This code was added in 21851c0673 but
looks completely broken:
- the status code is "206 OK" but "206" would be "Partial Content"
- the "Content-Length" header has a bogus value
- the "Content-RangeX" parameter has different bogus values (why
"Content-RangeX" anyway and not "Content-Range"?)
Apart from that, there are strange undocumented non-standard headers
which are probably there to work around bugs/expectations in one
broken proprietary client product. But these days, MPD doesn't bend
over to support broken clients. So let's kill this code.
Closes#304
For remote files (not streams), this downloads as quickly as possible
to a large buffer instead of throttling the stream during playback.
Throttling can make the server impatient and it may then disconnect.
This is what Qobuz and Tidal do, and this commit attempts to solve
this by not letting the Qobuz/Tidal server wait (closes#241).
Error message sent to client was "basic_string::_M_construct null not
valid" due to passing nullptr to the std::string constructor.
Regression caused by commit 386688b87a
When switching to another song manually, the player checks if the
decoder is already decoding that song; if so, it will attempt to reuse
it by seeking it to the new position. That however fails if the
decoder is not seekable (e.g. a radio stream) which leaves the user
unable to switch to that song with the bogus error message "Not
seekable".
Instead of stopping playback (due to seek time overflow), reject the
seek command. Closes#240
Relative negative values (with "seekcur") are still allowed, and MPD
will fix the resulting position if it turns out to be negative. But
the "seek" and "seekid" commands use an unsigned time stamp which must
not be negative.
With Grand Central Dispatch used in Main.cxx, debug builds on macOS
crash as the IsInside() assertion gets triggered in the event loop. As
a simple fix, usage of GCD is removed. Plugging and unplugging
headphones or changes of the default output device was tested without
issues. Whatever the original commit tried to fix by GCD probably does
not need fixing anymore.
From: Christian Kröner <ckroener@gmx.net>
This just copies the necessary bits and pieces from the ALSA plugin and applies them to OSXOutput based on dop config setting. It only changes the OSXOutput plugin as needed for DoP (further changes to support additionally e.g. integer mode or setting the physical device mode require rather a complete rewrite of the output plugin).
Fortunately the Core Audio API is by default bit perfect and supports DoP with minimal changes (setting the sampling rate accordingly after ensuring that the physical mode supports at least 24 bits per channel seems to be enough). This was tested on an Amanero Combo384 device hooked up to a ES9018 DAC.
USAGE (try only on DACs that support DoP):
- Add dop "yes" option to mpdconf
- Be sure to set at least 24bits per channel before playing some DSD file (using Audio-MIDI-Setup)
- Based on the dop setting, MPD will change the sample rate as required and output DoP signal to the DAC
- Hog mode is recommended to ensure that no other program will try to mix some output with the DoP stream (resulting in bad noise)
- Alternatively set the default output device to another device (e.g. the built-in output) to avoid having other audio interfere with DSD playback
support for chaining ogg opus streams to enable changing stream' metadata on the fly.
currently support is opt-in (enabled by additional option) because lots of clients can't handle this properly yet.
This addresses two problems:
1. the libFLAC write callback had to send an error status to its
caller when SubmitData() returned a command; this disrupted libFLAC
and the resulting command could not be used for anything;
2. the libFLAC function FLAC__stream_decoder_seek_absolute() also
calls the write callback, but its result cannot be used, because
seeking is still in progress, so we lose all data from one FLAC frame.
By moving the SubmitData() call until after CommandFinished(), we
avoid losing this data. This fixes another part of #113
Instead of passing whole chunks to the MusicPipe and checking the
end_time after each chunk, truncate the last chunk if it would exceed
the end_time. This requires keeping track of the absolute PCM frame
number.
This fixes a problem with gapless CUE song transitions: a small part
of the following song was always played twice.
Closes#113
Due to rounding errors, a slightly negative value can be passed to
set_normalized_volume(), which will make the log10() call fail.
Actually, volume 0 is already failing because log10(0) is illegal. So
let's fix this by implementing two corner cases: <=0 and >=100.
Closes#212
There is a POSIX definition for sched_setscheduler(), but Linux does
not implement that; instead of changing the process's scheduler, it
only affects one thread. This has caused some confusion among
application developers and C library developers.
