Many years ago, FAAD had a serious ABI bug: the NeAACDecInit()
prototype in its header declared the "samplerate" parameter to be
"unsigned long *", but internally, the function assumed it was
"uint32_t *" instead. On 32 bit machines, that was no difference, but
on 64 bit, this left one portion of the return value uninitialized;
and worse, on big-endian, the wrong word was filled. This bug had to
be worked around in MPD (commit 9c4e97a6).
A few months later, the bug was fixed in the FAAD CVS in commit 1.117
on file libfaad/decoder.c; the commit message was:
"Use public headers internally to prevent duplicate declarations"
The commit message was too brief at best; the problem was not
duplicate declarations, but a prototype mismatch. No mention of the
bug fix in the ChangeLog.
The MPD project never learned about this bug fix, and so MPD would
always pass a "uin32_t *" dressed up as a "unsigned long *". Nearly 6
years later, it's about time to fix this second ABI problem. Let's
kill the workaround!
On NetBSD, PTHREAD_MUTEX_INITIALIZER and PTHREAD_COND_INITIALIZER are
not compatible with C++11 "constexpr" (see Mantis ticket 0004110). As
a workaround, don't ues "constexpr", and use the functions
pthread_mutex_init(), pthread_mutex_destroy(), pthread_cond_init() and
pthread_cond_destroy() instead. This adds some runtime overhead, but
is portable to POSIX implementations that have awkward initializer
macros.
Casting std::numeric_limits<unsigned>::max() to "long" leads to an
overflow if sizeof(unsigned)==sizeof(long), and the result will be -1.
This happens on some 32 bit architectures, for example ARM and WIN32.
Workaround: use std::numeric_limits<int>::max(), which is the largest
signed integer. Since sizeof(long)>=sizeof(int), this will never
overflow.
Fixes Mantis ticket 0004080.
The IsActive() method returned true even if the timer was not active,
after it completed once. This broke the state file timer, and the
state file was not saved periodically.
When the data chunk size is not a multiple of the frame size, the last
partial frame lead to an endless loop. We fix this by checking
chunk_sze>=frame instead of chunk_sze>0. This way, the partial frame
is simply skipped.
Previously, MPD tried to slurp the whole song file, count the number
of frames and calculate the song duration from that. That however is
extremely expensive for remote files, and will delay playback for a
long time. Workaround: check only the first 128 frames and try to
extrapolate from here. Fixes Mantis ticket 0004035.
Implement a "bulk" edit mode that postpones both UpdateQueuedSong()
and OnModified(). This way, the playlist version gets incremented
only once. More importantly: when adding multiple songs to a queue
that consists of only one song, the first song that got added will
always be played next. By postponing this choice, all newly added
songs get a chance to become the next song. Fixes the second (and
last) part of Mantis ticket 0004005.
Don't restore the current song after shufflung when MPD is stopped
(but still remembers the current song internally). Fixes the first
part of Mantis ticket 0004005.
Turns out the lock-free code using atomics was not thread-safe. The
given callback could be invoked by GLib before the source_id attribute
was assigned. This commit changes the DeferredMonitor class to use a
Mutex to block the event loop until source_id is assigned. This bug
does not exist in the 0.19 branch because it does not use the GLib
main loop anymore.
This corrects a major mistake from commit 724a59aa - there was one
small thing that commit was supposed to do, and it failed.
AV_TIME_BASE is not a seek flag.
Max Kellermann