[CFRG] Re: Where should test vectors live?

["Bellebaum, Thomas" <[e-mail]>]

A few random thoughts...

Regarding long term availability: I am trying to think backwards, along the lines of "How would reference code for TCP/HTTP/RSA/... have to be written at the time to be useful today?".
I am uncertain that the obvious answer would have been python, although the question of which technology has been around for a while might not be equivalent to which technology will prevail going forward.
C89 with no dependencies besides the standard library seems quite stable, and will probably be around for a while. I can't tell whether this would be a crowd pleaser though.

As for where testvectors/reference code is to be stored, I like the idea of hosting it with the IETF, or, at the very least, provide updatable HTTP redirections (or links) to GitHub somewhere below ietf.org. RFCs should then reference ietf.org rather than GitHub directly. This way, if anything happens and GitHub disappears, one can simply relocate any artifacts.

There is one additional benefit to having test vectors within an RFC: To provide examples. At least I found them helpful a couple of times to gain a quicker understanding of how an algorithm functions. Of course, this is not an argument against providing one dedicated (primarily human readable rather than machine readable) example in the RFC, and delegating everything else to external resources.

-- TBB