* update docs for reworked latency API

* rename pa_latency_info to pa_timing_info, since that describes better what it is. Most people will only use pa_stream_get_time() anyway


git-svn-id: file:///home/lennart/svn/public/pulseaudio/trunk@651 fefdeb5f-60dc-0310-8127-8f9354f1896f
This commit is contained in:
Lennart Poettering 2006-04-07 01:29:33 +00:00
parent c0592bb27c
commit b8a729a00f
6 changed files with 142 additions and 94 deletions

View file

@ -80,22 +80,44 @@ typedef enum pa_stream_direction {
/** Some special flags for stream connections. \since 0.6 */
typedef enum pa_stream_flags {
PA_STREAM_START_CORKED = 1, /**< Create the stream corked, requiring an explicit pa_stream_cork() call to uncork it. */
PA_STREAM_INTERPOLATE_LATENCY = 2, /**< Interpolate the latency for
PA_STREAM_INTERPOLATE_TIMING = 2, /**< Interpolate the latency for
* this stream. When enabled,
* you can use
* pa_stream_interpolated_xxx()
* for synchronization. Using
* these functions instead of
* pa_stream_get_latency() has
* the advantage of not
* requiring a whole roundtrip
* for responses. Consider using
* this option when frequently
* requesting latency
* information. This is
* especially useful on long latency
* network connections. */
PA_STREAM_NOT_MONOTONOUS = 4, /**< Don't force the time to run monotonically */
* pa_stream_get_latency() and pa_stream_get_time()
* will try to estimate the
* current record/playback time
* based on the local time that
* passed since the last timing
* info update. In addition
* timing update requests are
* issued periodically
* automatically. Using this
* option has the advantage of
* not requiring a whole
* roundtrip when the current
* playback/recording time is
* needed. Consider using this
* option when requesting
* latency information
* frequently. This is
* especially useful on long
* latency network
* connections. */
PA_STREAM_NOT_MONOTONOUS = 4, /**< Don't force the time to
* increase monotonically. If
* this option is enabled,
* pa_stream_get_time() will not
* necessarily return always
* monotonically increasing time
* values on each call. This may
* confuse applications which
* cannot deal with time going
* 'backwards', but has the
* advantage that bad transport
* latency estimations that
* caused the time to to jump
* ahead can be corrected
* quickly, without the need to
* wait. */
} pa_stream_flags_t;
/** Playback and record buffer metrics */
@ -167,21 +189,23 @@ typedef enum pa_subscription_event_type {
/** Return one if an event type t matches an event mask bitfield */
#define pa_subscription_match_flags(m, t) (!!((m) & (1 << ((t) & PA_SUBSCRIPTION_EVENT_FACILITY_MASK))))
/** A structure for latency info. See pa_stream_get_latency(). The
/** A structure for all kinds of timing information of a stream. See
* pa_stream_update_timing_info() and pa_stream_get_timing_info(). The
* total output latency a sample that is written with
* pa_stream_write() takes to be played may be estimated by
* sink_usec+buffer_usec+transport_usec. The output buffer to which
* buffer_usec relates may be manipulated freely (with
* pa_stream_write()'s seek argument, pa_stream_flush() and friends),
* the buffers sink_usec/source_usec relates to is a first-in
* first-out buffer which cannot be flushed or manipulated in any
* the buffers sink_usec and source_usec relate to are first-in
* first-out (FIFO) buffers which cannot be flushed or manipulated in any
* way. The total input latency a sample that is recorded takes to be
* delivered to the application is:
* source_usec+buffer_usec+transport_usec-sink_usec. (Take care of
* sign issues!) When connected to a monitor source sink_usec contains
* the latency of the owning sink.*/
typedef struct pa_latency_info {
struct timeval timestamp; /**< The time when this latency info was current */
* the latency of the owning sink. The two latency estimations
* described here are implemented in pa_stream_get_latency().*/
typedef struct pa_timing_info {
struct timeval timestamp; /**< The time when this timing info structure was current */
int synchronized_clocks; /**< Non-zero if the local and the
* remote machine have synchronized
* clocks. If synchronized clocks are
@ -198,7 +222,14 @@ typedef struct pa_latency_info {
int playing; /**< Non-zero when the stream is currently playing. Only for playback streams. */
int write_index_corrupt; /**< Non-Zero if the write_index is not up to date because a local write command corrupted it */
int write_index_corrupt; /**< Non-zero if write_index is not
* up-to-date because a local write
* command that corrupted it has been
* issued in the time since this latency
* info was current . Only write
* commands with SEEK_RELATIVE_ON_READ
* and SEEK_RELATIVE_END can corrupt
* write_index. */
int64_t write_index; /**< Current write index into the
* playback buffer in bytes. Think twice before
* using this for seeking purposes: it
@ -213,9 +244,7 @@ typedef struct pa_latency_info {
* want to use it. Consider using
* PA_SEEK_RELATIVE_ON_READ
* instead. \since 0.8 */
uint32_t buffer_length; /* Current buffer length. This is usually identical to write_index-read_index. */
} pa_latency_info;
} pa_timing_info;
/** A structure for the spawn api. This may be used to integrate auto
* spawned daemons into your application. For more information see
@ -236,12 +265,12 @@ typedef struct pa_spawn_api {
* passed to the new process. */
} pa_spawn_api;
/** Seek type \since 0.8*/
/** Seek type for pa_stream_write(). \since 0.8*/
typedef enum pa_seek_mode {
PA_SEEK_RELATIVE = 0, /**< Seek relatively to the write index */
PA_SEEK_ABSOLUTE = 1, /**< Seek relatively to the start of the buffer queue */
PA_SEEK_RELATIVE_ON_READ = 2, /**< Seek relatively to the read index */
PA_SEEK_RELATIVE_END = 3, /**< Seek relatively to the current end of the buffer queue */
PA_SEEK_RELATIVE_ON_READ = 2, /**< Seek relatively to the read index. */
PA_SEEK_RELATIVE_END = 3, /**< Seek relatively to the current end of the buffer queue. */
} pa_seek_mode_t;
PA_C_DECL_END