Files
yabridge/src/common/communication/vst3.h
T
Robbert van der Helm 77d43e4f08 Increase the default serialization buffer size
Setting VST2 plugin parameters would sometimes result in allocations.
2021-05-23 16:45:01 +02:00

533 lines
23 KiB
C++

// yabridge: a Wine VST bridge
// Copyright (C) 2020-2021 Robbert van der Helm
//
// This program is free software: you can redistribute it and/or modify
// it under the terms of the GNU General Public License as published by
// the Free Software Foundation, either version 3 of the License, or
// (at your option) any later version.
//
// This program is distributed in the hope that it will be useful,
// but WITHOUT ANY WARRANTY; without even the implied warranty of
// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
// GNU General Public License for more details.
//
// You should have received a copy of the GNU General Public License
// along with this program. If not, see <https://www.gnu.org/licenses/>.
#pragma once
#include <future>
#include <variant>
#include "../logging/vst3.h"
#include "../serialization/vst3.h"
#include "common.h"
/**
* An instance of `AdHocSocketHandler` that encapsulates the simple
* communication model we use for sending requests and receiving responses. A
* request of type `T`, where `T` is in `{Control,Callback}Request`, should be
* answered with an object of type `T::Response`.
*
* See the docstrings on `Vst2EventHandler` and `AdHocSocketHandler` for more
* information on how this works internally and why it works the way it does.
*
* @note The name of this class is not to be confused with VST3's `IMessage` as
* this is very much just general purpose messaging between yabridge's two
* components. Of course, this will handle `IMessage` function calls as well.
*
* @tparam Thread The thread implementation to use. On the Linux side this
* should be `std::jthread` and on the Wine side this should be `Win32Thread`.
* @tparam Request Either `ControlRequest` or `CallbackRequest`.
*/
template <typename Thread, typename Request>
class Vst3MessageHandler : public AdHocSocketHandler<Thread> {
public:
/**
* Sets up a single main socket for this type of events. The sockets won't
* be active until `connect()` gets called.
*
* @param io_context The IO context the main socket should be bound to. A
* new IO context will be created for accepting the additional incoming
* connections.
* @param endpoint The socket endpoint used for this event handler.
* @param listen If `true`, start listening on the sockets. Incoming
* connections will be accepted when `connect()` gets called. This should
* be set to `true` on the plugin side, and `false` on the Wine host side.
*
* @see Sockets::connect
*/
Vst3MessageHandler(boost::asio::io_context& io_context,
boost::asio::local::stream_protocol::endpoint endpoint,
bool listen)
: AdHocSocketHandler<Thread>(io_context, endpoint, listen) {}
/**
* Serialize and send an event over a socket and return the appropriate
* response.
*
* As described above, if this function is currently being called from
* another thread, then this will create a new socket connection and send
* the event there instead.
*
* @param object The request object to send. Often a marker struct to ask
* for a specific object to be returned.
* @param logging A pair containing a logger instance and whether or not
* this is for sending host -> plugin control messages. If set to false,
* then this indicates that this `Vst3MessageHandler` is handling plugin
* -> host callbacks isntead. Optional since it only has to be set on the
* plugin's side.
* @param buffer The serialization and receiving buffer to reuse. This is
* optional, but it's useful for minimizing allocations in the audio
* processing loop.
*
* @relates Vst3MessageHandler::receive_messages
*/
template <typename T>
typename T::Response send_message(
const T& object,
std::optional<std::pair<Vst3Logger&, bool>> logging,
SerializationBufferBase& buffer) {
typename T::Response response_object;
receive_into(object, response_object, logging, buffer);
return response_object;
}
/**
* The same as the above, but with a small default buffer.
*
* @overload
*/
template <typename T>
typename T::Response send_message(
const T& object,
std::optional<std::pair<Vst3Logger&, bool>> logging) {
typename T::Response response_object;
receive_into(object, response_object, logging);
return response_object;
}
/**
* `Vst3MessageHandler::send_message()`, but deserializing the response into
* an existing object.
*
* @param response_object The object to deserialize into.
