Hi, Gabor. That's a weird result, and I can't reproduce it on my Mac.
However, the code that tracks buffer sizes is going to change pretty soon
(to support symbolic extents), so I'd hold off on testing this for now.
The test case was mine and the expected warnings deliberately short since
I wasn't sure if the text of the warning would change. I didn't do a great
job on that though -- the capitalization of "undefined" is different to
match the different messages.
Those test results do seem odd, though. Is it possible your target
architecture has a char bigger than 1 byte? The test case is not set up to
handle this right now.