PHP Code Organization – Clever Application Structures

My current employment is within an enterprise however the team size is small at 5 individuals. On top of that we are dispersed geographically, 2 in one state, 1 in another state, 1 in another state, and 1 in another country.

Our success thus far has hinged on using cloud based solutions within a Scrum methodology. We use tools such as ScrumWorks free offering (since it can be hosted within our domain however VersionOne has a free offering that can be up and running within their domain) as the base of the process. We then leverage Planning Poker which is also another free offering.

We follow the typical Scrum process complete with a daily stand-up meeting. The stand-up meeting has by far been the best approach in attempting to bring about cohesion throughout the team and attempt to remove the isolation that can exist in this type of atmosphere. The daily stand-up coupled with a live form of communication (we use Office Communicator in the enterprise however any IM client would suffice) are critical to insuring a sense of community.

With varying native tongues amongst the team the telephone is a prominent form of communication for in depth conversation however the benefit tends to vary as it can sometimes make the communication much more difficult; as a given accent can sometimes make it difficult however as with anything the more you do it the better you become.

On the flip side I have worked within a team that was located within the same building and no technology or methodology has provided the type of cohesion that close proximity can provide. There was a recent study that provided *empirical evidence to the fact that close proximity of team members is beneficial to the long term goal of the project.

Bottom line is that you will need to find out what tools work best for you. Make sure that the entire team speaks their mind on what is and is not working or isolation will surface very rapidly.

*I can not find the link at the moment but will post it once I do.

EDIT:

Based on comments...

Source control is via svn which is addressed by the enterprise via a custom solution leveraging CollabNet. This solution also takes care of defect management at a higher level as we then port those defects to ScrumWorks to get placed on the sprint backlog. In addition it also provides access to a build environment for CI and deployment.

Testing is attached to the stories from a unit/integration stance; ie...a stories point estimation should and will include testing. That is not a separate component; it is part of the development effort. The testing framework used is NUnit and tests are then ran within the IDE (Visual Studio) via a tie in into NUnit.

This allows the developer to constantly check in and run their tests as needed. Since we are a small team; ownership on problems within the code base rarely happen and we do not implement a strict only check-in when the code has passed all tests mentality as that is not the purpose of version control. Being dispersed it is critical that we check in frequently.

If you do not have access to an enterprise solution your next best bet would be to leverage a VPS hosting solution as they can be had for fairly cheap ($20.00 US/month) and it would allow you to centralize your source, provide version control, DR, deployment of ScrumWorks and other needs as they surfaced.

1. Yes. And also assemblies.
2. I'd separate by layers, then components.
3. Yes. There are different approaches to this, but I'd have an IDatabaseService (abstracting the various manners in which the database is called -- this can almost be a direct mapping of the ExecuteScalar/ExecuteNonQuery/ExecuteReader), and then various data access classes that partition by type of data. For example, you could have a UserDataAccess class that would have simple CRUD methods that create/modify/delete User objects. Another approach would be to have a User object that has the CRUD built right in.

4. No. That makes it much harder to unit test. You should use dependency injection to pass in dependencies to the constructor of each data access class (such as an IDatabaseService). You would then pass the data access objects into the business objects, like this:

   BusinessObject businessObject = new BusinessObject(new DataAccessObject(new DatabaseService())); businessObject.PerformOperation();

Each business object may need multiple data access objects. Your GUI code would also use one or more business objects. Some business objects may not need any data access objects but should never use the IDatabaseService directly.