https://bugs.kde.org/show_bug.cgi?id=380119
Ralf Habacker <ralf.habac...@freenet.de> changed:
What |Removed |Added
----------------------------------------------------------------------------
Status|UNCONFIRMED |RESOLVED
Resolution|--- |FIXED
--- Comment #1 from Ralf Habacker <ralf.habac...@freenet.de> ---
For the record:
The crash happens on returning from GWEN_Gui_ReadDialogPrefs.
GWEN_Gui_ReadDialogPrefs:
0x62d944d0 <+0>: mov 0x62dbb1bc,%eax
0x62d944d5 <+5>: test %eax,%eax
0x62d944d7 <+7>: je 0x62d94510 <GWEN_Gui_ReadDialogPrefs+64>
0x62d944d9 <+9>: mov 0x54(%eax),%edx
0x62d944dc <+12>: test %edx,%edx
0x62d944de <+14>: je 0x62d94510 <GWEN_Gui_ReadDialogPrefs+64>
0x62d944e0 <+16>: sub $0x1c,%esp
0x62d944e3 <+19>: mov 0x28(%esp),%ecx
0x62d944e7 <+23>: mov %eax,(%esp)
0x62d944ea <+26>: mov %ecx,0xc(%esp)
0x62d944ee <+30>: mov 0x24(%esp),%ecx
0x62d944f2 <+34>: mov %ecx,0x8(%esp)
0x62d944f6 <+38>: mov 0x20(%esp),%ecx
0x62d944fa <+42>: mov %ecx,0x4(%esp)
0x62d944fe <+46>: call *%edx --> calls AB_Gui_ReadDialogPrefs
0x62d94500 <+48>: sub $0x10,%esp --> clean up stack of called function
indicates stdcall
0x62d94503 <+51>: add $0x1c,%esp
0x62d94506 <+54>: ret --> crash happens here
0x62d94507 <+55>: mov %esi,%esi
0x62d94509 <+57>: lea 0x0(%edi,%eiz,1),%edi
0x62d94510 <+64>: mov $0xffffffbd,%eax
0x62d94515 <+69>: ret
It turned out that the call back functions in gwenhywfar were declared with
stdcall (calling function cleans up the stack) while aqbanking is compiled with
cdecl (stack is cleaned up inside the called function). Therefore on returning
from GWEN_Gui_ReadDialogPrefs the stack pointer points to a wrong address.
The fix is to remove stdcall from the call back function with the following
patch applied to the gwenhywfar package at
https://build.opensuse.org/package/show/home:rhabacker:branches:windows:mingw:win32:kmymoney/mingw32-gwenhywfar
--- a/src/gwenhywfarapi.h 2017-05-23 07:38:51.911107998 +0000
+++ b/src/gwenhywfarapi.h 2017-05-23 07:38:57.319127655 +0000
@@ -84,7 +84,7 @@
#endif
#if GWENHYWFAR_SYS_IS_WINDOWS
-# define GWENHYWFAR_CB __stdcall
+# define GWENHYWFAR_CB
#else
# define GWENHYWFAR_CB
#endif
A recompile with the patch applied indeed shows that the call to "sub
$0x10,%es" has been removed.
Dump of assembler code for function GWEN_Gui_ReadDialogPrefs:
0x62d943a0 <+0>: mov 0x62dbb1bc,%eax
0x62d943a5 <+5>: test %eax,%eax
0x62d943a7 <+7>: je 0x62d943d4 <GWEN_Gui_ReadDialogPrefs+52>
0x62d943a9 <+9>: mov 0x54(%eax),%edx
0x62d943ac <+12>: test %edx,%edx
0x62d943ae <+14>: je 0x62d943d4 <GWEN_Gui_ReadDialogPrefs+52>
0x62d943b0 <+16>: sub $0x1c,%esp
0x62d943b3 <+19>: mov 0x28(%esp),%ecx
0x62d943b7 <+23>: mov %eax,(%esp)
0x62d943ba <+26>: mov %ecx,0xc(%esp)
0x62d943be <+30>: mov 0x24(%esp),%ecx
0x62d943c2 <+34>: mov %ecx,0x8(%esp)
0x62d943c6 <+38>: mov 0x20(%esp),%ecx
0x62d943ca <+42>: mov %ecx,0x4(%esp)
0x62d943ce <+46>: call *%edx
0x62d943d0 <+48>: add $0x1c,%esp
0x62d943d3 <+51>: ret
0x62d943d4 <+52>: mov $0xffffffbd,%eax
0x62d943d9 <+57>: ret
--
You are receiving this mail because:
You are the assignee for the bug.
