May 2019 - Wine-Devel - list.winehq.org

Re: [PATCH 2/2] comctl32: Paint 32-bpp bitmaps with an alpha channel using GdiAlphaBlend.

by Dmitry Timoshkov

Alexandre Julliard <julliard(a)winehq.org> wrote: > >> > +static HBITMAP create_alpha_bitmap( HBITMAP hbitmap ) > >> > +{ > >> > + HBITMAP alpha = 0; > >> > + BITMAPINFO *info = NULL; > >> > + BITMAP bm; > >> > + HDC src, dst; > >> > + void *bits; > >> > + DWORD i; > >> > + const unsigned char *ptr; > >> > + BOOL has_alpha = FALSE; > >> > + > >> > + if (!GetObjectW( hbitmap, sizeof(bm), &bm )) return 0; > >> > + if (bm.bmBitsPixel != 32) return 0; > >> > + > >> > + if (!(src = CreateCompatibleDC( 0 ))) return 0; > >> > + if (!(dst = CreateCompatibleDC( src ))) goto done; > >> > + if (!(info = HeapAlloc( GetProcessHeap(), 0, FIELD_OFFSET( BITMAPINFO, bmiColors[256] )))) goto done; > >> > + info->bmiHeader.biSize = sizeof(BITMAPINFOHEADER); > >> > + info->bmiHeader.biWidth = bm.bmWidth; > >> > + info->bmiHeader.biHeight = -bm.bmHeight; > >> > + info->bmiHeader.biPlanes = 1; > >> > + info->bmiHeader.biBitCount = 32; > >> > + info->bmiHeader.biCompression = BI_RGB; > >> > + info->bmiHeader.biSizeImage = bm.bmWidth * bm.bmHeight * 4; > >> > + info->bmiHeader.biXPelsPerMeter = 0; > >> > + info->bmiHeader.biYPelsPerMeter = 0; > >> > + info->bmiHeader.biClrUsed = 0; > >> > + info->bmiHeader.biClrImportant = 0; > >> > + if (!(alpha = CreateDIBSection( dst, info, DIB_RGB_COLORS, &bits, NULL, 0 ))) goto done; > >> > + > >> > + SelectObject( src, hbitmap ); > >> > + SelectObject( dst, alpha ); > >> > + BitBlt(dst, 0, 0, bm.bmWidth, bm.bmHeight, src, 0, 0, SRCCOPY); > >> > >> BitBlt() is not guaranteed to preserve alpha, it's not the right > >> function to use here. > > > > I've heard that before but didn't see the reasoning, in which cases it's > > supposed to fail? Are you probably suggesting to use GdiAlphaBlend() or > > something else? > > GetDIBits() is a better choice. Though in general you'd have a DIB > section already, in which case you don't need to copy the bits at all. In this particular case there is no a DIB section, so I'd need to copy. I still fail to see why BitBlt() won't work in general when copying a bitmap with an alpah channel, and my experiance clearly shows that it does both under Windows and Wine. > It's not clear that an alpha channel is even meaningful with a DDB, did > you test that case? I'm testing with DIB sections, why would I need to test with a DDB? -- Dmitry.

May 31, 2019

Re: [PATCH v6 1/7] ntdll/tests: Move NtAllocateVirtualMemory tests that were in kernel32

by Dmitry Timoshkov

Alexandre Julliard <julliard(a)winehq.org> wrote: > Dmitry Timoshkov <dmitry(a)baikal.ru> writes: > > > Huw Davies <huw(a)codeweavers.com> wrote: > > > >> On Fri, May 31, 2019 at 10:31:10AM +0200, Rémi Bernon wrote: > >> > Is there a particular reason the tested functions are dynamically > >> > loaded in some test executables? > >> > >> The typical reason would be that the functions are not available on > >> all of the Windows versions that we test. > >> > >> As we retire testing on older Windows versions this can leave us > >> dynamically loading functions that we need not any more. > > > > Another reason is that sometimes PSDK doesn't have an appropriate import > > library. Alphough it tends to get added in newer PSDK versions, but that's > > not a good reason to break compilation with older PSDK versions by removing > > dynamic API loading (some people abuse this very badly recently). > > It's not unreasonable to require a recent PSDK, if that makes the code > simpler. Some of my test WMs running Widows XP, some Windows 7, it's pretty questionable to require installing a Windows 10 SDK there (if it will ever install under XP). On the other hand a test compiled under Windows 7 won't run under XP. Personally, I find testing under XP and Windows 7 very valueable, please don't ruin that ability. -- Dmitry.

