|
|
|
This documents presents best-practices for development and architecture related to The Open For Business Project.
It will not try to cover all of the practically limitless options available for using tools in OFBiz and the many related open source projects and standard APIs. It will also not try to cover all development best-practices, just the ones most closely related to creating and modifying OFBiz based components.
It will try to cover the best practices for every layer of architecture and the best tools to use, and for steps that should be taken prior to even laying out what code should be written. In some cases there will be close seconds to the best practices that will be presented because in some cases these second-best-practices will be more appropriate.
Note that this document assumes some knowledge of the OFBiz Core Framework. With some basic knowledge it can answer a lot of questions about how certain things should be used. If you are not familiar with the OFBiz Core Framework this will give you an introduction to it, but some things may not make a lot of sense to you.
There are various techniques that can be applied to this problem that produce good results. The most common, but least effective, practice is code generation. Because it is not something that I consider to be a best practice, I won't discuss it further here.
The best practices that are used in various places in the OFBiz framework are dynamic APIs such as the Entity and Service Engines, and special purpose languages like the Workflow and Rule Engines and the MiniLang library.
The dynamic API pattern is characterized by a generic API with simple operations that behave differently based on configuration and domain definition files. These are usually XML files. This is an alternative to code generation and the input or domain description files used for generating code can often be used unchanged to drive a dynamic API resulting in much less code and much more dynamic, ad-hoc control.
A special purpose language is used to create logic in a language or using a tool that fits the specific need more appropriately than a general purpose procedural language like Java. This reduces code because it is easier to describe what you want in a context that closely matches the problem that needs to be solved than it is to use a generic language. High level conepts can be expressed without the need to write a lot of code.
The best-practice tool to use in the data layer is the OFBiz Entity Engine. For most applications the Entity Engine will elegantly do the work for 99% of your database interaction needs. In the few cases where the Entity Engine is not sufficient I recommend using custom JDBC code for your queries or other commands. That would be one of the second-best-practices that are sometimes needed.
When using the Entity Engine refer to entity and field names using inline strings. This makes it much easier to read and maintain your code. If you need to prepare a large Map or EntityCondition to pass to an EE method it is generally cleaner to do it on a separate line, or on various separate lines before the actual EE method call.
Use a simple, normalized data model based on the needs of your applications. Usually the data model can be driven directly by the requirements for the functionality that will use the entities. This usually results in a highly normalized data model which will make your life much easier. When you need combined data for reporting, use the view-entity feature to accomplish joins and grouping and summarizing data.
Always use primary keys and avoid the use of generic sequenced primary keys when a more descriptive composite key is possible. Always use relationship definitions to document how entities are used together, to make it easier to get at related data, to constrain field by foreign keys, and to improve performance through automatic foreign key based indexes.
The best tool to use for invoking logic is the OFBiz Service Engine. Nearly all business logic should be implemented as a service to improve reusability and facilitate component based development.
Even though services are very flexible there are cases where the service model is not appropriate, even for business logic. In some cases calling a script or Java method directly is necessary and in those cases using the service model would not make sense and should not be used.
The Service Engine reduces code size by providing many ways to use logic implemented as a service. You can call your logic synchronously, asynchronously or on a schedule. When you are calling a service you don't need to know where it is located or how it is implemented. This makes it easy to effectively leverage remote services and services written using different languages. Being able to transparently call logic in different languages through the Service Engine is important for the effective use of special purpose languages.
Always implement your service using the easiest and most appropriate language or tool. You can implement services with many different languages including Java, BeanShell or any BSF compliant scripting language (Jython, Jacl, JavaScript, etc), OFBiz Workflow Engine processes, OFBiz MiniLang simple-methods, and various others. Additional languages can be supported by writing simple adapters.
Always call remote logic through the Service Engine. You can call remote services through various mechanisms including HTTP, SOAP, JMS, and others. You can also add remote service invocation mechanisms by creating a simple adapter.
Most of the time you will want to let the Service Engine automatically wrap your service call in a transaciton so that the whole thing will succeed, or the whole thing will fail. Note that if you call a service inside an existing transaction it will recognize the current transaction and use it instead of trying to create another one.
