CR: Fix crash in vidsite...

Synopsis
========

This change fixes several bugs filed against the site on
Cayenne as used by the windows player. The internal bug
numbers are:

193461  HTR: Crash in vidsite.dll trying to.....
193913  HTR: Crash in vidsite.dll after adding....
193949  Deaddlock playing remote SMIL with RV content
193976  HTR player crash after loading and playing....
194427  crash after.....

Each bug above seemed to die with a different stack trace.
I
believe however that they are all related. What was
happening
was that we had some timing issue going on where the
renderers
were actually blt'ing frames before the site got its first
WM_PAINT message. When the site gets its first WM_PAINT
message
it checks the resolution and bit depth of the screen to see
if
we need to re-create the DirectX surfaces. The problem comes
in
when a renderer is trying to use the site in one thread
while
we are tearing down and rebuilding the surfaces. This
results
in random crashes. Here is one such stack trace of a crash:

Thread 1:

	I420toRGB32_DBLROW2X_FAST_STRETCH2XPLUS()
  	IMAGE_STRETCH2XPLUS()
  	YUVtoRGB2(dest_format, ...)
  	I420toRGB32(dest_ptr, dest....)
  	ColorFuncAccess::ColorConvert(cidOut, dest_ptr...)
  	CBaseRootSurface::MinimalBlt(pImageData,
pBitmapInfo,....)
  	CBaseRootSurface::Blt(pImageData, pBitmapInfo,...)
  	CBaseSurface::Blt(pImageData, pBitmapInfo, rDestRect,
rSrcRect)
  	CWinSurface2::Blt(pImageData, pBitmapInfo, rDestRect,
rSrcRect)
  	CBaseSurface::OptimizedBlt(pImageBits, rDestRect,
rSrcRect)
  	CVideoRenderer::UpdateVideoSurface(pVideoSurface, ...)
  	CVideoRenderer::UpdateDisplay(pEvent, bSystemEvent,
bIsVisible)
  	CVideoRenderer::HandleEvent(pEvent)
  	CHXMultiInstanceSiteUser::HandleEvent(pEvent)
  	CHXSiteUserProxy::HandleEvent(pEvent)
  	CHXBaseSite::ForceRedraw()
  	CHXMultiInstanceSiteUserSupplier::ForceRedraw()
  	CVideoRenderer::ForceRefresh()
  	CVideoRenderer::SchedulerCallback(bIsScheduled,
bResched,....)
  	CVideoRenderer::BltIfNeeded()
  	CVideoFormat:ecodeFra
me(ulMaxExtraFrames)
  	CVideoRenderer::OnPace(ulId)
  	CVideoPaceMaker::ThreadRoutine(pArg)
  	HXWinThread::HXWinThreadStartRoutine(p)
  	_threadstartex(ptd)
  	7c80b683	

Thread 2:

	WinDraw2_IsDDAvailable(lpwd)
  	CWinBaseRootSurface::_IsHardwareAccelerationAvail()
  	CWinBaseRootSurface::_CreateCompositionSurface()
  	CBaseRootSurface::ReInitSurfaces()
  	CHXBaseSite::ReInitSurfaces()
  	CHXBaseSite::ChangeResolution()
  	CHXBaseSite::CheckDisplayMode(hdc)
  	CHXWinSite::HandleWndProc(hWnd, message, wParam, lParam,
lResult)
  	HXxWinHookSiteProc(hWnd, message, uParam, lParam)
  	77d48734	
  	77d48816	
  	77d4b4c0	
  	77d4b50c	
  	7c90eae3	
  	77d494d2	
  	77d4b530	
  	77d496c7	
  	CRCAEventLoopWin:oStep(th
isRunCount, bHandleQuitMsg)
  	CRCAEventLoopWin::Run()
  	CRPApp::RunRPApp()
  	WinMain(hInstance, hPrevInstance, lpCmdLine, nCmdShow)
  	WinMainCRTStartup()
  	7c816fd7	

Fix
===

The fix is simply to call CheckDisplayMode(NULL) earlier
then the first
WM_PAINT message. I ended up doing this in
BeginOptimizedBlt(). Each
renderer that will use the video surface will call this
method before
doing so. This fixes the problem by initialing the values
CheckDisplayMode() uses to decide if it will re-init the
surfaces:

             _GetDeviceCaps(hdc, newBitsPerPixel, newHozRes,
newVertRes);

             if (m_nOldBitsPerPixel != newBitsPerPixel ||
newHozRes != m_nOldHorzRes || newVertRes != m_nOldVertRes)
             {
                 m_nOldBitsPerPixel  = newBitsPerPixel;
                 m_nOldHorzRes           = newHozRes;
                 m_nOldVertRes           = newVertRes;
                 ChangeResolution();
             }

I also cleaned up a couple casting warnings....

Files Modified
==============
video/sitelib/basesurf.cpp

Branch(s)
=========
HEAD, 150Cay, 204Cay.

Platforms and Profiles Functionality verified
=============================================
windows/all-defines. Linux all defines.

QA Instructions
===============
I tested this on linux and windows. However, this CR does
change when surfaces are created (root's composition
surface). So, testing on the Mac is needed. I don't see
any problems but we should make sure.

--greg.

Index: basesurf.cpp
============================================================
=======
RCS file: /cvsroot/video/sitelib/basesurf.cpp,v
retrieving revision 1.18.4.7
diff -u -w -r1.18.4.7 basesurf.cpp
--- basesurf.cpp        24 Jan 2007 19:20:29 -0000     
1.18.4.7
+++ basesurf.cpp        9 May 2007 18:18:35 -0000
 -203,7
+203,7 

      IHXPreferences*    pPreferences    = NULL;
      IHXBuffer*         pBuffer         = NULL;
-    HXFLOAT         fTmp         = m_fScrollBarZoom;
+    HXFLOAT         fTmp         =
(float)m_fScrollBarZoom;
      HXBOOL          bYUY2First   = FALSE;

      ReadPrefBOOL(m_pContext,
"AllowOverlayShrinking", m_bAllowOverlayShrinking
);
 -218,7
+218,7 

m_pyuvInputMngr->SetOutputPriority(CID_I420, list, 4);
              }
      ReadPrefFLOAT(m_pContext, "SiteZoomFactor",
fTmp );
-    m_fScrollBarZoom = fTmp;
+    m_fScrollBarZoom = (double)fTmp;

m_pContext->QueryInterface(IID_IHXOverlayManager,
(void**)&m_pOverlayManager);

 -3991,6
+3991,7 
          return HXR_FAIL;
      }

+
      // Are the new parameters the same as the old
paramaters? If not
      // then delete the old surfaces.
      if (m_pOptimizedFormat && pBitmapInfo)
 -4007,6
+4008,12 

      m_pSite->_TLSLock();

+    //XXXgfw fixes several bugs. For one such bug and a
full
+    //description of why we need to check this here see
internal
+    //bug number 193976 or the commit mail for this
change.
+   
m_pSite->m_pTopLevelSite->CheckDisplayMode(NULL);

      HX_DELETE(m_pOptimizedFormat);
      m_pOptimizedFormat = new HXBitmapInfoHeader;

_______________________________________________
Video-dev mailing list
Video-devhelixcommunity.org
http://lists.helixcommunity.org/mailman/listinfo/video
-dev

CR: Fix crash in vidsite...

Thread: CR: Fix crash in vidsite...