Re: [boost] Re: Automata builder -- interface

From: Larry Evans (jcampbell3_at_[hidden])
Date: 2001-05-31 12:44:06

Douglas Gregor wrote:

> On Thursday 31 May 2001 11:45 am, you wrote:
> > --- In boost_at_y..., "joel de guzman" <isis-tech_at_m...> wrote:
> > > From: "Douglas Gregor"
> >
> > ...skipping
> >
> > > > Doug
> > >
> > > If we did it George Heintzelman's way, |= instead of = for new
> >
> > productions,
> >
> > > we don't have to pay for the feature if we don't use it. The penalty
> >
> > is the
> >
> > > added code to the = operator. If we let that as is (no redefines)
> >
> > and
> >
> > > use the |= for new productions (alternatives), then we have a
> >
> > win-win
> >
> > > situation: no penalty if the feature is not used and clarity in the
> > > imperative
> > > perspective.
> > >
> > > a = b;
> > > a |= c;
> > > a |= d;
> > > if (flag) a |= e;
> >
> > Someone seeing |= might think >>= is also defined.  Should
> > this also be provided?  Then what about &=?
>
> >>= and &= don't make sense in the context of a (E)BNF grammar. For instance:
>
> Rule<> A = B >> C
>                  | B >> D;
> A &= F;
>
> This would be equivalent to:
>
> Rule<> A = B >> C >> F
>                 | B >> D >> F;
>
> With |=, it merely adds another alternative production, instead of modifying
> each of the productions that already exist.

OK.  But now I'm wondering whether |= is a good idea.  For example:

   ChLit<> a('a');
   ChLit<> b('b');
   Rule<> As_b = b;
   As_b |= a>>A;

Does the final value of As_b represent the following BNF:

  As_b ::= 'a' As_b | 'b'

Or the following:

   As_b ::= 'a' 'b' | 'b'

?