ecc

a for ECDH, or information about the nonce in ECDSA.) It looks like the reduction in kinv alone would likely leak log2(k^-1) pretty directly in timing, so if you could capture some large number of signatures with a single key you could probably recover it, potentially even remotely. (Not trying to be harsh here, OpenSSL is just as bad for their generic code... but if someone is advertised as constant time, it ought to meet the strongest definition of it since the caller is not going to be a cryptographer and he's going to treat it as a black box and inevitably use it in the most exploitable way possible. :) ) General question about rust: Is there any way to reliably write constant time code in it? Or are you always at the mercy of the compiler undermining you? E.g. happily unrolling loops, optimizing out dummy operations, and turning branchless comparisons into branchy ones. (I wouldn't have expected that would be, since there isn't really in C; unless you're able to audit the compiler output ... and rust is even higher level).