Enable inferred callout-to-field bindings

[mosconi]

It would be nice to have an option to use inferred callouts also for field accesses. Instead of refactoring get/set field accesses into calls to get/set-method callouts in the source code, the compiler could perhaps perform such mapping behind the scene. (comparable to C# Properties).

[stephan]

Part of this issue has been addressed in

• OTJLD 3.5(h)
• 3.3.20-otjld-inferred-callout-to-field-1 ff.
• r17718

This part infers the use of an explicitly declared callout-to-field for what syntactically would be a field access.

For the remaining I'm uncertain how much automagic the compiler should implement:

1. From a getter/setter call within a role infer a corresponding callout-to-field.
2. From an unresolved field access infer both the callout-to-field and replace the field access with a getter/setter call.

If we would support all of these, should all be controlled by the same warning token (concerning @SuppressWarnings("inferredcallout") and concerning the global compiler option)?

[stephan]

I've uploaded reports of the new tests, but the link I gave is broken.

Correct links to tests are

• 3.3.20-otjld-inferred-callout-to-field-1
• 3.3.20-otjld-inferred-callout-to-field-2
• 3.3.20-otjld-inferred-callout-to-field-3
• 3.3.20-otjld-inferred-callout-to-field-4

[stephan]

Well, I did add some automagic (item 2. in the previous comment).

Test cases are

• 3.3.20-otjld-inferred-callout-to-field-5
• 3.3.20-otjld-inferred-callout-to-field-6

I believe this services the original intention. Still open issues:

• should method calls be analyzed by naming convention to infer a matching callout-to-field? E.g., given an unresolved call this.setVal("x"), should a callout binding void setVal(String v) -> set String val; be generated if the corresponding base field exists?
• What kinds of adaptations should be inferred / should be possible via explicit callout-to-field? Currently only those explicit callout-to-fields are used where field type and method signature use the same type.
• What levels of quickfix should be supported ("materialize inferred callout")? (I've just opened #32 for this).

I'm closing this issue in the light of the original intention. Please open new issues for things you'd like to see added.