The reflex library provides the foundation Classes and their implementation APIs to do Functional Reactive Programming. This is independent of the DOM creation code, and can be used to implement FRP architecture in non-web related apps also.

The Quick Ref provides a really nice overview of its APIs.

FRP Basics

In order to leverage the full power of reflex, one has to effectively use the ability to create an Event propagation graphs, and use it to model the business logic. This guide gives an overview of basics and various useful techniques.

Also see Reflex Basics




Is used to inject value in the reflex event-propagation-graph from outside using IO action:

newTriggerEvent :: TriggerEvent t m
  => m (Event t a    -- Event triggered by fun
       , a -> IO ()) -- fun

newTriggerEvent can also be used to break a big rec block.:

  ev1 <- widget1 evN


  evN <- widgetN evN_1

In this the widgetN and many other widgets in-between can be pulled outside the rec block:

(evN, evNIOAction) <- newTriggerEvent

ev1 <- widget1 evN


evN' <- widgetN evN_1

performEvent $ ((\v -> liftIO $ evNIOAction v) <$> evN')
From Dynamic

By calling updated on a Dynamic value one can obtain the event when its value changes.:

updated :: (Reflex t) => Dynamic t a -> Event t a
Repeating Events

Using APIs from Reflex.Time one can create repeating events.:

tickLossy :: (_)
  => NominalDiffTime -- in seconds
  -> UTCTime
  -> m (Event t TickInfo)

tickLossy will create an Event every n seconds. Though it is not guaranteed to always fire an Event after the elapsed time, especially if the value n is very small.

There are many more APIs in this module to generate repeating events based on more complex algorithms.

From DOM widgets

When doing DOM based programming using reflex-dom-core, a number of widgets provide Event in response to the external events.


Using these primary Events you can create secondary / derived events by

  1. Manipulating the value using Functor / fmap:

    -- inputValueEv :: Event t Int
    doubledInputValueEv = ffor inputValue (* 2)
  2. Filtering the value:

    -- inputValueEv :: Event t Int
    -- This Event will fire only if input value is even
    evenOnlyEv = ffilter even inputValueEv

    Use fmapMaybe fforMaybe for similar filtering

  3. Multiple events can be combined using

    Merges the value a

    <>         :: Semigroup a => Event a -> Event a -> Event a

    This fires the a event only when b is not firing at the same time:

    difference :: Event a -> Event b -> Event a

    Combine two separate events:

    align      ::                     Event a -> Event b -> Event (These a b)
    alignWith  :: (These a b -> c) -> Event a -> Event b -> Event c

    Combine a list of events:

    mergeWith  :: (a -> a -> a) -> [Event a] -> Event a
    mergeList  :: [Event a] -> Event (NonEmpty a)

    Drop all except the leftmost event:

    leftmost   :: [Event a] -> Event a

    Other APIs:

    mergeMap   :: Ord k => Map k (Event a) -> Event (Map k a)
    merge      :: GCompare k => DMap (WrapArg Event k) -> Event (DMap k)
  4. Tagging value of Dynamic or Behavior.

    Using these APIs, see Quick Ref

    gate                       ::                     Behavior Bool -> Event a -> Event a
    tag                        ::                        Behavior a -> Event b -> Event a
    tagPromptlyDyn             ::                         Dynamic a -> Event b -> Event a
    attach                     ::                        Behavior a -> Event b -> Event (a, b)
    attachPromptlyDyn          ::                         Dynamic a -> Event b -> Event (a, b)
    attachWith                 :: (a -> b ->       c) -> Behavior a -> Event b -> Event c
    attachPromptlyDynWith      :: (a -> b ->       c) ->  Dynamic a -> Event b -> Event c
    attachWithMaybe            :: (a -> b -> Maybe c) -> Behavior a -> Event b -> Event c
    attachPromptlyDynWithMaybe :: (a -> b -> Maybe c) ->  Dynamic a -> Event b -> Event c
    <@>                        ::                 Behavior (a -> b) -> Event a -> Event b
    <@                         ::                        Behavior a -> Event b -> Event a

    The below will create an event which will fire whenever the Dynamic changes and give the old value of the Dynamic.

    tag (current dyn) $ updated dyn


Behavior value can be tagged with an Event using tag or attach, or it can be sampled in a widget, when it is first created using sample.