While glibc implements Linux semantics, Musl has made their
sched_setscheduler() function an always-failing no-op, causing the
error message "sched_setscheduler failed: Function not implemented".
http://git.musl-libc.org/cgit/musl/commit/src/sched/sched_setscheduler.c?id=1e21e78bf7a5c24c217446d8760be7b7188711c2
Instead of relying on the C library which may be unreliable here, we
now roll our own system call wrapper.
Closes#218
This attribute shall be used only for IsInside() to make this safe
against a race condition described in #188:
> There is no requirement on the implementation that the ID of the
> created thread be available before the newly created thread starts
> executing.
http://pubs.opengroup.org/onlinepubs/009695399/functions/pthread_create.html):
This means that on some pthread implementations (e.g. Haiku), the
assert(thread.IsInside()) could fail.
Closes#188
This completes the bug fix commit
2065e3290452377b2931f3129b230c8cc536cbc8; if we clear "queued" then we
must clear "queued_song" as well, or another variant of the assertion
fails.
In current Android SDK releases, Ant support was removed. Move the
necessary build steps from the former Ant build system to our Makefile,
and call the required build tools from the Android SDK (aapt and dx),
Java SDK (javac) and Info-ZIP (zip) directly.
[mk: copied from Felix's commit
e52b906dba971a1173f9e8f83d32b52ee9f89af3 in the XCSoar project)
Supporting GCC 4.9 is becoming cumbersome as we refactor more code to
C++14.
This shouldn't be a problem for anybody, because Debian Stable
(Stretch) has version 6.3, and Ubuntu LTS (Xenial) has version 5.3.
Fixes deadlock because FileInputStream::Read() unlocks the mutex
(which was not locked) and then locks it, keeping it locked. This can
result in a deadlock. This happens because the archive and the file
mutex are different.
If a directory is a mount point, omit the "directory: " as well.
This bug is years old, but has become more visible now that mount
points are persistent in the state file.
These attributes are printed in the "outputs" response, and the new
command "outputset" allows setting new values.
No attributes are currently implemented.
If `SimpleDatabase::Visit` is called on a database that contains a mounted directry the URIs of the elements passed to the callbacks are not prefixed by the mountpoint path. This leads to lsinfo and add not working because they use the wrong URI. This pull request is using the `WalkMount` helper function to create prefixed versions of `VisitDirectory`, `VisitSong` and `VisitPlaylist` to add the correct prefix to the parameters of the callback functions.
[mk: the following text was copied from
https://github.com/MusicPlayerDaemon/MPD/pull/167]
For certain format (hi-res files) and normal buffer size hardware, The
hardware may at once consume most of the buffers. However, in Delay()
function, MPD is supposed to wait for 25 ms after the next try. it
will create a hiccup. The negative impact is much major than
increasing the latency.
I understand larger buffers come at a price. That's why in my earlier
commit last year I significantly reduced it. However, the buffer size
in CoreAudio is set according to the hardware, which is super small
latency. For instance, the system audio of 2015 generation of macbook
pro has maximum buffer size of 4096 samples, which is just 0.09s for
44.1k framerate, or 0.04s for 96k frames --- . compare to the 0.5 sec
latency alsa plugin has, even if we quadruple it, it's still super
tiny.
libupnp 1.6.24 added a few badly designed macros which break the MPD
build:
8177a4195a/
To work around this, we disable our emulation functions (from
714011c81e) on this libupnp version.
Closes#163
Without the flush, ReadPage() may not return any data, or not all
data. This may result in incomplete ddata the new "header" page,
corrupting streams with some encoders such as Vorbis.
Fixes#145
Don't require a quality or bitrate setting. If nothing is set, don't
fail startup - just go with a good default. A quality setting of 3 is
what "oggenc" defaults to as well.
InputStream::Read() and InputStream::Seek() are called with the mutex
locked. That means the implementation must not block, or unlock the
mutex before calling into blocking code.
Previously, a slow CD drive could stall the whole MPD process,
including the main thread, due to this problem.
Closes#149
Turns out that using CP_ACP is a lousy idea, because only very few
Unicode characters can be represented by it. Instead, switch to UTF-8
(which every sane person on other operating system already uses).
Closes#102
Our previous use of Queue::SwapOrders() could cause surprising
results:
- sometimes, the old "current" song would be played again (if the
newly selected song had not been played already)
- sometimes, the old "current" song would not be played again (if the
newly selected song had already been played)
This is inconsistent, because it should not depend on whether the
newly selected song had already been played.
So instead of Queue::SwapOrders() we now use Queue::MoveOrderAfter()
and Queue::MoveOrderBefore(), which is more expensive, but also more
consistent. It attempts to retain as much from the previous order
list as possible, and only moves the newly selected song around.
