A Tube is a computation which can

• yield an intermediate value downstream and suspend execution; and
• await a value from upstream, deferring execution until it is received.

Moreover, individual Tubes may be freely composed into larger ones, so long as their types match. Thus, one may write small, reusable building blocks and construct efficient stream process pipelines.

Since a much better engineered, more popular, and decidedly more mature library already uses the term "pipes" I have opted instead to think of my work as a series of tubes.

TubeF is the union of unary functions and binary products into a single type, here defined with a Boehm-Berarducci encoding.

This type is equivalent to the following:

       data TubeF a b k
           = Await (a -> k) -- :: (a -> k) -> TubeF a b k
           | Yield (b  , k) -- :: (b  , k) -> TubeF a b k
   

The type signatures for the two value constructors should bear a strong resemblance to the actual type signature of runT. Instead of encoding tubes as structures which build up when composed, a TubeF is a control flow mechanism which picks one of two provided continuations.

People using this library should never have to contend with these details but it is worth mentioning.

A computation which only yields and never awaits

A computation which only awaits and never yields.

run is shorter than runFreeT and who knows, maybe it'll change some day

Command to wait for a new value upstream

Command to send a value downstream

Convert a list to a Source

Enumerate yielded values into a continuation, creating a new Source

Infix version of for

Connect a task to a continuation yielding another task; see ><

Compose two tubes into a new tube.

Connects a Source to a Sink, finishing when either the Source is exhausted or the Sink terminates.

Insert a value into a Sink

This performs a neat trick: a Tube with a return type a will be turned into a new Tube containing the underlying TubeF value.

In this way the >< and >- functions can replace the () return value with a continuation and recursively traverse the computation until a final result is reached.

Continuously relays any values it receives. Iteratee identity.

Transforms all incoming values according to some function.

Refuses to yield the first n values it receives.

Relay only the first n elements of a stream.

Terminates the stream upon receiving a value violating the predicate

Yields only values satisfying some predicate.

Strict left-fold of a stream. Note that the actual return type of the source is not relevant, only the intermediate yield type.

Similar to each except it explicitly marks the stream as exhausted

Taps the next value from a source, maybe.

Source of Strings from stdin. This is mostly for debugging / ghci example purposes.

Sink for Strings to stdout. This is mostly for debugging / ghci example purposes.

A Pump is the dual to a Tube: where a Tube is a computation manipulating a stream of values, a Pump can be situated on either end of a tube to both insert values when requested and handle any yielded results.

One interesting use of a Pump is to feed data to a Tube, collecting the result as well as unused input:

   import Data.Functor.Identity

   p :: [a] -> Pump (Maybe a) x Identity [a]
   p inp = pump (return inp)
           (wa -> case (extract wa) of
               [] -> (Nothing, wa)
               x:xs -> (Just x, return xs))
           const

   -- a Sink that stops after 5 loops, or when input is exhausted
   add5 :: Sink (Maybe Int) IO Int
   add5 = loop 0 5 where
       loop acc ct = if 0 == ct
           then return acc
           else do
               mn <- await
               maybe (return acc)
                     (n -> loop (acc+n) (ct - 1))
                     mn

   result :: IO ([Int], Int)
   result = runPump (curry id) (p [1..10]) add5
   -- ([6,7,8,9,10],15)
   

Pumps are still being investigated by the author so if you come up with something interesting, please share!

Creates a Pump for a Tube using a comonadic seed value, a function to give it more data upon request, and a function to handle any yielded results.

Values received from the Tube may be altered and sent back into the tube, hence this mechanism does act like something of a pump.

Send a value into a Pump, effectively re-seeding the stream.

Pull a value from a Pump, along with the rest of the Pump.

Given a suitably matching Tube and Pump, you can use the latter to execute the former.

Lift a computation from the argument monad to the constructed monad.