Create a Dynamic which changes value when Event occurs:

holdDyn :: (MonadHold t m) => a -> Event t a -> m (Dynamic t a)

There are also a number of input APIs in reflex-dom-core which provide Dynamic values in the context of DOM. See DOM Input elements


Using some primary Dynamic values you can create secondary / derived values by

  • fmap - Simply use Functor instance when only one Dynamic value is being manipulated.

  • Combine multiple Dynamic values using:

    zipDyn :: Reflex t => Dynamic t a -> Dynamic t b -> Dynamic t (a, b)
    zipDynWith :: Reflex t => (a -> b -> c) -> Dynamic t a -> Dynamic t b -> Dynamic t c

    Zipping is useful when multiple Dynamic values have a common point of influence in the application.

    For example if you have two variable parameters like color and font of text. Then you can construct the dynamic attributes from these parameters by simply zipping them together.:

    -- textFont :: Dynamic t Text
    -- textColor :: Dynamic t Text
    getAttr (f,c) = ("style" =: ("font-family: " <> f "; color: " <> c))
    elDynAttr "div" (getAttr <$> (zipDyn textFont textColor)) $ text "Text"
  • Using Applicative:

    -- dInt1, dInt2, dInt3 :: Dynamic t Int
      eInt :: Dynamic t (Int, Int, Int)
      eInt = (,,) <$> dInt1 <*> dInt2 <*> dInt3

    Much more complicated things can be done using traverse/ sequenceA:

    -- mDyn :: Map k (Dynamic t Int)
      dMap :: Dynamic t (Map k Int)
      dMap = sequenceA mDyn


zipDynWith is more efficient than f <$> d1 <*> d2


The Reflex class provides the basic functionality for FRP. It provides the basic functions to efficiently handle the Event, Behavior and Dynamic values. All the pure APIs like tagDyn, zipDyn, etc are created using the functionality provided through Reflex class.

The other two most important features required for FRP are maintaining some state, and doing modifications based on events. This is provided from the two classes MonadHold and Adjustable.

Also see QuickRef


This is required to create any stateful computations with Reflex. It designates monads that can create new Behavior s based on Event s.:

hold :: a -> Event t a -> m (Behavior t a)


A Monad that supports adjustment over time. After an action has been run, if the given events fire, it will adjust itself so that its net effect is as though it had originally been run with the new value.:

runWithReplace :: m a -> Event t (m b) -> m (a, Event t b)

Event Propagation Graph

Simple Tree

Simply pass the Event/Dynamic values to input of functions. This will create kind of an event propagation flow from top to bottom. But no feedback-loops can be created, for that use RecursiveDo.


Is used to create a cyclic event propagation graph. Because the underlying mechanism of graph creation is monadic (using MonadHold, etc). To create feedback-loops we need to use MonadFix.

The actual usage is quite simple:

-- Required extension for rec style blocks
-- {-# LANGUAGE RecursiveDo #-}

    ev1 = f2 <$> ev2
  d1 <- widgetHold (w1Init) (w1 <$> ev1)
  ev2 <- viewD1Widget d1

in this example the ev1 is used to create a Dynamic value d1, which is then shown to the user using viewD1Widget. This widget can in turn modify the value using the Event ev2.

But there are some pitfalls too, especially if you use ‘Promptly’ APIs like tagPromptlyDyn, switchPromptlyDyn, attachPromptlyDyn, etc. All these APIs take a Dynamic value as input, and if used incorrectly they can cause problems like hang, stack overflow, etc.

In most cases you would want to use their corresponding APIs like tag, switch, attach, etc (which all work on the Behavior values), along with current :: Dynamic t a -> Behavior t a.

see debugging Hang / Stack Overflow

For more details checkout the articles on MonadFix / RecursiveDo

Maintaining State via fold

In order to store a state/data for your app (ie create a state machine) simply use foldDyn

-- State can be any arbitrary haskell data
stateDynVal :: Dynamic t MyState

-- ev can a collection of all events on which the state depends
-- For example all input events
ev :: Event t Inputs

-- This is a pure API which can process the input events and current state
-- to generate a new state.
eventHandler :: (Inputs -> MyState -> MyState)

-- foldDyn :: (a -> b -> b) -> b -> Event t a -> Dynamic t b
stateDynVal <- foldDyn eventHandler initState ev

Even nested state machines can be designed if your have a state with nested Dynamic value by using foldDynM

Use foldDynMaybe, foldDynMaybeM in cases where you want to filter input events, such that they don’t modify the state of application.

