[Bug 37892] GOTHIC 2 GOLD: Vdfs32e.exe crashes at startup

[wine-bugs at winehq.org]

https://bugs.winehq.org/show_bug.cgi?id=37892

--- Comment #4 from Marc Bessières <marc.bessieres at gmail.com> ---
Thank you both for looking into this bug.
Sorry for the delay in answering.

I checked again and here is what I find:
1.7.33 works
1.7.34 fails
1f7683777babab98197c39e5965ba6f70c01c8d0 works
ca51e113e4820f8b11016c13732b1a971d2b0054 fails

1f7683777babab98197c39e5965ba6f70c01c8d0 is the commit just before
ca51e113e4820f8b11016c13732b1a971d2b0054

Sorry for the complicated regression procedure, but after finding
ca51e113e4820f8b11016c13732b1a971d2b0054 and trying to revert it on top of
1.7.34 Vdfs32e.exe didn't work.
So I tried bisecting again and again, each reverting the previously found "bad"
commit(s).

As I see that Bela doesn't manage to make it work ever, I would agree with
Sebastian on the fact that it might be related to some corruption, luckily or
unluckily it is just repeatedly working here with the same commit, and failing
after the same one.

This gave me the idea to give valgrind a try:
export VALGRIND_OPTS="-q --trace-children=yes --track-origins=yes
--gen-suppressions=all --leak-check=full --num-callers=20
--workaround-gcc296-bugs=yes
--vex-iropt-register-updates=allregs-at-mem-access"

I started winefile without valgrind to start the wineserver then I ran:
valgring ./wine $WINEPREFIX//drive_c/Program\ Files/JoWooD\ Productions\
Software\ AG/Gothic\ II\ Gold/system/Vdfs32e.exe

And within the list of reports there is one that may be related, as part of it
(loader.c:2870) corresponds to part of the stack trace of the crash (I couldn't
find it in Austin English suppression files mentionned in the wiki)
==2388== Conditional jump or move depends on uninitialised value(s)
==2388==    at 0x4016C2: ???
==2388==    by 0x4B91B9B: ??? (in
/home/guest/wine-git/dlls/kernel32/kernel32.dll.so)
==2388==    by 0x4B92C42: start_process (process.c:1104)
==2388==    by 0x4871ADF: ??? (in /home/guest/wine-git/dlls/ntdll/ntdll.dll.so)
==2388==    by 0x4874C0C: call_thread_func (signal_i386.c:2723)
==2388==    by 0x4871ABD: ??? (in /home/guest/wine-git/dlls/ntdll/ntdll.dll.so)
==2388==    by 0x484675D: start_process (loader.c:2870)
==2388==    by 0x403FB9C: ??? (in
/home/guest/wine-git/libs/wine/libwine.so.1.0)
==2388==  Uninitialised value was created by a stack allocation
==2388==    at 0x401216: ???
==2388== 
{
   <insert_a_suppression_name_here>
   Memcheck:Cond
   obj:*
   obj:/home/guest/wine-git/dlls/kernel32/kernel32.dll.so
   fun:start_process
   obj:/home/guest/wine-git/dlls/ntdll/ntdll.dll.so
   fun:call_thread_func
   obj:/home/guest/wine-git/dlls/ntdll/ntdll.dll.so
   fun:start_process
   obj:/home/guest/wine-git/libs/wine/libwine.so.1.0
}

I just don't know why valgrind doesn't manage to decode some addresses,
especially the stack allocation one...
I may ask some valgrind people at FOSDEM.

While I was trying to launch Vdfs32e.exe several times to write that entry,
Vdfs32e.exe worked once.
So this is definitely not a regression.

Could someone with the right rights in Bugzilla remove the keyword and the sha1
entry?
Also at the same time may be put it also in NEW as Bela confirmed the failure
of Vdfs32e.exe

-- 
Do not reply to this email, post in Bugzilla using the
above URL to reply.
You are receiving this mail because:
You are watching all bug changes.