835 lines
20 KiB
C
835 lines
20 KiB
C
#include <AudioUnit/AudioUnit.h>
|
|
#include <CoreFoundation/CFString.h>
|
|
#include <CoreAudio/CoreAudio.h>
|
|
#include <unistd.h>
|
|
#include <errno.h>
|
|
|
|
#include <obs-module.h>
|
|
#include <util/threading.h>
|
|
#include <util/c99defs.h>
|
|
#include <util/apple/cfstring-utils.h>
|
|
|
|
#include "audio-device-enum.h"
|
|
|
|
#define PROPERTY_DEFAULT_DEVICE kAudioHardwarePropertyDefaultInputDevice
|
|
#define PROPERTY_FORMATS kAudioStreamPropertyAvailablePhysicalFormats
|
|
|
|
#define SCOPE_OUTPUT kAudioUnitScope_Output
|
|
#define SCOPE_INPUT kAudioUnitScope_Input
|
|
#define SCOPE_GLOBAL kAudioUnitScope_Global
|
|
|
|
#define BUS_OUTPUT 0
|
|
#define BUS_INPUT 1
|
|
|
|
#define MAX_DEVICES 20
|
|
|
|
#define set_property AudioUnitSetProperty
|
|
#define get_property AudioUnitGetProperty
|
|
|
|
#define TEXT_AUDIO_INPUT obs_module_text("CoreAudio.InputCapture");
|
|
#define TEXT_AUDIO_OUTPUT obs_module_text("CoreAudio.OutputCapture");
|
|
#define TEXT_DEVICE obs_module_text("CoreAudio.Device")
|
|
#define TEXT_DEVICE_DEFAULT obs_module_text("CoreAudio.Device.Default")
|
|
|
|
struct coreaudio_data {
|
|
char *device_name;
|
|
char *device_uid;
|
|
AudioUnit unit;
|
|
AudioDeviceID device_id;
|
|
AudioBufferList *buf_list;
|
|
bool au_initialized;
|
|
bool active;
|
|
bool default_device;
|
|
bool input;
|
|
bool no_devices;
|
|
|
|
uint32_t sample_rate;
|
|
enum audio_format format;
|
|
enum speaker_layout speakers;
|
|
|
|
pthread_t reconnect_thread;
|
|
os_event_t *exit_event;
|
|
volatile bool reconnecting;
|
|
unsigned long retry_time;
|
|
|
|
obs_source_t *source;
|
|
};
|
|
|
|
static bool get_default_output_device(struct coreaudio_data *ca)
|
|
{
|
|
struct device_list list;
|
|
|
|
memset(&list, 0, sizeof(struct device_list));
|
|
coreaudio_enum_devices(&list, false);
|
|
|
|
if (!list.items.num)
|
|
return false;
|
|
|
|
bfree(ca->device_uid);
|
|
ca->device_uid = bstrdup(list.items.array[0].value.array);
|
|
|
|
device_list_free(&list);
|
|
return true;
|
|
}
|
|
|
|
static bool find_device_id_by_uid(struct coreaudio_data *ca)
|
|
{
|
|
UInt32 size = sizeof(AudioDeviceID);
|
|
CFStringRef cf_uid = NULL;
|
|
CFStringRef qual = NULL;
|
|
UInt32 qual_size = 0;
|
|
OSStatus stat;
|
|
bool success;
|
|
|
|
AudioObjectPropertyAddress addr = {
|
|
.mScope = kAudioObjectPropertyScopeGlobal,
|
|
.mElement = kAudioObjectPropertyElementMaster};
|
|
|
|
if (!ca->device_uid)
|
|
ca->device_uid = bstrdup("default");
|
|
|
|
ca->default_device = false;
|
|
ca->no_devices = false;
|
|
|
|
/* have to do this because mac output devices don't actually exist */
|
|
if (astrcmpi(ca->device_uid, "default") == 0) {
|
|
if (ca->input) {
|
|
ca->default_device = true;
|
|
} else {
|
|
if (!get_default_output_device(ca)) {
|
|
ca->no_devices = true;
|
|
return false;
|
|
}
|
|
}
|
|
}
|
|
|
|
cf_uid = CFStringCreateWithCString(NULL, ca->device_uid,
|
|
kCFStringEncodingUTF8);
|
|
|
|
if (ca->default_device) {
|
|
addr.mSelector = PROPERTY_DEFAULT_DEVICE;
|
|
stat = AudioObjectGetPropertyData(kAudioObjectSystemObject,
|
|
&addr, qual_size, &qual,
|
|
&size, &ca->device_id);
|
|
success = (stat == noErr);
|
|
} else {
|
|
success = coreaudio_get_device_id(cf_uid, &ca->device_id);
|
|
}
|
|
|
|
if (cf_uid)
|
|
CFRelease(cf_uid);
|
|
|
|
return success;
|
|
}
|
|
|
|
static inline void ca_warn(struct coreaudio_data *ca, const char *func,
|
|
const char *format, ...)
