App-MHFS
view release on metacpan or search on metacpan
share/public_html/static/music_worklet_inprogress/decoder/deps/miniaudio/miniaudio.h view on Meta::CPAN
}
// Loop over each device info and do something with it. Here we just print the name with their index. You may want
// to give the user the opportunity to choose which device they'd prefer.
for (ma_uint32 iDevice = 0; iDevice < playbackCount; iDevice += 1) {
printf("%d - %s\n", iDevice, pPlaybackInfos[iDevice].name);
}
ma_device_config config = ma_device_config_init(ma_device_type_playback);
config.playback.pDeviceID = &pPlaybackInfos[chosenPlaybackDeviceIndex].id;
config.playback.format = MY_FORMAT;
config.playback.channels = MY_CHANNEL_COUNT;
config.sampleRate = MY_SAMPLE_RATE;
config.dataCallback = data_callback;
config.pUserData = pMyCustomData;
ma_device device;
if (ma_device_init(&context, &config, &device) != MA_SUCCESS) {
// Error
}
...
ma_device_uninit(&device);
ma_context_uninit(&context);
```
The first thing we do in this example is initialize a `ma_context` object with `ma_context_init()`.
The first parameter is a pointer to a list of `ma_backend` values which are used to override the
default backend priorities. When this is NULL, as in this example, miniaudio's default priorities
are used. The second parameter is the number of backends listed in the array pointed to by the
first parameter. The third parameter is a pointer to a `ma_context_config` object which can be
NULL, in which case defaults are used. The context configuration is used for setting the logging
callback, custom memory allocation callbacks, user-defined data and some backend-specific
configurations.
Once the context has been initialized you can enumerate devices. In the example above we use the
simpler `ma_context_get_devices()`, however you can also use a callback for handling devices by
using `ma_context_enumerate_devices()`. When using `ma_context_get_devices()` you provide a pointer
to a pointer that will, upon output, be set to a pointer to a buffer containing a list of
`ma_device_info` structures. You also provide a pointer to an unsigned integer that will receive
the number of items in the returned buffer. Do not free the returned buffers as their memory is
managed internally by miniaudio.
The `ma_device_info` structure contains an `id` member which is the ID you pass to the device
config. It also contains the name of the device which is useful for presenting a list of devices
to the user via the UI.
When creating your own context you will want to pass it to `ma_device_init()` when initializing the
device. Passing in NULL, like we do in the first example, will result in miniaudio creating the
context for you, which you don't want to do since you've already created a context. Note that
internally the context is only tracked by it's pointer which means you must not change the location
of the `ma_context` object. If this is an issue, consider using `malloc()` to allocate memory for
the context.
1.2. High Level API
-------------------
The high level API consists of three main parts:
* Resource management for loading and streaming sounds.
* A node graph for advanced mixing and effect processing.
* A high level "engine" that wraps around the resource manager and node graph.
The resource manager (`ma_resource_manager`) is used for loading sounds. It supports loading sounds
fully into memory and also streaming. It will also deal with reference counting for you which
avoids the same sound being loaded multiple times.
The node graph is used for mixing and effect processing. The idea is that you connect a number of
nodes into the graph by connecting each node's outputs to another node's inputs. Each node can
implement it's own effect. By chaining nodes together, advanced mixing and effect processing can
be achieved.
The engine encapsulates both the resource manager and the node graph to create a simple, easy to
use high level API. The resource manager and node graph APIs are covered in more later sections of
this manual.
The code below shows how you can initialize an engine using it's default configuration.
```c
ma_result result;
ma_engine engine;
result = ma_engine_init(NULL, &engine);
if (result != MA_SUCCESS) {
return result; // Failed to initialize the engine.
}
```
This creates an engine instance which will initialize a device internally which you can access with
`ma_engine_get_device()`. It will also initialize a resource manager for you which can be accessed
with `ma_engine_get_resource_manager()`. The engine itself is a node graph (`ma_node_graph`) which
means you can pass a pointer to the engine object into any of the `ma_node_graph` APIs (with a
cast). Alternatively, you can use `ma_engine_get_node_graph()` instead of a cast.
Note that all objects in miniaudio, including the `ma_engine` object in the example above, are
transparent structures. There are no handles to opaque structures in miniaudio which means you need
to be mindful of how you declare them. In the example above we are declaring it on the stack, but
this will result in the struct being invalidated once the function encapsulating it returns. If
allocating the engine on the heap is more appropriate, you can easily do so with a standard call
to `malloc()` or whatever heap allocation routine you like:
```c
ma_engine* pEngine = malloc(sizeof(*pEngine));
```
The `ma_engine` API uses the same config/init pattern used all throughout miniaudio. To configure
an engine, you can fill out a `ma_engine_config` object and pass it into the first parameter of
`ma_engine_init()`:
```c
ma_result result;
ma_engine engine;
ma_engine_config engineConfig;
engineConfig = ma_engine_config_init();
engineConfig.pResourceManager = &myCustomResourceManager; // <-- Initialized as some earlier stage.
result = ma_engine_init(&engineConfig, &engine);
if (result != MA_SUCCESS) {
return result;
}
```
This creates an engine instance using a custom config. In this particular example it's showing how
you can specify a custom resource manager rather than having the engine initialize one internally.
share/public_html/static/music_worklet_inprogress/decoder/deps/miniaudio/miniaudio.h view on Meta::CPAN
valid, in which case the sound acts as a node in the middle of the node graph. This means you can
connect other sounds to this sound and allow it to act like a sound group. Indeed, this is exactly
what a `ma_sound_group` is.
When loading a sound, you specify a set of flags that control how the sound is loaded and what
features are enabled for that sound. When no flags are set, the sound will be fully loaded into
memory in exactly the same format as how it's stored on the file system. The resource manager will
allocate a block of memory and then load the file directly into it. When reading audio data, it
will be decoded dynamically on the fly. In order to save processing time on the audio thread, it
might be beneficial to pre-decode the sound. You can do this with the `MA_SOUND_FLAG_DECODE` flag:
```c
ma_sound_init_from_file(&engine, "my_sound.wav", MA_SOUND_FLAG_DECODE, pGroup, NULL, &sound);
```
By default, sounds will be loaded synchronously, meaning `ma_sound_init_*()` will not return until
the sound has been fully loaded. If this is prohibitive you can instead load sounds asynchronously
by specificying the `MA_SOUND_FLAG_ASYNC` flag:
```c
ma_sound_init_from_file(&engine, "my_sound.wav", MA_SOUND_FLAG_DECODE | MA_SOUND_FLAG_ASYNC, pGroup, NULL, &sound);
```
This will result in `ma_sound_init_*()` returning quickly, but the sound won't yet have been fully
loaded. When you start the sound, it won't output anything until some sound is available. The sound
will start outputting audio before the sound has been fully decoded when the `MA_SOUND_FLAG_DECODE`
is specified.
