APT now build with OpenSSL (and sqv), hence sqv is the only thing
left pulling in nettle in the base system.
In Ubuntu I applied the following lines:
override_dh_auto_install:
dh_auto_install -- --features=crypto-openssl --no-default-features
(abusing that cargo is building in dh_auto_install only, and
dh_auto_build is a no-op)
You may have different preferences on how to express the same concept.
Hi, This is a very reasonable request and I strongly support it to be implemented for trixie. Having two libraries less in the base system is a welcome improvement. Uploaders, would you be okay with this being implemented? Thanks, Bastian
Hi I think that this is the correct thing to do. Only reason I haven't done it is a lack of free time. //Alex
you'd think so, yet it was a reasonable choice to choose nettle over openssl. no. certainly not now, as the trixie freeze has started. also please don't just change the crypto library just because, similarily like you also would not switch everything else to nettle, which would also mean one library less. we'll provide some longer explaination eventually.
FWIW, i have more confidence in Nettle as a cryptographic backend than i have in OpenSSL. If the project as a whole prefers to consolidate everything to OpenSSL, i am not going to block. But from my perspective Nettle is better engineered, with a more sensible API (as C APIs go), a more compact footprint, and a narrower scope of work than OpenSSL. Nettle will also be brought back into the end user system with many other common tools, including wget, qemu, etc. If it was just up to me, i'd be inclined to tag this wontfix for debian.
Unfortunatly using Nettle breaks package repos signed with ECDH Brainpool curves as these are not implemented in Nettle. Please reconsider switching to Openssl as a backend for sqv. Best regards Stefan Thöni
can you give examples of such repos?
Unfortunatly I can't, since the repo I'm using is non-public. The reason I chose Brainpool curves for that repo is the smartcards holding the keys don't support the obvious choices like ed25519, gnupg didn't support the NIST curves at that time and rsa4096 signatures take a really long to compute on the smartcards.