|
|
{
|
|
va_list args;
|
|
struct dstr str = {0};
|
|
|
|
va_start(args, format);
|
|
|
|
dstr_printf(&str, "[%s]:[device '%s'] ", func, ca->device_name);
|
|
dstr_vcatf(&str, format, args);
|
|
blog(LOG_WARNING, "%s", str.array);
|
|
dstr_free(&str);
|
|
|
|
va_end(args);
|
|
}
|
|
|
|
static inline bool ca_success(OSStatus stat, struct coreaudio_data *ca,
|
|
const char *func, const char *action)
|
|
{
|
|
if (stat != noErr) {
|
|
blog(LOG_WARNING, "[%s]:[device '%s'] %s failed: %d", func,
|
|
ca->device_name, action, (int)stat);
|
|
return false;
|
|
}
|
|
|
|
return true;
|
|
}
|
|
|
|
enum coreaudio_io_type {
|
|
IO_TYPE_INPUT,
|
|
IO_TYPE_OUTPUT,
|
|
};
|
|
|
|
static inline bool enable_io(struct coreaudio_data *ca,
|
|
enum coreaudio_io_type type, bool enable)
|
|
{
|
|
UInt32 enable_int = enable;
|
|
return set_property(ca->unit, kAudioOutputUnitProperty_EnableIO,
|
|
(type == IO_TYPE_INPUT) ? SCOPE_INPUT
|
|
: SCOPE_OUTPUT,
|
|
(type == IO_TYPE_INPUT) ? BUS_INPUT : BUS_OUTPUT,
|
|
&enable_int, sizeof(enable_int));
|
|
}
|
|
|
|
static inline enum audio_format convert_ca_format(UInt32 format_flags,
|
|
UInt32 bits)
|
|
{
|
|
bool planar = (format_flags & kAudioFormatFlagIsNonInterleaved) != 0;
|
|
|
|
if (format_flags & kAudioFormatFlagIsFloat)
|
|
return planar ? AUDIO_FORMAT_FLOAT_PLANAR : AUDIO_FORMAT_FLOAT;
|
|
|
|
if (!(format_flags & kAudioFormatFlagIsSignedInteger) && bits == 8)
|
|
return planar ? AUDIO_FORMAT_U8BIT_PLANAR : AUDIO_FORMAT_U8BIT;
|
|
|
|
/* not float? not signed int? no clue, fail */
|
|
if ((format_flags & kAudioFormatFlagIsSignedInteger) == 0)
|
|
return AUDIO_FORMAT_UNKNOWN;
|
|
|
|
if (bits == 16)
|
|
return planar ? AUDIO_FORMAT_16BIT_PLANAR : AUDIO_FORMAT_16BIT;
|
|
else if (bits == 32)
|
|
return planar ? AUDIO_FORMAT_32BIT_PLANAR : AUDIO_FORMAT_32BIT;
|
|
|
|
return AUDIO_FORMAT_UNKNOWN;
|
|
}
|
|
|
|
static inline enum speaker_layout convert_ca_speaker_layout(UInt32 channels)
|
|
{
|
|
switch (channels) {
|
|
case 1:
|
|
return SPEAKERS_MONO;
|
|
case 2:
|
|
return SPEAKERS_STEREO;
|
|
case 3:
|
|
return SPEAKERS_2POINT1;
|
|
case 4:
|
|
return SPEAKERS_4POINT0;
|
|
case 5:
|
|
return SPEAKERS_4POINT1;
|
|
case 6:
|
|
return SPEAKERS_5POINT1;
|
|
case 8:
|
|
return SPEAKERS_7POINT1;
|
|
}
|
|
|
|
return SPEAKERS_UNKNOWN;
|
|
}
|
|
|
|
static bool coreaudio_init_format(struct coreaudio_data *ca)
|
|
{
|
|
AudioStreamBasicDescription desc;
|
|
OSStatus stat;
|
|
UInt32 size = sizeof(desc);
|
|
struct obs_audio_info aoi;
|
|
int channels;
|
|
|
|
if (!obs_get_audio_info(&aoi)) {
|
|
blog(LOG_WARNING, "No active audio");
|
|
return false;
|
|
}
|
|
channels = get_audio_channels(aoi.speakers);
|
|
|
|
stat = get_property(ca->unit, kAudioUnitProperty_StreamFormat,
|
|
SCOPE_INPUT, BUS_INPUT, &desc, &size);
|
|
