- From: Andrew Thackrah <a.thackrah@opengroup.org>
- Date: Wed, 29 May 2002 10:17:15 +0100
- To: www-qa@w3.org
> Alex Rousskov writes: >> I believe we can come up with a simple addressing scheme if we limit >> the problem domain to, say, the extraction of single pieces from XML >> documents. > David Marston/Cambridge/IBM replied: > I think we need to recognize a smallish taxonomy. For example, the > "productions" used in some Recommendations should be marked as such. > Likewise, standalone testable sentences might be separately tagged > from sentences which are testable but have constraints on their > applicability which can be derived from a larger context. For example, > a spec may have several sequential paragraphs about how general XML is > to be handled, followed by several sequential paragraphs about XHTML, > then HTML, etc. The tags for testable sentences within those regions > should at least indicate that the sentences are not universal. > > Another situation is the presentation of normative instructions in > tabular form, such as the notorious table of data-type conversions. > In effect, each cell of the table is a testable assertion, even if it > isn't a standalone expression in text. I have wondered about the value of a specification in which each section has 3 parts 1. description 2. definitions 3. assertions The first part is informative natural language discussion to set the context - what is being defined and how it is to used. This section would not be normative. The second part provides definitions of any data structures, tables etc. This section is normative The third part contains assertions for the formal requirements of the section. This section is normative. Assertions can refer to definitions from the second part - a single assertion can refer to a table of values for example - this assertion can be decomposed into the appropriate number of unit tests. The advantage is that the spec writer has to write the assertions too (it's an advantage for test suite developers anyway..) If each assertion is uniquely identified then it can be referred directly from a test description. I think this is a little different from what is being discussed in the sense that the formal requirements are kept in a separate part of the section from the informal text. -AndrewT
Received on Wednesday, 29 May 2002 05:19:47 UTC