Re: Reflection on the special telco of CSVW

On 10 Sep 2014, at 14:09 , Andy Seaborne <andy@apache.org> wrote:

> On 10/09/14 12:42, Dan Brickley wrote:
>> On 10 September 2014 12:27, Ivan Herman <ivan@w3.org> wrote:
>>> 
>>> On 10 Sep 2014, at 12:21 , Andy Seaborne <andy@apache.org> wrote:
>>> 
>>>> One aspect of this choice is whether a transformation of a CSV file to be published on the web so other people (other than the data publisher) run it?  Or is it the input for a toolkit to generate format X and then a file with format X is put on the web?
>>>> 
>>>> If transforms are published, then there is a requirement for a programming-language, template-language independent solution.  I agree this is more work.
>>> 
>>> But the point is that such programming languages already exist. Several of them. One could say that we should not define yet another one.
>> 
>> I have the feeling that we're talking past each other some of the time
>> here. Reading Andy here, my first reading was that requiring a
>> "template-language independent" template/transform language seemed
>> like a contradiction. But I guess the idea is that we want portability
>> between specific template/transform software packages / projects. So
>> for example, emphatically blessing something as specific as Django
>> just wouldn't make sense for W3C, even though we might want to make it
>> possible in metadata for the Django-loving subset of the Web community
>> to declare associated templates via JSON-LD metadata.
> 
> Yes.  The enviroment affects the "hook".
> 
> Ivan said:
> """
> and add the hooks for further processing just like in Alternative2
> """
> 
> To a user, the whole process of CSV to their desired output is the task.  Not just some partial step.
> 
> We discuss {{foo.lowercase}} to mean make a call to lowercase the field.  That .lowercase is, in some proposals, as I understand them, a call to prog language-X.


That was not what I meant. What I mean was something like:

metadata...

...   template: {
	uri: http://ex1...
	format: "turtle"
      }
      transformation: {
	uri: http://ex2...
	format: "sparql"
      }

(Both are optional in the metadata.) Usage being:

  - templates are used to generate turtle (RDF) using whatever template language we define. It is an issue whether 'lowercase' is part of the language, I would prefer *not* to be the case.
  - *the result* of templating *may* be transformed, as a second step, using, in this case, a sparql transform

The template should/must be language independent. Should be simple, ie, such features as lowercase is, in my view a "no-no"

(Axel, I believe that is also an answer to you.)

Ivan

> 
>> 
>> The {{ mustache }} -based notations seem to be common enough that they
>> transcend any particular software project and programming language.
>> But there are some {{ }}-based template systems that slip rather
>> casually into assuming full Javascript capabilities, which is
>> something we ought to be very wary of doing. There is an interesting
>> design space between simple variable interpolation vs turing complete.
>> 
>> e.g.  http://www.polymer-project.org/docs/polymer/expressions.html
>> ("Polymer supports expressions in {{}} with a strict subset of the
>> JavaScript language. In order to use this feature, it’s important to
>> understand its behavior and limitations: The goal for inline
>> expressions is to allow the expression of simple value concepts and
>> relationships. It is generally bad practice to put complex logic into
>> your HTML (view).").
>> 
>> 
>>> If a template format is defined (complex or simple), then one can also publish the templates (e.g. [1,2]).
>>> 
>>> In other words, I am not sure I understand your point in terms of deciding whether we do templating or not.
>>> 
>>> [1] https://github.com/w3c/csvw/blob/gh-pages/experiments/simple-templates-jquery/simple_test/test-json.tmpl
>>> [2] https://github.com/w3c/csvw/blob/gh-pages/experiments/simple-templates-jquery/simple_test/test-turtle.tmpl
>>> 
>>> 
>>>> 
>>>> Assuming javascript is a possibility; while it is arguably the safest single choice, it does not work for many environments.  If you're in a lang-X programmer (e.g. R), you want to use lang-X skills.
>>>> 
>>>> Otherwise, if it's a tool-input and not published to be run elsewhere, it does not need this portability requirement.  A language or a basic-transform+improve style is more reasonable.  The tool space is weaker (transforms are tool specific).
>>>> 
>>> 
>>> 
>>> I do not understand. *If* we define a template language (simple or complex), it can be defined in different languages. I happened to have that done in Javascript, but it could have been done in Python without too much problems.
>> 
>> Ivan - when you say "it can be defined in different languages", can I
>> read "can be implemented" in different languages? Your implementation
>> seemed to be of a mustache-family language that *happened* to be
>> implemented in Javascript. I would hope that the same templates could
>> equally well be processed by Java, Python etc. implementations, and
>> that a group like ours should be capable of writing unit tests to
>> probe the behaviour of such implementations.
>> 
>> Dan
>> 
> 


----
Ivan Herman, W3C 
Digital Publishing Activity Lead
Home: http://www.w3.org/People/Ivan/
mobile: +31-641044153
GPG: 0x343F1A3D
WebID: http://www.ivan-herman.net/foaf#me