Thursday, January 17, 2013



This is the second version of MFlow [0], a deep-first effort in the development of a plataform for web applications at the higher level, by including as much haskell magic as possible. It is at the same time experimental and intended to be used in industry. I believe that haskell will not have its chance in the Web if it does not bring any unique advantage beyond speed and type safety. 

Deep-first means that the effort is more in adding higher level features rather than in being complete and bug free, with the conviction that feedback from real usage at the highest level is the best guide for development. 

Rather than to mimic other platforms in other languages, MFlow is as Haskell'ish as possible, and introduces new approaches like  statefulness,  event sourcing, back-execution, composable, active, self contained widgets and persistent STM. All of them collaborate to create a high level environment for web programming.

Statefulness means that the flow of execution is right in front of the eyes of the programmer, rather that in  the spaghetti code of the event-handling model,  the goto of the XXI century. it also means that the entire session is in a procedure, so the session data need no trickery.

But, like in the stateless case, any page in the flow could have a direct URL. [1]

Event sourcing in combination with statefulness means  automatic logging of all the events and automatic recovery of the execution state of the server procedure and its data,  thanks to the Workflow monad transformer [2].

So the server procedure can be stopped after timeout and restarted by the application server when the user invokes it. The programmer can determine the timeout and for how long the state is stored.

Back button management: the flow of execution can run back when the back button is pressed [3]. Thanks to a monad transformer. In combination with event sourcing, this means automatic rollback of user session data. So if the user want to undo his last item in the shopping cart, he only has to press the back button.

User interface: 

It is made of self contained pieces called widgets, made around the formlet concept, so they compose gracefully with applicative operators and other extra operators. A MFlow widget is more like the widgets of XWindows, ASP.NET or JavaServer Faces: they have their own HTML formatting, their own behaviours and interact with the server via AJAX, they execute actions when activated and they return statically typed data. They declare their required scripts, CSS links and server procedures in the form of requirements to the requirement service. They include the server and client code in the same piece of text. Being self contained and typed, they can be shared and plugged with a minimum of configuration and instructions.

The URL links are also widgets or elements of widgets and return data to the same flow, so no spaguetty event handling occur, unless the programmer need out-of-flow links.

This versions includes some  active widgets with server-side controls for creating them [4]. 

wai-warp, blaze-html, xhtml, HSP, hack.

Persistence and transactions: 
Global data can be transacted with any kind of persistence framework. MFlow internally uses TCache [5], which provides with persistent STM references (DBRefs). Their serialization can be configured. For rapid prototyping, they have default persistence in files. Just add a Read-Show or a Binary instance for your data and you can have a persistent, transactional DBRef to it. Use them like you use TVars. TCache can index text and register fields. It has a simple query language. Data can be accessed by key or by any indexed field.

The human mind learn by examples. There is a online demo for some of the features that you can play with and  you can download the code [6] 

Post a Comment