May 31, 2019

Re: [PATCH 2/2] comctl32: Paint 32-bpp bitmaps with an alpha channel using GdiAlphaBlend.

by Alexandre Julliard

Dmitry Timoshkov <dmitry(a)baikal.ru> writes: > Alexandre Julliard <julliard(a)winehq.org> wrote: > >> Dmitry Timoshkov <dmitry(a)baikal.ru> writes: >> >> > +static HBITMAP create_alpha_bitmap( HBITMAP hbitmap ) >> > +{ >> > + HBITMAP alpha = 0; >> > + BITMAPINFO *info = NULL; >> > + BITMAP bm; >> > + HDC src, dst; >> > + void *bits; >> > + DWORD i; >> > + const unsigned char *ptr; >> > + BOOL has_alpha = FALSE; >> > + >> > + if (!GetObjectW( hbitmap, sizeof(bm), &bm )) return 0; >> > + if (bm.bmBitsPixel != 32) return 0; >> > + >> > + if (!(src = CreateCompatibleDC( 0 ))) return 0; >> > + if (!(dst = CreateCompatibleDC( src ))) goto done; >> > + if (!(info = HeapAlloc( GetProcessHeap(), 0, FIELD_OFFSET( BITMAPINFO, bmiColors[256] )))) goto done; >> > + info->bmiHeader.biSize = sizeof(BITMAPINFOHEADER); >> > + info->bmiHeader.biWidth = bm.bmWidth; >> > + info->bmiHeader.biHeight = -bm.bmHeight; >> > + info->bmiHeader.biPlanes = 1; >> > + info->bmiHeader.biBitCount = 32; >> > + info->bmiHeader.biCompression = BI_RGB; >> > + info->bmiHeader.biSizeImage = bm.bmWidth * bm.bmHeight * 4; >> > + info->bmiHeader.biXPelsPerMeter = 0; >> > + info->bmiHeader.biYPelsPerMeter = 0; >> > + info->bmiHeader.biClrUsed = 0; >> > + info->bmiHeader.biClrImportant = 0; >> > + if (!(alpha = CreateDIBSection( dst, info, DIB_RGB_COLORS, &bits, NULL, 0 ))) goto done; >> > + >> > + SelectObject( src, hbitmap ); >> > + SelectObject( dst, alpha ); >> > + BitBlt(dst, 0, 0, bm.bmWidth, bm.bmHeight, src, 0, 0, SRCCOPY); >> >> BitBlt() is not guaranteed to preserve alpha, it's not the right >> function to use here. > > I've heard that before but didn't see the reasoning, in which cases it's > supposed to fail? Are you probably suggesting to use GdiAlphaBlend() or > something else? GetDIBits() is a better choice. Though in general you'd have a DIB section already, in which case you don't need to copy the bits at all. It's not clear that an alpha channel is even meaningful with a DDB, did you test that case? -- Alexandre Julliard julliard(a)winehq.org

May 31, 2019

[PATCH] user32: Fix some leaks on error path (Coverity).

by Sven Baars

Signed-off-by: Sven Baars <sven.wine(a)gmail.com> --- dlls/user32/rawinput.c | 2 ++ 1 file changed, 2 insertions(+) diff --git a/dlls/user32/rawinput.c b/dlls/user32/rawinput.c index 2085fd3f9f..56ea2497d1 100644 --- a/dlls/user32/rawinput.c +++ b/dlls/user32/rawinput.c @@ -141,6 +141,8 @@ static void find_hid_devices(void) if (!hid_devices) { ERR("Failed to allocate memory.\n"); + CloseHandle(file); + heap_free(path); goto done; } } -- 2.17.1

