music_pipe: added music_pipe_push()

Added music_pipe_allocate(), music_pipe_push() and
music_pipe_cancel().  Those functions allow the caller (decoder thread
in this case) to do its own chunk management.  The functions
music_pipe_flush() and music_pipe_tag() can now be removed.

src/decoder_internal.h

@@ -19,8 +19,11 @@
 #ifndef MPD_DECODER_INTERNAL_H
 #define MPD_DECODER_INTERNAL_H
 
+#include "decoder_command.h"
 #include "pcm_convert.h"
 
+struct input_stream;
+
 struct decoder {
 	struct pcm_convert_state conv_state;
 
@@ -31,6 +34,25 @@ struct decoder {
 
 	/** the last tag received from the decoder plugin */
 	struct tag *decoder_tag;
+
+	/** the chunk currently being written to */
+	struct music_chunk *chunk;
 };
 
+/**
+ * Returns the current chunk the decoder writes to, or allocates a new
+ * chunk if there is none.
+ */
+struct music_chunk *
+decoder_get_chunk(struct decoder *decoder);
+
+/**
+ * Flushes a chunk.  Waits for room in the music pipe if required.
+ *
+ * @return DECODE_COMMAND_NONE on success, any other command if we
+ * have received a decoder command while waiting
+ */
+enum decoder_command
+decoder_flush_chunk(struct decoder *decoder, struct input_stream *is);
+
 #endif