I've recently spent some time looking at XProc: An XML Pipeline Language. I'm starting an xproc tutorial, though it hasn't progressed very far as yet. As with many (most?) W3C specs, they are written for implementers, not users. I'm finding it hard going, especially as the implementations are still under development, so it's the usual question, my misunderstanding, or an implementation bug.

I'm pretty optimistic about xproc. It meets a clear need with XML. Just as DSDL has addressed the validation itch, so I'm hoping that xproc will meet the need to string multiple process steps together when dealing with a process chain, either one you've developed, or one that comes as part of a product or service. I have quite a few laying around. My blog. Aggregates using Python, sorts and selects using about 4 XSLT steps, then uploads to the site using ftp. The xproc write-up. It's Docbook, but in chapter level chunks. So I run the xInclude processor, validate using Jing, transform to HTML using the Docbook Stylesheets and again upload. This time I'm using Apache Ant, which is quite a favourite. With such chains of commands, I find it works well when there are no errors. Changing the process to deal with errors produces the oddities that I'm hopeful xproc will manage in a more regular manner. E.g. Jing isn't properly integrated into Ant, so when it discovers errors, ant continues processing, not what I want.

Anyway, that's my reasoning for spending time with xproc. Hopefully it will be a W3C rec within a few months.