EDT:Extensible Compiler Development

This page covers the work being done to make the compiler more extensible. Validation, for example, will be made extensible. A new model will be created so that bound AST will use IRs instead of bindings. The old binding model will still exist so that tooling written against it does not need to be rewritten. New tooling will be encouraged to use the new IR-based model.

Where to get the code

Git repository: ssh://<user>@git.eclipse.org/gitroot/edt/org.eclipse.edt.git

Projects:
org.eclipse.edt.compiler
org.eclipse.edt.mof
org.eclipse.edt.mof.egl

Branch:
extensibleCore

How to run the code

Right now you can only compile in SDK mode (the IDE builder will NOT work). Take a look at the "EGL2IR XML" Java Application launch configuration that's predefined. This will prompt you a couple times if you run it, or you can create a copy of this launch configuration and hard-code some values if you don't want to be prompted. Here's an example of hardcoded arguments:

-eglpath "/home/justin/.workspaces/edt-ext/test/EGLSource/" -irDestination "/home/justin/.workspaces/edt-ext/test/batchbin" -xmlOut "/home/justin/.workspaces/edt-ext/test/EGLSource/pkg/prog.egl" -extensions "org.eclipse.edt.mof.eglx.jtopen.IBMiExtension,org.eclipse.edt.mof.eglx.persistence.sql.SQLExtension"

Parser Updates

Several changes to the parser/lexer grammer files will be made:

1. Remove primitive types
2. Add ability for name types to include parameters. This is to support parameterizable types like eDecimal(5,2). This will requie the new expression to change, since the arguments to constructors will now be on the type.
3. Refactor the isNullable flag to be on fields, function returns and array elements, instead of being on the type.
4. Remove old keywords, such as FIELD and SQLNULLABLE from function parms

Annotation changes

To make the EGL code more consitant and readable, EDT will require all references to annotations to be preceded by a @. Previously, for annotations with a single value, we allowed the annotation to be specified without using the @. For example, consider the following annotation:

Record stuff type Annotation {targets = [fieldMember]}

    value boolean = false;

end

In RBD we allowed the following for this annotation:       myfield record1 {stuff = true}

However, this is confusing, since there is no indication if "stuff" is an annotation or a field in the record Record1.

To alleviate this, EDT will require that the user explicitly indicate all annotations. Therfore, the user will need to code the following to specify the annotation: 

    myfield record1 {@stuff{value = true}};

or

   myField record1 {@stuff{true}};

 

 

Another change to annotation definition is how stereotypes can be defined. Previously, to define a stereotype, you would write:

record myStereotype type annotation { @stereotype{}}

We will now support the following (in addition to the old syntax):

record myStereotype type stereotype {}

 

 

The old resolvable type definitions are being removed. These included: fieldRef, fieldInTargetRef, functionMemberRef, partRef, typeRef, recordRef, and serviceRef.

These will be replaced with the proxy types defined in egl.lang.reflect.ProxyTypes.egl. the following is the mapping fo the old value to the correct new values:

 

fieldRef -> egl.lang.reflect.FieldMbr

fieldInTargetRef -> egl.lang.reflect.FieldMbr // Note: we will no longer do special processing for what used to be functionMemberRef. This is the same as a fieldRef

functionMemberRef -> egl.lang.reflect.FunctionMbr  

partRef -> egl.lang.reflect.Part

typeRef -> egl.lang.reflect.TypeType

recordRef -> egl.lang.reflect.RecordPart

serviceRef -> egl.lang.reflect.ServicePart

Here is a complete list of all the proxy parts:

• TypeType
• Part
• ClassPart
• RecordPart
• DelegatePart
• HandlerPart
• ProgramPart
• LibraryPart
• InterfacePart
• ServicePart
• Member
• FieldMbr
• FunctionMbr
• ConstructorMbr
• SequenceType
• FixedPrecisionType
• TimestampPatternType
• IntervalType

 

 

Compiler updates

The following changes are made to the compiler:

1. Package name is now represented as a string (for example "pkg1,pkg2") instead of a string[]. This is so that we are not tied to object identity of the string[] and allow us to easily change to a case sensitive package name.
2. All references to string interning should be removed. Instead, we will go through a common class to convert strings (for now, this means just lowercasing the string).
3. There will be new binders/binding completors for AnnotationType and StereotypeType. This is because these are not represented as records in the IRs.

Validation Design

TBD

Serialization Design

TBD