October 2009 - Wine-Devel - list.winehq.org

Re: [comctl32/tests] Fix some test failures

by Paul Vriens

On 10/01/2009 02:58 PM, Nikolay Sivov wrote: > Yes, please. If a problem is only in rectangle orders it's very simple > to fix (test for yyyy in returned locale data). I suggest not to drop > tests yet, we'll probably want to drop them later when trying probable > dpi-dependent failures. Well I need to test for all possible occurrences of month and year. So something like: /* Get the format of the title */ len = GetLocaleInfo(LOCALE_USER_DEFAULT, LOCALE_SYEARMONTH, NULL, 0); yearmonth = HeapAlloc(GetProcessHeap(), 0, len); GetLocaleInfo(LOCALE_USER_DEFAULT, LOCALE_SYEARMONTH, yearmonth, len); /* Find out the order */ year = strstr(yearmonth, "y"); month = strstr(yearmonth, "M"); if (month < year) trace("Month before year\n"); else trace("Year before month\n"); HeapFree(GetProcessHeap(), 0, yearmonth); The 3 tests I removed with the first patch can be removed safely as all should be tested in the loop. Now I need to change the way we test to cope with: - order of year/month - text right after the prev button or right before the next button -- Cheers, Paul.

Oct. 1, 2009

Re: [comctl32/tests] Fix some test failures

by Nikolay Sivov

Paul Vriens wrote: > On 10/01/2009 02:03 PM, Nikolay Sivov wrote: >>> The first and last character show up 'underneath' the prev/next >>> buttons and the control is smaller than in US-English. This is because >>> the day acronyms are only 2 characters. >> You're testing on Windows system? > > Yes, this was on Windows XP. Ok, I'll try to fix it someday. > >> >> In git abbreviated day names use locale data already, so it should be >> smaller in Wine too. >>> >>> So the patch is not correct as it doesn't cover all cases (yet). >>> >> It could be fixed with GetLocaleInfo() but it complicates test of >> course. > > Well, yes. Are you going to go further with this or should I "perfect" > at least the tests? Yes, please. If a problem is only in rectangle orders it's very simple to fix (test for yyyy in returned locale data). I suggest not to drop tests yet, we'll probably want to drop them later when trying probable dpi-dependent failures.

Oct. 1, 2009

Re: [2/5] WineD3D: SetRenderTarget doesn't change the viewport in d3d7

by Stefan Dösinger

Am 01.10.2009 um 13:47 schrieb Alexandre Julliard: > Stefan Dösinger <stefandoesinger(a)gmx.at> writes: > >> Patches 2, 3 and 4 fix bug 19365. In the long run we want to get rid >> of the dxVersion parameter that is used in this patch, but we haven't >> agreed on a way to do this so far, so lets fix the bug first. > > It doesn't work here: > > ../../../tools/runtest -q -P wine -M ddraw.dll -T ../../.. -p > ddraw_test.exe.so visual.c && touch visual.ok > visual.c:600: Test failed: Offscreen failed: Got color 0x00ff00ff, > expected 0x00ffffff. > visual.c:788: Test failed: SRCALPHA on texture returned color > 0x007f0080, expected 0x00bf4000 > make: *** [visual.ok] Error 2 I see what's going on. The visual offscreen rendering test doesn't change the viewport(the test was ported from d3d9), so it ends up with a viewport that is bigger than the render target, and thus invalid. Trying to set the viewport just returned by GetViewport fails. The combination of the failing test and my new test shows that Windows clamps the currently set viewport to the render target size, if the viewport is bigger. GetViewport still returns the viewport from the stateblock. A test shows that this clamping occurs even in d3d9.