*
* @overload Vst3MessageHandler::send_message
*/
template <typename T>
typename T::Response& receive_into(
const T& object,
typename T::Response& response_object,
std::optional<std::pair<Vst3Logger&, bool>> logging,
SerializationBufferBase& buffer) {
using TResponse = typename T::Response;
// Since a lot of messages just return a `tresult`, we can't filter out
// responses based on the response message type. Instead, we'll just
// only print the responses when the request was not filtered out.
bool should_log_response = false;
if (logging) {
auto [logger, is_host_vst] = *logging;
should_log_response = logger.log_request(is_host_vst, object);
}
// A socket only handles a single request at a time as to prevent
// messages from arriving out of order. `AdHocSocketHandler::send()`
// will either use a long-living primary socket, or if that's currently
// in use it will spawn a new socket for us.
this->send([&](boost::asio::local::stream_protocol::socket& socket) {
write_object(socket, Request(object), buffer);
read_object<TResponse>(socket, response_object, buffer);
});
if (should_log_response) {
auto [logger, is_host_vst] = *logging;
logger.log_response(!is_host_vst, response_object);
}
return response_object;
}
/**
* The same function as above, but with a small default buffer.
*
* @overload
*/
template <typename T>
typename T::Response& receive_into(
const T& object,
typename T::Response& response_object,
std::optional<std::pair<Vst3Logger&, bool>> logging) {
SerializationBuffer<256> buffer{};
return receive_into(object, response_object, std::move(logging),
buffer);
}
/**
* Spawn a new thread to listen for extra connections to `endpoint`, and
* then start a blocking loop that handles messages from the primary
* `socket`.
*
* The specified function receives a `Request` variant object containing an
* object of type `T`, and it should then return the corresponding
* `T::Response`.
*
* @param logging A pair containing a logger instance and whether or not
* this is for sending host -> plugin control messages. If set to false,
* then this indicates that this `Vst3MessageHandler` is handling plugin
* -> host callbacks isntead. Optional since it only has to be set on the
* plugin's side.
* @param callback The function used to generate a response out of the
* request. See the definition of `F` for more information.
*
* @tparam F A function type in the form of `T::Response(T)` for every `T`
* in `Request`. This way we can directly deserialize into a `T::Response`
* on the side that called `receive_into(T, T::Response&)`.
* @tparam persistent_buffers If enabled, we'll reuse the buffers used for
* sending and receiving serialized data as well as the objects we're
* receiving into. This avoids allocations in the audio processing loop
* (after the first allocation of course). This is mostly relevant for the
* `YaProcessData` object stored inside of `YaAudioProcessor::Process`.
* These buffers are thread local and will also never shrink, but that
* should not be an issue with the `IAudioProcessor` and `IComponent`
* functions. Saving and loading state is handled on the main sockets,
* which don't use these persistent buffers.
*
* @relates Vst3MessageHandler::send_event
*/
template <bool persistent_buffers = false, typename F>
void receive_messages(std::optional<std::pair<Vst3Logger&, bool>> logging,
F&& callback) {
// Reading, processing, and writing back the response for the requests
// we receive works in the same way regardless of which socket we're
// using
const auto process_message =
[&](boost::asio::local::stream_protocol::socket& socket) {
// The persistent buffer is only used when the
// `persistent_buffers` template value is enabled, but we'll
// always use the thread local persistent object. Because of
// loading and storing state the buffer can grow a lot in size
// which is why we might not want to reuse that for tasks that
// don't need to be realtime safe, but the object has a fixed
// size. Normally reusing this object doesn't make much sense
// since it's a variant and it will likely have to be recreated
// every time, but on the audio processor side we store the
// actual variant within an object and we then use some hackery
// to always keep the large process data object in memory.
thread_local SerializationBuffer<256> persistent_buffer{};
thread_local Request persistent_object;
auto& request =
persistent_buffers
? read_object<Request>(socket, persistent_object,
persistent_buffer)
: read_object<Request>(socket, persistent_object);
// See the comment in `receive_into()` for more information
bool should_log_response = false;
if (logging) {
should_log_response = std::visit(
[&](const auto& object) {
auto [logger, is_host_vst] = *logging;
return logger.log_request(is_host_vst, object);
},
// In the case of `AudioProcessorRequest`, we need to
// actually fetch the variant field since our object
// also contains a persistent object to store process
// data into so we can prevent allocations during audio
// processing
get_request_variant(request));
}
// We do the visiting here using a templated lambda. This way we
// always know for sure that the function returns the correct
// type, and we can scrap a lot of boilerplate elsewhere.
std::visit(
[&]<typename T>(T object) {
typename T::Response response = callback(object);
if (should_log_response) {
auto [logger, is_host_vst] = *logging;
logger.log_response(!is_host_vst, response);
}
if constexpr (persistent_buffers) {
write_object(socket, response, persistent_buffer);
} else {
write_object(socket, response);
}
},
// See above
get_request_variant(request));
};
this->receive_multi(logging
? std::optional(std::ref(logging->first.logger))
: std::nullopt,
process_message);
}
};
/**
* Manages all the sockets used for communicating between the plugin and the
* Wine host when hosting a VST3 plugin.
