Martin Guy
I've trimmed down the sqrt2 example to a small source file which is attached and obligingly continues to work under hugs and dump core with nhc98.
Yes, I can reproduce the bug. I think it is likely due to nhc98 failing to check the internal representation size of large Integers correctly, causing some kind of uncaught heap overflow. Certainly, you can alter the number of digits calculated prior to the core dump by adjusting the heap size (e.g. with +RTS -H16M -RTS). Nasty, but at least you get the right answer before crashing!
I'm afraid that the wrong-answer of pi is rather hard to extract into a self-contained fragment as it uses the infinite-precision trig functions and they call in the rest of the world.
I suspect that the wrong value of pi is /not/ a manifestation of the same fault at all. If you want to send that code to me privately, I'm happy to investigate further. Regards, Malcolm