circular references in schemas

[Christine McGavran]

I have created an xml format for describing a custom user interface. My
application successfully parses and displays the UI described in xml files
created in that format.

Before today we were using Visual Studio .net 2002. I created a schema file
to describe my xml so that I could use the fancy xml editing capabilities in
the Visual Studio enviroment. With this schema referenced, I could
auto-complete while scripting xml for my user interface while editing in
DevStudio. A pretty neat feature, even though it wasn't really necessary.

Today we upgraded to Visual Studio .net 2003. One of the changes seems to be
that schema files are compiled instead of interpreted directly. The compiler
gives the following error on compile: "DataSet doesn't allow the circular
reference in the ComplexType named 'Window'". So no more fancy editing.

This isn't a Microsoft-specific problem, I get a similar error if I try to
use an application I downloaded via w3.org to validate xml using my schema.

My script allows you to create windows with child windows, which can also
have children. A pretty standard windowing system, I think. But it doesn't
seem legal xml to make a window element a child of a window element. I'm not
really familiar with all the logic and theory behind xml, I just thought it
was a convenient way to script our UI. I would like to understand why these
"circular references" are not allowed in the schema, and what alternatives a
proper xml guru might suggest for my situation.

Below is a simple schema that has this problem

<schema targetNamespace="test.xsd" elementFormDefault="qualified"
xmlns=http://www.w3.org/2001/XMLSchema xmlns:ui="test.xsd">
<element name="test">
<complexType>
<sequence>
<element name="window" type="ui:Window" />
</sequence>
</complexType>
</element>
<complexType name="Window">
<sequence>
<element name="children">
<complexType>
<sequence>
<element name="childWindow"
type="ui:Window" />
</sequence>
</complexType>
</element>
<element name="testString" type="string" />
</sequence>
</complexType>
</schema>

 

[C. M. Sperberg-McQueen]

Today we upgraded to Visual Studio .net 2003. One of the
changes seems to be that schema files are compiled instead
of interpreted directly. The compiler gives the following
error on compile: "DataSet doesn't allow the circular
reference in the ComplexType named 'Window'". So no more
fancy editing.

Is that the entirety of the error message?

This isn't a Microsoft-specific problem, I get a similar
error if I try to use an application I downloaded via w3.org
to validate xml using my schema.

More details, please? Do you mean XSV? If so, what exactly is the
error code XSV is giving you? I don't get any schema error from XSV
on the sample schema you provided (although I do get errors on any
document I provide, for reasons which will become clear below).

My script allows you to create windows with child windows,
which can also have children. A pretty standard windowing system,
I think. But it doesn't seem legal xml to make a window element a
child of a window element.

Ouch! No! Don't think that!

There is nothing in XML or in XML Schema that makes such
recursion illegal. (And to be fair, note that the error message
you quote from Visual Studio doesn't actually say it's illegal
XML, or a schema error. It says there is something there that
Visual Studio doesn't support.)

I'm not really familiar with all the logic and theory behind xml,
I just thought it was a convenient way to script our UI. I would
like to understand why these "circular references" are not allowed
in the schema, and what alternatives a proper xml guru might suggest
for my situation.

Below is a simple schema that has this problem

I'll take you at your word, though I suspect you did not
paste this in direct from a schema document that got the
same error message you give above. (I suspect it because
the xmlns attribute is missing its required quotation marks,
and you should have gotten a well-formedness error long
before anyone started worrying about circularity.)

<schema targetNamespace="test.xsd"
elementFormDefault="qualified"
xmlns=http://www.w3.org/2001/XMLSchema

er, "http://www.w3.org/2001/XMLSchema"

xmlns:ui="test.xsd">
<element name="test">
<complexType>
<sequence>
<element name="window" type="ui:Window" />
</sequence>
</complexType>
</element>
<complexType name="Window">
<sequence>
<element name="children">
<complexType>
<sequence>
<element name="childWindow"
type="ui:Window" />
</sequence>
</complexType>
</element>
<element name="testString" type="string" />
</sequence>
</complexType>
</schema>

If Visual Studio is complaining about the same problem
I see with this schema, I have to say I'm moderately
impressed with its acuity.

