && ! dlg->m_inst->m_Flags.Halt)
{
dlg->m_inst->ADCdbInquire(dlg->m_halfdata,hdlg,MSG_DRAW_SPECTRUM);
}

The ADCdbInquire()function contains a call to PostMessage()

bool CPXI::ADCdbInquire(short *halfbuffer, HWND hwnd ,int Message)
{
short iStatus, iHalfReady, iDAQstopped = 0;
unsigned long ulPtsTfr=1500; // number of points transferred

// Are you ready?
iStatus = DAQ\_DB\_HalfReady(m\_6052\_Device, &iHalfReady,
     &iDAQstopped);

if ((iHalfReady == 1) && (iDAQstopped == 0))
{
    iStatus = DAQ\_DB\_Transfer(m\_6052\_Device, halfbuffer,
    &ulPtsTfr, &iDAQstopped);

    NIDAQErrorHandler(iStatus, "DAQ\_DB\_Transfer",m\_SuppressErrors);

    if(Message != -1)
        PostMessage(hwnd,WM\_COMMAND,Message,NULL);

    return true;
}
return false;

}

and is sending a custom message defined in the header file. #define MSG_DRAW_SPECTRUM (WM_APP + 5) This custom messages refers to member function OnDrawSpectrum() where the following code exists to update the window.

void CDlg_SpectrumAnalyzer::OnDrawSpectrum()
{

int starti,finishi;
starti = m\_datacount \* m\_BuffSize / 2;
finishi = (m\_datacount + 1)\* m\_BuffSize / 2;

if(finishi > m\_AcqNPts)
    finishi = m\_AcqNPts;

int j = 0;
for(int i = starti; i < finishi; i++)
{
    m\_fftdata\[i\].re = m\_halfdata\[j++\];
    m\_fftdata\[i\].im = 0.;
}



m\_datacount++;

// if we are done collecting draw the graph
if((m\_datacount == m\_NBuffs))
{

    OnSpectrumStart();
    float freq = 0;
    float stepsize = ((float)(m\_MaxFreq)) / ((float)(m\_AcqNPts-1)/2.0);
    m\_datacount = 0;

    m\_Graph->ClearGraph(m\_graphNum,false);

    // do fft
    fftw\_one(m\_the\_plan, m\_fftdata, NULL);

    //m\_AcqNPts should be 2^n therefore /2 should be integer
    for(int i = m\_AcqNPts-1; i >= m\_AcqNPts/2;

Albert Holguin

Actually, SendMessage is synchronous... so if it's threaded, it really should be a PostMessage, otherwise you'll tie up the worker thread waiting on the GUI update.

Albert Holguin

Yeah that was pretty painful :)

Albert Holguin

That's a good question... I do believe the framework allows you to do it though... although now that you're questioning it, you're making me question whether I remember that correctly.

Albert Holguin

SendMessage() is a blocking call... that's what he was pointing out.

David Crow

AndrewG1231 wrote:

So, my question...in order to get this sequence to run several times, say 5 times, and update the window how would I use PostMessage()? Or if I have misunderstood something please let me know.

See here for more.

"One man's wage rise is another man's price increase." - Harold Wilson

"Fireproof doesn't mean the fire will never come. It means when the fire comes that you will be able to withstand it." - Michael Simmons

"Show me a community that obeys the Ten Commandments and I'll show you a less crowded prison system." - Anonymous

Chuck OToole

You know, some days I don't remember where my head is. Of course you can do it and, in fact, I do it all the time in certain apps. For example, if I have a "DO IT!" button, that usually creates a worker thread that does some stuff and updates a "log view" edit control in the main dialog. The GUI thread continues on and watches for the user clicking on "STOP DOING IT!". Meanwhile, the worker thread does a

TheDialog->m_editcontrol.SetText(logbuffer);

or something equivalent to update the subclassed edit window. Works just fine. I guess I forget this because the "SendMessage()" is buried in the SetText() function and not something I explicitly do but it is still "SendMessage()" from a thread that did not "own" or "create" the control. Oh well, must be getting old.

