C / C++ / MFC

12-Aug-02 9:27

I have a class where I have member variables.

I have declared a static member function to handle one single call to AfxBeginThread(...).

AfxBeginThread calls this static member function that calls a non-static member function in return that does all the background processing.

If I use global vars (declared outside the class scope) the application is fine. But as soon as I declare those globals as protected members of this class which calls AfxBeginThread, I get an access violation. My class looks like this:

class CReadIntoSybaseDlg : public CDialog
{
// Construction
public:
CReadIntoSybaseDlg(CWnd* pParent = NULL); // standard constructor
static UINT TestThread(LPVOID pParam);
UINT RunThread();

// Dialog Data
//{{AFX_DATA(CReadIntoSybaseDlg)
enum { IDD = IDD_READINTOSYBASE_DIALOG };
CButton m_OK;
CButton m_Cancel;
//}}AFX_DATA

// ClassWizard generated virtual function overrides
//{{AFX_VIRTUAL(CReadIntoSybaseDlg)
protected:
virtual void DoDataExchange(CDataExchange* pDX); // DDX/DDV support
//}}AFX_VIRTUAL

// Implementation
protected:
HICON m_hIcon;
CWinThread *m_pThWorker;
CMutex *cs;

CDatabase db;
CDaoDatabase db2;
CSybNames rs;
CDaoRecordset ds;
COleVariant oleVar;
CDaoTableDefInfo tb;
volatile BOOL running;

// Generated message map functions
//{{AFX_MSG(CReadIntoSybaseDlg)
virtual BOOL OnInitDialog();
afx_msg void OnSysCommand(UINT nID, LPARAM lParam);
afx_msg void OnPaint();
afx_msg HCURSOR OnQueryDragIcon();
virtual void OnOK();
virtual void OnCancel();
afx_msg void OnStop();
//}}AFX_MSG
DECLARE_MESSAGE_MAP()

};

I call AfxBeginThread(TestThread,0)
Then this TestThread is declared like this:

UINT TestThread(LPVOID pParam)
{
return ((CReadIntoSybaseDlg *)AfxGetApp())->RunThread();
}

In RunThread(), I am using the various CDatabase and CDaoDatabase variables as members of this class i.e. db, db2, rs,ds. If I declare them globally, its fine. But the moment I put my declaration in the class, it starts complaining. I know it has to do with threading but I have no clue. Could someone please help.

Thanks a lot in advance.

ashish

Bijesh12-Aug-02 10:32

Bijesh

12-Aug-02 10:32

achandra007 wrote:
UINT TestThread(LPVOID pParam)
{
return ((CReadIntoSybaseDlg *)AfxGetApp())->RunThread();
}

AfxGetApp() returns the main application object, which is a CWinApp - derived object. You should not cast that to your dialog class. You need to some other way to get the dialog object pointer ( if you app is a dialog based app then AfxGetMainWnd() might do it)

--------------------------------------------------

achandra00712-Aug-02 10:43

12-Aug-02 10:43

Thanks for the suggestion but my problem is with the threading. I can't seem to avoid the Access Violation when I use the CDatabase etc variables in my class and not as globals.

ashish

Daniel Lohmann12-Aug-02 10:41

Daniel Lohmann

12-Aug-02 10:41

achandra007 wrote:
UINT TestThread(LPVOID pParam)
{
return ((CReadIntoSybaseDlg *)AfxGetApp())->RunThread();
}

You are doing complete nonsense here! AfxGetApp() retuns a pointer to the one-and-only CWinApp object. This has nothing to do with your dialog and the result of your cast is just "undefined".

--

Daniel Lohmann

http://www.losoft.de
(Hey, this page is worth looking! You can find some free and handy NT tools there Big Grin | :-D

)

achandra00712-Aug-02 11:11

12-Aug-02 11:11

Thank you for politely replying to my question. Suprisingly, the "nonsense" works when I use global variables of type CDataset etc. But I won't waste your time in getting you to explain that. If you can answer my question (see the original post), then I thank you in advance. Otherwise, don't waste my time with these useless suggestions.

thanks
ashish

Daniel Lohmann12-Aug-02 12:23

Daniel Lohmann

12-Aug-02 12:23

Ashish,

Why do you try to blame the experienced people that just want to help you? Mad | :mad:

Surprisingly it works is really the right term here! This is definitively a huge bug in your code which, as I wrote, results in undefined behaviour. Undefined beheaviour means that anything could happen: Even that it surprisingly seems to work, but just stops working if you change something at a complete different position in your code.

To make it short: This bug has a good chance to be the reason for your access violations!

Fix it, check what happens and if you have still problems post the code again. It would be very kind if you enclose it with <pre> and </pre> tags next time, so it will appear well formatted in the forum. This would make it much better readable and therfore increases your chances to get a no-"nonsense" answer.

