[antlr-interest] bad prediction on common prefix despite enough lookahead depth

[Maurizio de Pascale]

Using the small grammar below ANTLR 2.7.7 reports no warning but the 
generated lexer fails to parse the input ">>":
in fact, the code generated for nextToken() uses only LA(1) and LA(2) to 
predict the C token, despite the lookahead depth set to 3, and then 
throws for the missing '='...

Here are the questions:
Is this a bug or am I missing something?
Is   D: (">>=")=> C | B | A;   the best solution?
Is this related to the linear approximate lookahead?
should it spit out at least a warning?
does ANTLR3 correctly handle this?


thank you in advance,
Maurizio de Pascale
mdepascale at dii.unisi.it



TEST.G:

options {language="Cpp";}

class TestLexer extends Lexer;
options {k=3;}

A : '>' ;
B : '=' ;
C : ">>=" ;