/* GStreamer * Copyright (C) 2011 Mark Nauwelaerts . * Copyright (C) 2011 Nokia Corporation. All rights reserved. * Contact: Stefan Kost * * This library is free software; you can redistribute it and/or * modify it under the terms of the GNU Library General Public * License as published by the Free Software Foundation; either * version 2 of the License, or (at your option) any later version. * * This library is distributed in the hope that it will be useful, * but WITHOUT ANY WARRANTY; without even the implied warranty of * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU * Library General Public License for more details. * * You should have received a copy of the GNU Library General Public * License along with this library; if not, write to the * Free Software Foundation, Inc., 59 Temple Place - Suite 330, * Boston, MA 02111-1307, USA. */ /** * SECTION:gstbaseaudioencoder * @short_description: Base class for audio encoders * @see_also: #GstBaseTransform * * This base class is for audio encoders turning raw audio samples into * encoded audio data. * * GstBaseAudioEncoder and subclass should cooperate as follows. * * * Configuration * * Initially, GstBaseAudioEncoder calls @start when the encoder element * is activated, which allows subclass to perform any global setup. * * * GstBaseAudioEncoder calls @set_format to inform subclass of the format * of input audio data that it is about to receive. Subclass should * setup for encoding and configure various base class context parameters * appropriately, notably those directing desired input data handling. * While unlikely, it might be called more than once, if changing input * parameters require reconfiguration. * * * GstBaseAudioEncoder calls @stop at end of all processing. * * * * As of configuration stage, and throughout processing, GstBaseAudioEncoder * provides a GstBaseAudioEncoderContext that provides required context, * e.g. describing the format of input audio data. * Conversely, subclass can and should configure context to inform * base class of its expectation w.r.t. buffer handling. * * * Data processing * * Base class gathers input sample data (as directed by the context's * frame_samples and frame_max) and provides this to subclass' @handle_frame. * * * If codec processing results in encoded data, subclass should call * @gst_base_audio_encoder_finish_frame to have encoded data pushed * downstream. Alternatively, it might also call to indicate dropped * (non-encoded) samples. * * * Just prior to actually pushing a buffer downstream, * it is passed to @pre_push. * * * During the parsing process GstBaseAudioEncoderClass will handle both * srcpad and sinkpad events. Sink events will be passed to subclass * if @event callback has been provided. * * * * * Shutdown phase * * GstBaseAudioEncoder class calls @stop to inform the subclass that data * parsing will be stopped. * * * * * * Subclass is responsible for providing pad template caps for * source and sink pads. The pads need to be named "sink" and "src". It also * needs to set the fixed caps on srcpad, when the format is ensured. This * is typically when base class calls subclass' @set_format function, though * it might be delayed until calling @gst_base_audio_encoder_finish_frame. * * In summary, above process should have subclass concentrating on * codec data processing while leaving other matters to base class, * such as most notably timestamp handling. While it may exert more control * in this area (see e.g. @pre_push), it is very much not recommended. * * In particular, base class will either favor tracking upstream timestamps * (at the possible expense of jitter) or aim to arrange for a perfect stream of * output timestamps, depending on #GstBaseAudioEncoder:perfect-ts. * However, in the latter case, the input may not be so perfect or ideal, which * is handled as follows. An input timestamp is compared with the expected * timestamp as dictated by input sample stream and if the deviation is less * than #GstBaseAudioEncoder:tolerance, the deviation is discarded. * Otherwise, it is considered a discontuinity and subsequent output timestamp * is resynced to the new position after performing configured discontinuity * processing. In the non-perfect-ts case, an upstream variation exceeding * tolerance only leads to marking DISCONT on subsequent outgoing * (while timestamps are adjusted to upstream regardless of variation). * While DISCONT is also marked in the perfect-ts case, this one optionally * (see #GstBaseAudioEncoder:hard-resync) * performs some additional steps, such as clipping of (early) input samples * or draining all currently remaining input data, depending on the direction * of the discontuinity. * * If perfect timestamps are arranged, it is also possible to request baseclass * (usually set by subclass) to provide additional buffer metadata (in OFFSET * and OFFSET_END) fields according to granule defined semantics currently * needed by oggmux. Specifically, OFFSET is set to granulepos (= sample count * including buffer) and OFFSET_END to corresponding timestamp (as determined * by same sample count and sample rate). * * Things that subclass need to take care of: * * Provide pad templates * * Set source pad caps when appropriate * * * Inform base class of buffer processing needs using context's * frame_samples and frame_bytes. * * * Set user-configurable properties to sane