...

In FFI code you have to be more careful [about killing threads], however.

It's clear there's a bunch of bad things could happen here - but not obvious what does happen.

Does killing a thread that is calling a C function kill it immediately (i.e., before the call completes) or wait until the C function finishes?

The thread can only receive an exception while running Haskell code. That means that in an 'unsafe' C call, the thread can't receive an exception until the call returns, but a thread doing a 'safe' C call may receive an exception if the call re-enters Haskell. Either way, you don't have to worry about being killed in the middle of C code, which would be completely impossible to deal safely with. Cheers, Simon