If an early exception gets caught (e.g. from
AllocatedPath::FromUTF8Throw()) before
DecoderControl::CommandFinishedLocked() is called, the decoder thread
would go in an endless loop, because DecoderCommand::START is still
set.
Closes#118
Our IcuCaseFold() fallback using strxfrm() is not actually case
insensitive. This commit fixes the problem by switching to
strcasecmp(). That function is not guaranteed to support UTF-8, but
it's the best we can do in this sparse situation.
Closes#111
RoarAudio's sndio emulation has been a source for annoyances. First,
their headers turned out to be broken with C++, due to their use of
the "new" keyword. Then they used a preprocessor macro to rename
"sio_hdl" to something else, effectively disallowing the use of
forward declarations. Enough is enough, and I'm removing support for
it.
RoarAudio users should better use the RoarAudio output plugin.
This commit is similar to 788e3b31e1,
and removes more "pure" attributes which were placed on functions that
could throw exceptions, which is illegal according to clang's
understanding of the attribute (but not according to GCC's). GitHub
issue #58 was most likely about StorageDirectoryReader::GetInfo() and
Storage::GetInfo(), which still had "pure" attributes.
Closes#58
Fixes build failure on OS X, closes#44. With the other plugins,
that's not critical, because those use the AudioOutputWrapper, which
hides this problem.
The "pure" and "const" attributes are not so well-defined, and a
recent clang version implements an optimization which pushes the
definition's boundary beyond what I believed it was. clang now
assumes that functions declared "pure" cannot throw exceptions, even
if they lack the "noexcept" specification.
When compiled with this new clang version, MPD will crash randomly if
an exception happens to get thrown by such as "pure" function
(https://github.com/MusicPlayerDaemon/MPD/issues/41).
This commit removes all such misplaced "pure" and "const" attributes,
closing #41.
An ino_t is usually a 64 bit integer, and some file systems (such as
Linux's kernel NFS client) really uses the upper 32 bit. This can
lead to false positives in the directory loop detection in
FindAncestorLoop(). Increasing these two attributes (in
StorageFileInfo and Directory) to 64 bit adds little overhead, but
makes the check a lot safer.
The TAG_MODIFIED handler (i.e. playlist::TagModified()) works only if
the modified song is the current song - something that is not updated
until SYNC_WITH_PLAYER is finished. This fixes tag updates right
after a new song is started.
https://bugs.musicpd.org/view.php?id=4656 describes a crash due to
division by zero because frame.samples==0. This should never happen,
but apparently can happen after seeking. The best we can do is to
just ignore this frame.
Fixes another buffer overflow: if the stream has a very long title or
URL, resulting in a metadata string of more than 2 kB, icy_string[0]
is a negative value, which gets casted to size_t - ouch!
https://bugs.musicpd.org/view.php?id=4652
Use SND_PCM_NONBLOCK, and perform all snd_pcm_writei() calls in the
IOThread. Use a lockless queue to copy data from the OutputThread to
the IOThread.
This rather major change aims to improve MPD's internal latency. All
waits are now under MPD's control, instead of blocking inside
libasound2.
As a side effect, an output's filter is now decoupled from the actual
device I/O, which solves a major latency problem with the conversion
filter on slow CPUs and small period buffers. See:
https://bugs.musicpd.org/view.php?id=3900
When rpc_reconnect_requeue() gets called from inside nfs_service(),
the NfsInputStream can stall completely because the old socket has
been unregistered from epoll automatically, but the new one has never
been registered. Therefore, nfs_service() will never be called again.
This kludge attempts to detect this condition by checking
nfs_which_events()==POLLOUT.
https://bugs.musicpd.org/view.php?id=4081
If the base class is not accessible, the "catching" the base class
won't work. This caused the fatal error:
terminate called after throwing an instance of 'LibmpdclientError'
Each close/open cycle resets the Filter's state, because a new Filter
instance is being created. That results in the serials
(replay_gain_serial and other_replay_gain_serial) being out of sync
with the internal ReplayGainFilter state.
So instead of initializing those serials once, we need to initialize
them each time we create new ReplayGainFilter instances, i.e. in
OpenFilter().
https://bugs.musicpd.org/view.php?id=4632
Previously, there was no special code to convert stereo to
multi-channel. The generic solution for this was to convert to mono,
and then copy the result to all channels. That's a pretty bad
solution, but at least something which always renders audio. MPD does
something, instead of failing.
