gstreamer的collectpad是一类特殊的pad,这类pad工作于收集模式,用于管理控制若干个pad组成的pad集合的数据同步处理。大部分的合成器(muxer)均使用collectpad来收集音视频数据,并根据可重载的收集条件判断函数对不同pad之间的数据进行处理(或同步)。
由于collectpad中大部分处理函数均可重载(set_func),因此本文只讨论默认的处理函数。
2. 默认流程:
collectpad的简单流程如下图:
不同的pad工作与不同的线程中,当某一个pad有数据到来时,会对所有pad进行判断,看看是否可以满足收集条件,如果满足收集条件就向对应的element推送数据。如果不满足收集条件,就会将该线程挂起,等待其他线程的数据。
当某个pad处于挂起时,其他pad收到数据后,一样会对收集条件进行判断,如果满足条件,会将所有pad的数据推送至element,同时广播条件变量,唤醒所有挂起中的其他pad(线程)。
简单的函数调用关系如下:
3. 数据结构:
数据结构如下:一个_GstCollectPads中维护了一个_GstCollectData的链表,每个pad对应一个_GstCollectData,其中记录了pad中的数据的时间戳,buffer,已经对应pad的状态(如锁、等待等标志位),GstCollectPadsPrivate中则记录了collectpad中注册的各种事件回调函数,这里的回调函数都有接口可以进行重载。此外,GstCollectPadsPrivate还维护了线程间同步用的锁和条件变量。
/**
* GstCollectPads:
* @data: (element-type GstBase.CollectData): #GList of #GstCollectData managed
* by this #GstCollectPads.
*
* Collectpads object.
*/
struct _GstCollectPads {
/* 基类。 */
GstObject object;
/*< public >*/ /* with LOCK and/or STREAM_LOCK */
/* 所有PAD的集合。 */
/*
* GstCollectData:
* @collect: owner #GstCollectPads
* @pad: #GstPad managed by this data
* @buffer: currently queued buffer.
* @pos: position in the buffer
* @segment: last segment received.
* @dts: the signed version of the DTS converted to running time. To access
* this memeber, use %GST_COLLECT_PADS_DTS macro. (Since 1.6)
*
* Structure used by the collect_pads.
struct _GstCollectData
{
/* with STREAM_LOCK of @collect */
/* 指向回collectpad。 */
GstCollectPads *collect;
GstPad *pad;
GstBuffer *buffer;
guint pos;
GstSegment segment;
/*< private >*/
/* state: bitfield for easier extension;
* eos, flushing, new_segment, waiting */
GstCollectPadsStateFlags state;
GstCollectDataPrivate *priv;
union {
struct {
/*< public >*/
gint64 dts;
/*< private >*/
} abi;
gpointer _gst_reserved[GST_PADDING];
} ABI;
};
*/
GSList *data; /* list of CollectData items */
/*< private >*/
GRecMutex stream_lock; /* used to serialize collection among several streams */
GstCollectPadsPrivate *priv;
gpointer _gst_reserved[GST_PADDING];
};
4. 代码分析:
4.1 主入口函数:
主入口函数gst_collect_pads_chain,不同pad工作于不同线程中。代码分析如下:
/* For each buffer we receive we check if our collected condition is reached
* and if so we call the collected function. When this is done we check if
* data has been unqueued. If data is still queued we wait holding the stream
* lock to make sure no EOS event can happen while we are ready to be
* collected
*/
static GstFlowReturn
gst_collect_pads_chain (GstPad * pad, GstObject * parent, GstBuffer * buffer)
{
GstCollectData *data;
GstCollectPads *pads;
GstFlowReturn ret;
GstBuffer **buffer_p;
guint32 cookie;
GST_DEBUG ("Got buffer for pad %s:%s", GST_DEBUG_PAD_NAME (pad));
/* some magic to get the managing collect_pads */
GST_OBJECT_LOCK (pad);
data = (GstCollectData *) gst_pad_get_element_private (pad);
if (G_UNLIKELY (data == NULL))
goto no_data;
ref_data (data);
GST_OBJECT_UNLOCK (pad);
pads = data->collect;
GST_COLLECT_PADS_STREAM_LOCK (pads);
/* 状态判断。 */
