Make array indexing immune to seg-faults

[simonpj]

As Spencer Janssen points out ( http://www.haskell.org/pipermail/libraries/2006-December/006539.html), it's possible for a bogus instance of Ix to cause a Haskell implementation to seg-fault, simply by returning an out-of-range index. This is definitely a Bad Thing.

The only way to avoid this possibility is to make (!) perform a bounds check after calling the index method of class Ix. GHC's current implementation (in GHC.Arr) is

(!) :: Ix i => Array i e -> i -> e
arr@(Array l u _) ! i = unsafeAt arr (index (l,u) i)

Instead we could have

(!) :: Ix i => Array i e -> i -> e
arr@(Array l u _) ! i = safeAt arr (index (l,u) i)

where safeAt performs a bounds check. But that would two bounds checks, one in index and one in safeAt. We could eliminate one by using unsafeIndex, which is a (usually hidden) method of GHC's Ix class definition. However, that might give rise to less-informative messages when the bounds check fails.

To implement safeAt, we'd need a new primop:

arraySize :: Array# a -> Int

There would need to be corresponding stuff for Data.Array.IArray and Data.Array.MArray.

[igloo]

See proposal #1610.

[igloo]

Fixed (proposal #1610)