This reverts commit a446ef28.

This rounds up to exactly 441 samples with 44.1kHz audio
This avoids a rounding error in drift calculation of 2.31 milliseconds
in the worst case of the buffers queue being full for the specific
case of 44.1kHz audio (such a small error is not significative).

For unknown reasons it also fixes A/V sync issues.

Also fixes a memory leak

macosx: only revert presentation options on fullscreen leave which are touched when entering fs for this particular window

This fixes menu / dock hide behaviour when handling around with multiple
vout windows which are set to fullscreen on multiple monitors, especially when
leaving fullscreen for one window with other windows still in fullscreen state.

Signed-off-by: Rémi Denis-Courmont <remi@remlab.net>

Signed-off-by: Jean-Baptiste Kempf <jb@videolan.org>

Signed-off-by: Rafaël Carré <funman@videolan.org>

Signed-off-by: Rafaël Carré <funman@videolan.org>

Signed-off-by: Rafaël Carré <funman@videolan.org>

Signed-off-by: Rafaël Carré <funman@videolan.org>

Signed-off-by: Rafaël Carré <funman@videolan.org>

Signed-off-by: Rafaël Carré <funman@videolan.org>

Signed-off-by: Rafaël Carré <funman@videolan.org>

Due to the way this struct is allocated, it can be extended later
without breaking ABI.
Signed-off-by: Rémi Denis-Courmont <remi@remlab.net>

Signed-off-by: Jean-Baptiste Kempf <jb@videolan.org>

Signed-off-by: Jean-Baptiste Kempf <jb@videolan.org>

The structure of the decode functions are like this:

Decode() {
    if (got output buffer) {
        consume output buffer;
    }
    wait for input buffer;
    if (got input buffer) {
        fill input buffer and send to the codec;
    }
}

This makes the functions prone to deadlock, if all input buffers
have been consumed and won't be returned by the codec until we
consume some of the output buffers (and we don't get the output
buffers until we already are waiting for an input buffer).

By using a timeout in the wait for input buffers, we get out of the
deadlock and can consume an output buffer, allowing the codec to
proceed. By making the timeout relatively large, it won't trigger
during normal decoding, but gets us out of possible deadlocks
(noticed at startup with qualcomm audio decoders).

The alternative to timeouts would be to push dummy/sentinel buffers
into the fifo (as we already do at reconfiguration) each time we
receive an output buffer. This would add quite a bit of overhead
though, having to process (and skip) one item in the input buffer
fifo for each received output buffer.
Signed-off-by: Martin Storsjö <martin@martin.st>

This way we don't need to define dummy GL_UNPACK_ROW_LENGHT for opengl es
compiles.

This proves ugly when changing skins

Signed-off-by: Rémi Denis-Courmont <remi@remlab.net>