On 10 Feb 2011, at 23:13, Andrew Deason wrote:
>> Premature optimization is the root of all evil.
>> Are you sure that the savings of a few bytes in, say, 
>> VL_GetAddrsNewAndShiny is worth the extra complexity?
> 
> No, but I also don't think there's much extra complexity. "Eh".

I think at this point we should be looking to solve the common case - we need 
to replace the use of afs_int32 for addresses. Anything more complex should 
really wait until we identify the need for it. Otherwise, we're going to end up 
constantly trying to design the better mousetrap.

>>> This would make things easier in many other situations as well, so
>>> I'm all for it. Would that mean we're diverging from regular XDR,
>>> though?
>> 
>> It would, but then, that's not new.  I proposed a new _primitive_ type
>> several days ago, though someone (you?) has since described a way to
>> do what we were talking about in terms of an opaque<>.
> 
> Yes, but IMHO, that way is very annoying.

Sadly, it's the way we're heading with many XDR structures, due to the lack of 
a flexible union type. It's actually not quite so bad in this case, because 
we're representing two types that natively are just comprised on a sequence of 
bytes. In other places in AFS, though, we've ended up using opaques which then 
go through a further layer of XDR demarshalling. This is obviously far from 
ideal.
> 
> So I would agree with a primitive type. Do we want to keep new primitive
> types prefixed with "afs" like afsUUID was? Call it... afsTLV ?

I'd like to see a primitive type too. I'm not sure where you're getting TLV 
from (yet another TLA?) - would something like afsAddress not be cleaner?

Cheers,

Simon.

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

Reply via email to