November 2008 - Wine-devel - list.winehq.org

dead stores in msxml3
by ricardo filipe 18 Nov '08

18 Nov '08

hi leslie. i'm tracking some dead stores with Clang and there are some in dlls/msxml3/saxreader.c i'm hoping you can tell me if the stores are to be removed for performance or if they should be checked in each iteration for correctness: file dlls/msxml3/saxreader.c lines: 1163 1166 1213 1216 regards, ricardo _________________________________________________________________ Confira vídeos com notícias do NY Times, gols direto do Lance, videocassetadas e muito mais no MSN Video! http://video.msn.com/?mkt=pt-br

2 1

Re: dlls/ntdll: fixed dead store (llvm/clang)
by Austin English 18 Nov '08

18 Nov '08

On Tue, Nov 18, 2008 at 4:22 PM, ricardo filipe <ricardo_barbano(a)hotmail.com> wrote: > > > ________________________________ > Conheça já o Windows Live Spaces, o site de relacionamentos do Messenger! > Crie já o seu! > > > You forgot the patch. -- -Austin

1 0

Re: dlls/oleaut32: add tests for OleLoadPictureFile(Ex)
by Jeremy Drake 18 Nov '08

18 Nov '08

On Tue, 18 Nov 2008, Michael Karcher wrote: > Looks OK to me, but what happened to your finding about > INET_E_RESOURCE_NOT_FOUND? Was it a red herring? I saw that on XP, but I couldn't reproduce it on Windows 2000. I need to investigate this further. Is there a mechansim to do windows-version-specific tests? > > I ended up having to add OleLoadPicutreFile and OleLoadPictureFileEx to > > the .spec and .c files as actual functions (for now just returning > > E_NOTIMPL) in order to get the test to even compile. > Might be that they don't appear in the ELF export table otherwise, and > as we don't crossbuild the tests, the imports are done by the ELF > linker/loader. BTW: In general, if you add stubs like this, please add a > FIXME call showing the parameters and saying stub. How do I show the VARIANT in the FIXME call? > This fragment repeats over and over, sometimes with PICTYPE_BITMAP > instead. Might be worth refactoring that into a function. todo_wine > works for function boundaries. If you refactor, you should pass that > function __LINE__ information and print it in your OK calls to know > which of the tests is failing if something fails. > > You do a lot of tests like this. Setting disp to "(void*)0xdeadbeef" > instead before allows you to differentiate between disp unmodified and > disp explicitly set to zero. Good ideas. I'll do that.

2 1

Re: cryptdlg(2/2): Implement FormatVerisignExtension
by Alexandre Julliard 18 Nov '08

18 Nov '08

"Juan Lang" <juan.lang(a)gmail.com> writes: > + if (dwFormatStrType & CRYPT_FORMAT_STR_NO_HEX) > + { > + static BOOL stringLoaded = FALSE; > + DWORD bytesNeeded; > + > + if (!stringLoaded) > + { > + LoadStringW(hInstance, IDS_VERISIGN_STATEMENT, verisign_statement, > + sizeof(verisign_statement) / sizeof(verisign_statement[0])); > + stringLoaded = TRUE; > + } There's no need to store a copy of the string. You should load it directly into the destination buffer. -- Alexandre Julliard julliard(a)winehq.org

2 2

Another source of hardware statistics!
by Dan Kegel 18 Nov '08

18 Nov '08

Unity3d is a game engine for casual gaming (not hardcore gaming). They have some good data on their user base's hardware at http://unity3d.com/webplayer/hwstats e.g. their breakdown of video cards for this quarter is at http://unity3d.com/webplayer/hwstats/pages/web-2008Q4-gfxcard.html Their top 3 graphics cards are Intel 945 (GMA 950) 13.3% Nvidia GeForce 6100 7.5% ATI Radeon X300/X550/Xpress200/1100 6.9% And believe it or not, 8% of their users still only have 256MB of RAM... - Dan

1 0

Re: fix a crash in widl on Vista x86
by Rob Shearman 18 Nov '08

18 Nov '08

2008/11/17 <netzimme(a)aim.com>: > Changelog: > Add a FIXME for the crash in widl on Vista x86. > > diff --git a/tools/widl/typegen.c b/tools/widl/typegen.c > index fcb141e..45467e5 100644 > --- a/tools/widl/typegen.c > +++ b/tools/widl/typegen.c > @@ -3137,7 +3137,7 @@ size_t get_size_procformatstring(const > statement_list_t *stmts, type_pred_t pred > size += get_size_procformatstring(stmt->u.lib->stmts, pred) - 1; > continue; > } > - else if (stmt->type != STMT_TYPE && stmt->u.type->type != > RPC_FC_IP) > + else if (stmt->type != STMT_TYPE || stmt->u.type->type != > RPC_FC_IP) > continue; > > iface = stmt->u.type; Looks good. I had the same fix queued up in my tree. However, your changelog message needs some work. The bug isn't specific to Vista x86, but is probably specific to the IDL file you are compiling. -- Rob Shearman