*
* On the plugin side this class should be initialized with `listen` set to
* `true` before launching the Wine VST host. This will start listening on the
* sockets, and the call to `connect()` will then accept any incoming
* connections.
*
* We'll have a host -> plugin connection for sending control messages (which is
* just a made up term to more easily differentiate between the two directions),
* and a plugin -> host connection to allow the plugin to make callbacks. Both
* of these connections are capable of spawning additional sockets and threads
* as needed.
*
* For audio processing (or anything that implement `IAudioProcessor` or
* `IComonent`) we'll use dedicated sockets per instance, since we don't want to
* do anything that could increase latency there.
*
* @tparam Thread The thread implementation to use. On the Linux side this
* should be `std::jthread` and on the Wine side this should be `Win32Thread`.
*/
template <typename Thread>
class Vst3Sockets : public Sockets {
public:
/**
* Sets up the sockets using the specified base directory. The sockets won't
* be active until `connect()` gets called.
*
* @param io_context The IO context the sockets should be bound to. Relevant
* when doing asynchronous operations.
* @param endpoint_base_dir The base directory that will be used for the
* Unix domain sockets.
* @param listen If `true`, start listening on the sockets. Incoming
* connections will be accepted when `connect()` gets called. This should
* be set to `true` on the plugin side, and `false` on the Wine host side.
*
* @see Vst3Sockets::connect
*/
Vst3Sockets(boost::asio::io_context& io_context,
const boost::filesystem::path& endpoint_base_dir,
bool listen)
: Sockets(endpoint_base_dir),
host_vst_control(io_context,
(base_dir / "host_vst_control.sock").string(),
listen),
vst_host_callback(io_context,
(base_dir / "vst_host_callback.sock").string(),
listen),
io_context(io_context) {}
// NOLINTNEXTLINE(clang-analyzer-optin.cplusplus.VirtualCall)
~Vst3Sockets() noexcept override { close(); }
void connect() override {
host_vst_control.connect();
vst_host_callback.connect();
}
void close() override {
// Manually close all sockets so we break out of any blocking operations
// that may still be active
host_vst_control.close();
vst_host_callback.close();
// This map should be empty at this point, but who knows
std::lock_guard lock(audio_processor_sockets_mutex);
for (auto& [instance_id, socket] : audio_processor_sockets) {
socket.close();
}
}
/**
* Connect to the dedicated `IAudioProcessor` and `IConnect` handling socket
* for a plugin object instance. This should be called on the plugin side
* after instantiating such an object.
*
* @param instance_id The object instance identifier of the socket.
*/
void add_audio_processor_and_connect(size_t instance_id) {
std::lock_guard lock(audio_processor_sockets_mutex);
audio_processor_sockets.try_emplace(
instance_id, io_context,
(base_dir / ("host_vst_audio_processor_" +
std::to_string(instance_id) + ".sock"))
.string(),
false);
audio_processor_sockets.at(instance_id).connect();
}
/**
* Create and listen on a dedicated `IAudioProcessor` and `IConnect`
* handling socket for a plugin object instance. The calling thread will
* block until the socket has been closed. This should be called from the
* Wine plugin host side after instantiating such an object.
*
* @param instance_id The object instance identifier of the socket.
* @param socket_listening_latch A promise we'll set a value for once the
* socket is being listened on so we can wait for it. Otherwise it can be
* that the native plugin already tries to connect to the socket before
* Wine plugin host is even listening on it.
* @param cb An overloaded function that can take every type `T` in the
* `AudioProcessorRequest` variant and then returns `T::Response`.
*
* @tparam F A function type in the form of `T::Response(T)` for every `T`
* in `AudioProcessorRequest::Payload`.
*/
template <typename F>
void add_audio_processor_and_listen(
size_t instance_id,
std::promise<void>& socket_listening_latch,
F&& callback) {
{
std::lock_guard lock(audio_processor_sockets_mutex);
audio_processor_sockets.try_emplace(
instance_id, io_context,
(base_dir / ("host_vst_audio_processor_" +
std::to_string(instance_id) + ".sock"))
.string(),
true);
}
socket_listening_latch.set_value();
audio_processor_sockets.at(instance_id).connect();
// This `true` indicates that we want to reuse our serialization and
// receiving buffers for all calls. This slightly reduces the amount of
// allocations in the audio processing loop.
audio_processor_sockets.at(instance_id)
.template receive_messages<true>(std::nullopt,
std::forward<F>(callback));
}
/**
* If `instance_id` is in `audio_processor_sockets`, then close its socket
* and remove it from the map. This is called from the destructor of
* `Vst3PluginProxyImpl` on the plugin side and when handling
* `Vst3PluginProxy::Destruct` on the Wine plugin host side.
