This document introduces the underlying concepts that inform Magpie's design. For a more practical how-to, see ***link***.
In the animal world, the Magpie (Pica pica) is an intelligent, mischievous bird with a reputation for pilfering shiny objects. In the enlightened world of Modern Perl, Magpie is an attempt to combine the best ideas (shiny objects) from the community's long history of Web development frameworks into a truly mature and intelligent environment.
Ambitious? You bet. And we're in good company, we think. Perl itself is a magpie and always has been. Originally designed to fuse the best features of a slew of *NIX utilities into cohesive whole that made easy things easy and hard things possible, Perl has continuously evolved by adopting and internalizing the best features from each new generation of languages and tools that have been invented since (many of which were themselves influenced by Perl). At the same time, the Perl community has constantly evolved through a similar process. Community support, development infrastructure, even what constitutes "good Perl code" have all evolved by an iterative process of adoption -> extension -> invention. This is why we're making Magpie. We believe that by combining the best features of various popular Web development frameworks-- together with our own bits of invention-- we can create a development environment where easy things are easy and hard things are possible.
Implemented as a Plack::Middleware component, Magpie provides a rapid application development framework for both browser-based Web applications and RESTful Web services by removing (or at least greatly reducing) the redundant aspects of day-to-day application coding. Magpie's design is heavily informed by the study of Finite State Machines.
Magpie's basic goals and design principles can be summed up as follows:
- The most important entity in Web development the Resource. By convention and explicit design, Magpie should be Resource-oriented.
- Magpie should provide a clear, well-defined interface for separating application state-detection from the event hander methods that are executed in response to a given state. This promotes fast, focussed, incremental development.
- In any given Magpie application, event hander methods may be divided across one or more of a series of Application Classes, each of which may have its own state/event mapping logic that determines which events will be fired. This encourages modularity and code reuse.
- In addition to the Application Class(es), a Magpie application pipeline will also contain one Output Class that generates the content for the requesting client. This encourages reusability by letting us expose the same application logic to different types of Web clients.
- All Magpie's base component and helper classes must be easily replaceable with user-defined classes; This promotes invention, user contribution and project longevity.
- Magpie's core code must be environment-agnostic, allowing developers to deploy applications under Mod_per1/Apache versions 1 and 2, as well as any Web server offering the Common Gateway Interface (CGI). User-defined application classes and custom components are free to favor one environment over another but Magpie's core must remain neutral.
- Magpie must be judiciously magical-- providing just enough Perlish wizardry to make writing applications easy, while not presuming or enforcing a One True Way(tm) that unduly limits developers' freedom.
When a client sends a request via HTTP to a given URL they are asking for a Representation of a Resource, not the Resource itself. Unfortunately, the first generation of Web frameworks that tried to implement practical RESTful architectures often mixed things up and made it seem like the URL was the Resource and the Resource was the Thing.
To illustrate the distinctions,
let's suppose that you want to implement a CRUD (Create,
Delete) interface to an
orders table in your database.
After some discussion,
you decide to use a RESTful API and so you reach for one of the popular tools that purports to make the process simple.
The early stages of the project are promising-- set-up is nearly trivial and in a very short time you have unit tests demonstrating a working HTTP endpoint for your order data.
a POST to the
/order endpoint requires more than a one-to-one relationship with the database's
You have to check product availability for each item in the order,
you have to access the
customer_addresses tables to grab the client's address to calculate shipping,
Where does all that extra logic go?
Do you cram it into the validation stage?
Do you create different Controller methods to handle those tasks?
you find that the hands-off "connect the database table to the Web" application you thought you were building is turning into a series of hacks,
The tool that was supposed to make things easier now actually ties your hands.
You're wondering why this all seems so hard and you begin to suspect that REST is just another in a long line of overcomplicated bondage fads that look great on paper but fall to pieces under their own complexity when you try to do anything substantial.
But the problem isn't REST-- that's just an architectural style-- the problem is that REST-on-MVC treats the
orders table as the Resource and tightly couples the Web API to that physical asset.
In general, an application can be seen as being in (or having) a series of application states. Consider the typical online registration application. First, the user is presented with an HTML form into which they type their desired username, password, personal details and other information. We can think of this as the "prompt state" since the core action involves prompting the user to sign up. Once the user has filled in the form, they hit the submit button to send the input to a URL on the server that implements an interface that is able to read that incoming data. Once the request if received, the data is often verified for fitness-- first by logic on the server side that verifies that the data is complete and appropriate; then by the user, who is given a read-only HTML page reflecting his or her input for review. Let's call this the "validation state". Presented with the information they have entered, the user may choose to return to the prompt state by clicking a "Make Changes" button, or to proceed with registration by clicking the "Register" button. (Note that the application itself often proactively returns to the prompt state if it finds the user's input to be unfit). When both the user and the system are satisfied with the input, the verified data is sent to the server by clicking the "Register" button presented during the validation state. The server receives the data, creates the new user account, and responds with an HTML document containing a polite message thanking them for registering. We'll call this the "complete state".
In short, we can say that the typical user registration application is a single entity that can be in one or another of three distinct states (prompt, validation, and complete).
---------- -------------- ------------ | Prompt | | Validation | | Complete | | State |-[User Input]->| State |-[Data Accepted]-->| State | ---------- -------------- ------------ ^------[Data Rejected]------| State Diagram for Online Registration
Despite the fact that this principle of state-based development and design is understood (if only intuitively) by most experienced Web developers, much of the code running on the Web today remains a mix of blocks of real application-level programming wrapped by largely redundant application state detection logic. We have learned through experience the benefits of separating an application's logic from its presentation, yet we persist in mixing application state detection with code that reacts to those states. Magpie exists because its developers and users have found that separating application state detection from behavioral logic offers similar benefits to those that come from drawing a clear line between application logic and presentation-- namely, the logical division of labor/time, and the ability to create uncluttered more maintainable code faster, and the freedom to reuse resources by decoupling distinct aspects of the application implementation.
Magpie works by mapping application states, determined by user input and other factors, to event handler methods, that implement the behavior associated with that state. In Magpie, developers need only register and implement the specific bits of code that react to a given application state, Magpie makes sure that correct event (or events) are fired.
Magpie's core distribution provides a simple Web application infrastructure along with a few commonly useful state-to-event mapping mechanisms that have proven themselves useful over time, but makes no other presumptions or proscriptions about the design or implementation of a given application. The point is to reduce brittleness and redundancies while ensuring that developers are free to implement their application in the way that makes most sense to them.