Hi,

I’m trying to come up with code completion for my MiniJava lab. On most of code positions i get NOCONTEXT(COMPLETION(…

For example for

someVar = this.

i get:

:: Completion triggered for: NOCONTEXT(COMPLETION(""))
   (context could not be parsed with this grammar; consider experimenting
   with additional productions for this partial construct, possibly
   marked with a {recover} annotation.) (candidate sorts: [VarDec, MethodDecl])

For things where i just need a variable such as:

someVar = ...

it works.

What am i doing wrong? The console out suggests to use recover rules, but very little documentation on that. I presume i should somehow modify my grammar to support partial constructs, but for trying something like:

VariableExpression "." ID "(" {Expression ","}* ")"				-> VariableExpression {cons("FunctionCall")}
VariableExpression "."	-> VariableExpression {cons("PartialFunctionCall"),recover}

doesn’t change anything.

I would appreciate any suggestions. Thank you.

Submitted by Vlad Vergu on 12 November 2010 at 20:37

As suggested by Lennart and Guido a valid context is given if the completion is triggered on a complete AST.

This is sufficient answer for me; i.e. question answered.

Details for others (MiniJava colleagues) running into this:

A complete AST in this case means a valid AST generated from (syntactically) valid MiniJava code:

MethodDecl(
    IntType()
  , "fac"
  , [Arg(IntType(), "num")]
  , []
  , []
  , ReturnStatement(
      FunctionCall(
        This()
      , "fa"
      , [MinusExpression(IntLit("1"), IntLit("1"))]
      )
    )
  )

If completion is triggered following “fa” in the function call a valid context will be given:
FunctionCall(This(),COMPLETION(“fa”),[MinusExpression(IntLit(“1”),IntLit(“1”))])

An incomplete AST would be something that is generated from syntactically invalid code. For example in my case triggering completion (at _) on:

class Fac extends _

gives NOCONTEXT whereas triggering on:

class Fac extends _ {
	
} 

gives:

Super(COMPLETION(""))

Thanks for posting your resolution here. This is an area for future improvement though; at some point we want to get rid of most of those NOCONTEXT cases or at least handle the case where the parser has insufficient information differently.