[Haskell-cafe] Boxed Mutable Arrays

[Bulat Ziganshin]

Hello Alberto,

Tuesday, December 15, 2009, 5:23:41 PM, you wrote:

ST monad is the same as IO monad modulo type trickery. tris trickery
allows to limit imperative operations available inside ST monad to
STRef/STArray manipulations which is guaranteed to be
referential-transparent as part of this trickery. STRef/STArray is,
again, equivalent to IORef/IOArray modulo this trickery

so basically we say "there is safe subset of IO operations, let's give
them acces in the way that prevents their unsafe use", makes
ST/STRef/STArray/runST newtypes wrapped around
IO/IORef/IOArray/unsafePerformIO and export
them to user w/o providing access to implementation details

ST monad doesn't contain any hidden state - at least when we say from
*implementation* point of view

> AFAIK, ST Arrays are pure data structures. They are not really
> mutable. They are destroyed and recreated on every update. The
> mutation is just simulated thanks to the hidden state in the state
> monad. Sure, the garbage collector must have a hard work in
> recycling all the "backbones" of the discarded arrays (not the elements).
>  
> 2009/12/14 Brad Larsen <brad.larsen at gmail.com>
>  Is anyone working on fixing ticket #650
>  <http://hackage.haskell.org/trac/ghc/ticket/650>?  In short, STArray
>  and the garbage collector don't play well together, resulting in array
>  updates being non-constant time operations.  This bug makes it very
>  difficult/impossible to write efficient array algorithms that depend
>  upon mutation in Haskell.
>  
>  On another note, does this (or perhaps better phrased, will this) bug
>  also affect Data Parallel Haskell?
>  
>  I would really like to see highly efficient, mutable, boxed arrays in
>  Haskell!  Unfortunately, I don't have the know-how to fix Ticket 650.
>  
>  Sincerely,
>  Brad
>  
> _______________________________________________
>  Haskell-Cafe mailing list
>  Haskell-Cafe at haskell.org
>  http://www.haskell.org/mailman/listinfo/haskell-cafe
>  


>   


-- 
Best regards,
 Bulat                            mailto:Bulat.Ziganshin at gmail.com