Av::Timestamp overflow

[nijiancong]

During an audio live stream transcoding, after utilizing AudioResampler::pop() for approximately 12 hours, the returned AudioSamples::pts() restarted from 0(

avcpp/src/audioresampler.cpp

Line 189 in 5e167a4

m_nextPts = dst.pts() + Timestamp{dst.samplesCount(), dst.timeBase()};

), prompting me to test the Timestamp addition operation.
After a certain number of iterations, the results obtained from "+=" and "+" operations differ.
"+=" continues to increment, whereas "+" operations result in overflow.

#include <avcpp/timestamp.h>
#include <avcpp/audioresampler.h>
#include <stdint.h>
#include <stdio.h>
#include <unistd.h>

int main(int argc, char* argv[])
{
    int64_t i = 0;
    av::Timestamp t(0, av::Rational{1, 48000});
    while (true) {
        t = t + av::Timestamp{1024, t.timebase()};
        // t += av::Timestamp{1024, t.timebase()};
        if (t.seconds() < 1) {
            printf("[%" PRId64 "]++++t=%0.3lf\n", i, t.seconds());
        }
        if (i % 1000000 == 0) {
            printf("t=%0.3lf\n", t.seconds());
        }
        i++;
    }
    return 0;
}

[h4tr3d]

Sounds like it is feature of the FFmpeg av_add_stable() and its usage in the operator+(a,b) implementation. For the first look, av_add_stable() always expects, that right value always less than left value. In the sample above, overflow will be reduced just by chanding operands order:

t = av::Timestamp{1024, t.timebase()} + t;

Yep, t is greater then 1024/tb: for the some reason of the operator+(a,b) implementattion rhs and lhs values swapped.

Will try to fix.