GnuPG and its alternatives: the state in 2021

The market of the software to replace GnuPG and the OpenPGP infrastructure appears to be quite a topic on itself, the precondition to replace GnuPG being its complexity (which has gone to levels so high the official library to interface with GnuPG, gpgme, is literally a command-line wrapper¹ to GnuPG), it being the de-facto implementation of OpenPGP, and its other problems, as commonly illustrated by this post by Latacora.

Theoretically, the program (or programs, thinking the Unix way) to replace GnuPG as a cryptography software suite should do these two things at the very minimum (GnuPG is capable of a lot more, including keyservers, Web of Trust, etc, but for simplicity let’s talk about the bare minimum):

• encrypt / decrypt data using (a)symmetric cryptography;
• generate digital signatures for data and be able to verify the generated signatures with public keys.

Encryption

One of the most well-known tools to do encryption is age, which, as of July 2021, isn’t out of release candidates just yet. age doesn’t even raise many concerns on its own, but it lacks even the simplest public/private key management capabilities (like key aliases or “use this private key as the default”).

It goes along well with the Unix way (one program per job to do it well), but can you seriously imagine an encryption tool where you provide your full recipient public key every time with a command-line key? Me neither. age seriously assumes you’re fine with entering something like this every time²:

% age -r age1h0ea8x07z33fc5lzkqj3kghup4g6jkzsaj6v86pgngwn7gs8q9es56lpwv \
    -o file.enc file.txt 

age also raises concerns by letting trivially discoverable bugs such as this one slip in the release candidates. This one is for sure easily discoverable, as I ran into it while writing akm in my first hours of trying age, and its presence seriously raised my eyebrows, as it’s been one of the tools recommended by supposedly competent cryptographers at Latacora long before the bug has been fixed.

reop can also do encryption for you, it will be discussed later.

Digital signatures

With digital signatures and the tools to verify them we have quite a number of options, with different problems / concerns surrounding them (remember, we want our tool to be at least somewhat reputable):

• The already mentioned reop is quite good on itself, as to be expected from the people around the OpenBSD project, but appears to be abandoned by its author, and so does its portable version for other operating systems.
• signify’s been made by the same person as reop, Ted Unangst, but is only available under OpenBSD, and Linux folks are supposed to enjoy one of the many available source ports (1, 2, etc). Which of those “sucks less” and will actively be maintained? Hard to tell. The quality of ports of doas, neither of which remember passwords under Linux, doesn’t enthuse much.
• There’s a portable alternative to signify called minisign by the author of libsodium, which is written in C. There’s little concerns about it, aside from the only random commit I noticed which got unrelated changes slipped into it (well, Git has revision history management tools for a reason?).
• There’s also sigtool, which for whatever reason has gone with YAML for its signature format and isn’t really easy to stumble upon, thanks in part to its name colliding with a ClamAV tool of the same name. The number of its users also leaves to be desired (remember, reputability).

Other problems

While out of these described above I have already gone with a pair of age/minisign keys for testing, there’s an inherent problem of having them integrated in my workflow. For instance, while you can sign commit/tag data with Git with GnuPG, there’s no way to sign those with the current set of Git hooks available (although there are hacks to do so using current Git GnuPG support).

Other tools, like NeoMutt, also appear to lack support for arbitrary digital signature / encryption tools unless you’re fine with hacking / gluing them together in ugly ways.

Conclusion

Unless Git and other software projects start abandoning GnuPG, GnuPG and OpenPGP are unlikely to go anywhere, as being tied too deep into software codebases and seeing no perfect simple alternatives just yet.

There’s also a new OpenPGP library in Rust, Sequoia, which solves a number of issues associated with GnuPG (namely the monolith problem), but it doesn’t look like any of the applications using GnuPG integrate with it just yet.

Anyways, you can now encrypt me messages with age (or akm), and verify my signatures with minisign with those keys:

age: age1h0ea8x07z33fc5lzkqj3kghup4g6jkzsaj6v86pgngwn7gs8q9es56lpwv
minisign: RWRltlKLStovfiGdhWNzla+GyANAL9ok1Bg15qCAq8oRPCGN6G4fjLj1