*
* @param instance_id The object instance identifier of the socket.
*
* @return Whether the socket was closed and removed. Returns false if it
* wasn't in the map.
*/
bool remove_audio_processor(size_t instance_id) {
std::lock_guard lock(audio_processor_sockets_mutex);
if (audio_processor_sockets.contains(instance_id)) {
audio_processor_sockets.at(instance_id).close();
audio_processor_sockets.erase(instance_id);
return true;
} else {
return false;
}
}
/**
* Send a message from the native plugin to the Wine plugin host to handle
* an `IAudioProcessor` or `IComponent` call. Since those functions are
* called from a hot loop we want every instance to have a dedicated socket
* and thread for handling those. These calls also always reuse buffers to
* minimize allocations.
*
* @tparam T Some object in the `AudioProcessorRequest` variant.
*/
template <typename T>
typename T::Response send_audio_processor_message(
const T& object,
std::optional<std::pair<Vst3Logger&, bool>> logging) {
typename T::Response response_object;
return receive_audio_processor_message_into(
object, response_object, object.instance_id, logging);
}
/**
* Overload for use with `MessageReference<T>`, since we cannot
* directly get the instance ID there.
*/
template <typename T>
typename T::Response send_audio_processor_message(
const MessageReference<T>& object_ref,
std::optional<std::pair<Vst3Logger&, bool>> logging) {
typename T::Response response_object;
return receive_audio_processor_message_into(
object_ref, response_object, object_ref.get().instance_id, logging);
}
/**
* Alternative to `send_audio_processor_message()` for use with
* `MessageReference<T>`, where we also want deserialize into an existing
* object to prevent allocations. Used during audio processing.q
*
* TODO: Think of a better name for this
*/
template <typename T>
typename T::Response& receive_audio_processor_message_into(
const MessageReference<T>& request_ref,
typename T::Response& response_ref,
std::optional<std::pair<Vst3Logger&, bool>> logging) {
return receive_audio_processor_message_into(
request_ref, response_ref, request_ref.get().instance_id, logging);
}
/**
* For sending messages from the host to the plugin. After we have a better
* idea of what our communication model looks like we'll probably want to
* provide an abstraction similar to `Vst2EventHandler`. For optimization
* reasons calls to `IAudioProcessor` or `IComponent` are handled using the
* dedicated sockets in `audio_processor_sockets`.
*
* This will be listened on by the Wine plugin host when it calls
* `receive_multi()`.
*/
Vst3MessageHandler<Thread, ControlRequest> host_vst_control;
/**
* For sending callbacks from the plugin back to the host. After we have a
* better idea of what our communication model looks like we'll probably
* want to provide an abstraction similar to `Vst2EventHandler`.
*/
Vst3MessageHandler<Thread, CallbackRequest> vst_host_callback;
private:
/**
* The actual implementation for `send_audio_processor_message` and
* `receive_audio_processor_message_into`. Here we keep a thread local
* static variable for our buffers sending.
*/
template <typename T>
typename T::Response& receive_audio_processor_message_into(
const T& object,
typename T::Response& response_object,
size_t instance_id,
std::optional<std::pair<Vst3Logger&, bool>> logging) {
thread_local SerializationBuffer<256> audio_processor_buffer{};
return audio_processor_sockets.at(instance_id)
.receive_into(object, response_object, logging,
audio_processor_buffer);
}
boost::asio::io_context& io_context;
/**
* Every `IAudioProcessor` or `IComponent` instance (which likely implements
* both of those) will get a dedicated socket. These functions are always
* called in a hot loop, so there should not be any waiting or additional
* thread or socket creation happening there.
*
* THe last `false` template arguments means that we'll disable all ad-hoc
* socket and thread spawning behaviour. Otherwise every plugin instance
* would have one dedicated thread for handling function calls to these
* interfaces, and then another dedicated thread just idling around.
*/
std::map<size_t, Vst3MessageHandler<Thread, AudioProcessorRequest>>
audio_processor_sockets;
std::mutex audio_processor_sockets_mutex;
};