i argue that the default way to structure a rails application is bad, and
subsequently encourages bad design because components of a logical unit are
spread all over the place: in the models
directory, the services
(the
mechanism the rails community seems to have settled on for bundling executive
code together), sometimes concerns
, etc.
look at discourse, my favorite rails project of all time, and it's almost
impossible to figure out the essence of the application. the entries in the
app/models
directory continues to grow with no end in sight. the
relationships between these models—who works with who and in what
capacity?—isn't immediately obvious.
categoria fixes this. it regroups data and executive code and separates them along domain lines. it's similar in spirit to phoenix's context. the point of departure is the following:
a category/domain doesn't concern itself with the web part of the application. that is because the web concerns tend to be cross-cutting (across domains). in a controller action, it's possible to invoke several commands from different domains in service of the request. the returned response might also be a combination of data from different domains. in order to allow this freedom, all app components that directly handle web requests can remain in their conventional locations.
what to call controllers? typically they have matched a given model. you have a
Document
model, here's the DocumentsController
. now that the document model
is subsumed under a domain of different name, the controller should be allowed
to float freely. it could be an interface to a domain, or to several domains at
the same time. an appropriate name should be chosen.
domains live under app/lib
of the rails application. internally, it is
organized into three main categories, represented by directories and ruby module
namespaces. all domains are sub-namespaced under the application's main
namespace. an initializer is added to override zeitwerk's default behavior of
using Object
as the root namespace for classes loaded from app/lib
.
the same structure is repeated under the test
directory, under universal
Test
namespace. just so constant loading, in non-test local and remote (prod)
environments don't load unnecessary code. the Test
namespace should only be
loaded in a test environment.