February 2016 - Wine-devel - list.winehq.org

Re: [v3 PATCH] gdi32: Support Begin/End Path for metafile DCs
by Huw Davies 16 Feb '16

16 Feb '16

On Tue, Feb 16, 2016 at 09:28:06PM +1100, Alistair Leslie-Hughes wrote: > From: Dmitry Timoshkov <dmitry(a)baikal.ru> > > Correct failed test. > Updated Author/signoffs. > v3 - Removed Dmitry signoff on request. > > Fixes https://bugs.winehq.org/show_bug.cgi?id=39185 > > Signed-off-by: Alistair Leslie-Hughes <leslie_alistair(a)hotmail.com> > --- > dlls/gdi32/enhmfdrv/dc.c | 4 +++- > dlls/gdi32/path.c | 20 +++++++++++++++++++- > dlls/gdi32/tests/metafile.c | 2 +- > 3 files changed, 23 insertions(+), 3 deletions(-) > > diff --git a/dlls/gdi32/path.c b/dlls/gdi32/path.c > index e09cd0b..166b6b2 100644 > --- a/dlls/gdi32/path.c > +++ b/dlls/gdi32/path.c ... > @@ -1513,6 +1523,14 @@ static BOOL pathdrv_ExtTextOut( PHYSDEV dev, INT x, INT y, UINT flags, const REC > struct path_physdev *physdev = get_path_physdev( dev ); > unsigned int idx, ggo_flags = GGO_NATIVE; > POINT offset = {0, 0}; > + DWORD type; > + > + type = GetObjectType(dev->hdc); > + if (type == OBJ_METADC || type == OBJ_ENHMETADC) > + { > + PHYSDEV next = GET_NEXT_PHYSDEV( dev, pExtTextOut ); > + return next->funcs->pExtTextOut( next, x, y, flags, lprc, str, count, dx ); > + } > > if (!count) return TRUE; > if (flags & ETO_GLYPH_INDEX) ggo_flags |= GGO_GLYPH_INDEX; Why the special treatment for ExtTextOut()? I can believe we may need to do something with Begin/EndPath, but why this? Dmitry, is this the reason you don't want to send it? Huw.

2 6

Re: [PATCH 5/5] ntdll: Add special handling for int $0x2d exceptions.
by Sebastian Lackner 16 Feb '16

