[aosd-discuss] Is return statement ignored in advice?

[Bram Adams]

Hi,

Op 17-aug-07, om 19:35 heeft Gregor Kiczales het volgende geschreven:

> In my experience, confusion about this comes up when people are  
> thinking
> about AspectJ advice using a 'pre-processor' or 'code injection'  
> semantic
> model.

Not necessarily. At the SPLAT workshop during AOSD '07, we presented  
a paper about an AOP implementation of idiom-based exception handling  
in a huge C system (http://www.aosd.net/workshops/splat/2007/papers/ 
adams.pdf). Our work was based around the notion of "local  
continuation join point", i.e. a new kind of join point representing  
the future of a program after conclusion of another join point J, but  
constrained to J's enclosing execution join point.

Consider a call to procedure g within a procedure f. The local  
continuation join point L of  the call to g models everything which  
will happen inside f after the call to g returns. If one advises L  
with around advice, then one is able to control whether the execution  
of g is completed or not. In the latter case, one effectively stops  
the execution of a procedure from within advice, but in a more fine- 
grained way than around-advising a normal call or execution join  
point allows. Major application of this is exception handling in C,  
because aspects enhanced with local continuation join points can  
express this in an easier to understand, more high-level way than  
e.g. setjmp/longjmp. At the same time, implementation of the local  
continuation join points is rather straightforward and can be made  
efficient (http://users.ugent.be/~badams/aspicere2/).

How does all this relate to this email thread? The combination "local  
continuation join point + around advice" can actually be expressed  
very succinctly by introducing a new keyword inside advice, inspired  
by "break". If this statement is reached, both the advice as well as  
the advised call join point's enclosing procedure are aborted (one  
does not need to write a separate advice anymore on the call join  
point's local continuation joinpoint). The guys of ACC (http:// 
aspectc.msrg.utoronto.ca/ACC) applied this idea with their preturn()  
construct, which is a return statement stopping both advice and base  
procedure.

Bottom line: attaching more semantics to a return statement inside  
advice makes sense from a pure AOP perspective (as opposed to a  
program transformation kind of view), especially in (but not limited  
to) older languages like C.

Kind regards,

Bram Adams
GH-SEL, INTEC, Ghent University (Belgium)