Now that MPD has proper support for multi-channel (by defining the
channel order), we can do better than that. It is a (somewhat) common
case to play back stereo music on a DAC which can only do
multi-channel. The best approach here is to copy the stereo channels
to front-left and front-right, and apply the "silence" pattern to all
other channels.
If the input AudioFormat changes but the out_audio_format doesn't
change (e.g. because there is a fixed "format" setting in this
"audio_output" section), the ConvertFilter needs to be reconfigured.
This didn't happen, resulting in awful static noise after changing
songs.
This method is used by DecoderControl::IsCurrentSong(), which is used
by the player thread to check whether the current decoder instance can
be reused to seek. When switching to another song in the same CUE
sheet, previously DetachedSong::IsSame() returned true, and thus the
old decoder instance was used for the new song, not considering the
new end_time. This led to the old decoder quickly quitting.
This way, we have four periods instead of the default of two. With
only two periods, we don't get woken up often enough, and we
frequently encounter buffer overruns. With four periods, we have more
time to breathe, and the buffer overruns magically disappear.
The byte order of DSD_U32 was wrong from the start. The oldest bits
must be in the MSB, not in the LSB, according to
snd_pcm_format_descriptions in alsa-lib.
DSD_U32 packs four bytes instead of one large "sample", thus the
sample rate is one quarter of the input sample rate. This fixes a
rather critical DSD_U32 playback problem.
Changed AlsaMixerPlugin to use the get and set normalized functions from volume_mapping of alsa-utils/alsamixer
Changed volume_mapping set volume to be for all channels and not per channel
added volume_mapping files to Makefile.am
Without this, the pipe would run empty very often, which may result in
an xrun if the roundtrip to the PlayerThread and back takes too long.
By waking up the PlayerThread before the pipe runs empty, we make MPD
much more latency tolerant, which is a major optimization.
The user unit omits the "ProtectKernelModules" setting which fails
with modular kernels:
Failed at step CAPABILITIES spawning /usr/bin/mpd: Operation not permitted
It is unfortunate that systemd (version 232) is unable to reduce its
own capabilities, because this requires us to split system and user
units.
https://bugs.musicpd.org/view.php?id=4608
This commit changes a minor queue priority design to something which
makes a little bit more sense.
Previously, a song that had already been played would only be
re-enqueued if its priority had just been raised above the current
song's. This means that if it was already above, it was not
re-enqueued. That is a surprising behavior, because users expect a
song to be played when its priority is raised.
Now the song is always re-enqueued if its priority is raised (and
above the current song's - no matter if it has already been above
before).
https://bugs.musicpd.org/view.php?id=4592
The ScopeExit library uses C++11 initializers, which gcc 4.6 does not
support. Let's kill support for this ancient incomplete C++11
compiler, nobody should be using it anymore.
The "seeking" flag is not set for the initial seek, and so
decoder_read() could be canceled when another SEEK was emitted during
initial seek.
This fixes several seek problems, for example the one reported for the
FLAC decoder plugin:
https://bugs.musicpd.org/view.php?id=4552
.. instead of doing it after seeking. After seeking, the command had
no effect, because CheckDecoderStartup() waits for all outputs to
finish. This caused a very long delay while seeking and switching
songs (https://bugs.musicpd.org/view.php?id=4534).
Source: mpd
Version: 0.19.14-2
Severity: important
Justification: fails to build form source (but built in the past)
Tags: patch
User: debian-alpha@lists.debian.org
Usertags: alpha
mpd FTBFS on Alpha with a failure in the test suite [1]:
FAIL: test/test_byte_reverse
============================
.F...
!!!FAILURES!!!
Test Results:
Run: 4 Failures: 1 Errors: 0
1) test: ByteReverseTest::TestByteReverse2 (F) line: 58 test/test_byte_reverse.cxx
assertion failed
- Expression: strcmp(result, (const char *)dest) == 0
This occurs because the test suite (in test/test_byte_reversal.cxx)
allocates static char arrays and passes the char arrays to functions
whose respective arguments were declared to be uint16_t *, etc., in
the main code.
This is in the realm of undefined behaviour on architectures with
strict memory alignment requirements. Although the test only fails
on Alpha (because Alpha has a particular CPU load instruction that
gcc likes to use to add bugs ..., ahem, optimise the code on the
assumption of alignment) it is potentially a latent bug for other
architectures with strict alignment requirements.
