[
Lists Home |
Date Index |
Thread Index
]
Michael Kay wrote:
>>Granted, dynamic validation (=at runtime) is always possible,
>>but it is
>>certainly not a big help. Errors are discovered after having gone to
>>production, doesn't sound good to me.
>>
>>
>
>Actually, it's an enormous help compared with not doing it at all.
>
>I agree that static checking, where possible, is even better. Saxon doesn't
>do any static checking against a schema yet.
>
>
>
Not even the payware-version?
>>I am sure you will not try to add exact static typing based
>>on Relax NG for XSLT in the next version of Saxon?
>>
>>
>
>No, but I wouldn't rule it out eventually.
>
>
>
>>There are papers that analyze
>>this for subsets of XSLT
>>
>>
>
>Can you give me a reference? I'm not aware of such papers.
>
>
>
A recent survey (not XSLT specific) that comes to my mind is "The Design
Space of Type Checkers for XML Transformation Languages" Anders Møller
and Michael I. Schwartzbach.
*Then there is "Towards static type checking for XSLT" Akihiko Tozawa,
which refers to the original work by Milo, Suciu and Vianu.
That one...
T. Milo, D. Suciu, and V. Vianu. Type-checking for XML
transformers. In Proceedings of the 19th ACM SIGACT-
SIGMOD-SIGART Symposium on Principles of Database
Systems, pages 1122, 2000.
should contain the proof of undecidability.
Most of these say **"type" and mean **regular tree languages. I am not
aware of many papers on XML Schema.
*
>>but doing it exactly, for the full
>>language,
>>is - as far as I remember - undecidable.
>>
>>
>
>Even with XML Schema, it's unprovable that a given stylesheet is guaranteed
>to create valid output (to do strict static typing, in the style of the
>XQuery static typing option, we would have to throw out
>xsl:apply-templates).
>
Not entirely sure whether recursive apply is the central problem: One
might be able to infer what type of the input is from the type of the
<xsl:template match="..."> and the input schema.
However, simple conditionals "if(condition) returnThis else returnThat"
will give the same limitations to type-checking as known from other
Turing complete languages.
Since PSVI infosets will always be in a subclass of the regular tree
languages (i.e. top-down deterministic regular tree language), I believe
it should be slightly easier to type-check against schemata than to
type-check against Relax NG. Of course, the treatment of substitution
groups might balance this out again...
>But pragmatically, I think it's possible to give
>static warnings for a good proportion of user errors. And I suspect that
>most of the heuristics to do this are equally applicable to DTDs or Relax
>NG.
>
>
>
The mentioned papers talk about "backwards type inference" and also
about approximating types conservatively(which might reject well-typed
stylesheets). The user then has to annotate (like with casts)
stylesheet fragments with type information when she is sure that her
stylesheet works out, and willing to "disable" type-checking in this way.
Hope this helps,
Burak
--
Burak Emir
http://lamp.epfl.ch/~buraq
|
