|
Hi.
I am working on my first "real" Windows program. I am at a point in the implementing process where I see a need to multithread one specific job. I will give a good example of something that is very similar to what I am to implement. I will then give my thoughts on a possible solution. I have no prior experience with multithreading, but Jeff Prosise's book does wonders.
I am design a dialog box with a progress bar *very similar* to
the one you see when you download a file using IE. When you download a large file (your download bandwidth is slow), you can see the progress of the download process. The process bar is easy to implement. The part that makes IE progress bar special is you can *still use* IE during the download process. The key is multithread.
Here is my design.
-----
-User activate a feature (via menu)
-View creates a dialogbox w/ progressbar (position 0)
-Dialogbox sends a message to MainFrm.
-MainFrm redirect message to View
-View calls function in Document to begin data processing
-View also passes the function a pointer to dialogbox
-Document starts a new thread (worker thread)
-Thread begin processing data
-Thread sends message back to Document when applicable
-Document calls updated progressbar via pointer to dialog
-----
Again, this is my first experiment with multithread. Is there anything wrong with the design above? I left some small features out. I will include it soon.
Thanks,
Kuphryn
|
|
|
|
|
Some details are missing to make a full evaluation of this approach, but IMHO the possibility of a deadlock is lurking in your design.
If your dialog features a cancel button, and you implement it like this:
void CProgressDialog::OnCancel()
{
signal the thread to die;
wait for the signal to die with WaitForMultipleObject or some similar API;
}then there's a possible deadlock here. Suppose the user presses the Canel button and CProgressDialog::OnCancel is called. If right after the thread is signalled to die the thread happens to be in an update call to the document, and the updating process involves sending (explicitly or implicitly) some message to CProgressDialog, then you've hit a deadlock, as the thread won't answer the termination signal until it's returned from the update call, and CProgressDialog won't process any further messages until it returns from CProgressDialog::OnCancel. You got it? In general, to avoid such deadlocks between a worker thread and the GUI part of a program, it is advisable to have the worker thread post messages to the GUI instead of sending them in a blocking fashion.
Joaquín M López Muñoz
Telefónica, Investigación y Desarrollo
|
|
|
|
|
Thanks.
Very point!
I do not quite understand what you meant with "sending them in a blocking fashion."
You were right on with OnCancel(). The way I planned to "signal the thread to die" is by using a boolean counter. OnCancel() would send a message to Doc. Doc would then set the counter to false. Meanwhile, the worker thread does its job only if counter is true.
Do you mean it is advisable to *post message* instead of *send message*?
Kuphryn
|
|
|
|
|
I do not quite understand what you meant with "sending them in a blocking fashion."
SendMessage, when issued from a worker thread, does not return until the GUI thread processes the message sent. So if the GUI is stuck, SendMessage will wait forever. PostMessage, on the other hand, only leaves the message in the GUI thread message queue and returns immediately.
You were right on with OnCancel(). The way I planned to "signal the thread to die" is by using a boolean counter. OnCancel() would send a message to Doc. Doc would then set the counter to false. Meanwhile, the worker thread does its job only if counter is true.
The deadlock will happen only if OnCancel waits for the thread to actually die before returning (which is the proper way to proceed, on the other hand).
Do you mean it is advisable to *post message* instead of *send message*?
Yes. Be careful with the parameters accompanying a posted message --they mustn't be pointers to temporary variables.
Joaquín M López Muñoz
Telefónica, Investigación y Desarrollo
|
|
|
|
|
Thanks for the valuable pointers.
I did not have time to work on that problem today (busy with other programs homeworks). I will try to work on it *soon* as in tomorrow or Friday at most.
I will post an update.
Kuphryn
|
|
|
|
|
Hi there,
I have a pure win32 dialog with textboxes and static objects on it. The code is all done without MFC.
Now, the fonts seem rather huge, so I tried to modify them. I called
createfont at the beginning of my Dialog proc with various numbers
for the height variable and use various item for parameter 5. I do
as shown below:
SelectObject(hdc, CreateFont(12,2, 0, 0, FW_THIN, 0,0,0,
DEFAULT_CHARSET, OUT_CHARACTER_PRECIS, CLIP_CHARACTER_PRECIS, DEFAULT_QUALITY,DEFAULT_PITCH | FF_DONTCARE, _T("TimesNewRoman")));
and at the end of the proc, I delete it as following:
DeleteObject(SelectObject(hdc, GetStockObject(SYSTEM_FONT)));
But nothing changes. What am I doing wrong?
|
|
|
|
|
You'll want to send a WM_SETFONT message to each control, specifying the font you wish to use.
