On Fri, Mar 11, 2011 at 6:51 PM, Simon Wilkinson <[email protected]> wrote:
>
> On 11 Mar 2011, at 23:27, Tom Keiser wrote:
>> So, what are we going to do about the GetCapabilities issue?  A
>> compliant implementation could cache this data forever, which
>> is.....preposterous.  Can we can handle this via the RFC errata
>> process?
>
> It's not an RFC, and we're not the IETF, so no.
>
> We've never really codified what happens next, beyond that being declared 
> "experimental" is the point that people can go away and get identifiers, and 
> release products containing the code. It's really important that this mean 
> something, which is why I don't think we should accept late challenges or 
> modifications to the document. People should be free to go away and write, 
> and release, code based on this groups consensus without worrying that weeks 
> later someone will go "hang on", and rewrite the standard under them.
>
> Sure, we may get it wrong. When we do, we have to accept that code may be 
> deployed containing our mistakes, and we have to consider how to fix those 
> mistakes without breaking the deployed base.
>
> In this case, there are two separate issues. The first is a document 
> clarification (GetCapabilities responses should only be cached for a certain 
> number of hours). I think that one is acceptable to bring back as a 
> clarfication when the document is next advanced (which is the point it will, 
> hopefully, become an RFC).

Hi Simon,

Ok.  That's perfectly fair.  I certainly don't want to hold up the
implementation process.  If we can introduce a TTL clarification
during the transition from afs3-stds experimental to informational
RFC, then I'd certainly feel a tad better.  However, if that would
complicate Derrick's implementation of the GetCaps client, then please
let me know and I'll forget I ever brought this up :)

>
> The second, however, is harder. I don't think that we can add fields to RPCs 
> that have reached experimental. So, we have to decide whether the possible 
> race is sufficiently serious to warrant creating a new RPC, with a new code 
> point, to implement this behaviour. In any event, it's perfectly acceptable 
> for people to release code and/or ship products that contain the RFC as 
> described within the current document - all we can do is propose a better 
> successor.
>

Indeed.  Conveniently enough, my second point is a total nit; if we
ever see this race in the wild, I'll honestly be a bit surprised.
Anyway, should we ever feel the need, introducing a RemoveAuthName2
RPC with the second argument in a future I-D is quite simple...

Cheers,

-Tom
_______________________________________________
AFS3-standardization mailing list
[email protected]
http://lists.openafs.org/mailman/listinfo/afs3-standardization

Reply via email to