Michael Scott

The ARM CryptoCell-310 may be a bit of a problem. See this thread https://devzone.nordicsemi.com/f/nordic-q-a/18578/arm-cryptocell-310-performance which suggests that for elliptic curve cryptography at least , the hardware is actually slower than...

Actually I am having some second thoughts! Figure 4 in the RFC shows [Application data] being sent from the Server to the Client (presumably in response to the early data...

Hello Nico, Yes! that works fine. My client first does a handshake retry (as you don't do X25519?), and then tries a session resumption with early data. And the server...

Will do. Yours is the most developer-friendly implementation I have come across.. BTW if you want to consider an alternative Swift crypto library (that does do X25519!), check out https://github.com/miracl/core...

Not actually a problem, in there by design. I guess the compiler is throwing an error only because you have a compiler flag like -fsanitize=undefined. In multi-precision arithmetic with careful...

Alas there is no such option -fno-strict-overflow And it is not wrap around that is wanted, we just need the bits to fall quietly off the end. A good compiler...

Furthermore if you insist on compiling with a flag that causes the compiler to do insert overflow checks on every integer operation, then the extra code inserted by the compiler...

Then how can you be sure that undefined behaviour will not occur in the release version? You may test with the sanitizer on, but conditions not tested for may occur...

So don't ever do negative shifts! On Wed, Mar 1, 2023 at 12:36 PM Matthias Görgens ***@***.***> wrote: > Here's the shift-example > > I mentioned. > > #include >...

In C a[NLEN_XXX - 1] = ((a[NLEN_XXX - 1] > (BASEBITS_XXX - n)); causes an apparent problem. Undefined behaviour. In Rust (release mode) self.w[NLEN - 1] = (self.w[NLEN - 1]...