Albert Holguin

Chuck O'Toole wrote:

Oh well, must be getting old.

Happens... Yeah, as a rule of thumb though, I like to tell people to PostMessage() when doing things between threads (since they're supposed to be independent anyway). As you probably know, the synchronicity of SendMessage() has the potential for deadlocks so it should only be used when you truly need it.

AndrewG1231

Thanks for the link to the article and I have tried to implement the PostMessage() in my code. However, I am still having problems trying to get it to execute more than once and update the window (looping the code is not working). After reading your suggested article, I am still not sure how to do this correctly, can you help? My code is as follows.

void CDlg_SpectrumAnalyzer::OnBtnClickedRun()
{
PostMessage(WM_COMMAND,IDC_SPECTRUM_START,0);
}

Yes, this is a "Run" button that sends the message and it executes correctly if called only a single time.

David Crow

AndrewG1231 wrote:

...it executes correctly if called only a single time.

What happens when the handler is called a second time?

"One man's wage rise is another man's price increase." - Harold Wilson

"Fireproof doesn't mean the fire will never come. It means when the fire comes that you will be able to withstand it." - Michael Simmons

"Show me a community that obeys the Ten Commandments and I'll show you a less crowded prison system." - Anonymous

AndrewG1231

When it is called the second time I encounter some code for checking the state of the instrument my program is controlling. You can see the whole block of code this belongs to in the first post of this thread.

if(!m_inst->StartOperation())
{
MessageBox("The Instrument is Currently in use by another function");
return;
}

I thought this indicated that the previous thread wasn't terminating before the message posted again and I tried to use Sleep() to delay the call, but this did not work. I encountered the same situation or the window did not update. I also tried to use the WaitForSingleObject(), but this failed as well.

David Crow

Look this over. It's from memory and completly untested so there may be a few small details left out, but it should give you good idea of how primary and secondary threads communicate.

BEGIN_MESSAGE_MAP(CMyDialog, CDiaog)
ON_BN_CLICKED(IDC_BUTTON, OnClick)
END_MESSAGE_MAP()

void CMyDialog::Acq_Data( void )
{
while (1)
{
// if a STOP event has been signaled [m_pEventStopRequested.SetEvent()]
if (WaitForSingleObject(m_pEventStopRequested->m_hObject, 0U) == WAIT_OBJECT_0)
{
// let OnClick() know that the secondary thread is not running
m_pEventThreadDone->SetEvent();

        return;
    }
    else
        ADCdbInquire(...); // this functon can post messages back to the primary thread
}

}

/* static */ UINT CMyDialog::Acq_Data( LPVOID lpVoid )
{
CMyDialog *pDlg = (CMyDialog *) lpVoid;

pDlg->Acq\_Data();

// once here, the secondary thread is done so post a message
// back to the primary thread to tell it to enable the button

return 0;

}

// put these in CMyDialog
CWinThread *m_pThread = NULL;
CEvent *m_pEventStopRequested = new CEvent(FALSE, TRUE); // stop requested (initially no)?
CEvent *m_pEventThreadDone = new CEvent(TRUE, TRUE); // thread done (initially yes)?

void CMyDialog::OnClick()
{
// if the secondary thread is not running
if (WaitForSingleObject(m_pEventThreadDone->m_hObject, 0U) == WAIT_OBJECT_0)
{
// at this point, you could disable the button that
// started this so that it could not be clicked again

    // set the STOP and DONE events to nonsignaled (haven't happened yet)
    m\_pEventStopRequested->ResetEvent();
    m\_pEventThreadDone->ResetEvent();

    // start the secondary thread
    m\_pThread = AfxBeginThread(Acq\_Data, this, THREAD\_PRIORITY\_HIGHEST);
}

}

"One man's wage rise is another man's price increase." - Harold Wilson

"Fireproof doesn't mean the fire will never come. It means when the fire comes that you will be able to withstand it." - Michael Simmons