--

Daniel Lohmann

http://www.losoft.de
(Hey, this page is worth looking! You can find some free and handy NT tools there Big Grin | :-D

)

achandra00712-Aug-02 12:27

12-Aug-02 12:27

Daniel

I am not trying to blame anyone. I asked a question and I think you just got off about the code. I admit it may have bugs - that is why I am asking the questions. But read my original post and then read my reply to Wes Jones. If you can still help, then thanks in advance. Otherwise, there is no need to pontificate. I will learn without it as well.

thanks
ashish

Wes Jones12-Aug-02 12:05

Wes Jones

12-Aug-02 12:05

Hi Ashish,

Not to be rude, but there are plenty of problems w/ the code.

The access violoation you are seeing is due to the fact that CYourApp does not have a RunThread() function.
When you call AfxGetApp() is is returning a pointer to a CWinApp derived class, which is really a pointer to your CYourApp class which the class wizard creates for you. The ptr returned by AfxGetApp() is totally unrelated to any instances of CReadIntoSybaseDlg. The reason it didn't blow up on you when you used global data is because there wasn't any code implicitly trying to access the 'this' pointer in the implementation of RunThread. Since you're casting the ptr to CReadIntoSybaseDlg, your tricking it into calling your CReadIntoSybaseDlg::RunThread function. Your new function is trying to access member data via the 'this' ptr, but since you've casted the pointer to an invalid type, the 'this' ptr is really invalid.

Here's the proper way to call it:

CReadIntoSybaseDlg::StartThreadFunc()
{
   m_pThWorker = AfxBeginThread(TestThread, this);//<-- note, passing this 
                                                  //ptr, not 0
}
UINT CReadIntoSybaseDlg::TestThread(LPVOID lpvParam)
{
   CReadIntoSybaseDlg * pThis = reinterpret_cast<CReadIntoSybaseDlg*>(lpvParam);
  ASSERT( pThis );//if this goes off, you passed the wrong param to 
                  //AfxBeginThread

  pThis->RunThread();
  return 0;
}

Other things that would make your code a little nicer would be to rename the member variables w/ the m_ prefix.

I'd recommend that you, or anyone else reading this, read the book "Programming Windows Applications" by Jeffrey Richter, or even the earlier version of the book titled "Advanced Window Applications" or something like that.
Also, get a copy of the big red MFC book from Wrox press by Mike Blazncack (sp?).

Without knowing the rest of your code, you're probably going to be better of, performance wise, if you use a CCriticalSection instead of a CMutex.

I also want to warn you to be careful not to access the database objects from different threads at the same time, which is probably what your doing w/ the CMutex anyway. Bad things can happen if one thread's doing an insert and another's doing a rollback.
HTH,
-Wes

Sonork ID 100.14017 wtheronjones

achandra00712-Aug-02 12:21

12-Aug-02 12:21

Wes

Thanks a lot for the information. In fact, I have done exactly as you have said. I have also changed the cast to use AfxGetMainWnd() instead of AfxGetApp().

This is really my first multi threading application and I am having a few problems with it. My goal is to have 1 worker and 1 UI thread. Both of them are launched successfully now but now I am having trouble with using CDaoDatabase based calls. I call the GetTableDefInfo and loop through checking the lAttribute (to check whether its a user table or a system table). The call to GetTableDefInfo goes through once but at the second call, the execution gets stuck at DAO_CHECK(pDAOTableDef->get_ValidationRule(&V_BSTR(&var))); call in void CDaoDatabase::FillTableDefInfo(DAOTableDef* pDAOTableDef,CDaoTableDefInfo& tabledefinfo, DWORD dwOptions) call.

I am pretty sure its related to the change in my code (from single threaded to dual-threaded). Could you please help.

Thanks
ashish

Wes Jones12-Aug-02 12:40

Wes Jones

12-Aug-02 12:40

Hi ashish,

Your function shouldn't cast to AfxGetMainWnd(), you should pass the this ptr to AfxGetMainWnd(). Your thread may work now, but that's only because this is probably a dialog based app where CReadIntoSybaseDlg happens to be your main window (AfxGetMainWnd() is returning a ptr to an actual CReadIntoSybaseDlg). You always want to know what parameter you're passing to your thread function, and AfxGetMainWnd isn't going to return a known type if you ever wanted to use this code in another program.
I don't have much experience w/ CDaoDatabase to know what the exact problem might be, but I suspect that maybe you're opening the record set in your worker thread, and moving through the record set in your main thread(or a similar scenario). If that's the case try to ensure that all access to the database related objects are taking place in the same thread because I suspect that CDaoDatabase doesn't like it if you're doing things from multiple threads. It's conceivable that MFC/CDaoDatabase is attaching itself to a CWnd object in your main thread since that's where it's probably being instantiated from. If MFC is attached to a CWnd in your main thread, and your main thread is blocking while waiting for a worker thread to finish something w/ the database, you could see a lockup condition.

Best advice would to be sure to do all your database accessing from the same thread that's using the db objects. Try doing all the processing in your main thread, and then try it again so that all the processing in the worker thread to see if it behaves differently.
Btw, Do you really need a worker thread for what you're trying to do? I use Sybase & I don't find the queries to take that long.
If you're designing your app so that the UI thread displays data when the worker thread signals that data is ready & your accessing the same db objects, & if CDaoDatabase really has an issue w/ being used this way, it's going to cause you to have to do some extra work.

