[antlr-interest] Early exit exception in positive closures

[Andreas Meyer]

Sam Harwell schrieb:
> When at all possible, avoid using semantic predicates, as they will
> greatly slow down your grammar relative to a properly left-factored
> grammar. 
Hmhmh ... but as long as they are just {...}? predicates instead of the 
gated semantic predicates {..}?=>, aren't they only executed when there 
is an ambiguity? Or do they prevent ANTLR from generating something more 
clever?

> If the negated set works, it is probably your best option for
> now. Maybe I can add a subtract set operation you can test, although I
> couldn't check it in officially since it deviates from the grammar
> language spec. Then you could write the following with no predicates (no
> predicates with short lookahead is optimal):
>
> // the keyword rule must be a set for this to work, which is "any single
> one of the following tokens"
> (identifier | keyword-(KW123|KW456))+ rule_with_keywords?
>   

Something like this would be great. But. if it deviates from the ANTLR 
grammar, I have another suggestion that would not violate the garmmar 
language spec: would it be possible to remove KW123 and KW456 
automatically, by letting ANTLR look at "rule_with_keywords"? When 
setting an option { greedy=false; } ? :-) So, maybe there is a certain 
token set computed, that decides if one should stay within the loop or 
exit the loop. With greedy=false, the token-set could be reduced by 
FIRST(rule_with_keywords) or any other rule that may follow the loop.

I also thought about something like:

id
    : ID
    | KW1 { avoid }
    | KW2 { avoid }
    | KW3 { avoid }
    .
    .
    ;

which would mean that, whenever during grammar analysis there is an 
ambiguity between a token with "avoid" and the same token without avoid, 
prefer the one without avoid. Does this sound reasonable? I find it 
interesting :-)


> You never responded to my offer to check out making the ANTLR Tool
> faster at analyzing your grammar. 20 seconds is a long time and I'm sure
> it can be improved. :)
>
>   
I really appreciate that offer and would be happy if I could help 
improve ANTLR, but sending the grammar itself is not possible, 
unfortunately ... is there some way you could send me a patched version 
of ANTLR that produces a logfile, which I can send to you?

Greetings,
Andreas Meyer