Multiple issues can appear when the Single Responsibility Principle (SRP) is violated.
Harder to read and understand
A controller violating SRP is harder to read and understand since its Cognitive Complexity is generally above average (see
S3776).
For example, a controller MediaController
that is in charge of both the "movies" and "photos" APIs would need to define all the
actions dealing with movies, alongside the ones dealing with photos, all defined in the same controller class.
The alternative is to define two controllers: a MovieController
and a PhotoController
, each in charge of a smaller
number of actions.
Harder to maintain and modify
Such complexity makes the controller harder to maintain and modify, slowing down new development and increasing the likelihood of bugs.
For example, a change in MediaController
made for the movies APIs may inadvertently have an impact on the photos APIs as well.
Because the change was made in the context of movies, tests on photos may be overlooked, resulting in bugs in production.
That would not be likely to happen when two distinct controllers, MovieController
and a PhotoController
, are
defined.
Harder to test
The controller also becomes harder to test since the test suite would need to define a set of tests for each of the
responsibilities of the controller, resulting in a large and complex test suite.
For example, the MediaController
introduced above would need to be tested on all movies-related actions, as well as on all
photos-related actions.
All those tests would be defined in the same test suite for MediaController
, which would be affected by the same issues of
cognitive complexity as the controller under test by the suite.
Harder to reuse
A controller that has multiple responsibilities is less likely to be reusable. Lack of reuse can result in code duplication.
For example, when a new controller wants to derive from an existing one, it’s less probable that the new controller requires all the behaviors
exposed by the reused controller.
Rather, it’s much more common that the new controller only needs to reuse a fraction of the existing one. When reuse is not possible, the only
valid alternative is to duplicate part of the logic in the new controller.
Higher likelihood of performance issues
A controller that has multiple responsibilities may end up doing more than strictly necessary, resulting in a higher likelihood of
performance issues.
To understand why, it’s important to consider the difference between ASP.NET application vs non-ASP.NET applications.
In a non-ASP.NET application, controllers may be defined as Singletons via a Dependency Injection library.
In such a scenario, they would typically be instantiated only once, lazily or eagerly, at application startup.
In ASP.NET applications, however, the default is that controllers are instantiated as many times as the number of requests that are served by
the web server. Each instance of the controller would need to resolve services independently.
While service instantiation is typically handled at application startup, service resolution happens every time an
instance of controller needs to be built, for each service declared in the controller.
Whether the resolution is done via Dependency Injection, direct static access (in the case of a Singleton), or a Service Locator, the cost of resolution needs to be paid at every single
instantiation.
For example, the movies-related APIs of the MediaController
mentioned above may require the instantiation of an
IStreamingService
, typically done via dependency injection. Such a service may not be relevant
for photos-related APIs.
Similarly, some of the photos-related APIs may require the instantiation of an IRedEyeRemovalService
, which may not work at all
with movies.
Having a single controller would force the developer to deal with both instantiations, even though a given instance of the controller may be
used only for photos, or only for movies.
More complex routing
A controller that deals with multiple concerns often has unnecessarily complex routing: the route template at controller level cannot factorize the
route identifying the concern, so the full route template needs to be defined at the action level.
For example, the MediaController
would have an empty route (or equivalent, e.g. /
or ~/
) and the actions
would need to define themselves the movie
or photo
prefix, depending on the type of media they deal with.
On the other hand, MovieController
and PhotoController
can factorize the movie
and photo
route respectively, so that the route on the action would only contain action-specific information.
What is the potential impact?
As the size and the responsibilities of the controller increase, the issues that come with such an increase will have a further impact on the
code.
- The increased complexity of reading and understanding the code may require the introduction of regions or partial classes, to be
able to visually separate the actions related to the different concerns. Those are patches, that don’t address the underlying issue.
- The increased complexity to maintain and modify will not give incentives to keep the architecture clean, leading to a more tightly
coupled and less modular system.
- The reduced reusability of the code may bring code duplication, which breaks the Don’t repeat
yourself (DRY) principle and which itself comes with a whole lot of issues, such as lack of maintainability and consistency.
- The performance penalty of conflating multiple responsibilities into a single controller may induce the use of techniques such as lazy service
resolution (you can find an example of this approach here). Those increase the
complexity of the code and make the runtime behavior less predictable.
Why MVC controllers are not in scope
Alongside attribute
routing, which is typical of web APIs, MVC controllers also come with [conventional routing].
In MVC, the file structure of controllers is important, since it drives conventional routing, which is specific to MVC,
as well as default view mapping.
For those reasons, splitting an MVC controller into smaller pieces may break core behaviors of the web application such as routing and views,
triggering a large refactor of the whole project.