There is a specific way of organizing a DFDL schema project that has been found to be helpful. It uses specific directory naming conventions and tree shape to manage name conflicts in a manner similar to how Java package names correspond to directory names.
This set of conventions provides a number of benefits:
- No name conflicts or ambiguity on classpath if multiple DFDL schemas are used together
- You can just copy a bunch of DFDL schemas into one directory tree and there will be no conflicts of file names.
- sbt can be used to
- Package the schema into a jar - The jar can then be on a classpath, become part of a larger application, etc.
- Auto-download all dependencies of the schema, including Daffodil itself.
- Run a suite of tests via 'sbt test'
- Publish a local version of the schema for use in other projects that also follow this layout.
- Eclipse IDE for development and test of the schema. Multiple such schemas all work together without conflict in the IDE.
- Encourages organizing DFDL schemas into reusable libraries.
- A DFDL schema project need not define a whole data format. It can define a library of pieces to be included/imported by other formats.
These conventions are actually usable for regular XML-schema projects, that is, they're not really DFDL-specific conventions. They're general conventions for organizing projects so as to achieve the above benefits.
We're using Tresys Technology (www.tresys.com) here as an example. Substitute your organization's details.
Let's assume the DFDL schema contains two files named main.dfdl.xsd, and format.dfdl.xsd, and that our format is named RFormat.
The standard file tree would be:
After you run 'sbt test', you'll notice a
lib_managed directory has been created. This directory is created by sbt to hold all the dependencies of the project.
Use the below template for the build.sbt file:
Edit the version of daffodil-tdml above to match the version you are using.
If you organize your DFDL schema project using the above conventions, and then run 'sbt compile', the lib_managed directory will be populated. Then if you create a new Eclipse scala project from the directory tree, Eclipse will see the lib_managed directory and construct a classpath containing all those jars.
DFDL schemas should have the ".dfdl.xsd" suffix to distinguish them from ordinary XML Schema files.
A DFDL schema should have a target namespace.
Stylistically, the XSD elementFormDefault="unqualified" is the preferred style for DFDL schemas.
Using a DFDL Schema
The xs:include or xs:import elements of a DFDL Schema can import/include a DFDL schema that follows these conventions like this:
The above is for using a DFDL schema as a library, from another different DFDL schema.
Within a DFDL schema, one DFDL schema file can reference another peer file that appears in the same directory (the src/main/resources/.../xsd directory) via:
That is, peer files need not carry the long "com/tresys/RFormat/xsd/" prefix that makes the reference globally unique.
However, if one schema wants to include another different schema, then this standard way of organizing schema projects insures that when packaged into Jar files, the /src/main/resources directory contents are at the "root" of the jar file so that the
schemaLocation of the
xs:include containing the fully qualified path (
"com/tresys/RFormat/xsd/main.dfdl.xsd") will be found on the CLASSPATH unambiguously. This convention is what allows the schema files themselves to have short names like
format.dfdl.xsd. Those names only need to be unique within a single schema project. Across schema projects our standard DFDL schema project layout insures unambiguous qualification is available.
Git Revision Control
You don't have to use Git version control, but many people do, and github.com is one of the reasons for this popularity.
Each DFDL schema should have its own Git repository if it is going to be revised independently. We encourage users to join the DFDLSchemas project on github and create repositories for, and publish schemas for any publicly-available formats there. For other formats that are not publicly available, one may want to put a placeholder for them on DFDLSchemas anyway (as IBM has done for some formats like Swift-MT.)
Jar File Packaging
A DFDL schema using the recommended file structure as described here, can be packaged into a jar for convenient import/include from other schemas.
The sbt command does all the work:
The resulting jar has the
src/main/resources directory in it at the root of the jar. If this jar is on the classpath, then other schemas containing XSD import or include statements will search the jar with the schema location.
That enables a different schema's build.sbt to contain a library dependency on our hypthetical dfdl-RFormat schema using a dependency like this:
That will result in the contents of the
src/main/resources directory above being on the classpath. XSD include and import statements search the classpath directories.