Always separate input processing logic, view data preparation logic, and view presentation templates. This will make it easy to reuse logic not only in web applications, but also for independent fat client applications. It will also make it easier to organize your code and find a specific piece of functionality when debugging or exploring to find out how a component works.
For each of the three separate pieces there are special best-practices tools to use.
Input processing logic should always be associated with a request in the controller.xml file and never with a view. Input processing logic should generally be implemented as a service and called through the service event handler which will automatically pull data from request parameters or attributes and convert it from a string to the object type defined in the service definition. This makes it easy to specify which parameters you care about processing just using the service definition, and let the framework get them ready for you.
There are various cases where input processing logic cannot be implemented as a service. There are various other types of event handlers for logic associated with requests that give you more to the request context and are not environment agnostic like services are. One example is receiving uploaded data. Another good example is doing special pre-processing and validation on parameters before passing them to a service for processing. Note that you can always call services from these custom events and wherever possible generic logic should be implemented in services.
Always let the Control Servlet configuration handle decisions about the appropriate response to take for a request given the result string from an event. In most cases the response will be the generation of a view, but sometimes it will make sense to chain requests together to acheive logic reuse or more advanced flow control.
View data preparation logic should always be associated with the view template it is meant to prepare data for. This should be done through JPublish in the page definition XML file by specifying an action. When a page is split up into multiple templates the data preparation action should be associated only with the individual small template that it prepares data for. This makes it easier to move templates and content pieces around and reuse them in many pages.
View data preparation logic should be implemented in a dynamic scripting language such as JavaScript, BeanShell, Jython or Jacl to make it easy to modify the user interface on the fly. Generic data retreival should be implemented as services which should be called from these dynamic action scripts. This makes it easier to share and reuse this functionality in multiple pages and in other types of user interfaces.
When preparing data in view actions you should make the data available to the view template by putting it in the "context" object. All attributes in the context object will be made avilable in the context of the template, if the template languge supports that.
Note that when using JSP for a view template you cannot use JPublish so the actions facility will not be available. Our recommendation for JSPs is to have a single scriplet at the top of the page that prepares the data. In this case try to call worker services or worker Java methods to do most of the work and to keep as much logic as possible out of the page.
The best-practice template engine that we recommend for HTML and other text generation is FreeMarker. It is like Velocity from Jakarta, but much more flexible and fits in nicely with other OFBiz Core Framework tools. Rather than running FreeMarker templates directly we strongly recommend using JPublish so that actions can be associated with pages and templates and pages can be decorated with common templates. We'll describe how to best use this below.
The view presenation templates should always be kept as simple as possible and common content such as headers, footers, sidebars, and so on should be added at run-time using the decoration pattern. The template file that should be used to decorate each page is specified in the JPublish page definition XML file.
Always use the view generation tool that most closely matches your needs. FreeMarker is the recommended tool for generating text output, but there are many situations where other tools are more appropriate. When you want to display report like views we recommend using Jasper Reports or DataVision and mounting those reports through a view-map in the controller.xml file. If you want to use other text generation tools such as Velocity or XSLT we recommend you do so through JPublish, especially if you want it to be decorated and have actions run to prepare for the templates.
If you have UI patterns that are repeated frequently such as forms, query data displays, tab or menu bars, expanding tree views, and so forth we currently recommend using an XML file to describe the UI pattern and then transform it using XSLT into a FreeMarker template which will be run to merge the resulting template with data from the context. Note that this recommendation may change as other methods are tried and refined. This is a problem with many possible solutions.
When using FreeMarker is not possible or practical we recommend using another dynamic templating language such as Velocity. When that is also not possible or practical we recommend using JSPs. But, note that when using JSPs you cannot take advantage of the actions or decoration templates because you cannot run it through JPublish. This is thanks to limitations in the JSP specification. Even through you cannot use the decorator patter, you can use the composite view pattern with the OFBiz Regions framework. Regions are specified in the regions.xml file. Note that these are not as easy to use as JPublish composite views, and they do not support actions. But the Regions framework does offer a lot of flexibility and is very useful in many cases.
