Re: ocl conformance test



Date view Thread view Subject view Author view Attachment view

From: Jim Steel (Jim.Steel@irisa.fr)
Date: Mon 16 Feb 2004 - 08:30:19 GMT


I think part of the problem with XMI is the wide variety of choice. As 
the old saying goes, the best thing about standards is that there are so 
many to choose from. For UML tools, they can have any combination of UML 
1.3, 1.4, 1.5, 2.0 (such as it is), and XMI 1.0, 1.1, 1.2, and 2.0 (the 
latter two of which are both still current) and quite legitimately claim 
conformance.

(Of course, I don't suggest for a second that all the products are 
conformant - I tried a similar thing with MOF tools a couple of years 
back with similar results).

Its a problem, but its a symptom of standardising against developing 
technologies like XML - e.g. there was no XML Schema when XMI began, and 
even MDA itself, through churn in the MOF and UML standards. It makes it 
complicated for a reference implementation, too, since it has to 
understand all the combinations (or at least a number of them).

Jim.


Laurence Tratt wrote:
> On Sun, Feb 15, 2004 at 09:57:17PM +0100, Stefan Haustein wrote:
> 
> Dear Stefan,
> 
> 
>>I agree that support for XMI in tools is still far from perfect.
>>
>>However, most tools claim to be able to read XMI. Perhaps the availability
>>of a vendor-neutral OCL conformance test file would encourage vendors to
>>fix their XMI support---at least to a level that allows to read the test
>>file. As you say, part of the problem is that there is no such test.
> 
> 
> Tests are good as far as they go. However in practise they rarely cover all
> that is needed. For that you really need to have a decent reference
> implementation available. [For those of you unsure what I mean here: a
> reference implementation can be defined as a (generally non-production
> quality) freely available implementation which correctly implements all
> mandatory parts of a standard and successfully executes all tests. It is
> therefore available both as a demonstration of to how to implement a
> standard, and also a way of testing interoperability between it and another
> tool.] In the QVT world, the QVT-Partners - of which I am a member -
> strongly advocated the concept of reference implementations as a way of
> ensuring a higher quality end standard. Reference implementations have a
> number of precedents outside the UML/OCL/QVT type world, and have been
> highly successful at promoting real interoperability between tools
> conforming to a standard. Without a reference implementation to back up the
> suggested OCL tests, the cynic in me worries that conformance to the tests
> will barely be worth the paper it is written upon.
> 
> To bring this back to XMI, the fact that most tools *claim* to be able read
> XMI is nearly - but not quite - as laughable as the fact that some tools
> have been claiming to support UML2 since even before it entered its current
> stage of finalization. Marketing men aren't noted for their strict adherence
> to the truth ;) Given that many (most?) tool vendors play little more than
> lip service to OCL, I have to doubt whether an OCL test suite on its own
> would encourage them in any way to improve their XMI support.
> 
> I do see a potential solution here to the XMI problem: a reference
> implementation of XMI import/export would I think force vendors to improve
> their own XMI support. I have heard vague whisperings that the OMG might
> embark on such a course - I wouldn't hold your breath though!
> 
> Yours,
> 
> 
> Laurie

Date view Thread view Subject view Author view Attachment view