16 Feb '16

Thanks for working on this. I assume you already saw the new test failures, those should be fixed of course. Besides that I have some more remarks on the patch: On 12.02.2016 18:52, Henri Verbeet wrote: > Signed-off-by: Henri Verbeet <hverbeet(a)codeweavers.com> > --- > dlls/ntdll/signal_i386.c | 14 +++++++++++++- > dlls/ntdll/tests/exception.c | 38 +++++++++++++++++++++++--------------- > 2 files changed, 36 insertions(+), 16 deletions(-) > > diff --git a/dlls/ntdll/signal_i386.c b/dlls/ntdll/signal_i386.c > index a3abbff..46f932a 100644 > --- a/dlls/ntdll/signal_i386.c > +++ b/dlls/ntdll/signal_i386.c > @@ -2072,8 +2072,20 @@ static void segv_handler( int signal, siginfo_t *siginfo, void *sigcontext ) > case TRAP_x86_PROTFLT: /* General protection fault */ > case TRAP_x86_UNKNOWN: /* Unknown fault code */ > { > + CONTEXT *win_context = get_exception_context( rec ); > WORD err = get_error_code(context); > - if (!err && (rec->ExceptionCode = is_privileged_instr( get_exception_context(rec) ))) break; > + if (!err && (rec->ExceptionCode = is_privileged_instr( win_context ))) break; > + if ((err & 7) == 2 && (err >> 3) == 0x2d) /* int $0x2d */ > + { Int 0x2d is not just a special way to raise a breakpoint, its actually a kernel interface to send commands to the debugger. EAX contains the command value, other registers contain the arguments. An proper implementation is not necessary I guess, but at least for small values of EAX which are assigned to commands we would need better test coverage. Most anti- debugger tricks set EAX = 0 which is not tested at all so far. > + win_context->Eip += 3; > + rec->ExceptionCode = EXCEPTION_BREAKPOINT; > + rec->ExceptionAddress = (char *)win_context->Eip; > + rec->NumberParameters = is_wow64 ? 1 : 3; > + rec->ExceptionInformation[0] = EAX_sig(context); > + rec->ExceptionInformation[1] = ECX_sig(context); > + rec->ExceptionInformation[2] = EDX_sig(context); > + break; > + } > rec->ExceptionCode = EXCEPTION_ACCESS_VIOLATION; > rec->NumberParameters = 2; > rec->ExceptionInformation[0] = 0; > diff --git a/dlls/ntdll/tests/exception.c b/dlls/ntdll/tests/exception.c > index d6a610e..57b201a 100644 > --- a/dlls/ntdll/tests/exception.c > +++ b/dlls/ntdll/tests/exception.c > @@ -219,6 +219,11 @@ static const struct exception > > { { 0xf1, 0x90, 0xc3 }, /* icebp; nop; ret */ > 1, 1, FALSE, STATUS_SINGLE_STEP, 0 }, > + { { 0xb8, 0xb8, 0xb8, 0xb8, 0xb8, /* mov $0xb8b8b8b8, %eax */ > + 0xb9, 0xb9, 0xb9, 0xb9, 0xb9, /* mov $0xb9b9b9b9, %ecx */ > + 0xba, 0xba, 0xba, 0xba, 0xba, /* mov $0xbabababa, %edx */ > + 0xcd, 0x2d, 0xc3 }, /* int $0x2d; ret */ > + 17, 0, FALSE, STATUS_BREAKPOINT, 3, { 0xb8b8b8b8, 0xb9b9b9b9, 0xbabababa } }, > }; Could you maybe also provide a testcase when a debugger is attached? I would expect that it behaves different then. Actually it should not even produce an exception then because the exception is transparently translated into a debug event, like wine already does for debug strings for example. > > static int got_exception; > @@ -473,7 +478,7 @@ static DWORD handler( EXCEPTION_RECORD *rec, EXCEPTION_REGISTRATION_RECORD *fram > CONTEXT *context, EXCEPTION_REGISTRATION_RECORD **dispatcher ) > { > const struct exception *except = *(const struct exception **)(frame + 1); > - unsigned int i, entry = except - exceptions; > + unsigned int i, parameter_count, entry = except - exceptions; > > got_exception++; > trace( "exception %u: %x flags:%x addr:%p\n", > @@ -482,20 +487,23 @@ static DWORD handler( EXCEPTION_RECORD *rec, EXCEPTION_REGISTRATION_RECORD *fram > ok( rec->ExceptionCode == except->status || > (except->alt_status != 0 && rec->ExceptionCode == except->alt_status), > "%u: Wrong exception code %x/%x\n", entry, rec->ExceptionCode, except->status ); > - ok( rec->ExceptionAddress == (char*)code_mem + except->offset, > - "%u: Wrong exception address %p/%p\n", entry, > - rec->ExceptionAddress, (char*)code_mem + except->offset ); > - > - if (except->alt_status == 0 || rec->ExceptionCode != except->alt_status) > - { > - ok( rec->NumberParameters == except->nb_params, > - "%u: Wrong number of parameters %u/%u\n", entry, rec->NumberParameters, except->nb_params ); > - } > + ok( context->Eip == (DWORD_PTR)code_mem + except->offset, > + "%u: Unexpected eip %#x/%#lx\n", entry, > + context->Eip, (DWORD_PTR)code_mem + except->offset ); > + ok( rec->ExceptionAddress == (char*)context->Eip || > + (rec->ExceptionCode == STATUS_BREAKPOINT && rec->ExceptionAddress == (char*)context->Eip + 1), I would prefer an exact test for ExceptionAddress here, off-by-one errors in exception handling would be kinda odd. > + "%u: Unexpected exception address %p/%p\n", entry, > + rec->ExceptionAddress, (char*)context->Eip ); > + > + if (except->status == STATUS_BREAKPOINT && is_wow64) > + parameter_count = 1; > + else if (except->alt_status == 0 || rec->ExceptionCode != except->alt_status) > + parameter_count = except->nb_params; > else > - { > - ok( rec->NumberParameters == except->alt_nb_params, > - "%u: Wrong number of parameters %u/%u\n", entry, rec->NumberParameters, except->nb_params ); > - } > + parameter_count = except->alt_nb_params; > + > + ok( rec->NumberParameters == parameter_count, > + "%u: Unexpected parameter count %u/%u\n", entry, rec->NumberParameters, parameter_count ); > > /* Most CPUs (except Intel Core apparently) report a segment limit violation */ > /* instead of page faults for accesses beyond 0xffffffff */ > @@ -530,7 +538,7 @@ static DWORD handler( EXCEPTION_RECORD *rec, EXCEPTION_REGISTRATION_RECORD *fram > > skip_params: > /* don't handle exception if it's not the address we expected */ > - if (rec->ExceptionAddress != (char*)code_mem + except->offset) return ExceptionContinueSearch; > + if (context->Eip != (DWORD_PTR)code_mem + except->offset) return ExceptionContinueSearch; > > context->Eip += except->length; > return ExceptionContinueExecution; >

