Re: Remove useless associativity/precedence from parsers

Tom,

> Le 23 mai 2019 à 06:00, Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us> a écrit :
>
> Robert Haas <robertmhaas(at)gmail(dot)com> writes:
>> Another thing is that it would be nice to have a better way of
>> resolving conflicts than attaching precedence declarations. Some
>> problems can't be solved that way at all, and others can only be
>> solved that way at the risk of unforeseen side effects.
>
> Yeah, we've definitely found that resolving shift/reduce conflicts via
> precedence declarations has more potential for surprising side-effects
> than one would think.

That's why in recent versions of Bison we also provide a means
to pure %expect directives on the rules themselves, to be more
precise about what happens.

> It feels to me that there's something basically
> wrong with that concept, or at least wrong with the way we've used it.

I'm trying to find means to scope the prec/assoc directives, because
they are too powerful, and that's dangerous. This is also why I try
to remove the useless ones.

Some people don't trust assoc/prec directives at all and use only
unambiguous grammars. But this can be very verbose...

I agree something is not so cool about these directives. GLR parsers
have a clear concept of in-between-rules precedence (%dprec). Something
similar for LR (hence fully static) would be nice, but it remains to
be invented.