From lieber@ccs.neu.edu Thu Jan 14 16:23:29 1999
Received: from stockberg.ccs.neu.edu (lieber@stockberg.ccs.neu.edu [129.10.116.114])
	by amber.ccs.neu.edu (8.9.1a/8.9.1) with ESMTP id QAA21895;
	Thu, 14 Jan 1999 16:23:28 -0500 (EST)
Received: (from lieber@localhost)
	by stockberg.ccs.neu.edu (8.9.1a/8.9.1) id QAA20375;
	Thu, 14 Jan 1999 16:23:27 -0500 (EST)
Date: Thu, 14 Jan 1999 16:23:27 -0500 (EST)
From: Karl Lieberherr <lieber@ccs.neu.edu>
Message-Id: <199901142123.QAA20375@stockberg.ccs.neu.edu>
To: johan@ccs.neu.edu, lieber@ccs.neu.edu
Subject: Re: TAO etc.
Status: R

Hi Johan:

>From johan@ccs.neu.edu Thu Jan 14 15:06:18 1999
>To: Karl Lieberherr <lieber@ccs.neu.edu>
>Subject: Re: TAO etc. 
>From: Johan Ovlinger <johan@ccs.neu.edu>
>
> It works, however, only for the most common cases and is not general
> because it does not construct the traversal graph.
>
>Could you expand, please? I acknowledge that it has a more restricted

Your algorithm does not do an NFA intersection. It breaks with zigzags.
Read the paper with Boaz:
http://www.ccs.neu.edu/research/demeter/biblio/strategies.html

>input language than dem does, but I think that you exaggerate the
>differences.  If it has to be 100 % the same as demjava, that could be
>arranged too, as could using doug's AP library. However, what TAO does
>(did) do, it does correctly! This makes it better than demjava, (or at
>least made it better than demjava for a long time; those bugs may have
>been fixed now), so it is unfair to imply that TAO is not
>general. Maybe not complete, but fully general. 
>
>In general, we have two avenues to interpreting traversals; we can
>either be fully dynamic and use reflection, or be less dynamic and
>force the user to have his code analysed before he/she can use runtime
>traversals. 
>
>In neither case does the user's code NEED to be recompiled (it is just
>an artifact of an early design decision that TAO needs this). 
>
>However, if speed is of any concern, I suspect that the reflective
>approach will be unsatisfactory. 
>
>Also, it is not so much of a burden to have the program analysed
>before it can be traversed- it needs to be compiled anyway, so this
>will not be too burdensome. RMI needs this too, so it is an accepted
>burden by the user. 
>
>I propose that instead of using reflection, modify TAO to generate all
>its methods into one class (which is added to the user's package to
>allow access to private classes). 
>
>I'm not certain that TAO is a better solution than reflection, but I
>do think that it deserves a more in depth analysis than I think it has
>recieved. I once again volunteer to describe the changes I think need
>to be made to it.
>
>Johan
>

That is useful. How does that work: generate all
its methods into one class?

I would like to use TAO ideas as we move from the first version
of Demeter/J to improved versions.

-- Karl

-- Karl