/* if not started, bail out */
if (G_UNLIKELY (!pads->priv->started))
goto not_started;
/* check if this pad is flushing */
if (G_UNLIKELY (GST_COLLECT_PADS_STATE_IS_SET (data,
GST_COLLECT_PADS_STATE_FLUSHING)))
goto flushing;
/* pad was EOS, we can refuse this data */
if (G_UNLIKELY (GST_COLLECT_PADS_STATE_IS_SET (data,
GST_COLLECT_PADS_STATE_EOS)))
goto eos;
/* see if we need to clip */
/* 数据前处理。 */
if (pads->priv->clip_func) {
GstBuffer *outbuf = NULL;
ret =
pads->priv->clip_func (pads, data, buffer, &outbuf,
pads->priv->clip_user_data);
buffer = outbuf;
if (G_UNLIKELY (outbuf == NULL))
goto clipped;
if (G_UNLIKELY (ret == GST_FLOW_EOS))
goto eos;
else if (G_UNLIKELY (ret != GST_FLOW_OK))
goto error;
}
GST_DEBUG_OBJECT (pads, "Queuing buffer %p for pad %s:%s", buffer,
GST_DEBUG_PAD_NAME (pad));
/* One more pad has data queued */
// 如果当前collectpad处于WAITING状态会将queuedpads增加
if (GST_COLLECT_PADS_STATE_IS_SET (data, GST_COLLECT_PADS_STATE_WAITING))
pads->priv->queuedpads++;
buffer_p = &data->buffer;
gst_buffer_replace (buffer_p, buffer);
/* update segment last position if in TIME */
/* 更新当前pad上对应的时间信息,后续用于重新计算等待状态需要用到。 */
if (G_LIKELY (data->segment.format == GST_FORMAT_TIME)) {
GstClockTime timestamp;
timestamp = GST_BUFFER_DTS_OR_PTS (buffer);
if (GST_CLOCK_TIME_IS_VALID (timestamp))
data->segment.position = timestamp;
}
/* While we have data queued on this pad try to collect stuff */
do {
/* Check if our collected condition is matched and call the collected
* function if it is */
/* 主要处理函数,判断收集条件是否满足,后续分析。 */
ret = gst_collect_pads_check_collected (pads);
/* when an error occurs, we want to report this back to the caller ASAP
* without having to block if the buffer was not popped */
/* 数据流处理异常,进入异常处理分支。 */
if (G_UNLIKELY (ret != GST_FLOW_OK))
goto error;
/* data was consumed, we can exit and accept new data */
/* 当buffer在check_collected函数中被消费,会在其中减少引用次数,释放buffer。
* 数据被处理后退出循环,等待下一次buffer到来调用chain函数。 */
if (data->buffer == NULL)
break;
/* 数据未被处理,未满足数据收集条件,本pad对应线程将进行唤醒等待。 */
/* Having the _INIT here means we don't care about any broadcast up to here
* (most of which occur with STREAM_LOCK held, so could not have happened
* anyway). We do care about e.g. a remove initiated broadcast as of this
* point. Putting it here also makes this thread ignores any evt it raised
* itself (as is a usual WAIT semantic).
*/
GST_COLLECT_PADS_EVT_INIT (cookie);
/* pad could be removed and re-added */
unref_data (data);
GST_OBJECT_LOCK (pad);
if (G_UNLIKELY ((data = gst_pad_get_element_private (pad)) == NULL))
goto pad_removed;
ref_data (data);
GST_OBJECT_UNLOCK (pad);
GST_DEBUG_OBJECT (pads, "Pad %s:%s has a buffer queued, waiting",
GST_DEBUG_PAD_NAME (pad));
/* wait to be collected, this must happen from another thread triggered
* by the _chain function of another pad. We release the lock so we
* can get stopped or flushed as well. We can however not get EOS
* because we still hold the STREAM_LOCK.
