The mighty version attribute

Hi folks,

I've been thinking for a long time about adding a version
attribute to Mallard. It seems easy enough, but I wanted
to define it in a way that allowed validators to know how
to validate a page. And I wanted that to include extensions.

I thought I had a really great idea on how to do this, but
then ran across the "name classes aren't patterns" limitation
in RELAX NG:

http://tech.groups.yahoo.com/group/rng-users/message/1214

Turns out, I still think I have a good way to do this, but it
requires working around this limitation. I've just done a basic
proof-of-concept implementation of this idea, and I'd like to
share it and get feedback.

We'd have a version attribute on the page element. This attribute
takes a space-separated list, where the first element identifies
the primary schema, and the others identify extension schemas.
Version token are mapped to URIs of RNG files as follows:

 * If the token has a colon, it's taken to be an absolute URI.
 * If the token has a slash (a/b), then we construct this URI:
   http://projectmallard.org/a/b/a-b.rng
 * Otherwise, we construct the URI from the token v like this:
   http://projectmallard.org/v/mallard-v.rng

Then we construct a new RNG file by slurping in the contents
of each of these, copying the child elements of each grammar
element, being careful not to screw up element definitions
that rely on the default namespace.

Except (and this is the workaround), we need to rewrite the
bits that allow external namespace elements and attributes.
So we collect a list of namespaces by looking at the ns
attribute of the grammar element of each of the RNG files.
Then, whenever we encounter anyName/except, we replace it
with anyName except any of the collected namespaces, plus
the local namespace.

It's not nearly as complicated as it sounds. I've attached
an XSLT file that takes a Mallard file and outputs an RNG
file that will validate the page according to what I just
outlined.

I wish the name class hack weren't necessary, but it is.
This technique works nicely with the ways Mallard allows
external namespace extensions.

--
Shaun

-------------- next part --------------
A non-text attachment was scrubbed...
Name: mal2rng.xsl
Type: application/xslt+xml
Size: 2977 bytes
Desc: not available
URL: <http://projectmallard.org/pipermail/mallard-list/attachments/20110220/295b3f1c/attachment.xsl>