Re: Reductions during Bison error handling

[Paul Eggert]

> Date: Tue, 21 May 2002 13:12:37 -0600 (MDT)
> From: Richard Stallman <address@hidden>
> 
> We certainly must not assume nobody depends on this merely because
> it isn't documented.
 
Paul H. is suggesting that we alter Bison to match yacc's behavior, so
this will hurt only programs that require Bison and do not work with
yacc.  I just checked a wide variety of free programs, and the only
ones that I found required Bison were PostgreSQL, getdate.y and
plural.y (the last two are used in several GNU utilities).  However,
none of the grammars used in these programs mention the 'error'
nonterminal, so the problem can't occur with them.

So it appears to be safe to assume that nobody depends on the current
Bison behavior in this area.

I'd like to see the Paul H.'s fix before it gets installed, but from
what I've seen so far it looks like a good idea.