Hi Nick,

Thanks for providing some feedback.

On Thursday, April 16th, 2026 at 15:57, Nick Hilliard <[email protected]> wrote:

> So, yes. RPSL did try the set theory thing and in theory this idea could be 
> workable, maybe.
> 

> Flicking back through rfc2622, I'm reminded of the horrors of mbrs-by-ref. I 
> know it's inexcusably poor taste, but I can't help wondering whether you 
> should also include a excl-mbrs-by-ref? Probably better to leave this for an 
> April 1 RFC.

I have a split opinion on this - it seems like a valid request to me, but few 
initial thoughts are:

1. I can use the proposed exclude members field to exclude anything, included 
sets included by mbrs-by-ref (although not as efficiently, I would have to 
specify each RPLS key to excluded, so it would just be an efficiency gain to 
excl-mbrs-by-ref, but that leads me to 2...)

2. Some sort of excl-mbrs-by-ref might exclude too much, I only want to exclude 
some stuff covered by the maintainer so I'll still need to fall back to the 
proposed exclude members field to do a partial exclude?

3. My understanding is that mbrs-by-ref isn't used that much, so is this worth 
it?

This is off the top of my head and not thought through in depth. I can try and 
pull up some numbers on mbrs-by-ref usage if you think it worthwhile?

> Firstly, any output is going to be acutely dependent on how 
> draft-romijn-grow-rpsl-registry-scoped-members operates. I haven't got to 
> that yet in my ietf draft consumption backlog, but my first thoughts would be 
> that the source:: scope would need to be applied aggressively to child 
> objects (maybe it is already).

Due to being a co-author of draft-romijn-grow-rpsl-registry-scoped-members, 
that is fully catered for in this document.

> There's plenty of scope for exclusion to create operational problems. E.g. 
> what happens when you have different sets which reference each other and then 
> one of them is configured to include the other as an exclusion? There's no 
> shortage of foot-guns here.

I have tried to think about this; I couldn't come up with a foot gun scenario, 
and I had already thought through stuff like order-of-operations type stuff 
that might arise. See section "3.3. Cumulative Excludes", if I've missed some 
scenario which leads to a new foot gun I haven't foreseen, please let me know, 
I obviously want to avoid that.

> A complete set theory implementation would ideally include operators for 
> union, intersection, complement and all that. You can synthesize it kinda 
> with a simple exclusion mechanism, but I wonder about data consistency and 
> general flexibility.
> 

> There are two main problems with RPSL. One is grammar flexibility, of which 
> there is very little, which is why any extension needs a new key mechanism 
> defined. The second is that it inherently requires symbolic expressions, 
> which increases the complexity of the parser.
> 

> So, if I were approaching this - which I'm not - I'd might be tempted to look 
> at this from the point of view of pure set theory and design a key which 
> allowed complex set manipulation grammar. I.e. create a full mechanism for 
> symbolic manipulation of as-sets. Any existing code implementation already 
> needs to deal with a good deal of symbolic manipulation, so this may not be 
> as problematic as it seems at the outset.

I'm split on this, like the earlier point - I see it as a valid point, but like 
the earlier point, it's not clear to me how worthwhile this would be. I think 
most people are only interested in including everything in set X, and for some 
of us, to then exclude a few bits (hence the document). I'm not aware of people 
with more syntactically advanced use cases (I'm happy to be corrected here).

> I'd also ditch extending route-set objects - no-one uses them really. Also 
> draft-romijn-grow-rpsl-registry-scoped-members does not apply to route-sets, 
> so you'd have a lot of work to do if you wanted to make this work properly.

Happy to :D My opinion was that route sets are basically dead, but I felt that 
if I left them out, the first feedback I'd get is from the one network still 
using them "WHY U NO ROUTE SETS!!!".

With kind regards,
James.

Attachment: signature.asc
Description: OpenPGP digital signature

_______________________________________________
GROW mailing list -- [email protected]
To unsubscribe send an email to [email protected]

Reply via email to