For example in a shopping cart if the user has not selected any items, the “add to cart” button should do nothing. This kind of behavior can be implemented by returning Nothing from the eventHandler.


getPostBuild :: PostBuild t m => m (Event t ())

This Event will fire once at the start of an action / DOM widget is created. Also each time that part of the DOM gets re-created (like if it is created from scratch via widgetHold). This can be used to do communication with server or do some FFI.

Note that the Event fires when the build action completes, but the fragment may not yet be in the browser DOM. So you might have to add some delay to this before accessing the DOM via some FFI.

Doing IO via performEvent


doneEv <- performEvent (ffor triggerEv $ \val -> liftIO $ do
  putStrLn "Doing some action"
  someIOAction val)

widgetHold (text "Waiting for action to complete")
  (showResultOfAction <$> doneEv)


Does the doneEv always occur in the frame after triggerEv?

Debounce, Delay, BatchOccurence

Reflex.Time provides a set of useful APIs which come handy when you need to do real life event handling.:

debounce :: (_) => NominalDiffTime -> Event t a -> m (Event t a)

-- Wait for user to stop typing for 0.5 sec, and then send a search request to server

searchTextEv <- debounce 0.5 (_textInput_input someTextInput)

When doing FFI calls delay may be required:

delay :: (_) => NominalDiffTime -> Event t a -> m (Event t a)

performEvent (abort <$ stopAndRestartEv)
delayedEv <- delay 0.2 stopAndRestartEv
performEvent (start <$ delayedEv)

When handling a set of events from external sources many times the sequence of events is not deterministic, or perhaps we want a debounce kind of functionality but dont want to miss any Event. In such cases we need to use batchOccurrences to properly model the logic.

batchOccurrences :: (_) => NominalDiffTime -> Event t a -> m (Event t (Seq a))

Higher order FRP

Nested Values and flattening

When you model real world Dynamic values many times you end up with nested structures.

For example, if the value of items in a shopping cart depends on the shipping method chosen, then you can end up with a value total' :: Dynamic t [Dynamic t Int]:

selectedItems :: Dynamic t [Item]
isExpeditedShipping :: Dynamic t Bool

total' = Dynamic t [Dynamic t Int]
total' = ffor selectedItems
          (map getItemPrice)

getItemPrice :: Item -> Dynamic t Int
getItemPrice itm = ffor isExpeditedShipping
                      True -> (itemPrice itm) + (shippingCharges itm)
                      False -> itemPrice itm)

In such cases in order to get a total value Dynamic t Int, you need to use flattening APIs. In case of Dynamic it is simply join from Control.Monad (since Dynamic has an instance of Monad):

total'' :: Dynamic t (Dynamic t Int)
total'' = foldr1 (\a b -> (+) <$> a <*> b) <$> total'

total :: Dynamic t Int
total = join total''

See QuickRef for details on other flattening APIs.

Dynamic widgets on Dynamic Collections

In order to model complex flows of events or dynamically changing data collection, we need to use higher order containers like lists ([]) or Maps (Data.Map).

To effectively work with such Dynamic collections, Reflex.Collection provides a bunch of APIs.

See Quickref for a summary of these APIs


Provides these APIs. If you look closely they are the equivalent of dyn and widgetHold, but work in non-DOM applications.:

networkView :: (Reflex t, NotReady t m, Adjustable t m, PostBuild t m)
  => Dynamic t (m a) -> m (Event t a)

networkHold :: (Reflex t, Adjustable t m, MonadHold t m)
  => m a -> Event t (m a) -> m (Dynamic t a)

EventWriter and DynamicWriter

EventWriter allows you to send events “upwards” in your widget hierarchy, much like Elm’s update propagation.:

-- Main APIs
runEventWriterT :: (Reflex t, Monad m, Semigroup w) => EventWriterT t w m a -> m (a, Event t w)
tellEvent :: EventWriter t w m => Event t w -> m ()

-- Example usage
body :: MonadWidget t m => m ()
body = do
    (_, ev) <- runEventWriterT ewbs
    dy <- foldDyn (:) ["bar"] ev
    simpleList dy dynText
  return ()

ewbs :: MonadWidget t m => EventWriterT t Text m ()
ewbs = do
  evClick <- button "Click Me"
  tellEvent ("foo" <$ evClick)
  return ()


Requester lets you make requests and receive responses anywhere within your widgets, and automatically collect/distribute them as necessary.

