- From: Ray Whitmer <rayw@netscape.com>
- Date: Wed, 23 May 2001 11:49:29 -0700
- To: David Brownell <david-b@pacbell.net>
- CC: Dimitris Dimitriadis <dimitris.dimitriadis@improve.se>, www-dom-ts@w3.org
David Brownell wrote: >> In any case I cannot see that as an argument to kepp the metadata separate, >> if the tests change, they remain one test; passsing them through a process >> does not produce more tests or more metadata. > > > Actually, in the interest of reducing the number of divergent > implementations, it can be better to turn such "problematic" > tests into multiple tests. That also lets "legacy" behaviors (such > as "the way IE4 does it", just an example) be tested for in > their own right -- sometimes they're essential conformance > points for products, though one hopes not for W3C (which > should flag all conformance bugs). In this particular case, you either get the right number of attributes in the list, or you do not, and I think that needs to be preserved. Adding more tests could do two things: 1. Provide pass/fail better aligned with use cases deemed most important (tests that align better with use cases). 2. Determine exactly why the tests failed (more detailed tests). For example, Mozilla and IE would both pass the following tests, which might be better aligned with real use cases: 1. Does the attribute list array indexed by number contain each specified attribute a serializer would need to write (the main use for enumerating attributes by number)? 2. Can I do a getAttribute or namedItem on each attribute (default and specified) that is expected to be there and get the right value? But this is arguably tailored for known bugs in the browsers implementations. I am all for adding tests, but not for eliminating them just because they discover non-compliance. Eliminating them is fine if we discover that they do not validate implementation of the spec. Neither browser would pass more detailed tests that I would like to see: 1. Examine the existence, type, and value of every standardized property and method of the returned attribute list and the nodes it contains, as well as individually-requested attribute nodes, making sure they are correct. There would be quite suprising failures. I think the browsers need to pass this more stringent test for real interoperability. With small fixes, I think Netscape would pass. 2. Test failure cases for proper results when an attribute does not exist. Ray Whitmer rayw@netscape.com
Received on Wednesday, 23 May 2001 14:46:03 UTC