If TransformView_Create returns an error, it should not return a pointer
that table_load_transform will try to free.

Fixes double frees in read_components and read_dependency. The double
frees could also be fixed by using free instead of free_dependency in
those functions, but zeroing out the pointers is easier to understand
because it ensures that alloc_dependency can always be paired with
free_dependency.

Fix memory leaks on error paths, don't call memset unnecessarily, and
ensure that all output fields are initialized.

We don't need the window anymore, it was only used to send ClipCursor
notifications. This improves cursor clipping performance a lot as it
avoids re-creating a window every time.

If the focus changes between Wine windows, the wineserver logic will
decide to reset the clipping rectangle. However winex11 also needs to
support the case when focus changes to a host window, in virtual desktop
mode, and in this case the foreground window doesn't actually change.

To fix this, in virtual desktop mode, release the cursor on focus out
events, and reapply the cursor clipping rect when the virtual desktop
window is focused again.

We can use the same logic on NotifyGrab events, when the WM grabs the
keyboard, and later reapply the Wine clipping rect when we are notified
about the keyboard ungrab.

Wine-Bug: https://bugs.winehq.org/show_bug.cgi?id=55046

This might be the case when in virtual desktop mode for instance, where
we don't change Wine foreground window when the virtual desktop is
focused out.

Wine-Bug: https://bugs.winehq.org/show_bug.cgi?id=55046

Wine-Bug: https://bugs.winehq.org/show_bug.cgi?id=55047

Wine-Bug: https://bugs.winehq.org/show_bug.cgi?id=55047

Signed-off-by: Fabian Maurer <dark.shadow4@web.de>

It's more elegant to put ole32 in the DELAYIMPORTS.

Signed-off-by: Bernhard Kölbl <besentv@gmail.com>

Programs might expect objects inside the queued toplogy to be freed,
before they eventually call release on the session itself.

This fixes reference leaks to stored objects in queued topology nodes,
even when IMFMediaSession_Shutdown() was called.
Signed-off-by: Bernhard Kölbl <besentv@gmail.com>

Wine-Bug: https://bugs.winehq.org/show_bug.cgi?id=55043

Signed-off-by: Nikolay Sivov <nsivov@codeweavers.com>

Also introduce async helpers.

The buffers cannot be freed directly in wave_out_proc, because calling
waveOut related functions in the callback could cause a deadlock.