W3C home > Mailing lists > Public > public-forms@w3.org > March 2008

Pros and cons of different solutions: three slightly different versions of the PO example

From: Erik Bruchez <ebruchez@orbeon.com>
Date: Wed, 5 Mar 2008 10:46:17 -0800
Message-Id: <D8FEBB62-24EC-4090-AC6F-E60C39C8A01E@orbeon.com>
To: "Forms WG (new)" <public-forms@w3.org>
All,

Following our discussion today about simplified syntax and how you
refer to control values in expressions, I attach 3 different examples
of a simplified the view of the PO order John sent.

Note that I have not considered so far the issue of submission or
insert/delete.

I think it is primarily important to consider what the code looks like
to the form author. It has to be easily understandable. After all,
that is our primary goal in simplification.

The 3 files look almost entirely the same, except for the way you
refer to elements. But this has subtle and potentially far-reaching
implications:

1. By referring directly to the implicit data model's nodes
-----------------------------------------------------------

The benefit is that it's super-plain XPath. The drawback is that you
need, as a form author, to be aware of:

* The implied structure of the underlying data model.

* "where" you currently are in that model, i.e. a notion of what the
   current XPath context is.

So is this too much XPath knowledge to ask from form authors for a
simplified syntax? For us old-timers it is not that difficult, but for
simple use cases like John's PO it is disappointing to require that
much from the form author.

2. By using a "control()" XPath function
----------------------------------------

This functions allows referring to the value associated with the
control, specified by name (or id, see below). Note that I say
"associated", which means that it could be the actual value of the
control, or it could be pointing to instance data through implicit
binds. The point is that it doesn't matter much as we define what
"control()" means.

Note that control() could return a node-set (or even a sequence of
values in XPath 2.0) when the control is within a repeat. This allows
the sum() function to work as expected.

The big benefit is that you don't worry about the current XPath
context anymore. It is unambiguous to which control you are referring
to, assuming you use unique control names.

This would be easier to enforce if we actually referred to the control
by id instead of name. In HTML, there is a distinction between the two
though, where the id is, well, the id of the element, and the name
defines something in the data model (name/value in HTML, element name
in XForms). But it is a pain to have to define a name AND an
id. Anyway, this is probably something that can be figured out.

3. By using XPath variables
---------------------------

The use of variables (i.e. the $foo syntax) has the same benefits as
the "control()" function.

The benefit over the "control()" function is that the syntax is
simpler, and you use a well-understood XPath construct instead of a
custom function.

Here my main worry is that of scoping and future-proofing. It is
important IMO that we do not mess up how we use XPath variables in
order to leave open the specification of locally-scoped variables in
the future.

But:

* IF <xf:input name="foo"> implies an <xf:bind>
* AND that <xf:bind> is considered in scope where the variable is used

THEN the scoping rule becomes palatable. Certainly, it will be if you
look at the corresponding canonical XForms.

But this still raises a few questions:

Q1: When writing canonical XForms, do binds in a model always define
     XPath variables, and if so, do they expose them by id?

The issue here is that ids are global in a document, while currently
in XForms models are scoped with the @model attribute. If you don't
use an id, then you can have variables with the same name defined in
multiple models. This enhances reusability.

In most languages, variables can be scoped in a similar way. The same
I think would be desirable for XForms, therefore using global ids to
expose variable names is a serious drawback IMO (even though
currently, binds are exposed this way).

Do we need to introduce <xforms:bind name="foo"> ?

Q2: Do we want to call our future "variables" "binds" instead?

I like the idea of calling a cat a cat and naming variables
<xforms:variable>, as this is a notion familiar to every developer on
the planet and every mainstream programming language has them and
calls them "variables".

Binds are familiar to long-time XForms people, but not so much to
programmers who come from other languages.

So the day we introduce locally-scoped variables a la eXforms [1],
will we call them <xforms:variable>, or <xforms:bind>? You know what
my vote is.

Q3: Further, a variable in XPath has a name and a value. Binds do not
     quite work this way.

Instead, they are currently defined to bind to one or more nodes (and
possibly, when using the @calculate MIP, assign a value to the
node(s)).

This does not quite map in my mind to the notion of a variable
today. Yes, we can define the behavior of <xforms:bind> as exposing
the value of the node as the value of the variable, as a node-set for
example.

But that is not enough for generalized variables. So how do we get
there? Certainly, we could modify <xforms:bind> to get closer to the
XSLT notion of <xforms:variable>, as a holder for any XPath type. But
then why not introduce <xforms:variable> instead?

One solution could look like this:

* Only <xforms:bind> elements having a "name" attribute expose
   themselves as variables. This solves the issue of scoping by model.

* We do not extend <xforms:bind> in the future to look more like
   variables. Instead, we introduce <xforms:variable name="foo"/>. So
   you can expose variables to XPath either with binds, or with actual
   variables which are more generic and more author-friendly IMO, when
   actually used as variables. This is especially important for
   locally-scoped variables, whether in the model, in the view, or in
   actions.

The bottom line for me at this point remains that I still have quite a
few concerns about the use of the $foo syntax. I believe that we need
to fully understand the implications of this before going forward with
this solution.

BUT if we find a consistent, future-proof solution for the $foo syntax,
then I would favor it over other solutions because the benefits are
so appealing.

-Erik

[1] http://www.exforms.org/variable.html

--
Orbeon Forms - Web Forms for the Enterprise Done the Right Way
http://www.orbeon.com/




Received on Wednesday, 5 March 2008 18:46:56 UTC

This archive was generated by hypermail 2.3.1 : Tuesday, 1 October 2013 22:06:47 UTC