Hi Eelco:

I just encountered your very interesting OOPSLA 2004 paper.

Our 3D OOPSLA 2003 paper seems closely related. Do you agree with 
the comparison below?

-- Karl
============================================
Concrete Syntax for Objects
Domain-Specific Language Embedding and Assimilation without Restrictions
by Martin Bravenboer and Eelco Visser
OOPSLA 2004

MetaBorg falls into the strategic programming line of work.

In the Demeter world we have used concrete syntax for objects
since 1988
http://www.ccs.neu.edu/research/demeter/biblio/class-dictionaries.html
The new work has an assimilation component that is new
in its generality.

In Demeter we would write:
JPanel panel = Jpanel.parse("some string");

while in MetaBorg we can write:

JPanel panel = panel of border layout {
  north = label "Welcome"
  ....
}

with the compile-time checking advantage.

It is refreshing to see that this paper says that we use the wrong tools
like JavaCC (or ANTLR). They stress the importance of scannerless parsing.

Our work closest to MetaBorg is our OOPSLA 3D Track paper in 2003:
http://www.ccs.neu.edu/research/demeter/biblio/XAspects.html

The Metaborg method has the following characteristics:

Syntactic:

Embedded code fragments are checked syntactically at compile-time.
Also the case in XAspects if we introduce a new aspect (e.g., traversal aspect).

No restrictions on syntax definition:

XAspects is LL(k) based.

Interaction with host language:

Embedded code fragments should be able to refer to artifacts in the host program
and vice versa:

XAspects also allows such interaction in both directions.
For example, a Traversal aspects refers to classes and method
bodies refer to the Traversal aspect.

Combination of extensions:

It is possible to combine multiple domain notations.
XAspects also has this property.

No restrictions on assimilation:

The translation of embedded fragments to the host language should
not be limited.
It should allow global analysis.
In XAspects, translation of the traversal fragment requires global
analysis of the class graph (as an example).

Three of the four Metaborg properties mentioned also apply to XAspects.
It would be very useful to have a better understanding of when
Metaborg and when XAspects should be used.

One key difference between Metaborg and XAspects is that in XAspects
we don't write the embedded fragments into the Java program but into
a separate aspect. The aspect defines a new entity (e.g., a method)
that is referred to in the Java program.

How can we implement XAspects using Metaborg? What are the benefits?

-- Karl