Death of irrtoolset?

[Nick Hilliard]

On 11/10/2009 13:36, S.P.Zeidler wrote:
> The actual front end code is not that terribly hairy (as far as the
> backends let it). The problem is the libraries irrtoolset contains that
> all apps use. They'd need a rewrite.

rtconfig is quite simple really.  And peval is just a CLI front-end to the 
libs.  As you say, the real problem is in the back-end machinery.  I 
recently spent some time looking at properly implementing a peering-set 
iterator for rtconfig, but gave up in frustration - hence the talk at RIPE59.

Having said that, with a suitable back-end library or else good quality 
hooks, rtconfig could be rewritten in perl / Template::Toolkit in a day or 
two.  C++ is too low-level.

> My own preference, btw, would be to have a core C or C++ tool that
> replaces peval, but spits out something more suited for being fed into
> other pieces of code (like XML), and replace everything else with
> perl modules. This way one could continue using lex&  yacc, which makes
> catering to features of RPSL a lot easier, and if a rabid fan of, say, lua
> wants stuff in their language they can reuse the hard part easily.

The RPSL rfcs are troublesome to implement because of their rich syntax and 
in particular, their support for regexps.  Also, as many people have 
pointed out, most people don't use most of the stuff in RPSLng, so there is 
limited value in reimplementing the whole lot.  I would see value in 
reimplementing:

- a sanely licensed and extensible rpsl expression evaluation library
- an extensible config generator
- a web based front-end tool which can create and manipulate rpsl objects
- an update to the rpsl rfcs to support mpls / vrfs.

I don't see a need to slavishly reimplement rtconfig, but peval is quite 
useful in itself and once you have a functional back-end library, there is 
no reason not to jam on a tiny cli front-end which could act as replacement 
for this tool.

Most of the reason that rpsl is so unsuccessful is that people don't have 
any simple tools to build constructions using it, and once the information 
is in the registry, there are no good tools to turn the policy into 
something which does anything.  But the idea is actually useful, and has 
lots of good quality niche value.

Nick