intertwingly

Don Box: Today, the data needed to initialize an object graph is hidden behind imperative statements in your programming language of choice. That makes it difficult to tease out of the rest of your program without weird markers in the source code to delimit the structure of your code into recognizable pieces.

I buy this part.  I don't buy the Man vs Machine angle.  If you want to write some queries over this data, the schema had better be understandable by mere humans.  And if you want to write a transformation, you had better be prepared to wallow in the instance data, as that's how you will have to debug it.

My previous post on this subject had more to it, but I chickened out and didn't post it.  What the heck, I could be wrong, but... here's essentially what I said.

What I can piece together is that somebody was designing a visual composition editor targeting the traditional VB crowd.  And ran into limits in the ability to round trip generated source code that had been modified by humans.  Attributes helped, but weren't enough.  What was needed was some real structure.

So, that person chose to serialize that data separately.  Such is not new, other visual composition editors have done this for years.  However, given that it is 2003, the choice was made to serialize that data as XML.  This too has been done before.  Furthermore, I gather that that person wasn't exactly a connoisseur of XML, but merely viewed it as just another data format. 

However, an inevitable consequence of having chosen an XML format is that the development team will receive a lot of "help" in designing the format.  Both from within Microsoft and from the peanut gallery.  Whether they want it or not.

I've been on both sides of that particular equation.

I could be wrong about some or all of this.  Even if I am largely right, I can never be 100% sure.  However, if my guesses are correct and some of the internal discussions are going the way I expect that they have been, all I can say to Don is:

You go girl.