Both XSV and Xerces J believe, as I do, that the schema
document you give is valid, once the xmlns attribute gets
its quotation marks. The only problem with it is that
there are no finite documents which conform to it.
The problem is not the recursion per se, it's the fact
that there is no bottom to it, just as in the following
instructions there is no way to reach step 5.

0 Start. Go to step 1.

1 You have a top-level element named 'test', which has
a required child named 'window'. Go to step 2.

2 The 'window' element, being of type ui:Window, has
required children named 'children' and 'testString'.
Go to step 3.

3 The 'children' element has exactly one required child
named 'childWindow'. Go to step 4.

4 The 'childWindow' element, being of type ui:Window, has
required children named 'children' and 'testString'.
Go to step 3.

5 Done.

If I change the schema so that the childWindow element is
optional, then I can create a finite document which conforms
to the schema. In practice, I suspect you also want to
specify that a window can have more than one childWindow
among its children. One way to do this is to change the
declaration for 'childWindow' thus:

<element name="childWindow"
minOccurs="0"
maxOccurs="unbounded"
type="ui:Window" />

There are other ways, of course.

Is this the problem with your real schema, or is the
absence of the minOccurs and maxOccurs an artifact of
the way you cut down the problem for presentation in
your post?

You asked what alternatives others might suggest. My
first suggestion is to correct the occurrence information.
My second would be to change the name of the child
window to 'window', to make the recursion even more
obvious, unless there is a strong reason to distinguish
top-level windows from others (in which case, I'd
probably use the names 'topWindow' and 'window'
rather than 'window' and 'childWindow' -- but naming
is a matter of taste, and you should suit yourself and
your collaborators, not me).

I hope this helps,

-C. M. Sperberg-McQueen
World Wide Web Consortium

 

[Christine McGavran]

Is that the entirety of the error message?

yes

Do you mean XSV?

I'm not really sure what Microsoft uses, but I don't think it's XSV. The
tool is called "XML Data Proxy Generator."

I don't get any schema error from XSV
on the sample schema you provided
There is nothing in XML or in XML Schema that makes such
recursion illegal.

Then maybe it's a Microsoft issue after all.. I've also posted to the
Microsoft group.

I'll take you at your word, though I suspect you did not
paste this in direct from a schema document that got the
same error message you give above. (I suspect it because
the xmlns attribute is missing its required quotation marks,

You know I did paste it directly, but the document I pasted from has the
quotation marks. Beats me why they got left out... probably some Outlook
thing.

If I change the schema so that the childWindow element is
optional, then I can create a finite document which conforms
to the schema. In practice, I suspect you also want to
specify that a window can have more than one childWindow
among its children. One way to do this is to change the
declaration for 'childWindow' thus:

<element name="childWindow"
minOccurs="0"
maxOccurs="unbounded"
type="ui:Window" />

Doing this doesn't change my results at all.

Is this the problem with your real schema, or is the
absence of the minOccurs and maxOccurs an artifact of
the way you cut down the problem for presentation in
your post?

I knew about the minOccurs/maxOccurs thing, but Microsoft seemed to ignore
them, at least in the old version of DevStudio, so I had left them out of my
test.

My second would be to change the name of the child
window to 'window', to make the recursion even more
obvious

My actual schema (not the test) does this, actually. I was just trying to be
clear for the test.

I hope this helps,

Wish it did. I do thank you for letting me know I'm not trying to do
something entirely weird.

 

[Christine McGavran]

Just to follow-up, I got the response below from the Microsoft group.

I did finally manage to get the smart editor to work on my test schema
without the compile step, so I guess the compile step isn't needed after all
for the editing feature (which is all I really cared about). My real schema
doesn't work yet, so I guess I just rebuild it step-by-step until I figure
out what it doesn't like, when I get some time.

--

that is a drawback of the Data View in vs.net, it would never lend
itself well to this type of implementation.

"The Visual Studio .NET XML Designer applies the Microsoft ADO.NET
schema inference rules to determine the schema information for the
source XML when it tries to generate Data view. The ADO.NET schema
inference rules in the RTM release of the Microsoft .NET Framework do
not permit a single table to be a child table in more than one
DataRelation. "