[java] Are there best practices for (Java) package organization?

A little while ago, I saw a question answered here regarding the fine-grained organization of java packages. For example, my.project.util, my.project.factory, my.project.service, etc.

I can't find it now, so I may as well ask the question.

Are there best practices with regards to the organization of packages in Java and what goes in them?

How do you organize your classes in your Java project?

For instance, a project I'm working on with a few people has a package called beans. It started out being a project containing simple beans, but has ended up (through poor experience and lack of time) containing everything (almost). I've cleaned them up a little, by putting some factory classes in a factory package (classes with static methods that create beans) but we have other classes that do business logic and others that do simple processing (not with business logic) like retrieving a message for a code from a properties file.

Your thoughts and comments are appreciated.

I organize packages by feature, not by patterns or implementation roles. I think packages like:

• beans
• factories
• collections

are wrong.

I prefer, for example:

• orders
• store
• reports

so I can hide implementation details through package visibility. Factory of orders should be in the orders package so details about how to create an order are hidden.

Short answer: One package per module/feature, possibly with sub-packages. Put closely related things together in the same package. Avoid circular dependencies between packages.

Long answer: I agree with most of this article

Are there best practices with regards to the organisation of packages in Java and what goes in them?

Not really no. There are lots of ideas, and lots opinions, but real "best practice" is to use your common sense!

(Please read No best Practices for a perspective on "best practices" and the people who promote them.)

However, there is one principal that probably has broad acceptance. Your package structure should reflect your application's (informal) module structure, and you should aim to minimize (or ideally entirely avoid) any cyclic dependencies between modules.

(Cyclic dependencies between classes in a package / module are just fine, but inter-package cycles tend to make it hard understand your application's architecture, and can be a barrier to code reuse. In particular, if you use Maven you will find that cyclic inter-package / inter-module dependencies mean that the whole interconnected mess has to be one Maven artifact.)

I should also add that there is one widely accepted best practice for package names. And that is that your package names should start with your organization's domain name in reverse order. If you follow this rule, you reduce the likelihood of problems caused by your (full) class names clashing with other peoples'.

I'm not aware of standard practices for package organization. I generally create packages that cover some reasonably broad spectrum, but I can differentiate within a project. For example, a personal project I'm currently working on has a package devoted to my customized UI controls (full of classes subclassing swing classes). I've got a package devoted to my database management stuff, I've got a package for a set of listeners/events that I've created, and so on.

On the other hand I've had a coworker create a new package for almost everything he did. Each different MVC he wanted got its own package, and it seemed a MVC set was the only grouping of classes allowed to be in the same package. I recall at one point he had 5 different packages that each had a single class in them. I think his method is a little bit on the extreme (and the team forced him to reduce his package count when we simply couldn't handle it), but for a nontrivial application, so would putting everything in the same package. It's a balance point you and your teammates have to find for yourself.

One thing you can do is try to step back and think: if you were a new member introduced to the project, or your project was released as open source or an API, how easy/difficult would it be to find what you want? Because for me, that's what I really want out of packages: organization. Similar to how I store files in folder on my computer, I expect to be able to find them again without having to search my entire drive. I expect to be able to find the class I want without having to search the list of all classes in the package.

I prefer feature before layers, but I guess it depends on you project. Consider your forces:

• Dependencies
  Try minimize package dependencies, especially between features. Extract APIs if necessary.
• Team organization
  In some organizations teams work on features and in others on layers. This influence how code is organized, use it to formalize APIs or encourage cooperation.
• Deployment and versioning
  Putting everything into a module make deployment and versioning simpler, but bug fixing harder. Splitting things enable better control, scalability and availability.
• Respond to change
  Well organized code is much simpler to change than a big ball of mud.
• Size (people and lines of code)
  The bigger the more formalized/standardized it needs to be.
• Importance/quality
  Some code is more important than other. APIs should be more stable then the implementation. Therefore it needs to be clearly separated.
• Level of abstraction and entry point
  It should be possible for an outsider to know what the code is about, and where to start reading from looking at the package tree.

Example:

com/company/module
  + feature1/
    - MainClass          // The entry point for exploring
    + api/               // Public interface, used by other features
    + domain/
      - AggregateRoot
      + api/             // Internal API, complements the public, used by web
      + impl/ 
    + persistence/       
    + web/               // presentation layer 
    + services/          // Rest or other remote API 
    + support/            
  + feature2/
  + support/             // Any support or utils used by more than on feature
    + io
    + config
    + persistence
    + web

This is just an example. It is quite formal. For example it defines 2 interfaces for feature1. Normally that is not required, but could be a good idea if used differently by different people. You may let the internal API extend the public.

I do not like the 'impl' or 'support' names, but they help separate the less important stuff from the important (domain and API). When it comes to naming I like to be as concrete as possible. If you have a package called 'utils' with 20 classes, move StringUtils to support/string, HttpUtil to support/http and so on.

I've seen some people promote 'package by feature' over 'package by layer' but I've used quite a few approaches over many years and found 'package by layer' much better than 'package by feature'.

Further to that I have found that a hybrid: 'package by module, layer then feature' strategy works extremely well in practice as it has many advantages of 'package by feature':

• Promotes creation of reusable frameworks (libraries with both model and UI aspects)
• Allows plug and play layer implementations - virtually impossible with 'package by feature' because it places layer implementations in same package/directory as model code.
• Many more...

I explain in depth here: Java Package Name Structure and Organization but my standard package structure is:

revdomain.moduleType.moduleName.layer.[layerImpl].feature.subfeatureN.subfeatureN+1...

Where:

revdomain Reverse domain e.g. com.mycompany

moduleType [app*|framework|util]

moduleName e.g. myAppName if module type is an app or 'finance' if its an accounting framework

layer [model|ui|persistence|security etc.,]

layerImpl eg., wicket, jsp, jpa, jdo, hibernate (Note: not used if layer is model)

feature eg., finance

subfeatureN eg., accounting

subfeatureN+1 eg., depreciation

*Sometimes 'app' left out if moduleType is an application but putting it in there makes the package structure consistent across all module types.