This fixes extracting of the tarball when master has progressed
from the specified hash.
Signed-off-by: Martin Storsjö <martin@martin.st>

Ref #8850

Don't ask me why the original developer needed to use mismatched type
names here.

Fixes: #8850

We now output 20 and 24 bits PCM into S32N
Don't use swab, as it is forbidden for an odd number of channels

Fixes: #8982

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

The streams lists were just moved out from direct.js to streams.js.

Streams still use crappy proprietary formats with ugly resolution,
can't do anything for this... If only they used a fixed playlist...
Signed-off-by: François Revol <revol@free.fr>
Signed-off-by: Rafaël Carré <funman@videolan.org>

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

Ref: #8983

The previous value was quite bogus.

This new implementation opportunistically tries to read data before
invoking poll(). This reduces latency when data is already waiting in
buffers, allows receiving pending data in non-blocking fashion even if
the object has been killed. This also gives GnuTLS a chance to dequeue
data from internal buffers.

Also the corner case of 0 read should now be handled correctly.

The corner case of an error after success read is no longer handled.
This was race-prone and useless.