*/
/* 等待条件变量被唤醒。 */
GST_COLLECT_PADS_STREAM_UNLOCK (pads);
GST_COLLECT_PADS_EVT_WAIT (pads, cookie);
GST_COLLECT_PADS_STREAM_LOCK (pads);
GST_DEBUG_OBJECT (pads, "Pad %s:%s resuming", GST_DEBUG_PAD_NAME (pad));
/* 唤醒后的状态判断。 */
/* after a signal, we could be stopped */
if (G_UNLIKELY (!pads->priv->started))
goto not_started;
/* check if this pad is flushing */
if (G_UNLIKELY (GST_COLLECT_PADS_STATE_IS_SET (data,
GST_COLLECT_PADS_STATE_FLUSHING)))
goto flushing;
}
while (data->buffer != NULL);
unlock_done:
GST_COLLECT_PADS_STREAM_UNLOCK (pads);
/* data is definitely NULL if pad_removed goto was run. */
if (data)
unref_data (data);
if (buffer)
gst_buffer_unref (buffer);
return ret;
/* 异常状态处理。 */
pad_removed:
{
GST_WARNING ("%s got removed from collectpads", GST_OBJECT_NAME (pad));
GST_OBJECT_UNLOCK (pad);
ret = GST_FLOW_NOT_LINKED;
goto unlock_done;
}
/* ERRORS */
no_data:
{
GST_DEBUG ("%s got removed from collectpads", GST_OBJECT_NAME (pad));
GST_OBJECT_UNLOCK (pad);
gst_buffer_unref (buffer);
return GST_FLOW_NOT_LINKED;
}
not_started:
{
GST_DEBUG ("not started");
gst_collect_pads_clear (pads, data);
ret = GST_FLOW_FLUSHING;
goto unlock_done;
}
flushing:
{
GST_DEBUG ("pad %s:%s is flushing", GST_DEBUG_PAD_NAME (pad));
gst_collect_pads_clear (pads, data);
ret = GST_FLOW_FLUSHING;
goto unlock_done;
}
eos:
{
/* we should not post an error for this, just inform upstream that
* we don't expect anything anymore */
GST_DEBUG ("pad %s:%s is eos", GST_DEBUG_PAD_NAME (pad));
ret = GST_FLOW_EOS;
goto unlock_done;
}
clipped:
{
GST_DEBUG ("clipped buffer on pad %s:%s", GST_DEBUG_PAD_NAME (pad));
ret = GST_FLOW_OK;
goto unlock_done;
}
error:
{
/* we print the error, the element should post a reasonable error
* message for fatal errors */
GST_DEBUG ("collect failed, reason %d (%s)", ret, gst_flow_get_name (ret));
gst_collect_pads_clear (pads, data);
goto unlock_done;
}
}
4.2 框架上的收集条件判断
在check函数,首先对collectpads上面的pad状态进行检查,只有当有数据的pads和总的pads数满足一定条件时候,才会执行第二重的收集条件判断。函数为gst_collect_pads_check_collected,代码分析如下:
static GstFlowReturn
gst_collect_pads_check_collected (GstCollectPads * pads)
{
GstFlowReturn flow_ret = GST_FLOW_OK;
GstCollectPadsFunction func;
gpointer user_data;
