Merge pull request #4245 from uklotzde/lp1934785-soundsourceffmpeg

lp1934785 SoundSourceFFmpeg: Ignore inaudible samples before start of stream

2 files changed, 50 insertions, 4 deletions

diff --git a/src/sources/readaheadframebuffer.cpp b/src/sources/readaheadframebuffer.cpp
index 407074963c..f2db73a920 100644
--- a/src/sources/readaheadframebuffer.cpp
+++ b/src/sources/readaheadframebuffer.cpp
@@ -3,10 +3,17 @@
 #include "util/logger.h"
 #include "util/sample.h"
 
+// Override or set to `true` to enable verbose debug logging.
 #if !defined(VERBOSE_DEBUG_LOG)
 #define VERBOSE_DEBUG_LOG false
 #endif
 
+// Override or set to `true` to break with a debug assertion
+// if an overlap or gap in the audio stream has been detected.
+#if !defined(DEBUG_ASSERT_ON_DISCONTINUITIES)
+#define DEBUG_ASSERT_ON_DISCONTINUITIES false
+#endif
+
 namespace mixxx {
 
 namespace {
@@ -135,6 +142,9 @@ ReadableSampleFrames ReadAheadFrameBuffer::fillBuffer(
                 << bufferedRange()
                 << "and input buffer"
                 << inputRange;
+#if DEBUG_ASSERT_ON_DISCONTINUITIES
+        DEBUG_ASSERT(!"Unexpected gap");
+#endif
         switch (discontinuityGapMode) {
         case DiscontinuityGapMode::Skip:
             reset(inputRange.start());
@@ -314,6 +324,9 @@ WritableSampleFrames ReadAheadFrameBuffer::consumeAndFillBuffer(
                     << outputRange
                     << "with input buffer"
                     << inputRange;
+#if DEBUG_ASSERT_ON_DISCONTINUITIES
+            DEBUG_ASSERT(!"Unexpected overlap");
+#endif
             switch (discontinuityOverlapMode) {
             case DiscontinuityOverlapMode::Ignore:
                 break;
@@ -345,6 +358,9 @@ WritableSampleFrames ReadAheadFrameBuffer::consumeAndFillBuffer(
                     << bufferedRange()
                     << "with input buffer"
                     << inputRange;
+#if DEBUG_ASSERT_ON_DISCONTINUITIES
+            DEBUG_ASSERT(!"Unexpected overlap");
+#endif
             switch (discontinuityOverlapMode) {
             case DiscontinuityOverlapMode::Ignore:
                 break;
@@ -397,6 +413,9 @@ WritableSampleFrames ReadAheadFrameBuffer::consumeAndFillBuffer(
                         << outputRange
                         << "and input buffer"
                         << inputRange;
+#if DEBUG_ASSERT_ON_DISCONTINUITIES
+                DEBUG_ASSERT(!"Unexpected gap");
+#endif
                 switch (discontinuityGapMode) {
                 case DiscontinuityGapMode::Skip:
                     break;
diff --git a/src/sources/soundsourceffmpeg.cpp b/src/sources/soundsourceffmpeg.cpp
index d76851f585..e1b48196b2 100644
--- a/src/sources/soundsourceffmpeg.cpp
+++ b/src/sources/soundsourceffmpeg.cpp
@@ -60,7 +60,7 @@ constexpr int64_t kavStreamDecoderFrameDelayAAC = 2112;
 // Use 0-based sample frame indexing
 constexpr SINT kMinFrameIndex = 0;
 
-constexpr SINT kSamplesPerMP3Frame = 1152;
+constexpr SINT kMaxSamplesPerMP3Frame = 1152;
 
 const Logger kLogger("SoundSourceFFmpeg");
 
