...

Our feeling over here is that this is an unforced change, so in isolation it probably wouldn't be worthwhile. It'll break some code, and the awkward squad paper will have to be updated to comply (although we just noticed it is already wrong about the behaviour of Prelude.catch).

However, if there's concensus that folk would prefer the alternate definition of IOError, then we're happy to go along with it.

There have been three responses, all in favour of change. What's your view now? I would rather see a confusing interface fixed than force the other implementations to change their Preludes to implement it, but then I have no investment in the old interface.

Fine, I'm happy to go along with the change. Cheers, Simon