COBOL.now VS Monocypher

Oof, I remember being in school and they taught us COBOL for 3 semesters (this was late 1990s...) I had to go look what Github has that's COBOL and this is what I found.

Then don’t forget https://monocypher.org as well. Bigger than libhydrogen but still small enough for many targets, faster across the board, and compatible with libsodium. If you can spare a couple more KB of flash in your microcontroller, you can get very good performance on the device and scale like crazy with top performance on the server side.

Compilers don’t find all the optimisations. Last time I saw this was when someone noticed that my code was 5% slower than the reference implementation. This patch fixed it.

If encryption performance is not that important (especially on the client side, which I expect won't use too much bandwidth), but you value minimising dependencies, consider using Monocypher instead of libsodium. Monocypher is a single-file library that has absolutely zero dependency (not even libc). The price to pay for that is (i) right now it's slower than libsodium, and (ii) it doesn't provide an RNG, you'll have to call your OS's RNG manually.

I'm not yelling at you for your choice. See here for how hypocritical it would be of me.

We kind of are though. And in some circles this pendulum has even swung too far the other way, with people scolding me for writing and distributing a cryptographic library. Because it's dangerous, and users expect security, and I'm breaking a social contract if I release crap, no matter what's written in the licence (the no liability legalese bit).

I believe my coding style is best shown by example. Some people have called it impressive. Some others have called it the worst they've ever seen. This may or may not come from the domain: cryptographic code tends to be pathologically straightline. At the very least it tend to produce longer functions than other domains.

Hey, maybe I can actually write my own crypto library?

Ouch. I would never do that without first thoroughly studying SQLite. I would need one hell of an iron-clad rationale to even contemplate writing my own DB code, and that's coming from someone who implemented his own crypto.

And before you accuse me of being part of the propaganda: I have never written a single line of Rust, and I'm actively working on a C cryptographic library of all things. That library is responsible for teaching me how insane UB is in C by the way. There is no way I ever willingly develop anything C or C++ ever again without getting it through all the sanitisers I can think of. (Fuzzing and property based tests, of course, are a given.) And by the way I highly recomend the TIS interpreter (or TIS-ci, which kindly gave me an account.)