defaults for format and * implementing codec at hand, e.g. those controlling timestamp behaviour * and discontinuity processing. * * * Accept data in @handle_frame and provide encoded results to * @gst_base_audio_encoder_finish_frame. * * * */ #ifdef HAVE_CONFIG_H # include "config.h" #endif #include "gstbaseaudioencoder.h" #include #include #include #include GST_DEBUG_CATEGORY_STATIC (gst_base_audio_encoder_debug); #define GST_CAT_DEFAULT gst_base_audio_encoder_debug #define GST_BASE_AUDIO_ENCODER_GET_PRIVATE(obj) \ (G_TYPE_INSTANCE_GET_PRIVATE ((obj), GST_TYPE_BASE_AUDIO_ENCODER, \ GstBaseAudioEncoderPrivate)) enum { PROP_0, PROP_PERFECT_TS, PROP_GRANULE, PROP_HARD_RESYNC, PROP_TOLERANCE }; #define DEFAULT_PERFECT_TS FALSE #define DEFAULT_GRANULE FALSE #define DEFAULT_HARD_RESYNC FALSE #define DEFAULT_TOLERANCE 40000000 struct _GstBaseAudioEncoderPrivate { /* activation status */ gboolean active; /* input base/first ts as basis for output ts; * kept nearly constant for perfect_ts, * otherwise resyncs to upstream ts */ GstClockTime base_ts; /* corresponding base granulepos */ gint64 base_gp; /* input samples processed and sent downstream so far (w.r.t. base_ts) */ guint64 samples; /* currently collected sample data */ GstAdapter *adapter; /* offset in adapter up to which already supplied to encoder */ gint offset; /* mark outgoing discont */ gboolean discont; /* to guess duration of drained data */ GstClockTime last_duration; /* subclass provided data in processing round */ gboolean got_data; /* subclass gave all it could already */ gboolean drained; /* subclass currently being forcibly drained */ gboolean force; /* output bps estimatation */ /* global in samples seen */ guint64 samples_in; /* global bytes sent out */ guint64 bytes_out; /* context storage */ GstBaseAudioEncoderContext ctx; /* pending serialized sink events, will be sent from finish_frame() */ GList *pending_events; }; static void do_init (GType gtype) { const GInterfaceInfo preset_interface_info = { NULL, /* interface_init */ NULL, /* interface_finalize */ NULL /* interface_data */ }; g_type_add_interface_static (gtype, GST_TYPE_PRESET, &preset_interface_info); } GST_BOILERPLATE_FULL (GstBaseAudioEncoder, gst_base_audio_encoder, GstElement, GST_TYPE_ELEMENT, do_init); static void gst_base_audio_encoder_finalize (GObject * object); static void gst_base_audio_encoder_reset (GstBaseAudioEncoder * enc, gboolean full); static void gst_base_audio_encoder_set_property (GObject * object, guint prop_id, const GValue * value, GParamSpec * pspec); static void gst_base_audio_encoder_get_property (GObject * object, guint prop_id, GValue * value, GParamSpec * pspec); static gboolean gst_base_audio_encoder_sink_activate_push (GstPad * pad, gboolean active); static gboolean gst_base_audio_encoder_sink_event (GstPad * pad, GstEvent * event); static gboolean gst_base_audio_encoder_sink_setcaps (GstPad * pad, GstCaps * caps); static GstFlowReturn gst_base_audio_encoder_chain (GstPad * pad, GstBuffer * buffer); static gboolean gst_base_audio_encoder_src_query (GstPad * pad, GstQuery * query); static gboolean gst_base_audio_encoder_sink_query (GstPad * pad, GstQuery * query); static const GstQueryType *gst_base_audio_encoder_get_query_types (GstPad * pad); static GstCaps *gst_base_audio_encoder_sink_getcaps (GstPad * pad); static void gst_base_audio_encoder_class_init (GstBaseAudioEncoderClass * klass) { GObjectClass *gobject_class; gobject_class = G_OBJECT_CLASS (klass); GST_DEBUG_CATEGORY_INIT (gst_base_audio_encoder_debug, "baseaudioencoder", 0, "baseaudioencoder element"); g_type_class_add_private (klass, sizeof (GstBaseAudioEncoderPrivate)); gobject_class->set_property = gst_base_audio_encoder_set_property; gobject_class->get_property = gst_base_audio_encoder_get_property; gobject_class->finalize = GST_DEBUG_FUNCPTR (gst_base_audio_encoder_finalize); /* properties */ g_object_class_install_property (gobject_class, PROP_PERFECT_TS, g_param_spec_boolean ("perfect-ts", "Perfect Timestamps", "Favour perfect timestamps over tracking upstream timestamps", DEFAULT_PERFECT_TS, G_PARAM_READWRITE | G_PARAM_STATIC_STRINGS)); g_object_class_install_property (gobject_class, PROP_GRANULE, g_param_spec_boolean ("granule", "Granule Marking", "Apply granule semantics to buffer metadata (implies perfect-ts)", DEFAULT_GRANULE, G_PARAM_READABLE | G_PARAM_STATIC_STRINGS)); g_object_class_install_property (gobject_class, PROP_HARD_RESYNC, g_param_spec_boolean ("hard-resync", "Hard Resync", "Perform clipping and sample flushing upon discontinuity", DEFAULT_HARD_RESYNC, G_PARAM_READWRITE | G_PARAM_STATIC_STRINGS)); g_object_class_install_property (gobject_class, PROP_TOLERANCE, g_param_spec_int64 ("tolerance", "Tolerance", "Consider discontinuity if timestamp jitter/imperfection exceeds tolerance (ns)", 0, G_MAXINT64, DEFAULT_TOLERANCE, G_PARAM_READWRITE | G_PARAM_STATIC_STRINGS)); } static void gst_base_audio_encoder_base_init (gpointer g_class) { } static void gst_base_audio_encoder_init (GstBaseAudioEncoder * enc, GstBaseAudioEncoderClass * bclass) { GstPadTemplate *pad_template; GST_DEBUG_OBJECT (enc, "gst_base_audio_encoder_init"); enc->priv = GST_BASE_AUDIO_ENCODER_GET_PRIVATE (enc); /* only push mode supported */ pad_template = gst_element_class_get_pad_template (GST_ELEMENT_CLASS (bclass), "sink"); g_return_if_fail (pad_template != NULL); enc->sinkpad = gst_pad_new_from_template (pad_template, "sink"); gst_pad_set_event_function (enc->sinkpad, GST_DEBUG_FUNCPTR (gst_base_audio_encoder_sink_event)); gst_pad_set_setcaps_function (enc->sinkpad, GST_DEBUG_FUNCPTR (gst_base_audio_encoder_sink_setcaps)); gst_pad_set_getcaps_function (enc->sinkpad, GST_DEBUG_FUNCPTR (gst_base_audio_encoder_sink_getcaps)); gst_pad_set_query_function (enc->sinkpad, GST_DEBUG_FUNCPTR (gst_base_audio_encoder_sink_query)); gst_pad_set_chain_function (enc->sinkpad, GST_DEBUG_FUNCPTR (gst_base_audio_encoder_chain)); gst_pad_set_activatepush_function (enc->sinkpad, GST_DEBUG_FUNCPTR (gst_base_audio_encoder_sink_activate_push)); gst_element_add_pad (GST_ELEMENT (enc), enc->sinkpad); GST_DEBUG_OBJECT (enc, "sinkpad created"); /* and we don't mind upstream traveling stuff that much ... */ pad_template = gst_element_class_get_pad_template (GST_ELEMENT_CLASS (bclass), "src"); g_return_if_fail (pad_template != NULL); enc->srcpad = gst_pad_new_from_template (pad_template, "src"); gst_pad_set_query_function (enc->srcpad, GST_DEBUG_FUNCPTR (gst_base_audio_encoder_src_query)); gst_pad_set_query_type_function (enc->srcpad, GST_DEBUG_FUNCPTR (gst_base_audio_encoder_get_query_types)); gst_pad_use_fixed_caps (enc->srcpad); gst_element_add_pad (GST_ELEMENT (enc), enc->srcpad); GST_DEBUG_OBJECT (enc, "src created"); enc->priv->adapter = gst_adapter_new (); enc->ctx = &enc->priv->ctx; g_static_rec_mutex_init (&enc->stream_lock); /* property default */ enc->perfect_ts = DEFAULT_PERFECT_TS; enc->hard_resync = DEFAULT_HARD_RESYNC; enc->tolerance = DEFAULT_TOLERANCE; /* init state */ gst_base_audio_encoder_reset (enc, TRUE); GST_DEBUG_OBJECT (enc, "init ok"); } static void gst_base_audio_encoder_reset (GstBaseAudioEncoder * enc, gboolean full) { GST_BASE_AUDIO_ENCODER_STREAM_LOCK (enc); if (full) { enc->priv->active = FALSE; enc->priv->samples_in = 0; enc->priv->bytes_out = 0; g_free (enc->ctx->state.channel_pos); memset (enc->ctx, 0, sizeof (enc->ctx)); g_list_foreach (enc->priv->pending_events, (GFunc) gst_event_unref, NULL); g_list_free (enc->priv->pending_events); enc->priv->pending_events = NULL; } gst_segment_init (&enc->segment, GST_FORMAT_TIME); gst_adapter_clear (enc->priv->adapter); enc->priv->got_data = FALSE; enc->priv->drained = TRUE; enc->priv->offset = 0; enc->priv->base_ts = GST_CLOCK_TIME_NONE; enc->priv->base_gp = -1; enc->priv->samples = 0; enc->priv->discont = FALSE; GST_BASE_AUDIO_ENCODER_STREAM_UNLOCK (enc); } static void gst_base_audio_encoder_finalize (GObject * object) { GstBaseAudioEncoder *enc = GST_BASE_AUDIO_ENCODER (object); g_object_unref (enc->priv->adapter); g_static_rec_mutex_free (&enc->stream_lock); G_OBJECT_CLASS (parent_class)->finalize (object); } /** * gst_base_audio_encoder_finish_frame: * @enc: a #GstBaseAudioEncoder * @buffer: encoded data * @samples: number of samples (per channel) represented by encoded data * * Collects encoded data and/or pushes encoded data downstream. * Source pad caps must be set when this is called. Depending on the nature * of the (framing of) the format, subclass can decide whether to push * encoded data directly or to collect various "frames" in a single buffer. * Note that the latter behaviour is recommended whenever the format is allowed, * as it incurs no additional latency and avoids otherwise generating a * a multitude of (small) output buffers. If not explicitly pushed, * any available encoded data is pushed at the end of each processing cycle, * i.e. which encodes as much data as available input data allows. * * If @samples < 0, then best estimate is all samples provided to encoder * (subclass) so far. @buf may be NULL, in which case next number of @samples * are considered discarded, e.g. as a result of discontinuous transmission, * and a discontinuity is marked (note that @buf == NULL => push == TRUE). * * Returns: a #GstFlowReturn that should be escalated to caller (of caller) */ GstFlowReturn gst_base_audio_encoder_finish_frame (GstBaseAudioEncoder * enc, GstBuffer * buf, gint samples) { GstBaseAudioEncoderClass *klass; GstBaseAudioEncoderPrivate *priv; GstBaseAudioEncoderContext *ctx; GstFlowReturn ret = GST_FLOW_OK; klass = GST_BASE_AUDIO_ENCODER_GET_CLASS (enc); priv = enc->priv; ctx = enc->ctx; /* subclass should know what it is producing by now */ g_return_val_if_fail (GST_PAD_CAPS (enc->srcpad) != NULL, GST_FLOW_ERROR); /* subclass should not hand us no data */ g_return_val_if_fail (buf == NULL || GST_BUFFER_SIZE (buf) > 0, GST_FLOW_ERROR); GST_BASE_AUDIO_ENCODER_STREAM_LOCK (enc); GST_LOG_OBJECT (enc, "accepting %d bytes encoded data as %d samples", buf ? GST_BUFFER_SIZE (buf) : -1, samples); /* mark subclass still alive and providing */ priv->got_data = TRUE; if (priv->pending_events) { GList *pending_events, *l; pending_events = priv->pending_events; priv->pending_events = NULL; GST_DEBUG_OBJECT (enc, "Pushing pending events"); for (l = priv->pending_events; l; l = l->next) gst_pad_push_event (enc->srcpad, l->data); g_list_free (pending_events); } /* remove corresponding samples from input */ if (samples < 0) samples = (enc->priv->offset / ctx->state.bpf); if (G_LIKELY (samples)) { /* track upstream ts if so configured */ if (!enc->perfect_ts) { guint64 ts, distance; ts = gst_adapter_prev_timestamp (priv->adapter, &distance); g_assert (distance % ctx->state.bpf == 0); distance /= ctx->state.bpf; GST_LOG_OBJECT (enc, "%" G_GUINT64_FORMAT " samples past prev_ts %" GST_TIME_FORMAT, distance, GST_TIME_ARGS (ts)); GST_LOG_OBJECT (enc, "%" G_GUINT64_FORMAT " samples past base_ts %" GST_TIME_FORMAT, priv->samples, GST_TIME_ARGS (priv->base_ts)); /* when draining adapter might be empty and no ts to offer */ if (GST_CLOCK_TIME_IS_VALID (ts) && ts != priv->base_ts) { GstClockTimeDiff diff; GstClockTime old_ts, next_ts; /* passed into another buffer; * mild check for discontinuity and only mark if so */ next_ts = ts + gst_util_uint64_scale (distance, GST_SECOND, ctx->state.rate); old_ts = priv->base_ts + gst_util_uint64_scale (priv->samples, GST_SECOND, ctx->state.rate); diff = GST_CLOCK_DIFF (next_ts, old_ts); GST_LOG_OBJECT (enc, "ts diff %d ms", (gint) (diff / GST_MSECOND)); /* only mark discontinuity if beyond tolerance */ if (G_UNLIKELY (diff < -enc->tolerance || diff > enc->tolerance)) { GST_DEBUG_OBJECT (enc, "marked discont"); priv->discont = TRUE; } GST_LOG_OBJECT (enc, "new upstream ts %" GST_TIME_FORMAT " at distance %" G_GUINT64_FORMAT, GST_TIME_ARGS (ts), distance); /* re-sync to upstream ts */ priv->base_ts = ts; priv->samples = distance; } } /* advance sample view */ if (G_UNLIKELY (samples * ctx->state.bpf > priv->offset)) { if (G_LIKELY (!priv->force)) { /* no way we can let this pass */ g_assert_not_reached (); /* really no way */ goto overflow; } else { priv->offset = 0; if (samples * ctx->state.bpf >= gst_adapter_available (priv->adapter)) gst_adapter_clear (priv->adapter); else gst_adapter_flush (priv->adapter, samples * ctx->state.bpf); } } else { gst_adapter_flush (priv->adapter, samples * ctx->state.bpf); priv->offset -= samples * ctx->state.bpf; /* avoid subsequent stray prev_ts */ if (G_UNLIKELY (gst_adapter_available (priv->adapter) == 0)) gst_adapter_clear (priv->adapter); } /* sample count advanced below after buffer handling */ } /* collect output */ if (G_LIKELY (buf)) { GST_LOG_OBJECT (enc, "taking %d bytes for output", GST_BUFFER_SIZE (buf)); buf = gst_buffer_make_metadata_writable (buf); /* decorate */ gst_buffer_set_caps (buf, GST_PAD_CAPS (enc->srcpad)); if (G_LIKELY (GST_CLOCK_TIME_IS_VALID (priv->base_ts))) { /* FIXME ? lookahead could lead to weird ts and duration ? * (particularly if not in perfect mode) */ /* mind sample rounding and produce perfect output */ GST_BUFFER_TIMESTAMP (buf) = priv->base_ts + gst_util_uint64_scale (priv->samples - ctx->lookahead, GST_SECOND, ctx->state.rate); GST_DEBUG_OBJECT (enc, "out samples %d", samples); if (G_LIKELY (samples > 0)) { priv->samples += samples; GST_BUFFER_DURATION (buf) = priv->base_ts + gst_util_uint64_scale (priv->samples - ctx->lookahead, GST_SECOND, ctx->state.rate) - GST_BUFFER_TIMESTAMP (buf); priv->last_duration = GST_BUFFER_DURATION (buf); } else { /* duration forecast in case of handling remainder; * the last one is probably like the previous one ... */ GST_BUFFER_DURATION (buf) = priv->last_duration; } if (priv->base_gp >= 0) { /* pamper oggmux */ /* FIXME: in longer run, muxer should take care of this ... */ /* offset_end = granulepos for ogg muxer */ GST_BUFFER_OFFSET_END (buf) = priv->base_gp + priv->samples - enc->ctx->lookahead; /* offset = timestamp corresponding to granulepos for ogg muxer */ GST_BUFFER_OFFSET (buf) = GST_FRAMES_TO_CLOCK_TIME (GST_BUFFER_OFFSET_END (buf), ctx->state.rate); } else { GST_BUFFER_OFFSET (buf) = priv->bytes_out; GST_BUFFER_OFFSET_END (buf) = priv->bytes_out + GST_BUFFER_SIZE (buf); } } priv->bytes_out += GST_BUFFER_SIZE (buf); if (G_UNLIKELY (priv->discont)) { GST_LOG_OBJECT (enc, "marking discont"); GST_BUFFER_FLAG_SET (buf, GST_BUFFER_FLAG_DISCONT); priv->discont = FALSE; } if (klass->pre_push) { /* last chance for subclass to do some dirty stuff */ ret = klass->pre_push (enc, &buf); if (ret != GST_FLOW_OK || !buf) { GST_DEBUG_OBJECT (enc, "subclass returned %s, buf %p", gst_flow_get_name (ret), buf); if (buf) gst_buffer_unref (buf); goto exit; } } GST_LOG_OBJECT (enc, "pushing buffer of size %d with ts %" GST_TIME_FORMAT ", duration %" GST_TIME_FORMAT, GST_BUFFER_SIZE (buf), GST_TIME_ARGS (GST_BUFFER_TIMESTAMP (buf)), GST_TIME_ARGS (GST_BUFFER_DURATION (buf))); ret = gst_pad_push (enc->srcpad, buf); GST_LOG_OBJECT (enc, "buffer pushed: %s", gst_flow_get_name (ret)); } else { /* merely advance samples, most work for that already done above */ priv->samples += samples; } exit: GST_BASE_AUDIO_ENCODER_STREAM_UNLOCK (enc); return ret; /* ERRORS */ overflow: { GST_ELEMENT_ERROR (enc, STREAM, ENCODE, ("received more encoded samples %d than provided %d", samples, priv->offset / ctx->state.bpf), (NULL)); if (buf) gst_buffer_unref (buf); ret = GST_FLOW_ERROR; goto exit; } } /* adapter tracking idea: * - start of adapter corresponds with what has already been encoded * (i.e. really returned by encoder subclass) * - start + offset is what needs to be fed to subclass next */ static GstFlowReturn gst_base_audio_encoder_push_buffers (GstBaseAudioEncoder * enc, gboolean force) { GstBaseAudioEncoderClass *klass; GstBaseAudioEncoderPrivate *priv; GstBaseAudioEncoderContext *ctx; gint av, need; GstBuffer *buf; GstFlowReturn ret = GST_FLOW_OK; klass = GST_BASE_AUDIO_ENCODER_GET_CLASS (enc); g_return_val_if_fail (klass->handle_frame != NULL, GST_FLOW_ERROR); priv = enc->priv; ctx = enc->ctx; while (ret == GST_FLOW_OK) { buf = NULL; av = gst_adapter_available (priv->adapter); g_assert (priv->offset <= av); av -= priv->offset; need = ctx->frame_samples_min > 0 ? ctx->frame_samples_min * ctx->state.bpf : av; GST_LOG_OBJECT (enc, "available: %d, needed: %d, force: %d", av, need, force); if ((need > av) || !av) { if (G_UNLIKELY (force)) { priv->force = TRUE; need = av; } else { break; } } else { priv->force = FALSE; } if (ctx->frame_samples_max > 0) need = MIN (av, ctx->frame_samples_max * ctx->state.bpf); if (ctx->frame_samples_min == ctx->frame_samples_max) { /* if we have some extra metadata, * provide for integer multiple of frames to allow for better granularity * of processing */ if (ctx->frame_samples_min > 0 && need) { if (ctx->frame_max > 1) need = need * MIN ((av / need), ctx->frame_max); else if (ctx->frame_max == 0) need = need * (av / need); } } if (need) { buf = gst_buffer_new (); GST_BUFFER_DATA (buf) = (guint8 *) gst_adapter_peek (priv->adapter, priv->offset + need) + priv->offset; GST_BUFFER_SIZE (buf) = need; } GST_LOG_OBJECT (enc, "providing subclass with %d bytes at offset %d", need, priv->offset); /* mark this already as consumed, * which it should be when subclass gives us data in exchange for samples */ priv->offset += need; priv->samples_in += need / ctx->state.bpf; priv->got_data = FALSE; ret = klass->handle_frame (enc, buf); if (G_LIKELY (buf)) gst_buffer_unref (buf); /* no data to feed, no leftover provided, then bail out */ if (G_UNLIKELY (!buf && !priv->got_data)) { priv->drained = TRUE; GST_LOG_OBJECT (enc, "no more data drained from subclass"); break; } } return ret; } static GstFlowReturn gst_base_audio_encoder_drain (GstBaseAudioEncoder * enc) { if (enc->priv->drained) return GST_FLOW_OK; else return gst_base_audio_encoder_push_buffers (enc, TRUE); } static void gst_base_audio_encoder_set_base_gp (GstBaseAudioEncoder * enc) { GstClockTime ts; if (!enc->granule) return; /* use running time for granule */ /* incoming data is clipped, so a valid input should yield a valid output */ ts = gst_segment_to_running_time (&enc->segment, GST_FORMAT_TIME, enc->priv->base_ts); if (GST_CLOCK_TIME_IS_VALID (ts)) { enc->priv->base_gp = GST_CLOCK_TIME_TO_FRAMES (enc->priv->base_ts, enc->ctx->state.rate); GST_DEBUG_OBJECT (enc, "new base gp %" G_GINT64_FORMAT, enc->priv->base_gp); } else { /* should reasonably have a valid base, * otherwise start at 0 if we did not already start there earlier */ if (enc->priv->base_gp < 0) { enc->priv->base_gp = 0; GST_DEBUG_OBJECT (enc, "new base gp %" G_GINT64_FORMAT, enc->priv->base_gp); } } } static GstFlowReturn gst_base_audio_encoder_chain (GstPad * pad, GstBuffer * buffer) { GstBaseAudioEncoder *enc; GstBaseAudioEncoderPrivate *priv; GstBaseAudioEncoderContext *ctx; GstFlowReturn ret = GST_FLOW_OK; gboolean discont; enc = GST_BASE_AUDIO_ENCODER (GST_OBJECT_PARENT (pad)); priv = enc->priv; ctx = enc->ctx; GST_BASE_AUDIO_ENCODER_STREAM_LOCK (enc); /* should know what is coming by now */ if (!ctx->state.bpf) goto not_negotiated; GST_LOG_OBJECT (enc, "received buffer of size %d with ts %" GST_TIME_FORMAT ", duration %" GST_TIME_FORMAT, GST_BUFFER_SIZE (buffer), GST_TIME_ARGS (GST_BUFFER_TIMESTAMP (buffer)), GST_TIME_ARGS (GST_BUFFER_DURATION (buffer))); /* input shoud be whole number of sample frames */ if (GST_BUFFER_SIZE (buffer) % ctx->state.bpf) goto wrong_buffer; #ifndef GST_DISABLE_GST_DEBUG { GstClockTime duration; GstClockTimeDiff diff; /* verify buffer duration */ duration = gst_util_uint64_scale (GST_BUFFER_SIZE (buffer), GST_SECOND, ctx->state.rate * ctx->state.bpf); diff = GST_CLOCK_DIFF (duration, GST_BUFFER_DURATION (buffer)); if (GST_BUFFER_DURATION (buffer) != GST_CLOCK_TIME_NONE && (diff > GST_SECOND / ctx->state.rate / 2 || diff < -GST_SECOND / ctx->state.rate / 2)) { GST_DEBUG_OBJECT (enc, "incoming buffer had incorrect duration %" GST_TIME_FORMAT ", expected duration %" GST_TIME_FORMAT, GST_TIME_ARGS (GST_BUFFER_DURATION (buffer)), GST_TIME_ARGS (duration)); } } #endif discont = GST_BUFFER_FLAG_IS_SET (buffer, GST_BUFFER_FLAG_DISCONT); if (G_UNLIKELY (discont)) { GST_LOG_OBJECT (buffer, "marked discont"); enc->priv->discont = discont; } /* clip to segment */ /* NOTE: slightly painful linking -laudio only for this one ... */ buffer = gst_audio_buffer_clip (buffer, &enc->segment, ctx->state.rate, ctx->state.bpf); if (G_UNLIKELY (!buffer)) { GST_DEBUG_OBJECT (buffer, "no data after clipping to segment"); goto done; } GST_LOG_OBJECT (enc, "buffer after segment clipping has size %d with ts %" GST_TIME_FORMAT ", duration %" GST_TIME_FORMAT, GST_BUFFER_SIZE (buffer), GST_TIME_ARGS (GST_BUFFER_TIMESTAMP (buffer)), GST_TIME_ARGS (GST_BUFFER_DURATION (buffer))); if (!GST_CLOCK_TIME_IS_VALID (priv->base_ts)) { priv->base_ts = GST_BUFFER_TIMESTAMP (buffer); GST_DEBUG_OBJECT (enc, "new base ts %" GST_TIME_FORMAT, GST_TIME_ARGS (priv->base_ts)); gst_base_audio_encoder_set_base_gp (enc); } /* check for continuity; * checked elsewhere in non-perfect case */ if (enc->perfect_ts) { GstClockTimeDiff diff = 0; GstClockTime next_ts = 0; if (GST_BUFFER_TIMESTAMP_IS_VALID (buffer) && GST_CLOCK_TIME_IS_VALID (priv->base_ts)) { guint64 samples; samples = priv->samples + gst_adapter_available (priv->adapter) / ctx->state.bpf; next_ts = priv->base_ts + gst_util_uint64_scale (samples, GST_SECOND, ctx->state.rate); GST_LOG_OBJECT (enc, "buffer is %" G_GUINT64_FORMAT " samples past base_ts %" GST_TIME_FORMAT ", expected ts %" GST_TIME_FORMAT, samples, GST_TIME_ARGS (priv->base_ts), GST_TIME_ARGS (next_ts)); diff = GST_CLOCK_DIFF (next_ts, GST_BUFFER_TIMESTAMP (buffer)); GST_LOG_OBJECT (enc, "ts diff %d ms", (gint) (diff / GST_MSECOND)); /* if within tolerance, * discard buffer ts and carry on producing perfect stream, * otherwise clip or resync to ts */ if (G_UNLIKELY (diff < -enc->tolerance || diff > enc->tolerance)) { GST_DEBUG_OBJECT (enc, "marked discont"); discont = TRUE; } } /* do some fancy tweaking in hard resync case */ if (discont && enc->hard_resync) { if (diff < 0) { guint64 diff_bytes; GST_WARNING_OBJECT (enc, "Buffer is older than expected ts %" GST_TIME_FORMAT ". Clipping buffer", GST_TIME_ARGS (next_ts)); diff_bytes = GST_CLOCK_TIME_TO_FRAMES (-diff, ctx->state.rate) * ctx->state.bpf; if (diff_bytes >= GST_BUFFER_SIZE (buffer)) { gst_buffer_unref (buffer); goto done; } buffer = gst_buffer_make_metadata_writable (buffer); GST_BUFFER_DATA (buffer) += diff_bytes; GST_BUFFER_SIZE (buffer) -= diff_bytes; GST_BUFFER_TIMESTAMP (buffer) += diff; /* care even less about duration after this */ } else { /* drain stuff prior to resync */ gst_base_audio_encoder_drain (enc); } } /* now re-sync ts */ priv->base_ts += diff; gst_base_audio_encoder_set_base_gp (enc); priv->discont |= discont; } gst_adapter_push (enc->priv->adapter, buffer); /* new stuff, so we can push subclass again */ enc->priv->drained = FALSE; ret = gst_base_audio_encoder_push_buffers (enc, FALSE); done: GST_LOG_OBJECT (enc, "chain leaving"); GST_BASE_AUDIO_ENCODER_STREAM_UNLOCK (enc); return ret; /* ERRORS */ not_negotiated: { GST_ELEMENT_ERROR (enc, CORE, NEGOTIATION, (NULL), ("encoder not initialized")); gst_buffer_unref (buffer); ret = GST_FLOW_NOT_NEGOTIATED; goto done; } wrong_buffer: { GST_ELEMENT_ERROR (enc, STREAM, ENCODE, (NULL), ("buffer size %d not a multiple of %d", GST_BUFFER_SIZE (buffer), ctx->state.bpf)); gst_buffer_unref (buffer); ret = GST_FLOW_ERROR; goto done; } } static gboolean gst_base_audio_encoder_sink_setcaps (GstPad * pad, GstCaps * caps) { GstBaseAudioEncoder *enc; GstBaseAudioEncoderClass *klass; GstBaseAudioEncoderContext *ctx; GstAudioState *state; gboolean res = TRUE, changed = FALSE; enc = GST_BASE_AUDIO_ENCODER (GST_PAD_PARENT (pad)); klass = GST_BASE_AUDIO_ENCODER_GET_CLASS (enc); /* subclass must do something here ... */ g_return_val_if_fail (klass->set_format != NULL, FALSE); ctx = enc->ctx; state = &ctx->state; GST_BASE_AUDIO_ENCODER_STREAM_LOCK (enc); GST_DEBUG_OBJECT (enc, "caps: %" GST_PTR_FORMAT, caps); if (!gst_caps_is_fixed (caps)) goto refuse_caps; /* adjust ts tracking to new sample rate */ if (GST_CLOCK_TIME_IS_VALID (enc->priv->base_ts) && state->rate) { enc->priv->base_ts += GST_FRAMES_TO_CLOCK_TIME (enc->priv->samples, state->rate); enc->priv->samples = 0; } if (!gst_base_audio_parse_caps (caps, state, &changed)) goto refuse_caps; if (changed) { GstClockTime old_min_latency; GstClockTime old_max_latency; /* drain any pending old data stuff */ gst_base_audio_encoder_drain (enc); /* context defaults */ enc->ctx->frame_samples_min = 0; enc->ctx->frame_samples_max = 0; enc->ctx->frame_max = 0; enc->ctx->lookahead = 0; /* element might report latency */ GST_OBJECT_LOCK (enc); old_min_latency = ctx->min_latency; old_max_latency = ctx->max_latency; GST_OBJECT_UNLOCK (enc); if (klass->set_format) res = klass->set_format (enc, state); /* notify if new latency */ GST_OBJECT_LOCK (enc); if ((ctx->min_latency > 0 && ctx->min_latency != old_min_latency) || (ctx->max_latency > 0 && ctx->max_latency != old_max_latency)) { GST_OBJECT_UNLOCK (enc); /* post latency message on the bus */ gst_element_post_message (GST_ELEMENT (enc), gst_message_new_latency (GST_OBJECT (enc))); GST_OBJECT_LOCK (enc); } GST_OBJECT_UNLOCK (enc); } else { GST_DEBUG_OBJECT (enc, "new audio format identical to configured format"); } exit: GST_BASE_AUDIO_ENCODER_STREAM_UNLOCK (enc); return res; /* ERRORS */ refuse_caps: { GST_WARNING_OBJECT (enc, "rejected caps %" GST_PTR_FORMAT, caps); goto exit; } } /** * gst_base_audio_encoder_proxy_getcaps: * @enc: a #GstBaseAudioEncoder * @caps: initial * * Returns caps that express @caps (or sink template caps if @caps == NULL) * restricted to channel/rate combinations supported by downstream elements * (e.g. muxers). * * Returns: a #GstCaps owned by caller */ GstCaps * gst_base_audio_encoder_proxy_getcaps (GstBaseAudioEncoder * enc, GstCaps * caps) { const GstCaps *templ_caps; GstCaps *allowed = NULL; GstCaps *fcaps, *filter_caps; gint i, j; /* we want to be able to communicate to upstream elements like audioconvert * and audioresample any rate/channel restrictions downstream (e.g. muxer * only accepting certain sample rates) */ templ_caps = caps ? caps : gst_pad_get_pad_template_caps (enc->sinkpad); allowed = gst_pad_get_allowed_caps (enc->srcpad); if (!allowed || gst_caps_is_empty (allowed) || gst_caps_is_any (allowed)) { fcaps = gst_caps_copy (templ_caps); goto done; } GST_LOG_OBJECT (enc, "template caps %" GST_PTR_FORMAT, templ_caps); GST_LOG_OBJECT (enc, "allowed caps %" GST_PTR_FORMAT, allowed); filter_caps = gst_caps_new_empty (); for (i = 0; i < gst_caps_get_size (templ_caps); i++) { GQuark q_name; q_name = gst_structure_get_name_id (gst_caps_get_structure (templ_caps, i)); /* pick rate + channel fields from allowed caps */ for (j = 0; j < gst_caps_get_size (allowed); j++) { const GstStructure *allowed_s = gst_caps_get_structure (allowed, j); const GValue *val; GstStructure *s; s = gst_structure_id_empty_new (q_name); if ((val = gst_structure_get_value (allowed_s, "rate"))) gst_structure_set_value (s, "rate", val); if ((val = gst_structure_get_value (allowed_s, "channels"))) gst_structure_set_value (s, "channels", val); gst_caps_merge_structure (filter_caps, s); } } fcaps = gst_caps_intersect (filter_caps, templ_caps); gst_caps_unref (filter_caps); done: gst_caps_replace (&allowed, NULL); GST_LOG_OBJECT (enc, "proxy caps %" GST_PTR_FORMAT, fcaps); return fcaps; } static GstCaps * gst_base_audio_encoder_sink_getcaps (GstPad * pad) { GstBaseAudioEncoder *enc; GstBaseAudioEncoderClass *klass; GstCaps *caps; enc = GST_BASE_AUDIO_ENCODER (gst_pad_get_parent (pad)); klass = GST_BASE_AUDIO_ENCODER_GET_CLASS (enc); g_assert (pad == enc->sinkpad); if (klass->getcaps) caps = klass->getcaps (enc); else caps = gst_base_audio_encoder_proxy_getcaps (enc, NULL); gst_object_unref (enc); GST_LOG_OBJECT (enc, "returning caps %" GST_PTR_FORMAT, caps); return caps; } static gboolean gst_base_audio_encoder_sink_eventfunc (GstBaseAudioEncoder * enc, GstEvent * event) { GstBaseAudioEncoderClass *klass; gboolean handled = FALSE; klass = GST_BASE_AUDIO_ENCODER_GET_CLASS (enc); switch (GST_EVENT_TYPE (event)) { case GST_EVENT_NEWSEGMENT: { GstFormat format; gdouble rate, arate; gint64 start, stop, time; gboolean update; gst_event_parse_new_segment_full (event, &update, &rate, &arate, &format, &start, &stop, &time); if (format == GST_FORMAT_TIME) { GST_DEBUG_OBJECT (enc, "received TIME NEW_SEGMENT %" GST_TIME_FORMAT " -- %" GST_TIME_FORMAT ", time %" GST_TIME_FORMAT ", rate %g, applied_rate %g", GST_TIME_ARGS (start), GST_TIME_ARGS (stop), GST_TIME_ARGS (time), rate, arate); } else { GST_DEBUG_OBJECT (enc, "received NEW_SEGMENT %" G_GINT64_FORMAT " -- %" G_GINT64_FORMAT ", time %" G_GINT64_FORMAT ", rate %g, applied_rate %g", start, stop, time, rate, arate); GST_DEBUG_OBJECT (enc, "unsupported format; ignoring"); break; } GST_BASE_AUDIO_ENCODER_STREAM_LOCK (enc); /* finish current segment */ gst_base_audio_encoder_drain (enc); /* reset partially for new segment */ gst_base_audio_encoder_reset (enc, FALSE); /* and follow along with segment */ gst_segment_set_newsegment_full (&enc->segment, update, rate, arate, format, start, stop, time); GST_BASE_AUDIO_ENCODER_STREAM_UNLOCK (enc); break; } case GST_EVENT_FLUSH_START: break; case GST_EVENT_FLUSH_STOP: GST_BASE_AUDIO_ENCODER_STREAM_LOCK (enc); /* discard any pending stuff */ /* TODO route through drain ?? */ if (!enc->priv->drained && klass->flush) klass->flush (enc); /* and get (re)set for the sequel */ gst_base_audio_encoder_reset (enc, FALSE); g_list_foreach (enc->priv->pending_events, (GFunc) gst_event_unref, NULL); g_list_free (enc->priv->pending_events); enc->priv->pending_events = NULL; GST_BASE_AUDIO_ENCODER_STREAM_UNLOCK (enc); break; case GST_EVENT_EOS: GST_BASE_AUDIO_ENCODER_STREAM_LOCK (enc); gst_base_audio_encoder_drain (enc); GST_BASE_AUDIO_ENCODER_STREAM_UNLOCK (enc); break; default: break; } return handled; } static gboolean gst_base_audio_encoder_sink_event (GstPad * pad, GstEvent * event) { GstBaseAudioEncoder *enc; GstBaseAudioEncoderClass *klass; gboolean handled = FALSE; gboolean ret = TRUE; enc = GST_BASE_AUDIO_ENCODER (gst_pad_get_parent (pad)); klass = GST_BASE_AUDIO_ENCODER_GET_CLASS (enc); GST_DEBUG_OBJECT (enc, "received event %d, %s", GST_EVENT_TYPE (event), GST_EVENT_TYPE_NAME (event)); if (klass->event) handled = klass->event (enc, event); if (!handled) handled = gst_base_audio_encoder_sink_eventfunc (enc, event); if (!handled) { /* Forward non-serialized events and EOS/FLUSH_STOP immediately. * For EOS this is required because no buffer or serialized event * will come after EOS and nothing could trigger another * _finish_frame() call. * * For FLUSH_STOP this is required because it is expected * to be forwarded immediately and no buffers are queued anyway. */ if (!GST_EVENT_IS_SERIALIZED (event) || GST_EVENT_TYPE (event) == GST_EVENT_EOS || GST_EVENT_TYPE (event) == GST_EVENT_FLUSH_STOP) { ret = gst_pad_event_default (pad, event); } else { GST_BASE_AUDIO_ENCODER_STREAM_LOCK (enc); enc->priv->pending_events = g_list_append (enc->priv->pending_events, event); GST_BASE_AUDIO_ENCODER_STREAM_UNLOCK (enc); ret = TRUE; } } GST_DEBUG_OBJECT (enc, "event handled"); gst_object_unref (enc); return ret; } static gboolean gst_base_audio_encoder_sink_query (GstPad * pad, GstQuery * query) { gboolean res = TRUE; GstBaseAudioEncoder *enc; enc = GST_BASE_AUDIO_ENCODER (gst_pad_get_parent (pad)); switch (GST_QUERY_TYPE (query)) { case GST_QUERY_FORMATS: { gst_query_set_formats (query, 3, GST_FORMAT_TIME, GST_FORMAT_BYTES, GST_FORMAT_DEFAULT); res = TRUE; break; } case GST_QUERY_CONVERT: { GstFormat src_fmt, dest_fmt; gint64 src_val, dest_val; gst_query_parse_convert (query, &src_fmt, &src_val, &dest_fmt, &dest_val); if (!