Path: utzoo!utgpu!jarvis.csri.toronto.edu!cs.utexas.edu!wuarchive!decwrl!shelby!unix!hplabs!hp-ses!hpcuhb!hpcllla!hpclisp!hpclscu!shankar
From: shankar@hpclscu.HP.COM (Shankar Unni)
Newsgroups: comp.lang.c++
Subject: Re: zortech problem with lex
Message-ID: <58170007@hpclscu.HP.COM>
Date: 1 Mar 90 22:11:26 GMT
References: <6300008@ux1.cso.uiuc.edu>
Organization: Hewlett-Packard Calif. Language Lab
Lines: 23

> I saw that comment about C++ not being LALR(1) in the g++ distribution,
> and at the time believed it.  However, my biggest thrill at OOPSLA '89
> was sitting at the same table as Bjarne Stroustrup for a few hours drinking
> free beer.  He said that he IS USING A LALR(1) GRAMMER for C++ 2.0.  
> He appeared sober & honest, therefore I assume that C++ is LALR(1), but 
> non-trivial.

No, it becomes LALR(1) only by a trick: a feedback mechanism by which the
lexer can:

   (a) consult a symbol table to see if an identifier is really a
       typedef name, and
   (b) an ambiguity resolver called by the parser to handle any remaining
       ambiguities.

The whole thing is done by adding another layer between the lexer and the
parser, with these smarts built in.

Using the classical model of a table-driven scanner and parser, it is
impossible to write a LALR(1) description for C++ (or for that matter, C).
It could be LR(2), though; I can't be sure..
----
Shankar Unni.