Since the code is compiled with the c++11 standard I attach a patch
that modifies the test suite to align the non-compliant strings with
the alignas() attribute. The test suite now passes on Alpha with
that patch.
Cheers
Michael
[1] https://buildd.debian.org/status/fetch.php?pkg=mpd&arch=alpha&ver=0.19.14-2&stamp=1461542099
> In file included from src/decoder/DecoderBuffer.cxx:21:0:
> src/decoder/DecoderBuffer.hxx:41:20: error: 'uint8_t' was not declared in this scope
> DynamicFifoBuffer<uint8_t> buffer;
> ^
> src/decoder/DecoderBuffer.hxx:41:27: error: template argument 1 is invalid
> DynamicFifoBuffer<uint8_t> buffer;
> ^
> src/decoder/DecoderBuffer.hxx: In member function 'void DecoderBuffer::Clear()':
> src/decoder/DecoderBuffer.hxx:61:10: error: request for member 'Clear' in '((DecoderBuffer*)this)->DecoderBuffer::buffer', which is of non-class type 'int'
> buffer.Clear();
> ^
> src/decoder/DecoderBuffer.hxx: In member function 'size_t DecoderBuffer::GetAvailable() const':
> src/decoder/DecoderBuffer.hxx:78:17: error: request for member 'GetAvailable' in '((const DecoderBuffer*)this)->DecoderBuffer::buffer', which is of non-class type 'const int'
> return buffer.GetAvailable();
> ^
> src/decoder/DecoderBuffer.hxx: In member function 'ConstBuffer<void> DecoderBuffer::Read() const':
> src/decoder/DecoderBuffer.hxx:87:19: error: request for member 'Read' in '((const DecoderBuffer*)this)->DecoderBuffer::buffer', which is of non-class type 'const int'
> auto r = buffer.Read();
> ^
> src/decoder/DecoderBuffer.hxx:88:27: error: could not convert '{<expression error>, <expression error>}' from '<brace-enclosed initializer list>' to 'ConstBuffer<void>'
> return { r.data, r.size };
> ^
> src/decoder/DecoderBuffer.hxx: In member function 'void DecoderBuffer::Consume(size_t)':
> src/decoder/DecoderBuffer.hxx:105:10: error: request for member 'Consume' in '((DecoderBuffer*)this)->DecoderBuffer::buffer', which is of non-class type 'int'
> buffer.Consume(nbytes);
> ^
This seems to be caused by a lacking include, fixed by the below patch.
I'm unsure what made this appear now, though, compiler and toolchain
libraries seem to be the same upstream versions that built 0.19.14-1
just fine in late March.
When a reference counter is at its limit, don't allocate a new
TagPoolSlot - that would result in many TagPoolSlot instances with
ref==1. This in turn would make the linked list very very large,
which means quadratic runtime for many operations.
Apparently all other C libraries are not compatible with "constexpr".
Those which are not will get a performance penalty, but at least they
work at all.
MPD does not really take advantage of memory-mapped I/O by generating
data right into the ALSA buffer; using plain snd_pcm_mmap_writei() has
no advantage compared to snd_pcm_writei(). Let's kill this
non-feature.
The initgroups() manpage says we need to check for _BSD_SOURCE. The
thing is that glibc deprecated this macro, and doesn't define it
anymore, effectively breaking all MPD supplementary groups.
The real fix is to check for initgroups() availability at configure
time, instead of relying on the deprecated _BSD_SOURCE macro.
Apply padding only to the fseek(), not to the chunk size. This fixes
bogus "failed to read riff chunk" messages when the last chunk has an
odd size.
See http://bugs.musicpd.org/view.php?id=4486
systemd does not understand LimitRTTIME=-1. For no limit we have to use
the string 'infinity' (see systemd.exec(5)).
Signed-off-by: Christian Hesse <mail@eworm.de>
This reverts commit d7d9dbd2c2 by
reimplementing it with the current MPD API.
3 years ago, I was wrong about the "embcue" plugin being able to
replace this one, because "embcue" reads a tag named "CUESHEET", while
this plugin reads the "CUESHEET" FLAC metablock. There's an important
difference between those two!
Allocate the buffer dynamically using av_malloc(), and free
AVIOContext.buffer in the destructor, as mandated by the libavformat
documentation.
Fixes http://bugs.musicpd.org/view.php?id=4446
Wildcard matches are directly applied to all filenames in
subdirectories without any attempt at matching relative paths.
