Thanks for this.  But if you want to recreate my problem then my zipped 
thread in my sample app.  Any ideas?
I couldn't recreate this problem.
What I did was create a CWinThread class, in it's InitInstance, I called 
Sleep(5000); and then AfxMessageBox("Box1");
Then from the CWinApp::InitInstance, I created the thread, called 
Sleep(20000);, and then AfxMessgaeBox("Box2);
Box1 showed up after ~ 5 seconds, and Box2 after 20
I personally would have used waitable timers in the thread instead of 
sleep.
#pragma once
#include <afxtempl.h>
typedef CMap<HANDLE,HANDLE,HANDLE,HANDLE> TimerMap;
// CDialogThread
class CDialogThread : public CWinThread
{
DECLARE_DYNCREATE(CDialogThread)
protected:
CDialogThread();           // protected constructor used by dynamic 
creation
virtual ~CDialogThread();
public:
virtual BOOL InitInstance();
virtual int ExitInstance();
protected:
  virtual BOOL OnIdle(LONG lCount);
DECLARE_MESSAGE_MAP()
private:
  HANDLE SetTimer(LONGLONG Milliseconds);
  void KillTimer(HANDLE &Handle);
  void OnTimer(HANDLE Handle);
  HANDLE               m_TimerHandle;
  TimerMap             m_Timers;
};
// DialogThread.cpp : implementation file
//
#include "stdafx.h"
#include "DelayedStartTest.h"
#include "DialogThread.h"
// CDialogThread
IMPLEMENT_DYNCREATE(CDialogThread, CWinThread)
CDialogThread::CDialogThread()
{
}
CDialogThread::~CDialogThread()
{
}
BOOL CDialogThread::InitInstance()
{
  m_TimerHandle = SetTimer(5000);
  return TRUE;
}
int CDialogThread::ExitInstance()
{
// TODO:  perform any per-thread cleanup here
return CWinThread::ExitInstance();
}
BEGIN_MESSAGE_MAP(CDialogThread, CWinThread)
END_MESSAGE_MAP()
// CDialogThread message handlers
HANDLE CDialogThread::SetTimer(LONGLONG Milliseconds)
{
  HANDLE Handle = CreateWaitableTimer(NULL,TRUE,NULL);
  LARGE_INTEGER liDueTime;
  liDueTime.QuadPart= Milliseconds * -10000;
  if (!SetWaitableTimer(Handle,&liDueTime,0,NULL,NULL,TRUE))
  {
     CloseHandle(Handle);
     ASSERT(FALSE);
     return NULL;
  }
  m_Timers.SetAt(Handle,Handle);
  return Handle;
}
void CDialogThread::KillTimer(HANDLE &Handle)
{
  if (Handle != NULL)
  {
     CancelWaitableTimer(Handle);
     CloseHandle(Handle);
     m_Timers.RemoveKey(Handle);
     Handle = NULL;
  }
}
BOOL CDialogThread::OnIdle(LONG lCount)
{
  HANDLE *pTimers = new HANDLE[m_Timers.GetCount()];
  HANDLE Handle;
  POSITION Pos = m_Timers.GetStartPosition();
  int i = 0;
  while (Pos)
  {
     m_Timers.GetNextAssoc(Pos,Handle,Handle);
     pTimers[i++] = Handle;
  }
  DWORD Ret = 
WaitForMultipleObjects((DWORD)m_Timers.GetCount(),pTimers,FALSE,0);
  DWORD Err = GetLastError();
  if (Ret != WAIT_TIMEOUT && 
m_Timers.Lookup(pTimers[Ret-WAIT_OBJECT_0],Handle))
  {
     OnTimer(Handle);
  }
  delete [] pTimers;
  CWinThread::OnIdle(lCount);
  return TRUE;
}
void CDialogThread::OnTimer(HANDLE Handle)
{
  if (m_TimerHandle == Handle)
  {
     KillTimer(m_TimerHandle);
     AfxMessageBox("The wait dialog in the thread");
  }
}
AliR.
"Peter Boulton" <peter@data*nospam*perceptions.co.uk> wrote in message 
news:fpkndt$6tg$2$830fa17d@news.demon.co.uk...
I am trying to develop a small window which can appear after a few seconds 
delay if my app needs to wait to open a sql server recordset because some 
of the records it needs are exclusively locked by another user.
In this situation the ado code blocks till the appointed connection 
timeout period and then throws an exception. Rather than having my users 
stare at an hourglass till this happens, I want a message window to 
appear telling them it's waiting to access the data.  On the other hand, 
if I make it appear every time and the response from the database is 
instant then I don't want to show it - it will just flash momentarily and 
disappear, which is ugly.
So my idea is this:
- Launch a user interface thread.  The thread goes to sleep for, say, 5 
seconds.
- Open the recordset
- When the recordset 'opens' we then kill the user interface thread. 
(Therefore if the open takes more than 5 secs the user sees the message 
window, otherwise the thread which creates it gets killed before its 
window is displayed.)
The problem I'm seeing is that where the main thread blocks waiting for 
the recordset to open the user interface thread seems to also block, so 
the window never appears.
I'm starting the new thread with:
m_pMyLongProcess = 
(CLongProcessThread*)AfxBeginThread(RUNTIME_CLASS(CLongProcessThread), 
THREAD_PRIORITY_NORMAL, 0, CREATE_SUSPENDED);
Then I'm initialising some member variables of m_pMyLongProcess and 
calling m_pMyLongProcess->ResumeThread(); to get the thread running.
In my test code I simulate blocking with:
StartNewThread();    // This fn contains AfxBeginThread, the member 
initialisations for the thread class, and ResumeThread()
Sleep(10000);
AfxMessageBox("Awake!");
(The AfxMessageBox serves no function other than to alert me that the 
main thread has finished sleeping.  The user interface thread window 
appears simultaneously with the message box.)
On this code, the window launched in the user-interface thread appears 
after the Sleep(10000) period rather than the 5 seconds (or whatever) 
Sleep() in my user-interface thread.  Which seems wrong - surely the 
Sleep() should apply to the thread and not the process?  (I hope I've 
explained this clearly!  I fear I'm missing something fairly fundamental 
as I try to avoid threads as much as possible!)  For anyone interested 
enough, you can download my full test project at:
http://www.dataper.demon.co.uk/misc/LongProcessWndExample.zip
It's a small 135k zip.  The thread is launched from the document.  The 
code for the user interface thread is in LongProcessThread.cpp/h.  There 
is a main menu option to mess around with dynamically changing the text 
on the window.
I'm looking for enlightenment!  Thanks.
Pete