Re: Permanently sinking output ports

> The problem with your proposal is that the XProc standards doesn't say that you read the pipeline sequentially (because of the dependency graph)

I know, but the dependency graph is precisely why  the author or the runtime can know statically when some ports will no longer be read. What I suggest is for the author to be able to give a hint to the implementation, so that it's easier to implement such optimization without requiring to perform a complete graph analysis.

Romain.

On 17 nov. 2012, at 12:51, QuiXProc XProc <quixproc@gmail.com> wrote:

> Hi Romain,
> 
> At QuiXProc, we tried to find some way to implement such features
> 
> The problem with your proposal is that the XProc standards doesn't say that you read the pipeline sequentially (because of the dependency graph)
> 
> I think the only way to implement such feature nicely is that implementation detect that they have finished to read the stream and drop it
> 
> You may have a look at QuiXProc Open 1.1.0 ( http://code.google.com/p/quixproc/ ) and tells us if it deals better with resources
> 
> Regards,
> 
> The QuiXProc Team
> 
> 
> On Fri, Nov 16, 2012 at 1:46 AM, Romain Deltour <rdeltour@gmail.com> wrote:
> Wouldn't it be useful to be able to specify statically at authoring time when a port will no longer be used after a certain point in a pipeline ? I'm thinking of an attribute in the p:sink element that would say "when you reach this p:sink, these ports will no longer be read".
> 
> It would help implementors to free the memory during at execution time, without needing some advanced static optimization.
> 
> Just a thought…
> 
> Romain.
> 
> 
>