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.
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>
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
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.