(res = gst_base_audio_raw_audio_convert (&enc->ctx->state, src_fmt, src_val, &dest_fmt, &dest_val))) goto error; gst_query_set_convert (query, src_fmt, src_val, dest_fmt, dest_val); break; } default: res = gst_pad_query_default (pad, query); break; } error: gst_object_unref (enc); return res; } static const GstQueryType * gst_base_audio_encoder_get_query_types (GstPad * pad) { static const GstQueryType gst_base_audio_encoder_src_query_types[] = { GST_QUERY_POSITION, GST_QUERY_DURATION, GST_QUERY_CONVERT, GST_QUERY_LATENCY, 0 }; return gst_base_audio_encoder_src_query_types; } /* FIXME ? are any of these queries (other than latency) an encoder's business * also, the conversion stuff might seem to make sense, but seems to not mind * segment stuff etc at all * Supposedly that's backward compatibility ... */ static gboolean gst_base_audio_encoder_src_query (GstPad * pad, GstQuery * query) { GstBaseAudioEncoder *enc; GstPad *peerpad; gboolean res = FALSE; enc = GST_BASE_AUDIO_ENCODER (GST_PAD_PARENT (pad)); peerpad = gst_pad_get_peer (GST_PAD (enc->sinkpad)); GST_LOG_OBJECT (enc, "handling query: %" GST_PTR_FORMAT, query); switch (GST_QUERY_TYPE (query)) { case GST_QUERY_POSITION: { GstFormat fmt, req_fmt; gint64 pos, val; if ((res = gst_pad_peer_query (enc->sinkpad, query))) { GST_LOG_OBJECT (enc, "returning peer response"); break; } if (!peerpad) { GST_LOG_OBJECT (enc, "no peer"); break; } gst_query_parse_position (query, &req_fmt, NULL); fmt = GST_FORMAT_TIME; if (!(res = gst_pad_query_position (peerpad, &fmt, &pos))) break; if ((res = gst_pad_query_convert (peerpad, fmt, pos, &req_fmt, &val))) { gst_query_set_position (query, req_fmt, val); } break; } case GST_QUERY_DURATION: { GstFormat fmt, req_fmt; gint64 dur, val; if ((res = gst_pad_peer_query (enc->sinkpad, query))) { GST_LOG_OBJECT (enc, "returning peer response"); break; } if (!peerpad) { GST_LOG_OBJECT (enc, "no peer"); break; } gst_query_parse_duration (query, &req_fmt, NULL); fmt = GST_FORMAT_TIME; if (!(res = gst_pad_query_duration (peerpad, &fmt, &dur))) break; if ((res = gst_pad_query_convert (peerpad, fmt, dur, &req_fmt, &val))) { gst_query_set_duration (query, req_fmt, val); } break; } case GST_QUERY_FORMATS: { gst_query_set_formats (query, 2, GST_FORMAT_TIME, GST_FORMAT_BYTES); res = TRUE; break; } case GST_QUERY_CONVERT: { GstFormat src_fmt, dest_fmt; gint64 src_val, dest_val; gst_query_parse_convert (query, &src_fmt, &src_val, &dest_fmt, &dest_val); if (!(res = gst_base_audio_encoded_audio_convert (&enc->ctx->state, enc->priv->bytes_out, enc->priv->samples_in, src_fmt, src_val, &dest_fmt, &dest_val))) break; gst_query_set_convert (query, src_fmt, src_val, dest_fmt, dest_val); break; } case GST_QUERY_LATENCY: { if ((res = gst_pad_peer_query (enc->sinkpad, query))) { gboolean live; GstClockTime min_latency, max_latency; gst_query_parse_latency (query, &live, &min_latency, &max_latency); GST_DEBUG_OBJECT (enc, "Peer latency: live %d, min %" GST_TIME_FORMAT " max %" GST_TIME_FORMAT, live, GST_TIME_ARGS (min_latency), GST_TIME_ARGS (max_latency)); GST_OBJECT_LOCK (enc); /* add our latency */ if (min_latency != -1) min_latency += enc->ctx->min_latency; if (max_latency != -1) max_latency += enc->ctx->max_latency; GST_OBJECT_UNLOCK (enc); gst_query_set_latency (query, live, min_latency, max_latency); } break; } default: res = gst_pad_query_default (pad, query); break; } gst_object_unref (peerpad); return res; } static void gst_base_audio_encoder_set_property (GObject * object, guint prop_id, const GValue * value, GParamSpec * pspec) { GstBaseAudioEncoder *enc; enc = GST_BASE_AUDIO_ENCODER (object); switch (prop_id) { case PROP_PERFECT_TS: if (enc->granule && !g_value_get_boolean (value)) GST_WARNING_OBJECT (enc, "perfect-ts can not be set FALSE"); else enc->perfect_ts = g_value_get_boolean (value); break; case PROP_HARD_RESYNC: enc->hard_resync = g_value_get_boolean (value); break; case PROP_TOLERANCE: enc->tolerance = g_value_get_int64 (value); break; default: G_OBJECT_WARN_INVALID_PROPERTY_ID (object, prop_id, pspec); break; } } static void gst_base_audio_encoder_get_property (GObject * object, guint prop_id, GValue * value, GParamSpec * pspec) { GstBaseAudioEncoder *enc; enc = GST_BASE_AUDIO_ENCODER (object); switch (prop_id) { case PROP_PERFECT_TS: g_value_set_boolean (value, enc->perfect_ts); break; case PROP_GRANULE: g_value_set_boolean (value, enc->granule); break; case PROP_HARD_RESYNC: g_value_set_boolean (value, enc->hard_resync); break; case PROP_TOLERANCE: g_value_set_int64 (value, enc->tolerance); break; default: G_OBJECT_WARN_INVALID_PROPERTY_ID (object, prop_id, pspec); break; } } static gboolean gst_base_audio_encoder_activate (GstBaseAudioEncoder * enc, gboolean active) { GstBaseAudioEncoderClass *klass; gboolean result = FALSE; klass = GST_BASE_AUDIO_ENCODER_GET_CLASS (enc); g_return_val_if_fail (!enc->granule || enc->perfect_ts, FALSE); GST_DEBUG_OBJECT (enc, "activate %d", active); if (active) { if (!enc->priv->active && klass->start) result = klass->start (enc); } else { /* We must make sure streaming has finished before resetting things * and calling the ::stop vfunc */ GST_PAD_STREAM_LOCK (enc->sinkpad); GST_PAD_STREAM_UNLOCK (enc->sinkpad); if (enc->priv->active && klass->stop) result = klass->stop (enc); /* clean up */ gst_base_audio_encoder_reset (enc, TRUE); } GST_DEBUG_OBJECT (enc, "activate return: %d", result); return result; } static gboolean gst_base_audio_encoder_sink_activate_push (GstPad * pad, gboolean active) { gboolean result = TRUE; GstBaseAudioEncoder *enc; enc = GST_BASE_AUDIO_ENCODER (gst_pad_get_parent (pad)); GST_DEBUG_OBJECT (enc, "sink activate push %d", active); result = gst_base_audio_encoder_activate (enc, active); if (result) enc->priv->active = active; GST_DEBUG_OBJECT (enc, "sink activate push return: %d", result); gst_object_unref (enc); return result; }