A not-harmful change to MIDI timing for Windows and Mac...

... and hoping it is positively helpful for Linux.

In AudioIO::MidiTime(), compute one of the terms by different means.

Use PaStreamInfo::outputLatency.

Do not use the difference of PaStreamCallbackTimeInfo::outputBufferDacTime
and PaStreamCallbackTimeInfo::currentTime.

Which debugging shows is very nearly the same value for Windows and Mac.

But we suspect the PaStreamCallbackTimeInfo fields are not correctly reported
on Linux.

src/AudioIO.cpp

@@ -315,6 +315,13 @@ wxArrayLong AudioIO::mCachedSampleRates;
 double AudioIO::mCachedBestRateIn = 0.0;
 double AudioIO::mCachedBestRateOut;
 
+enum {
+   // This is the least positive latency we can
+   // specify to Pm_OpenOutput, 1 ms, which prevents immediate
+   // scheduling of events:
+   MIDI_MINIMAL_LATENCY_MS = 1
+};
+
 #ifdef EXPERIMENTAL_SCRUBBING_SUPPORT
 
 #include "tracks/ui/Scrubbing.h"
@@ -1525,6 +1532,7 @@ bool AudioIO::StartPortAudioStream(double sampleRate,
    int  userData = 24;
    int* lpUserData = (captureFormat_saved == int24Sample) ? &userData : NULL;
 
+   mMidiTimeCorrection = 0;
    mLastPaError = Pa_OpenStream( &mPortStreamV19,
                                  useCapture ? &captureParameters : NULL,
                                  usePlayback ? &playbackParameters : NULL,
@@ -1538,6 +1546,10 @@ bool AudioIO::StartPortAudioStream(double sampleRate,
    Px_SetInputVolume(mPortMixer, oldRecordVolume);
 #endif
    if (mPortStreamV19 != NULL && mLastPaError == paNoError) {
+      auto info = Pa_GetStreamInfo(mPortStreamV19);
+      if (info)
+         mMidiTimeCorrection =
+            info->outputLatency - (MIDI_MINIMAL_LATENCY_MS / 1000.0);
       #ifdef __WXMAC__
       if (mPortMixer) {
          if (Px_SupportsPlaythrough(mPortMixer)) {
@@ -2110,7 +2122,6 @@ bool AudioIO::StartPortMidiStream()
    if (nTracks == 0)
       return false;
 
-   mMidiLatency = 1; // arbitrary, but small
    //printf("StartPortMidiStream: mT0 %g mTime %g\n",
    //       gAudioIO->mT0, gAudioIO->mTime);
 
@@ -2141,7 +2152,7 @@ bool AudioIO::StartPortMidiStream()
                                 0,
                                 &::MidiTime,
                                 NULL,
-                                mMidiLatency);
+                                MIDI_MINIMAL_LATENCY_MS);
    if (mLastPmError == pmNoError) {
       mMidiStreamActive = true;
       mMidiPaused = false;
@@ -3829,7 +3840,7 @@ void AudioIO::OutputEvent()
 
    // 0.0005 is for rounding
    double time = eventTime + 0.0005 -
-                 ((mMidiLatency + mSynthLatency) * 0.001);
+                 (mSynthLatency * 0.001);
 
    time += 1; // MidiTime() has a 1s offset
    // state changes have to go out without delay because the
@@ -4034,17 +4045,25 @@ PmTimestamp AudioIO::MidiTime()
    // note: the extra 0.0005 is for rounding. Round down by casting to
    // unsigned long, then convert to PmTimeStamp (currently signed)
 
-   // PRL:  Bug1714 happened because PaUtil_GetTime() and
-   // mAudioCallbackOutputDacTime could be very widely different, causing
-   // integer overlows.
-   // Portaudio documentation says the origin for
-   // the times passed to audacityAudioCallback is unspecified.
    // See long comments at the top of the file for the explanation of this
    // calculation; we must use PaUtil_GetTime() here and also in the audio
    // callback, to change the origin of times from portaudio, so the diffence of
    // now and then is small, as the long comment assumes.
+
+   // PRL: the time correction is really Midi latency achieved by different
+   // means than specifiying it to Pm_OpenStream.  The use of the accumulated
+   // sample count generated by the audio callback (in AudioTime()) might also
+   // have the virtue of keeping the Midi output synched with audio, even though
+   // pmlinuxalsa.c does not implement any synchronization of its own.
+   // We are now using mMidiTimeCorrection, computed once after opening the
+   // portaudio stream, rather than the difference of dac and current
+   // times reported to the audio callback; because the answer is very nearly
+   // the same on Windows and macOs, doing no harm, whereas on Linux with ALSA,
+   // the dac and current times were not reliable, and that caused irregular
+   // timing of Midi playback because latency was effectively zero.
    auto clockChange = PaUtil_GetTime() - mAudioCallbackClockTime;
-   auto offset = mAudioCallbackOutputDacTime - mAudioCallbackOutputCurrentTime;
+   // auto offset = mAudioCallbackOutputDacTime - mAudioCallbackOutputCurrentTime;
+   auto offset = mMidiTimeCorrection;
    return (PmTimestamp) ((unsigned long) (1000 * (
       AudioTime() + 1.0005 -
       mAudioFramesPerBuffer / mRate +
@@ -4283,8 +4302,13 @@ int audacityAudioCallback(const void *inputBuffer, void *outputBuffer,
 #ifdef EXPERIMENTAL_MIDI_OUT
    /* GSW: Save timeInfo in case MidiPlayback needs it */
    gAudioIO->mAudioCallbackClockTime = PaUtil_GetTime();
+
+   // PRL:  We no longer trust these numbers on Linux / ALSA:
+   /*
    gAudioIO->mAudioCallbackOutputDacTime = timeInfo->outputBufferDacTime;
    gAudioIO->mAudioCallbackOutputCurrentTime = timeInfo->currentTime;
+    */
+
    // printf("in callback, mAudioCallbackOutputDacTime %g\n", gAudioIO->mAudioCallbackOutputDacTime); //DBG
    gAudioIO->mAudioFramesPerBuffer = framesPerBuffer;
    if(gAudioIO->IsPaused())

src/AudioIO.h

@@ -528,20 +528,27 @@ private:
    //   MIDI_PLAYBACK:
    PmStream        *mMidiStream;
    PmError          mLastPmError;
-   /// Latency value for PortMidi
+   PaTime           mMidiTimeCorrection; // seconds
-   long             mMidiLatency;
    /// Latency of MIDI synthesizer
-   long             mSynthLatency;
+   long             mSynthLatency; // ms
 
    // These fields are used to synchronize MIDI with audio:
 
    /// PortAudio's clock time
    volatile double  mAudioCallbackClockTime;
-   /// PortAudio's currentTime -- its origin is unspecified!  So that's why
+
-   /// we also record the above
+   /// PRL: no longer using the next two because they are not reliably
+   /// reported to the audio callback on Linux.
+   /// Compute the approximate difference of them by other means now:
+   /// use PaStreamInfo::outputLatency.
+
+   /*
+   /// PortAudio's currentTime -- its origin is unspecified!
    volatile double  mAudioCallbackOutputCurrentTime;
    /// PortAudio's outTime
    volatile double  mAudioCallbackOutputDacTime;
+    */
+
    /// Number of frames output, including pauses
    volatile long    mNumFrames;
    /// How many frames of zeros were output due to pauses?