Turn the music_pipe into a simple music_chunk queue. The music_chunk
allocation code is moved to music_buffer, and is now managed with a
linked list instead of a ring buffer. Two separate music_pipe objects
are used by the decoder for the "current" and the "next" song, which
greatly simplifies the cross-fading code.
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.
Now that I've found this nice function in the GLib docs, we can
finally remove our custom sleep function. Still all those callers of
g_usleep() have to be migrated one day to use events, instead of
regular polling.
The decoder_plugin struct is used by both the MPD core and the decoder
plugin implementations. Move it to a shared header file, to minimize
header dependencies.
One of the previous patches made MPD consume 100% CPU in a busy wait:
when the music_pipe was full, it did not wait (with notify_wait()) for
free chunks, because a variable has a different meaning now. Always
pass "true" as the "wait" parameter.
This patch fixes a minor memory leak: when decoder_tag() attempted to
send a merged tag object (created by tag_add_stream_tags()), and was
interrupted by a decoder command, it did not free the temporary merged
tag object.
It is illegal to pass an empty audio buffer around. pcm_resample()
sometimes seems to result in 0 samples, maybe related to
libsamplerate. To work around that problem, add special checks after
both pcm_convert() invocations. Removed the pcm_resample()==0 checks
from pcm_convert().
When a global audio format is configured (setting
"audio_output_format"), decoder_data() overwrote the "length"
parameter with the size of the output buffer (result of
pcm_convert_size()). Declare a separate variable for the output
buffer length.
Seeking was somewhat broken in some decoder plugins because they sent
empty chunks, and never got a command. Check the decoder command
before doing anything else in decoder_data().
The assertion on dc.state in decoder_read() was too strict: when a
decoder tried to call decoder_read() from tag_dup(), the decoder state
was NONE. Allow this special case.
If an input stream provides tags (e.g. from an icecast server), send
them in the decoder_data() and decoder_tag() methods. Removed the
according code from the mp3 and oggvorbis plugins - decoders shouldn't
have to care about stream tags.
This patch also adds the missing decoder_tag() invocation to the mp3
plugin.
Don't pass the "seekable" flag with every decoder_data() invocation.
Since that flag won't change within the file, it is enough to pass it
to decoder_initialized() once per file.
.. and rename dc.audioFormat to dc.in_audio_format. The music pipe
does not need to know the audio format, and its former "audioFormat"
property indicated the format of the most recently added chunk, which
might be confusing when you are reading the oldest chunks.
No CamelCase in the file name. The output_buffer struct is going to
be renamed to music_pipe. There are so many buffer levels in MPD, and
calling this one "output buffer" is wrong, because it's not the last
buffer before the music reaches the output devices.
A decoder_flush() invocation was missing in the FLAC plugin, resulting
in casual assertion failures due to a wrong assumption about the last
chunk's audio format. It's much easier to remove that decoder_flush()
function and make the decoder thread call ob_flush().
Instead of manually calling memset(0) on the pcm_convert_state struct,
client code should use a library function from pcm_utils.c. This way,
we can change the semantics of the struct easily.