How does splitting / joining work in mkvmerge?

[resonator]

Hey!

I've just posted a thread in the audio forum and I realized I know little to nothing about how splitting in mkvmerge actually works. AFAIK AC3 files (as well as pretty much all other audio formats for that matter) have frames with a certain amount of ms of audio stored in them, am I correct? Now, when mkvmerge splits on video keyframes those audio frames don't necessarily have to coincide with them, or do they? Just what exactly happens there? Is the audio simply cut on the next (audio) frame and then a delay time is applied to compensate for the milliseconds that had to be dropped? If so, then what happens with that one (or maybe more?) audio frame that's been dropped? Is it still there in the first file? Or is gone?