- From: David Hull <dmh@tibco.com>
- Date: Wed, 21 Sep 2005 01:51:01 -0400
- To: public-ws-desc-comments@w3.org
- Message-id: <4330F4C5.1010509@tibco.com>
This email collects several minor editorial points regarding the WSDL Core specification, some more minor than others. I apologize for the somewhat late submission, which was due to several extraneous circumstances too tedious to mention. * In the paragraph at the beginning of section 1.3 ending "... the functionality implied by that extension is only optional to the client. But it needs to be supported by the Web Service." the last sentence might better read "It must be supported by the Web Service." * In section 2.3.1, the paragraph beginning "Note that faults other than the ones described in the Interface ..." seems like more than a mere note. * In the same section, the rather long note beginning "Despite having a {name} property, Interface Fault components cannot be identified ..." seems to say, basically, that separately included interfaces must be consistent with each other to the extent they overlap. If this is so, something to the effect of "In other words, interfaces must be consistent where they overlap." might make the situation clearer. Also, my first reaction to this was "Ah, the diamond inheritance problem." But diamond inheritance is not a problem, since the shared ancestor is automatically consistent with itself (see the note on idempotency below). If I've got this right, it might be worth noting. * The above note appears in very similar form in other sections. I'm sure that factoring this out has been suggested, but for what it's worth, I'd like to weigh in in favor of having at least the bulk of it appear only once, possibly with a short reference where the mostly-repeated text presently appears * In section 2.4.1.1, in "... the rules implied by that IRI (such as rules that govern the schemas) MUST be followed or it is an error", it was not clear to me what exactly should happen should this error occur. The text appears to mean that the combination simply would not work, just as in-scope properties with an empty intersection won't work. The question here is, what entity, if any, needs to report this error (and how, etc.). Also, the net effect of all this seems to be the conjunction of the (possibly empty) set of constraints specified by the IRIs, but the separation into cases obscures this somewhat. If there is a general agreement that this paragraph could be made clearer, I would be willing to attempt a clearer wording. * In section 2.7.1, second paragraph. Consecutive sentences start with "Thus," and "Hence," This might flow better as "Thus, by definition, every SOAP 1.2 abstract feature is also a WSDL 2.0 feature, and there is no need to define a separate WSDL 2.0 feature ..." * In section 2.8.1, first paragraph, the last sentence asserts "Properties, and hence property values, can be shared amongst features/bindings/modules, and are named with IRIs precisely to allow this type of sharing." This seems to emphasize that the names are not QNames, but it wasn't apparent to me why QNames wouldn't have worked equally well. I have the distinct feeling the answer is blindingly obvious and I'm just missing it. * In 2.8.1.1, the paragraph beginning "Many of the component types in the component model contain a {properties} property ..." produced parse errors on the first couple of readings. The root problem is that "property" is badly overloaded, at least in this context. Unfortunately, it's not clear how to avoid this painlessly, as both senses of "property" are integral to the spec. The upshot seems to be to distinguish between properties (in the F&P sense) that are directly attached (declared properties) and all applicable properties, whether directly attached or pulled in by composition (in-scope properties). If this is the intent, the wording could be clearer (and again, I could probably have a go at rewording it). * In 2.9.1, paragraph 6, the text "Thus, it is an error for a binding component not to define bindings ..." again leads me to wonder what reports the error and when. Also, the "Thus," made me wonder if I'd missed a step (quite likely). * In 2.9.2, in the XML representation of Binding components, the word "whgse" was clearly inserted to test whether reviewers were awake. I'm happy to report that I might well be. * Section 2.11.2.2 seems like boilerplate in the same sense as the note on inheritance and overlap mentioned above. * In the tables in sections 2.12.3 and 2.13.3, the descriptions of {interface message|fault reference}seem rather lengthy for a table entry. And again, they seem to be near duplicates. The general problem with such duplication is of course that it's hard to tell without careful inspection what's duplicated and what's not. * In section 2.15.1, the {address} is given as optional. Is there any guidance on when one might want to not include it? * In section 2.17, the definition of equivalence is oddly asymmetric. Speaking as a math major, I'd be more comfortable if "and the second component has no additional properties" were simply "and vice versa." Similarly, "set-based values are considered equivalent if they contain corresponding equivalent values, without regard to order." feels a bit odd since on the one hand, sets are unordered by definition, and on the other hand they are also duplicate-free by definition but this is not mentioned. I might reword this as "Set-based values are considered equivalent if for each value in the first, there is an equivalent value in the second, and vice versa," parallel with the proposed rewording for components as a whole (and the usual set-theoretic definition that each is a subset of the other). * In the same section, in the first bullet point (simple types), does "contain the same value" mean "contain the same actual value" in the XSD sense? If not, what does it mean? * In section 2.19, first paragraph, a "tuple, consisting of two parts" is generally called an "ordered pair". * In the following paragraph, "QName references are resolved by looking in the appropriate property," the appropriate property should be spelled out, even if the result is a bit tedious. * In 4.1, the paragraph beginning "A mutual include is direct inclusion by one WSDL 2.0 document of another WSDL 2.0 document ..." the upshot seems to be that inclusion is idempotent (because a component is equivalent to itself and so multiple inclusion doesn't redefine anything). It would be better either to mention idempotency explicitly, or, perhaps better, to rephrase this paragraph in terms of the transitive closure (which is mentioned earlier) to explain that everything in the transitive closure of the inclusion relation is thrown into the same flat space. * In section 4.1, the location attribute doesn't have to be dereferenceable, but this may lead to unresolvable QNames down the line. It would be nice to say that implementations SHOULD report broken location links as a root cause of the QName failures they cause, assuming that this is not burdensome to determine. * A.3 (security considerations) is remarkably short. This is just a comment -- I don't think it needs to be lengthened.
Received on Wednesday, 21 September 2005 05:51:18 UTC