> crypto stuff also, my assumption here would be that the built-ins don't have any comparison problems since they fetch the carry bit from a cpu flag - ie any time you're doing "crypto stuff", you kind of make assumptions about the compiler you're working with given the lack of standardization around constant-time-ness.
Given the state of compilers, it's perhaps slightly beyond the reach of the primitives here to test for that, although it would indeed be a huge advantage if it was verified in unit tests. LLVM for example tests the assembly output of a given program in certain cases, it would actually be pretty cool to have a test that ensures that for a given platform, a "construct" compiles to a certain instruction so that "crypto stuff" can be built more reliably. This also reminds me of C++ atomics where the standard provides a compile-time constant that describes whether they are true atomics or emulated with locks on any given platform - something similar would perhaps be interesting here, ie that on "verified" platforms, it would be set to `true` to signal to the calling library that it's "safe". This would again be trivial to implement in `nlvm` \- in `nim`, it's a bit more tricky because of the interaction with the underlying C compiler zoo.
