None of the call sites use the input thread and wait pipe.

If net_Write() returns a short count, send() is not going to help. It
means the connection was closed.

Allows using tls transport and does
non persistent connection fallback.

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

Libass does the same, and I think it's a good idea. But their method is IMO overly complicated.
Or else I'm over-simplifying things. But this seems to work very well.
Signed-off-by: Jean-Baptiste Kempf <jb@videolan.org>

It should be LayoutLine(). To be consistent with LayoutParagraph() and LayoutText().
Signed-off-by: Jean-Baptiste Kempf <jb@videolan.org>

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

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

FixGlyph() expects to be called immediately after FT_Load_Glyph(), which is no longer the case.
Signed-off-by: Jean-Baptiste Kempf <jb@videolan.org>

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

Those patches prevent an infinite loop when specifying an interface name, and
avoid using a down interface as the defautl interface.
fix #14526

With waitall set to false, net_Read() is identical to recv()
outside the input thread. This avoids allocating a bogus waitpipe.

net_Read() is identical to recv() outside the input thread and if
waitall is false.