If you need to wait for an asynchronously loaded sound to be fully loaded, you can use a fence. A
fence in miniaudio is a simple synchronization mechanism which simply blocks until it's internal
counter hit's zero. You can specify a fence like so:
```c
ma_result result;
ma_fence fence;
ma_sound sounds[4];
result = ma_fence_init(&fence);
if (result != MA_SUCCES) {
return result;
}
// Load some sounds asynchronously.
for (int iSound = 0; iSound < 4; iSound += 1) {
ma_sound_init_from_file(&engine, mySoundFilesPaths[iSound], MA_SOUND_FLAG_DECODE | MA_SOUND_FLAG_ASYNC, pGroup, &fence, &sounds[iSound]);
}
// ... do some other stuff here in the mean time ...
// Wait for all sounds to finish loading.
ma_fence_wait(&fence);
```
If loading the entire sound into memory is prohibitive, you can also configure the engine to stream
the audio data:
```c
ma_sound_init_from_file(&engine, "my_sound.wav", MA_SOUND_FLAG_STREAM, pGroup, NULL, &sound);
```
When streaming sounds, 2 seconds worth of audio data is stored in memory. Although it should work
fine, it's inefficient to use streaming for short sounds. Streaming is useful for things like music
tracks in games.
When you initialize a sound, if you specify a sound group the sound will be attached to that group
automatically. If you set it to NULL, it will be automatically attached to the engine's endpoint.
If you would instead rather leave the sound unattached by default, you can can specify the
`MA_SOUND_FLAG_NO_DEFAULT_ATTACHMENT` flag. This is useful if you want to set up a complex node
graph.
Sounds are not started by default. To start a sound, use `ma_sound_start()`. Stop a sound with
`ma_sound_stop()`.
Sounds can have their volume controlled with `ma_sound_set_volume()` in the same way as the
engine's master volume.
Sounds support stereo panning and pitching. Set the pan with `ma_sound_set_pan()`. Setting the pan
to 0 will result in an unpanned sound. Setting it to -1 will shift everything to the left, whereas
+1 will shift it to the right. The pitch can be controlled with `ma_sound_set_pitch()`. A larger
value will result in a higher pitch. The pitch must be greater than 0.
The engine supports 3D spatialization of sounds. By default sounds will have spatialization
enabled, but if a sound does not need to be spatialized it's best to disable it. There are two ways
to disable spatialization of a sound:
```c
// Disable spatialization at initialization time via a flag:
ma_sound_init_from_file(&engine, "my_sound.wav", MA_SOUND_FLAG_NO_SPATIALIZATION, NULL, NULL, &sound);
// Dynamically disable or enable spatialization post-initialization:
ma_sound_set_spatialization_enabled(&sound, isSpatializationEnabled);
```
By default sounds will be spatialized based on the closest listener. If a sound should always be
spatialized relative to a specific listener it can be pinned to one:
```c
ma_sound_set_pinned_listener_index(&sound, listenerIndex);
```
Like listeners, sounds have a position. By default, the position of a sound is in absolute space,
but it can be changed to be relative to a listener:
```c
ma_sound_set_positioning(&sound, ma_positioning_relative);
```
Note that relative positioning of a sound only makes sense if there is either only one listener, or
the sound is pinned to a specific listener. To set the position of a sound:
```c
ma_sound_set_position(&sound, posX, posY, posZ);
```
The direction works the same way as a listener and represents the sound's forward direction:
```c
ma_sound_set_direction(&sound, forwardX, forwardY, forwardZ);
```
Sound's also have a cone for controlling directional attenuation. This works exactly the same as
listeners:
share/public_html/static/music_worklet_inprogress/decoder/deps/miniaudio/miniaudio.h view on Meta::CPAN
// Fade in over 1 second.
ma_sound_set_fade_in_milliseconds(&sound, 0, 1, 1000);
// ... sometime later ...
// Fade out over 1 second, starting from the current volume.
ma_sound_set_fade_in_milliseconds(&sound, -1, 0, 1000);
```
By default sounds will start immediately, but sometimes for timing and synchronization purposes it
can be useful to schedule a sound to start or stop:
```c
// Start the sound in 1 second from now.
ma_sound_set_start_time_in_pcm_frames(&sound, ma_engine_get_time(&engine) + (ma_engine_get_sample_rate(&engine) * 1));
// Stop the sound in 2 seconds from now.
ma_sound_set_stop_time_in_pcm_frames(&sound, ma_engine_get_time(&engine) + (ma_engine_get_sample_rate(&engine) * 2));
```
Note that scheduling a start time still requires an explicit call to `ma_sound_start()` before
anything will play.
The time is specified in global time which is controlled by the engine. You can get the engine's
current time with `ma_engine_get_time()`. The engine's global time is incremented automatically as
audio data is read, but it can be reset with `ma_engine_set_time()` in case it needs to be
resynchronized for some reason.
To determine whether or not a sound is currently playing, use `ma_sound_is_playing()`. This will
take the scheduled start and stop times into account.
Whether or not a sound should loop can be controlled with `ma_sound_set_looping()`. Sounds will not
be looping by default. Use `ma_sound_is_looping()` to determine whether or not a sound is looping.
Use `ma_sound_at_end()` to determine whether or not a sound is currently at the end. For a looping
sound this should never return true.
Internally a sound wraps around a data source. Some APIs exist to control the underlying data
source, mainly for convenience:
```c
ma_sound_seek_to_pcm_frame(&sound, frameIndex);
ma_sound_get_data_format(&sound, &format, &channels, &sampleRate, pChannelMap, channelMapCapacity);
ma_sound_get_cursor_in_pcm_frames(&sound, &cursor);
ma_sound_get_length_in_pcm_frames(&sound, &length);
```
Sound groups have the same API as sounds, only they are called `ma_sound_group`, and since they do
not have any notion of a data source, anything relating to a data source is unavailable.
Internally, sound data is loaded via the `ma_decoder` API which means by default in only supports
file formats that have built-in support in miniaudio. You can extend this to support any kind of
file format through the use of custom decoders. To do this you'll need to use a self-managed
resource manager and configure it appropriately. See the "Resource Management" section below for
details on how to set this up.
6. Resource Management
======================
Many programs will want to manage sound resources for things such as reference counting and
streaming. This is supported by miniaudio via the `ma_resource_manager` API.
The resource manager is mainly responsible for the following:
* Loading of sound files into memory with reference counting.
* Streaming of sound data
When loading a sound file, the resource manager will give you back a `ma_data_source` compatible
object called `ma_resource_manager_data_source`. This object can be passed into any
`ma_data_source` API which is how you can read and seek audio data. When loading a sound file, you
specify whether or not you want the sound to be fully loaded into memory (and optionally
pre-decoded) or streamed. When loading into memory, you can also specify whether or not you want
the data to be loaded asynchronously.
The example below is how you can initialize a resource manager using it's default configuration:
```c
ma_resource_manager_config config;
ma_resource_manager resourceManager;
config = ma_resource_manager_config_init();
result = ma_resource_manager_init(&config, &resourceManager);
if (result != MA_SUCCESS) {
ma_device_uninit(&device);
printf("Failed to initialize the resource manager.");
return -1;
}
```
You can configure the format, channels and sample rate of the decoded audio data. By default it
will use the file's native data format, but you can configure it to use a consistent format. This
is useful for offloading the cost of data conversion to load time rather than dynamically
converting at mixing time. To do this, you configure the decoded format, channels and sample rate
like the code below:
```c
config = ma_resource_manager_config_init();
config.decodedFormat = device.playback.format;
config.decodedChannels = device.playback.channels;
config.decodedSampleRate = device.sampleRate;
```
In the code above, the resource manager will be configured so that any decoded audio data will be
pre-converted at load time to the device's native data format. If instead you used defaults and
the data format of the file did not match the device's data format, you would need to convert the
data at mixing time which may be prohibitive in high-performance and large scale scenarios like
games.
