Re: img_border_percent , object_border_perc and object_border_pixel testcases

On 10/01/2011 09:40 PM, public-html-testsuite@gtalbot.org wrote:
>> Hi Gérard,
>>
>
> Hello Ms2ger,
>
> Thank you for your reply.
>
>> On 09/29/2011 08:36 PM, public-html-testsuite@gtalbot.org wrote:
>>> http://w3c-test.org/html/tests/approved/xhtml5/img_border_percent.xhtml
>>> http://w3c-test.org/html/tests/approved/xhtml5/object_border_perc.xhtml
>>> http://w3c-test.org/html/tests/approved/xhtml5/object_border_pixel.xhtml
>>   >
>>> So, the logical question would be how far back should backward
>>> compatibility go in HTML5?
>>
>> As specified in the Rendering section, in this case. See [1] in
>> particular.
>>
>>> All those testcases have no link to their authors. Is that normal? I
>>> mean,
>>> submitted testcases do not have to indicate their author or sources?
>>
>> That's normal, yes.
>
>
> I do not agree with such policy. There is enough anonymity on the web as
> it is.

Can we please not have arguments like this? I don't think it contributes 
positively to the tone of the list.

> People who want their testcases to be taken seriously should have
> their real names mentioned and a way to reach them.

People who want their testcases to be taken seriously should contribute 
high quality testcases that match the spec and find bugs in browsers. 
Having a contact for the testcase author is sometimes helpful but is 
unnecessary; testcases often outlive their author's involvement in web 
standards. We have to take collective responsibility for the content of 
testcases. If people other than the author can't verify that they are 
correct they shouldn't pass review.

> In my opinion, the rules, process, method of parsing and rendering such
> testcases should be indicated with an adequate<meta assert="" name="">
> (added in the testcase) and adequate<link rel="help" href="" title="">
> (added in the testcase) to link precisely to relevant section of the spec.

That doesn't work for HTML. We can't require specific markup because we 
are testing the markup layer. This is not CSS. We could, and perhaps 
should, have out of band metadata but I think the most compelling use 
case would be an annotated version of the spec that showed where we have 
tests. So far no one has come forward to build this in a way that would 
work well with the HTML5 spec, so there has been no serious push to add 
all the metadata to tests (it would also be a serious inconvenience to 
some methods of writing tests e.g. the attribute reflection tests where 
thousands of tests are generated from relatively small amounts of data).

> Especially if the expected result of such testcase is not so obvious or
> intuitive or indicate a clear divergence with previous specification and
> parsing.

Previous specifications are irrelevant.

> I still does not understand how, on one hand, the spec speaks of
> "non-conforming", "obsolete" almost everywhere when border attribute is
> specified for<img>  and<object>  but still honors it anyway as in those
> testcases. It's a double message, it's a confusing message: it's like
> saying "Don't do this ... but, if you do, we'll honor it anyway".

In general the spec has requirements for authors and different 
requirements for implementors. You need to understand this basic 
distinction to make any sense of the spec at all. Our testcases are all 
aimed at implementors and so all follow the implementation requirements. 
The author requirements are irrelevant.

Received on Monday, 3 October 2011 08:01:18 UTC