If END is omitted, mpd should use the highest possible value instead of raising an error.
This partially reverts 52e9cab1

When all outputs fail, MPD automatically pauses playback, but did not
emit the IDLE_PLAYER event.  Thus, clients were not woken up.  (Mantis
3665)

This file is generated during configuration and should be ignored.

- Add mandatory "name" setting for winmm/openal examples.

- Fix the language in openal sample (a -> an)

- Added replaygain_missing_preamp and replaygain_limit settings

- Added information about "auto" replay gain mode

At least in my configuration libgme lacks .pc file
thus could not be detected with pkg-config.

Track whether _finish() has been called, and deliver all partial
results then.  Fixes Mantis ticket 0003621.

Rationale: vanilla libid3tag does not have any pkg-config stuff
and fails to detect because symbols from libz are not found.

- fix potential memory leak of system_path

  'Potential' because currently g_get_system_config_dirs()
  returns single entry on Windows, but that might change.

- remove incorrect g_free() call

  It's not required at all because
  g_get_system_config_dirs() returns GLib owned memory.

- remove extra semicolon

Since some configurations use the "bind_to_address" option in their
/etc/mpd.conf, the systemd service file must wait for the
network.target or else mpd will start before it and thus fail due to
no iface.

The song's URI must be UTF-8, not filesystem character set.

Fixes minor Windows compatibility problem.

Implements Mantis feature request 3582.

Interleaving was completely wrong.  This code was never used, so it
didn't have an effect.

Works around assertion failure in decoder_timestamp().

Use av_get_sample_fmt_string() to obtain a human-readable string.

API simplification.

This is necessary for Ogg packets that span more than one page.

We can only check for negative values if the variable is signed.

Make sure the file header gets written at the beginning, before
_write() gets called.