This change is based on the following feature request:
http://bugs.musicpd.org/view.php?id=3729
Use the first INDEX in each TRACK section, instead of the last, for the
start time. This preserves the original CD layout (including gaps
between tracks), and avoids skipping sections of songs in more exotic
cuesheets (eg musical suite tracks).
Fixes 0004355 and 0003359
If the song tag comes from a stream, and MPD playback restarts, MPD
would believe the tag should override the newly received tag. This
makes the previous tag appear stuck. This change passes the song tag
only if it's authoritative - i.e. if it's a song file.
Right after booting, the monotonic clock starts with a very small
value, and AudioOutput::LockUpdate() may believe that the fail_timer
has not recovered yet.
The Connect method can be called between Schedule and lock. In that case, when
locked, the state is already set to CONNECTING of READY and the condition won't
be signaled anymore.
Not initialising granulepos leads to it having arbitrary values in the
encoded stream including possibly negative values which are not valid
and confuse opusdec. Explicitly initialise opus_encoder::granulepos
to avoid that problem.
Requiring this prefix makes the client's intention very clear, but it
was too hard to understand why this prefix was needed. Initially, my
intention was to differentiate from broken clients which prefix relate
URIs with a slash; once MPD allowed that. In the past few years
however, MPD has disallowed that, and there was no significant
breakage (except for the "add /" special case which some clients
apparently still do). So I figure it's about time to define that an
URI that begins with a slash points to an arbitrary file on the file
system.
The file handle is never reset to INVALID_HANDLE_VALUE, and thus the
destructor will assume the operation shall be cancelled and will
delete the temporary file.
This was a major breakage for saving the database file and the state
file.
Build a table of pre-existing tag types before adding new items. The
old way would check HasType() each time, which would return true after
the first instance of that tag type had been added, preventing
duplicate tag types to be merged.
This broke duplicate tag types loaded from the state file, because
this code path uses TagBuilder::Complement().
This is Darwin specific: the previous implementation was causing an integer
overflow when base.numer is very large. On PPC Darwin, the timebase info is 1000000000/33330116 and this is too large for integer arithmetic.
This is Darwin specific: the previous implementation was causing an integer
overflow when base.numer is very large. On PPC Darwin, the timebase info is 1000000000/33330116 and this is too large for integer arithmetic.
The Linux feature allows writing new files to an invisible file, and
then replace the old file. This preserves the old file if we get
interrupted by some event.
Fixes a problem with the "curl" input plugin: IsEOF() always returns
true because the "open" flag was cleared by
CurlInputStream::RequestDone() when end-of-stream was reached. This
flag stays false even when seeking to another position has succeeded.
This patch resets the "open" flag to true after seeking successfully.
NetBSD's pthread_setname_np() prototype is incompatible with the rest
of the world, and it requires to pass the string argument as a
non-const pointer. Instead of working around this misdesign, I hereby
disable the feature on NetBSD.
Here's a change to dynamically allocate the DSD ID3 tag buffer.
Pretty much anything with cover art is going to exceed the existing,
static 4k limit... Here's a change to dynamically allocate the buffer
and sanity check it at some upper limit. I rather arbitrarily pulled
256k out of thin air just to keep a corrupt file from causing it to
trying to allocate a buffer larger than available memory.
When mounting had not yet finished, SocketMonitor::IsDefined() was
always false, due to the workaround at the beginning of the function
that calls SocketMonitor::Steal(). This commit drops the IsDefined()
check because it was never necessary and breaks reconnect.
nfs_destroy_context() will invoke all pending callbacks with
err==-EINTR. In CancellableCallback::Callback(), this will invoke
NfsConnection::DeferClose(), which however is only designed to be
called from nfs_service(). In non-debug mode, this will leak memory
because nfs_close_async() is never called.
Workaround: before nfs_destroy_context(), invoke nfs_close_async() on
all pending file handles.
The method NfsConnection::CancellableCallback::Callback() will always
invoke NfsConnection::Close() on the file handle, even if the void
pointer is not a nfsfh. This can happen if the Open() was not
successful, e.g. when the file does not exist.
MPD used both "album artist" and "albumartist" tags and mapped them to one tag.
This could lead to issues, if a file had both tags, causing MPD to send
a list of albumartists instead of a single one.
Since "album artist" is not a standard tag anyway and even its originators
started to use the proper alternative, its time to say goodbye!
