wmic currently crashes on, e. g., VT_EMPTY or VT_ARRAY types returned.

Otherwise enumeration fails on Windows.

The result access is not forward only.

Wine-Bug: https://bugs.winehq.org/show_bug.cgi?id=53814Signed-off-by: Nikolay Sivov <nsivov@codeweavers.com>

Apps that issues many small draw calls (e.g. Earth 2150) may cause
frequent DISCARD maps, which are significantly slower than NOOVERWRITE
ones.

Wine-Bug: https://bugs.winehq.org/show_bug.cgi?id=45273

Wine-Bug: https://bugs.winehq.org/show_bug.cgi?id=45273

This will end up moving window around only if the old virtual screen
origin is different from the new one. Don't send unnecessary messages
if it didn't change.

This helps speeding up D3D tests on the testbot VMs with multiple
displays and hopefully solve the timeout problems.

When f7332ab4 switched load_so_dll
over to a proper Unix syscall, ntdll.so started providing the
symbol __wine_unix_call_funcs.

When a DLL calls NtQueryVirtualMemory(MemoryWineUnixFuncs), it can
do so speculatively, not knowing whether the corresponding unix
lib actually does provide anything.

When ntdll.so started providing this symbol, then such speculative
calls to NtQueryVirtualMemory(MemoryWineUnixFuncs) would end up
returning ntdll's __wine_unix_call_funcs if the intended DLL's
unix library lacked one.

This is the case with bcrypt, where the unix library doesn't
provide __wine_unix_call_funcs if gnutls is unavailable.

This fixes crashes after f7332ab4,
in builds without gnutls.
Signed-off-by: Martin Storsjö <martin@martin.st>

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

In ept_map, rpcss uses TowerConstruct to allocate tower array elements.

The array is later released by the generated stub code, and each element
is ultimately freed using MIDL_user_free.

TowerConstruct always allocate memory using I_RpcAllocate, we must use
I_RpcFree to free it.