Re: [xliff-comment] Section 2.5.4 Validating Documents with Extensions - needs revision

["Asgeir Frimannsson" <asgeirf@gmail.com>]

Doug, all,

(for us reading the comments list, this is a reply to
http://www.oasis-open.org/archives/xliff/200610/msg00027.html )

On 10/26/06, Doug Domeny <ddomeny@ektron.com> wrote:
> After analysis, I now recommend processContents="strict" for the strict
> schema. All modified files are attached.
>
>                         strict         lax w/ xsd      lax w/o xsd
>                     valid  invalid    valid  invalid    valid
> StylusStudio 2006  correct correct   correct valid     correct
> MSXML 4 DOM        correct correct   correct valid     correct
> MSXML 6 DOM        correct correct   correct valid     correct
> Saxonica 8.7       correct correct   correct valid     correct
> .NET XML           correct correct   correct correct   error
> Xerces-J 2.5.1     correct correct   correct valid     correct
> XSV 2.10-1         correct correct   correct valid     error†
>
> † if URI returns error 404
>
> As you can see, "lax" for most of the parsers is the same as "skip" and
> therefore useless except in .NET parser, which acts like "strict".

I'm not sure these tests serves 'lax' the full justice it deserves.
This might be because you are using StylusStudio to validate files
using the above validators?

I did some further investigation and came up with 16 test cases based
on the following 4 (boolean) attributes:
1) Invalid Content - If the <any> type contains elements defined in
foreign schema 'other' and the contents of the element is not valid
according to its specification.
2) Undeclared Elements - If the <any> type contains elements *NOT*
defined in foreign schema 'other'.
3) schemaLocation - If the XML contains a schemaLocation hint for the
foreign schema 'other'
4) Schema available - If the foreign schema 'other' is available in
the file system path

Please see the attachment for the test data and results (CompareValidators.xls).

Before starting this test, I had a hypothesis that attribute 2
(undeclared elements) above should not affect 'lax' validation, and
the majority of validators tested proved this true. If an element
which is not declared in a foreign namespace occurs in an <any> type,
even if *a* schema for this namespace is resolved, it will pass 'lax'
validation. This seems to be the common interpretation of the Schema
specification. I was however unable to find a good reference to a
discussion on this. I have a personal theory that this is because of
the way namespaces are resolved - they do not according to the Schema
spec have to be from a single file or location, a schema can e.g. use
the <xsd:include /> element and if this second file is not available,
the xml processor would 'laxly' ignore this and still validate 'if it
can'. But as I said, this is just a theory.

Some notes:

XSV:
 - Produces an error when schemaLocation is available for a namespace
processed with 'lax' when the schema file could not be found in the
file system. This is most probably a bug in the validator, or a rather
strict interpretation of the schemaLocation attribute.

Validating in StylusStudio:
- The tool does not report schema validity errors correctly, as it
simply prints out all messages and does not distinguish between
informational messages, warnings and errors. It is practically
impossible to interpret if a schema is valid or invalid based on this.
This is evident particularly in the xerces and xsv validator, as I
have these tools also outside of StylusStudio. I cannot comment on the
.NET validator used in StylusStudio, as I do not have a .NET 1.x
validator readily available - but I suspect some of the 'errors' in
this validator are really warnings, hence the question mark in the
test results.

I hope this helps clear up a few things, and bring my good friend Lax
some justice.

regards,
asgeir

--
Asgeir Frimannsson
PhD Candidate
School of Software Engineering and Data Communications
Queensland University of Technology
126 Margaret Street, Level 3
Brisbane QLD 4001, Australia

Phone: (+61) 7 3138 9332 Mob: (+61) 405 412 696
Email: a.frimannsson@qut.edu.au




-- 
Asgeir Frimannsson
PhD Candidate
School of Software Engineering and Data Communications
Queensland University of Technology
126 Margaret Street, Level 3
Brisbane QLD 4001, Australia

Phone: (+61) 7 3138 9332 Mob: (+61) 405 412 696
Email: a.frimannsson@qut.edu.au

schema validation.zip