Oct. 1, 2009

Re: [comctl32/tests] Fix some test failures

by Paul Vriens

On 10/01/2009 02:03 PM, Nikolay Sivov wrote: > Paul Vriens wrote: >> On 10/01/2009 01:32 PM, Nikolay Sivov wrote: >>> Paul Vriens wrote: >>>> On 10/01/2009 12:29 PM, Paul Vriens wrote: >>>>> Hi, >>>>> >>>>> Instead of checking fixed points in the title we now check for the >>>>> order >>>>> of items only. >>>>> >>>>> I didn't cater for locale differences on Windows (as I don't have a >>>>> way >>>>> to test), so I'm going to rely on test.winehq.org after this patch has >>>>> been applied (if, of course). >>>>> >>>>> The test could be easily extended if we know for a fact that items >>>>> should be at a fixed place (or by calculating text width, button width >>>>> an such). >>>>> >>>>> Changelog >>>>> Fix some test failures >>>>> >>>> >>>> FWIW, changing the regional settings on a Windows box does change the >>>> title of the calendar. Setting it to Portuguese shows that the name of >>>> the month is longer than expected and hence the hittests don't work as >>>> it was before. >>>> >>>> This patch also succeeds on the Portuguese settings (tested on W2K3). >>>> >>> So, back to previous question, after a brief search for this string >>> 'LOCALE_SYEARMONTH "yyyy' I found some locales that use that. >>> Now the question is does this control respect this locale parameter. >>> I'll to run native comctl32 for one of such locales today later, or will >>> force this string for default ENU. After that it'll be clear what to do >>> with our monthcal. >> >> If I set the regional settings to Lithuanian I can see the title as >> "2007 m. balandis" which corresponds to LOCALE_SYEARMONTH in the nls >> file. >> >> The first and last character show up 'underneath' the prev/next >> buttons and the control is smaller than in US-English. This is because >> the day acronyms are only 2 characters. > You're testing on Windows system? Yes, this was on Windows XP. > > In git abbreviated day names use locale data already, so it should be > smaller in Wine too. >> >> So the patch is not correct as it doesn't cover all cases (yet). >> > It could be fixed with GetLocaleInfo() but it complicates test of course. Well, yes. Are you going to go further with this or should I "perfect" at least the tests? -- Cheers, Paul.

Oct. 1, 2009

Re: wininet(3/4): Set callbacks needed by OpenSSL for multithreaded use

by Alexandre Julliard

Juan Lang <juan.lang(a)gmail.com> writes: > +static HANDLE *ssl_locks; > + > +static unsigned long ssl_thread_id(void) > +{ > + return GetThreadId(GetCurrentThread()); > +} You want GetCurrentThreadId(); > +static void ssl_lock_callback(int mode, int type, const char *file, int line) > +{ > + if (mode & CRYPTO_LOCK) > + WaitForSingleObject(ssl_locks[type], INFINITE); > + else > + ReleaseMutex(ssl_locks[type]); > +} You should use critical sections instead, it will be faster. -- Alexandre Julliard julliard(a)winehq.org

Oct. 1, 2009

Re: [2/9] include: Add MAPI_USE_DEFAULT and MDB_NO_DIALOG definitions to mapix.h

by Alexandre Julliard

Owen Rudge <orudge(a)codeweavers.com> writes: > @@ -51,6 +51,9 @@ typedef ULONG FLAGS; > #ifndef MAPI_EXTENDED > #define MAPI_EXTENDED 0x00000020 > #endif > +#ifndef MAPI_USE_DEFAULT > +#define MAPI_USE_DEFAULT 0x00000040 > +#endif > #ifndef MAPI_FORCE_DOWNLOAD > #define MAPI_FORCE_DOWNLOAD 0x00001000 > #endif > @@ -65,6 +68,9 @@ typedef ULONG FLAGS; > #define MAPI_SIMPLE_DEFAULT (MAPI_LOGON_UI|MAPI_ALLOW_OTHERS|MAPI_FORCE_DOWNLOAD) > #define MAPI_SIMPLE_EXPLICIT (MAPI_NEW_SESSION|MAPI_EXPLICIT_PROFILE|MAPI_FORCE_DOWNLOAD) > > +/* IMAPISession::OpenMessageStore() flags */ > +#define MDB_NO_DIALOG 0x00000001 > + These things apparently belong to mapidefs.h. -- Alexandre Julliard julliard(a)winehq.org

Oct. 1, 2009

Re: [comctl32/tests] Fix some test failures

by Nikolay Sivov

Paul Vriens wrote: > On 10/01/2009 01:32 PM, Nikolay Sivov wrote: >> Paul Vriens wrote: >>> On 10/01/2009 12:29 PM, Paul Vriens wrote: >>>> Hi, >>>> >>>> Instead of checking fixed points in the title we now check for the >>>> order >>>> of items only. >>>> >>>> I didn't cater for locale differences on Windows (as I don't have a >>>> way >>>> to test), so I'm going to rely on test.winehq.org after this patch has >>>> been applied (if, of course). >>>> >>>> The test could be easily extended if we know for a fact that items >>>> should be at a fixed place (or by calculating text width, button width >>>> an such). >>>> >>>> Changelog >>>> Fix some test failures >>>> >>> >>> FWIW, changing the regional settings on a Windows box does change the >>> title of the calendar. Setting it to Portuguese shows that the name of >>> the month is longer than expected and hence the hittests don't work as >>> it was before. >>> >>> This patch also succeeds on the Portuguese settings (tested on W2K3). >>> >> So, back to previous question, after a brief search for this string >> 'LOCALE_SYEARMONTH "yyyy' I found some locales that use that. >> Now the question is does this control respect this locale parameter. >> I'll to run native comctl32 for one of such locales today later, or will >> force this string for default ENU. After that it'll be clear what to do >> with our monthcal. > > If I set the regional settings to Lithuanian I can see the title as > "2007 m. balandis" which corresponds to LOCALE_SYEARMONTH in the nls > file. > > The first and last character show up 'underneath' the prev/next > buttons and the control is smaller than in US-English. This is because > the day acronyms are only 2 characters. You're testing on Windows system? In git abbreviated day names use locale data already, so it should be smaller in Wine too. > > So the patch is not correct as it doesn't cover all cases (yet). > It could be fixed with GetLocaleInfo() but it complicates test of course.