if (!ca_success(stat, ca, "coreaudio_init_format", "get input format"))
|
|
return false;
|
|
|
|
/* Certain types of devices have no limit on channel count, and
|
|
* there's no way to know the actual number of channels it's using,
|
|
* so if we encounter this situation just force to what is defined in output */
|
|
if (desc.mChannelsPerFrame > 8) {
|
|
desc.mChannelsPerFrame = channels;
|
|
desc.mBytesPerFrame = channels * desc.mBitsPerChannel / 8;
|
|
desc.mBytesPerPacket =
|
|
desc.mFramesPerPacket * desc.mBytesPerFrame;
|
|
}
|
|
|
|
stat = set_property(ca->unit, kAudioUnitProperty_StreamFormat,
|
|
SCOPE_OUTPUT, BUS_INPUT, &desc, size);
|
|
if (!ca_success(stat, ca, "coreaudio_init_format", "set output format"))
|
|
return false;
|
|
|
|
if (desc.mFormatID != kAudioFormatLinearPCM) {
|
|
ca_warn(ca, "coreaudio_init_format", "format is not PCM");
|
|
return false;
|
|
}
|
|
|
|
ca->format = convert_ca_format(desc.mFormatFlags, desc.mBitsPerChannel);
|
|
if (ca->format == AUDIO_FORMAT_UNKNOWN) {
|
|
ca_warn(ca, "coreaudio_init_format",
|
|
"unknown format flags: "
|
|
"%u, bits: %u",
|
|
(unsigned int)desc.mFormatFlags,
|
|
(unsigned int)desc.mBitsPerChannel);
|
|
return false;
|
|
}
|
|
|
|
ca->sample_rate = (uint32_t)desc.mSampleRate;
|
|
ca->speakers = convert_ca_speaker_layout(desc.mChannelsPerFrame);
|
|
|
|
if (ca->speakers == SPEAKERS_UNKNOWN) {
|
|
ca_warn(ca, "coreaudio_init_format",
|
|
"unknown speaker layout: "
|
|
"%u channels",
|
|
(unsigned int)desc.mChannelsPerFrame);
|
|
return false;
|
|
}
|
|
|
|
return true;
|
|
}
|
|
|
|
static bool coreaudio_init_buffer(struct coreaudio_data *ca)
|
|
{
|
|
UInt32 buf_size = 0;
|
|
UInt32 size = 0;
|
|
UInt32 frames = 0;
|
|
OSStatus stat;
|
|
|
|
AudioObjectPropertyAddress addr = {
|
|
kAudioDevicePropertyStreamConfiguration,
|
|
kAudioDevicePropertyScopeInput,
|
|
kAudioObjectPropertyElementMaster};
|
|
|
|
stat = AudioObjectGetPropertyDataSize(ca->device_id, &addr, 0, NULL,
|
|
&buf_size);
|
|
if (!ca_success(stat, ca, "coreaudio_init_buffer", "get list size"))
|
|
return false;
|
|
|
|
size = sizeof(frames);
|
|
stat = get_property(ca->unit, kAudioDevicePropertyBufferFrameSize,
|
|
SCOPE_GLOBAL, 0, &frames, &size);
|
|
if (!ca_success(stat, ca, "coreaudio_init_buffer", "get frame size"))
|
|
return false;
|
|
|
|
/* ---------------------- */
|
|
|
|
ca->buf_list = bmalloc(buf_size);
|
|
|
|
stat = AudioObjectGetPropertyData(ca->device_id, &addr, 0, NULL,
|
|
&buf_size, ca->buf_list);
|
|
if (!ca_success(stat, ca, "coreaudio_init_buffer", "allocate")) {
|
|
bfree(ca->buf_list);
|
|
ca->buf_list = NULL;
|
|
return false;
|
|
}
|
|
|
|
for (UInt32 i = 0; i < ca->buf_list->mNumberBuffers; i++) {
|
|
size = ca->buf_list->mBuffers[i].mDataByteSize;
|
|
ca->buf_list->mBuffers[i].mData = bmalloc(size);
