Permalink
Find file Copy path
Fetching contributors…
Cannot retrieve contributors at this time
152 lines (114 sloc) 5.47 KB

Design Notes

Some of the use cases that influenced Morepath's design are documented here.

Publish any model

It should be possible to publish any model object to the web on a readable URL. This includes model objects that are retrieved from a relational database and were created with a ORM.

Allowing individual models to be published on separate URLs avoids the god object antipattern where all web operations are routed through a single object. Instead each model, through view objects, can handle model-specific requests and operations. This encourages a more modular and reusable application design.

Routing

It should be easy to declare explicit routes to model. A route consists of a routing pattern with zero or more variables. The variables are used to identify the model, for instance using a relational database query.

Having routes makes it easier to reason about the URL structure of an application. Routes also make it easier to expose models that are retrieved using a query or are constructed on the fly, without imposing a specific structure on the models.

Traversal

It should be possible to associate routes with specific models in the application, not just to the root. This way models with sub-paths to sub-components can be made available as reusable components; an example of this could be a container. If the model is published, its sub-components are then exposed as well.

This allows for increased reuse of not just models but relationships between models, and lets the developer publish nested structures that cannot be specified using routing alone.

Linking

If a model is published, it should be possible to automatically generate a link to a model instance in the form of a URL.

This way there is no need to construct URLs manually, and there is no need to have to refer to routes explicitly in order to construct URLs. The system knows which route to use and how to construct the parameters that go into the route itself, given the model.

This is useful when creating RESTful web services (where hypermedia is the engine of application state), or to construct rich client-side applications that get all their URLs from the server from a REST-style web service.

Model is web-agnostic

Model classes should not have to have any web knowledge; no particular base classes are required, and no methods or attributes need to be implemented in order to publish instances of that model to the web. In case of an ORM, the ORM does not need to be reconfigured in order to publish ORM-mapped classes to the web. Models do not receive any request object and do not have to generate a response object.

Instead this knowledge is external to the models. Models should be optimized for programmatic use in general.

View/model separation

View objects are responsible for translating the model to the web and web operations to operations on the model. Views receive the request object and generate the response object. This is again to avoid giving the models knowledge about the web. This is a kind of model/view separation where the view is the intermediary between the model and the web.

Isolation between applications

The system allows multiple applications to be published at the same time. Applications work in isolation from each other by default. For instance, publishing a model on a URL does not affect another application, and publishing a view for a model does not make that view available in the other application.

Sharing between applications

In order to support reusable components, it should be possible to explicitly break application isolation and make routes to models and views globally available. Each application will share this information.

[Morepath in fact now allows more controlled sharing; only Morepath itself is globally shared]

Models can be published once per application

Per application a model can be exposed on a single URL pattern. So, the same instance could be published once per application, in a URL structure optimal for each application.

Again this supports applications working in isolation - they may treat database models differently than other applications do.

Linking to another application

It should be possible to construct URLs to models in the context of another application, if this application is given explicitly during link time.

Reusable components

It should be possible to define a base class (or interface) for a model that automatically pulls in (globally registered) views and sub-paths when you subclass from it. This lets a framework developer define APIs that an application developer can implement. By doing so, the application developer automatically gets a whole set of views for their models.

Declarative

It should be possible to register the components in a declarative way. This avoids spaghetti registration code, and also makes it possible to more easily reason about registrations (for instance to do overrides or detect conflicts).

Conflicts

If you try to do the same registration multiple times, the system should fail explicitly, as otherwise this would lead to subtle errors.

Overrides

It should be possible to override one registration with another one. This should either be an explicit operation, or the result of overriding in a different registry that has precedence over the defaults.