Internally the resource manager uses the `ma_decoder` API to load sounds. This means by default it
only supports decoders that are built into miniaudio. It's possible to support additional encoding
formats through the use of custom decoders. To do so, pass in your `ma_decoding_backend_vtable`
vtables into the resource manager config:
```c
ma_decoding_backend_vtable* pCustomBackendVTables[] =
{
&g_ma_decoding_backend_vtable_libvorbis,
&g_ma_decoding_backend_vtable_libopus
};
...
share/public_html/static/music_worklet_inprogress/decoder/deps/miniaudio/miniaudio.h view on Meta::CPAN
config = ma_resource_manager_config_init();
config.pVFS = &vfs;
```
This is particularly useful in programs like games where you want to read straight from an archive
rather than the normal file system. If you do not specify a custom VFS, the resource manager will
use the operating system's normal file operations. This is default.
To load a sound file and create a data source, call `ma_resource_manager_data_source_init()`. When
loading a sound you need to specify the file path and options for how the sounds should be loaded.
By default a sound will be loaded synchronously. The returned data source is owned by the caller
which means the caller is responsible for the allocation and freeing of the data source. Below is
an example for initializing a data source:
```c
ma_resource_manager_data_source dataSource;
ma_result result = ma_resource_manager_data_source_init(pResourceManager, pFilePath, flags, &dataSource);
if (result != MA_SUCCESS) {
// Error.
}
// ...
// A ma_resource_manager_data_source object is compatible with the `ma_data_source` API. To read data, just call
// the `ma_data_source_read_pcm_frames()` like you would with any normal data source.
result = ma_data_source_read_pcm_frames(&dataSource, pDecodedData, frameCount, &framesRead);
if (result != MA_SUCCESS) {
// Failed to read PCM frames.
}
// ...
ma_resource_manager_data_source_uninit(pResourceManager, &dataSource);
```
The `flags` parameter specifies how you want to perform loading of the sound file. It can be a
combination of the following flags:
```
MA_RESOURCE_MANAGER_DATA_SOURCE_FLAG_STREAM
MA_RESOURCE_MANAGER_DATA_SOURCE_FLAG_DECODE
MA_RESOURCE_MANAGER_DATA_SOURCE_FLAG_ASYNC
MA_RESOURCE_MANAGER_DATA_SOURCE_FLAG_WAIT_INIT
```
When no flags are specified (set to 0), the sound will be fully loaded into memory, but not
decoded, meaning the raw file data will be stored in memory, and then dynamically decoded when
`ma_data_source_read_pcm_frames()` is called. To instead decode the audio data before storing it in
memory, use the `MA_RESOURCE_MANAGER_DATA_SOURCE_FLAG_DECODE` flag. By default, the sound file will
be loaded synchronously, meaning `ma_resource_manager_data_source_init()` will only return after
the entire file has been loaded. This is good for simplicity, but can be prohibitively slow. You
can instead load the sound asynchronously using the `MA_RESOURCE_MANAGER_DATA_SOURCE_ASYNC` flag.
This will result in `ma_resource_manager_data_source_init()` returning quickly, but no data will be
returned by `ma_data_source_read_pcm_frames()` until some data is available. When no data is
available because the asynchronous decoding hasn't caught up, `MA_BUSY` will be returned by
`ma_data_source_read_pcm_frames()`.
For large sounds, it's often prohibitive to store the entire file in memory. To mitigate this, you
can instead stream audio data which you can do by specifying the
`MA_RESOURCE_MANAGER_DATA_SOURCE_FLAG_STREAM` flag. When streaming, data will be decoded in 1
second pages. When a new page needs to be decoded, a job will be posted to the job queue and then
subsequently processed in a job thread.
For in-memory sounds, reference counting is used to ensure the data is loaded only once. This means
multiple calls to `ma_resource_manager_data_source_init()` with the same file path will result in
the file data only being loaded once. Each call to `ma_resource_manager_data_source_init()` must be
matched up with a call to `ma_resource_manager_data_source_uninit()`. Sometimes it can be useful
for a program to register self-managed raw audio data and associate it with a file path. Use the
`ma_resource_manager_register_*()` and `ma_resource_manager_unregister_*()` APIs to do this.
`ma_resource_manager_register_decoded_data()` is used to associate a pointer to raw, self-managed
decoded audio data in the specified data format with the specified name. Likewise,
`ma_resource_manager_register_encoded_data()` is used to associate a pointer to raw self-managed
encoded audio data (the raw file data) with the specified name. Note that these names need not be
actual file paths. When `ma_resource_manager_data_source_init()` is called (without the
`MA_RESOURCE_MANAGER_DATA_SOURCE_FLAG_STREAM` flag), the resource manager will look for these
explicitly registered data buffers and, if found, will use it as the backing data for the data
source. Note that the resource manager does *not* make a copy of this data so it is up to the
caller to ensure the pointer stays valid for it's lifetime. Use
`ma_resource_manager_unregister_data()` to unregister the self-managed data. You can also use
`ma_resource_manager_register_file()` and `ma_resource_manager_unregister_file()` to register and
unregister a file. It does not make sense to use the `MA_RESOURCE_MANAGER_DATA_SOURCE_FLAG_STREAM`
flag with a self-managed data pointer.
6.1. Asynchronous Loading and Synchronization
---------------------------------------------
When loading asynchronously, it can be useful to poll whether or not loading has finished. Use
`ma_resource_manager_data_source_result()` to determine this. For in-memory sounds, this will
return `MA_SUCCESS` when the file has been *entirely* decoded. If the sound is still being decoded,
`MA_BUSY` will be returned. Otherwise, some other error code will be returned if the sound failed
to load. For streaming data sources, `MA_SUCCESS` will be returned when the first page has been
decoded and the sound is ready to be played. If the first page is still being decoded, `MA_BUSY`
will be returned. Otherwise, some other error code will be returned if the sound failed to load.
In addition to polling, you can also use a simple synchronization object called a "fence" to wait
for asynchronously loaded sounds to finish. This is called `ma_fence`. The advantage to using a
fence is that it can be used to wait for a group of sounds to finish loading rather than waiting
for sounds on an individual basis. There are two stages to loading a sound:
* Initialization of the internal decoder; and
* Completion of decoding of the file (the file is fully decoded)
You can specify separate fences for each of the different stages. Waiting for the initialization
of the internal decoder is important for when you need to know the sample format, channels and
sample rate of the file.
The example below shows how you could use a fence when loading a number of sounds:
```c
// This fence will be released when all sounds are finished loading entirely.
ma_fence fence;
ma_fence_init(&fence);
// This will be passed into the initialization routine for each sound.
ma_resource_manager_pipeline_notifications notifications = ma_resource_manager_pipeline_notifications_init();
notifications.done.pFence = &fence;
// Now load a bunch of sounds:
for (iSound = 0; iSound < soundCount; iSound += 1) {
ma_resource_manager_data_source_init(pResourceManager, pSoundFilePaths[iSound], flags, ¬ifications, &pSoundSources[iSound]);
}
// ... DO SOMETHING ELSE WHILE SOUNDS ARE LOADING ...