2 3

Re: advapi32/tests: Use todo_wine_if() in tests
by Marvin 16 Feb '16

16 Feb '16

Hi, While running your changed tests on Windows, I think I found new failures. Being a bot and all I'm not very good at pattern recognition, so I might be wrong, but could you please double-check? Full results can be found at https://testbot.winehq.org/JobDetails.pl?Key=20532 Your paranoid android. === w1064 (32 bit security) === security.c:3753: Test failed: MACHINE\Software owner SID (S-1-5-18) != Administrators SID (S-1-5-32-544). security.c:3794: Test failed: Builtin Users ACE has unexpected flags (0x12 != 0xa) security.c:3798: Test failed: Builtin Users ACE has unexpected mask (0x20019 != 0x80000000) security.c:3807: Test failed: Builtin Admins ACE has unexpected flags (0x12 != 0x0)

1 0

Re: user32: Select text control content while setting initial dialog focus
by Dmitry Timoshkov 16 Feb '16

16 Feb '16

Piotr Caban <piotr(a)codeweavers.com> wrote: > --- a/dlls/user32/dialog.c > +++ b/dlls/user32/dialog.c > @@ -707,7 +707,11 @@ static HWND DIALOG_CreateIndirect( HINSTANCE hInst, LPCVOID dlgTemplate, > focus = GetNextDlgTabItem( hwnd, 0, FALSE ); > if (!focus) focus = GetNextDlgGroupItem( hwnd, 0, FALSE ); > if (focus) > + { > + if (SendMessageW( focus, WM_GETDLGCODE, 0, 0 ) & DLGC_HASSETSEL) > + SendMessageW( focus, EM_SETSEL, 0, -1 ); > SetFocus( focus ); > + } > } Selecting the text on SetFocus() is the job of DefDlgProc(), it's worth to investigate why it doesn't work. -- Dmitry.

2 4

Re: [PATCH] mscoree: Implement ICorRuntimeHost::CreateDomain{, Ex}(). (try 3)
by Vincent Povirk 16 Feb '16

16 Feb '16

Sadly, we can't add any tests that use Mono, because it doesn't work in 64-bit Wine.

2 1

Re: [v2 PATCH] gdi32: Support Begin/End Path for metafile DCs
by Dmitry Timoshkov 16 Feb '16

16 Feb '16

Alistair Leslie-Hughes <leslie_alistair(a)hotmail.com> wrote: > From: Dmitry Timoshkov <dmitry(a)baikal.ru> > > Correct failed test. > Updated Author/signoffs. > > Fixes https://bugs.winehq.org/show_bug.cgi?id=39185 > > Signed-off-by: Dmitry Timoshkov <dmitry(a)baikal.ru> > Signed-off-by: Alistair Leslie-Hughes <leslie_alistair(a)hotmail.com> Please resend the patch without my sign-off. -- Dmitry.

