Re: [dispatch] New version for the Alert-InfoURNs draft: draft-liess-dispatch-alert-info-urns-00.txt

["Spencer Dawkins" <[e-mail]>]

I agree with Adam's point here (matrix isn't sparse, but a lot of the 
n-dimensional points are really boring because they have the same value as 
other n-dimensional points).

I was seeing this the way Paul was seeing it previously, so this was a 
helpful insight.

Thanks,

Spencer

>> I will take this as an open issue for the dispatch meeting slides. I
>> hope Adam will be there because it was his proposal to use combinations
>> of discrete items rather than hierarchy.
>
> Sorry for not following this more closely.
>
> At this point, I think we have two different things to tease apart: (1) 
> The semantics for combining different dimensions of ringback, and (2) the 
> syntax for combining these dimensions.
>
> We really need to understand the first one before we start quibbling over 
> the second one. So, while I do have some thoughts about the syntax, I'm 
> not going to go into them -- even a little bit -- until we have a better 
> grip on the semantic problem.
>
> Here's an important point that Paul raises:
>> It seems that what you propose is a multidimensional space, where every 
>> point in that space is potentially rendered uniquely. But in practice 
>> values will often be supplied for some but not all of the dimensions. In 
>> that case we end up specifying some "slice" of the array. I guess in this 
>> case you could look to see what you would render for each point in the 
>> slice. If those are all identical, then you are in good shape. But if 
>> they are not, then what? Do you pick one of the points? That amounts to 
>> choosing a default value for each unspecified dimension. Is that the 
>> right answer?
>
> I agree. We're effectively defining an n-dimensional space, and assuming 
> that a point can be selected with fewer than n ordinals. And, yes, I think 
> the correct answer is simply defining a default value for every new 
> dimension defined -- preferably within the specification of the dimension 
> itself (although some things, like country indication, would probably be 
> better handled as local policy). Then, you can specify a value for as few 
> as zero of the dimensions, and still have the answer be a point.
>
> The other side of that coin is that clients receiving this information 
> will not necessarily understand (or care to support) all of the dimensions 
> being presented. I believe that this is handled on the recipient's part by 
> ignoring any dimensions it does not understand.
>
> However, I think that Paul's characterization of the difference between 
> the previous approach and the current approach isn't correct:
>> It seems like what you must be doing is replacing a hierarchy with an 
>> N-dimensional matrix which is sparsely populated. That seems harder to 
>> deal with.
>
> I don't agree that it is sparsely populated. Many of the points might 
> result in the same answer, but that doesn't mean it's sparsely populated. 
> I think you could take the current draft and (modulo the issues with 
> 7.3.3) pick any value from each of the three dimensions at random, and 
> have a semantically sensible ringtone. Here; I'll roll the dice a couple 
> of times: Internal short Brazilian ringback. External priority Pakistani 
> ringback. You can play along at home -- find me an empty space in the 
> matrix, and I'll concede the point.
>
> So, quite to the contrary, I think the two approaches are *semantically* 
> identical, and the only difference between what was proposed before and 
> what is proposed now is a *superficial* and *syntactic* difference. If you 
> need proof, I'll write a quick perl script that converts between the two 
> syntactic representations. :)
>
> /a