|
|
}
|
|
|
|
return true;
|
|
}
|
|
|
|
static void buf_list_free(AudioBufferList *buf_list)
|
|
{
|
|
if (buf_list) {
|
|
for (UInt32 i = 0; i < buf_list->mNumberBuffers; i++)
|
|
bfree(buf_list->mBuffers[i].mData);
|
|
|
|
bfree(buf_list);
|
|
}
|
|
}
|
|
|
|
static OSStatus input_callback(void *data,
|
|
AudioUnitRenderActionFlags *action_flags,
|
|
const AudioTimeStamp *ts_data, UInt32 bus_num,
|
|
UInt32 frames, AudioBufferList *ignored_buffers)
|
|
{
|
|
struct coreaudio_data *ca = data;
|
|
OSStatus stat;
|
|
struct obs_source_audio audio;
|
|
|
|
stat = AudioUnitRender(ca->unit, action_flags, ts_data, bus_num, frames,
|
|
ca->buf_list);
|
|
if (!ca_success(stat, ca, "input_callback", "audio retrieval"))
|
|
return noErr;
|
|
|
|
for (UInt32 i = 0; i < ca->buf_list->mNumberBuffers; i++)
|
|
audio.data[i] = ca->buf_list->mBuffers[i].mData;
|
|
|
|
audio.frames = frames;
|
|
audio.speakers = ca->speakers;
|
|
audio.format = ca->format;
|
|
audio.samples_per_sec = ca->sample_rate;
|
|
audio.timestamp = ts_data->mHostTime;
|
|
|
|
obs_source_output_audio(ca->source, &audio);
|
|
|
|
UNUSED_PARAMETER(ignored_buffers);
|
|
return noErr;
|
|
}
|
|
|
|
static void coreaudio_stop(struct coreaudio_data *ca);
|
|
static bool coreaudio_init(struct coreaudio_data *ca);
|
|
static void coreaudio_uninit(struct coreaudio_data *ca);
|
|
|
|
static void *reconnect_thread(void *param)
|
|
{
|
|
struct coreaudio_data *ca = param;
|
|
|
|
ca->reconnecting = true;
|
|
|
|
while (os_event_timedwait(ca->exit_event, ca->retry_time) ==
|
|
ETIMEDOUT) {
|
|
if (coreaudio_init(ca))
|
|
break;
|
|
}
|
|
|
|
blog(LOG_DEBUG, "coreaudio: exit the reconnect thread");
|
|
ca->reconnecting = false;
|
|
return NULL;
|
|
}
|
|
|
|
static void coreaudio_begin_reconnect(struct coreaudio_data *ca)
|
|
{
|
|
int ret;
|
|
|
|
if (ca->reconnecting)
|
|
return;
|
|
|
|
ret = pthread_create(&ca->reconnect_thread, NULL, reconnect_thread, ca);
|
|
if (ret != 0)
|
|
blog(LOG_WARNING,
|
|
"[coreaudio_begin_reconnect] failed to "
|
|
"create thread, error code: %d",
|
|
ret);
|
|
}
|
|
|
|
static OSStatus
|
|
notification_callback(AudioObjectID id, UInt32 num_addresses,
|
|
const AudioObjectPropertyAddress addresses[], void *data)
|
|
{
|
|
struct coreaudio_data *ca = data;
|
|
|
|
coreaudio_stop(ca);
|
|
coreaudio_uninit(ca);
|
|
|
|
if (addresses[0].mSelector == PROPERTY_DEFAULT_DEVICE)
|
|
ca->retry_time = 300;
|
|
else
|
|
ca->retry_time = 2000;
|
|
|
|
blog(LOG_INFO,
|
|
"coreaudio: device '%s' disconnected or changed. "
|
|
"attempting to reconnect",
|
|
ca->device_name);
|
|
|
|
coreaudio_begin_reconnect(ca);
|
|
|
|
UNUSED_PARAMETER(id);
|
|
UNUSED_PARAMETER(num_addresses);
|
|
|
|
return noErr;
|
|
}
|
|
|
|
static OSStatus add_listener(struct coreaudio_data *ca, UInt32 property)