1 0

Re: [PATCH] gdi32: Support Begin/End Path for metafile DCs
by Dmitry Timoshkov 16 Feb '16

16 Feb '16

Hi Alistair, Alistair Leslie-Hughes <leslie_alistair(a)hotmail.com> wrote: > Based off Dmitry Timoshkov patch. > > Fixes https://bugs.winehq.org/show_bug.cgi?id=39185 It's an exact copy of my patch with just one hunk removed (without which the tests won't pass under Wine). So it's not just "based", it's a plain copy. It would be fine if you sent it in two parts: first under my name (but in that case you need at least my sign-off), and next patch with your test case. -- Dmitry.

1 1

Re: [PATCH 6/7] ntdll: Add special handling for int $0x2d exceptions.
by Sebastian Lackner 16 Feb '16

16 Feb '16

On 15.02.2016 18:18, Henri Verbeet wrote: > Signed-off-by: Henri Verbeet <hverbeet(a)codeweavers.com> > --- > dlls/ntdll/signal_i386.c | 29 ++++++++++++++++++++++++++++- > dlls/ntdll/tests/exception.c | 38 +++++++++++++++++++++++--------------- > 2 files changed, 51 insertions(+), 16 deletions(-) > If possible please group only related patches in the future. According to the testbot, it still introduces test failures: http://source.winehq.org/patches/data/119233.testfail Also, I'm not sure if you saw https://www.winehq.org/pipermail/wine-devel/2016-February/111675.html where I gave some additional feedback. If you have any questions, feel free to ask. Best regards, Sebastian

1 0

Re: ws2_32/tests: Add tests for getsockopt() with SO_SNDBUF and SO_RCVBUF optname.
by Bruno Jesus 16 Feb '16

16 Feb '16

On Tue, Feb 16, 2016 at 9:30 AM, Sebastian Lackner <sebastian(a)fds-team.de> wrote: > Signed-off-by: Sebastian Lackner <sebastian(a)fds-team.de> > --- > > @Bruno: As far as I remember you were working on a fix for this, what is > the progress? > Hi, Sebastian. Unfortunately since December I'm buried by real life work and I thought I could restart working in Wine on February but that was not possible. My solution to this is still theoretical, I didn't write any code yet. The solution is to cache the values of SO_RCVBUF, SO_SNDBU, SO_REUSEADDR and SO_ACCEPTCONN. Why? SO_RCVBUF and SO_SNBUF are messed by Linux kernel. So we would cache the requested value and return a min() between what we have in the cache and what we read back from the kernel. This would solve this problem and not require #ifdef. SO_REUSEADDR because Windows does not have TIME_WAIT at all, so all sockets would be created with SO_REUSEADDR in the server and we would take care of it manually. There is a long standing uTorrent bug about this. SO_ACCEPTCONN is bugged in OSX, it is never updated by the kernel so it is useless and always fail. To solve this we would set the cache to 1 when listen() is called and return the cache instead of asking for the kernel. I'm still reading all bugs and emails so I'm aware of your divide by 2 solution and I really don't mind if it gets accepted as it is ready and mine is still months to come. Best wishes, Bruno

1 0

Re: ws2_32/tests: Add tests for getsockopt() with SO_SNDBUF and SO_RCVBUF optname.
by Marvin 16 Feb '16

16 Feb '16

Hi, While running your changed tests on Windows, I think I found new failures. Being a bot and all I'm not very good at pattern recognition, so I might be wrong, but could you please double-check? Full results can be found at https://testbot.winehq.org/JobDetails.pl?Key=20513 Your paranoid android. === w7pro64 (32 bit sock) === sock.c:7750: Test failed: bytesReturned isn't supposed to be 1 sock.c:7774: Test failed: GetOverlappedResult failed, error 0

2 1