Re: A Variables Example together with a Worked Example

[david_marston@us.ibm.com]

I like the newest example for variables, where geopolitical-location 
(a.k.a. locale?) is the key for more than one assertion. It may be longer 
than you wanted for just introducing the subject, but it shows the point 
that the variable can be reused. The part where the WEIGHTs are set make 
this use of variables look a little OCL-ish, which I consider a bad thing, 
but maybe there are operational efficiencies that offset the aesthetic 
concerns. There are "exogenous" values, such as properties that cannot be 
determined by testing, so maybe that concept can be brought into the mix. 
In this case, geopolitical-location is an exogenous variable, while the 
WEIGHTs are the other kind, endogenous variables. Before you can run tests 
derived from these TAs, you need values assigned to all the exogenous 
variables. The endogenous ones will be calculated when "resolving" the 
TAs, which is a combination of running tests and/or resolving TAs such as 
widget-TA100-7a. (Notice that widget-TA100-7a does not require an actual 
unit-under-test, only that you know which geopolitical-location you're 
testing.)

This leads to the further question, which can be deferred until a later 
edition. Should the collection of TAs for a particular spec be accompanied 
by a complete list of all the properties? Or maybe it needs a list of all 
the exogenous variables. See Appendix F of the XSLT 2.0 specification [1] 
for a real-life example of a list of properties that must be reset for 
each unit-under-test.

SDG>There are here three types of TA
>1. normal
>2. property definition
>3. variable definition

Interesting. I wonder if the Prescription Level should be used to earmark 
type 2 or 3 TAs. To continue the example, since widget-TA100-7a does not 
require an actual unit-under-test, it is not really "mandatory" because it 
doesn't impose a constraint that a non-conformant unit-under-test can 
violate. It's more like a "definition" as Stephen observed. I think 
Prescription Level needs careful engineering.
.................David Marston
IBM Research

[1] 
http://www.w3.org/TR/2007/REC-xslt20-20070123/#implementation-defined-features