[Haskell-cafe] Pure and monadic functions using the Repa arrays

[Michael Orlitzky]

On 08/21/12 09:19, felipe zapata wrote:
> Hi Haskellers,
> 
> I have been playing with the Repa functions and trying the
> Repa-examples. In order to gain experience with the Repa functions I
> have written some small linear algebra utilities and import this module
> to a bigger project. In the beginning of my project I used the mmultP
> function from the repa-examples to calculate a big matrix, therefore I
> have and array of type:
> 
> 
> arr :: Monad m => m (Array U DIM2 Double)
> 
> 
> Then I carried this array in a lot of functions which become Monadic
> function and then it is necessary to introduce the monadic machinery for
> manipulating this functions . The Question is then if there is the
> possibility to work with a pure function in place of the monadic version?
> 
> There is something like a runRepa function?
> 
> runRepa :: Monad m => m (Array U DIM2 Double) - > Array U DIM2 Double
> 
> 
> or could I used the unsafePerformIO function ?
> 
> 
> or the evaluation of the parallel arrays must be postponed until the
> Repa.Array is called in the main function?

When this change was introduced (there wasn't always the arbitrary monad
m around everything), I remember I just wrapped my one big repa function
in the identity monad and it worked fine. For example,

  -- Grid.hs
  import Control.Monad.Identity (Identity)
  ...
  zoom :: Values3D -> ScaleFactor -> Identity Values3D

  -- Main.hs
  import Control.Monad.Identity (runIdentity)
  ...
  let output = runIdentity $ zoom dbl_data zoom_factor

This gives you the warm/fuzzy of knowing that the function is pure, but
I think I eventually just let it run in IO to avoid introducing mtl as a
dependency.