EventF describes the context of a TransientIO computation:

Run a "non transient" computation within the underlying state monad, so it is guaranteed that the computation neither can stop neither can trigger additional events/threads.

Run a transient computation with a default initial state

Run a transient computation with a given initial state

Get the continuation context: closure, continuation, state, child threads etc

Run the closure and the continuation using the state data of the calling thread

Run the closure and the continuation using its own state data.

Warning: Radically untyped stuff. handle with care

Compose a list of continuations.

Run the closure (the x in 'x >>= f') of the current bind operation.

Run the continuation (the f in 'x >>= f') of the current bind operation with the current state.

Save a closure and a continuation (x and f in 'x >>= f').

Save a closure and continuation, run the closure, restore the old continuation. | NOTE: The old closure is discarded.

Restore the continuations to the provided ones. | NOTE: Events are also cleared out.

Run a chain of continuations. WARNING: It is up to the programmer to assure that each continuation typechecks with the next, and that the parameter type match the input of the first continuation. NOTE: Normally this makes sense to stop the current flow with stop after the invocation.

stop the current computation and does not execute any alternative computation

Constraint type synonym for a value that can be logged.

Dynamic serializable data for logging.

A synonym of empty that can be used in a monadic expression. It stops the computation, which allows the next computation in an Alternative (<|>) composition to run.

Run b once, discarding its result when the first task in task set a has finished. Useful to start a singleton task after the first task has been setup.

Set the current closure and continuation for the current statement

Reset the closure and continuation. Remove inner binds than the previous computations may have stacked in the list of continuations. resetEventCont :: Maybe a -> EventF -> StateIO ()

Total variant of tail that returns an empty list when given an empty list.

Sets the maximum number of threads that can be created for the given task set. When set to 0, new tasks start synchronously in the current thread. New threads are created by parallel, and APIs that use parallel.

Terminate all the child threads in the given task set and continue execution in the current thread. Useful to reap the children when a task is done.

Add a label to the current passing threads so it can be printed by debugging calls like showThreads

Show the tree of threads hanging from the state.

Return the state of the thread that initiated the transient computation topState :: TransIO EventF

find the first computation state which match a filter in the subthree of states

Return the state variable of the type desired for a thread number

execute all the states of the type desired that are created by direct child threads

Add n threads to the limit of threads. If there is no limit, the limit is set.

Ensure that at least n threads are available for the current task set.

Disable tracking and therefore the ability to terminate the child threads. By default, child threads are terminated automatically when the parent thread dies, or they can be terminated using the kill primitives. Disabling it may improve performance a bit, however, all threads must be well-behaved to exit on their own to avoid a leak.

Enable tracking and therefore the ability to terminate the child threads. This is the default but can be used to re-enable tracking if it was previously disabled with freeThreads.

Kill all the child threads of the current thread.

Kill the current thread and the childs.

Kill the childs and the thread of an state

Same as getSData but with a more general type. If the data is found, a Just value is returned. Otherwise, a Nothing value is returned.

Retrieve a previously stored data item of the given data type from the monad state. The data type to retrieve is implicitly determined by the data type. If the data item is not found, empty is executed, so the alternative computation will be executed, if any, or Otherwise, the computation will stop.. If you want to print an error message or a default value, you can use an Alternative composition. For example:

getSData <|> error "no data of the type desired"
getInt = getSData <|> return (0 :: Int)

Same as getSData

setData stores a data item in the monad state which can be retrieved later using getData or getSData. Stored data items are keyed by their data type, and therefore only one item of a given type can be stored. A newtype wrapper can be used to distinguish two data items of the same type.

import Control.Monad.IO.Class (liftIO)
import Transient.Base
import Data.Typeable

data Person = Person
   { name :: String
   , age :: Int
   } deriving Typeable

main = keep $ do
     setData $ Person Alberto  55
     Person name age <- getSData
     liftIO $ print (name, age)

Accepts a function which takes the current value of the stored data type and returns the modified value. If the function returns Nothing the value is deleted otherwise updated.

