Re: [xml-dev] XML data interchange format: Flatter is better

[Thomas Passin <list1@tompassin.net>]

On 10/27/2014 1:37 PM, Costello, Roger L. wrote:
> Hi Folks,
>
> Designing data interchange formats can be quick, easy, and
> inexpensive.
>
> The comments made in this message are intended for this environment:
>
> The XML is distributed to a broad spectrum of consumers, each
> consumer might perform widely different operations on the XML.
>
> I recommend making flat XML. Design your XML to just have a root
> element, containing a linear sequence of elements.
In my experience, working with a flat list of data elements like this is 
much harder than you make out.  That's because you have to create and 
track a lot of state for yourself that otherwise the xml (or xslt) 
parser could do for you.  You have to figure out, and write code for, 
the start and end of grouped sections.  In your vineyard example, "lot 
1" data ends when you spot the next <lot-number> element.  Nothing else 
indicates what lot the ripe-grapes, picker, etc are associated with.

So the first thing your code is probably going to do is to associate the 
lots with their ripe-grapes, etc.

If you don't need such an association, if you don't need to associate 
them with their lots, you can still pull out that information with ease 
from more hierarchical xml, using e.g. xslt.  OTOH, the more 
hierarchical levels there are in the unflattened xml, the harder it is 
to work out their relationships when you receive the flattened data.

To put it another way, when you flatten a hierarchical structure, you 
throw away knowledge that you already know about the data.  The data 
user will have to reconstruct some or all of it, which may or may not be 
feasible but will usually be harder than necessary.

Of course, in real life there's usually  a place for everything, so 
there could be some cases for which a flattened design would be good. 
But not in general.

So I disagree with your recommendation.

> Be a markup minimalist.
>
> Here are the reasons for my recommendation:
>
> 1. Consumers of flat XML can apply powerful parsing techniques to the
> linear sequence of elements. Thus, consumers can add whatever
> structure is appropriate for their particular applications to process
> the data efficiently.
>
> 2. XML Schema design and implementation of flat XML is trivial:
> simply create an XML Schema with a sequential list of element
> declarations.
>
> 3. If, on the other hand, you were to design the XML to have lots of
> structure, it is a near certainty that the structure will not be
> suitable for many of your consumer's applications. Further, its
> structure is likely to hamper powerful parsing techniques.
>
> The "flatter is better" philosophy may be summarized this way:
>
> I can't predict how my consumers will want the data structured, so I
> won't try to predict. I will let them apply their own structure to
> the data.
>
> Let's take an example. Suppose that you want to model a grape
> vineyard, with pickers scattered about on the various lots. The
> following XML is not flat. It's probably a design that most people
> would come up with. I assert it is a bad design.
>
> <Vineyard> <Lot id="1"> <ripe-grapes>4</ripe-grapes> <Picker
> id="John"> <metabolism>2</metabolism>
> <grape-wealth>20</grape-wealth> </Picker> </Lot> <Lot id="2">
> <ripe-grapes>3</ripe-grapes> </Lot> ... </Vineyard>
>
> That design is well-suited to operations such as this:
>
> What Pickers are on lot 23?
>
> But it is horrible for operations such as this:
>
> Move Picker John to Lot 2.
>
> Don't design XML that way. Design XML to be flat, like this:
>
> <Vineyard> <lot-number>1</lot-number> <ripe-grapes>4</ripe-grapes>
> <picker>John</picker> <metabolism>2</metabolism>
> <grape-wealth>20</grape-wealth> <lot-number>1</lot-number>
> <ripe-grapes>3</ripe-grapes> ... </Vineyard>
>
> That's a beautiful design. It enables powerful parsing techniques to
> be applied to it. For instance, one consumer may parse it to generate
> the above structuring. Another consumer may parse it to generate this
> radically different structuring:
>
> <Lot id="1"> <ripe-grapes>4</ripe-grapes> </Lot> <Lot id="2">
> <ripe-grapes>3</ripe-grapes> </Lot> <Picker id="John" locatedOn="1">
> <metabolism>2</metabolism> <grape-wealth>20</grape-wealth> </Picker>
> ...
>
> And another consumer may parse it to generate still another
> structuring.
>
> Each consumer parses the flat XML to create a structuring that is
> well-suited to their particular application processing.
>
> Are you creating XML "for the long haul"?
>
> Are you creating XML "for a broad, diverse set of clients"?
>
> Are you a manager and don't want to dump a lot of time and money into
> creating "the perfect XML Schema design"?
>
> Then create flat XML.
>
> Flatter is better!
>
> For more info on parsing flat XML see my recent posts:
>
> Recursive Descent Parsing for XML Developers:
> http://lists.xml.org/archives/xml-dev/201410/msg00017.html
>
> Bottom-up Parsing for XML Developers:
> http://lists.xml.org/archives/xml-dev/201409/msg00016.html
>
> Comments welcome.
>
> /Roger