Irregular fmp4 fragment durations with "progressive" setting

[mcassagnes]

What do you want to do with Hls.js?

Hi, given a HLS VOD with fragmented mp4s (6s fragments, 2s chunks) using byterange requests, I'm wondering why the loaded fragments are of different durations when using the fetch-loader vs xhr-loader.

Given this stream: https://hlsjs.video-dev.org/demo/?src=https%3A%2F%2Fd3bao5x7qmqzgm.cloudfront.net%2Fchunked%2Fmaster.m3u8&demoConfig=eyJlbmFibGVTdHJlYW1pbmciOnRydWUsImF1dG9SZWNvdmVyRXJyb3IiOnRydWUsInN0b3BPblN0YWxsIjpmYWxzZSwiZHVtcGZNUDQiOmZhbHNlLCJsZXZlbENhcHBpbmciOi0xLCJsaW1pdE1ldHJpY3MiOi0xfQ==

Example with xhr-loader showing consistent ~6s fragments for audio and video in the Timeline:

{
  "debug": true,
  "enableWorker": true,
  "backBufferLength": 90,
  "progressive": false
}

Example with fetch-loader showing fragments of varying durations in the Timeline:

{
  "debug": true,
  "enableWorker": true,
  "backBufferLength": 90,
  "progressive": true
}

The reason I am interested in using fetch-loader is to reduce startup time by allowing playback as soon as a chunk is downloaded rather than a full fragment. Is there a less experimental way to achieve this that you recommend?

Thank you

What have you tried so far?

• Different chunk durations (500ms, 1s, 2s)
• A custom fetch-loader, but the magic comes from more than just using fetch it is actually appending the chunks as they come which I have not re-implemented

I wanted to try next:

• fmp4 fragments of equal length as the chunk (6s fragments, 6s chunk with single moof+mdat)

[robwalch]

I'm guessing this is an issue with parseSegmentIndex or getDuration in mp4-tools and/or the progressive pipeline expecting only one moof in each segment such that it can always get the complete duration of a segment on first usable chunk.

HLS.js updates the parsed playlist model to reflect the duration of parsed media. We don't want to change that, but we do want to represent the duration of segments correctly whether or not they are loaded and parsed in "progressive" mode.

Let me know if this specific bug is something you'd be willing to debug further and provide a fix for. If so I can help root-cause to point you in the right direction.