Re: [closed] Re: comments on XProc last-call draft

>We've been content so far to leave the question of whether steps are
>executed in a parallel or serial fashion as an implementation issue.
>Can you provide an example where user-level control over this behavior
>is necessary?

Following is an "adaptation" of an use case I have currently.
How can you guarantee spec interoperability for it?
<p:pipeline name="main">
	<p:group>
		<p:xslt />
		<acme:delete-file file="the-input-file" />
	</p:group>
	<p:xslt >
		<p:input port="source">
                    <p:pipe step="main" port="source" />
                </p:source>
	</p:xslt>
</p:pipeline>
Apparently this is a broken processing from non-parallel point of view, and quite fine from parallel.

My opinion is that parallelism is too big issue for implementations to deal alone. Language support is needed as to which use cases are supported and which not.

Also, this is a question about cpu scalability as well. Currently it is very unclear to me what this spec would be capable of covering for enterprise systems.


> | 8) "p:serialize" : I'd happy to see also "exclude-prefixes" (after XSLT).
>
> Can you live without it? :-)
>   
The workaround is to use xslt for serialization. Frankly it would be 
very annoying ;(

Nikolai

Received on Wednesday, 30 January 2008 12:47:32 UTC