When the decoder finishes the "queued" song very quickly (before the
"current" song finishes playing), an assertion in do_play() fails
because it thinks that it should start decoding the queued song,
although that has in fact just finished.
While paused, the player thread re-locks its mutex and waits for a
signal. This is racy: when the command is set while the thread is
waiting for the lock, it may wait forever. This patch adds another
command check before player_wait().
Use GMutex/GCond instead of the notify library. Manually lock the
player_control object before accessing the protected attributes. Use
the GCond object to notify the player thread and the main thread.
Right after seeking and song change, the elapsed_time shows old
information, because the output thread didn't finish a full chunk
yet. This patch re-adds a second elapsed_time variable, and keeps
track of a fallback value, in case the output thread can't provide a
reliable value.
Don't set the error in play_chunk(); do all the error handling in the
caller. The errored_song attribute isn't set anymore; it doesn't make
sense for PLAYER_ERROR_AUDIO.
With these methods, an output plugin can allocate some global
resources only if it is actually enabled. The method enable() is
called after daemonization, which allows for more sophisticated
resource allocation during that method.
When the audio output fails to open, MPD pauses playback, but doesn't
reset player.play_audio_format. This leads to an assertion failure in
audio_output_all_check() on the next REFRESH command, because no audio
output is open.
Tracking the "elapsed" time from the chunks which we have sent to the
output pipe is very imprecise: since we have implemented the music
pipe, we're sending large number of chunks at once, giving the
"elapsed" time stamp a resolution of usually more than a second.
This patch changes the source of this information to the outputs. If
a chunk has been played by all outputs, the "elapsed" time stamp is
updated.
The new command PLAYER_COMMAND_REFRESH makes the player thread update
its status information: it tells the outputs to update the chunk time
stamp. After that, player_control.elapsed_time is current.
Replace decoder_control.notify with decoder_control.mutex and
decoder_control.cond. Lock the mutex on all accesses to
decoder_control.command and decoder_control.state.
Calculate the total play time with the audio_format object each time,
using audio_format_time_to_size(). The function
audioFormatSizeToTime() is not needed anymore, and will be removed
with this patch.
Do all the software volume stuff inside each output thread, not in the
player thread. This allows one software mixer per output device, and
also allows the user to configure the mixer type (hardware or
software) for each audio output.
This moves the global "mixer_type" setting into the "audio_output"
section, deprecating the "mixer_enabled" flag.
When the decoder is finished, break out of the player loop only after
another player.pipe check. We did check the pipe size a few lines
above, but that check was kind of racy.
When a music_chunk only contains a tag but no PCM data, play_chunk()
returns true without freeing the chunk. The caller now assumes that
the chunk is moved into some music_pipe and does not bother to free it
either.
To check for leaked music_chunk objects, free the music buffer on
CLOSE_AUDIO. This invokes an assertion check which ensures that all
chunks have been returned to the buffer.
When all audio outputs have been closed due to failures, pause the
playback instead of stopping it. This way, the user may resume
at the current position after the problem has been dealt with.
When no audio outputs could be opened while seeking, leave MPD seeked
at that position and pause playback. The user may continue from this
point at any time, as soon as the audio outputs are fixed. The old
behaviour triggered an assertion failure: the failure wasn't passed
properly to the do_play() function, which attempted to play audio
chunks.
When the decoder initialization has not been completed yet, all calls
to dc_seek() will fail, because dc.seekable is not initialized yet.
Wait for the decoder to complete its initialization, i.e. until it has
called decoder_initialized().
This updates the copyright header to all be the same, which is
pretty much an update of where to mail request for a copy of the GPL
and the years of the MPD project. This also puts all committers under
'The Music Player Project' umbrella. These entries should go
individually in the AUTHORS file, for consistancy.
After a seek, wait until enough new chunks are decoded before starting
playback. If this takes too long, send silence chunks to the audio
outputs meanwhile.
When playback is unpaused, pass the audio_format to
audio_output_all_open(). Don't assume that output_all.c remembers the
previous audio format. Also check if there has been an audio format
yet.
