If you send the "quit" or "stop" command with the client API, it will
now attempt to kill network I/O immediately (same as normal input in the
previous commits).
This makes the player wait until each script is loaded. Do this to give
the script a chance to setup all its event handlers. It might also be
useful to allow a script to change options that matter for playback.
While waiting for a script to be loaded, the player actually accepts
input. This is needed because the scripts can execute player commands
anyway while they are being "loaded". The player won't react to most
commands though: it can't quit or navigate the playlist in this state.
For deciding whether a script is finally loaded, we use a cheap hack: if
mpv_wait_event() is called, it's considered loaded. Let's hope this is
good enough. I think it's better than introducing explicit API for this.
Although I'm sure this will turn out as too simplistic some time in the
future, the same would probably happen with a more explicit API.
The event was copied early, and wasn't released if it was rejected
instead of being added to the event queue. Fix by copying the event at a
point when it's certainly added to the event queue.
The dup_event_data() function is merely moved.
Add a mechanism to the client API code, which allows the player core to
query whether a client API event is needed at all. Use it for the cache
update.
In this case, this is probably a pure microoptimization; but the
mechanism will be useful for other things too.
Until recently, vo_opengl could be accessed from a single thread only,
due to the OpenGL API context being thread-specific. This issue doesn't
exist anymore, because VOs run on their own thread. This means we can
simply lock/unlock the playloop instead of doing something complicated
to get the playloop thread to execute our code.
Internally, there are two mechanisms which can trigger property
notification as used with "observed" properties in the client API.
The first mechanism associates events with a group of properties that
are potentially changed by a certain event. mp_event_property_change[]
declares these associations, and maps each event to a set of strings.
When an event happens, the set of strings is matched against the list of
observed properties of each client. Make this more efficient by
comparing bitsets of events instead. This way, only a bit-wise "and" is
needed for each observed property. Even better, we can completely skip
clients which have no observed properties that match.
The second mechanism just updates individual properties explicitly by
name. Optimize this by using the property index instead. It would be
nice if we could reuse the first mechanism for the second one, but
there are too many properties to fit into a 64 bit mask.
(Though the limit on 64 events might get us into trouble later...)
This is perfectly allowed, but was ignored, because it's a corner case.
It doesn't actually wait for other clients to be destroyed, but on the
other hand I think there's no way to have other clients before
initialization.
CC: @mpv-player/stable
While I'm not very fond of "const", it's important for declarations
(it decides whether a symbol is emitted in a read-only or read/write
section). Fix all these cases, so we have writeable global data only
when we really need.
Since redundant wakeups are avoided now, it's easy to miss a wakeup when
creating/setting the pipe/callback after the client API was signalled.
If the client API is signalled, need_wakeup is set to true, and
wakeup_client skips writing to the pipe or calling the client API. That
this can happen is not very obvious to the client API, so trigger a
wakeup right on start in order to remove this special case.
Until now, availability of new log messages (through the mechanism
associated with mpv_request_log_messages()) did not wakeup the client
API properly. Commit 3b7402b5 was basically a hack to improve that
somewhat, but it wasn't a solution.
The main problem is that the client API itself is producing messages, so
the message callback would attempt to lock the client API lock,
resulting in a deadlock. Even if the lock was recursive, we'd run into
lock-order issues.
Solve this by using a separate lock for waiting and wakeup. Also, since
it's a natural addition, avoid redundant wakeups. This means the wakeup
callback as well as the wakeup pipe will be triggered only once until
the next mpv_wait_event() call happens.
This might make the wakeup callback be invoked in a reentrant way for
the first time, for example if a mpv_* function prints to a log. Adjust
the docs accordingly. (Note that non-reentrant beheavior was never
guaranteed - basically the wakeup callback is somewhat dangerous and
inconvenient.)
Also remove some traces of unneeded code. ctx->shutdown for one was
never set, and probably a leftover of an abandoned idea.
A bit verbose, but less misleading. In most cases, the API user probably
actually wants mpv_terminate_destroy() instead, so the less-useful
function shouldn't have a simnpler name anyway.
mpv_destroy() should perhaps better be called mpv_detach(), because it
destroys only the handle, not necessarily the player. The player is only
terminated if a quit command is sent.
This function quits automatically, and additionally waits until the
player is completely destroyed. It removes the possibility that the
player core is still uninitializing, while all client handles are
already destroyed. (Although in practice, the difference is usually not
important.)
Now a negative timeout mean an infinite timeout. This is similar to the
poll() system call. Apparently this is more intuitive and less confusing
than specifying a "very high" value as timeout if you want to wait
forever.
For callers that never pass negative timeouts, nothing changes.
Listening on messages currently uses polling (every time
mpv_wait_event() has no new events, the message buffer is polled and a
message event is possibly created). Improve this situation a bit, and
call the user-supplied wakeup callback.
This will increase the frequency with which the wakeup callback is
called, but the client is already supposed to be able to deal with this
situation. Also, as before, calling mpv_wait_event() from the wakeup
callback is forbidden, so the client can't read new messages from the
callback directly.
The wakeup pipe is written either. Since the wakeup pipe is created
lazily, we can't access the pipe handle without creating a race
condition or a deadlock. (This is actually very silly, since in practice
the race condition won't matter, but for now let's keep it clean.)
If an API user calls mpv_wait_event() and mpv_observe_property() before
mpv_initialize(), it could happen that a property was accessed before
initialization, which is not ok.
This was always intended this way, and even documented in client.h. Due
to an oversight it was never actually implemented.
