Bug 413942

Description Markus Keller 2013-07-29 14:41:35 EDT

The ASTParser has various problems where it generates wrong source ranges with nontrivial annotations. While investigating these problems I found the problem with the wrong nesting of ArrayType nodes (bug 413569).

The problem is that various ASTConverter#retrieve*(..) methods make wrong assumptions about the possible form of annotations. E.g. retrieveExtraDimension(..) and retrieveEndOfElementTypeNamePosition(..) assume they can just skip "@AnnotationName", but that strategy fails as soon as the annotation has arguments. retrieveProperRightBracketPosition(..) and its callers also use a similarly doomed strategy in trying to find the brackets of an array type starting from the beginning of the array type.

The loops in the retrieve*(..) methods should only skip over TokenNameCOMMENT_* tokens, and otherwise give up when they encounter an unknown token id.

The strategy should not be to scan over annotations, but to use the annotation nodes' source ranges to update the start position and then only scan for the brackets after the annotation node.


Test case with all kinds of problems (but syntactically correct):

public class ArrayTypeAnnotations extends @NonNull(int[].class) Object {
    Object field = new ArrayList< @NonEmpty(Bla) int @NonNull(Bla) [] @NonEmpty(Blu) [ ]>() ;
    
    int[][] xxx;
    ArrayList<int[]> [][] yyy; // source ranges already broken in AST.JLS4
    ArrayList<int[][]> [][][][] zzz;
    ArrayList<Float> [][][] zzz2;
    Object a = new ArrayList< @TakeType(int[].class) int @TakeType(float.class) [] @TakeType(double.class) []>() ;
    Object b = new @NonNull(Math.PI) ArrayList< >() ;
    Object c = new ArrayList<@NonNull(Math.PI) Object[]>() ;

    void foo(int i) {
        for (String tab@TakeType(int[].class) [] = null;; ++i) { break; }
        for (@Deprecated String tab@TakeType(int[].class) [][]  = null;; ++i) {}
    }
}