Matt Sicker

I tried using curve25519_32.c for comparison with the Java variant. One of the confusing aspects is that the field elements are unpacked from int arrays rather than byte arrays as...

I just noticed that [OpenBSD](https://github.com/openbsd/src/blob/master/sys/crypto/curve25519.c) is using fiat-crypto, though it's somewhat transitive since it's related to WireGuard code.

Here's a fairly simple combination of operations which give different results in Java and C: https://gist.github.com/jvz/0bd0ff04b47524da8b8aedc16e79a4eb I've seen similar discrepancies in a few other operations similarly.

Thanks for the contribution! Can you fix the tests?

The test failures appear to be on Windows only, so that could be part of the problem.

Almost sounds like the test failure might be unrelated then?

Yeah I think this might be backwards. I'd expect a build with a build step to bind its credential parameter bindings to the invoked build.

Oh that's interesting! Let me verify this works equivalently to what I had in mind before I can merge this. Should be within the next few days.

It might be useful to highlight any form fields that should be paid attention to for updates in general. The full week extension rule idea sounds like a great UX...

@daniel-beck can you use your superpowers again? I haven't gotten around to enabling CD here yet.