Re: [abc] Concrete operational semantics

What I really want is for someone to check what Eric has written as
his understanding of the implementation of tracematches and make sure
it corresponds with what we really have. Eric will look at Julian's
point and will also see how his op semantics relates to the one in
the OOPSLA paper.

If we have a common understanding, then we can all take a look at the
analysis Eric proposes and ensure that it is sound. He is also
working on more benchmarks. There will certainly be a class of
tracematches for which his analysis is not yet precise enough due to
the lack of a must points-to analysis. However, I am hoping we
can show that we have an interesting first step that is already
useful in some cases and can be used as a framework for further
improvements by incorporating (1) more precise points-to analyses and
(2) different kinds of context.

Cheers, Laurie

+-----------------------------------------------------------------
| Laurie Hendren --- laurie.hendren@mcgill.ca
| Associate Dean (Academic), Faculty of Science,
| Dawson Hall, McGill University, 853 Sherbrooke St W,
| Montreal QC H3A 2T6 Canada, 514-398-7179, fax 514-398-1774
+----------------------------------------------------------------
| For contact and home page info as Professor, Computer Science:
| http://www.sable.mcgill.ca/~hendren --- hendren@cs.mcgill.ca
| Research: http://www.sable.mcgill.ca http://aspectbench.org
+----------------------------------------------------------------

Oege de Moor wrote:
>
> Wow, that's great news, Eric!
>
> On Fri, 6 Oct 2006, Eric Bodden wrote:
>
>>> As I understand it, none of the existing benchmarks has that property.
>>
>> It seems that for all benchmarks contained in the DaCapo benchmark
>> suites, we are able to remove all (!) of the shadows for the
>> SafeEnumeration tracematch. The analysis never took longer than a few
>> minutes, always spending more than 90% in constructing call graph and
>> points-to information. For other patterns, results will be worse, but
>> still I think this is a fairly good result.
>>
>> So I would not worry too much about the benchmarks, although in general
>> you are of course right: It's all worth nothing if there is no benefit
>> from it.
>>
>> Eric
>>
>>
>
Received on Fri Oct 06 21:04:27 2006