# Algorithms: Flood Fill in Haskell - Abstracting the common

In the comments of my blog post describing the flood fill algorithm in Haskell David Turner pointed out that the way I was passing the grid around was quite error prone.

```
floodFill :: Array (Int, Int) Colour -> (Int, Int) -> Colour -> Colour -> Array (Int, Int) Colour
floodFill grid point@(x, y) target replacement =
if((not $ inBounds grid point) || grid ! (x,y) /= target) then grid
else
gridNorth
where grid' = replace grid point replacement
gridEast = floodFill grid' (x+1, y) target replacement
gridWest = floodFill gridEast (x-1, y) target replacement
gridSouth = floodFill gridWest (x, y+1) target replacement
gridNorth = floodFill gridSouth (x, y-1) target replacement
```

I actually did pass the wrong grid variable around while I was writing it and ended up quite confused as to why it wasn’t working as I expected.

David’s suggestion based on the above version of the algorithm was that I might want to use the state monad to thread state rather than explicitly passing it around like I am here.

I had a go at writing the function using the state monad but realised while doing so that rather than doing that I could fold over a cells neighbours rather than using the state monad to thread state.

I didn’t see the ‘neighbours’ concept the first time I wrote it but David wrote a version of the algorithm in which he introduced that.

This is the resulting function:

```
floodFill :: Array (Int, Int) Colour -> (Int, Int) -> Colour -> Colour -> Array (Int, Int) Colour
floodFill grid point target replacement =
if(target == replacement) then
grid
else
let gridWithSquareReplaced = if onGrid point then grid // [(point, replacement)] else grid
validNeighbours = filter (onGrid `and` sameAsTarget) neighbours in
foldl (\grid point -> floodFill grid point target replacement) gridWithSquareReplaced validNeighbours
where neighbours = let (x,y) = point in [(x+1, y), (x-1, y), (x, y+1), (x, y-1)]
sameAsTarget point = grid ! point == target
onGrid = inRange $ bounds grid
and :: (a -> Bool) -> (a -> Bool) -> a -> Bool
and f g x = f x && g x
```

I initially get all of the neighbouring squares but then filter those out if they don’t fit on the grid or aren’t of the required colour. We can then use ‘foldl’ to iterate over each neighbour passing the grid along as the accumulator.

David also wrote a more succinct function to work out whether or not a value fits on the grid so I’ve stolen that as well!

I’m still not sure I totally understand when I should be using ‘let’ and ‘where’ but in this example the functions I’ve put in the ‘where’ section are helper functions whereas the values I defined in the ‘let’ section are domain concepts.

Whether that’s the correct way to think of the two I’m not sure!