Instead of passing individual buffers to audio_output_all_play(), pass
music_chunk objects. Append all those chunks asynchronously to a
music_pipe instance. All output threads may then read chunks from
this pipe. This reduces MPD's internal latency by an order of
magnitude.
When a PAUSE command is received while the decoder starts, don't open
the audio device when the decoder becomes ready. It's pointless,
because MPD will close if after that.
In !NDEBUG, remember which audio_format is stored in every chunk and
every pipe. Check the audio_format of every new data block appended
to the music_chunk, and the format of every new chunk appended to the
music_pipe.
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.
The warning message "problems opening audio device while playing ..."
does not help at all, and should be removed. At this point, the real
error message has already been logged by the output thread.
When a file is not seekable, MPD dropped the audio buffers before even
attempting to seek. This caused noticable sound corruption. Fix:
first attempt to seek, and only if that succeeds, call
audio_output_all_cancel().
The crossfading code shouldn't depend on the audio output code. Pass
the current audio format to cross_fade_calc() and let it compare
directly, instead of using isCurrentAudioFormat().
When we reset pc.next_song if there is no song queued, this might
cause a race condition: the next song to be played is cleared, while
pc.command was already set. Clear the "next_song" only if there is a
song queued for the current do_play() invocation.
After a player command (successful or not), reset pc.next_song,
because the queue is supposed to be empty then. Otherwise,
playlist.queued and pc.next_song may disagree, which triggers an
assertion failure.
Break from the loop instead of returning the function. This calls
player_stop_decoder(), which in turn emits the PLAYLIST event. This
allows the playlist to re-start the player.
The player_thread loop requests the next song from the playlist as
soon as the decoder finishes the song which is currently being played.
This is superfluous, and can lead to synchronization errors and wrong
results. The playlist already knows when the player starts playing
the next song (player_wait_for_decoder() triggers the PLAYLIST event),
and will then trigger the scheduler to provide the next song.
The "TAG" event is emitted by the player thread when the current
song's tag has changed. Split this event from "PLAYLIST" and make it
a separate callback, which is more efficient.
When the decoder of the new song is not fast enough, the player thread
has to wait for it for a moment. However the variable "nextChunk" was
reset to -1 during that, making the next loop iteration assume that
cross-fading has not begun yet. This patch overwrites it with "0"
while waiting.
Commit b3e2635a introduced a regression: when a stream tag was
changed, the playlist version had to be updated. This was done in
syncCurrentPlayerDecodeMetadata(), called by syncPlayerAndPlaylist().
After b3e2635a, this was not called anymore. Fix this by emitting
PIPE_EVENT_PLAYLIST.
There is only one location using PIPE_EVENT_SIGNAL: to synchronize
player_command() with player_command_finished(). Use the "notify"
library instead of the event_pipe here.
Make the event_pipe (formerly main_notify) send/receive a set of
events, with a callback for each one.
The default event PIPE_EVENT_SIGNAL does not have a callback. It
is still there for waking up the main thread, when it is waiting for
the player thread.
We are going to migrate away from the concept of notifying the main
thread. There should be events sent to it instead. This patch starts
a series to implement that.
I have found something that looks like a bug in MPD:
- When a song is finished, the next one is played and the 'player'
event is emitted.
- When the client sends the status command just after this event, the
songid is the new one but the 'elapsed' time is not reseted to 0.
This is problem because I have implemented the solution using a timer
on client side to compute the elapsed time but with this bug the
elapsed time continues to be incremented on a new song.
Don't send a "next song" request to the main thread when the current
song hasn't started playing yet, i.e. there are already two different
songs in the music pipe. This would erase information about the song
boundary within the music pipe, and thus triggered an assertion
failure. The bug could occur when playing very short songs which fit
into the pipe as a whole.
Fix a deadlock: when the decoder waited for buffer space, the player
could enter a deadlock situation because it waits for more chunks for
crossfading chunks. Signal the decoder before entering notify_wait().
When a tag is updated, the old tag was freed before the new one was
created. Reverse the order to be sure that other threads always see a
valid pointer.
This still leaves a possible race condition, but it will be addressed
later.
The player did not care about the exact error value, it only checked
whether an error has occured. This could fit well into
decoder_control.state - introduce a new state "DECODE_STATE_ERROR".