|
|
{
|
|
AudioObjectPropertyAddress addr = {property,
|
|
kAudioObjectPropertyScopeGlobal,
|
|
kAudioObjectPropertyElementMaster};
|
|
|
|
return AudioObjectAddPropertyListener(ca->device_id, &addr,
|
|
notification_callback, ca);
|
|
}
|
|
|
|
static bool coreaudio_init_hooks(struct coreaudio_data *ca)
|
|
{
|
|
OSStatus stat;
|
|
AURenderCallbackStruct callback_info = {.inputProc = input_callback,
|
|
.inputProcRefCon = ca};
|
|
|
|
stat = add_listener(ca, kAudioDevicePropertyDeviceIsAlive);
|
|
if (!ca_success(stat, ca, "coreaudio_init_hooks",
|
|
"set disconnect callback"))
|
|
return false;
|
|
|
|
stat = add_listener(ca, PROPERTY_FORMATS);
|
|
if (!ca_success(stat, ca, "coreaudio_init_hooks",
|
|
"set format change callback"))
|
|
return false;
|
|
|
|
if (ca->default_device) {
|
|
AudioObjectPropertyAddress addr = {
|
|
PROPERTY_DEFAULT_DEVICE,
|
|
kAudioObjectPropertyScopeGlobal,
|
|
kAudioObjectPropertyElementMaster};
|
|
|
|
stat = AudioObjectAddPropertyListener(kAudioObjectSystemObject,
|
|
&addr,
|
|
notification_callback,
|
|
ca);
|
|
if (!ca_success(stat, ca, "coreaudio_init_hooks",
|
|
"set device change callback"))
|
|
return false;
|
|
}
|
|
|
|
stat = set_property(ca->unit, kAudioOutputUnitProperty_SetInputCallback,
|
|
SCOPE_GLOBAL, 0, &callback_info,
|
|
sizeof(callback_info));
|
|
if (!ca_success(stat, ca, "coreaudio_init_hooks", "set input callback"))
|
|
return false;
|
|
|
|
return true;
|
|
}
|
|
|
|
static void coreaudio_remove_hooks(struct coreaudio_data *ca)
|
|
{
|
|
AURenderCallbackStruct callback_info = {.inputProc = NULL,
|
|
.inputProcRefCon = NULL};
|
|
|
|
AudioObjectPropertyAddress addr = {kAudioDevicePropertyDeviceIsAlive,
|
|
kAudioObjectPropertyScopeGlobal,
|
|
kAudioObjectPropertyElementMaster};
|
|
|
|
AudioObjectRemovePropertyListener(ca->device_id, &addr,
|
|
notification_callback, ca);
|
|
|
|
addr.mSelector = PROPERTY_FORMATS;
|
|
AudioObjectRemovePropertyListener(ca->device_id, &addr,
|
|
notification_callback, ca);
|
|
|
|
if (ca->default_device) {
|
|
addr.mSelector = PROPERTY_DEFAULT_DEVICE;
|
|
AudioObjectRemovePropertyListener(kAudioObjectSystemObject,
|
|
&addr, notification_callback,
|
|
ca);
|
|
}
|
|
|
|
set_property(ca->unit, kAudioOutputUnitProperty_SetInputCallback,
|
|
SCOPE_GLOBAL, 0, &callback_info, sizeof(callback_info));
|
|
}
|
|
|
|
static bool coreaudio_get_device_name(struct coreaudio_data *ca)
|
|
{
|
|
CFStringRef cf_name = NULL;
|
|
UInt32 size = sizeof(CFStringRef);
|
|
char *name = NULL;
|
|
|
|
const AudioObjectPropertyAddress addr = {
|
|
kAudioDevicePropertyDeviceNameCFString,
|
|
kAudioObjectPropertyScopeInput,
|
|
kAudioObjectPropertyElementMaster};
|
|
|
|
OSStatus stat = AudioObjectGetPropertyData(ca->device_id, &addr, 0,
|
|
NULL, &size, &cf_name);
|
|
