It seems the "Simple IO Regions" are insecure:
> {-# OPTIONS -fglasgow-exts #-}
>
> module BreakIORegions where
>
> import IORegions
> import Control.Monad
In particular, we can build actions involving a handle inside its scope,
and execute them outside.
Look closely at the type error from Oleg's test 8:
IORegionsTest.lhs:36:53:
Could not deduce (IORegions.IN mark marks1)
from the context (IORegions.IN mark marks)
arising from use of `qGetChar' at IORegionsTest.lhs:36:53-60
Probable fix:
add (IORegions.IN mark marks1) to the expected type of an expression
In the first argument of `return', namely `(qGetChar q)'
In a lambda abstraction: \ q -> return (qGetChar q)
In the second argument of `withFile', namely `(\ q -> return
(qGetChar q))'
The problem isn't anything about which marks are present - it's because
the application qGetChar q somehow ended up with a different type
variable "marks1" than the marks variable used in this instance of IOM.
I don't understand why trying to use the returned action later on isn't
enough to force the types to unify, but maybe GHC is a little too eager
to figure out whether class constraints are satisfied.
Anyway, we can leak an action if we give it a little help in unifying
the types, here with lexically scope type variables
> test1 = let (body :: IOM marks Char) =
> do a <- withFile "/etc/motd"
> (\q -> return (qGetChar q)
> :: IOM marks (IOM marks Char))
> a
> in body
> test1r = runIOM test1 >>= print
Or more compactly with join:
> test = runIOM (join (withFile "/etc/motd" (return . qGetChar))) >>= print
Brandon Moore
_______________________________________________
Haskell mailing list
[email protected]
http://www.haskell.org/mailman/listinfo/haskell
