[antlr-interest] LL(1), LL(k), LL(*), predicated decisions

[Terence Parr]

Hi Jeff,

Thanks for the pointer.  Patterns are just names for stuff  
programmers know how to do.  In this case, I know how to build DFAs  
reasonably efficiently.  I was just trying to get java to listen to  
me. ;)  The other case of generating state machine DFA for even fixed  
lookahead decisions was a question of size, performance really not  
one of which DFA pattern to use.  If Java were more like C when it  
came to arrays, more opportunities would present themselves.  I have  
decided to keep these tables around for all decisions so target  
developers for C and friends have the choice.

I don't know anything about patterns...I just code.  If you could  
state simply the opportunity I'm missing, I'd appreciate it.  Might  
be a cool trick to learn in this area.

Thanks,
Ter

On Jun 5, 2006, at 9:25 PM, Jeff Barnes wrote:

> Terence,
>
> I've watched you talk about the problems of the
> implementation you chose for the cyclic DFAs for
> months now... Perhaps you should consider another
> alternative. Seems like you're trying to brute-force
> the Cargill pattern in the name of performance.
>
> http://www.geocities.com/jbarnesweb/fsmsurvey.pdf
>
> Consider the Knowledge pattern...
>
> Jeff