For what it's worth, I compiled Daniel Wengelin's program with Janus/Ada
(with Tucker's output lines), and got the same results as Tucker did. This
apparently is a GNAT bug.
As far as the criticality of it goes, if you doing anything mission-critical
with a compiler, you have to be using a supported product. When users have a
critical bug with one of our products, we send out a repair via E-Mail as
soon as possible. Most other Ada vendors make interim fixes available
quickly as well (including ACT).
It is understandable that not everyone can justify buying an Ada compiler
(although compilers for common platforms don't cost much); but in that case
you have to recognize that other people are paying the freight, and you
ought to be happy to get anything at all. You get what you pay for; there
are no critical bugs in something free.