W3C home > Mailing lists > Public > www-forms@w3.org > August 2000

Data Model vs. Data Validation

From: Chris Hubick <chris@hubick.com>
Date: Fri, 25 Aug 2000 11:23:36 -0400
Message-ID: <012401c00ea8$71717750$64dc4e0c@chlap>
To: "XForms" <www-forms@w3.org>
I have spent the last year and a half building a proprietary
architecture which shares many design goals with XForms.  Our
system is written in Java and uses XML bindings for constructing
controls and validations upon them.  So naturally, when I learned
of XForms a few months ago, I became very interested.  I would
naturally like to take advantage of the synergy and migrate our
system to track the specification.  I have recently started to
take a closer look at just how I would do this, and it has raised
some concerns.

It is stated that XForms will be split into three layers, Data,
Logic, and Presentation.  Naturally I think this is fantastic.

My issue is that there seems to be no distinction/separation
between the "Data Model" and the "Validation" upon that model.
These are decidedly different concepts and should be separated
out.  I would argue that these validations should be part of the
"Logic" (or some other?) layer, rather than the part of the data
model itself.

<number name="count" min="1" integer="true"/>

It is up to the data model to specify how one defines a number,
as above.  An attributes such as "min" is not part of the data
model, it is at a higher level.  The "min" attribute works /with/
a number..."min" in no way actually defines what you are working
with /as/ a number.

The data model should be just that, a data model...it should
restrict itself to defining the datatypes.  This can be a fine
line.  I don't think masking, for example, is a validation...I
view a mask the same way I do the "integer" attribute above...as
'subclassing' a general type into a more specific type.  For
example, I would view a phone number mask as /defining/ the
pattern of digits that constitutes a phone number...a validation
would be something that says something like "the area code must
be 204"..it works /with/ the phone number, but does not define
what a phone number is, and should thus be separate.

Mixing data and validation leads to problems later on when you
want to add more (complex) validations.  I initially made the
same mistake in the system I built.  I have to deal with
validation such as "If the user selects 'item 1' from the
'droplist field', the 'number field' must be 'between 1 and 5'.
If they select 'item 2' from the list, the number must be between
'6 and 10', etc, etc".  I quickly got to the point where the form
author needed the ability to add custom form specific
validations.  At this point is where you start to see the need
for an independent and extensible way to define validations.  And
the best thing of all...we already have a basis for this...it's
called MathML....

For the above number example:

<apply>
  <min/>
  <ci>count</ci>
  <cn>1</cn>
</apply>

or instead of:

<string name="foo" min="5"/>

you could:

<apply>
  <min/>
  <apply>
    <strlen/>
    <ci>foo</ci>
  </apply>
  <cn>5</cn>
</apply>

Doing things like this allows one to build significantly more
complex validations...and you can create custom ones just as you
create custom functions in MathML.

---
Chris Hubick
mailto:chris@hubick.com
http://www.hubick.com/
Received on Friday, 25 August 2000 11:23:34 GMT

This archive was generated by hypermail 2.2.0+W3C-0.50 : Saturday, 10 March 2012 06:21:48 GMT