if (stat != noErr) {
|
|
blog(LOG_WARNING,
|
|
"[coreaudio_get_device_name] failed to "
|
|
"get name: %d",
|
|
(int)stat);
|
|
return false;
|
|
}
|
|
|
|
name = cfstr_copy_cstr(cf_name, kCFStringEncodingUTF8);
|
|
if (!name) {
|
|
blog(LOG_WARNING, "[coreaudio_get_device_name] failed to "
|
|
"convert name to cstr for some reason");
|
|
return false;
|
|
}
|
|
|
|
bfree(ca->device_name);
|
|
ca->device_name = name;
|
|
|
|
if (cf_name)
|
|
CFRelease(cf_name);
|
|
|
|
return true;
|
|
}
|
|
|
|
static bool coreaudio_start(struct coreaudio_data *ca)
|
|
{
|
|
OSStatus stat;
|
|
|
|
if (ca->active)
|
|
return true;
|
|
|
|
stat = AudioOutputUnitStart(ca->unit);
|
|
return ca_success(stat, ca, "coreaudio_start", "start audio");
|
|
}
|
|
|
|
static void coreaudio_stop(struct coreaudio_data *ca)
|
|
{
|
|
OSStatus stat;
|
|
|
|
if (!ca->active)
|
|
return;
|
|
|
|
ca->active = false;
|
|
|
|
stat = AudioOutputUnitStop(ca->unit);
|
|
ca_success(stat, ca, "coreaudio_stop", "stop audio");
|
|
}
|
|
|
|
static bool coreaudio_init_unit(struct coreaudio_data *ca)
|
|
{
|
|
AudioComponentDescription desc = {
|
|
.componentType = kAudioUnitType_Output,
|
|
.componentSubType = kAudioUnitSubType_HALOutput};
|
|
|
|
AudioComponent component = AudioComponentFindNext(NULL, &desc);
|
|
if (!component) {
|
|
ca_warn(ca, "coreaudio_init_unit", "find component failed");
|
|
return false;
|
|
}
|
|
|
|
OSStatus stat = AudioComponentInstanceNew(component, &ca->unit);
|
|
if (!ca_success(stat, ca, "coreaudio_init_unit", "instance unit"))
|
|
return false;
|
|
|
|
ca->au_initialized = true;
|
|
return true;
|
|
}
|
|
|
|
static bool coreaudio_init(struct coreaudio_data *ca)
|
|
{
|
|
OSStatus stat;
|
|
|
|
if (ca->au_initialized)
|
|
return true;
|
|
|
|
if (!find_device_id_by_uid(ca))
|
|
return false;
|
|
if (!coreaudio_get_device_name(ca))
|
|
return false;
|
|
if (!coreaudio_init_unit(ca))
|
|
return false;
|
|
|
|
stat = enable_io(ca, IO_TYPE_INPUT, true);
|
|
if (!ca_success(stat, ca, "coreaudio_init", "enable input io"))
|
|
goto fail;
|
|
|
|
stat = enable_io(ca, IO_TYPE_OUTPUT, false);
|
|
if (!ca_success(stat, ca, "coreaudio_init", "disable output io"))
|
|
goto fail;
|
|
|
|
stat = set_property(ca->unit, kAudioOutputUnitProperty_CurrentDevice,
|
|
SCOPE_GLOBAL, 0, &ca->device_id,
|
|
sizeof(ca->device_id));
|
|
if (!ca_success(stat, ca, "coreaudio_init", "set current device"))
|
|
goto fail;
|
|
|
|
if (!coreaudio_init_format(ca))
|
|
goto fail;
|
|
if (!coreaudio_init_buffer(ca))
|
|
goto fail;
|
|
if (!coreaudio_init_hooks(ca))
|
|
goto fail;
|
|
|
|
stat = AudioUnitInitialize(ca->unit);
|
|
if (!ca_success(stat, ca, "coreaudio_initialize", "initialize"))
|
|
goto fail;
|
|
|
|
if (!coreaudio_start(ca))
|
|
goto fail;
|
|
|
|
