Re: Minutes for RELAX NG TC 2001-06-14

[Kohsuke KAWAGUCHI <kohsuke.kawaguchi@eng.sun.com>]

> clarification. Murata-san was using the term in light of regular tree
> languages that can generate acceptable sentences that can then generate
> trees. [Clarify?]

Here is my take:

A regular expression R (like "a+,b") is said to "generate a language"
because R defines a set of strings that are accepted by R. "Language"
means "set" here.

So the language generated by "a+,b" is  {"ab","aab","aaab", ... }

Here, We'd like to extend this to a RELAX NG pattern. A RELAX NG pattern
P is said to "generate a language" because P defines a set of trees that
are accepted by P.

So the language generated by 

<element name="foo">
  <data type="integer"/>
</element>

is

{ <foo>0</foo>, <foo>1</foo>, ... }


Now consider the following pattern:

<element name="foo">
  <attribute name="a"/>
  <attribute name="a"/>
</element>

What is the language generated by the above pattern?

Is it {<foo a="..." a="..."/>}?

The only reason we want to remove <foo a="..." a="..."/> from the above
set is, it is no longer XML.

So look at this in this way: the pattern can "generate" something that
is no longer XML.


Murata-san was thinking about this and wanted to add restrictions so
that no legal RELAX NG pattern generates "non-XML".

James then argued that this objective is actually already accomplished
by the current inference rules.

I said that is true but that is only because the current inference rule
is too smart.

--
Kohsuke KAWAGUCHI                          +1 650 786 0721
Sun Microsystems                   kohsuke.kawaguchi@sun.com