g_return_val_if_fail (GST_IS_COLLECT_PADS (pads), GST_FLOW_ERROR);
/* 获取回调数据。 */
GST_OBJECT_LOCK (pads);
func = pads->priv->func;
user_data = pads->priv->user_data;
GST_OBJECT_UNLOCK (pads);
g_return_val_if_fail (pads->priv->func != NULL, GST_FLOW_NOT_SUPPORTED);
/* check for new pads, update stats etc.. */
/* 主要是对等待唤醒的pad的cookie进行校验。 */
gst_collect_pads_check_pads (pads);
/* 所有pad都是EOS状态。直接处理剩余的所有数据。 */
if (G_UNLIKELY (pads->priv->eospads == pads->priv->numpads)) {
/* If all our pads are EOS just collect once to let the element
* do its final EOS handling. */
GST_DEBUG_OBJECT (pads, "All active pads (%d) are EOS, calling %s",
pads->priv->numpads, GST_DEBUG_FUNCPTR_NAME (func));
if (G_UNLIKELY (g_atomic_int_compare_and_exchange (&pads->priv->seeking,
TRUE, FALSE))) {
GST_INFO_OBJECT (pads, "finished seeking");
}
do {
flow_ret = func (pads, user_data);
} while (flow_ret == GST_FLOW_OK);
} else {
/* 有pad处于非EOS状态。 */
gboolean collected = FALSE;
/* We call the collected function as long as our condition matches. */
/* 只有满足(有数据的有效pad数 + 无效pad数 >= 总的pad数)时,才可以进入下一步的
* 条件判断,这个判断是框架级别的判断,总是存在,其余重载的判断函数(func)都在这个循环中处理。
* 如果函数不执行,则buffer一定不会被消费,在外层会走入线程挂起等待唤醒的流程。 */
while (((pads->priv->queuedpads + pads->priv->eospads) >=
pads->priv->numpads)) {
GST_DEBUG_OBJECT (pads,
"All active pads (%d + %d >= %d) have data, " "calling %s",
pads->priv->queuedpads, pads->priv->eospads, pads->priv->numpads,
GST_DEBUG_FUNCPTR_NAME (func));
if (G_UNLIKELY (g_atomic_int_compare_and_exchange (&pads->priv->seeking,
TRUE, FALSE))) {
GST_INFO_OBJECT (pads, "finished seeking");
}
/* 具体数据的收集条件判断。 */
flow_ret = func (pads, user_data);
collected = TRUE;
/* 数据处理异常或者已经没有有数据的pad了,中断循环。 */
/* break on error */
if (flow_ret != GST_FLOW_OK)
break;
/* Don't keep looping after telling the element EOS or flushing */
if (pads->priv->queuedpads == 0)
break;
}
if (!collected)
GST_DEBUG_OBJECT (pads, "Not all active pads (%d) have data, continuing",
pads->priv->numpads);
}
return flow_ret;
}
4.3 默认的第二重收集条件判断
第二重收集条件的函数是可以进行重载的,可以使用gst_collect_pads_set_function进行设置,这里只分析默认的函数gst_collect_pads_default_collected。
代码分析如下:
/*
* Default collect callback triggered when #GstCollectPads gathered all data.
*
* Called with STREAM_LOCK.
