strerror_r() will still crash as setlocale() is used regardless of
ENABLE_NLS.

This reverts commit 61d0ac25.

If mac users can test ... (evenif it's really simple)

and doesn't seem to affect any regression on demuxing.

Signed-off-by: Rémi Denis-Courmont <rem@videolan.org>

Assume RTP converys video (which is fine for MPEG-TS), as we cannot
know if it's only audio anyway.
Signed-off-by: Rémi Denis-Courmont <rem@videolan.org>

Signed-off-by: Rémi Denis-Courmont <rem@videolan.org>

Signed-off-by: Rémi Denis-Courmont <rem@videolan.org>

but NOT from var_OptionParse().

Rationale: At a shallow level, this breaks the command line use badly.
At a deeper level: We still do security enforcement in
var_OptionParse(). In practice, the config chain strings are always
coming from (part of) the value of string configuration variable,
which is parsed by var_OptionParse(). Hence, as long as these variables
are all tagged as "unsafe", only trusted input can ever reach
config_ChainParse().

There are other (intractable?) issues with checking configuration chain
strings: First, we would also need to check the module name in front of
the {var1=val1,...} stuff. Second, some modules parse their
configuration chain manually, i.e. they don't call config_ChainParse()
and sometimes do not register their variables as configuration items
(e.g. the sout duplicate module).

If you have a super-duper idea on how to improve this, you're welcome,
but in the mean time...
Signed-off-by: Rémi Denis-Courmont <rem@videolan.org>
Vaguely-acked-by: Pierre d'Herbemont <pdherbemont@free.fr>

But its still very time critical, because paout gets sometimes empty
so that aout_OutputNextBuffer(..) doesn't deliver new audiobuffers,
sometimes it happens - that these buffers arrive very late for output.
Reasons: -changed CPU load, (extra running application, also fast forward
seeking and and jumping may lead to this situation.)

Enhancement #897: added an option to let the user  choose is prefered audio device. (needs restart of VLC to get applied)

Signed-off-by: Jean-Paul Saman <jpsaman@videolan.org>

Signed-off-by: Jean-Paul Saman <jpsaman@videolan.org>

Don't leak memory when filling in p_stream->p_decoder_specific_info for subtitles. The p_stream->p_decoder_specific_info pointer was already allocated just before checking for VLC_FOURCC('s','u','b','t'). The fix moves the non-subtitles case to an else.
Signed-off-by: Jean-Paul Saman <jpsaman@videolan.org>

This reverts commit 9438646c.
wrong file in my list I'am sorry, and wrong file modes

(but its still critical because it happens - aout_OutputNextBuffer(...) has no buffers
available - sometime the output doesn't know until some ms before the last block
has finished to play! so each little disturbtion of the CPU or load change
leads to studdering/hickups)

#897 enhancement: give the use the option to choose its prefered audio device,
instead of useing Windows default