Re: how to test parsers?

On 30 Jul 2009, at 21:54, Michael Schneider wrote:
[snip]
>> As I recall, we talked about this under the subject of parser/ 
>> serializer
>> conformance.  Maybe my best bet is to make sure the two ontologies  
>> each
>> entail each other....  Is that good enough?  Is there anything  
>> simpler I
>> can do?
>
> I think, that's a sufficient but also a necessary approach, at least  
> in
> general (there may be special treatment for special scenarios).

We had a big fuss about this a ways back. My strong viewis that it's  
neither necessary nor sufficient (as Sandro pointed out in his next  
message).

In this particular case, I believe the two axioms were structurally  
equivalent. SE is not ideal but is *much* better than mutual  
entailment (the false positives are much easier to inspect).

For your example, Sandro, there are several possible reactions:

1) Manual inspect (not hugely unreasonable, actually)
2) Report it as an annoying serialization bug (not becuase it doesnt  
conform, but because it's uboptimal)
3) Do a structural equivalence test.

You could approximate the later in a lot of cases by, whenever a set  
appears in the UML, asking for a determinate sort on the  
serialization. Indeed, I suspect that that's what's happening here  
(Nothing being lexically prior to thing...easy enough to test).

Hope this helps.

Cheers,
Bijan.

Received on Thursday, 30 July 2009 20:51:39 UTC