-Wes

Sonork ID 100.14017 wtheronjones

achandra00713-Aug-02 2:52

13-Aug-02 2:52

Wes

Thanks for the advice. I will make the appropriate changes.

You may be right about CDaoDatabase. I am opening it in the main thread (Open call) and then passing the main processing function to AfxBeginThread so that all the processing (multiple databases are being used) can be done in the background. The reason I have to use two threads is that I need GUI control over the dialog window.

I read some of the DAO documentation and it does not support multiple threads. However, I had thought if I did everything in the worker thread, it should be OK. Apparently, it does not like being opened with the main thread and then being worked on by another. I will probably have to move the code around.

Thanks a lot
ashish

achandra00713-Aug-02 2:58

13-Aug-02 2:58

Wes,

On another note, since you mentioned you are using Sybase, I too am using Sybase but with the MFC ODBC classes. It is too slow, especially with inserts and updates. Any hints?

Thanks
ashish

Wes Jones13-Aug-02 5:31

Wes Jones

13-Aug-02 5:31

Hey,

I don't find the ODBC to be slow at all for inserts & updates... I am using more of the ODBC API instead of CDatabase, but that's cause most of my classes work w/ CE (no CDatabase on CE). There are some things you can do to increase performance such as not using CRecordset::AddNew() and then CRecordst::Update to insert a new row. If you know the values, just write an insert statement... it's only one trip to the database instead of two, (unless you need retrieve the new primary key). If you're doing more than a certain # of inserts to the same table(a batch operation let's say), it's best to use a prepared statement & bound parameters at more of an API level than to use a bunch of single insert statements or using AddNew/Update. Just guessing, but i'd think that the odbc classes will be faster than the dao classes.

-Wes

Sonork ID 100.14017 wtheronjones

achandra00713-Aug-02 7:50

13-Aug-02 7:50

I meant the ODBC interface in MFC implemented through CDatabase. I find that to be slow. But its only slow with Sybase and not with SQL Server, which is my other data source.

thanks
ashish

Memory dump

Hugo Hallman12-Aug-02 8:09

Hugo Hallman

12-Aug-02 8:09

In VS6 you could see a live memory-dump in a window. I can't find this feature in VC7. Is it left and hidden, or removed?

Re: Memory dump

PJ Arends12-Aug-02 15:09

PJ Arends

12-Aug-02 15:09

Debug -> Windows -> Memory -> pick on of 4 windows

CPUA 0x5041

Sonork 100.11743 Chicken Little

"So it can now be written in stone as a testament to humanities achievments "PJ did Pi at CP"." Colin Davies

Within you lies the power for good - Use it!

printf question

Shawn Horton12-Aug-02 8:10

Shawn Horton

12-Aug-02 8:10

I am converting a UNIX application to Windows using VC++ and MFC. There are a large number of printf statements in the code, so I re-defined the printf function, with vsprintf, so that all text will go to a richedit control.

The problem I face is that when a data file is read, it is possible that one of the fields will have a "%" sign as part of its value. When this value is displayed, it looks as though vsprintf is behaving as documented and printing the value without the % sign. Does anyone know of an easy, efficient way to make sure the % sign is printed as well?

Thanks,

Shawn

Tomasz Sowinski12-Aug-02 8:21

Tomasz Sowinski

12-Aug-02 8:21

Shawn Horton wrote:
Does anyone know of an easy, efficient way to make sure the % sign is printed as well?

I'm not sure what exactly do you want to achieve. You've mentioned that there are fields in data file and vsprintf replaces % sequences with actual values. Do you want to keep a percent sign before the value? If this is the case, try appending %% in front - this will result in one percent sign being printed.

Tomasz Sowinski -- http://www.shooltz.com

What is "scratch" and why can everything be made from it?

Shawn Horton12-Aug-02 8:39

Shawn Horton

12-Aug-02 8:39

Maybe this will help:

char data_name[DATANAMELEN];

read_function(data_name);

//data_name now contains the value "%recovered"

printf("we read in %s", data_name);

The printf statement to the Richedit window is now
we read in recovered

notice the missing %?

Maybe that is a bit clearer.

My hope is that I will not have to look for a % in character in each variable, and then insert a % in front of that.

Thanks for the help.

Tomasz Sowinski12-Aug-02 8:54

Tomasz Sowinski

12-Aug-02 8:54

Shawn Horton wrote:
Maybe that is a bit clearer.

Totally clear now. I don't know about any format modifier which could cause [v]sprintf to literally copy % sign.

Tomasz Sowinski -- http://www.shooltz.com

What is "scratch" and why can everything be made from it?

Shawn Horton12-Aug-02 9:01

Shawn Horton

12-Aug-02 9:01

I should just post something like "Duh I am having a problem" on my first post, because the second one usually makes more sense.