*/
static GstFlowReturn
gst_collect_pads_default_collected (GstCollectPads * pads, gpointer user_data)
{
GstCollectData *best = NULL;
GstBuffer *buffer;
GstFlowReturn ret = GST_FLOW_OK;
GstCollectPadsBufferFunction func;
gpointer buffer_user_data;
g_return_val_if_fail (GST_IS_COLLECT_PADS (pads), GST_FLOW_ERROR);
/* 获取回调数据。 */
GST_OBJECT_LOCK (pads);
func = pads->priv->buffer_func;
buffer_user_data = pads->priv->buffer_user_data;
GST_OBJECT_UNLOCK (pads);
g_return_val_if_fail (func != NULL, GST_FLOW_NOT_SUPPORTED);
/* Find the oldest pad at all cost */
/* 寻找最合适的pad,并计算最早的数据和时间戳。
* 最后返回gst_collect_pads_recalculate_waiting的返回值,
* TRUE表示从非等待状态变为等待状态。
* 在默认场景下,只有使用了set_wait为FALSE时候才会标记为non-waiting状态。
* 因此如果在默认框架下主动设置了non-waiting状态,需要留意时间比较函数。
* 否则这里会进入一个
* FLOW_OK -> 数据没有POP -> pad_num没有变化 -> gst_collect_pads_check_collected主循环中死循环的问题。 */
if (gst_collect_pads_recalculate_full (pads)) {
/* waiting was switched on,
* so give another thread a chance to deliver a possibly
* older buffer; don't charge on yet with the current oldest */
ret = GST_FLOW_OK;
goto done;
}
best = pads->priv->earliest_data;
/* No data collected means EOS. */
/* 在waiting状态下,但是没有最新的数据包,因此认为这个pad已经进入EOS状态了。无法接收数据。
* 注意,这里设置non-waiting以后并修改了时间比较函数后,其他地方调用默认函数,也会导致一个问题:
* 由于non-waiting增加了queuedpad,因此如果总的有两个pad,且两个pad都设置了non-waiting后,在函数
* gst_collect_pads_check_collected中条件判断总是成立,且queuedpad在non-waiting状态下无法自减,
* 第一次进入时候就会把所有的pad的数据直接处理,及时处理完所有数据后,依旧走到这里进行判断,
* 但是这时候buffer已经为空,导致collectpad认为这个pad的数据已经进入EOS状态,处理异常。 */
if (G_UNLIKELY (best == NULL)) {
ret = func (pads, best, NULL, buffer_user_data);
if (ret == GST_FLOW_OK)
ret = GST_FLOW_EOS;
goto done;
}
/* make sure that the pad we take a buffer from is waiting;
* otherwise popping a buffer will seem not to have happened
* and collectpads can get into a busy loop */
gst_collect_pads_set_waiting (pads, best, TRUE);
/* Send buffer */
/* 使用pop弹出buffer,并将buffer发送给buffer_func进行处理。 */
buffer = gst_collect_pads_pop (pads, best);
ret = func (pads, best, buffer, buffer_user_data);
/* maybe non-waiting was forced to waiting above due to
* newsegment events coming too sparsely,
* so re-check to restore state to avoid hanging/waiting */
gst_collect_pads_recalculate_full (pads);
done:
return ret;
}
注意,这里如果对某些函数进行重载或者设置了非等待状态,有两个潜在的异常流程。
4.3.1 异常流程1:
当使用默认的时间比较函数,且设置了非等待状态的pad有数据到来时,在函数gst_collect_pads_recalculate_waiting,当earliest_data检测到本PAD时,这时候时间戳应该是相等的,但是这时候如果处于非等待状态,无论是否加锁最后都会返回TRUE,这时候gst_collect_pads_default_collected函数中的第一个判断总会直接返回GST_FLOW_OK,但是实际并没有弹出任何buffer,但是gst_collect_pads_check_collected的循环条件并没有改变,导致这个线程会一直在这里循环。如果其他pad没有数据进入,则这个pad会进入死循环。
4.3.2 异常流程2:
当所有的pads都设置了non-waiting状态,则在框架的收集条件判断函数gst_collect_pads_check_collected中的pads数量比较循环总是成立,且所有pads数据弹出时都不会减少当前的queuedpad数,因此当有一个buffer弹出后,会持续弹出所有buffer,当buffer为空时,循环条件依旧成立,在处理空buffer时,认为这个pad已经进入了EOS状态,从而导致异常。异常流程如下图:
4.4 寻找最优的可用buffer和pad
这个函数流程比较简单,就是遍历collectpads中的所有pad,并和earliest_time进行比较,寻找最早的时间点的buffer。
这里涉及到时间比较的函数,这里的默认时间比较函数比较简单,就是单纯判断时间点的大小,相等返回0,第一个时间点大于第二个返回1,小于返回-1。
/**
* gst_collect_pads_find_best_pad:
* @pads: the collectpads to use
* @data: returns the collectdata for earliest data
* @time: returns the earliest available buffertime
*
* Find the oldest/best pad, i.e. pad holding the oldest buffer and
* and return the corresponding #GstCollectData and buffertime.
