accentuated chars in lexer makes gcj failed

Hello,

the following Lexer grammar :
CHAR  : ('a'..'z'|'A'..'Z'|'_'| '-'
         | '=E9' | '=E8' | '=EA' | '=EB'
         | '=E1' | '=E0' | '=E2' | '=E4'
         | '=FA' | '=F9' | '=FB' | '=FC'          | '=EE' | '=EF'
         | '=F4' | '=F6' );

Get translated in the following pieces of code
               case '-':
               {
                        match('-');
                        break;
               }
               case '\u00e9':
               {
                        match('=E9');
                        break;
               }

 It works well when compiling in java bytecode (javac), But when using
gcj, gcj complains about the '=E9' accentuated char :
   GraphesLexer.java:671: erreur: unrecognized character in input
stream.

 If I replace '=E9' by '\u00e9', it works like a charm.
 Are their  any reason why '=E9' is used instead of 'u00e9' ?
 (It is perhaps a bug of gcj, but the difference between the case
parameter and the match() paramter looks strange to the newbie I am)...


 -mat