It's not just a serialization issue. The data model is different; a tree 
consisting of a root element owning lots of row elements owning lots of cell 
elements each of which contains a text node which is the string representation 
of a number is not the same as a two-dimensional array of numbers.

Michael Kay
Saxonica

On 3 Feb 2014, at 10:16, jean-marc Mercier <[email protected]> wrote:

> Michael, you're right, and I do now understand that an efficient 
> serialization / deserialization of a matrix structure could be more 
> straightforward using JSON than XML.
> However, without considering this potential overhead due to serializing / 
> deserializing procedure, that should be or order "epsilon" comparing to 
> linear algebra algorithms complexity, once loaded into memory, I don't think 
> that a XQUERY coded matrix algorithm will be more efficient using XML or JSON 
> as serialization choice. 
> 
> Am I right ?
> 
> 
> 2014-02-02 Michael Kay <[email protected]>:
> 
> On 2 Feb 2014, at 20:33, jean-marc Mercier <[email protected]> wrote:
> 
> > N-dimensional representation of arrays are quite straightforward with XML 
> > too. Is there any incentive to expect better performances with a JSON 
> > matrix representation rather than an XML one ?
> >
> 
> I think that if you had an XML schema for an XML representation of 
> N-dimensional arrays, and if the XPath processor recognized that schema and 
> used a custom tree representation for its instances, then arrays could be 
> represented using XML just as efficiently as using JSONiq arrays. But if you 
> use a general tree representation that allow any element names, namespaces, 
> base URIs, mixed content, attributes, and all the other paraphernalia of XML, 
> then it is likely to be significantly less efficient.
> 
> For example:
> 
> * XML is text-oriented, and using XML for numeric values invariably involves 
> string-to-number conversion, which is expensive
> 
> * Numeric subscripts when addressing XML (as in para[3]) are likely to have 
> O(n) performance rather than constant performance, because the tree structure 
> is likely to be optimized for scanning all the children rather than locating 
> an individual child by its index.
> 
> Michael Kay
> Saxonica
> 
> 

_______________________________________________
[email protected]
http://x-query.com/mailman/listinfo/talk

Reply via email to