Oct. 1, 2009

Re: [comctl32/tests] Fix some test failures

by Paul Vriens

On 10/01/2009 01:32 PM, Nikolay Sivov wrote: > Paul Vriens wrote: >> On 10/01/2009 12:29 PM, Paul Vriens wrote: >>> Hi, >>> >>> Instead of checking fixed points in the title we now check for the order >>> of items only. >>> >>> I didn't cater for locale differences on Windows (as I don't have a way >>> to test), so I'm going to rely on test.winehq.org after this patch has >>> been applied (if, of course). >>> >>> The test could be easily extended if we know for a fact that items >>> should be at a fixed place (or by calculating text width, button width >>> an such). >>> >>> Changelog >>> Fix some test failures >>> >> >> FWIW, changing the regional settings on a Windows box does change the >> title of the calendar. Setting it to Portuguese shows that the name of >> the month is longer than expected and hence the hittests don't work as >> it was before. >> >> This patch also succeeds on the Portuguese settings (tested on W2K3). >> > So, back to previous question, after a brief search for this string > 'LOCALE_SYEARMONTH "yyyy' I found some locales that use that. > Now the question is does this control respect this locale parameter. > I'll to run native comctl32 for one of such locales today later, or will > force this string for default ENU. After that it'll be clear what to do > with our monthcal. If I set the regional settings to Lithuanian I can see the title as "2007 m. balandis" which corresponds to LOCALE_SYEARMONTH in the nls file. The first and last character show up 'underneath' the prev/next buttons and the control is smaller than in US-English. This is because the day acronyms are only 2 characters. So the patch is not correct as it doesn't cover all cases (yet). -- Cheers, Paul.

Oct. 1, 2009

Re: [2/5] WineD3D: SetRenderTarget doesn't change the viewport in d3d7

by Alexandre Julliard

Stefan Dösinger <stefandoesinger(a)gmx.at> writes: > Patches 2, 3 and 4 fix bug 19365. In the long run we want to get rid > of the dxVersion parameter that is used in this patch, but we haven't > agreed on a way to do this so far, so lets fix the bug first. It doesn't work here: ../../../tools/runtest -q -P wine -M ddraw.dll -T ../../.. -p ddraw_test.exe.so visual.c && touch visual.ok visual.c:600: Test failed: Offscreen failed: Got color 0x00ff00ff, expected 0x00ffffff. visual.c:788: Test failed: SRCALPHA on texture returned color 0x007f0080, expected 0x00bf4000 make: *** [visual.ok] Error 2 -- Alexandre Julliard julliard(a)winehq.org

Oct. 1, 2009

Re: [comctl32/tests] Fix some test failures

by Nikolay Sivov

Paul Vriens wrote: > On 10/01/2009 12:29 PM, Paul Vriens wrote: >> Hi, >> >> Instead of checking fixed points in the title we now check for the order >> of items only. >> >> I didn't cater for locale differences on Windows (as I don't have a way >> to test), so I'm going to rely on test.winehq.org after this patch has >> been applied (if, of course). >> >> The test could be easily extended if we know for a fact that items >> should be at a fixed place (or by calculating text width, button width >> an such). >> >> Changelog >> Fix some test failures >> > > FWIW, changing the regional settings on a Windows box does change the > title of the calendar. Setting it to Portuguese shows that the name of > the month is longer than expected and hence the hittests don't work as > it was before. > > This patch also succeeds on the Portuguese settings (tested on W2K3). > So, back to previous question, after a brief search for this string 'LOCALE_SYEARMONTH "yyyy' I found some locales that use that. Now the question is does this control respect this locale parameter. I'll to run native comctl32 for one of such locales today later, or will force this string for default ENU. After that it'll be clear what to do with our monthcal.

Oct. 1, 2009