*
* This function should be called with STREAM_LOCK held,
* such as in the callback.
*/
static void
gst_collect_pads_find_best_pad (GstCollectPads * pads,
GstCollectData ** data, GstClockTime * time)
{
GSList *collected;
GstCollectData *best = NULL;
GstClockTime best_time = GST_CLOCK_TIME_NONE;
g_return_if_fail (data != NULL);
g_return_if_fail (time != NULL);
/* 遍历所有pads,对所有pads中的数据与当前的earliest_time进行比较,
* 寻找时间最靠前的buffer及其对应的pad。 */
for (collected = pads->data; collected; collected = g_slist_next (collected)) {
GstBuffer *buffer;
GstCollectData *data = (GstCollectData *) collected->data;
GstClockTime timestamp;
buffer = gst_collect_pads_peek (pads, data);
/* if we have a buffer check if it is better then the current best one */
if (buffer != NULL) {
timestamp = GST_BUFFER_DTS_OR_PTS (buffer);
gst_buffer_unref (buffer);
if (best == NULL || pads->priv->compare_func (pads, data, timestamp,
best, best_time, pads->priv->compare_user_data) < 0) {
best = data;
best_time = timestamp;
}
}
}
/* set earliest time */
*data = best;
*time = best_time;
GST_DEBUG_OBJECT (pads, "best pad %s, best time %" GST_TIME_FORMAT,
best ? GST_PAD_NAME (((GstCollectData *) best)->pad) : "(nil)",
GST_TIME_ARGS (best_time));
}
3.5 重新计算等待状态函数
函数gst_collect_pads_recalculate_waiting会根据earliest_time和所有pad上的数据进行比较,计算collectpad是否需要重新进入等待状态,返回TRUE表示从非等待状态进入等待状态。
这里如果设置了non-waiting状态,则要小心4.3中出现的异常。
/* General overview:
* - only pad with a buffer can determine earliest_data (and earliest_time)
* - only segment info determines (non-)waiting state
* - ? perhaps use _stream_time for comparison
* (which muxers might have use as well ?)
*/
/*
* Function to recalculate the waiting state of all pads.
*
* Must be called with STREAM_LOCK.
*
* Returns %TRUE if a pad was set to waiting
* (from non-waiting state).
*/
static gboolean
gst_collect_pads_recalculate_waiting (GstCollectPads * pads)
{
GSList *collected;
gboolean result = FALSE;
/* If earliest time is not known, there is nothing to do. */
/* 没有数据可以比较。 */
if (pads->priv->earliest_data == NULL)
return FALSE;
/* 遍历所有pads。 */
for (collected = pads->data; collected; collected = g_slist_next (collected)) {
GstCollectData *data = (GstCollectData *) collected->data;
int cmp_res;
GstClockTime comp_time;
/* check if pad has a segment */
/* 检查本pad上对应的时间信息。 */
if (data->segment.format == GST_FORMAT_UNDEFINED) {
GST_WARNING_OBJECT (pads,
"GstCollectPads has no time segment, assuming 0 based.");
gst_segment_init (&data->segment, GST_FORMAT_TIME);
GST_COLLECT_PADS_STATE_SET (data, GST_COLLECT_PADS_STATE_NEW_SEGMENT);
}
/* check segment format */
if (data->segment.format != GST_FORMAT_TIME) {
GST_ERROR_OBJECT (pads, "GstCollectPads can handle only time segments.");
continue;
}
/* check if the waiting state should be changed */