Either modify according with the first parameter or insert according with the second, depending on if the data exist or not. It returns the old value or the new value accordingly.

runTransient $ do                   modifyData' (\h -> h ++ " world") "hello new" ;  r <- getSData ; liftIO $  putStrLn r   -- > "hello new"
runTransient $ do setData "hello" ; modifyData' (\h -> h ++ " world") "hello new" ;  r <- getSData ; liftIO $  putStrLn r   -- > "hello world"

Same as modifyData

Same as setData

Delete the data item of the given type from the monad state.

Same as delData

mutable state reference that can be updated (similar to STRef in the state monad) They are identified by his type. Initialized the first time it is set.

Run an action, if it does not succeed, undo any state changes that it might have caused and allow aternative actions to run with the original state

Executes the computation and reset the state either if it fails or not.

generates an identifier that is unique within the current program execution

Generator of identifiers that are unique within the current monadic sequence They are not unique in the whole program.

StreamData represents a task in a task stream being generated.

A task stream generator that produces an infinite stream of tasks by running an IO computation in a loop. A task is triggered carrying the output of the computation. See parallel for notes on the return value.

Run an IO computation asynchronously carrying the result of the computation in a new thread when it completes. If there are no threads available, the async computation and his continuation is executed before any alternative computation. See parallel for notes on the return value.

Force an async computation to run synchronously. It can be useful in an Alternative composition to run the alternative only after finishing a computation. Note that in Applicatives it might result in an undesired serialization.

spawn = freeThreads . waitEvents

An stream generator that run an IO computation periodically at the specified time interval. The task carries the result of the computation. A new result is generated only if the output of the computation is different from the previous one. See parallel for notes on the return value.

Run an IO action one or more times to generate a stream of tasks. The IO action returns a StreamData. When it returns an SMore or SLast a new result is returned with the result value. If there are threads available, the res of the computation is executed in a new thread. If the return value is SMore, the action is run again to generate the next result, otherwise task creation stop.

Unless the maximum number of threads (set with threads) has been reached, the task is generated in a new thread and the current thread returns a void task.

Execute the IO action and the continuation

kill all the child threads associated with the continuation context

Make a transient task generator from an asynchronous callback handler.

The first parameter is a callback. The second parameter is a value to be returned to the callback; if the callback expects no return value it can just be a return (). The callback expects a setter function taking the eventdata as an argument and returning a value to the callback; this function is supplied by react.

Callbacks from foreign code can be wrapped into such a handler and hooked into the transient monad using react. Every time the callback is called it generates a new task for the transient monad.

Runs the rest of the computation in a new thread. Returns empty to the current thread

listen stdin and triggers a new task every time the input data matches the first parameter. The value contained by the task is the matched value i.e. the first argument itself. The second parameter is a message to the user for the user. The label is displayed in the console when the option match.

inputf after sucessful or not identifier default value proc

Waits on stdin and return a value when a console input matches the predicate specified in the first argument. The second parameter is a string to be displayed on the console before waiting.

input with a default value

Wait for the execution of exit and return the result or the exhaustion of thread activity

Runs the transient computation in a child thread and keeps the main thread running until all the user threads exit or some thread exit.

The main thread provides facilities for accepting keyboard input in a non-blocking but line-oriented manner. The program reads the standard input and feeds it to all the async input consumers (e.g. option and input). All async input consumers contend for each line entered on the standard input and try to read it atomically. When a consumer consumes the input others do not get to see it, otherwise it is left in the buffer for others to consume. If nobody consumes the input, it is discarded.

A / in the input line is treated as a newline.

When using asynchronous input, regular synchronous IO APIs like getLine cannot be used as they will contend for the standard input along with the asynchronous input thread. Instead you can use the asynchronous input APIs provided by transient.

