Re: More test suite problems

I have long thought the SAX API is broken in this regard.

Unless we want to change the Infoset spec and break the
ability to do proper catalog resolution, I think we should
leave things as is and have a note that says that people who 
use a tool that uses the broken SAX API will get different
results.  And we should tell the Xerces folks that their
parser is broken.  After all, that is what a test suite
is for, isn't it?

paul

At 09:03 2002 05 28 -0400, Glenn Marcy wrote:

>Glenn writes:
>>There is also the following problem discovered by the Xerces
>>parser that I am not sure how to solve.
>>
>>Output Tests:
>>ibm-valid-P28-ibm28v02.xml
>>ibm-valid-P29-ibm29v01.xml
>>ibm-valid-P29-ibm29v02.xml
>>ibm-valid-P57-ibm57v01.xml
>>ibm-valid-P58-ibm58v01.xml
>>ibm-valid-P58-ibm58v02.xml
>>ibm-valid-P70-ibm70v01.xml
>>
>>Xerces resolves and returns the complete path of systemIds for
>>notation decls, where as the actual outputs do not and can not
>>contain complete resolved paths for systemIds.  Xerces appears
>>to behave in accordance with the SAX API...
>>
>>    notationDecl
>>    .....
>>    If a system identifier is present, and it is a URL, the
>>    SAX parser must resolve it fully before passing it to the
>>    application through this event.
>>
>>I am not sure what we can do at this point to support expected
>>results files for documents that have absolute path names in
>>their infoset.
>>
>>Does anyone have any thoughts or suggestions for how this
>>issue can be addressed?  Otherwise, I think that we would need
>>to remove the expected results files for these tests.
>
>One thing that I should have mentioned is that SAX is certainly
>inconsistant with the Infoset specification on this point given
>this description from Notation information item:
>
>  2. [system identifier] The system identifier of the notation,
>     as it appears in the declaration of the notation, without
>     any additional URI escaping applied by the processor. If
>     no system identifier was specified, this property has no
>     value.
>
>However, if most people are testing conformance using the SAX
>API and the SAX implementations are "correct" then we cannot
>get the correct infoset to test expected results conformance.
>
>One idea that came to me would be to add a new attribute to
>the test suite descriptions similar to NAMESPACE that could
>be used by a SAX based test harness to identify those tests
>that cannot be compared with the expected output.
>
>Best Regards,
>Glenn

Received on Monday, 3 June 2002 15:59:35 UTC