// Wait for loading of sounds to finish.
ma_fence_wait(&fence);
```
In the example above we used a fence for waiting until the entire file has been fully decoded. If
you only need to wait for the initialization of the internal decoder to complete, you can use the
`init` member of the `ma_resource_manager_pipeline_notifications` object:
```c
notifications.init.pFence = &fence;
```
If a fence is not appropriate for your situation, you can instead use a callback that is fired on
an individual sound basis. This is done in a very similar way to fences:
```c
typedef struct
{
ma_async_notification_callbacks cb;
void* pMyData;
} my_notification;
void my_notification_callback(ma_async_notification* pNotification)
{
my_notification* pMyNotification = (my_notification*)pNotification;
// Do something in response to the sound finishing loading.
}
...
my_notification myCallback;
myCallback.cb.onSignal = my_notification_callback;
myCallback.pMyData = pMyData;
ma_resource_manager_pipeline_notifications notifications = ma_resource_manager_pipeline_notifications_init();
notifications.done.pNotification = &myCallback;
ma_resource_manager_data_source_init(pResourceManager, "my_sound.wav", flags, ¬ifications, &mySound);
```
In the example above we just extend the `ma_async_notification_callbacks` object and pass an
instantiation into the `ma_resource_manager_pipeline_notifications` in the same way as we did with
the fence, only we set `pNotification` instead of `pFence`. You can set both of these at the same
time and they should both work as expected. If using the `pNotification` system, you need to ensure
your `ma_async_notification_callbacks` object stays valid.
6.2. Resource Manager Implementation Details
--------------------------------------------
Resources are managed in two main ways:
* By storing the entire sound inside an in-memory buffer (referred to as a data buffer)
* By streaming audio data on the fly (referred to as a data stream)
A resource managed data source (`ma_resource_manager_data_source`) encapsulates a data buffer or
data stream, depending on whether or not the data source was initialized with the
`MA_RESOURCE_MANAGER_DATA_SOURCE_FLAG_STREAM` flag. If so, it will make use of a
`ma_resource_manager_data_stream` object. Otherwise it will use a `ma_resource_manager_data_buffer`
object. Both of these objects are data sources which means they can be used with any
`ma_data_source_*()` API.
Another major feature of the resource manager is the ability to asynchronously decode audio files.
This relieves the audio thread of time-consuming decoding which can negatively affect scalability
due to the audio thread needing to complete it's work extremely quickly to avoid glitching.
Asynchronous decoding is achieved through a job system. There is a central multi-producer,
multi-consumer, fixed-capacity job queue. When some asynchronous work needs to be done, a job is
posted to the queue which is then read by a job thread. The number of job threads can be
configured for improved scalability, and job threads can all run in parallel without needing to
worry about the order of execution (how this is achieved is explained below).
When a sound is being loaded asynchronously, playback can begin before the sound has been fully
decoded. This enables the application to start playback of the sound quickly, while at the same
time allowing to resource manager to keep loading in the background. Since there may be less
threads than the number of sounds being loaded at a given time, a simple scheduling system is used
to keep decoding time balanced and fair. The resource manager solves this by splitting decoding
into chunks called pages. By default, each page is 1 second long. When a page has been decoded, a
new job will be posted to start decoding the next page. By dividing up decoding into pages, an
individual sound shouldn't ever delay every other sound from having their first page decoded. Of
course, when loading many sounds at the same time, there will always be an amount of time required
to process jobs in the queue so in heavy load situations there will still be some delay. To
determine if a data source is ready to have some frames read, use
`ma_resource_manager_data_source_get_available_frames()`. This will return the number of frames
available starting from the current position.
6.2.1. Job Queue
----------------
The resource manager uses a job queue which is multi-producer, multi-consumer, and fixed-capacity.
This job queue is not currently lock-free, and instead uses a spinlock to achieve thread-safety.
Only a fixed number of jobs can be allocated and inserted into the queue which is done through a
lock-free data structure for allocating an index into a fixed sized array, with reference counting
for mitigation of the ABA problem. The reference count is 32-bit.
For many types of jobs it's important that they execute in a specific order. In these cases, jobs
are executed serially. For the resource manager, serial execution of jobs is only required on a
per-object basis (per data buffer or per data stream). Each of these objects stores an execution
counter. When a job is posted it is associated with an execution counter. When the job is
processed, it checks if the execution counter of the job equals the execution counter of the
owning object and if so, processes the job. If the counters are not equal, the job will be posted
back onto the job queue for later processing. When the job finishes processing the execution order
of the main object is incremented. This system means the no matter how many job threads are
executing, decoding of an individual sound will always get processed serially. The advantage to
having multiple threads comes into play when loading multiple sounds at the same time.
The resource manager's job queue is not 100% lock-free and will use a spinlock to achieve
thread-safety for a very small section of code. This is only relevant when the resource manager
uses more than one job thread. If only using a single job thread, which is the default, the
lock should never actually wait in practice. The amount of time spent locking should be quite
short, but it's something to be aware of for those who have pedantic lock-free requirements and
need to use more than one job thread. There are plans to remove this lock in a future version.
In addition, posting a job will release a semaphore, which on Win32 is implemented with
`ReleaseSemaphore` and on POSIX platforms via a condition variable:
share/public_html/static/music_worklet_inprogress/decoder/deps/miniaudio/miniaudio.h view on Meta::CPAN
/*
Copies a channel map if one is specified, otherwise copies the default channel map.
The output buffer must have a capacity of at least `channels`. If not NULL, the input channel map must also have a capacity of at least `channels`.
*/
MA_API void ma_channel_map_copy_or_default(ma_channel* pOut, size_t channelMapCapOut, const ma_channel* pIn, ma_uint32 channels);
/*
Determines whether or not a channel map is valid.
A blank channel map is valid (all channels set to MA_CHANNEL_NONE). The way a blank channel map is handled is context specific, but
is usually treated as a passthrough.
Invalid channel maps:
- A channel map with no channels
- A channel map with more than one channel and a mono channel
The channel map buffer must have a capacity of at least `channels`.
*/
MA_API ma_bool32 ma_channel_map_is_valid(const ma_channel* pChannelMap, ma_uint32 channels);
/*
Helper for comparing two channel maps for equality.
This assumes the channel count is the same between the two.
Both channels map buffers must have a capacity of at least `channels`.
*/
MA_API ma_bool32 ma_channel_map_is_equal(const ma_channel* pChannelMapA, const ma_channel* pChannelMapB, ma_uint32 channels);
/*
Helper for determining if a channel map is blank (all channels set to MA_CHANNEL_NONE).
The channel map buffer must have a capacity of at least `channels`.
*/
MA_API ma_bool32 ma_channel_map_is_blank(const ma_channel* pChannelMap, ma_uint32 channels);
/*
Helper for determining whether or not a channel is present in the given channel map.
The channel map buffer must have a capacity of at least `channels`.
