Pipes.Lift

Note that runWriterP and execWriterP will keep the accumulator in weak-head-normal form so that folds run in constant space when possible.

This means that until transformers adds a truly strict WriterT, you should consider unwrapping WriterT first using runWriterP or execWriterP before running your Proxy. You will get better performance this way and eliminate space leaks if your accumulator doesn't have any lazy fields.

Probably the most useful functionality in this module is lifted error handling. Suppose that you have a Pipes.Pipe whose base monad can fail using ExceptT:

import Control.Monad.Trans.Error
import Pipes

example :: Monad m => Pipe Int Int (ExceptT String m) r
example = for cat $ \n ->
    if n == 0
    then lift $ throwError "Zero is forbidden"
    else yield n

Without the tools in this module you cannot recover from any potential error until after you compose and run the pipeline:

>>> import qualified Pipes.Prelude as P>>> runExceptT $ runEffect $ P.readLn >-> example >-> P.print42<Enter>421<Enter>10<Enter>Zero is forbidden>>> 

This module provides catchError, which lets you catch and recover from errors inside the Pipe:

 import qualified Pipes.Lift as Lift

 caught :: Pipe Int Int (ExceptT String IO) r
 caught = example `Lift.catchError` \str -> do
     liftIO (putStrLn str)
     caught

This lets you resume streaming in the face of errors raised within the base monad:

>>> runExceptT $ runEffect $ P.readLn >-> caught >-> P.print0<Enter>Zero is forbidden42<Enter>420<Enter>Zero is forbidden1<Enter>1...

Another common use case is running a base monad before running the pipeline. For example, the following contrived Producer uses StateT gratuitously to increment numbers:

import Control.Monad (forever)
import Control.Monad.Trans.State.Strict
import Pipes

numbers :: Monad m => Producer Int (StateT Int m) r
numbers = forever $ do
    n <- lift get
    yield n
    lift $ put $! n + 1

You can run the StateT monad by supplying an initial state, before you ever compose the Producer:

import Pipes.Lift

naturals :: Monad m => Producer Int m r
naturals = evalStateP 0 numbers

This deletes StateT from the base monad entirely, give you a completely pure Pipes.Producer:

>>> Pipes.Prelude.toList naturals[0,1,2,3,4,5,6...]

Note that the convention for the StateT run functions is backwards from transformers for convenience: the initial state is the first argument.

All of these functions internally use distribute, which can pull out most monad transformers from the base monad. For example, evalStateP is defined in terms of distribute:

evalStateP s p = evalStateT (distribute p) s

Therefore you can use distribute to run other monad transformers, too, as long as they implement the MFunctor type class from the mmorph library.