Sure, EOF is checked every time you read rather than being cached after the first time it occurs (actually this is the case in the new I/O library in 5.02, I'm not sure about 5.00.2). So it's reasonable to do a sleep/read loop similar to 'tail -f'.

Odd! Not only does it not have that behaviour in ghc-4.0[48], it also didn't seem to work when I tried coding it that way in C. Maybe tells you more about my C skills... ;-) I'll wait until the stable ghc-5.02, and then try this again, just in case this has indeed changed significantly recently, rather than it just being me missing something obvious. Cheers&thanks, Alex.