Christopher Thielen cthielen@gmail.com writes:
Thank you Alexandre, you're absolutely right.
Below is a further analysis on toolbar, trackbar, and button:
Toolbar: Assumes a WM_CAPTURECHANGED messages implies a toolbar button is no longer pressed. The proposed patch would therefore break existing behavior if and only if the toolbar code also calls SetCapture() on itself while it already has capture. Toolbar currently only calls SetCapture() in the case of a LButtonDown on itself. Notably, another piece of code could in theory call SetCapture() on the toolbar's hWnd as well.
Trackbar: Assumes a WM_CAPTURECHANGED message implies a TB_ENDTRACK, again, only breaking current behavior if trackbar calls SetCapture() while it already has mouse capture. Like Toolbar above, Trackbar only calls SetCapture() if it receives a LButtonDown. Again, another piece of code could in theory call SetCapture() and pass the trackbar's hWnd even when it already has mouse focus.
dlls/user32/button.c: Causes a button with mouse focus and BUTTON_BTNPRESSED to lose its highlighted state as documented at https://msdn.microsoft.com/en-us/library/aa931528.aspx. Like the two cases above, button would need to call SetCapture() on itself for this patch's behavioral changes to be relevant. Button calls SetCapture() on KEYDOWN, LBUTTONDBCLK, and LBUTTONDOWN.
I would be grateful if a seasoned Wine developer could chime in as to whether or not the above scenarios are possible, or what further research they would recommend. Otherwise, one safe course of action could be to emulate the existing Wine behavior of requiring the newly mouse-focused hWnds to be different from the existing hWnd in order for these three code paths to run (thus maintaining the current behavior for these specific cases). That would correct the SetCapture() behavior, fix the documented bug, and prevent any side effects in this code, though I'd prefer to better understand the above three code paths than rely on that idea.
As you said, other code may call SetCapture, or the controls themselves could be modified to call it at some later date. So the window procs should be updated to verify that the capture has really changed before proceeding.