"Show me a community that obeys the Ten Commandments and I'll show you a less crowded prison system." - Anonymous

AndrewG1231

Thanks for the suggestions! However, this piece of code is causing me to get the error, "A nonstatic member reference must be relative to a specific object". I tried to use this-> before m_pEventStopRequested->m_hObject but it did not solve the problem. I have the declaration for the CEvent in the constructor, but I am not sure to what object it is referring.

if (WaitForSingleObject(m_pEventStopRequested->m_hObject, 0U) == WAIT_OBJECT_0)
{
// let OnClick() know that the secondary thread is not running
m_pEventThreadDone.SetEvent();
return;
}

Would you happen to know what object it is referring to?

David Crow

AndrewG1231 wrote:

However, this piece of code is causing me to get the error, "A nonstatic member reference must be relative to a specific object".

What is the exact error number?

AndrewG1231 wrote:

m_pEventThreadDone.SetEvent();

Should be:

m_pEventThreadDone->SetEvent();

That, and adding a return value, is all I had to do to get the code to compile.

"One man's wage rise is another man's price increase." - Harold Wilson

"Fireproof doesn't mean the fire will never come. It means when the fire comes that you will be able to withstand it." - Michael Simmons

"Show me a community that obeys the Ten Commandments and I'll show you a less crowded prison system." - Anonymous

AndrewG1231

Here is the exact compiler message of the error.

Dlg_SpectrumAnalyzer.cpp(150): error C2227: left of '->m_hObject' must point to class/struct/union/generic type
1>Dlg_SpectrumAnalyzer.cpp(153): error C2227: left of '->SetEvent' must point to class/struct/union/generic type
1>Dlg_SpectrumAnalyzer.cpp(154): error C2561: 'CDlg_SpectrumAnalyzer::Acq_Data' : function must return a value

However, I implemented you suggestion a bit differently. I did not use void Acq_DAta(void) instead I have incorporated everything in the following member.

UINT CDlg_SpectrumAnalyzer::Acq_Data(LPVOID pParam)
{

CDlg\_SpectrumAnalyzer \*dlg =(CDlg\_SpectrumAnalyzer\*) pParam;
HWND hdlg = dlg->GetSafeHwnd();
// start data ACQ
while(1)
{
     if (WaitForSingleObject(m\_pEventStopRequested->m\_hObject, 0U) == WAIT\_OBJECT\_0)
         {
        // let OnClick() know that the secondary thread is not running
        m\_pEventThreadDone->SetEvent();
        return;
     }
     else
     {
		
	dlg->m\_inst->ADCdbRead(dlg->m\_AcqChan,dlg->m\_MaxFreq\*2,dlg->m\_BuffSize,dlg->m\_fulldata,dlg->m\_halfdata);
	while(dlg->m\_Continue && ! dlg->m\_inst->m\_Flags.Halt)
		dlg->m\_inst->ADCdbInquire(dlg-   >m\_halfdata,hdlg,MSG\_DRAW\_SPECTRUM);

		dlg->m\_inst->ADCdbStop();
		//release instrument
		dlg->m\_inst->EndOperation();
		
		return 1;
     }
	 
}

}

Here is the button control code.

void CDlg_SpectrumAnalyzer::OnBnClickedRun()
{
// TODO: Add your control notification handler code here
if (WaitForSingleObject(m_pEventThreadDone->m_hObject, 0U) == WAIT_OBJECT_0)
{
// at this point, you could disable the button that
// started this so that it could not be clicked again

    // set the STOP and DONE events to nonsignaled (haven't happened yet)
    m\_pEventStopRequested->ResetEvent();
    m\_pEventThreadDone->ResetEvent();

    // start the secondary thread
    m\_pThread = AfxBeginThread(Acq\_Data, this, THREAD\_PRIORITY\_HIGHEST);
}

}

Have I done something foolish in the way I was trying to use your suggestion?