May 31, 2019

Re: [PATCH v6 2/7] ntdll/tests: Add broken results NtAllocateVirtualMemory tests on w1064v1809

by Rémi Bernon

On Thu, 2019-05-30 at 13:21 +0100, Huw Davies wrote: > On Wed, May 29, 2019 at 03:13:08PM +0200, Rémi Bernon wrote: > > if (status == STATUS_SUCCESS) ok(VirtualFree(addr2, 0, > > MEM_RELEASE), "VirtualFree failed\n"); > > > > /* AT_ROUND_TO_PAGE flag is not supported for > > NtAllocateVirtualMemory */ > > @@ -66,7 +69,9 @@ static void test_AllocateVirtualMemory(void) > > status = pNtAllocateVirtualMemory(GetCurrentProcess(), &addr2, > > 0, &size, MEM_RESERVE | > > MEM_COMMIT | > > AT_ROUND_TO_PAGE, PAGE_EXECUTE_READWRITE); > > todo_wine > > - ok(status == STATUS_INVALID_PARAMETER_5, > > "NtAllocateVirtualMemory returned %08x\n", status); > > + ok(status == STATUS_INVALID_PARAMETER_5 || > > + broken(status == STATUS_INVALID_PARAMETER) /* w1064v1809 > > */, > > + "NtAllocateVirtualMemory returned %08x\n", status); > > And again. > > Huw. Here Wine returns STATUS_INVALID_PARAMETER, like w1064v1809, but not like previous Windows version, therefore the todo_wine. Should we detect the emulated windows version to be sure the returned status is consistent with it, or can the todo_wine be dropped? Rémi.

May 31, 2019

[PATCH 2/2] d3dx9: Print a FIXME when D3DXSaveSurfaceToFileInMemory() is called with an empty rect.

by Matteo Bruni

Signed-off-by: Matteo Bruni <mbruni(a)codeweavers.com> --- dlls/d3dx9_36/surface.c | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) diff --git a/dlls/d3dx9_36/surface.c b/dlls/d3dx9_36/surface.c index 57d9818cd91..531aeeb90d9 100644 --- a/dlls/d3dx9_36/surface.c +++ b/dlls/d3dx9_36/surface.c @@ -2129,7 +2129,8 @@ HRESULT WINAPI D3DXSaveSurfaceToFileInMemory(ID3DXBuffer **dst_buffer, D3DXIMAGE { if (src_rect->left == src_rect->right || src_rect->top == src_rect->bottom) { - WARN("Invalid rectangle with 0 area\n"); + FIXME("Empty rectangle.\n"); + /* This should return a proper file header. */ return D3DXCreateBuffer(64, dst_buffer); } if (src_rect->left < 0 || src_rect->top < 0) -- 2.21.0

May 31, 2019

[PATCH 1/2] wined3d: Zero-initialize the swapchain context array.

by Matteo Bruni

Wine-Bug: https://bugs.winehq.org/show_bug.cgi?id=47255 Signed-off-by: Matteo Bruni <mbruni(a)codeweavers.com> --- Fixes a regression from commit db201072655946662c041a66ee434c30c245e5b0. dlls/wined3d/swapchain.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/dlls/wined3d/swapchain.c b/dlls/wined3d/swapchain.c index fbedd7470df..48d0920554d 100644 --- a/dlls/wined3d/swapchain.c +++ b/dlls/wined3d/swapchain.c @@ -892,7 +892,7 @@ static HRESULT swapchain_init(struct wined3d_swapchain *swapchain, struct wined3 if (!(device->wined3d->flags & WINED3D_NO3D)) { - if (!(swapchain->context = heap_alloc(sizeof(*swapchain->context)))) + if (!(swapchain->context = heap_alloc_zero(sizeof(*swapchain->context)))) { ERR("Failed to create the context array.\n"); hr = E_OUTOFMEMORY; -- 2.21.0

