catch (fail "...")

[Brian Huffman]

Hi,

Here is a problem I came across recently, revealing a difference between ghci 
and Hugs:

hugs> catch (error "not ok") (\e -> putStrLn "ok")
Program error: not ok
ghci> catch (error "not ok") (\e -> putStrLn "ok")
*** Exception: not ok

hugs> catch (ioError (userError "not ok")) (\e -> putStrLn "ok")
ok
ghci> catch (ioError (userError "not ok")) (\e -> putStrLn "ok")
ok

hugs> catch (fail "not ok") (\e -> putStrLn "ok")
Program error: not ok
ghci> catch (fail "not ok") (\e -> putStrLn "ok")
ok

As I expected, neither of them can catch "error", and both of them can catch 
"ioError". But only ghci can catch the "fail". It seems to me that ghc has 
made a convenient choice here, because if I need to raise an IO exception, it 
is easier to type "fail s" than "ioError (userError s)".

However, the online Haskell report says that for IO, fail s = error s, which 
is what Hugs does. 
(http://www.haskell.org/onlinereport/standard-prelude.html#$iMonad$$IO)

Does anyone know the reasoning behind the design choice made in the report, 
or why the ghc implementation is different? I am mainly concerned about 
portability issues: particularly, is it a good idea to use fail for raising 
IO exceptions?

Thanks
- Brian Huffman