> I agree to have typical ones like <title> or <documentation> explicit,
> but <header> is very useful by itself. So I want to keep that, too.
>
> For example, depending on the situation, I sometimes exclude some of the
> test cases. I'd like to use meta information in the header for this.
I agree that the set of metadata elements needs to be extensible, but I
still don't think we need header to get this. For example, we can do what
we do in RELAX NG and allow arbitrary namespaced qualified elements.
> As for explicit metadata elements, I'd like to propose <title> (or
> <name>), which is supposed to have a short string that represents its
> name.
I think <author> would be useful as well.
I noticed you have a metadata element for saying whether it depends on a
XML Schema namespaces. How about something like
<requires datatypeLibrary="http://..."/>
Any others?
>> - I made resources local to each test case, and made the directory
>> structure explicit
>
> Some of my test cases share a common resource between test cases. So I'd
> appreciate if we can keep resources a bit more reusable.
I think that's an over optimization. It makes test suites harder to
maintain and harder to understand. I think it is highly desirable to be
able to make each testCase self-contained. It's also not a very effective
optimization in my experience: I have found it much more common for
instances to be common between test cases than external resources.
> I guess we can make resources <testSuite>-local; resources can be
> written only under <testSuite> and test cases directly contained in that
> test suite can reference them. In this way, when we extract files into a
> directory structure, one test suite can go into one directory and
> everything will work file.
Allowing resources both in <testCase>s and in <testSuite>s would make
things rather complicated.
> Also, how do you use <dir> in resources? Is this necessary? Are you
> using it to test the xml:base handling?
<testCase>
<section>4.6</section>
<documentation>Same value of href before resolution, but not a
loop.</documentation>
<correct>
<externalRef xmlns="http://relaxng.org/ns/structure/0.9" href="sub/x"/>
</correct>
<dir name="sub">
<resource name="x">
<externalRef xmlns="http://relaxng.org/ns/structure/0.9" href="sub/x"/>
</resource>
<dir name="sub">
<resource name="x">
<element xmlns="http://relaxng.org/ns/structure/0.9" name="foo">
<empty/>
</element>
</resource>
</dir>
</dir>
<valid>
<foo/>
</valid>
<invalid>
<bar/>
</invalid>
</testCase>
Using dir has two advantages:
- it avoids having an attribute whose value is restricted to a relative
URI; this just feels wrong to me
- it makes it slightly easier to create the necessary directories if you
split the test suite into separate files
James
