Thou shalt have no other gods before the ANSI C standard 1639

Thomas lovein

The mathematical function is defined for any length. The software API is limited to a gigabyte. I haven't seen any applications where this is a problem; security requires a separate authenticator on every packet.

If it turns out that people need Poly1305 software that can compute an authenticator of a 10-gigabyte disk file, I'll write software to do the job. Of course, the length should then be measured by int64, not some CPU-dependent notion such as sizet.

Even if that's true, who cares? The reader doesn't say to himself ``Wow, I'm not sure what signed right shifts mean for a 32-bit twos-complement number, because the ANSI C standard hasn't told me what it means!''

The notion that the English language is defined by ANSI is even sillier than the notion that real-world C programming is defined by ANSI. I see no evidence of any actual ambiguity here.

``I think your code wasn't portable to a machine I had five years ago, and I think this modification would have made it work, and I could have tried it five years ago'' doesn't even come close to meeting my standards for code tests.

I often release software that handles computer variations by conditional compilation. The poly1305 library has several separate asm versions! But this also means extra tests---which I make sure to do.

There was an Alpha at UIC, and perhaps it's still running, but it always ran Linux; the C environment was pretty much the same on that machine as it was on, say, a 64-bit SPARC. Yes, it's conceivable that I can find a user still stuck with Tru64, but I can reduce the porting costs by waiting for that user to send me email.

---D. J. Bernstein, buttociate Professor, Department of Mathematics, Statistics, and Computer Science, University of Illinois at Chicago