Re: Localized precedence declarations

[Hans Aberg]

On 23 Jun 2009, at 20:29, Michiel Helvensteijn wrote:

>  The operator precedences in my language are
> static. I just need a way to isolate the different groupings of  
> precedence
> declarations, so that no shift/reduce conflict is resolved by two
> declarations from different groupings.

My method can be worked by hand. This is in fact the safe part; I also  
made a part where LR is directly modified, and I need to check Knuth's  
paper on it - on the todo list :-).

> Take this example:
>
> %nonassoc IF_THEN_RULE
> %nonassoc "else"
>
> %left '+' '-'
> %left '*'
>
> It generates the following relative orders:
>
> IF_THEN_RULE < "else"
> IF_THEN_RULE < '+'
> IF_THEN_RULE < '-'
> IF_THEN_RULE < '*'
> "else" < '+'
> "else" < '-'
> "else" < '*'
> '+' < '*'
> '- < '*'
>
> But I would prefer if only the following relations were understood:
>
> IF_THEN_RULE < "else"
> '+' < '*'
> '- < '*'
>
> All the others were unintentional, and might silently (and badly)  
> resolve a
> shift/reduce conflict that I introduced by accident.

What about adding a '%nonassoc "then"', which resolve the dangling  
else. The other things should follow.

If that does not help, I can send you my paper (and also if you just  
want it :-)).

  Hans