[Ur] More questions: sql_injectable, reactivity and interaction with CSS

[Adam Chlipala]

Artyom Shalkhakov wrote:
> First, Is there any way to write an sql_injectable instance?

Yes, but only by building on the instances included in [Basis].

> I'd like
> to serialize and deserialize a tree data structure (xbody in
> particular) to store it in the database.
>    

The Ur/Web version in the public Mercurial repo includes a new type 
family [Basis.serialized], such that any [serialized t] is 
[sql_injectable].  See basis.urs for the details.  I should make a new 
release in the next few days, and it will contain this functionality 
(and you can use the repo's version for now, if you can't wait).

> So is Ur "reactive" in the sense that modification of
> sources is automatically propagated to every dependent dyn tag?
>    

That's right.  None of the event-y stuff associated with traditional FRP 
is included.  Perhaps I should change the way I describe these features 
in documentation.

> I'm trying to write a simple in-place editor for text. Given some text
> on a webpage, users can double-click on a paragraph, say, which turns
> the text into a text box and two buttons (apply and cancel). I don't
> know how to do this.
>    

Well, you could, say, create a [source bool] that is changed by the 
[onclick] handler of the paragraph.  The paragraph content would be a 
<dyn> that does different things based on the source's value.

> Third, what is the approach to styling elements? ATM, the only way to
> communicate with CSS is via classes, but is full CSS embedding worth
> it?
>    

I think many would say that it is "good programming practice" to 
encapsulate all styles inside classes, and so we have at least that one 
benefit of the present architecture. ;-)

Supporting embedded CSS safely would be non-trivial, as CSS supports 
executable code (for instance, in a property like url(javascript:...)).  
Support for 'style' attributes is not on my near-term to-do list.

> And finally, can you compare Ur to OPA of<www.mlstate.com>?
>    

I think you can get a good sense for the two languages' aesthetic feels 
by comparing the demos that are online.  That is my only source of 
information on OPA, as I have not tried it myself.

I believe OPA is more or less "an ML," while Ur has a much more 
expressive type system.  I first started working on Ur to support new 
kinds of metaprogramming, most of which are impossible in OPA or any 
other essentially ML-like language.  OPA also seems to lack support for 
relational programming, which Ur/Web gives through interfacing with an 
external DBMS.

There are probably some other important differences, but MLstate is 
being too secretive thus far for me to find them easily.