David T. Lewis wrote:
>>> No, the compiler got it exactly right. It was faithfully optimizing  
>>> the chosen definition of #ifNotNil: - whoever changed the  
>>> definition "got it wrong" since the definition should not be  
>>> changed without also changing the optimization in the compiler.
>>>
>>> But what's *really* annoying to me is that this change went unnoticed.
>> Indeed I think that we should invest in tests and a testserver
> 
> Tests are good, but they do not detect this kind of problem.

Why not? If the purpose of a test is to detect a problem then the 
inability to load a simple test like:

testIfNotNilArg
   self assert: (3 ifNotNil:[:val| val = 3])

is precisely pointing out that there is a problem, e.g., the test 
detected the problem and served its intended purposes.

Cheers,
   - Andreas