--------
Well actually they are sort of interesting Nish, on Nudes
|
|
|
|
|
Further to that, I have also created an edit control as following:
hwndEdit = CreateWindow( TEXT("edit"), tcLabel,
WS_CHILD | WS_VISIBLE | WS_BORDER | ES_LEFT | ES_AUTOHSCROLL ,
cxChar + 100, cyChar * (2 * i ) /* * (1 + 2 * i) */, 20 * cxChar + 100,
cyChar * (1.25), hwndDlg, (HMENU) i,
((LPCREATESTRUCT) lParam)->hInstance, NULL)
However, the following catch inside my dialog's WM_COMMAND catch statement, no longer works after creating my edit box :
if (HIWORD(wParam) == EN_CHANGE ||
HIWORD(wParam) == CBN_SELCHANGE)
When I click inside the text boxes, a couple of messages are generated, but they don't correspond to "EN_CHANGE". How would I find out what those messages are within VC btw? Any idea what this problem maybe?
thanks
|
|
|
|
|
In Visual Studio .net, if you add a file to a project, then that file will appear at the end of the project, rather than appearing in alphabetical order. (VS6 alphabetizes the files.) Is there a property somewhere that can be set to fix this?
|
|
|
|
|
Isn't it just that VC7 also takes into account the case of the filename? If so, this is a known bug.
source: nntp://microsoft.public.dotnet.langages.vc
IIRC it "will be fixed in a later release" but there were no promises of an sp fix (even that it would problably just adding an 'i' in the middle of a strcmp).
|
|
|
|
|
I have noticed the same thing, but if I close my project, and opens it again, the files are sorted correctly.
- Anders
Money talks, but all mine ever says is "Goodbye!"
|
|
|
|
|
OK, then this is another bug. You should probably report it to the NG I mentioned in my post.
|
|
|
|
|
I would like to learn something about virtual desktops, but it's hard to find some articles or msdn info on it.
Is there some sort of function in Windows that makes it possible to switch desktops? As far as I know windows does support it completely, so it shouldn't be hard to code.
Got any links or information?
Cheers!
|
|
|
|
|
It's not fully supported (f.ex. creating a new desktop still gets SystemParametersInfo to return the working area of a desktop with an Explorer bar), and you can't move windows from one desktop to another like you'd expect from any other OS, but you should have a look at the Win32 SDK sample "switcher". It's using CreateDesktop & co.
Note: You [obviously] need an NT kernel based OS for this.
|
|
|
|
|
Thanks for the feedback, but I was looking for something that would work on both NT and 9x.
There are quite many apps out there that work on 95/98, so it must be possible. As the apps are fairly small I guess it doesn't take an awful lot of code to make it work.
I'll keep searching, if you got an idea please tell me.
Thanks in advance
|
|
|
|
|
In Win9x it's completely unsupported. I've seen references to apps being _really_ ugly and to "switch desktop" they just grab a handle to any other apps window and move it to e.g. "appWnd.currpos - screenWidth", but I'd say yo get no friends from that approach even that it appears to work.
Now you know how some others do it. I just hope you don't do it yourself...
And even if you do, you'd have to track e.g. ALT+TAB to either not display the apps on "another desktop", or to swith desktop when alt-tabbing to an app on "another desktop".
|
|
|
|
|
Okay, thanks.
I guess I have to do it in the dirty way, as there appears to be no proper solution.
But as the desktop is a class of some sort it should be possible to create an array of them. I wonder why it isn't an array of desktops in the first place....
|
|
|
|
|
Hello there,
This isn't so much a problem (because I've got a solution for it) but I would like to know why it has happened in the first place. I've made use of a CPropertySheet control in one of my applications, it has two tabs (CPropertyPages) on it inherited from dialogues as CPropPage1 and CPropPage2 etc., pretty standard MFC via MSDN stuff. The sheet is constructed using AddPage (I think it's called that) not member variables in the CPropertySheet, everything works just perfectly, it constructs, shows and all the controls work great.