@@ -106,7 +106,7 @@ inline int64_t getStreamStartTime(const AVStream& avStream) {
             // using the default start time.
             // Not all M4A files encode the start_time correctly, e.g.
             // the test file cover-test-itunes-12.7.0-aac.m4a has a valid
-            // start_time of 0. Unfortunately, this special case is cannot
+            // start_time of 0. Unfortunately, this special case cannot be
             // detected and compensated.
             start_time = math_max(kavStreamDefaultStartTime, kavStreamDecoderFrameDelayAAC);
             break;
@@ -135,6 +135,7 @@ inline int64_t getStreamEndTime(const AVStream& avStream) {
 }
 
 inline SINT convertStreamTimeToFrameIndex(const AVStream& avStream, int64_t pts) {
+    DEBUG_ASSERT(pts != AV_NOPTS_VALUE);
     // getStreamStartTime(avStream) -> 1st audible frame at kMinFrameIndex
     return kMinFrameIndex +
             av_rescale_q(
@@ -185,7 +186,7 @@ SINT getStreamSeekPrerollFrameCount(const AVStream& avStream) {
         // slight deviations from the exact signal!
         DEBUG_ASSERT(avStream.codecpar->channels <= 2);
         const SINT mp3SeekPrerollFrameCount =
-                9 * (kSamplesPerMP3Frame / avStream.codecpar->channels);
+                9 * (kMaxSamplesPerMP3Frame / avStream.codecpar->channels);
         return math_max(mp3SeekPrerollFrameCount, defaultSeekPrerollFrameCount);
     }
     case AV_CODEC_ID_AAC:
@@ -1051,12 +1052,33 @@ ReadableSampleFrames SoundSourceFFmpeg::readSampleFramesClamped(
 #if VERBOSE_DEBUG_LOG
                 avTrace("Received decoded frame", *m_pavDecodedFrame);
 #endif
-                DEBUG_ASSERT(m_pavDecodedFrame->pts != AV_NOPTS_VALUE);
+                VERIFY_OR_DEBUG_ASSERT(
+                        (m_pavDecodedFrame->flags &
+                                (AV_FRAME_FLAG_CORRUPT |
+                                        AV_FRAME_FLAG_DISCARD)) == 0) {
+                    av_frame_unref(m_pavDecodedFrame);
+                    continue;
+                }
                 const auto decodedFrameCount = m_pavDecodedFrame->nb_samples;
                 DEBUG_ASSERT(decodedFrameCount > 0);
                 auto streamFrameIndex =
                         convertStreamTimeToFrameIndex(
                                 *m_pavStream, m_pavDecodedFrame->pts);
+                // Only audible samples are counted, i.e. any inaudible aka
+                // "priming" samples are not included in nb_samples!
+                // https://bugs.launchpad.net/mixxx/+bug/1934785
+                if (streamFrameIndex < kMinFrameIndex) {
+#if VERBOSE_DEBUG_LOG
+                    const auto inaudibleFrameCountUntilStartOfStream =
+                            kMinFrameIndex - streamFrameIndex;
+                    kLogger.debug()
+                            << "Skipping"
+                            << inaudibleFrameCountUntilStartOfStream
+                            << "inaudible sample frames before the start of the stream";
+#endif
+                    streamFrameIndex = kMinFrameIndex;
+                }
+                DEBUG_ASSERT(streamFrameIndex >= kMinFrameIndex);
                 decodedFrameRange = IndexRange::forward(
                         streamFrameIndex,
                         decodedFrameCount);
@@ -1203,6 +1225,11 @@ ReadableSampleFrames SoundSourceFFmpeg::readSampleFramesClamped(
             // Housekeeping before next decoding iteration
             av_frame_unref(m_pavDecodedFrame);
             av_frame_unref(m_pavResampledFrame);
+
+            // The first loop condition (see below) should always be true
+            // and has only been added to prevent infinite looping in case
+            // of unexpected result values.
+            DEBUG_ASSERT(avcodec_receive_frame_result == 0);
         } while (avcodec_receive_frame_result == 0 &&
                 m_frameBuffer.isValid());
     }