On schema quality and schema limitations

Hi,

  On http://www.w3.org/QA/2002/04/Web-Quality:

[...]
  Note: Some documents are valid with regards to the DTD and still
        incorrect with regards to the HTML specification. In the near
        future, we will present a list of possible errors not detected
        by the HTML validator.
[...]

This has not happend yet. Unfortunately. In

  http://www.w3.org/mid/3f9ab490.140875568@smtp.bjoern.hoehrmann.de
  http://www.w3.org/mid/3faa04b3.226926613@smtp.bjoern.hoehrmann.de

I have stressed why such lists and proper conformance terminology are
important. Modularization of XHTML 1.0 Second Edition

  http://www.w3.org/TR/2004/WD-xhtml-modularization-20040218/

make such lists even more important. It is quite difficult to properly
review the XML Schema modules without information about what is covered
in those schemas, what is not, and most importantly why it is not
covered. It seems for example that

  <font color='orange'>...</font>

is not invalidated by those schemas, while

  <font color='#XXXXXX'>...</font>

is. Now I wonder why. Is it because XML Schema does not allow to say
the value must match the PCRE

  /^(black|white|...)|(#[0-9A-F]{6})$/i

or is this necessary for extensibility if a host language chooses to
allow color='orange' or is this intentional because everyone is using
color='orange' anyway? Or what about the ContentType (text/html, etc.),
it is defined as

  <!-- media type, as per [RFC2045] -->
  <xs:simpleType name="ContentType">
    <xs:list itemType="xs:string"/>
  </xs:simpleType>

Looks familiar to

  <!-- comma-separated list of media types, as per [RFC2045] -->
  <xs:simpleType name="ContentTypes">
    <xs:list itemType="xs:string"/>
  </xs:simpleType>

Why are both xs:lists? http://www.w3.org/TR/xmlschema-0/ suggests that
xs:list is for (white-?)space separated lists, yet neither type allows
spaces (or maybe ContentTypes does before and/or after the comma, I've
been unable to find explicit information in this regard) so this seems
to be an inappropriate type for both. 

  http://www.w3.org/TR/xhtml1-schema/

defines them as

  <xs:simpleType name="ContentType">
    <xs:restriction base="xs:string"/>
  </xs:simpleType>

  <xs:simpleType name="ContentTypes">
    <xs:restriction base="xs:string"/>
  </xs:simpleType>

That seems a bit more accurate, but

  <form accept="Karl Dubost" ...>

even though it would probably be fun to upload Karl somewhere, it seems
that this should not be allowed. At the very least I would expect that
ContentType is required to contain a slash. Maybe it is not due to one
of the reasons cited above. So continue to look at other types,
FrameTarget is next. The PCRE for FrameTarget is (derived from HTML4)

  /^(_(blank|self|parent|top))|([A-Z].*)$/i

(note in particular that target='_new' is forbidden). Ok, now

  http://www.w3.org/TR/xhtml1-schema/

has

  <xs:simpleType name="FrameTarget">
    <xs:restriction base="xs:NMTOKEN">
      <xs:pattern value="_(blank|self|parent|top)|[A-Za-z]\c*"/>
    </xs:restriction>
  </xs:simpleType>

This appears to be more restrictive than my PCRE. Hmm, [A-Za-z] suggests
that this pattern is case-sensitive,

  <a target='_BLANK' ...>

would thus be considered invalid. It is not according to HTML4. Just
like

  <a target='X$' ...>

is allowed in HTML 4. But then I have never understood the model behind
changes (and lack thereof) for the lexical space of such attributes
between HTML 4 and XHTML 1. Looking at the XHTML M12N SE WD again, it is
defined as

  <xs:simpleType name="FrameTarget">
    <xs:restriction base="xs:string"/>
  </xs:simpleType>  

This appears to allow any of

  <a target='_new' ...>
  <a target='_BLANK' ...>
  <a target='X$' ...>

So maybe this changed again? Or there is some good reason for this
aswell. Oh, by the way, http://www.w3.org/TR/xhtml1-schema/ defines the
color type as