Before Starting - Methodology Recommendations
Before you can build something that something has to be defined. Every detail has to be decided at some point. Because of the difficulty of communication between individuals and the difficulties that are inherently involved in complex systems we recommend various practices.
While these techniques are used for some components in the core Open For Business functionality they are primarily meant for those who are creating derivative works based on OFBiz. The reason for this is that OFBiz is a generic framework, set of application components and suite of applications. In many cases these applications can be used as is but they are not meant to be everything to everyone. Because of this work will likely be needed for derivative works that are targeted at a specific industry, type of business, or specific end user.
Methodology: Keep it simple, but not too simple
There are dozens of different software development methodologies, and more are being introducted all the time. There are lots of good ideas presented in the hundreds, or even thousands, of documents available about different methodologies. In general we recommend the "agile" approach, but that needs to be qualified given the different perspectives that have been attached to that name. In general we recommend keeping it as simple as possible, but not too simple. For more specifics on agile software development see the Agile Alliance web site.
So the big question is generally: what is the minimum that we need to handle this project? The more methodology you have the more time and money you will spend on it. The less methodology you have the greater the chance that the requirements of the project will not be satisfied.
There are really only two roles necessary for any project: a buyer and a seller. In the case of software projects, and other service offerings, the more common terms are customer and developer. Those are the terms that will be used in this document. The term developer is used loosely here and refers to anyone that produces actual deliverables. The term customer is also used loosely here and refers to anyone that specifies what will be done and pays for it to be done. In some cases the customer and developer will be the same person, or the same group of people.
It should be noted here that an important part of the "customer" role is the actual target user(s) of the software. If possible reviewing stories and use cases (as described below) with users should be done. Interviewing those who are currently performing tasks that will in the future involve interaction with the new system can make it much easier to acheive complete and effective results. This is true on many levels. For instance, the best usability can be acheived by reviewing screen/page prototypes with real users and then validating final deliverables with real users. If the user group is large this may require a lot of work to get a good sampling of the different users of the system.
In many projects the two roles, customer and developer, will consist of groups of people and other roles will be introduced within the groups to clarify responsibilities. In these cases the methodology may also have to be more significant to keep things from deteriorating into chaos.
One practice that can reduce the need for a complicated and detailed methodology is to have one person from each group act as the agent for the entire group. Of course, that agent should discuss things with the rest of the group, but if the number of voices for interaction between the groups can be consolidated to one then communication and management of details will be much easier.
Note that for large projects having one agent for each side will not be practical. In those cases dividing the scope of the project and likewise dividing the customer and developer groups will be necessary.
Another variation on this is where a group creates a product targeted at hundreds or thousands of customers. In this case the "customer" that defines the product will be part of the same organization that produces the product. Given the definition of customer above the sales, marketing, and testing groups should be in the customer group, and an agent from that combined group should coordinate with the development group.
The biggest trick with communication is to find the balance between to much and too little communication. The burden for this is generally on the customer because it is customer that specifies what will be build. Like I said at the beginning of this section everything that will be built MUST be defined sooner or later. Anything that is not specified by the customer can be handled in one of two ways. The developer can decide what will fit into the gaps or the developer can request clarification and details from the customer.
To facilitate communication and make sure that the final result satisfies the needs requested by the customer the following process is recommended.
One important aspect of this process is that it should be repeated frequently and the scope for each pass should be kept small so that when miscommunications occur they can be identified and resolved quickly.
While this process is simple and involves simple roles and artifacts it can be used as a skeleton for a much larger and more detailed process. Each step can be divided into many steps and extra steps to compensate for the needs of a specific project can be added.
Also note that this general process is the basis of the Work Effort components that are part of OFBiz and those tools, as well as many other available tools, can be used to make artifacts and other information easier to manage.
You will notice that this process involves a great deal of customer interaction. If done right the customer and developer will interact on nearly a daily basis. For coordination and management periodic meetings should be held to review new requests or changes to existing requests and to adjust priorities as necessary. These meeting should be held once every one or two weeks for most projects.
After looking at this list you may have questions about what constitutes a story, use case, and task.
A story is essentially an informal free form document from the customer that describes what is desired in complete sentences or as an outline. Precise details are not needed at this stage. Note that this document may change to be more precise or to include additional aspects. This document will be used to create more formal use case(s).
