On 05.01.2011, at 22:28, Bernhard Pieber wrote:

> Hi Bert,
> 
> Thanks for your answer!
> 
> Am 05.01.2011 um 16:36 schrieb Bert Freudenberg:
>> On 05.01.2011, at 12:03, Bernhard Pieber wrote:
>> It just occurred to me that the broken tests can be fixed if the following is put to the trunk as a postscript or preamble:
>>> 
>>> PackageOrganizer default
>>> 	unregisterPackageNamed: 'PackageInfo';
>>> 	unregisterPackageNamed: 'ToolBuilder';
>>> 	unregisterPackageNamed: 'Morphic-TrueType'
> Would you be opposed to put this to the trunk? I cannot see any adverse effect on anyone's image.

I wouldn't be opposed. AFAICT there are no preambles/postscripts in these instances so we would not loose anything. I just don't see the point either.  E.g., at least in my trunk image there are 81 registered package infos but only 60 working copies. So why delete these 3?

>> Maybe PackageDependencyTest should be made less whiny? By default it should not complain about those packages. You only want to ensure that a release image is clean, not every working image.
> Good idea. One way to make it less whiny would be to only take into account PackageInfos associated with a Monticello MCWorkingCopy. I *think* this was the intention behind the PackageDependencyTest anyway.
> 
>> Perhaps unit tests aren't the best way to ensure a clean release. They're supposed to exercise code. ReleaseBuilder does some sanity checks already, maybe it should also check for clean packages.
> I don't think so. I kind of like this test. I think it is useful for improving the package structure. And it should only affect trunk packages, not your code. Once the obsolete PackageInfos are cleaned it will be more useful. We can then use it to flag unwanted dependencies, e.g. PackageInfo-Base should not depend on Morphic.

It certainly is useful. But why does it have to be a unit test?

- Bert -