  <xs:simpleType name="Color">
    <xs:restriction base="xs:string">
      <xs:pattern value="[A-Za-z]+|#[0-9A-Fa-f]{3}|#[0-9A-Fa-f]{6}"/>
    </xs:restriction>
  </xs:simpleType>

while M12N has

    <!-- sixteen color names or RGB color expression-->
    <xs:simpleType name="Color">
        <xs:union memberTypes="xs:NMTOKEN">    
           <xs:simpleType>                      
              <xs:restriction base="xs:token">
                 <xs:pattern value="#[0-9a-fA-F]{6}"/>
              </xs:restriction>
           </xs:simpleType>         
        </xs:union>
    </xs:simpleType>

Seems like in one XHTML schema I can write

  <font color = 'Hazaël-Massieux'>...

while the other invalidates it. More importantly, very common constructs
such as

  <body bgcolor = 'ffffff' ...>

would apparently validate. Maybe this is useful. Puzzling. Another
example, the class attribute is defined in HTML 4 as CDATA, in XHTML
1.0 it is still defined as CDATA but in XHTML 1.1 is is NMTOKENS. If
I remember correctly, I have been told that XHTML 1.0 was supposed
to be as close as possible to HTML 4. But the target attribute changed
from CDATA to NMTOKEN. So, if there really is a rule, it seems that it
is not consistently applied.

Oh, great, XML 1.0 was changed to allow empty xml:lang attributes, I
told the HTML WG a year ago and asked them to incorporate this change
into their DTDs

  http://www.w3.org/mid/3e856679.216018678@smtp.bjoern.hoehrmann.de

Let's see

  <!-- a language code, as per [RFC3066] -->
  <!ENTITY % LanguageCode.datatype "NMTOKEN" >

Hmm, they never got back to me on this one, ... aha, here

  http://hades.mn.aptest.com/cgi-bin/voyager-issues/Modularization-DTDs?user=guest;selectid=6298

"Updated in XHTML Modularization SE". Note quite, no? It would also be
interesting to know whether <form name='...' ...> and <a name='...' ...>
are allowed in XHTML 1.0 SE Strict. HTML 4.0 Strict does not allow it,
HTML 4.01 Strict allows it, XHTML 1.0 FE Strict does not, XHTML 1.0 SE
Strict does not either. I have asked in July 2003

  http://www.w3.org/mid/3f4a2211.377098258@smtp.bjoern.hoehrmann.de

According to

  http://hades.mn.aptest.com/cgi-bin/voyager-issues/XHTML-1.0?user=guest;selectid=6504

they still need to figure this one out. Of course

  http://www.w3.org/2002/08/REC-xhtml1-20020801-errata/

says

  Known errors
  None at this time.

Back to M12N SE XML Schemas, http://www.w3.org/TR/xhtml1/#prohibitions
notes that e.g.

  <a ...><span><a ...>...</a></span></a>

is not allowed but this cannot be expressed using XML DTDs (while SGML
DTDs allow it and it is defined that way in the HTML 4 DTDs). Will XML
Schema Validators catch this? And if, is this constrained spellt out
in the relevant schemas?

What about anchors, there must be a unique anchor <=> element
relationship, is this requirement covered by those schemas? I asked
whether this is possible

  http://www.w3.org/mid/402c46ea.490712317@smtp.bjoern.hoehrmann.de

but

  http://lists.w3.org/Archives/Public/xmlschema-dev/2004Jan/0073.html

suggests it is not. HTML 4 also says that <input type='reset' ...> is
allowed to omit the name attribute, while <input type='password' ...>
is not, I believe this is neither possible to spell out in XML Schema.

I do not know, I am not a XML Schema expert. I did not even manage to
figure out how a xs:list is supposed to be separated from the
specification. So it seems I won't become one either. Well...