The primary API which will be used to initiate a request and get a response is:

requesting :: Event t (Request m a) -> m (Event t (Response m a))

This requires defining two type constructors Request m and Response m.

The API to actually collect all the requests and provide response to each request is:

runRequesterT :: (Reflex t, Monad m)
  => RequesterT t request response m a
  -> Event t (RequesterData response)
  -> m (a, Event t (RequesterData request))

As you can see all the requests are bundled up in the RequesterData request, and the responses are also provided in a similar event of type RequesterData response.

The RequesterData is like a Map structure where the keys are some arbitrary values corresponding to the origin of request, and the values are the actual request data.

to provide a response one can use these APIs:

traverseRequesterData :: forall m request response. Applicative m
  => (forall a. request a -> m (response a))
  -> RequesterData request
  -> m (RequesterData response)

can be used to provide response to all the request by specifying a request handler.

But if you want access to each request separately and provide the responses in independent manner (in case you are doing XHR/ websocket requests for each request separately).

Then you can convert this into a list of key value pairs (DSum), provide the response to each request by using the same key with singletonRequesterData to recreate the RequesterData:

requesterDataToList :: RequesterData f -> [DSum RequesterDataKey f]

singletonRequesterData :: RequesterDataKey a -> f a -> RequesterData f


Reflex.Workflow provides a specialised API:

newtype Workflow t m a = Workflow { unWorkflow :: m (a, Event t (Workflow t m a))}

workflow :: forall t m a. (Reflex t, Adjustable t m, MonadFix m, MonadHold t m)
  => Workflow t m a -> m (Dynamic t a)

The working of this API can be easily explained using a DOM based widget example:

-- A DOM based example of Workflow
page1, page2, page3 :: (MonadWidget t m) => Workflow t m Text
page1 = Workflow . el "div" $ do
  el "div" $ text "This is page 1"
  pg2 <- button "Switch to page 2"
  return ("Page 1", page2 <$ pg2)

page2 = Workflow . el "div" $ do
  el "div" $ text "This is page 2"
  pg3 <- button "Switch to page 3"
  pg1 <- button "No wait, I want to go back to page 1"
  return ("Page 2", leftmost [page3 <$ pg3, page1 <$ pg1])

page3 = Workflow . el "div" $ do
  el "div" $ text "You have arrived on page 3"
  pg1 <- button "Start over"
  return ("Page 3", page1 <$ pg1)

main = mainWidget $ do
  r <- workflow page1
  el "div" $ do
    text "Current page is: "
    dynText r



UniqDynamic is useful to eliminate redundant update events from a Dynamic.:

uniqDynamic :: Reflex t => Dynamic t a -> UniqDynamic t a

fromUniqDynamic :: (Reflex t, Eq a) => UniqDynamic t a -> Dynamic t a

Internally, UniqDynamic uses pointer equality as a heuristic to avoid unnecessary update propagation; this is much more efficient than performing full comparisons. However, when the UniqDynamic is converted back into a regular Dynamic, a full comparison is performed.

In order to maintain this constraint, the value inside a UniqDynamic is always evaluated to weak head normal form.

Also see the documentation of Reflex.Dynamic.Uniq

Patch and Incremental

An Incremental is a more general form of a Dynamic. Instead of always fully replacing the value, only parts of it can be patched. This is only needed for performance critical code via mergeIncremental to make small changes to large values.

Reflex.Patch.* provides a number of data structures which have the ability to do incremental updates.

Cheap / Fast variants of APIs



A frame is the atomic time unit

  • Frame begins with, say, a mouse click
  • Mouse click event fires
  • Events fmapped from that event fire
  • All other events depending on those events fire
  • Repeat until there are no more event firings
  • Frame ends

Spider Timeline