[antlr-interest] How to make a syntactic predicate exit a rule completely

[Naveen Chawla]

Dear Jim,

I've solved it!

I did so as follows:

 (simplified)

grammar conjunctive;    //using ANTLRWorks

englishSentence : (adverbPhrase ','?)* simpleSentence;
           adverbPhrase: 'con' simpleSentence | prepPhr;

           simpleSentence : nounPhrase verbPhrase;

                       nounPhrase: ('noun' prepPhr)=> 'noun' prepPhr |
'noun';

                                  prepPhr: 'prep' nounPhrase;

                       //rule changed
                       verbPhrase: ('verb' object object? verbPhrase)=>
'verb' adverbPhrase? | 'verb' complement? adverbPhrase?;

                                   complement: (object object)=> object
object | object;

                                               object: nounPhrase;

Which is to say that "if the verb is followed by an object (or two) but then
followed by a verbPhrase, discount the fact that there may be a complement
for that verb, otherwise act normally as though there might".  It doesn't
handle two *objects* properly but I can easily do that now.

My entire English structure grammar is done now I believe, subject to
further tests. You helped clarify some things.


2009/10/18 Naveen Chawla <naveen.chwl at googlemail.com>

> Dear Jim,
>
> I agree with you. I am on what I believe to be the last "hurdle" of my
> grammar:
>
> (simplified)
>
> grammar conjunctive;    //using ANTLRWorks
>
> englishSentence : (adverbPhrase ','?)* simpleSentence;
>            adverbPhrase: 'con' simpleSentence | prepPhr;
>
>            simpleSentence : nounPhrase verbPhrase;
>
>                        nounPhrase: ('noun' prepPhr)=> 'noun' prepPhr |
> 'noun';
>
>                                   prepPhr: 'prep' nounPhrase;
>
>                        verbPhrase: 'verb' complement? adverbPhrase?;
>
>                                    complement: (object object)=> object
> object | object;
>
>                                                object:
> nounPhrase;
>
> Problem sentence: "con noun verb noun verb" SHOULD give adverbPhrase
> simpleSentence (but doesn't, instead consuming the second "noun" as a
> *complement* to the verbPhrase, leaving an impossible simpleSentence)
>
> My proposed solution was: Make sure that for something to be an "object",
> it must not be followed by a verbPhrase. This is so that it is not consumed
> by a possible *complement?*, IF it is part of a new simpleSentence (i.e. if
> it is the subject of a sentence).
> I believe my entire grammar is completed, as soon as this problem is fixed.
>
> 2009/10/19 Jim Idle <jimi at temporal-wave.com>
>
>>    I really think you are beginning to make this way more complicated
>> than it really is. You need to step away from your current pursuit of
>> predicates and so on and go back to fundamentals. You are trying to get the
>> parser to do way more than parser should do. I know it is tempting because
>> it seems that the parser will do a lot of validation for you, but it is
>> (generally) the wrong way. You want your parser to be as simple as possible,
>> accepting anything that is good syntax from a pure string together of tokens
>> point of view, THEN apply semantics to see if anything fails. This way you r
>> parser rules can accept all the clauses that might be correct (as in all
>> syntactically sound combinations), then reject the incorrect ones by
>> semantic analysis. Your parser will be simpler, your users will thank you
>> for the improved error messages and the whole thing will be easier to
>> maintain.
>>
>>
>>
>> Basically if you find yourself in the nitty-gritty like this, you are
>> probably heading down the wrong path or flogging a dead horse ;-) You could
>> use backtracking mode for this, but then your error messages will be even
>> more arbitrary.
>>
>>
>>
>> For the predicate, don’t forget that the prediction will still select a
>> rule and try the predicate. Perhaps you could use gated predicates on the
>> ‘invocation’ of the rule so that it is not ‘called’ if it won’t match, but
>> once you are in a rule you have to do something like this;
>>
>> r : (pred1)=>rule1
>>
>>  | (pred2)=>rule2
>>
>>  | // neither of the above, match nothing
>>
>>  ;
>>
>>
>>
>> But then you start need to set state flags, test the flags, etc. You soon
>> end up with a lot more work than just using a semantic analysis phase after
>> the initial parse. You will also have to deal with the absence of flags due
>> to syntax errors and so on, whereas if you defer to semantic analysis as a
>> separate phase, then you will be able to rely on a sound syntax.
>>
>>
>>
>> So:
>>
>>
>>
>> 1)      Don’t try to influence the paths the parser takes – let the
>> tokens do that;
>>
>> 2)      Afterwards, in the tree walk, determine whether the various
>> combinations of clauses are correct, issue good errors here if not;
>>
>> 3)      Generate code (or whatever it is you are doing) in the knowledge
>> that it is all good from here.
>>
>>
>>
>> Jim
>>
>>
>>
>> *From:* antlr-interest-bounces at antlr.org [mailto:
>> antlr-interest-bounces at antlr.org] *On Behalf Of *Naveen Chawla
>> *Sent:* Saturday, October 17, 2009 10:45 PM
>> *To:* antlr-interest at antlr.org
>> *Subject:* [antlr-interest] How to make a syntactic predicate exit a rule
>> completely
>>
>>
>>
>> Ok, so even if all syntactic predicates have failed in a particular
>> rule, that rule is still included as "matched" in that analysis! Is this
>> true or are my tests wrong? This is not what I hoped. I was hoping that upon
>> complete failure of all syntactic predicates in a rule, the rule becomes
>> disregarded, and that "disregarding" is cascaded as far up as possible, and
>> an alternative path is attempted. Is this how it's works, or am I missing
>> something? Or is there a way, using actions, to "force" the exiting of a
>> particular rule as if it had not been matched at all?
>>
>>
>>
>> Many regards,
>>
>> N
>>
>>
>> List: http://www.antlr.org/mailman/listinfo/antlr-interest
>> Unsubscribe:
>> http://www.antlr.org/mailman/options/antlr-interest/your-email-address
>>
>>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://www.antlr.org/pipermail/antlr-interest/attachments/20091019/58353daf/attachment.html