[antlr-interest] Why doesn't this work?

[Indhu Bharathi]

Hi Gavin,

Thanks for the reply. I had previously fixed it with gated symantic 
predicates with using LA(...). Instead using syntactic predicates as you 
have suggested looks like a more elegant option. Will try that. Thanks 
again.

Though the problem is solved now, I'm interested to know why the 
original example didn't work. That will produce a more readable grammar 
file, ofcourse will a little sacrifice on performance. One difference in 
the generated code I see is, INT_FLOAT_PATTERN() function is not using 
synpred*** at all. It is just guessing the alts using a dfa. However the 
dfa has something called special state transition which checks just one 
of the three syntactic predicates. Shouldn't INT_FLOAT_PATTERN() use all 
three synpred*** directly? I feel there is something going wrong in the 
code generation part.

Thanks, Indhu

Gavin Lambert wrote:
> At 03:12 9/04/2009, Indhu Bharathi wrote:
>> INT_FLOAT_PATTERN
>>     :    (NUMBER DOT NUMBER LETTER ) => NUMBER DOT NUMBER LETTER
>>         { $type=PATTERN; }
>>
>>     |    ( NUMBER DOT NUMBER ) =>  NUMBER DOT NUMBER
>>         { $type=FLOAT_LIT; }
>>
>>     |    (NUMBER) => NUMBER
>>         { $type=INT_LIT; }
>>
>>     ;
>
> I'm not sure why that isn't working, but you might want to left-factor 
> it a bit more than you have:
>
> INT_FLOAT_PATTERN /* or you could just make this INT_LIT */
>   : NUMBER { $type = INT_LIT; }
>     ( (DOT NUMBER) => DOT NUMBER { $type = FLOAT_LIT; }
>       (LETTER { $type = PATTERN; })?
>     )?
>   ;
>
> This should be more efficient (because it will only do the 
> lookahead-backtracking once), and hopefully it'll actually work 
> (though I haven't tested it).
>
>