*/
MA_API ma_bool32 ma_channel_map_contains_channel_position(ma_uint32 channels, const ma_channel* pChannelMap, ma_channel channelPosition);
/************************************************************************************************************************************************************
Conversion Helpers
************************************************************************************************************************************************************/
/*
High-level helper for doing a full format conversion in one go. Returns the number of output frames. Call this with pOut set to NULL to
determine the required size of the output buffer. frameCountOut should be set to the capacity of pOut. If pOut is NULL, frameCountOut is
ignored.
A return value of 0 indicates an error.
This function is useful for one-off bulk conversions, but if you're streaming data you should use the ma_data_converter APIs instead.
*/
MA_API ma_uint64 ma_convert_frames(void* pOut, ma_uint64 frameCountOut, ma_format formatOut, ma_uint32 channelsOut, ma_uint32 sampleRateOut, const void* pIn, ma_uint64 frameCountIn, ma_format formatIn, ma_uint32 channelsIn, ma_uint32 sampleRateIn);
MA_API ma_uint64 ma_convert_frames_ex(void* pOut, ma_uint64 frameCountOut, const void* pIn, ma_uint64 frameCountIn, const ma_data_converter_config* pConfig);
/************************************************************************************************************************************************************
Ring Buffer
************************************************************************************************************************************************************/
typedef struct
{
void* pBuffer;
ma_uint32 subbufferSizeInBytes;
ma_uint32 subbufferCount;
ma_uint32 subbufferStrideInBytes;
MA_ATOMIC(4, ma_uint32) encodedReadOffset; /* Most significant bit is the loop flag. Lower 31 bits contains the actual offset in bytes. Must be used atomically. */
MA_ATOMIC(4, ma_uint32) encodedWriteOffset; /* Most significant bit is the loop flag. Lower 31 bits contains the actual offset in bytes. Must be used atomically. */
ma_bool8 ownsBuffer; /* Used to know whether or not miniaudio is responsible for free()-ing the buffer. */
ma_bool8 clearOnWriteAcquire; /* When set, clears the acquired write buffer before returning from ma_rb_acquire_write(). */
ma_allocation_callbacks allocationCallbacks;
} ma_rb;
MA_API ma_result ma_rb_init_ex(size_t subbufferSizeInBytes, size_t subbufferCount, size_t subbufferStrideInBytes, void* pOptionalPreallocatedBuffer, const ma_allocation_callbacks* pAllocationCallbacks, ma_rb* pRB);
MA_API ma_result ma_rb_init(size_t bufferSizeInBytes, void* pOptionalPreallocatedBuffer, const ma_allocation_callbacks* pAllocationCallbacks, ma_rb* pRB);
MA_API void ma_rb_uninit(ma_rb* pRB);
MA_API void ma_rb_reset(ma_rb* pRB);
MA_API ma_result ma_rb_acquire_read(ma_rb* pRB, size_t* pSizeInBytes, void** ppBufferOut);
MA_API ma_result ma_rb_commit_read(ma_rb* pRB, size_t sizeInBytes);
MA_API ma_result ma_rb_acquire_write(ma_rb* pRB, size_t* pSizeInBytes, void** ppBufferOut);
MA_API ma_result ma_rb_commit_write(ma_rb* pRB, size_t sizeInBytes);
MA_API ma_result ma_rb_seek_read(ma_rb* pRB, size_t offsetInBytes);
MA_API ma_result ma_rb_seek_write(ma_rb* pRB, size_t offsetInBytes);
MA_API ma_int32 ma_rb_pointer_distance(ma_rb* pRB); /* Returns the distance between the write pointer and the read pointer. Should never be negative for a correct program. Will return the number of bytes that can be read before the read pointer hi...
MA_API ma_uint32 ma_rb_available_read(ma_rb* pRB);
MA_API ma_uint32 ma_rb_available_write(ma_rb* pRB);
MA_API size_t ma_rb_get_subbuffer_size(ma_rb* pRB);
MA_API size_t ma_rb_get_subbuffer_stride(ma_rb* pRB);
MA_API size_t ma_rb_get_subbuffer_offset(ma_rb* pRB, size_t subbufferIndex);
MA_API void* ma_rb_get_subbuffer_ptr(ma_rb* pRB, size_t subbufferIndex, void* pBuffer);
typedef struct
{
ma_rb rb;
ma_format format;
ma_uint32 channels;
} ma_pcm_rb;
MA_API ma_result ma_pcm_rb_init_ex(ma_format format, ma_uint32 channels, ma_uint32 subbufferSizeInFrames, ma_uint32 subbufferCount, ma_uint32 subbufferStrideInFrames, void* pOptionalPreallocatedBuffer, const ma_allocation_callbacks* pAllocationCallba...
MA_API ma_result ma_pcm_rb_init(ma_format format, ma_uint32 channels, ma_uint32 bufferSizeInFrames, void* pOptionalPreallocatedBuffer, const ma_allocation_callbacks* pAllocationCallbacks, ma_pcm_rb* pRB);
MA_API void ma_pcm_rb_uninit(ma_pcm_rb* pRB);
MA_API void ma_pcm_rb_reset(ma_pcm_rb* pRB);
MA_API ma_result ma_pcm_rb_acquire_read(ma_pcm_rb* pRB, ma_uint32* pSizeInFrames, void** ppBufferOut);
MA_API ma_result ma_pcm_rb_commit_read(ma_pcm_rb* pRB, ma_uint32 sizeInFrames);
MA_API ma_result ma_pcm_rb_acquire_write(ma_pcm_rb* pRB, ma_uint32* pSizeInFrames, void** ppBufferOut);
MA_API ma_result ma_pcm_rb_commit_write(ma_pcm_rb* pRB, ma_uint32 sizeInFrames);
MA_API ma_result ma_pcm_rb_seek_read(ma_pcm_rb* pRB, ma_uint32 offsetInFrames);
MA_API ma_result ma_pcm_rb_seek_write(ma_pcm_rb* pRB, ma_uint32 offsetInFrames);
MA_API ma_int32 ma_pcm_rb_pointer_distance(ma_pcm_rb* pRB); /* Return value is in frames. */
share/public_html/static/music_worklet_inprogress/decoder/deps/miniaudio/miniaudio.h view on Meta::CPAN
ma_uint64 totalLengthInPCMFrames; /* This is calculated when first loaded by the MA_JOB_TYPE_RESOURCE_MANAGER_LOAD_DATA_STREAM. */
ma_uint32 relativeCursor; /* The playback cursor, relative to the current page. Only ever accessed by the public API. Never accessed by the job thread. */
