Commit 44beae519d caused the MPD
executable to grow by more than 1 MB because the `slots` array
suddenly was not zero-initialized anymore, because the doubly-linked
list is circular, while the singly-linked list is zero-terminated.
This option moves the array back to the `bss` section.
Fixes UBSanitizer warning:
src/util/IntrusiveList.hxx:108:10: runtime error: downcast of address 0xdeadbeef which does not point to an object of type 'Foo'
This is a big simplification of all IntrusiveHashSet users: instead of
having to implement multiple overloads of Hash and Equal, the
IntrusiveHashSet class can first extract the effective key from an
item and then pass it to the Hash/Equal functions.
Preparing to add a key extraction function. Without this "Operators"
template parameter, we'd have even more template parameters, and that
parameter list would grow too complex. Better wrap it in one single
template that contains all operators.
This is an API change which all callers need to adjust to, but it will
be worth it.
The list head is a stable value that is guaranteed to be still valid
when insert_commit() gets called.
This fixes a linked list corruption bug in class StaticCache which
occurs when the cache item pointed to by the iterator gets evicted
between insert_check() and insert_commit().
6d91b5c7b2 ("fix double promotions") changed
how LAME peak values are decoded, producing large incorrect values that
cause some MP3 files to play silently.
Restore the original decode from MAD fixed-point format to double and
document what it's doing.
Fixes#1823
GNU libstdc++ declares cmath functions as constexpr, but libc++ does
not. The C++ standards state the standard library functions should not
be declared constexpr unless explicitly required.
This eliminates all `Storage::GetInfo()` calls from
`UpdateWalk::PurgeDeletedFromDirectory()` and instead uses a "marker"
field to mark items that have been visited; later, all unmarked items
can be deleted.
This eliminates a lot of redundant I/O which is noticable with the
`curl` storage plugin (i.e. WebDAV).
Let OnHeaders() check the status.
The status checking code was added by commit 4f021cbced in 2011,
but in 2008, commit a8e81326d0 enabled `CURLOPT_FAILONERROR`, which
means the status checking code never had any effect.
This allows `LoadExcludeListOrLog()` to hide boring "404 Not Found"
log messages via `IsFileNotFound()`.
The `assert(!quit)` can fail if the `EventThread` gets stopped before
it enters `EventLoop::Run()`. There is a similar problem with `alive`,
which gets reset by `EventThread::Stop()`.
If that happens, then `EventLoop::Run()` should return immediately
without handling any events.
Without clearing all `in_playlist` flags, the songs will never be
revealed again if they were hidden once by a CUE sheet, not even after
the CUE sheet gets deleted or modified.
Closes https://github.com/MusicPlayerDaemon/MPD/issues/1784
This adds `MakeFfmpegError()` to the executable and fixes a linker
failure when `libavutil` is available, but `libavformat` and
`libavcodec` are not.
Closes https://github.com/MusicPlayerDaemon/MPD/issues/1786
libfmt 10 doesn't know how to format a StringView, and doesn't cast to
std::string_view anymore. The StringView class has been removed from
MPD 0.24 completely, and this is a stable-branch-only workaround.
Closes https://github.com/MusicPlayerDaemon/MPD/pull/1814
This should prevent ffmpeg from taking priority over the gme plugin.
The ffmpeg plugin is more buggy than gme.
One of the prominent bugs of preferring ffmpeg over gme is that ffmpeg
cannot seek SAP files while gme can. This should prevent that from
happening.
libfmt version 10 apparently doesn't know how to format a
`StringBuffer`, failing the MPD build. Since we have a formatter
specialization for `AudioFormat`, let's use that - it's better and
easier to use.
Closes https://github.com/MusicPlayerDaemon/MPD/issues/1807
Fixes hide_playlist_targets not working after server restart
Currently, `hide_playlists_targets` works by skipping songs with
`in_playlist` value set to true in
[`Directory::Walk`](a57bcd0238/src/db/plugins/simple/Directory.cxx (L237)). But
`in_playlist` is not stored and only updated in
[`UpdateWalk::PurgeDanglingFromPlaylists`](a57bcd0238/src/db/update/Playlist.cxx (L139)),
which will only be executed while updating DB.
This causes the problem that playlist target songs are correctly
hidden after database update, but will remain visible after mpd server
restarted. This pr solves the problem by storing `in_playlist` value
of songs into the `SimpleDatabase` file.
By default, if the parent of a process dies, the process gets SIGHUP
and is supposed to shut down. This however doesn't work for MPD,
because MPD redefines SIGHUP with a different meaning (like most
daemons do).
To work around this, we configure the kernel to send SIGTERM instead
of SIGHUP.
Closes https://github.com/MusicPlayerDaemon/MPD/issues/1706
Fixes a busy loop in BufferingInputStream::RunThreadLocked() because
the method never learns that seeking is ignored, even though the HTTP
stream is already broken and can never be read; nobody cared to check
for errors.
Closes https://github.com/MusicPlayerDaemon/MPD/issues/1727
libfmt version 10 apparently doesn't know how to format a
`StringBuffer`, failing the MPD build. Since we have a formatter
specialization for `AudioFormat`, let's use that - it's better and
easier to use.
Closes https://github.com/MusicPlayerDaemon/MPD/issues/1807
This should prevent ffmpeg from taking priority over the gme plugin.
The ffmpeg plugin is more buggy than gme.
One of the prominent bugs of preferring ffmpeg over gme is that ffmpeg
cannot seek SAP files while gme can. This should prevent that from
happening.
Some ALSA capture devices can have very large buffers, holding 10
seconds or more audio. Using the maximum buffer size with such
devices leads to unacceptably large, and unnecessary, latency.
Also, some ALSA drivers (e.g. HDA Intel PCH) report an invalid
maximum period size, and the period size that mpd calculates from
the maximum buffer size results in "Invalid argument" error when
applying the hw_params. Note that the "default" capture device on
many cards includes the "dsnoop" plugin which imposes a buffer
size of 16384 frames, so that "alsa://" works OK but
"alsa://plughw" or "alsa://hw" both fail.
Limit the maximum buffer time for ALSA input devices to a more useable
2 seconds, thereby avoiding both the above problems.
Since using libfmt (commit: dfc5b49) ICY MetaData was broken. Eg MP3 streams encoded with LAME were affected resulting in missing stream titles on VLC or strange noises on Sonos hardware. This commit fixes the icy-metaint field.
Add an `always_off` option to outputs that causes them to never start
playback even if they're enabled.
This allows placeholder `null` outputs to be defined for the purpose
of having an external client react to the enabled state without the
side effects of real outputs. Like an external mixer, the client can
perform some action when an output is enabled.
Normally `null` outputs can be used for playback so it's possible for
MPD to continue playback silently if a problem occurs with all the real
outputs (or there are none enabled).
Max Kellermann