This shouldn't happen unless the codec is buggy

Do not start live playback when there is not enough data available. There
should at least be 3 times the target duration of data segments.

The httplive stream_filter ignored HTTP port numbers other then 80 and 443.
These where dectected by HTTP access input. Other port numbers were simply
ignored.
The patch now uses the original portnumber and carries it along when
handling relative_URI() and download URI in AccessOpen() functions.

This reverts commit 01dbb084.

I must have had something really wrong going on in scratchbox for this
commit to fix things.

Untested

This simplifies memory handling in most cases.

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

The attribute name is valid until the next attribute or next node.
This simplifies the code a bit: fewer error paths, and free() calls.

A few potential memory leaks were fixed too.

Rename preferences entries, put in the right categories, etc...

This is used in a very particular situation
(language code changing continuously on a single PID, signaled by a
change in teletext data), but maybe other uses can be inspired from it.
Signed-off-by: Jean-Baptiste Kempf <jb@videolan.org>

Work based on Christophe Massiot setlang module, merged by /me

Sorry, I failed my merge ;)

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

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

 * modules/access/linsys/*: add support for SDI and HD-SDI cards of
   Linear Systems/Computer Modules
Signed-off-by: Jean-Baptiste Kempf <jb@videolan.org>

It is behaving a little bit different from the libxml parser which has
caused weird bugs in the past. Anyway, everybody has libxml (I am
planning to write an Expat parser too but not there yet).

commit id: c5f5039b defined
a shadow variable psz_path which content is needed to correctly
construct a relative_URI().