The intention is that mpv embedded in applications and "real mpv" don't
conflict. An API user can undo this by setting the "config" option to
"yes", if using the user's mpv config is desired.
Until now, an error was reported only if the command couldn't be parsed.
Attempt to do more fine-grained reporting. This is not necessarily
perfect, but it's an improvement.
This is probably a good idea, because it would make it easier for
software embedding mpv to configure the mpv parts, without requiring the
host program to provide explicit mechanisms for this (other than calling
mpv_load_config_file()).
The code paths for setting options by string and by direct "raw" value
were too different, which resulted in some weird code. Make the code
paths closer to each other.
Also, use this to remove the weirdness in the mpv_set_option()
implementation.
Use the time as returned by mp_time_us() for mpthread_cond_timedwait(),
instead of calculating the struct timespec value based on a timeout.
This (probably) makes it easier to wait for a specific deadline.
If a property is notified as changed, and then again (before the change
notification is returned to the client), and the second change is a
sporadic change (i.e. nothing actually changed) and the change
notification was associated with with a data type, it could happen that
a change was "overlooked", because it would detect no change on the
second notification.
This is actually a pretty annoying corner case, due to the annoying way
we do things, so just store both the previously returned _and_ the newly
obtained property value. then we always compare with the user value to
check for a change, excluding any possibility of a missed change.
Note that we don't (can't/shouldn't) care if a value changes, and then
changes back; it's fine if that doesn't generate a notification. This is
due to how property notifications are supposed to be coalesced.
This was part of osdep/threads.c out of laziness. But it doesn't contain
anything OS dependent. Note that the rest of threads.c actually isn't
all that OS dependent either (just some minor ifdeffery to work around
the lack of clock_gettime() on OSX).
This should fix some issues, such as not being able to set the
"no-video" option with MPV_FORMAT_FLAG.
Note that this changes semantics a bit. Now setting an option strictly
overwrite it, even if the corresponding command line option does not.
For example, if we change --sub to append by default, then setting the
"sub" option via the client API would still never append. (Oddly, this
also applies to --vf-add, which will overwrite the old value when using
the client API.)
I'm doing this because there's no proper separation between the command
line parser and setting an option using the MPV_FORMAT_STRING format.
Maybe the solution to this mess would be adding format aware code (i.e.
m_option_set_node) to every option type, and falling back to strings
only if needed - but this would mean that you couldn't set e.g. an
integer option using MPV_FORMAT_STRING, which doesn't seem to be ideal
either.
In conclusion, the current approach seems to be most robust, but I'm
open to suggestions should someone find that these semantics are a
problem.
And slightly adjust the semantics of MPV_EVENT_PAUSE/MPV_EVENT_UNPAUSE.
The real pause state can now be queried with the "core-idle" property,
the user pause state with the "pause" property, whether the player is
paused due to cache with "paused-for-cache", and the keep open event can
be guessed with the "eof-reached" property.
Otherwise, the client API user could not know why playback was stopped.
Regarding the fact that 0 is used both for normal EOF and EOF on error:
this is because mplayer traditionally did not distinguish these, and in
general it's hard to tell the real reason. (There are various weird
corner cases which make it hard.)
This turned out ridiculously complex. I think it will have to be
simplified some day. Main reason for the complexity are:
- filtering properties by forcing clients to observe individual
properties explicitly
(to avoid spamming clients with changes they don't want)
- optional retrieval of property value with the notification
(the basic idea was that this is more user friendly)
- allowing to the client to specify a format in which the value
should be retrieved
(because if a property changes its type, the client API couldn't
convert it properly, and compatibility would break)
I don't know yet which of these are important, and everything could
change. In particular, the interface and semantics should be adjusted
to reduce the implementation complexity.
While I consider the API complete, there could (and probably will) be
bugs left. Also while the implementation is complete, it's inefficient.
The complexity of the property matching is O(a*b*c) with a clients,
b observed properties, and c properties changing at once. I threw away
an earlier implementation using bitmasks, because it was too unwieldy.
Remove the use of mp_ring and use a simple array and a bunch of
variables instead. This is way less awkwad.
The change in reserve_reply fixes incorrect tracking of free events.
Change script_message to broadcast the message to all clients. Add a new
script_message_to command, which does what the old script_message
command did.
This is intended as simplification, although it might lead to chaos too.
Sending an asynchronous request and then calling mpv_destroy() would
crash the player when trying to send the reply to the removed client.
Fix this by waiting until all remaining replies have been sent.
Always map MPV_FORMAT_STRING to setting property value directly through
M_PROPERTY_SET_STRING, instead of trying to go through
M_PROPERTY_SET_NODE.
This treats a direct MPV_FORMAT_STRING query differently from a
MPV_FORMAT_STRING wrapped in a mpv_node. This was already the case in
mpv_get_property(). The reason for all this is that mpv_node is supposed
to be the exact type, while a direct MPV_FORMAT_STRING goes through all
possible conversions.
Not sure if these semantics are good.
1. Cannot set option after initialized: it seems that this bug has
existed since libmpv was introduced first. Maybe just a typo.
2. Crash when setting property with native format: mpv_set_property
just causes a crash when using a native format. I found an invalid
casting and fixed it.
3. Wrong error value for mpv_get_property: when an error occurred,
mpv_get_property always returns wrong format error because every
error for property except M_PROPERTY_NOT_IMPLEMENTED is just ignored.
Signed-off-by: wm4 <wm4@nowhere>
Closes pull request #593. Does not incldue the first fix, which was not
correct. The underlying bug will be fixed by a later commit.
Commit message extracted from pull request and slightly edited.
wm4
xylosper