discusion about a message loop in listview.c

[Dietrich Teickner]

I have some weeks before reported, FlashFXP v3.02 loops with a 
message-loop in listview.c after login. It does this in Odin and in Wine.
In Odin it stops in smaller time (deep 1000). Wine has a bigger stack, 
and so needs wine more time for this, bat I can see in the loog, the 
args for the function are more and more locatet at lower stack adresses.
I have written this workarout for testing this problem in wine and in 
odin. with this FlashFXH 3.02 will display the contence of the server 
dir very fast at both environment. In the log I can see, that the 
message-loop was reported and stopped. I thing, we need a flag inside 
any  item to report every function, that can called recursive, this item 
is in work (function-separart, or global ?), that can stop recursions of 
this type in the future or many programms. I have also found this try 
does also helps Agent 2.0 in the groups->defailt propertys (only 
odin-tested).
This does also every crashs in earlyer builds with stack overflow in odin.

Dietrich

Index: wine/dlls/comctl32/listview.c
===================================================================
RCS file: /home/wine/wine/dlls/comctl32/listview.c,v
retrieving revision 1.406
diff -u -r1.406 listview.c
--- wine/dlls/comctl32/listview.c	25 Mar 2005 20:49:00 -0000	1.406
+++ wine/dlls/comctl32/listview.c	29 Mar 2005 16:31:48 -0000
@@ -139,6 +139,7 @@
   *   WM_CREATE does not issue WM_QUERYUISTATE and associated registry
   *     processing for "USEDOUBLECLICKTIME".
   */
+#define __T__

  #include "config.h"
  #include "wine/port.h"
@@ -7050,10 +7051,23 @@
   *   SUCCESS : TRUE
   *   FAILURE : FALSE
   */
+#ifdef __T__
+static int zIndex = -2;
+typedef struct {
+     PVOID lTol;
+     HWND cHwnd;
+     HWND cHwndParent;
+     INT cItem;
+     } SetItemStateLink, *PSetItemStateLink;
+#endif
  static BOOL LISTVIEW_SetItemState(LISTVIEW_INFO *infoPtr, INT nItem, 
const LVITEMW *lpLVItem)
  {
      BOOL bResult = TRUE;
      LVITEMW lvItem;
+#ifdef __T__
+    int saveLastError = GetLastError();
+    SetItemStateLink cTol = { NULL, infoPtr->hwndSelf, 
infoPtr->hwndNotify, nItem};
+#endif

      lvItem.iItem = nItem;
      lvItem.iSubItem = 0;
@@ -7062,6 +7076,34 @@
      lvItem.stateMask = lpLVItem->stateMask;
      TRACE("lvItem=%s\n", debuglvitem_t(&lvItem, TRUE));

+#ifdef __T__
+    if (-2 == zIndex)
+    {
+       zIndex = TlsAlloc();
+       SetLastError(saveLastError);
+    } /* endif */
+    if (zIndex >= 0)
+    {
+       cTol.lTol = TlsGetValue(zIndex);
+       if (cTol.lTol < (PVOID)&cTol)
+       {
+          cTol.lTol = NULL;
+       } /* endif */
+       SetLastError(saveLastError);
+       if (cTol.lTol
+       &&  (((PSetItemStateLink)cTol.lTol)->cHwnd == infoPtr->hwndSelf)
+       &&  (((PSetItemStateLink)cTol.lTol)->cHwndParent == 
infoPtr->hwndNotify)
+       &&  (((PSetItemStateLink)cTol.lTol)->cItem == nItem))
+       {
+         TRACE_(listview)("(DT) Hwnd:%x,%x Parent:%x,%x Item:%d,%d, 
stop the recursion\n", infoPtr->hwndSelf, 
((PSetItemStateLink)cTol.lTol)->cHwnd,
+                infoPtr->hwndNotify, 
((PSetItemStateLink)cTol.lTol)->cHwndParent,
+                nItem, ((PSetItemStateLink)cTol.lTol)->cItem);
+         return TRUE;
+       }
+       TlsSetValue(zIndex,(LPVOID)&cTol);
+       SetLastError(saveLastError);
+    }
+#endif
      if (nItem == -1)
      {
      	/* apply to all items */
@@ -7070,6 +7112,14 @@
      }
      else
  	bResult = LISTVIEW_SetItemT(infoPtr, &lvItem, TRUE);
+#ifdef __T__
+    if (zIndex >= 0)
+    {
+       saveLastError = GetLastError();
+       TlsSetValue(zIndex,(LPVOID)cTol.lTol);
+       SetLastError(saveLastError);
+    }
+#endif

      /*
       *update selection mark

[Dimitrie O. Paun]

On Tue, Mar 29, 2005 at 06:52:07PM +0200, Dietrich Teickner wrote:
> I have some weeks before reported, FlashFXP v3.02 loops with a 
> message-loop in listview.c after login. It does this in Odin and in Wine.
> In Odin it stops in smaller time (deep 1000). Wine has a bigger stack, 
> and so needs wine more time for this, bat I can see in the loog, the 
> args for the function are more and more locatet at lower stack adresses.
> I have written this workarout for testing this problem in wine and in 
> odin. with this FlashFXH 3.02 will display the contence of the server 
> dir very fast at both environment. In the log I can see, that the 
> message-loop was reported and stopped. I thing, we need a flag inside 
> any  item to report every function, that can called recursive, this item 
> is in work (function-separart, or global ?), that can stop recursions of 
> this type in the future or many programms. I have also found this try 
> does also helps Agent 2.0 in the groups->defailt propertys (only 
> odin-tested).
> This does also every crashs in earlyer builds with stack overflow in odin.

Dietrich,

You need to create a small example app that exhibits this behaviour
such that it crashes on Wine/Odin, but runs in Windows. As things
stand, I can not test with either app since they don't install under
Wine. In any event, a small example app would be better, since the
source would be available too.

-- 
Dimi.