Hi Johan:

handling dynamically changing class graphs in a distributed environment is
difficult. But companies already must deal with this: they need
continuously running applications
while changing their business models.
(for example banks) 

Mitch was telling me that you have a Java code generator for traversal
automata. Will it be possible to use it as part of Demeter/Java
for strategic traversal automata?

-- Karl

>From johan@ccs.neu.edu Sat May  9 16:45:15 1998
>To: Karl Lieberherr <lieber@ccs.neu.edu>
>cc: dem@ccs.neu.edu
>Subject: Re: Prting demeter to python 
>Date: Sat, 09 May 1998 16:45:10 -0400
>From: Johan Ovlinger <johan@ccs.neu.edu>
>
>Karl pontificates:
>  I think that it would also make sense to have a dynamically 
>  changing class graph together with TAO while using general
>  traversal strategies. I think positive constraints are more
>  natural.
>
>This is at best really tricky to do. 
>
>Claim: if you want positve constraints, the traversal expander MUST
>       know the entire class graph. 
>
>There are 3 ways I can think of to figure this out:
>1) the user tells it, via a special file. This kind of disqualifies
>   the 'dynamic' claim.  
>2) We instrument the set and constructor methods for every class to verify and if
>   needed update a central classgraph somewhere. One problem here is
>   versioning of classes, but that seems more bookkeeping than
>   difficult. Details details. Getting it right is going to be hard.
>   This approach of course has the problem that a set can force
>   recomputation of all traversals, even those in progress. Forget any
>   real time constraints (this I think is bad news for the intended uses
>   of this technology, karl?). 
>3) search the object graph. Yuck!
>
>I point out that Mitch's & my traversal automata are independet of
>classgraphs, and so seem to be a better candidate for situations with
>dynamic class graphs.
>
>Johan
>