Re: How to change default outcome of shift/reduce conflict?

[Anthony DeRobertis]

Hans Aberg writes: 

> I thought "OR" was a terminal, and "expr" a nonterminal that may expand to
> contain "OR" by the rule you listed as
>   expr -> expr OR expr

Yes, that's correct. But that's not the full story. I've posted it before, 
but here it is again: 

answer_statement: ANSWER expr answer_btn_list
                ...
              ; 

answer_btn_list: WITH answer_btn_list_oneplus
             | /* empty */
             ; 

answer_btn_list_oneplus: answer_btn_list_oneplus OR expr
                     | expr     %prec PSEUDO_MAX
                     ; 

expr contains a lot of things, expr OR expr is one of them. 

> Which is causing the shift/reduce conflicts. Your answer gives me the
> impression that you may not understand how a grammar works.

Correct me if I'm wrong, but given: 

  answer "Hello" with "Cancel" or "OK" 

the shift-reduce conflict is on the token "OR" after "Cancel". Bison could, 
due to the ambiguous grammar, either shift OR (leading to reduction by expr 
OR expr -> expr) or reduce (expr -> answer_btn_list_oneplus). 

By default, it decided on shift. I didn't like that, so I now have that 
%prec in there, which tells bison to reduce. 


> The question is really: How do you want the rules
>   expr -> expr OR expr
>   x -> ANSWER expr WITH expr OR expr
> to be applied when the expression
>   ANSWER expr WITH expr OR expr OR expr OR expr
> appears on top of the stack?

I hope the above answers that. 

> But you have not told us whether
>   answer "Hello" OR "world" with "Cancel" or "OK" or "Bye"
> is legal.

Well, syntacticly (is that a word?) it's legal. It would display a dialogue 
with three buttons, except for the runtime error from "hello" and "world" 
not being boolean values. 

> So then you must have yet another rule you have not told us, either
>   x -> ANSWER expr WITH "(" expr ")"
>   x -> ANSWER expr WITH expr

Actually, it's one of the many things in expr: '(' expr ')'. 

> I can only give my hunch of what the grammar might be (leaving it to Akim
> finding an efficient one):

I've tried changing the expr above to expr_orless and then adding a seperate 
expr: expr_orless | expr OR expr, but that led to reduce-reduce conflicts. I 
honestly didn't spend much time thinking about why.

pgpiOUhLjfkNe.pgp
Description: PGP signature