[antlr-interest] NoViableAltException in optional alternative

[Terence Parr]

i have a bug to look at optional subrules on end of rules. use another  
rule that invokes nubmers EOF.
Ter
On May 24, 2008, at 11:04 AM, Brent Yates wrote:

> If the prediction DFA for an alternative fails but the alternative  
> is optional, should the rule throw a NoViableAltException?  It seems  
> to me that it should not, but the code generated by Antlr 3.1b1 does  
> throw the exception.  Here is a short silly grammar that shows the  
> problem.  If you runs this in AntlrWorks and test just the numbers  
> rule with an input of '5', you will get NoViableAltExceptions.
>
> grammar Bug;
>
> options
>     {
>     language= Java;
>     }
>
> start
>     :    declarations EOF
>     ;
>
> declarations
>     :    numbers ';'
>     ;
>
> numbers
>     :    '5'
>         (
>             '.'
>         |    '&' '%' '%'
>         |    '&' '%' '^'
>         |    '%' '$'
>         |    '%' '^'
>         )?
>     ;
>
> If one calls the 'numbers' rule in isolation (for example during  
> unit testing or if you are building an interpreter) and feeds a  
> single '5' to it followed by EOF (or anything other than semicolon),  
> the alternative prediction DFA inside the numbers function fails and  
> a NoViableAltException is thrown.  I belive this is incorrect as the  
> numbers rule should have succeeded since the final alternative is  
> optional.  The extra input should be passed up the chain for other  
> rules to verify.
>
> Antlr is also not consistent in this behavior.  Making minor changes  
> to the grammar will sometimes cause the generated code to NOT use a  
> DFA for the alternative check and in this case no exception is  
> thrown when the alternative does not match - reguardless of the  
> reason for not matching (i.e., extra input that is not in the follow  
> set).
>
> The inconsistency makes it hard to craft gUnit tests as the rule  
> tests work sometimes and then later fail (due to  
> NoViableAltExceptions) after unrelated changes to the grammar.  I  
> also think this could cause problems for applications where the  
> rules need to be called independently.
>
> Thanks,
>
> Brent Yates
> brent.yates at gmail.com
>
>
>
>
>
>