As I point out in

  http://www.w3.org/mid/3faa04b3.226926613@smtp.bjoern.hoehrmann.de

this might all get worse. If a specification ships with schemas in DTD,
RNG, WXS 1.0, WXS 1.1, Schematron, ... and they all combined still don't
cover certain aspects of validity... Who is reviewing these schemas? 
Unless I miss an important (probably undocumented) aspect of XHTML M12N
SE, it strikes me as most obvious that these schemas are not quite what
they should be. They are on the Recommendation Track for more than three
years now, am I the only one who looks at them? That seems a bit
unlikely. But this appears to underscore of what I suggested for SpecGL.
Though that might be insufficient. Maybe the QA Activity should have a
Schema Expert who reviews schemas as part of the QA review. That would
of course be most difficult if machine-reportable errors are hard to
discover in the specification.

  http://www.w3.org/TR/xhtml1/#prohibitions

is good practise as is

  http://www.w3.org/TR/xhtml1-schema/#diffs

That is at least something. Insufficient, but helpful. XHTML M12N SE
lacks such a section.

  http://www.w3.org/QA/WG/2003/09/qaframe-spec-extech-20030912

suggests that the QA Activity like XHTML M12N a lot. I do not. I do not
like statements such as

[...]
  When the user agent claims to support facilities defined within this
  specification or required by this specification through normative
  reference, it must do so in ways consistent with the facilities'
  definition.
[...]

Especially not if "facilities" is undefined. Of course

  http://www.w3.org/2001/04/REC-xhtml-modularization-20010410-errata

  Known errors
  None at this time. 

At least they fixed this unknown error in the SE draft in response to

  http://www.w3.org/mid/3f650077.197501031@smtp.bjoern.hoehrmann.de

But that does not help such statements. Lets have a look at M12N SE
again,

[...]
  3.4. XHTML Family Document Conformance

  A conforming XHTML family document is a valid instance of an
  XHTML Host Language Conforming Document Type.
[...]

What a "valid instance" is I do not know. Btw., they happen to like
this facilities speak very much, 

http://www.w3.org/TR/xhtml-print/

[...]
  2.1. Document Conformance

    A conforming XHTML-Print document is a document that requires only
    the facilities described as mandatory in this specification. 
[...]

What it means for a document to require something, or what these
facilities are, or which of them are described as mandatory, I do not
know. At least they improved the text discussed in 

  http://www.w3.org/mid/407fad85.21285647@smtp.bjoern.hoehrmann.de

a little...

Hmm, it seems I got drifted a bit...

But back to <http://www.w3.org/QA/2002/04/Web-Quality>, documenting the
current limitations of the W3C MarkUp Validator is quite simple, doing
so for all these schemas probably not. But it is apparently necessary to
do this even if only to improve the quality of the schemas. It is also
much simpler to write a little add-on script (or XSLT) that could be
plugged into an existing ACME schema validator to cover uncovered
aspects. Multiple tools or schemas don't help as the tools lack
functionality to share validation information. It is also helpful for
the community if such information is available as they would provide
better understanding of the issues involved. It makes them aware where
to trust validators and where not. It also helps to make them aware
about certain constraints. And tool developers would get less bed
feedback about "changing the rules" and all that. Seriously, if the
MarkUp Validator is improved to check whether %URI; attributes really
contain legal URIs, I am certain there would be negative feedback from
the I18N WG/IG about invalidating their

  * http://www.w3.org/International/tests/test-idn.html
  * http://www.w3.org/International/tests/sec-idn-1.html
  * http://www.w3.org/International/tests/sec-idn-2.html

conformance test pages. But test suite should be

  http://www.w3.org/mid/Pine.LNX.4.58.0403121204370.23385@dhalsim.dreamhost.com

valid, no? But maybe these are error recovery tests. I do not know, they
do not mention error recovery. Maybe I miss something. The QA Activity
apparently does not have the resources to document limitations of tools
and schemas. Who knows best about conformance requirments? The WG
publishing the spec. And who knows best about limitations in published
schemas? The editors of these schemas. Hence they should be required to
edit schemas and their limitations. And WGs as I suggested for SpecGL.

regards.

Received on Friday, 9 April 2004 00:49:50 UTC