[precis] PRECIS Go implementation

Sam Whited <sam@samwhited.com> Fri, 05 February 2016 16:40 UTC

MIME-Version: 1.0
From: Sam Whited <sam@samwhited.com>
Date: Fri, 05 Feb 2016 10:40:15 -0600
Message-ID: <CAHbk4RKGsEXk6VCeyqL+OBx-1aocfoqUHr45qQ+93G3Jenrc4A@mail.gmail.com>
To: precis@ietf.org
Content-Type: text/plain; charset="UTF-8"
Archived-At: <http://mailarchive.ietf.org/arch/msg/precis/QS_PkuU98h6CgGiIvyB3JYgjYAU>
Cc: Peter Saint-Andre <peter@andyet.net>
Subject: [precis] PRECIS Go implementation
Precedence: list

Hi all,

I've mentioned that I was working on a Golang implementation of PRECIS
on this list before, and now that it's been merged into the
experimental part of the Go standard library I wanted to provide an
update and announce it's availability:

https://godoc.org/golang.org/x/text/secure/precis

There are still a few hiccups and TODO's, mostly due to the fact that
there's not currently a Go implementation of the BIDI rule, but those
should be smoothed out over time and I'm convinced that it's mostly
compatible with other PRECIS implementations like Christian Schedt's
recent Java version (from which I stole many tests, thanks Christian).

The source can be found here, for anyone who is interested:
https://github.com/golang/text/tree/master/secure/precis

An interesting thing to note about this implementation is that we
decided that it is fast enough to NOT implement the prepare step.
Instead, clients and servers alike are expected perform enforcement.
To give a quick example, here is a benchmark of enforcing the string
"Malvolio" (including setup and teardown, not reusing constructed
profiles):

PASS
BenchmarkUsernameCaseMapped-8             300000              5727
ns/op           14064 B/op         20 allocs/op
BenchmarkUsernameCasePreserved-8          300000              5428
ns/op           14064 B/op         20 allocs/op
BenchmarkOpaqueString-8                   300000              4736
ns/op            9888 B/op         19 allocs/op
BenchmarkNickname-8                       300000              4874
ns/op            9888 B/op         19 allocs/op
ok      _/home/sam/src/golang-x-text/secure/precis      6.452s

While this doesn't look great at first glance, it's worth noting that
this is a naive implementation with minimal optimizations beyond
pre-generating a trie of derived properties. It will be quite easy to
go back and get those 20 allocations per operation down to a more
reasonable number (even after adding BIDI support), which should
increase the speed per operation significantly.

Best,
Sam


-- 
Sam Whited
pub 4096R/54083AE104EA7AD3
https://blog.samwhited.com

[precis] PRECIS Go implementation Sam Whited