Egeria provides technology for an open standard that seeks to improve the processing and protection of data across organizations. For its developers, this carries the benefit that their work receives high recognition, but also additional responsibilities to ensure its wide applicability and longevity.
For example, Egeria seeks a broad audience - from developers to adopting vendors to consuming users. Building this audience and allowing the community to scale requires clarity in the way the software is written, documented, packaged and used. Many of the guidelines seek to make it easier for someone new to pick up the software, at the expense of maybe a little more work, or a little less freedom of action for the original developer.
As such, these guidelines exist to remind us of these broader responsibilities.
The minimum level required to build & run Egeria is Java 8. The code is also buildable under Java 11, but is configured for Java 8 language level and bytecode. It is also runnable under Java 11.
Non-LTS releases are not tested.
The Java distributions we are using include:
- Azul zulu - as provided by Azure Pipelines.
- openjdk - from Ubuntu distributions
- Adoptopenjdk openjdk (with hotspot)
- Adoptopenjdk openjdk (with J9)
All automated builds are performed on Linux (Ubuntu) only using the Azul vm provided by Azure pipelines.
Most developers use MacOS, but Windows should work also.
Problems with any of these should be raised as issues.
Build output should be checked for any warnings ie ‘[WARNING]’ and these should be eliminated.
For example the java compiler is set to use ‘-Xlint:all’ and may report warnings about deprecated function, unsafe casts, unchecked conversions etc which should be addressed.
Other tools used in the build may also result in warnings which should also be addressed, whilst testcases should ensure output is captured to avoid such warnings appear in the build logs.
License text in files
All files for Egeria should have a license included. We are using the Apache 2.0 license, which protects our code whilst still allowing commercial exploitation of the code. There is an example of the license text at the top of this file. The following files in the License-Example-Files directory have the correct license information formatted for different file types to make it easy to use.
Notice that the license information is coded using SPDX.
Although all code for Egeria should be clear and easy to read, the code itself can only describe what it is doing. It can rarely describe why it is doing it. Also, the Egeria codebase is quite large and hard to digest in one go. Having summaries of its behaviour and philosophy helps people to understand its capability faster.
README markdown files
Each directory (particularly code modules) should have a
README.md file that describes the
content of the directory. These files are displayed automatically by GitHub when the
directory is accessed and this helps someone navigating through the directory structures.
The exception is that directories representing Java packages do not need README files because they are covered by Javadoc.
Javadoc is used to build a code reference for our public site. It is generated as part of the build. There are three places where Javadoc should be provided by the developer of Java code:
- Every Java source file should begin with a header javadoc tag just before the start of the class/interface/enum, which explains the purpose and responsibilities of the code.
- All public methods should have a clear Javadoc header describing the purpose, parameters and results (including exceptions). This includes test cases.
- Each Java package should include a
package-info.javafile describing the purpose of the package and its content.
Java code files may have additional comments, particularly where the processing is complex. The most useful comments are those that describe the purpose, or intent of the code, rather than a description of what each line of code is doing.
The output from a build should be checked to ensure there are no javadoc warnings - for example about undocumented parameters or exceptions.
New dependencies must only be introduced with the agreement of the broader community. These include frameworks, utility classes, annotations and external packages. This may seem annoying but there are good reasons for this:
- The Egeria code needs to be embeddable in many different vendor products. This is made easier by keeping the code libraries we are dependent on to the minimum in order to avoid conflicts with libraries a consuming vendor may have already chosen.
- As developers, we have legal obligations to ensure we only use appropriately licensed software in our work and part of the discussion related a new dependency is to understand its license.
- Some projects may provide useful functionality but are only supported by one person who may get bored with it, or no longer have the time to support it. We should aim to build on dependent libraries that are backed by a strong community or vendor.
- Each library function, or set of annotations, adds to the learning curve of new people joining the team. By only bringing in the really beneficial libraries we ensure that the complexity they see relates only to the complexity of the problem space, rather than the additional complexity we have introduced in pursuit of playing with new functions.
If a developer wishes to introduce a new dependency to the Egeria project, they should prepare a short guide (in a markdown file) that explains the value of the new library, how it is to be used and links to more information. They should then present their recommendation to the community and and if agreed by the community, store the guide in the developer resources.
Once in place, the dependency should be maintained across the smallest appropriate number of modules, and should be consistent throughout. - particularly when it may impact consuming technologies.
For more on how dependencies are managed in the codebase refer to Dependency Management.
Coding style and layout
There are many coding and layout styles that provide clear and readable code. Developers can choose the layout they prefer but with the following restrictions/suggestions:
- Try to use full words rather than abbreviations or shortened versions of a word for names such as class names, method names and variable names. Cryptic names create more effort for the reader to follow the code.
- Use the same style throughout a file.
If changing an existing file, use the same style and layout as the original developer. Don’t impose your own style in the middle of the code since the inconsistency that you introduce makes the whole file harder to read. It should not be possible to see where you have made the changes once the code is committed into git.
- For java unit tests use /src/test/java folder of the module (standard maven location), and postfix java file names with “Test”.
All code submissions should be accompanied by automated tests that validate the essential behaviour of the code. These automated tests should be incorporated in the build so that they run either at the test or verify stages of the build.
License: CC BY 4.0, Copyright Contributors to the ODPi Egeria project.