Re: A "processing model" proposal

> I have no problem with:
> 
>   <p:stage name=3D"someFunkyThing">
>     <p:output name=3D"p1"/>
>     <p:output name=3D"p2"/>
>   </p:stage>
> 
> and then some other stages later on:
> 
>   <p:stage name=3D"someOtherFunkyThing">
>     <p:input name=3D"p1"/>
>   </p:stage>
> 
>   <p:stage name=3D"someOtherOtherFunkyThing">
>     <p:input name=3D"p2"/>
>   </p:stage>

Yes, that's the sort of thing I was thinking of.

> | I also want a single output to be
> | connectable to multiple steps' inputs, but we could have (conceptually
> | at least) a "tee" component that produces multiple copies of its
> | input.
> 
> I think that's fine as long as they're labeled. I think we should make
> infosets immutable. That is, a process can transform A to produce B,
> but A still exists and hasn't been changed.

Conceptually, yes.  If two steps have the same infoset as input, then
one step shouldn't be able to change what the other one sees.  In terms
of APIs, an implementation might want to make infosets mutable, in which
case it must provide separate copies.  Sending one infoset to two steps
also has implications for streaming, of course.

-- Richard

Received on Thursday, 16 February 2006 22:37:29 UTC