blog(LOG_INFO, "coreaudio: device '%s' initialized", ca->device_name);
|
|
return ca->au_initialized;
|
|
|
|
fail:
|
|
coreaudio_uninit(ca);
|
|
return false;
|
|
}
|
|
|
|
static void coreaudio_try_init(struct coreaudio_data *ca)
|
|
{
|
|
if (!coreaudio_init(ca)) {
|
|
blog(LOG_INFO,
|
|
"coreaudio: failed to find device "
|
|
"uid: %s, waiting for connection",
|
|
ca->device_uid);
|
|
|
|
ca->retry_time = 2000;
|
|
|
|
if (ca->no_devices)
|
|
blog(LOG_INFO, "coreaudio: no device found");
|
|
else
|
|
coreaudio_begin_reconnect(ca);
|
|
}
|
|
}
|
|
|
|
static void coreaudio_uninit(struct coreaudio_data *ca)
|
|
{
|
|
if (!ca->au_initialized)
|
|
return;
|
|
|
|
if (ca->unit) {
|
|
coreaudio_stop(ca);
|
|
|
|
OSStatus stat = AudioUnitUninitialize(ca->unit);
|
|
ca_success(stat, ca, "coreaudio_uninit", "uninitialize");
|
|
|
|
coreaudio_remove_hooks(ca);
|
|
|
|
stat = AudioComponentInstanceDispose(ca->unit);
|
|
ca_success(stat, ca, "coreaudio_uninit", "dispose");
|
|
|
|
ca->unit = NULL;
|
|
}
|
|
|
|
ca->au_initialized = false;
|
|
|
|
buf_list_free(ca->buf_list);
|
|
ca->buf_list = NULL;
|
|
}
|
|
|
|
/* ------------------------------------------------------------------------- */
|
|
|
|
static const char *coreaudio_input_getname(void *unused)
|
|
{
|
|
UNUSED_PARAMETER(unused);
|
|
return TEXT_AUDIO_INPUT;
|
|
}
|
|
|
|
static const char *coreaudio_output_getname(void *unused)
|
|
{
|
|
UNUSED_PARAMETER(unused);
|
|
return TEXT_AUDIO_OUTPUT;
|
|
}
|
|
|
|
static void coreaudio_shutdown(struct coreaudio_data *ca)
|
|
{
|
|
if (ca->reconnecting) {
|
|
os_event_signal(ca->exit_event);
|
|
pthread_join(ca->reconnect_thread, NULL);
|
|
os_event_reset(ca->exit_event);
|
|
}
|
|
|
|
coreaudio_uninit(ca);
|
|
|
|
if (ca->unit)
|
|
AudioComponentInstanceDispose(ca->unit);
|
|
}
|
|
|
|
static void coreaudio_destroy(void *data)
|
|
{
|
|
struct coreaudio_data *ca = data;
|
|
|
|
if (ca) {
|
|
coreaudio_shutdown(ca);
|
|
|
|
os_event_destroy(ca->exit_event);
|
|
bfree(ca->device_name);
|
|
bfree(ca->device_uid);
|
|
bfree(ca);
|
|
}
|
|
}
|
|
|
|
static void coreaudio_update(void *data, obs_data_t *settings)
|
|
{
|
|
struct coreaudio_data *ca = data;
|
|
|
|
coreaudio_shutdown(ca);
|
|
|
|
bfree(ca->device_uid);
|
|
ca->device_uid = bstrdup(obs_data_get_string(settings, "device_id"));
|
|
|
|
coreaudio_try_init(ca);
|
|
}
|
|
|
|
static void coreaudio_defaults(obs_data_t *settings)
|
|
{
|
|
obs_data_set_default_string(settings, "device_id", "default");
|
|
}
|
|
|
|
static void *coreaudio_create(obs_data_t *settings, obs_source_t *source,
|
|
bool input)
|
|
{
|
|
struct coreaudio_data *ca = bzalloc(sizeof(struct coreaudio_data));
|
|
|
|
if (os_event_init(&ca->exit_event, OS_EVENT_TYPE_MANUAL) != 0) {
|
|
blog(LOG_ERROR,
|
|
"[coreaudio_create] failed to create "
|
|
"semephore: %d",
