Date: Tue, 15 Feb 2005 21:28:48 +0900 (JST) From: "Michael Leonhard" Subject: [SDL] resize bug on Win32 and patch This is my first post to this mailing list. In this email I will detail a bug in the behavior of resizable SDL windows on Win32. Then I will explain the solution and provide a patch. Symptoms: Under Windows, an SDL display created with the SDL_RESIZABLE flag exhibits quirky behavior when being maximized. The window is resized to the proper size, but it is shifted upwards about half the height of the title bar. Similarly, a window whose origin is above the top of the screen will spontaneously move its upper-left origin upon being resized. After two such resize-induced moves, the title bar will be entirely off the top edge of the screen. Subsequently, when the mouse is clicked and released on the window border, the window will shrink its height spontaneously. This height shrinkage occurs even if the user did not resize the border. To observe this curious situation, please invoke: SDL-1.2.8/test/testwm.exe -resize Cause: A pair of integers, SDL_windowX and SDL_windowY, are defined in video/wincommon/SDL_sysevents.c. They are used by the DirectX video driver and the DIB video driver: video/windx5/SDL_dx5video.c video/windib/SDL_dibvideo.c As I understand the source code, the primary use of these variables is to create a rectangle that represents the surface area in CLIENT SPACE. Client space refers to a coordinate system that originates at the upper left corner of a Win32 Window's drawable area. This is just inside the window border and title bar. This client space rectangle, called bounds, is subsequently converted to screen space with a call to AdjustWindowRectEx. The problem is found in SDL's handling of the WM_WINDOWPOSCHANGED message. According to MSDN, "The WM_WINDOWPOSCHANGED message is sent to a window whose size, position, or place in the Z order has changed as a result of a call to the SetWindowPos function or another window-management function." I have confirmed that this message is indeed being sent to the SDL window when the mouse is clicked on the window border, even if the window border is not dragged. In video/wincommon/SDL_sysevents.c, on line 464, in response to the WM_WINDOWPOSCHANGED message, the (potentially) new client rectangle is obtained. This rectangle is translated into screen coordinates and THEN assigned to the SDL_windowX and Y variables. Thus screen coordinates are being assigned to client coordinate variables. Once this is understood, the solution is apparent: assign SDL_windowX and Y before translating the rectangle to screen coordinates. This is accomplished by the following patch. -Mike_L

Wish list for the 1.3 development branch:

 * Use /etc/fb.modes, if available, like GGI does
 * Add mousewheel events (new unified event architecture?)
 * DirectInput joystick support needs to be implemented
 * Be able to enumerate and select available audio and video drivers
 * Fullscreen video mode support for MacOS X
 * Explicit vertical retrace wait (maybe separate from SDL_Flip?)
 * Shaped windows, windows without borders
 * Multiple windows, multiple display support
 * SDL_INIT_EVENTTHREAD on Windows and MacOS?
 * Add a timestamp to events
 * Use RDTSC for timer resolution on x86 hardware
 * Add audio input API
 * Add hardware accelerated scaled blit
 * Add hardware accelerated alpha blits
 * Redesign blitting architecture to allow blit plugins

In the jump from 1.2 to 1.3, we should change the SDL_Rect members to
int and evaluate all the rest of the datatypes.  This is the only place
we should do it though, since the 1.2 series should not break binary
compatibility in this way.

Requests:
 * PCM and CDROM volume control (deprecated, but possible)