Re: Determining LR(k)-ness

[Hans Aberg]

On 18 Nov 2009, at 23:41, John Levine wrote:

> > > For an arbitrary .y file, I'd like to know if it's possible to get
> > > some report detail that would tell me (when using the %glr-parser
> > > option) what was the maximum 'k' lookahead used for that grammar.
> >
> > Unless it has been changed lately, it just uses LALR(1) and splits  
> > the
> > parse when ambiguous.
>
> That's what it does.  Bison can tell you whether a grammar is LALR(1),
> but if it's not it can't tell you what class of language it is.

Only that there is some work going on to do a LR(1). But it may not  
have reached the %glr feature then.

> It is my impression that most practical GLR grammars have parses that
> collapse pretty quickly after they split, but I don't have any
> statistics to back that up.

That is supposed to be true about natural languages, too. Perhaps that  
is the reason - human create both. But human languages have context  
dependent tokenization, which cannot be handled by a traditional Flex- 
Bison setup.

  Hans