May 31, 2019

Re: [PATCH v6 1/7] ntdll/tests: Move NtAllocateVirtualMemory tests that were in kernel32

by Alexandre Julliard

Dmitry Timoshkov <dmitry(a)baikal.ru> writes: > Huw Davies <huw(a)codeweavers.com> wrote: > >> On Fri, May 31, 2019 at 10:31:10AM +0200, Rémi Bernon wrote: >> > Is there a particular reason the tested functions are dynamically >> > loaded in some test executables? >> >> The typical reason would be that the functions are not available on >> all of the Windows versions that we test. >> >> As we retire testing on older Windows versions this can leave us >> dynamically loading functions that we need not any more. > > Another reason is that sometimes PSDK doesn't have an appropriate import > library. Alphough it tends to get added in newer PSDK versions, but that's > not a good reason to break compilation with older PSDK versions by removing > dynamic API loading (some people abuse this very badly recently). It's not unreasonable to require a recent PSDK, if that makes the code simpler. -- Alexandre Julliard julliard(a)winehq.org

May 31, 2019

Re: [PATCH v6 1/7] ntdll/tests: Move NtAllocateVirtualMemory tests that were in kernel32

by Dmitry Timoshkov

Huw Davies <huw(a)codeweavers.com> wrote: > On Fri, May 31, 2019 at 10:31:10AM +0200, Rémi Bernon wrote: > > Is there a particular reason the tested functions are dynamically > > loaded in some test executables? > > The typical reason would be that the functions are not available on > all of the Windows versions that we test. > > As we retire testing on older Windows versions this can leave us > dynamically loading functions that we need not any more. Another reason is that sometimes PSDK doesn't have an appropriate import library. Alphough it tends to get added in newer PSDK versions, but that's not a good reason to break compilation with older PSDK versions by removing dynamic API loading (some people abuse this very badly recently). -- Dmitry.

May 31, 2019

Re: [PATCH 2/2] comctl32: Paint 32-bpp bitmaps with an alpha channel using GdiAlphaBlend.

by Dmitry Timoshkov

Alexandre Julliard <julliard(a)winehq.org> wrote: > Dmitry Timoshkov <dmitry(a)baikal.ru> writes: > > > +static HBITMAP create_alpha_bitmap( HBITMAP hbitmap ) > > +{ > > + HBITMAP alpha = 0; > > + BITMAPINFO *info = NULL; > > + BITMAP bm; > > + HDC src, dst; > > + void *bits; > > + DWORD i; > > + const unsigned char *ptr; > > + BOOL has_alpha = FALSE; > > + > > + if (!GetObjectW( hbitmap, sizeof(bm), &bm )) return 0; > > + if (bm.bmBitsPixel != 32) return 0; > > + > > + if (!(src = CreateCompatibleDC( 0 ))) return 0; > > + if (!(dst = CreateCompatibleDC( src ))) goto done; > > + if (!(info = HeapAlloc( GetProcessHeap(), 0, FIELD_OFFSET( BITMAPINFO, bmiColors[256] )))) goto done; > > + info->bmiHeader.biSize = sizeof(BITMAPINFOHEADER); > > + info->bmiHeader.biWidth = bm.bmWidth; > > + info->bmiHeader.biHeight = -bm.bmHeight; > > + info->bmiHeader.biPlanes = 1; > > + info->bmiHeader.biBitCount = 32; > > + info->bmiHeader.biCompression = BI_RGB; > > + info->bmiHeader.biSizeImage = bm.bmWidth * bm.bmHeight * 4; > > + info->bmiHeader.biXPelsPerMeter = 0; > > + info->bmiHeader.biYPelsPerMeter = 0; > > + info->bmiHeader.biClrUsed = 0; > > + info->bmiHeader.biClrImportant = 0; > > + if (!(alpha = CreateDIBSection( dst, info, DIB_RGB_COLORS, &bits, NULL, 0 ))) goto done; > > + > > + SelectObject( src, hbitmap ); > > + SelectObject( dst, alpha ); > > + BitBlt(dst, 0, 0, bm.bmWidth, bm.bmHeight, src, 0, 0, SRCCOPY); > > BitBlt() is not guaranteed to preserve alpha, it's not the right > function to use here. I've heard that before but didn't see the reasoning, in which cases it's supposed to fail? Are you probably suggesting to use GdiAlphaBlend() or something else? I'm using BitBlt() in all my test apps that play with alpha and I didn't see a single case where alpha is not preserved. -- Dmitry.

May 31, 2019