Charles Durham
Charles Durham

Reputation: 1707

Lazily evaluate monadic functions in Haskell

I can't seem to figure out a workaround for this issue i'm having.

I have something like this:

  getFilePathForDay :: Day -> IO (Maybe FilePath)

  getFilePathForDays date days = do
      files <- mapM getFilePathForDay $ iterate (addDays 1) date
      return . take days . catMaybes $ files

I am trying to get x amount of valid file paths where x is the number of days I want, but the above code just runs forever. I have ran into this problem before with monads and I was wondering if there is a workaround available for this issue I am having.

Thanks!

Upvotes: 8

Views: 321

Answers (2)

Don Stewart
Don Stewart

Reputation: 138007

Sometimes lazy IO is the right approach. Since you have an infinite, lazy sequence of days, and you're sampling, lazily, some (dynamic) range from that sequence, we should also hit the filesystem only on demand.

This can be done with unsafeInterleaveIO:

import System.IO.Unsafe         ( unsafeInterleaveIO )

-- return a lazy list of filepaths, pull as many as you need
getFilePathForDays :: Day -> Int -> IO [FilePath]
getFilePathForDays date days = do

    let go (d:ds) = unsafeInterleaveIO $ do
                        f  <- getFilePathForDay d
                        fs <- go ds
                        return (f:fs)

    -- an infinite, lazy stream of filepaths
    fs <- go (iterate (addDays 1) date)

    return . take days . catMaybes $ fs

Where go returns a lazy stream of results, and we take as many as we need. Easy.

Upvotes: 4

hammar
hammar

Reputation: 139930

The code runs forever because you are trying to sequence an infinite number of side-effecting computations in your call to mapM.

Since you don't know ahead of time how many of these computations you will need to run (as indicated by your use of catMaybes), you will have to interleave the calls to getFilePathForDay with the rest of your computation. This could be done using unsafeInterleaveIO, but as the name suggests, this is not a recommended approach.

You could implement this manually as:

getFilePathForDays _ 0 = return []
getFilePathForDays date days = do
    mpath <- getFilePathForDay date
    case mpath of
        Just path -> (path :) <$> remaining (days-1)
        Nothing   -> remaining days
  where
    remaining days = getFilePathForDays (addDays 1 date) days

There is probably a more elegant way, but it's not coming to me right now :)

Upvotes: 8

Related Questions