A built-in interactive command handler also reads the stdin asynchronously. All available options waiting for input are displayed when the program is run. The following commands are available:

1. ps: show threads
2. log: inspect the log of a thread
3. end, exit: terminate the program

An input not handled by the command handler can be handled by the program.

The program's command line is scanned for -p or --path command line options. The arguments to these options are injected into the async input channel as keyboard input to the program. Each line of input is separated by a /. For example:

 foo  -p  ps/end

Same as keep but does not read from the standard input, and therefore the async input APIs (option and input) cannot be used in the monad. However, keyboard input can still be passed via command line arguments as described in keep. Useful for debugging or for creating background tasks, as well as to embed the Transient monad inside another computation. It returns either the value returned by exit. or Nothing, when there are no more threads running

Exit the main thread with a result, and thus all the Transient threads (and the application if there is no more code)

If the first parameter is Nothing return the second parameter otherwise return the first parameter..

Delete all the undo actions registered till now for the given track id.

backCut for the default track; equivalent to backCut ().

Run the action in the first parameter and register the second parameter as the undo action. On undo (back) the second parameter is called with the undo track id as argument.

onBack for the default track; equivalent to onBack ().

Register an undo action to be executed when backtracking. The first parameter is a "witness" whose data type is used to uniquely identify this backtracking action. The value of the witness parameter is not used.

For a given undo track type, stop executing more backtracking actions and resume normal execution in the forward direction. Used inside an undo action.

forward for the default undo track; equivalent to forward ().

Abort finish. Stop executing more finish actions and resume normal execution. Used inside onFinish actions.

Start the undo process for a given undo track identifier type. Performs all the undo actions registered for that type in reverse order. An undo action can use forward to stop the undo process and resume forward execution. If there are no more undo actions registered, execution stop

a backpoint is a location in the code where callbacks can be installed and will be called when the backtracing pass trough that point. Normally used for exceptions.

install a callback in a backPoint

back for the default undo track; equivalent to back ().

Clear all finish actions registered till now. initFinish= backCut (FinishReason Nothing)

Register an action that to be run when finish is called. onFinish can be used multiple times to register multiple actions. Actions are run in reverse order. Used in infix style.

Run the action specified in the first parameter and register the second parameter as a finish action to be run when finish is called. Used in infix style.

Execute all the finalization actions registered up to the last initFinish, in reverse order and continue the execution. Either an exception or Nothing can be

trigger finish when the stream of data ends

Install an exception handler. Handlers are executed in reverse (i.e. last in, first out) order when such exception happens in the continuation. Note that multiple handlers can be installed for the same exception type.

The semantic is, thus, very different than the one of onException

set an exception point. Thi is a point in the backtracking in which exception handlers can be inserted with onExceptionPoint it is an specialization of backPoint for exceptions.

When an exception backtracking reach the backPoint it executes all the handlers registered for it.

Use case: suppose that when a connection fails, you need to stop a process. This process may not be directly involved in the connection. Perhaps it was initiated after the socket is being read so an exception will not backtrack trough the process, since it is downstream, not upstream. The process may be even unrelated to the connection, in other branch of the computation.

in this case you only need to create a exceptionPoint before stablishin the connection, and use onExceptionPoint to set a handler that will be called when the connection fail.

Delete all the exception handlers registered till now.

Use it inside an exception handler. it stop executing any further exception handlers and resume normal execution from this point on.

catch an exception in a Transient block

The semantic is the same than catch but the computation and the exception handler can be multirhreaded

throw an exception in the Transient monad there is a difference between throw and throwt since the latter preserves the state, while the former does not. Any exception not thrown with throwt does not preserve the state.

main= keep  $ do
     onException $ \(e:: SomeException) -> do
                 v <- getState <|> return "hello"
                 liftIO $ print v
     setState "world"
     throw $ ErrorCall "asdasd"

the latter print "hello". If you use throwt instead, it prints "world"