Exoplayer: SeekMap.getPosition implementations have side effects

[tfgstudios]

See the bolded line of code:
_Mp4Extractor.java_

@Override
  public long getPosition(long timeUs) {
    long earliestSamplePosition = Long.MAX_VALUE;
    for (Mp4Track track : tracks) {
      TrackSampleTable sampleTable = track.sampleTable;
      int sampleIndex = sampleTable.getIndexOfEarlierOrEqualSynchronizationSample(timeUs);
      if (sampleIndex == C.INDEX_UNSET) {
        // Handle the case where the requested time is before the first synchronization sample.
        sampleIndex = sampleTable.getIndexOfLaterOrEqualSynchronizationSample(timeUs);
      }
      **track.sampleIndex = sampleIndex;**

      long offset = sampleTable.offsets[sampleIndex];
      if (offset < earliestSamplePosition) {
        earliestSamplePosition = offset;
      }
    }
    return earliestSamplePosition;
  }

A getter changes the state of the object?

I've spent 2 hours trying to figure out why CacheEvictor.OnSpanTouched(Cache, CacheSpan, CacheSpan)was called after I try to retrieve the current position in bytes (by casting my current Extractor to a seekMap and calling SeekMap .getPosition(long))

Is this by design? If yes, then I hope it'll be rethinked.. and meanwhile, how can I get the current position as the offset in bytes? I can modify the source file, but is there a more elegant way?

Thanks

[ojw28]

That doesn't look right, so I suspect it's a bug. @andrewlewis can confirm.

[tfgstudios]

I just put another method getPosition_NoStateChange() for now with that line removed and everything's working properly.
I almost finished an Evictor that removes the _furthest_ CacheSpans from the current position, rather than the _oldest_. It's way better than the default Lru approach. I can make a pull request for it if you think it'll be useful for others, but for now it only works with Mp4Extractor(in fact, it only needs the SeekMap.getPosition(long) implementation); and it requires someone to tell it the current position (currently, I'm using the exoplayer instance for this)

[ojw28]

FlacOggSeeker and OggSeekMap also have unwanted side effects. We'll be merging fixes for these two (and also for Mp4Extractor) into dev-v2 shortly. The changes will not be backported to dev-v1 because it requires touching quite a lot of files.

[ojw28]

Fixed in dev-v2.