Hi Niklas On Tue, Mar 24, 2026 at 04:03:19PM +0100, Niklas Haas via ffmpeg-devel wrote: > On Tue, Mar 24, 2026, at 3:23 PM, Michael Niedermayer via ffmpeg-devel wrote: > > If you mean that checkasm would be hosted on code.ffmpeg.org, then yes, > > that would address the main concern I raised. > > > > If you have a more specific setup in mind, please elaborate. > > > > thx > > Code is now at https://code.ffmpeg.org/FFmpeg/checkasm
Good, that addresses the hosting point.
>
> However, I'm not the best at CI, so I would appreciate if somebody else could
> port the CI checks to run locally on foregejo (if possible). Either way, this
> is still a strict improvement over the status quo, even without tests, so I
> don't think missing local testing should be a blocker anymore?
>
> Regarding the preferred method of integrating this into FFmpeg/FFmpeg, I will
> start a GA vote in 48 hours with the following options:
>
> * A: Plain code dump (as v1 of this PR) of only relevant files
> * B: Git subtree with full history
> * C: Git subtree with squashed history
> * D: Git subtree with squashed history, amended to remove unnecessary files
> * D: Git submodules
I think we should discuss this further before going to a vote.
Voting is useful when we fail to reach consensus. Here, this still looks like a
technical integration question where we may be able to reach broad agreement
first.
Also, many of us are not deeply familiar with the tradeoffs of all these
choices, and 48 hours is not much time for everyone to stop other work and
evaluate them properly.
For what it is worth, I have consistently had bad experiences with submodules
as a user.
Consider a simple case: a user hits a FATE failure in a checkasm-related test.
bisecting that needs to be straightforward. I think we all agree on that.
My current view is that the obvious default choice is:
subtree
+ full history
+ changes intended for checkasm go to code.ffmpeg.org/FFmpeg/checkasm
and then are merged as part of a PR that needs them in FFmpeg
+ changes in the FFmpeg repository's subtree copy of checkasm should be
limited to changes needed specifically for integration into FFmpeg, ideally none
The last two points matter because they keep the maintenance flow clear and
one-directional.
Again, I think this should be discussed before voting, because adding
submodules would be a significant structural change for FFmpeg.
thx
[...]
--
Michael GnuPG fingerprint: 9FF2128B147EF6730BADF133611EC787040B0FAB
Complexity theory is the science of finding the exact solution to an
approximation. Benchmarking OTOH is finding an approximation of the exact
signature.asc
Description: PGP signature
_______________________________________________ ffmpeg-devel mailing list -- [email protected] To unsubscribe send an email to [email protected]
