[VDD 2025] Technical Discussions summary

VLC4

libVLC API rework

• no blocker, everyone agrees on what needs to be done (@deyayush6).
• need to sync wrappers (libvlcpp, etc..)

Core Regressions & Changes

Clock core

• clock-start branch (@tguillem and @alexandre-janniaux)
• any sample with audio timestamps (or rate) wrong are glitching, notably for old codecs
  • waiting issue and samples from @fcartegnie

Frame-previous

• @tguillem has some kinda working branch, he should propose it to be merged. AND BE PRAGMATIC

MP4 seeking too much regression

• #29426
• #28966
• @DoctorWho will test on samba / http shares and provide samples

SPU regression issues

• Master issue: #28908
• Subtitle perf scaling improvement MR need to be rebased and merged (@alexandre-janniaux ?)
  • Check the 4K problem with derek file (@jbk need to provide one)

Packaging

Remove ActiveX

• NPAPI has been removed, can we remove also ActiveX ? 👍
• ActriveX is now considered as UNMAINTAINED
• The code will be kept on master, but it will be removed from the CI (@robUx4)

Features

Preparser process (almost merged, no blocker)

• Needs rebase and buildsystem (@gabriel_lt)

whisper/STT

• external process in progress
• after the external process is merged, we can ship what is working right now (simple solution) (@gabriel_lt)
• Even if not perfect, need to be sure core changes don't introduce regressions (@tguillem)

IAMF

• detect IAMF track in mp4 => fallback to avformat
• FFmpeg headers to vlc_iamf.h
• ES calling ES recursively ?
• Master ES with all sub ES in extradata ? Use json serializer from Preparser ? => No core changes
• @tguillem and @ePirat

click to pause

• #29240
  • A group that wants precise pause. (Already handled !7208 (merged) (merged))
  • A group that does not care about the 300ms delay, however does not want fullscreen-toggle lag.
  • A group that uses click to pause so rapidly that fullscreen toggle might annoy them.
• => The window provider handle the click forward whether it has focus or not
• Option to disable
• @deyayush6: could you work on it after libvlc 4 API ?

Unsecure network auth (libdsm)

• Downgrade attack smb2 -> smb
• add a dialog warning YES/no/don't care when logging to smb1. (@tguillem)
• Switch to smb2 by default on all ports (-> @Aza , @fkuehne )

VLC release process

• macOS release built from CI
• Windows signature process
  • the main issue right now is the person signing the binaries and packages doesn't have a lot of time
  • we also need to have the signing script to be more flexible
• MSI signing
  • this is 100% related to the signing script flexibility
  • as soon as the person(s) able to sign have more time, we can fix this pretty quickly (3.0.22 stable).

Windows update process

• If we want to handle updates properly on Windows, we will need at some point: OS (precise) version, GPU information , CPU arch, CPU extensions
• any user profile related info CANNOT be sent to the server (locales, user personal info, etc.)
• any hardware or pure-system related info can be used, AS LONG AS we inform the user (for example during the install/reinstall/update wizard).
• Agreed on doing the update selection process at the server side with the hardware information (so a very simple process on VLC side sending info, and getting the proper update manifest from the server.
• @Garf on it

Gitlab

• Add a tag to differentiate between pending backports and backported Merge Requests
• contrib tarballs caching

Medialibrary preview:

• Preview implementation (cf. https://code.videolan.org/videolan/medialibrary/-/issues/493))
• POC, but not mandatory for VLC 4.0 (Gabriel and Francois)
• The medialib already has a way to manage metadata, but the issue is how to publish the preview picture AND the timestamp/crop metadata externally (ie. having users a way to generate that outside of VLC)
• As adaptive already has some standard for that, we should try to:
  • get some extreme simplified version of the format metadata (manifest?)
  • have a dedicated module/submodule from the adaptive access/demux
  • use the module as an input slave for the player to use the info.
• It doesn't seem a good idea to enable that at real time (during the playback) given the potential hardware limitations of the VLC host.
• Having a way to properly inject this information in a format like MKV (the standard way) could be valuable. #29393
• Whether the preparser/medialib will trigger the previewer in batch will be decided by every application port (desktop, android, iOS, etc.)
• Depending where the medialibrary stores the previews, a new api may be needed for the ports to manage this cache.
  This cache management could be merged with the cache generated by the download API (basically for podcasts)
• @gabriel_lt and @fcartegnie on it

macOS

• Remove growl in 3.0 ? (at least on arm64) 👍 #29432 (closed)
  • Keep it for intel
• Sparkle version in 4.0
  • keep it
• update CI machines (use macOS 26 VMs ?)
  • which Xcode version we use ?
    • For intel: use latest xcode version compatible
    • For arm: use latest code version to have up to date UI
  • we need to select xcode from the job tag (There seems to have some environment variable for the xcode selection)
  • we need to have multiple versions of XCode on the existing CI machines
• @robUx4 and @fkuehne on it

LLM/AI USAGE

• Say no but VLC developers with commits access can use it
• They need to understand and review it
• Copy QEMU AI rule ?
• @tguillem on it

Youtube

• remove in lua 3.0 as it can't work (no js) !7984
• use yt-dlp executable (vlc 4 code up to be published) (@Courmisch on it)
• NewPipe from vlc-android ?
  • VLC android will not embed NewPipe.
  • There should be no direct link between VLC and NewPipe

System testing and CI

• work from @wlalaoui, whether and how we are going to integrate this in the CI
• run it if qt folder is modified
• one requirement is having a Windows CI machine (@robUx4 ?)

How to integrate Coverity CI job

• Closed source: does not belong on VideoLAN
• need to be run manually

RISC-V

• TBA