The nondeterminism warning comes from "linear approximate lookahead".
Antlr may still generate the right code. if so you can either shutdown
the warning or workaround the grammar(left factor).
Gokhan Caglar wrote:
> Here are two grammars which I think are semantically equal, look at
> the do loop in the generated code. I know I can left factor DO's but
> I am trying to demonstrate a behavior. The first one reports a non
> determinism:
>
> For the grammar:
>
> options {
> k = 2;
> }
>
> batch
> :
> (statement1)*
> (statement2)?
> ;
>
> statement1
> :
> DO THIS
> | BREAK
> ;
>
> statement2
> : DO THAT
> ;
>
>
> This code gets generated:
>
> public final void batch() throws RecognitionException,
> TokenStreamException {
> try { // for error handling
> {
> _loop6:
> do {
> if
> ((LA(1)==DO||LA(1)==BREAK) && (_tokenSet_1.member(LA(2)))) {
> statement();
> }
> else {
> break _loop6;
> }
> } while (true);
> }
> {
> switch ( LA(1)) {
> case DO:
> {
> statement2();
> break;
> }
> case EOF:
> case GO:
> {
> break;
> }
> default:
> {
> throw new
> NoViableAltException(LT(1), getFilename());
> }
> }
> }
> }
> catch (RecognitionException ex) {
> reportError(ex);
> recover(ex,_tokenSet_2);
> }
> }
>
>
>
> For the grammar:
>
> options {
> k = 2;
> }
>
> batch
> :
> (statement1 | statement3)*
> (statement2)?
> ;
>
> statement1
> :
> DO THIS
> ;
>
> statement3
> :
> BREAK
> ;
>
> statement2
> : DO THAT
> ;
>
>
> This code gets generated:
>
> public final void batch() throws RecognitionException,
> TokenStreamException {
> try { // for error handling
> {
> _loop6:
> do {
> if ((LA(1)==DO) &&
> (LA(2)==THIS)) {
> statement1();
> }
> else if ((LA(1)==BREAK)) {
> statement3();
> }
> else {
> break _loop6;
> }
> } while (true);
> }
> {
> switch ( LA(1)) {
> case DO:
> {
> statement2();
> break;
> }
> case EOF:
> case GO:
> {
> break;
> }
> default:
> {
> throw new
> NoViableAltException(LT(1), getFilename());
> }
> }
> }
> }
> catch (RecognitionException ex) {
> reportError(ex);
> recover(ex,_tokenSet_1);
> }
> }
>
> It goes back to the first generated code, and reports non-determinism
> if we do this:
> batch
> :
> ((statement1 | statement3))* // Adding an extra paranthesis
> (statement2)?
> ;
>
>
> Thanks,
> Gokhan
>
--
Xue Yong Zhi
http://seclib.blogspot.com