MA_ATOMIC(8, ma_uint64) absoluteCursor; /* The playback cursor, in absolute position starting from the start of the file. */
ma_uint32 currentPageIndex; /* Toggles between 0 and 1. Index 0 is the first half of pPageData. Index 1 is the second half. Only ever accessed by the public API. Never accessed by the job thread. */
MA_ATOMIC(4, ma_uint32) executionCounter; /* For allocating execution orders for jobs. */
MA_ATOMIC(4, ma_uint32) executionPointer; /* For managing the order of execution for asynchronous jobs relating to this object. Incremented as jobs complete processing. */
/* Written by the public API, read by the job thread. */
MA_ATOMIC(4, ma_bool32) isLooping; /* Whether or not the stream is looping. It's important to set the looping flag at the data stream level for smooth loop transitions. */
/* Written by the job thread, read by the public API. */
void* pPageData; /* Buffer containing the decoded data of each page. Allocated once at initialization time. */
MA_ATOMIC(4, ma_uint32) pageFrameCount[2]; /* The number of valid PCM frames in each page. Used to determine the last valid frame. */
/* Written and read by both the public API and the job thread. These must be atomic. */
MA_ATOMIC(4, ma_result) result; /* Result from asynchronous loading. When loading set to MA_BUSY. When initialized set to MA_SUCCESS. When deleting set to MA_UNAVAILABLE. If an error occurs when loading, set to an error code. */
MA_ATOMIC(4, ma_bool32) isDecoderAtEnd; /* Whether or not the decoder has reached the end. */
MA_ATOMIC(4, ma_bool32) isPageValid[2]; /* Booleans to indicate whether or not a page is valid. Set to false by the public API, set to true by the job thread. Set to false as the pages are consumed, true when they are filled. */
MA_ATOMIC(4, ma_bool32) seekCounter; /* When 0, no seeking is being performed. When > 0, a seek is being performed and reading should be delayed with MA_BUSY. */
};
struct ma_resource_manager_data_source
{
union
{
ma_resource_manager_data_buffer buffer;
ma_resource_manager_data_stream stream;
} backend; /* Must be the first item because we need the first item to be the data source callbacks for the buffer or stream. */
ma_uint32 flags; /* The flags that were passed in to ma_resource_manager_data_source_init(). */
MA_ATOMIC(4, ma_uint32) executionCounter; /* For allocating execution orders for jobs. */
MA_ATOMIC(4, ma_uint32) executionPointer; /* For managing the order of execution for asynchronous jobs relating to this object. Incremented as jobs complete processing. */
};
typedef struct
{
ma_allocation_callbacks allocationCallbacks;
ma_log* pLog;
ma_format decodedFormat; /* The decoded format to use. Set to ma_format_unknown (default) to use the file's native format. */
ma_uint32 decodedChannels; /* The decoded channel count to use. Set to 0 (default) to use the file's native channel count. */
ma_uint32 decodedSampleRate; /* the decoded sample rate to use. Set to 0 (default) to use the file's native sample rate. */
ma_uint32 jobThreadCount; /* Set to 0 if you want to self-manage your job threads. Defaults to 1. */
ma_uint32 jobQueueCapacity; /* The maximum number of jobs that can fit in the queue at a time. Defaults to MA_JOB_TYPE_RESOURCE_MANAGER_QUEUE_CAPACITY. Cannot be zero. */
ma_uint32 flags;
ma_vfs* pVFS; /* Can be NULL in which case defaults will be used. */
ma_decoding_backend_vtable** ppCustomDecodingBackendVTables;
ma_uint32 customDecodingBackendCount;
void* pCustomDecodingBackendUserData;
} ma_resource_manager_config;
MA_API ma_resource_manager_config ma_resource_manager_config_init(void);
struct ma_resource_manager
{
ma_resource_manager_config config;
ma_resource_manager_data_buffer_node* pRootDataBufferNode; /* The root buffer in the binary tree. */
#ifndef MA_NO_THREADING
ma_mutex dataBufferBSTLock; /* For synchronizing access to the data buffer binary tree. */
ma_thread jobThreads[MA_RESOURCE_MANAGER_MAX_JOB_THREAD_COUNT]; /* The threads for executing jobs. */
#endif
ma_job_queue jobQueue; /* Multi-consumer, multi-producer job queue for managing jobs for asynchronous decoding and streaming. */
ma_default_vfs defaultVFS; /* Only used if a custom VFS is not specified. */
ma_log log; /* Only used if no log was specified in the config. */
};
/* Init. */
MA_API ma_result ma_resource_manager_init(const ma_resource_manager_config* pConfig, ma_resource_manager* pResourceManager);
MA_API void ma_resource_manager_uninit(ma_resource_manager* pResourceManager);
MA_API ma_log* ma_resource_manager_get_log(ma_resource_manager* pResourceManager);
/* Registration. */
MA_API ma_result ma_resource_manager_register_file(ma_resource_manager* pResourceManager, const char* pFilePath, ma_uint32 flags);
MA_API ma_result ma_resource_manager_register_file_w(ma_resource_manager* pResourceManager, const wchar_t* pFilePath, ma_uint32 flags);
MA_API ma_result ma_resource_manager_register_decoded_data(ma_resource_manager* pResourceManager, const char* pName, const void* pData, ma_uint64 frameCount, ma_format format, ma_uint32 channels, ma_uint32 sampleRate); /* Does not copy. Increments t...
MA_API ma_result ma_resource_manager_register_decoded_data_w(ma_resource_manager* pResourceManager, const wchar_t* pName, const void* pData, ma_uint64 frameCount, ma_format format, ma_uint32 channels, ma_uint32 sampleRate);
MA_API ma_result ma_resource_manager_register_encoded_data(ma_resource_manager* pResourceManager, const char* pName, const void* pData, size_t sizeInBytes); /* Does not copy. Increments the reference count if already exists and returns MA_SUCCESS....
