[PATCH 0/1] MR6482: win32u: Leave window surface alpha bits to -1 when unset.

[Rémi Bernon]

Fixes: 05d727a935b9009f188cc74ca77e14afc44aef30
Wine-Bug: https://bugs.winehq.org/show_bug.cgi?id=57160

-- 
https://gitlab.winehq.org/wine/wine/-/merge_requests/6482

[PATCH 1/1] win32u: Leave window surface alpha bits to -1 when unset.

[Rémi Bernon]

From: Rémi Bernon <rbernon(a)codeweavers.com>

Fixes: 05d727a935b9009f188cc74ca77e14afc44aef30
Wine-Bug: https://bugs.winehq.org/show_bug.cgi?id=57160
---
 dlls/win32u/window.c | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)

diff --git a/dlls/win32u/window.c b/dlls/win32u/window.c
index 77471213431..ab58cabb40e 100644
--- a/dlls/win32u/window.c
+++ b/dlls/win32u/window.c
@@ -1949,14 +1949,14 @@ static struct window_surface *get_window_surface( HWND hwnd, UINT swp_flags, BOO
 
     if (new_surface && !is_layered)
     {
-        DWORD lwa_flags = 0;
+        DWORD lwa_flags = 0, alpha_bits = -1;
         COLORREF key;
         BYTE alpha;
 
         if (!NtUserGetLayeredWindowAttributes( hwnd, &key, &alpha, &lwa_flags )) lwa_flags = 0;
-        if (!(lwa_flags & LWA_ALPHA)) alpha = 255;
+        if (lwa_flags & LWA_ALPHA) alpha_bits = alpha << 24;
         if (!(lwa_flags & LWA_COLORKEY)) key = CLR_INVALID;
-        window_surface_set_layered( new_surface, key, alpha << 24, 0 );
+        window_surface_set_layered( new_surface, key, alpha_bits, 0 );
     }
 
     return new_surface;
-- 
GitLab

https://gitlab.winehq.org/wine/wine/-/merge_requests/6482

[Dmitry Timoshkov (＠dmitry)]

Dmitry Timoshkov (@dmitry) commented about dlls/win32u/window.c:
> if (new_surface && !is_layered)
>      {
> -        DWORD lwa_flags = 0;
> +        DWORD lwa_flags = 0, alpha_bits = -1;

This either shouldn't be a DWORD or -1. What about using \~0?

-- 
https://gitlab.winehq.org/wine/wine/-/merge_requests/6482#note_81994

[Rémi Bernon]

On Thu Sep 12 08:15:18 2024 +0000, Dmitry Timoshkov wrote:
> This either shouldn't be a DWORD or -1. What about using \~0?
> It does the same thing, I don't think it matters and alpha_bits is initialized with -1 everywhere else it's used.

Using ~0 makes it very tempting to use ~0u, which is wrong in various cases (`UINT64 bla = ~0u`, isn't the same value as `UINT64 bla = ~0` / `UINT64 bla = -1` which both are equal).

-- 
https://gitlab.winehq.org/wine/wine/-/merge_requests/6482#note_81998

[Dmitry Timoshkov (＠dmitry)]

On Thu Sep 12 08:23:17 2024 +0000, Rémi Bernon wrote:
> It does the same thing, I don't think it matters and alpha_bits is
> initialized with -1 everywhere else it's used.
> Using ~0 makes it very tempting to use ~0u, which is wrong in various
> cases (`UINT64 bla = ~0u`, isn't the same value as `UINT64 bla = ~0` /
> `UINT64 bla = -1` which both are equal).
> Of course it's the same thing, hence the suggestion. It's clearly wrong to assign a signed value to a variable of an unsigned type.

-- 
https://gitlab.winehq.org/wine/wine/-/merge_requests/6482#note_82001

[Rémi Bernon]

On Thu Sep 12 08:31:57 2024 +0000, Dmitry Timoshkov wrote:
> Of course it's the same thing, hence the suggestion. It's clearly wrong
> to assign a signed value to a variable of an unsigned type.
> Well, if you want to be absolutely strict, ~0 is still signed, so you have to use ~0u (which is inconvenient because then you have to use the exactly right suffix for the variable type).

-- 
https://gitlab.winehq.org/wine/wine/-/merge_requests/6482#note_82002