A use case is a document that contains detailed information in a more formal structure and represents the complete specification corresponding to one or more stories. A use case should include the following:
Note that for simplicity scenarios should be kept small and include only the minimal basic process. Variations on process should be defined separately. Exceptions and errors in the process should be defined separately along with their results.
Defining tasks is generally fairly simple and should be done by the developer in a form and using terms that the developer will understand. There is a lot of meta-data that is generally associated with such a task. Examples include esimated and actual time and cost, start and end dates, all parties performing or otherwise involved with the task, and so on.
Some basic preparations should always be done before getting into the implementation. Some basic principles and a basic process has been presented. Because of the nature of how most projects work all of the steps in the process will be done in one way or another. Following this process does not require the creation of any written documents, although in many cases that will help quite a lot. It can be done through verbal conversation, or even a series of thoughts inside one's head.
The process described is meant to be customized to your needs. Following it as is should do for most medium sized projects. When there is only one or two individuals involved, or tens or hundreds of individuals, changes will obviously be desired. When modifying the process keep in mind the basic principles presented and be especially careful about leaving out communication steps, espcially the communication verification steps.
Managing Your Source Differences
When customizing OFBiz components for your specific business requirements you will most likely come across situations where small changes to files are needed and it doesn't make sense to replace a component completely or attach logic to it through Service ECA rules or other means. So, how should you manage the differences between your code and the OFBiz baseline code so that updating from the OFBiz baseline will be as painless as possible?
The first thing you should always do is isolate your chagnes and comment them so that it is easy to distringuish what you changed from the OFBiz baseline code. The final goal is to make it easy to merge your changes with future changes in the OFBiz baseline.
There are various ways to structure your internal CVS or other code repository. The most flexible, but laborious, way is to create a complete copy of the OFBiz CVS tree in a module in your own CVS repository. This could perhaps be called "ofbizbase". This should contain the revisions of the code that you are currently working with and allows you to selectively incorporate specific changes from the OFBiz baseline. These files should be direct copies from a revision in the OFBiz CVS repository. If you make changes in this module that are not part from the OFBiz baseline they should be temporary changes that are meant to be overwritten and replaced with future versions from the OFBiz baseline.
Once your ofbizbase snapshot module as described above is in place you should create another module, maybe called "ofbizdiff". This will contain ONLY the files that you have changed relative to your ofbizbase module. When updating the version of a file in your ofbizbase repository from the OFBiz baseline you should do a three way diff/merge to make sure the changes in the OFBiz baseline make it into the version of the file in your ofbizdiff module as well as the file in your ofbizbase module.
The three way diff/merge will be done with the following three files: the old file from your ofbizbase module as the base, your changed file in your ofbizdiff module as the merge target, and the new file from the OFBiz baseline as the merge source.
The build process for this sort of a structure is very simple. Create a temporary build directory. Copy the entire ofbizbase tree into the directory. Copy the entire ofbizdiff tree into that directory, replacing files as needed. In the root of the temporary build directory run the Ant build file. What could be more simple?
When editing files in your ofbizdiff module you should ALWAYS edit them in place in the ofbizdiff directory tree and copy them to the temporary build directory for testing. Doing this you will always be able to do a CVS update to see which files you have modified as you are working.
Many variations on this technique can be effectively used. If you don't plan to make temporary changes to ofbizbase files or update individual ofbizbase files selectively, you may do just fine with an archive (like a zip or tar.gz file) instead of a CVS module for your ofbizbase tree. You could keep the current version of this archive in your ofbizdiff CVS module.
We have heard a lot of questions about using simple merge scripts to apply differences to source in the ofbizdiff tree so that full files do not have to be stored ant maintained there. This is fine, but it is a LOT harder to perform the diff/merge or other code synchronization when you don't have the complete files to do a three way diff/merge. So, we don't recommend using minimal diff files that can be applied as part of preparation for the build process. Just copy the file from ofbizbase to ofbizdiff and make the changes you need as recommended above.
For information about Open For Business contact David E. Jones at [email protected] or Andy Zeneski at [email protected].