|
|
errno);
|
|
bfree(ca);
|
|
return NULL;
|
|
}
|
|
|
|
ca->device_uid = bstrdup(obs_data_get_string(settings, "device_id"));
|
|
ca->source = source;
|
|
ca->input = input;
|
|
|
|
if (!ca->device_uid)
|
|
ca->device_uid = bstrdup("default");
|
|
|
|
coreaudio_try_init(ca);
|
|
return ca;
|
|
}
|
|
|
|
static void *coreaudio_create_input_capture(obs_data_t *settings,
|
|
obs_source_t *source)
|
|
{
|
|
return coreaudio_create(settings, source, true);
|
|
}
|
|
|
|
static void *coreaudio_create_output_capture(obs_data_t *settings,
|
|
obs_source_t *source)
|
|
{
|
|
return coreaudio_create(settings, source, false);
|
|
}
|
|
|
|
static obs_properties_t *coreaudio_properties(bool input)
|
|
{
|
|
obs_properties_t *props = obs_properties_create();
|
|
obs_property_t *property;
|
|
struct device_list devices;
|
|
|
|
memset(&devices, 0, sizeof(struct device_list));
|
|
|
|
property = obs_properties_add_list(props, "device_id", TEXT_DEVICE,
|
|
OBS_COMBO_TYPE_LIST,
|
|
OBS_COMBO_FORMAT_STRING);
|
|
|
|
coreaudio_enum_devices(&devices, input);
|
|
|
|
if (devices.items.num)
|
|
obs_property_list_add_string(property, TEXT_DEVICE_DEFAULT,
|
|
"default");
|
|
|
|
for (size_t i = 0; i < devices.items.num; i++) {
|
|
struct device_item *item = devices.items.array + i;
|
|
obs_property_list_add_string(property, item->name.array,
|
|
item->value.array);
|
|
}
|
|
|
|
device_list_free(&devices);
|
|
return props;
|
|
}
|
|
|
|
static obs_properties_t *coreaudio_input_properties(void *unused)
|
|
{
|
|
UNUSED_PARAMETER(unused);
|
|
|
|
return coreaudio_properties(true);
|
|
}
|
|
|
|
static obs_properties_t *coreaudio_output_properties(void *unused)
|
|
{
|
|
UNUSED_PARAMETER(unused);
|
|
|
|
return coreaudio_properties(false);
|
|
}
|
|
|
|
struct obs_source_info coreaudio_input_capture_info = {
|
|
.id = "coreaudio_input_capture",
|
|
.type = OBS_SOURCE_TYPE_INPUT,
|
|
.output_flags = OBS_SOURCE_AUDIO | OBS_SOURCE_DO_NOT_DUPLICATE,
|
|
.get_name = coreaudio_input_getname,
|
|
.create = coreaudio_create_input_capture,
|
|
.destroy = coreaudio_destroy,
|
|
.update = coreaudio_update,
|
|
.get_defaults = coreaudio_defaults,
|
|
.get_properties = coreaudio_input_properties,
|
|
.icon_type = OBS_ICON_TYPE_AUDIO_INPUT,
|
|
};
|
|
|
|
struct obs_source_info coreaudio_output_capture_info = {
|
|
.id = "coreaudio_output_capture",
|
|
.type = OBS_SOURCE_TYPE_INPUT,
|
|
.output_flags = OBS_SOURCE_AUDIO | OBS_SOURCE_DO_NOT_DUPLICATE |
|
|
OBS_SOURCE_DO_NOT_SELF_MONITOR,
|
|
.get_name = coreaudio_output_getname,
|
|
.create = coreaudio_create_output_capture,
|
|
.destroy = coreaudio_destroy,
|
|
.update = coreaudio_update,
|
|
.get_defaults = coreaudio_defaults,
|
|
.get_properties = coreaudio_output_properties,
|
|
.icon_type = OBS_ICON_TYPE_AUDIO_OUTPUT,
|
|
};
|