Previously, if the user set a custom size in the options, it wasn't respected but the snapshot was stored at the native video size

This fixes a regression introduced in e3b809c1

the SwapChain dimensions are read from shared private data in the DXGI swap chain
Signed-off-by: Jean-Baptiste Kempf <jb@videolan.org>

Fixes #14909

_Ios_Openmode is an implementation detail of the GNU c++ stdlib whereas ios_base::openmode is the portable way to go

This option was previously set by androidsurface.

This reverts commit 59678ec0.

Since vlc_atomic_float is only used for C files, another option would be
to disable this block when building in C++ mode.
Related to https://gcc.gnu.org/bugzilla/show_bug.cgi?id=64658

Signed-off-by: Hugo Beauzée-Luyssen <hugo@beauzee.fr>
cf. https://gcc.gnu.org/bugzilla/show_bug.cgi?id=60940

This prevents us from providing our atomic types definitions, as they
would conflict with the ones in std::

This commits remove the android_surface module. The old android surface API is
now used by the android_window module via native_window_api_t.

So far no keep-alive messages are sent when GET_PARAMETERS is not supported by
the server. Change this to fall back to sending OPTIONS as keep-alive message,
which is the default keep-alive message in the SAT>IP specification. As
RFC2326 does not enforce a specific command as keep-alive message this shall
be valid for any compliant RTSP server.
Signed-off-by: Julian Scheel <julian@jusst.de>
Signed-off-by: Jean-Baptiste Kempf <jb@videolan.org>

Register a satip-access, which uses a hardcoded default SDP dataset instead of
querying it via DESCRIBE. While some SAT>IP servers implement a common
DESCRIBE the spec is allowing servers to only support DESCRIBE of already
configured sessions, which requires SETUP to be executed before DESCRIBE can
be used at all.
Signed-off-by: Julian Scheel <julian@jusst.de>
Signed-off-by: Jean-Baptiste Kempf <jb@videolan.org>

This fixes mismatched binary interface when either:
 - the versions of C and C++ were different within the LibVLC build, or
 - an out-of-tree plugin used a different language versin than LibVLC.

The LibVLC implementation for ISO C atomics with Intel intrinsics was
not generally binary compatible with the native implementation of the
compiler. In other words, only one atomics implementation can be used
in one LibVLC installation. The One Definition Rule from C++ imposes
that native implementation is used, so the LibVLC implementation is
hereby removed.

Regression from edf1c760.

Also error out if necessary.

Lock is not needed anymore, as this variable is only accessed
from the main thread.

The vlc target only creates a symlink, so it makes no sense to
depend on that in any way.

This also fixes unneeded recreation of the package with every
make invocation.

According to docu, application:willFinishLaunching: will be called
as one of the first steps of [NSApp run], before the main loop
will be actually started. Thus, it should make no performance
difference if we load the main window already in openIntf().

For some unknown reasons, a direct call to makeKeyAndOrderFront
works in run(), but not in app:willFinishLaunching:, though.

This moves the main loop out of the macosx interface module.
Instead, the open callback only initializes the interface and the
close callback cleans up stuff. The mainloop is now started in
darwinvlc.m. In case the mac interface is not used, a CoreFoundation
mainloop is solely started to wait for termination events.

Additionally, this cleans up darwinvlc.m, and fixes signal
handling, which was dead code so far (in case the interface was used).
Now, GCD is used to catch SIGINT and SIGTERM in order to allow
ordinary shutdown.

refs #14362
close #6354

As we do not want to break other os versions, the only way to
select the proper linker is to use another target for osx.

Still, "make install" should produce a binary named "vlc" in the
end, thus the binary is renamed at install stage.