Now, when a button on the first property page is clicked, some extended processing needs to occur so I launch a worker thread and do some work. This works great too. Recently I added some options on the second property page and needed to make use of them during the extended processing thread on the first property page - still with me?
I wrote code (quickly guestimated) something like this:
void PropPage1::ButtonClick()
{
AfxBeginThread( Thread, this );
}
UINT Thread(LPVOID pParam)
{
CPropPage1 *caller = (CPropPage1 *)pParam;
CPropPage2 *options = NULL;
CPropertySheet *parent = NULL;
if(caller)
{
parent = (CPropertySheet *)GetParent();
if(parent) options = parent->GetPage(1);
... on and on - the rest doesn't matter.
}
return( preset_errorcode_from_somewhere );
}
Does that look OK? The reason I stop here is because at the "..." I get an Access Violation from CPropertySheet::GetPage(...) - the exact violation is from a CPtrArray I believe. No matter how many variations I do on this code, the same thing always happens. I might add that GetPageCount() correctly returns 2 indicating that there are two pages available.
The EXACT same code does not crash, if it is called from the CPropPage1::ButtonClick() handler instead of the worker thread. Now I'm aware that the code isn't exactly thread safe  but should it really be crashing? I aren't aware of the behind the scenes operation, but obviously something is crossing a boundary somewhere. Why? My solution was to pass pointers to the PropertyPages via a structure into the thread and that works fine. The problem would seem to be a fault with the way GetPage(...) works because all of the other CPropertyPage members appear to function perfectly.
Just curious to know if anyone has a potential explanation for this behaviour?
Jon.
|
|
|
|
|
The problem is a very frequent one, and stems from the fact that CWnd-derived objects do funny things across threads. The explanation is a little long, please bear with me:
Given a Cwnd, its associated HWND handle is simply stored in the m_hWnd member. The other way around, obtaining the associated CWnd from a given HWND, is not that simple. There's no easy way to have Win32 store some pointer to the CWnd inside the data hold for a particualr HWND. To work around this problem, MFC guys decided to keep a global map of HWNDs to CWnd pointers. Now comes the key point: for historical and performance reasons, this HWND to CWnd map is not unique, but instead each thread has its own, stored in so called thread local storage (TLS). In your particular case, the worker thread is given a pointer to CPropPage. So far so good. But when GetParent is called, the following sequence of events happen:- Win32
::GetParent is called, returning a valid HWND.
- The associated
CWnd is sought for, and as there is none local to this thread, a mock-up one is created (and inserted into the map). This object is not the CPropertySheet you began with.
And the rest is easy. GetPage fails because it is issued on a CWnd constructed out of nothing, with no information about the original CPropertySheet (apart from being associated to the same HWND.)
Calls to methods of caller will work OK nevertheless, as this time caller is actually the real object.
The moral of this story is: don't pass CWnd pointers across threads.
Joaquín M López Muñoz
Telefónica, Investigación y Desarrollo
|
|
|
|
|
How can I make a Combo Box uneditable. ie The user will not be allowed to type characters into it, only select something from the list
|
|
|
|
|
use the CBS_DROPDOWNLIST style instead of CBS_DROPDOWN.
--------
Well actually they are sort of interesting Nish, on Nudes
|
|
|
|
|
|
Hello,
The MSDN docs don't say how to respond when the serialization tries to serialized an unsupported version (see code snippet). Any ideas?
void MyObj::Serialize(CArchive& ar)
{
if (ar.IsLoading() == TRUE) {
int nVersion = ar.GetObjectSchema();
switch (nVersion) {
case 1:
ar >> m_id;
break;
// unknown version
default:
--> WHAT TO DO HERE????
break;
}
}
else {
ar << m_id;
}
}
|
|
|
|
|
WHAT TO DO HERE????
Throw an exception?
|
|
|
|
|
MSDN doc about CArchive::Serialize features this interface:
virtual void Serialize( CArchive& ar );
throw( CMemoryException );
throw( CArchiveException );
throw( CFileException ); which means that the framework expects the method to throw any of the exceptions listed above. In your particular case, CArchiveException is the appropriate one to throw, with the m_cause member set to CArchiveException::badSchema.
Joaquín M López Muñoz
Telefónica, Investigación y Desarrollo
|
|
|
|
|
General 
News 
Suggestion 
Question 
Bug 
Answer 
Joke 
Praise 
Rant 
Admin 
Submit an article or tip