MA_API ma_result ma_resource_manager_register_encoded_data_w(ma_resource_manager* pResourceManager, const wchar_t* pName, const void* pData, size_t sizeInBytes);
MA_API ma_result ma_resource_manager_unregister_file(ma_resource_manager* pResourceManager, const char* pFilePath);
MA_API ma_result ma_resource_manager_unregister_file_w(ma_resource_manager* pResourceManager, const wchar_t* pFilePath);
MA_API ma_result ma_resource_manager_unregister_data(ma_resource_manager* pResourceManager, const char* pName);
MA_API ma_result ma_resource_manager_unregister_data_w(ma_resource_manager* pResourceManager, const wchar_t* pName);
/* Data Buffers. */
MA_API ma_result ma_resource_manager_data_buffer_init_ex(ma_resource_manager* pResourceManager, const ma_resource_manager_data_source_config* pConfig, ma_resource_manager_data_buffer* pDataBuffer);
MA_API ma_result ma_resource_manager_data_buffer_init(ma_resource_manager* pResourceManager, const char* pFilePath, ma_uint32 flags, const ma_resource_manager_pipeline_notifications* pNotifications, ma_resource_manager_data_buffer* pDataBuffer);
MA_API ma_result ma_resource_manager_data_buffer_init_w(ma_resource_manager* pResourceManager, const wchar_t* pFilePath, ma_uint32 flags, const ma_resource_manager_pipeline_notifications* pNotifications, ma_resource_manager_data_buffer* pDataBuffer);
MA_API ma_result ma_resource_manager_data_buffer_init_copy(ma_resource_manager* pResourceManager, const ma_resource_manager_data_buffer* pExistingDataBuffer, ma_resource_manager_data_buffer* pDataBuffer);
MA_API ma_result ma_resource_manager_data_buffer_uninit(ma_resource_manager_data_buffer* pDataBuffer);
MA_API ma_result ma_resource_manager_data_buffer_read_pcm_frames(ma_resource_manager_data_buffer* pDataBuffer, void* pFramesOut, ma_uint64 frameCount, ma_uint64* pFramesRead);
MA_API ma_result ma_resource_manager_data_buffer_seek_to_pcm_frame(ma_resource_manager_data_buffer* pDataBuffer, ma_uint64 frameIndex);
MA_API ma_result ma_resource_manager_data_buffer_get_data_format(ma_resource_manager_data_buffer* pDataBuffer, ma_format* pFormat, ma_uint32* pChannels, ma_uint32* pSampleRate, ma_channel* pChannelMap, size_t channelMapCap);
MA_API ma_result ma_resource_manager_data_buffer_get_cursor_in_pcm_frames(ma_resource_manager_data_buffer* pDataBuffer, ma_uint64* pCursor);
MA_API ma_result ma_resource_manager_data_buffer_get_length_in_pcm_frames(ma_resource_manager_data_buffer* pDataBuffer, ma_uint64* pLength);
MA_API ma_result ma_resource_manager_data_buffer_result(const ma_resource_manager_data_buffer* pDataBuffer);
MA_API ma_result ma_resource_manager_data_buffer_set_looping(ma_resource_manager_data_buffer* pDataBuffer, ma_bool32 isLooping);
MA_API ma_bool32 ma_resource_manager_data_buffer_is_looping(const ma_resource_manager_data_buffer* pDataBuffer);
MA_API ma_result ma_resource_manager_data_buffer_get_available_frames(ma_resource_manager_data_buffer* pDataBuffer, ma_uint64* pAvailableFrames);
/* Data Streams. */
MA_API ma_result ma_resource_manager_data_stream_init_ex(ma_resource_manager* pResourceManager, const ma_resource_manager_data_source_config* pConfig, ma_resource_manager_data_stream* pDataStream);
MA_API ma_result ma_resource_manager_data_stream_init(ma_resource_manager* pResourceManager, const char* pFilePath, ma_uint32 flags, const ma_resource_manager_pipeline_notifications* pNotifications, ma_resource_manager_data_stream* pDataStream);
MA_API ma_result ma_resource_manager_data_stream_init_w(ma_resource_manager* pResourceManager, const wchar_t* pFilePath, ma_uint32 flags, const ma_resource_manager_pipeline_notifications* pNotifications, ma_resource_manager_data_stream* pDataStream);
MA_API ma_result ma_resource_manager_data_stream_uninit(ma_resource_manager_data_stream* pDataStream);
MA_API ma_result ma_resource_manager_data_stream_read_pcm_frames(ma_resource_manager_data_stream* pDataStream, void* pFramesOut, ma_uint64 frameCount, ma_uint64* pFramesRead);
MA_API ma_result ma_resource_manager_data_stream_seek_to_pcm_frame(ma_resource_manager_data_stream* pDataStream, ma_uint64 frameIndex);
MA_API ma_result ma_resource_manager_data_stream_get_data_format(ma_resource_manager_data_stream* pDataStream, ma_format* pFormat, ma_uint32* pChannels, ma_uint32* pSampleRate, ma_channel* pChannelMap, size_t channelMapCap);
MA_API ma_result ma_resource_manager_data_stream_get_cursor_in_pcm_frames(ma_resource_manager_data_stream* pDataStream, ma_uint64* pCursor);
MA_API ma_result ma_resource_manager_data_stream_get_length_in_pcm_frames(ma_resource_manager_data_stream* pDataStream, ma_uint64* pLength);
MA_API ma_result ma_resource_manager_data_stream_result(const ma_resource_manager_data_stream* pDataStream);
MA_API ma_result ma_resource_manager_data_stream_set_looping(ma_resource_manager_data_stream* pDataStream, ma_bool32 isLooping);
MA_API ma_bool32 ma_resource_manager_data_stream_is_looping(const ma_resource_manager_data_stream* pDataStream);
MA_API ma_result ma_resource_manager_data_stream_get_available_frames(ma_resource_manager_data_stream* pDataStream, ma_uint64* pAvailableFrames);
/* Data Sources. */
MA_API ma_result ma_resource_manager_data_source_init_ex(ma_resource_manager* pResourceManager, const ma_resource_manager_data_source_config* pConfig, ma_resource_manager_data_source* pDataSource);
MA_API ma_result ma_resource_manager_data_source_init(ma_resource_manager* pResourceManager, const char* pName, ma_uint32 flags, const ma_resource_manager_pipeline_notifications* pNotifications, ma_resource_manager_data_source* pDataSource);
MA_API ma_result ma_resource_manager_data_source_init_w(ma_resource_manager* pResourceManager, const wchar_t* pName, ma_uint32 flags, const ma_resource_manager_pipeline_notifications* pNotifications, ma_resource_manager_data_source* pDataSource);
MA_API ma_result ma_resource_manager_data_source_init_copy(ma_resource_manager* pResourceManager, const ma_resource_manager_data_source* pExistingDataSource, ma_resource_manager_data_source* pDataSource);
MA_API ma_result ma_resource_manager_data_source_uninit(ma_resource_manager_data_source* pDataSource);
MA_API ma_result ma_resource_manager_data_source_read_pcm_frames(ma_resource_manager_data_source* pDataSource, void* pFramesOut, ma_uint64 frameCount, ma_uint64* pFramesRead);
MA_API ma_result ma_resource_manager_data_source_seek_to_pcm_frame(ma_resource_manager_data_source* pDataSource, ma_uint64 frameIndex);
share/public_html/static/music_worklet_inprogress/decoder/deps/miniaudio/miniaudio.h view on Meta::CPAN
is done with `pa_context_new()`, it's not actually connected to anything. When you connect, you call `pa_context_connect()`. However, if
you remember, PulseAudio is an asynchronous API. That means you cannot just assume the context is connected after `pa_context_context()`
has returned. You instead need to wait for it to connect. To do this, you need to either wait for a callback to get fired, which you can
set with `pa_context_set_state_callback()`, or you can continuously poll the context's state. Either way, you need to run this in a loop.
All objects from here out are created from the context, and, I believe, you can't be creating these objects until the context is connected.
This waiting loop is therefore unavoidable. In order for the waiting to ever complete, however, the main loop needs to be running. Before
attempting to connect the context, the main loop needs to be started with `pa_threaded_mainloop_start()`.
The reason for this asynchronous design is to support cases where you're connecting to a remote server, say through a local network or an
internet connection. However, the *VAST* majority of cases don't involve this at all - they just connect to a local "server" running on the
host machine. The fact that this would be the default rather than making `pa_context_connect()` synchronous tends to boggle the mind.
Once the context has been created and connected you can start creating a stream. A PulseAudio stream is analogous to miniaudio's device.
The initialization of a stream is fairly standard - you configure some attributes (analogous to miniaudio's device config) and then call
`pa_stream_new()` to actually create it. Here is where we start to get into "operations". When configuring the stream, you can get
information about the source (such as sample format, sample rate, etc.), however it's not synchronous. Instead, a `pa_operation` object
is returned from `pa_context_get_source_info_by_name()` (capture) or `pa_context_get_sink_info_by_name()` (playback). Then, you need to
run a loop (again!) to wait for the operation to complete which you can determine via a callback or polling, just like we did with the
context. Then, as an added bonus, you need to decrement the reference counter of the `pa_operation` object to ensure memory is cleaned up.
