Bug#1141438: RFS: rnnoise/0.2+ds-1 [ITP] -- Recurrent neural network for audio noise reduction (development files)

On 05.07.2026 13:44:35 +0300 you wrote:
> Test are optional but recommended, and you are right, they can be
> considered as trivial.
> Would it be possible to do more without downloading data or adding DFSG
> compliant data ?
> Do you have some suggestion for a better test ?

Upstream testsuite is usually a good starting point for autopkgtests which
should run against the installed package. RNNoise unfortunately does not have
tests with reasonable code coverage. It is another labor-intensive task to
write non-superficial tests covering each aspect of the library, and it is
unrelated to the packaging task. So, we does not have a better option here.

That does not mean that "build-against-librnnoise" is a bad test. It may still
catch simple packaging issues. But it should be marked as superficial according
to the specification [1]. It does not provide significant coverage and can
pass even if the library is not functional. Worth noting is that if a package
does not have autopkgtests or all tests are superficial, then Britney does not
give a discount for migrating packages to testing, and the Release Team will
demand unblock requests when the Hard Freeze begins.

 [1]: https://salsa.debian.org/ci-team/autopkgtest/blob/master/doc/README.package-tests.rst

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

iHUEABYIAB0WIQQRm7llN8yxifaG60cF2qh9JI3wlQUCakqmjQAKCRAF2qh9JI3w
lbaQAQDj2RkxACAitWbXoE0Z88GsB2O2kNK6zwGE1ilYzs2QMgEAjoUZi9A6iBby
2pz/xi0AOxaq4LGR4+YxGIw/3eKl/QI=
=cFZg
-----END PGP SIGNATURE-----