/* 将earliest_time和当前pad上的时间信息进行比较。
* 当cmp_res为1,表示本pad的时间比earliest_time晚,这时候数据可以消费,不需要等待。
* 将返回FALSE,在函数gst_collect_pads_default_collected执行buffer_func消费buffer。
* 否则表示本pad时间比earliest_time早或者相等,如果这时候是在非等待状态,则要设置成等待状态,
* 同时返回TRUE,并在gst_collect_pads_default_collected不处理buffer,返回GST_FLOW_OK,重新计算best。
* 这里要注意设置了non-waiting后的第一个包,第一个包的时间有可能是相同的,即0:00 == 0:00
*/
comp_time = data->segment.position;
cmp_res = pads->priv->compare_func (pads, data, comp_time,
pads->priv->earliest_data, pads->priv->earliest_time,
pads->priv->compare_user_data);
if (cmp_res > 0)
/* stop waiting */
gst_collect_pads_set_waiting (pads, data, FALSE);
else {
if (!GST_COLLECT_PADS_STATE_IS_SET (data, GST_COLLECT_PADS_STATE_WAITING)) {
/* start waiting */
gst_collect_pads_set_waiting (pads, data, TRUE);
result = TRUE;
}
}
}
return result;
}
4.6 锁和等待状态
collectpad提供了接口gst_collect_pads_set_waiting可以给其他组件设置某个pad为等待或者非等待状态。其设置与锁GST_COLLECT_PADS_STATE_LOCKED标志位有关系。
默认情况下的pad(注意,这里的pad为单独的一个pad,并非整个collectpad,这些状态为单个pad私有,并不是collectpad的属性)均为等待状态,而锁的初始化则根据element调用collect添加pad的函数gst_collect_pads_add_pad的最后一个参数决定。
/**
* gst_collect_pads_set_waiting:
* @pads: the collectpads
* @data: the data to use
* @waiting: boolean indicating whether this pad should operate
* in waiting or non-waiting mode
*
* Sets a pad to waiting or non-waiting mode, if at least this pad
* has not been created with locked waiting state,
* in which case nothing happens.
*
* This function should be called with @pads STREAM_LOCK held, such as
* in the callback.
*
* MT safe.
*/
void
gst_collect_pads_set_waiting (GstCollectPads * pads, GstCollectData * data,
gboolean waiting)
{
g_return_if_fail (pads != NULL);
g_return_if_fail (GST_IS_COLLECT_PADS (pads));
g_return_if_fail (data != NULL);
GST_DEBUG_OBJECT (pads, "Setting pad %s to waiting %d, locked %d",
GST_PAD_NAME (data->pad), waiting,
GST_COLLECT_PADS_STATE_IS_SET (data, GST_COLLECT_PADS_STATE_LOCKED));
/* Do something only on a change and if not locked */
/* 修改等待状态标志位需要在没有上锁的情况下处理,
* 可以通过GST_COLLECT_PADS_STATE_SET (data, GST_COLLECT_PADS_STATE_LOCKED);方式加解锁。
* 如果设置为非等待,则会把对应的queuedpad自增,当所有pad都处于非等待状态,则框架收集条件总是满足。
* 可能存在4.3.2的问题。
*/
if (!GST_COLLECT_PADS_STATE_IS_SET (data, GST_COLLECT_PADS_STATE_LOCKED) &&
(GST_COLLECT_PADS_STATE_IS_SET (data, GST_COLLECT_PADS_STATE_WAITING) !=
! !waiting)) {
/* Set waiting state for this pad */
if (waiting)
GST_COLLECT_PADS_STATE_SET (data, GST_COLLECT_PADS_STATE_WAITING);
else
GST_COLLECT_PADS_STATE_UNSET (data, GST_COLLECT_PADS_STATE_WAITING);
/* Update number of queued pads if needed */
if (!data->buffer &&
!GST_COLLECT_PADS_STATE_IS_SET (data, GST_COLLECT_PADS_STATE_EOS)) {
if (waiting)
pads->priv->queuedpads--;
else
pads->priv->queuedpads++;
}
/* signal waiters because something changed */
GST_COLLECT_PADS_EVT_BROADCAST (pads);
}
}