Re: Proposed resolution for #19: attrInAttr

[James Clark <jjc@jclark.com>]

I have a similar concern with this.  If the spec is going to say that
implementations should issue a warning, it needs to define precisely what
the condition is under which they should issue a warning. I don't think
defining this precisely is trivial. You can have attribute patterns contains
attribute or element patterns that can match something.  For example, should
it give a warning for:

<attribute name="foo">
  <choice>
    <data type="xsd:string">
    <element name="bar">...</element>
  </choice>
</attribute>

? Why should it not be an error to have something like:

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

?  Is it because it is too hard to detect?  If it's too hard to detect, then
I don't want to say that implementations "should" detect it.

----- Original Message -----
From: "MURATA Makoto" <mura034@attglobal.net>
To: <trex@lists.oasis-open.org>
Sent: Friday, May 25, 2001 5:18 PM
Subject: Proposed resolution for #19: attrInAttr


> After studying GNF, I have changed my mind.  Attribute patterns containing
> attribute patterns look strange, but I do not see other technical reasons.
>
> Here is my proposal.
>
> This syntax allows attribute patterns to  directly or indirectly contain
> other attribute or element patterns, which never match anything.
Implementations
> SHOULD issue warning for such embedded patterns, but this is not an error.
>
> Cheers,
>
> ----
> MURATA Makoto  mura034@attglobal.net
>
>
>