Re: libavfilter conflict, technical resolution and maintainership

--===============4870983813479402871==
Content-Type: multipart/signed; micalg=pgp-sha512;
	protocol="application/pgp-signature"; boundary="Y80QIIGobut5bQXI"
Content-Disposition: inline


--Y80QIIGobut5bQXI
Content-Type: text/plain; charset=utf-8
Content-Disposition: inline
Content-Transfer-Encoding: quoted-printable

Hi Nicolas

I wrote this before seeing the "Reply-To: ffmpeg-devel"
But i wanted to send a mail to ffmpeg-devel anyway, so
its this mail

I think we largely agree on the technical direction

On Fri, Jul 31, 2026 at 11:51:19AM +0200, Nicolas George wrote:
> Michael Niedermayer (HE12026-07-28):
> > * Nicolas, are you willing to continue maintaining the libavfilter fram=
ework?
>=20
> Thanks for your confidence. I would like to continue maintaining it, I
> need to be able to do so, so I have two conditions:
>=20
> 1. Technical: That the official tree is the way it needs to be to work
> and evolve properly: only one conversion filter per media type (scale
> for video) and negotiation defaults to not supporting premultiplied.

1a. Regarding having only one conversion filter, I believe there is already
agreement on this with Nikklas, if i understood him correctly last wednsday,
he intends to implement premultiplied alpha processing directly in swscale =
at
some unspecified future point in time.

1b. Yes a filter that doesnt specify support should be assumed not to suppo=
rt
premultiplied.


>=20
> 2. Political: That it is made absolutely clear that a maintainer is 100%
> within their prerogatives to revert on sight commits pushed despite
> objections or requests for more time.

Yes, I intend to write a mail about this to the mailing list. (even better =
reply-to makes this that mail)

There are, however, a few things that must also be clear:
A. "Nothing" outside libavfilter that worked before the original commit tha=
ts
    reverted, can break.
A1 If code has been in the tree for months, reverting it must also not break
   code that has worked with it for months.

B. A maintainer needs to be active. he cannot have the authority to revert
   other people=E2=80=99s commits while contributing only eight commits to =
libavfilter
   in two years. (of course author or committer counts are just numbers tha=
t have
   limited meaning)
   But I think authority and active maintainership need to be connected. Th=
e maintainer
   does not need to be in the top 3 committers, but he should be active.

C. A maintainer should help other contributors resolve technical disagreeme=
nts.
   In the previous premultiplied-alpha discussions, both sides drifted away
   from trying to find an actual solution. The detailed information and test
   image you sent me regarding my pull request are a good example of how su=
ch
   feedback should work.


>=20
> (Maintainer's authority of course subject to global project authority,
> of course, preferably from a trusted and competent long-term leader
> rather than a committee elected by Anton's sock puppets.)
>=20

> I briefly considered asking for excuses, but that would be childish and
> unproductive. But these three should not expect any help from me in the
> foreseeable future.
>=20

> I strongly think the burden of fixing the tree to achieve 1 should be
> Niklas' task, as he broke it and he was paid for it, but I am not making
> it a condition.

Maintainers can reject code and gently push contributors in a particular
direction. They can also do the work themselves.

I am not sure how you expect to get Niklas to work on a particular task.
=46rom my own experience, it is often difficult to convince other people
to work on what one considers important.

Niklas already wants eventual premultiplied-alpha support in swscale,
so that part should not require convincing.

I also know that Niklas is interested in continuing to work on libavfilter
and improving its design, as he has said so himself.

thx

[...]
--=20
Michael     GnuPG fingerprint: 9FF2128B147EF6730BADF133611EC787040B0FAB

Arguing that you don=E2=80=99t care about the right to privacy because you =
have
nothing to hide is no different than saying you don=E2=80=99t care about fr=
ee speech
because you have nothing to say - Edward Snowden

--Y80QIIGobut5bQXI
Content-Type: application/pgp-signature; name="signature.asc"

-----BEGIN PGP SIGNATURE-----

iF0EABEKAB0WIQSf8hKLFH72cwut8TNhHseHBAsPqwUCam+swgAKCRBhHseHBAsP
q3AqAJ9u4GWHAqSuVujWkqxnO6m0T5HCIQCdF/V9zr3PUs+h3wdlA67ex6DaoZ0=
=YgzU
-----END PGP SIGNATURE-----

--Y80QIIGobut5bQXI--

--===============4870983813479402871==
Content-Type: text/plain; charset="us-ascii"
MIME-Version: 1.0
Content-Transfer-Encoding: 7bit
Content-Disposition: inline

_______________________________________________
ffmpeg-devel mailing list -- [email protected]
To unsubscribe send an email to [email protected]

--===============4870983813479402871==--