ffmpeg: Compensate non monotonicity errors in the audio TSs.

Audio timestamps are passed through by default and when the input
doesnt contain clean timestamps this can lead to non monotonicity
errors. (rounding to a course timebase can cause this too)

Print a warning when the errors in the timestamps are large

Fixes Ticket1167 (regression since timestamps are passed through)

This is a generic workaround that is intended to handle
slightly incorrect input files. It is very possible that some
demuxers contain bugs that lead to wrong timestamps, these demuxers
should of course still be fixed even if this change happens to
hide the issue.
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>

ffmpeg.c

@@ -1132,8 +1132,14 @@ static int encode_audio_frame(AVFormatContext *s, OutputStream *ost,
     if (got_packet) {
         if (pkt.pts != AV_NOPTS_VALUE)
             pkt.pts      = av_rescale_q(pkt.pts,      enc->time_base, ost->st->time_base);
-        if (pkt.dts != AV_NOPTS_VALUE)
+        if (pkt.dts != AV_NOPTS_VALUE) {
+            int64_t max = ost->st->cur_dts + !(s->oformat->flags & AVFMT_TS_NONSTRICT);
             pkt.dts      = av_rescale_q(pkt.dts,      enc->time_base, ost->st->time_base);
+            if (ost->st->cur_dts && ost->st->cur_dts != AV_NOPTS_VALUE &&  max > pkt.dts) {
+                av_log(s, max - pkt.dts > 2 ? AV_LOG_WARNING : AV_LOG_DEBUG, "Audio timestamp %"PRId64" < %"PRId64" invalid, cliping\n", pkt.dts, max);
+                pkt.pts = pkt.dts = max;
+            }
+        }
         if (pkt.duration > 0)
             pkt.duration = av_rescale_q(pkt.duration, enc->time_base, ost->st->time_base);