All of that just to retrieve basic information about a device!
Once the basic information about the device has been retrieved, miniaudio can now create the stream with `ma_stream_new()`. Like the
context, this needs to be connected. But we need to be careful here, because we're now about to introduce one of the most horrific design
choices in PulseAudio.
PulseAudio allows you to specify a callback that is fired when data can be written to or read from a stream. The language is important here
because PulseAudio takes it literally, specifically the "can be". You would think these callbacks would be appropriate as the place for
writing and reading data to and from the stream, and that would be right, except when it's not. When you initialize the stream, you can
set a flag that tells PulseAudio to not start the stream automatically. This is required because miniaudio does not auto-start devices
straight after initialization - you need to call `ma_device_start()` manually. The problem is that even when this flag is specified,
PulseAudio will immediately fire it's write or read callback. This is *technically* correct (based on the wording in the documentation)
because indeed, data *can* be written at this point. The problem is that it's not *practical*. It makes sense that the write/read callback
would be where a program will want to write or read data to or from the stream, but when it's called before the application has even
requested that the stream be started, it's just not practical because the program probably isn't ready for any kind of data delivery at
that point (it may still need to load files or whatnot). Instead, this callback should only be fired when the application requests the
stream be started which is how it works with literally *every* other callback-based audio API. Since miniaudio forbids firing of the data
callback until the device has been started (as it should be with *all* callback based APIs), logic needs to be added to ensure miniaudio
doesn't just blindly fire the application-defined data callback from within the PulseAudio callback before the stream has actually been
started. The device state is used for this - if the state is anything other than `ma_device_state_starting` or `ma_device_state_started`, the main data
callback is not fired.
This, unfortunately, is not the end of the problems with the PulseAudio write callback. Any normal callback based audio API will
continuously fire the callback at regular intervals based on the size of the internal buffer. This will only ever be fired when the device
is running, and will be fired regardless of whether or not the user actually wrote anything to the device/stream. This not the case in
PulseAudio. In PulseAudio, the data callback will *only* be called if you wrote something to it previously. That means, if you don't call
`pa_stream_write()`, the callback will not get fired. On the surface you wouldn't think this would matter because you should be always
writing data, and if you don't have anything to write, just write silence. That's fine until you want to drain the stream. You see, if
you're continuously writing data to the stream, the stream will never get drained! That means in order to drain the stream, you need to
*not* write data to it! But remember, when you don't write data to the stream, the callback won't get fired again! Why is draining
important? Because that's how we've defined stopping to work in miniaudio. In miniaudio, stopping the device requires it to be drained
before returning from ma_device_stop(). So we've stopped the device, which requires us to drain, but draining requires us to *not* write
data to the stream (or else it won't ever complete draining), but not writing to the stream means the callback won't get fired again!
This becomes a problem when stopping and then restarting the device. When the device is stopped, it's drained, which requires us to *not*
write anything to the stream. But then, since we didn't write anything to it, the write callback will *never* get called again if we just
resume the stream naively. This means that starting the stream requires us to write data to the stream from outside the callback. This
disconnect is something PulseAudio has got seriously wrong - there should only ever be a single source of data delivery, that being the
callback. (I have tried using `pa_stream_flush()` to trigger the write callback to fire, but this just doesn't work for some reason.)
Once you've created the stream, you need to connect it which involves the whole waiting procedure. This is the same process as the context,
only this time you'll poll for the state with `pa_stream_get_status()`. The starting and stopping of a streaming is referred to as
"corking" in PulseAudio. The analogy is corking a barrel. To start the stream, you uncork it, to stop it you cork it. Personally I think
it's silly - why would you not just call it "starting" and "stopping" like any other normal audio API? Anyway, the act of corking is, you
guessed it, asynchronous. This means you'll need our waiting loop as usual. Again, why this asynchronous design is the default is
absolutely beyond me. Would it really be that hard to just make it run synchronously?
Teardown is pretty simple (what?!). It's just a matter of calling the relevant `_unref()` function on each object in reverse order that
they were initialized in.
That's about it from the PulseAudio side. A bit ranty, I know, but they really need to fix that main loop and callback system. They're
embarrassingly unpractical. The main loop thing is an easy fix - have synchronous versions of all APIs. If an application wants these to
run asynchronously, they can execute them in a separate thread themselves. The desire to run these asynchronously is such a niche
requirement - it makes no sense to make it the default. The stream write callback needs to be change, or an alternative provided, that is
constantly fired, regardless of whether or not `pa_stream_write()` has been called, and it needs to take a pointer to a buffer as a
parameter which the program just writes to directly rather than having to call `pa_stream_writable_size()` and `pa_stream_write()`. These
changes alone will change PulseAudio from one of the worst audio APIs to one of the best.
*/
/*
It is assumed pulseaudio.h is available when linking at compile time. When linking at compile time, we use the declarations in the header
to check for type safety. We cannot do this when linking at run time because the header might not be available.
*/
#ifdef MA_NO_RUNTIME_LINKING
/* pulseaudio.h marks some functions with "inline" which isn't always supported. Need to emulate it. */
#if !defined(__cplusplus)
#if defined(__STRICT_ANSI__)
#if !defined(inline)
#define inline __inline__ __attribute__((always_inline))
#define MA_INLINE_DEFINED
#endif
#endif
#endif
#include <pulse/pulseaudio.h>
#if defined(MA_INLINE_DEFINED)
#undef inline
#undef MA_INLINE_DEFINED
#endif
#define MA_PA_OK PA_OK
#define MA_PA_ERR_ACCESS PA_ERR_ACCESS
#define MA_PA_ERR_INVALID PA_ERR_INVALID
#define MA_PA_ERR_NOENTITY PA_ERR_NOENTITY
#define MA_PA_ERR_NOTSUPPORTED PA_ERR_NOTSUPPORTED
#define MA_PA_CHANNELS_MAX PA_CHANNELS_MAX
#define MA_PA_RATE_MAX PA_RATE_MAX
typedef pa_context_flags_t ma_pa_context_flags_t;
#define MA_PA_CONTEXT_NOFLAGS PA_CONTEXT_NOFLAGS
#define MA_PA_CONTEXT_NOAUTOSPAWN PA_CONTEXT_NOAUTOSPAWN
#define MA_PA_CONTEXT_NOFAIL PA_CONTEXT_NOFAIL
typedef pa_stream_flags_t ma_pa_stream_flags_t;
#define MA_PA_STREAM_NOFLAGS PA_STREAM_NOFLAGS
#define MA_PA_STREAM_START_CORKED PA_STREAM_START_CORKED
#define MA_PA_STREAM_INTERPOLATE_TIMING PA_STREAM_INTERPOLATE_TIMING
#define MA_PA_STREAM_NOT_MONOTONIC PA_STREAM_NOT_MONOTONIC
#define MA_PA_STREAM_AUTO_TIMING_UPDATE PA_STREAM_AUTO_TIMING_UPDATE
#define MA_PA_STREAM_NO_REMAP_CHANNELS PA_STREAM_NO_REMAP_CHANNELS
( run in 1.069 second using v1.01-cache-2.11-cpan-75ffa21a3d4 )