On 12 juin 05, at 20:28, Richard Staehli wrote:

> I think you are right that a numbering convention could identify  
> the relation between a package version and the correct test  
> version, but your example demonstrates how this convention might be  
> confusing or wrongly applied:
>
> On Sunday, June 12, 2005, Stephane Ducasse wrote:
>
>
>> yes this is what we learned the hard way (impossible to go in the
>> past and identify the tests
>> covering the right version old version).
>>
>> I have the impression that having the same number for tests than for
>> the package works
>>
>> PA1             PAT1
>> PA1.2           PAT1 "nothing changes in the tests
>> PA2             PAT1 "nothing
>> PA3
>> PA4             PAT4 "tests changes to cover version 4"
>> PA4             PAT4.1 "fixed tests alone
>> PA5             PAT4.1
>> PA6             PAT4.1
>> PA7             PAT7  "tests changed and are in sync with package7
>>
>
> If you look for tests for PA3, how do you know that PAT1 is the  
> right version.  It seems likely that PA1 might have been changed  
> (twice) and that the developers simply forgot to update the tests.
>
> The question of what tests are intended for a particular version of  
> a package is, or should be, the same as the question of what type  
> (functional behavior) the package is supposed to implement.  While  
> we don't need a static type system in the Smalltalk language, we  
> already have a behavioral testing system in the Smalltalk  
> development environment.  We can make do with informal naming and  
> version numbering conventions to express the relationship between  
> package and tests, or we could assert that package tests define  
> (albeit informally and incompletely)  the types in a package and  
> require that a package declare its type by including a reference to  
> the correct test version (test package name and major version number).

yes may be a direct reference is indeed better.