Re: For loop over a collection, multiple

[John Snelson]

Martin Probst wrote:
>> There are plenty of essential use cases of this kind (most of the ones I see
>> on a daily basis...)
>>
>> In the absence of such a scripting extension you have to go back and forth
>> between XQuery and other language (Java, Ruby, etc). That's what people
>> do,  that is very expensive in productivity and performance (thanks the
>> wonderful XML APIs..), and not maintainable.
> 
> I wouldn't agree. I have experimented several times with writing
> complete applications in XQuery, and it always feels like writing PHP
> code in 1999. No polymorphy, no abstraction mechanisms, no data types,
> and - to be quite honest - a relatively clunky syntax for a general
> purpose programming language. All of those are not problems when using
> XQuery as a query and/or XML manipulation language.

The higher order functions support in XQuery 1.1 should make a massive 
difference to that. It also allows you to create "sequences of 
sequences", and to therefore build data structures. I've already built a 
red/black tree using it, which can be used for sorted sets and maps.

> Of course it depends on your system setup, but you neither have to pay
> a performance penalty for moving from XQuery to another language, nor
> do XML APIs have to be ugly; at least for the part that you would do
> in an imperative language.

But nothing is as simple as XQuery's tight integration with XPath, and 
straightforward inline XML construction.

> I also wouldn't agree with the maintenance thing. The worst
> maintenance nightmares I've seen were usually written in a language
> that wasn't fit for the purpose, thus all kinds of problems ensued.
> Rather use the right tool for the job in a polyglot programming
> environment.

I don't disagree. It's just that there's a lot of applications who need 
to manipulate XML, query databases and/or produce XML/HTML. In that 
case, as many eXist and MarkLogic customers are finding out, XQuery is a 
great fit.

>> That's WAY worse then an XQuery scripting extension: at least XQuery-SE
>> programs can be compiled and optimized globally, but if the logic is split into
>> multiple languages, it's much harder, if not impossible to optimize globally.
> 
> That is actually a big concern for me, that scripting will make it
> impossible to optimize and compile XQuery programs globally. I
> probably don't have enough insight into the spec right now, but it
> seems to be like you wouldn't be able to do any reordering, shoving
> conditions around etc beyond the scope of a single sequential
> expression, would you?

Sure - it will be hard to optimize the imperative logic, just as it's 
hard to optimize Haskell's monads, or Scheme's assignments. But then 
again, 95% of code isn't on the critical path and doesn't need optimizing.

I don't want to sound too dismissive - I think optimization and 
parallalization is really important. But so is interacting with the real 
world and the 4th dimension. XQuery will always have at it's core a 
declarative functional language which will give users the tools they 
need to write optimizable code, and a query origin which I hope will 
encourage the solving of problems in a functional manner.

>> I do not know a single functional programming language that didn't get into
>> "dirty" scripting extensions when it reached a certain audience, and a
>> certain maturity.
> 
> Haskell? Depends on how you see Monads, I guess.

Monads are imperative, no question. In fact, I'm almost certain you 
could prove that XQuery Scripting Extensions is equivalent to 
introducing a monad called "sequential".

John

-- 
John Snelson, Oracle Corporation               http://twitter.com/jpcs
Berkeley DB XML:            http://oracle.com/database/berkeley-db/xml
XQilla:                                  http://xqilla.sourceforge.net