1 0

NT driver address space
by Alexander Morozov 18 Nov '08

18 Nov '08

Hi, Under Wine NT drivers are running in user address space. Some drivers do not work correctly because of this. For example Guardant USB token driver crashes when loading but this problem can be fixed with such patch (really I tested with our wine fork): diff --git a/dlls/ntdll/virtual.c b/dlls/ntdll/virtual.c index 78e38dc..f607f22 100644 --- a/dlls/ntdll/virtual.c +++ b/dlls/ntdll/virtual.c @@ -904,6 +904,7 @@ static NTSTATUS map_image( HANDLE hmapping, int fd, char *base, SIZE_T total_siz server_enter_uninterrupted_section( &csVirtual, &sigset ); + if (0x10000 == base) base = 0x81000000; if (base >= (char *)0x110000) /* make sure the DOS area remains free */ status = map_view( &view, base, total_size, mask, FALSE, VPROT_COMMITTED | VPROT_READ | VPROT_EXEC | VPROT_WRITECOPY | VPROT_IMAGE ); diff --git a/libs/wine/mmap.c b/libs/wine/mmap.c index ff117cc..4c18683 100644 --- a/libs/wine/mmap.c +++ b/libs/wine/mmap.c @@ -346,7 +346,7 @@ void mmap_init(void) #if defined(__i386__) && !defined(__FreeBSD__) && !defined(__FreeBSD_kernel__) /* commented out until FreeBSD gets fixed */ char stack; char * const stack_ptr = &stack; - char *user_space_limit = (char *)0x7ffe0000; + char *user_space_limit = (char *)0xaffe0000; reserve_malloc_space( 8 * 1024 * 1024 ); I see two ways of fixing this. 1. Modify wine to run winedevice.exe in kernel address space (> 0x80000000) by checking name of executable. 2. Modify ExAllocatePool and some other functions to use kernel address space, modify functions which is used by drivers (ntoskrnl.exe exports etc.) to use only this memory allocation functions, modify winedevice.exe to load driver and create stack in kernel address space. What is better decision? Thanks, Alexander Morozov

1 0

Re: Patch for [Bug 16087] wine spews innumerable "Unable to check compatibility for Format..." errors in some games
by Henri Verbeet 18 Nov '08

18 Nov '08

2008/11/18 Stefan Dösinger <stefan(a)codeweavers.com>: > I think it would be better f getColorBits returned a proper error to the > caller, and the caller handles these errors without spamming the FIXME or > ERR channel. That would seem better to me than prefiltering the parameters > It does, it returns FALSE for "unsupported" formats. The only thing the patch fixes is the FIXME & ERR output. I think there are places where we'd like getColorBits() to work properly for float formats though.

1 0

Re: dlls/cabinet: fix dead stores (llvm/clang)
by Dmitry Timoshkov 18 Nov '08

18 Nov '08

"ricardo filipe" <ricardo_barbano(a)hotmail.com> wrote: > --- a/dlls/cabinet/fci.c > +++ b/dlls/cabinet/fci.c > @@ -462,7 +462,7 @@ static cab_ULONG fci_get_checksum(const void *pv, UINT cb, CHECKSUM seed) > case 2: > ul |= (((ULONG)(*pb++)) << 8); > case 1: > - ul |= *pb++; > + ul |= *pb; > default: > break; > } Is there any reason that you ignored 2 similar cases above? -- Dmitry.

3 2

Re: Patch for [Bug 16087] wine spews innumerable "Unable to check compatibility for Format..." errors in some games
by Henri Verbeet 18 Nov '08

18 Nov '08

> + switch(Format) > + { > + case WINED3DFMT_X8R8G8B8: > + case WINED3DFMT_R8G8B8: > + case WINED3DFMT_A8R8G8B8: > + case WINED3DFMT_A2R10G10B10: > + case WINED3DFMT_X1R5G5B5: > + case WINED3DFMT_A1R5G5B5: > + case WINED3DFMT_R5G6B5: > + case WINED3DFMT_X4R4G4B4: > + case WINED3DFMT_A4R4G4B4: > + case WINED3DFMT_R3G3B2: > + case WINED3DFMT_A8P8: > + case WINED3DFMT_P8: > + /* taken from utils.c/getColorBits() */ > + break; > + default: > + /* getColorBits doesn't work here, so avoid the silly error messages. */ > + return FALSE; > + } As you mention in the bug report, this is not a proper fix. We should probably fix the format table to use a shift and a bitcount.

1 0